In this document we report on a user-scenario-based evaluation aiming at assessing the performance of machine translation (MT) systems in a real context of use. We describe a sequel of experiments that has been performed to estimate the usefulness of MT and to test if improvements of MT technology lead to better performance in the usage scenario. One goal is to find the best methodology for evaluating the eventual benefit of a machine translation system in an application. The evaluation is based on the QTLeap corpus, a novel multilingual language resource that was collected through a real-life support service via chat. It is composed of naturally occurring utterances produced by users while interacting with a human technician providing answers. The corpus is available in eight different languages: Basque, Bulgarian, Czech, Dutch, English, German, Portuguese and Spanish.
Out-of-vocabulary words (OOVs) are a ubiquitous and difficult problem in statistical machine translation (SMT). This paper studies different strategies of using BabelNet to alleviate the negative impact brought about by OOVs. BabelNet is a multilingual encyclopedic dictionary and a semantic network, which not only includes lexicographic and encyclopedic terms, but connects concepts and named entities in a very large network of semantic relations. By taking advantage of the knowledge in BabelNet, three different methods ― using direct training data, domain-adaptation techniques and the BabelNet API ― are proposed in this paper to obtain translations for OOVs to improve system performance. Experimental results on English―Polish and English―Chinese language pairs show that domain adaptation can better utilize BabelNet knowledge and performs better than other methods. The results also demonstrate that BabelNet is a really useful tool for improving translation performance of SMT systems.
The present work is an overview of the TraMOOC (Translation for Massive Open Online Courses) research and innovation project, a machine translation approach for online educational content. More specifically, videolectures, assignments, and MOOC forum text is automatically translated from English into eleven European and BRIC languages. Unlike previous approaches to machine translation, the output quality in TraMOOC relies on a multimodal evaluation schema that involves crowdsourcing, error type markup, an error taxonomy for translation model comparison, and implicit evaluation via text mining, i.e. entity recognition and its performance comparison between the source and the translated text, and sentiment analysis on the students’ forum posts. Finally, the evaluation output will result in more and better quality in-domain parallel data that will be fed back to the translation engine for higher quality output. The translation service will be incorporated into the Iversity MOOC platform and into the VideoLectures.net digital library portal.
While an increasing number of (automatic) metrics is available to assess the linguistic quality of machine translations, their interpretation remains cryptic to many users, specifically in the translation community. They are clearly useful for indicating certain overarching trends, but say little about actual improvements for translation buyers or post-editors. However, these metrics are commonly referenced when discussing pricing and models, both with translation buyers and service providers. With the aim of focusing on automatic metrics that are easier to understand for non-research users, we identified Edit Distance (or Post-Edit Distance) as a good fit. While Edit Distance as such does not express cognitive effort or time spent editing machine translation suggestions, we found that it correlates strongly with the productivity tests we performed, for various language pairs and domains. This paper aims to analyse Edit Distance and productivity data on a segment level based on data gathered over some years. Drawing from these findings, we want to then explore how Edit Distance could help in predicting productivity on new content. Some further analysis is proposed, with findings to be presented at the conference.
We present a freely available corpus containing source language texts from different domains along with their automatically generated translations into several distinct morphologically rich languages, their post-edited versions, and error annotations of the performed post-edit operations. We believe that the corpus will be useful for many different applications. The main advantage of the approach used for creation of the corpus is the fusion of post-editing and error classification tasks, which have usually been seen as two independent tasks, although naturally they are not. We also show benefits of coupling automatic and manual error classification which facilitates the complex manual error annotation task as well as the development of automatic error classification tools. In addition, the approach facilitates annotation of language pair related issues.
Existing Arabic sentiment lexicons have low coverage―with only a few thousand entries. In this paper, we present several large sentiment lexicons that were automatically generated using two different methods: (1) by using distant supervision techniques on Arabic tweets, and (2) by translating English sentiment lexicons into Arabic using a freely available statistical machine translation system. We compare the usefulness of new and old sentiment lexicons in the downstream application of sentence-level sentiment analysis. Our baseline sentiment analysis system uses numerous surface form features. Nonetheless, the system benefits from using additional features drawn from sentiment lexicons. The best result is obtained using the automatically generated Dialectal Hashtag Lexicon and the Arabic translations of the NRC Emotion Lexicon (accuracy of 66.6%). Finally, we describe a qualitative study of the automatic translations of English sentiment lexicons into Arabic, which shows that about 88% of the automatically translated entries are valid for English as well. Close to 10% of the invalid entries are caused by gross mistranslations, close to 40% by translations into a related word, and about 50% by differences in how the word is used in Arabic.
Sentiment Analysis systems aims at detecting opinions and sentiments that are expressed in texts. Many approaches in literature are based on resources that model the prior polarity of words or multi-word expressions, i.e. a polarity lexicon. Such resources are defined by teams of annotators, i.e. a manual annotation is provided to associate emotional or sentiment facets to the lexicon entries. The development of such lexicons is an expensive and language dependent process, making them often not covering all the linguistic sentiment phenomena. Moreover, once a lexicon is defined it can hardly be adopted in a different language or even a different domain. In this paper, we present several Distributional Polarity Lexicons (DPLs), i.e. large-scale polarity lexicons acquired with an unsupervised methodology based on Distributional Models of Lexical Semantics. Given a set of heuristically annotated sentences from Twitter, we transfer the sentiment information from sentences to words. The approach is mostly unsupervised, and experimental evaluations on Sentiment Analysis tasks in two languages show the benefits of the generated resources. The generated DPLs are publicly available in English and Italian.
In this paper, we analyze the sentiments derived from the conversations that occur in social networks. Our goal is to identify the sentiments of the users in the social network through their conversations. We conduct a study to determine whether users of social networks (twitter in particular) tend to gather together according to the likeness of their sentiments. In our proposed framework, (1) we use ANEW, a lexical dictionary to identify affective emotional feelings associated to a message according to the Russell’s model of affection; (2) we design a topic modeling mechanism called Sent_LDA, based on the Latent Dirichlet Allocation (LDA) generative model, which allows us to find the topic distribution in a general conversation and we associate topics with emotions; (3) we detect communities in the network according to the density and frequency of the messages among the users; and (4) we compare the sentiments of the communities by using the Russell’s model of affect versus polarity and we measure the extent to which topic distribution strengthen likeness in the sentiments of the users of a community. This works contributes with a topic modeling methodology to analyze the sentiments in conversations that take place in social networks.
A key point in Sentiment Analysis is to determine the polarity of the sentiment implied by a certain word or expression. In basic Sentiment Analysis systems this sentiment polarity of the words is accounted and weighted in different ways to provide a degree of positivity/negativity. Currently words are also modelled as continuous dense vectors, known as word embeddings, which seem to encode interesting semantic knowledge. With regard to Sentiment Analysis, word embeddings are used as features to more complex supervised classification systems to obtain sentiment classifiers. In this paper we compare a set of existing sentiment lexicons and sentiment lexicon generation techniques. We also show a simple but effective technique to calculate a word polarity value for each word in a domain using existing continuous word embeddings generation methods. Further, we also show that word embeddings calculated on in-domain corpus capture the polarity better than the ones calculated on general-domain corpus.
Emotion Recognition (ER) is an important part of dialogue analysis which can be used in order to improve the quality of Spoken Dialogue Systems (SDSs). The emotional hypothesis of the current response of an end-user might be utilised by the dialogue manager component in order to change the SDS strategy which could result in a quality enhancement. In this study additional speaker-related information is used to improve the performance of the speech-based ER process. The analysed information is the speaker identity, gender and age of a user. Two schemes are described here, namely, using additional information as an independent variable within the feature vector and creating separate emotional models for each speaker, gender or age-cluster independently. The performances of the proposed approaches were compared against the baseline ER system, where no additional information has been used, on a number of emotional speech corpora of German, English, Japanese and Russian. The study revealed that for some of the corpora the proposed approach significantly outperforms the baseline methods with a relative difference of up to 11.9%.
We address the task of automatically estimating the missing values of linguistic features by making use of the fact that some linguistic features in typological databases are informative to each other. The questions to address in this work are (i) how much predictive power do features have on the value of another feature? (ii) to what extent can we attribute this predictive power to genealogical or areal factors, as opposed to being provided by tendencies or implicational universals? To address these questions, we conduct a discriminative or predictive analysis on the typological database. Specifically, we use a machine-learning classifier to estimate the value of each feature of each language using the values of the other features, under different choices of training data: all the other languages, or all the other languages except for the ones having the same origin or area with the target language.
We present a study of the adequacy of current methods that are used for POS-tagging historical Dutch texts, as well as an exploration of the influence of employing different techniques to improve upon the current practice. The main focus of this paper is on (unsupervised) methods that are easily adaptable for different domains without requiring extensive manual input. It was found that modernising the spelling of corpora prior to tagging them with a tagger trained on contemporary Dutch results in a large increase in accuracy, but that spelling normalisation alone is not sufficient to obtain state-of-the-art results. The best results were achieved by training a POS-tagger on a corpus automatically annotated by projecting (automatically assigned) POS-tags via word alignments from a contemporary corpus. This result is promising, as it was reached without including any domain knowledge or context dependencies. We argue that the insights of this study combined with semi-supervised learning techniques for domain adaptation can be used to develop a general-purpose diachronic tagger for Dutch.
In this paper we present a language resource for German, composed of a list of 1,021 unique errors extracted from a collection of texts written by people with dyslexia. The errors were annotated with a set of linguistic characteristics as well as visual and phonetic features. We present the compilation and the annotation criteria for the different types of dyslexic errors. This language resource has many potential uses since errors written by people with dyslexia reflect their difficulties. For instance, it has already been used to design language exercises to treat dyslexia in German. To the best of our knowledge, this is first resource of this kind in German.
In this paper, we present the CItA corpus (Corpus Italiano di Apprendenti L1), a collection of essays written by Italian L1 learners collected during the first and second year of lower secondary school. The corpus was built in the framework of an interdisciplinary study jointly carried out by computational linguistics and experimental pedagogists and aimed at tracking the development of written language competence over the years and students’ background information.
Part-of-speech (POS) induction is one of the most popular tasks in research on unsupervised NLP. Various unsupervised and semi-supervised methods have been proposed to tag an unseen language. However, many of them require some partial understanding of the target language because they rely on dictionaries or parallel corpora such as the Bible. In this paper, we propose a different method named delexicalized tagging, for which we only need a raw corpus of the target language. We transfer tagging models trained on annotated corpora of one or more resource-rich languages. We employ language-independent features such as word length, frequency, neighborhood entropy, character classes (alphabetic vs. numeric vs. punctuation) etc. We demonstrate that such features can, to certain extent, serve as predictors of the part of speech, represented by the universal POS tag.
The SpeDial consortium is sharing two datasets that were used during the SpeDial project. By sharing them with the community we are providing a resource to reduce the duration of cycle of development of new Spoken Dialogue Systems (SDSs). The datasets include audios and several manual annotations, i.e., miscommunication, anger, satisfaction, repetition, gender and task success. The datasets were created with data from real users and cover two different languages: English and Greek. Detectors for miscommunication, anger and gender were trained for both systems. The detectors were particularly accurate in tasks where humans have high annotator agreement such as miscommunication and gender. As expected due to the subjectivity of the task, the anger detector had a less satisfactory performance. Nevertheless, we proved that the automatic detection of situations that can lead to problems in SDSs is possible and can be a promising direction to reduce the duration of SDS’s development cycle.
This paper describes a method to automatically create dialogue resources annotated with dialogue act information by reusing existing dialogue corpora. Numerous dialogue corpora are available for research purposes and many of them are annotated with dialogue act information that captures the intentions encoded in user utterances. Annotated dialogue resources, however, differ in various respects: data collection settings and modalities used, dialogue task domains and scenarios (if any) underlying the collection, number and roles of dialogue participants involved and dialogue act annotation schemes applied. The presented study encompasses three phases of data-driven investigation. We, first, assess the importance of various types of features and their combinations for effective cross-domain dialogue act classification. Second, we establish the best predictive model comparing various cross-corpora training settings. Finally, we specify models adaptation procedures and explore late fusion approaches to optimize the overall classification decision taking process. The proposed methodology accounts for empirically motivated and technically sound classification procedures that may reduce annotation and training costs significantly.
In this paper, we present a taxonomy of stories told in dialogue. We based our scheme on prior work analyzing narrative structure and method of telling, relation to storyteller identity, as well as some categories particular to dialogue, such as how the story gets introduced. Our taxonomy currently has 5 major dimensions, with most having sub-dimensions - each dimension has an associated set of dimension-specific labels. We adapted an annotation tool for this taxonomy and have annotated portions of two different dialogue corpora, Switchboard and the Distress Analysis Interview Corpus. We present examples of some of the tags and concepts with stories from Switchboard, and some initial statistics of frequencies of the tags.
PentoRef is a corpus of task-oriented dialogues collected in systematically manipulated settings. The corpus is multilingual, with English and German sections, and overall comprises more than 20000 utterances. The dialogues are fully transcribed and annotated with referring expressions mapped to objects in corresponding visual scenes, which makes the corpus a rich resource for research on spoken referring expressions in generation and resolution. The corpus includes several sub-corpora that correspond to different dialogue situations where parameters related to interactivity, visual access, and verbal channel have been manipulated in systematic ways. The corpus thus lends itself to very targeted studies of reference in spontaneous dialogue.
Spoken conversation corpora often adapt existing Dialogue Act (DA) annotation specifications, such as DAMSL, DIT++, etc., to task specific needs, yielding incompatible annotations; thus, limiting corpora re-usability. Recently accepted ISO standard for DA annotation – Dialogue Act Markup Language (DiAML) – is designed as domain and application independent. Moreover, the clear separation of dialogue dimensions and communicative functions, coupled with the hierarchical organization of the latter, allows for classification at different levels of granularity. However, re-annotating existing corpora with the new scheme might require significant effort. In this paper we test the utility of the ISO standard through comparative evaluation of the corpus-specific legacy and the semi-automatically transferred DiAML DA annotations on supervised dialogue act classification task. To test the domain independence of the resulting annotations, we perform cross-domain and data aggregation evaluation. Compared to the legacy annotation scheme, on the Italian LUNA Human-Human corpus, the DiAML annotation scheme exhibits better cross-domain and data aggregation classification performance, while maintaining comparable in-domain performance.
This paper presents WikiCoref, an English corpus annotated for anaphoric relations, where all documents are from the English version of Wikipedia. Our annotation scheme follows the one of OntoNotes with a few disparities. We annotated each markable with coreference type, mention type and the equivalent Freebase topic. Since most similar annotation efforts concentrate on very specific types of written text, mainly newswire, there is a lack of resources for otherwise over-used Wikipedia texts. The corpus described in this paper addresses this issue. We present a freely available resource we initially devised for improving coreference resolution algorithms dedicated to Wikipedia texts. Our corpus has no restriction on the topics of the documents being annotated, and documents of various sizes have been considered for annotation.
The aim of distributional semantics is to model the similarity of the meaning of words via the words they occur with. Thereby, it relies on the distributional hypothesis implying that similar words have similar contexts. Deducing meaning from the distribution of words is interesting as it can be done automatically on large amounts of freely available raw text. It is because of this convenience that most current state-of-the-art-models of distributional semantics operate on raw text, although there have been successful attempts to integrate other kinds of―e.g., syntactic―information to improve distributional semantic models. In contrast, less attention has been paid to semantic information in the research community. One reason for this is that the extraction of semantic information from raw text is a complex, elaborate matter and in great parts not yet satisfyingly solved. Recently, however, there have been successful attempts to integrate a certain kind of semantic information, i.e., co-reference. Two basically different kinds of information contributed by co-reference with respect to the distribution of words will be identified. We will then focus on one of these and examine its general potential to improve distributional semantic models as well as certain more specific hypotheses.
This paper presents the adaptation of an Entity Centric Model for Portuguese coreference resolution, considering 10 named entity categories. The model was evaluated on named e using the HAREM Portuguese corpus and the results are 81.0% of precision and 58.3% of recall overall, the resulting system is freely available
This paper presents a data-driven co-reference resolution system for German that has been adapted from IMS HotCoref, a co-reference resolver for English. It describes the difficulties when resolving co-reference in German text, the adaptation process and the features designed to address linguistic challenges brought forth by German. We report performance on the reference dataset TüBa-D/Z and include a post-task SemEval 2010 evaluation, showing that the resolver achieves state-of-the-art performance. We also include ablation experiments that indicate that integrating linguistic features increases results. The paper also describes the steps and the format necessary to use the resolver on new texts. The tool is freely available for download.
This paper describes a coreference annotation scheme, coreference annotation specific issues and their solutions through our proposed annotation scheme for Hindi. We introduce different co-reference relation types between continuous mentions of the same coreference chain such as “Part-of”, “Function-value pair” etc. We used Jaccard similarity based Krippendorff‘s’ alpha to demonstrate consistency in annotation scheme, annotation and corpora. To ease the coreference annotation process, we built a semi-automatic Coreference Annotation Tool (CAT). We also provide statistics of coreference annotation on Hindi Dependency Treebank (HDTB).
We present coreference annotation on parallel Czech-English texts of the Prague Czech-English Dependency Treebank (PCEDT). The paper describes innovations made to PCEDT 2.0 concerning coreference, as well as coreference information already present there. We characterize the coreference annotation scheme, give the statistics and compare our annotation with the coreference annotation in Ontonotes and Prague Dependency Treebank for Czech. We also present the experiments made using this corpus to improve the alignment of coreferential expressions, which helps us to collect better statistics of correspondences between types of coreferential relations in Czech and English. The corpus released as PCEDT 2.0 Coref is publicly available.
We describe challenges and advantages unique to coreference resolution in the biomedical domain, and a sieve-based architecture that leverages domain knowledge for both entity and event coreference resolution. Domain-general coreference resolution algorithms perform poorly on biomedical documents, because the cues they rely on such as gender are largely absent in this domain, and because they do not encode domain-specific knowledge such as the number and type of participants required in chemical reactions. Moreover, it is difficult to directly encode this knowledge into most coreference resolution algorithms because they are not rule-based. Our rule-based architecture uses sequentially applied hand-designed “sieves”, with the output of each sieve informing and constraining subsequent sieves. This architecture provides a 3.2% increase in throughput to our Reach event extraction system with precision parallel to that of the stricter system that relies solely on syntactic patterns for extraction.
Characters form the focus of various studies of literary works, including social network analysis, archetype induction, and plot comparison. The recent rise in the computational modelling of literary works has produced a proportional rise in the demand for character-annotated literary corpora. However, automatically identifying characters is an open problem and there is low availability of literary texts with manually labelled characters. To address the latter problem, this work presents three contributions: (1) a comprehensive scheme for manually resolving mentions to characters in texts. (2) A novel collaborative annotation tool, CHARLES (CHAracter Resolution Label-Entry System) for character annotation and similiar cross-document tagging tasks. (3) The character annotations resulting from a pilot study on the novel Pride and Prejudice, demonstrating the scheme and tool facilitate the efficient production of high-quality annotations. We expect this work to motivate the further production of annotated literary corpora to help meet the demand of the community.
In most international industries, English is the main language of communication for technical documents. These documents are designed to be as unambiguous as possible for their users. For international industries based in non-English speaking countries, the professionals in charge of writing requirements are often non-native speakers of English, who rarely receive adequate training in the use of English for this task. As a result, requirements can contain a relatively large diversity of lexical and grammatical errors, which are not eliminated by the use of guidelines from controlled languages. This article investigates the distribution of errors in a corpus of requirements written in English by native speakers of French. Errors are defined on the basis of grammaticality and acceptability principles, and classified using comparable categories. Results show a high proportion of errors in the Noun Phrase, notably through modifier stacking, and errors consistent with simplification strategies. Comparisons with similar corpora in other genres reveal the specificity of the distribution of errors in requirements. This research also introduces possible applied uses, in the form of strategies for the automatic detection of errors, and in-person training provided by certification boards in requirements authoring.
We present a novel method to automatically improve the accurracy of part-of-speech taggers on learner language. The key idea underlying our approach is to exploit the structure of a typical language learner task and automatically induce POS information for out-of-vocabulary (OOV) words. To evaluate the effectiveness of our approach, we add manual POS and normalization information to an existing language learner corpus. Our evaluation shows an increase in accurracy from 72.4% to 81.5% on OOV words.
We present a new resource for Swedish, SweLL, a corpus of Swedish Learner essays linked to learners’ performance according to the Common European Framework of Reference (CEFR). SweLL consists of three subcorpora ― SpIn, SW1203 and Tisus, collected from three different educational establishments. The common metadata for all subcorpora includes age, gender, native languages, time of residence in Sweden, type of written task. Depending on the subcorpus, learner texts may contain additional information, such as text genres, topics, grades. Five of the six CEFR levels are represented in the corpus: A1, A2, B1, B2 and C1 comprising in total 339 essays. C2 level is not included since courses at C2 level are not offered. The work flow consists of collection of essays and permits, essay digitization and registration, meta-data annotation, automatic linguistic annotation. Inter-rater agreement is presented on the basis of SW1203 subcorpus. The work on SweLL is still ongoing with more that 100 essays waiting in the pipeline. This article both describes the resource and the “how-to” behind the compilation of SweLL.
The paper introduces SVALex, a lexical resource primarily aimed at learners and teachers of Swedish as a foreign and second language that describes the distribution of 15,681 words and expressions across the Common European Framework of Reference (CEFR). The resource is based on a corpus of coursebook texts, and thus describes receptive vocabulary learners are exposed to during reading activities, as opposed to productive vocabulary they use when speaking or writing. The paper describes the methodology applied to create the list and to estimate the frequency distribution. It also discusses some characteristics of the resulting resource and compares it to other lexical resources for Swedish. An interesting feature of this resource is the possibility to separate the wheat from the chaff, identifying the core vocabulary at each level, i.e. vocabulary shared by several coursebook writers at each level, from peripheral vocabulary which is used by the minority of the coursebook writers.
Automated grammatical error detection, which helps users improve their writing, is an important application in NLP. Recently more and more people are learning Chinese, and an automated error detection system can be helpful for the learners. This paper proposes n-gram features, dependency count features, dependency bigram features, and single-character features to determine if a Chinese sentence contains word usage errors, in which a word is written as a wrong form or the word selection is inappropriate. With marking potential errors on the level of sentence segments, typically delimited by punctuation marks, the learner can try to correct the problems without the assistant of a language teacher. Experiments on the HSK corpus show that the classifier combining all sets of features achieves an accuracy of 0.8423. By utilizing certain combination of the sets of features, we can construct a system that favors precision or recall. The best precision we achieve is 0.9536, indicating that our system is reliable and seldom produces misleading results.
We present a light-weight machine learning tool for NLP research. The package supports operations on both discrete and dense vectors, facilitating implementation of linear models as well as neural models. It provides several basic layers which mainly aims for single-layer linear and non-linear transformations. By using these layers, we can conveniently implement linear models and simple neural models. Besides, this package also integrates several complex layers by composing those basic layers, such as RNN, Attention Pooling, LSTM and gated RNN. Those complex layers can be used to implement deep neural models directly.
This study examines two possibilities of using the FLELex graded lexicon for the automated assessment of text complexity in French as a foreign language learning. From the lexical frequency distributions described in FLELex, we derive a single level of difficulty for each word in a parallel corpus of original and simplified texts. We then use this data to automatically address the lexical complexity of texts in two ways. On the one hand, we evaluate the degree of lexical simplification in manually simplified texts with respect to their original version. Our results show a significant simplification effect, both in the case of French narratives simplified for non-native readers and in the case of simplified Wikipedia texts. On the other hand, we define a predictive model which identifies the number of words in a text that are expected to be known at a particular learning level. We assess the accuracy with which these predictions are able to capture actual word knowledge as reported by Dutch-speaking learners of French. Our study shows that although the predictions seem relatively accurate in general (87.4% to 92.3%), they do not yet seem to cover the learners’ lack of knowledge very well.
We argue that the field of spoken CALL needs a shared task in order to facilitate comparisons between different groups and methodologies, and describe a concrete example of such a task, based on data collected from a speech-enabled online tool which has been used to help young Swiss German teens practise skills in English conversation. Items are prompt-response pairs, where the prompt is a piece of German text and the response is a recorded English audio file. The task is to label pairs as “accept” or “reject”, accepting responses which are grammatically and linguistically correct to match a set of hidden gold standard answers as closely as possible. Initial resources are provided so that a scratch system can be constructed with a minimal investment of effort, and in particular without necessarily using a speech recogniser. Training data for the task will be released in June 2016, and test data in January 2017.
Dialogue robots are attractive to people, and in language learning systems, they motivate learners and let them practice conversational skills in more realistic environment. However, automatic speech recognition (ASR) of the second language (L2) learners is still a challenge, because their speech contains not just pronouncing, lexical, grammatical errors, but is sometimes totally disordered. Hence, we propose a novel robot assisted language learning (RALL) system using two robots, one as a teacher and the other as an advanced learner. The system is designed to simulate multiparty conversation, expecting implicit learning and enhancement of predictability of learners’ utterance through an alignment similar to “interactive alignment”, which is observed in human-human conversation. We collected a database with the prototypes, and measured how much the alignment phenomenon observed in the database with initial analysis.
We present OSMAN (Open Source Metric for Measuring Arabic Narratives) - a novel open source Arabic readability metric and tool. It allows researchers to calculate readability for Arabic text with and without diacritics. OSMAN is a modified version of the conventional readability formulas such as Flesch and Fog. In our work we introduce a novel approach towards counting short, long and stress syllables in Arabic which is essential for judging readability of Arabic narratives. We also introduce an additional factor called “Faseeh” which considers aspects of script usually dropped in informal Arabic writing. To evaluate our methods we used Spearman’s correlation metric to compare text readability for 73,000 parallel sentences from English and Arabic UN documents. The Arabic sentences were written with the absence of diacritics and in order to count the number of syllables we added the diacritics in using an open source tool called Mishkal. The results show that OSMAN readability formula correlates well with the English ones making it a useful tool for researchers and educators working with Arabic text.
Enabling users of intelligent systems to enhance the system performance by providing feedback on their errors is an important need. However, the ability of systems to learn from user feedback is difficult to evaluate in an objective and comparative way. Indeed, the involvement of real users in the adaptation process is an impediment to objective evaluation. This issue can be solved by using an oracle approach, where users are simulated by oracles having access to the reference test data. Another difficulty is to find a meaningful metric despite the fact that system improvements depend on the feedback provided and on the system itself. A solution is to measure the minimal amount of information needed to correct all system errors. It can be shown that for any well defined non interactive task, the interactively supervised version of the task can be evaluated by combining such an oracle-based approach and a minimum supervision rate metric. This new evaluation protocol for adaptive systems is not only expected to drive progress for such systems, but also to pave the way for a specialisation of actors along the value chain of their technological development.
This paper addresses the problem of quantifying the differences between entity extraction systems, where in general only a small proportion a document should be selected. Comparing overall accuracy is not very useful in these cases, as small differences in accuracy may correspond to huge differences in selections over the target minority class. Conventionally, one may use per-token complementarity to describe these differences, but it is not very useful when the set is heavily skewed. In such situations, which are common in information retrieval and entity recognition, metrics like precision and recall are typically used to describe performance. However, precision and recall fail to describe the differences between sets of objects selected by different decision strategies, instead just describing the proportional amount of correct and incorrect objects selected. This paper presents a method for measuring complementarity for precision, recall and F-score, quantifying the difference between entity extraction approaches.
Opinion Mining is a topic which attracted a lot of interest in the last years. By observing the literature, it is often hard to replicate system evaluation due to the unavailability of the data used for the evaluation or to the lack of details about the protocol used in the campaign. In this paper, we propose an evaluation protocol, called DRANZIERA, composed of a multi-domain dataset and guidelines allowing both to evaluate opinion mining systems in different contexts (Closed, Semi-Open, and Open) and to compare them to each other and to a number of baselines.
Web corpora are often constructed automatically, and their contents are therefore often not well understood. One technique for assessing the composition of such a web corpus is to empirically measure its similarity to a reference corpus whose composition is known. In this paper we evaluate a number of measures of corpus similarity, including a method based on topic modelling which has not been previously evaluated for this task. To evaluate these methods we use known-similarity corpora that have been previously used for this purpose, as well as a number of newly-constructed known-similarity corpora targeting differences in genre, topic, time, and region. Our findings indicate that, overall, the topic modelling approach did not improve on a chi-square method that had previously been found to work well for measuring corpus similarity.
This contribution presents the background, design and results of a study of users of three oral corpus platforms in Germany. Roughly 5.000 registered users of the Database for Spoken German (DGD), the GeWiss corpus and the corpora of the Hamburg Centre for Language Corpora (HZSK) were asked to participate in a user survey. This quantitative approach was complemented by qualitative interviews with selected users. We briefly introduce the corpus resources involved in the study in section 2. Section 3 describes the methods employed in the user studies. Section 4 summarizes results of the studies focusing on selected key topics. Section 5 attempts a generalization of these results to larger contexts.
In this paper we describe a corpus of automatic translations annotated with both error type and quality. The 300 sentences that we have selected were generated by Google Translate, Systran and two in-house Machine Translation systems that use Moses technology. The errors present on the translations were annotated with an error taxonomy that divides errors in five main linguistic categories (Orthography, Lexis, Grammar, Semantics and Discourse), reflecting the language level where the error is located. After the error annotation process, we accessed the translation quality of each sentence using a four point comprehension scale from 1 to 5. Both tasks of error and quality annotation were performed by two different annotators, achieving good levels of inter-annotator agreement. The creation of this corpus allowed us to use it as training data for a translation quality classifier. We concluded on error severity by observing the outputs of two machine learning classifiers: a decision tree and a regression model.
This paper presents an approach for automatic evaluation of the readability of text simplification output for readers with cognitive disabilities. First, we present our work towards the development of the EasyRead corpus, which contains easy-to-read documents created especially for people with cognitive disabilities. We then compare the EasyRead corpus to the simplified output contained in the LocalNews corpus (Feng, 2009), the accessibility of which has been evaluated through reading comprehension experiments including 20 adults with mild intellectual disability. This comparison is made on the basis of 13 disability-specific linguistic features. The comparison reveals that there are no major differences between the two corpora, which shows that the EasyRead corpus is to a similar reading level as the user-evaluated texts. We also discuss the role of Simple Wikipedia (Zhu et al., 2010) as a widely-used accessibility benchmark, in light of our finding that it is significantly more complex than both the EasyRead and the LocalNews corpora.
Word embeddings have been successfully used in several natural language processing tasks (NLP) and speech processing. Different approaches have been introduced to calculate word embeddings through neural networks. In the literature, many studies focused on word embedding evaluation, but for our knowledge, there are still some gaps. This paper presents a study focusing on a rigorous comparison of the performances of different kinds of word embeddings. These performances are evaluated on different NLP and linguistic tasks, while all the word embeddings are estimated on the same training data using the same vocabulary, the same number of dimensions, and other similar characteristics. The evaluation results reported in this paper match those in the literature, since they point out that the improvements achieved by a word embedding in one task are not consistently observed across all tasks. For that reason, this paper investigates and evaluates approaches to combine word embeddings in order to take advantage of their complementarity, and to look for the effective word embeddings that can achieve good performances on all tasks. As a conclusion, this paper provides new perceptions of intrinsic qualities of the famous word embedding families, which can be different from the ones provided by works previously published in the scientific literature.
In this paper, we claim that the CAMOMILE collaborative annotation platform (developed in the framework of the eponymous CHIST-ERA project) eases the organization of multimedia technology benchmarks, automating most of the campaign technical workflow and enabling collaborative (hence faster and cheaper) annotation of the evaluation data. This is demonstrated through the successful organization of a new multimedia task at MediaEval 2015, Multimodal Person Discovery in Broadcast TV.
This paper discusses a methodology to measure the usability of machine translated content by end users, comparing lightly post-edited content with raw output and with the usability of source language content. The content selected consists of Online Help articles from a software company for a spreadsheet application, translated from English into German. Three groups of five users each used either the source text - the English version (EN) -, the raw MT version (DE_MT), or the light PE version (DE_PE), and were asked to carry out six tasks. Usability was measured using an eye tracker and cognitive, temporal and pragmatic measures of usability. Satisfaction was measured via a post-task questionnaire presented after the participants had completed the tasks.
This project assesses the resources necessary to make oral history searchable by means of automatic speech recognition (ASR). There are many inherent challenges in applying ASR to conversational speech: smaller training set sizes and varying demographics, among others. We assess the impact of dataset size, word error rate and term-weighted value on human search capability through an information retrieval task on Mechanical Turk. We use English oral history data collected by StoryCorps, a national organization that provides all people with the opportunity to record, share and preserve their stories, and control for a variety of demographics including age, gender, birthplace, and dialect on four different training set sizes. We show comparable search performance using a standard speech recognition system as with hand-transcribed data, which is promising for increased accessibility of conversational speech and oral history archives.
Odin is an information extraction framework that applies cascades of finite state automata over both surface text and syntactic dependency graphs. Support for syntactic patterns allow us to concisely define relations that are otherwise difficult to express in languages such as Common Pattern Specification Language (CPSL), which are currently limited to shallow linguistic features. The interaction of lexical and syntactic automata provides robustness and flexibility when writing extraction rules. This paper describes Odin’s declarative language for writing these cascaded automata.
Breaking news on economic events such as stock splits or mergers and acquisitions has been shown to have a substantial impact on the financial markets. As it is important to be able to automatically identify events in news items accurately and in a timely manner, we present in this paper proof-of-concept experiments for a supervised machine learning approach to economic event detection in newswire text. For this purpose, we created a corpus of Dutch financial news articles in which 10 types of company-specific economic events were annotated. We trained classifiers using various lexical, syntactic and semantic features. We obtain good results based on a basic set of shallow features, thus showing that this method is a viable approach for economic event detection in news text.
Predictive modeling, often called “predictive analytics” in a commercial context, encompasses a variety of statistical techniques that analyze historical and present facts to make predictions about unknown events. Often the unknown events are in the future, but prediction can be applied to any type of unknown whether it be in the past or future. In our case, we present some experiments applying predictive modeling to the usage of technical terms within the NLP domain.
This paper presents the Gavagai Living Lexicon, which is an online distributional semantic model currently available in 20 different languages. We describe the underlying distributional semantic model, and how we have solved some of the challenges in applying such a model to large amounts of streaming data. We also describe the architecture of our implementation, and discuss how we deal with continuous quality assurance of the lexicon.
Social media outlets are providing new opportunities for harvesting valuable resources. We present a novel approach for mining data from Twitter for the purpose of building transliteration resources and systems. Such resources are crucial in translation and retrieval tasks. We demonstrate the benefits of the approach on Arabic to English transliteration. The contribution of this approach includes the size of data that can be collected and exploited within the span of a limited time; the approach is very generic and can be adopted to other languages and the ability of the approach to cope with new transliteration phenomena and trends. A statistical transliteration system built using this data improved a comparable system built from Wikipedia wikilinks data.
Many emerging documents usually contain temporal information. Because the temporal information is useful for various applications, it became important to develop a system of extracting the temporal information from the documents. Before developing the system, it first necessary to define or design the structure of temporal information. In other words, it is necessary to design a language which defines how to annotate the temporal information. There have been some studies about the annotation languages, but most of them was applicable to only a specific target language (e.g., English). Thus, it is necessary to design an individual annotation language for each language. In this paper, we propose a revised version of Koreain Time Mark-up Language (K-TimeML), and also introduce a dataset, named Korean TimeBank, that is constructed basd on the K-TimeML. We believe that the new K-TimeML and Korean TimeBank will be used in many further researches about extraction of temporal information.
Hypernymy relations (those where an hyponym term shares a “isa” relationship with his hypernym) play a key role for many Natural Language Processing (NLP) tasks, e.g. ontology learning, automatically building or extending knowledge bases, or word sense disambiguation and induction. In fact, such relations may provide the basis for the construction of more complex structures such as taxonomies, or be used as effective background knowledge for many word understanding applications. We present a publicly available database containing more than 400 million hypernymy relations we extracted from the CommonCrawl web corpus. We describe the infrastructure we developed to iterate over the web corpus for extracting the hypernymy relations and store them effectively into a large database. This collection of relations represents a rich source of knowledge and may be useful for many researchers. We offer the tuple dataset for public download and an Application Programming Interface (API) to help other researchers programmatically query the database.
The objective of this paper was to evaluate the performance of two statistical machine translation (SMT) systems within a cross-language information retrieval (CLIR) architecture and examine if there is a correlation between translation quality and CLIR performance. The SMT systems were KantanMT, a cloud-based machine translation (MT) platform, and Moses, an open-source MT application. First we trained both systems using the same language resources: the EMEA corpus for the translation model and language model and the QTLP corpus for tuning. Then we translated the 63 queries of the OHSUMED test collection from Greek into English using both MT systems. Next, we ran the queries on the document collection using Apache Solr to get a list of the top ten matches. The results were compared to the OHSUMED gold standard. KantanMT achieved higher average precision and F-measure than Moses, while both systems produced the same recall score. We also calculated the BLEU score for each system using the ECDC corpus. Moses achieved a higher BLEU score than KantanMT. Finally, we also tested the IR performance of the original English queries. This work overall showed that CLIR performance can be better even when BLEU score is worse.
This work analyses a corpus made of the titles of research projects belonging to the last four European Commission Framework Programmes (FP4, FP5, FP6, FP7) during a time span of nearly two decades (1994-2012). The starting point is the idea of creating a corpus of titles which would constitute a terminological niche, a sort of “cluster map” offering an overall vision on the terms used and the links between them. Moreover, by performing a terminological comparison over a period of time it is possible to trace the presence of obsolete words in outdated research areas as well as of neologisms in the most recent fields. Within this scenario, the minimal purpose is to build a corpus of titles of European projects belonging to the several Framework Programmes in order to obtain a terminological mapping of relevant words in the various research areas: particularly significant would be those terms spread across different domains or those extremely tied to a specific domain. A term could actually be found in many fields and being able to acknowledge and retrieve this cross-presence means being able to linking those different domains by means of a process of terminological mapping.
The paper investigates the extent of the support semi-automatic analysis can provide for the specific task of assigning Hohfeldian relations of Duty, using the General Architecture for Text Engineering tool for the automated extraction of Duty instances and the bearers of associated roles. The outcome of the analysis supports scholars in identifying Hohfeldian structures in legal text when performing close reading of the texts. A cyclic workflow involving automated annotation and expert feedback will incrementally increase the quality and coverage of the automatic extraction process, and increasingly reduce the amount of manual work required of the scholar.
The emergence of the web has necessitated the need to detect and correct noisy consumer-generated texts. Most of the previous studies on English spelling-error extraction collected English spelling errors from web services such as Twitter by using the edit distance or from input logs utilizing crowdsourcing. However, in the former approach, it is not clear which word corresponds to the spelling error, and the latter approach requires an annotation cost for the crowdsourcing. One notable exception is Rodrigues and Rytting (2012), who proposed to extract English spelling errors by using a word-typing game. Their approach saves the cost of crowdsourcing, and guarantees an exact alignment between the word and the spelling error. However, they did not assert whether the extracted spelling error corpora reflect the usual writing process such as writing a document. Therefore, we propose a new correctable word-typing game that is more similar to the actual writing process. Experimental results showed that we can regard typing-game logs as a source of spelling errors.
The paper describes automatic definition finding implemented within the leading corpus query and management tool, Sketch Engine. The implementation exploits complex pattern-matching queries in the corpus query language (CQL) and the indexing mechanism of word sketches for finding and storing definition candidates throughout the corpus. The approach is evaluated for Czech and English corpora, showing that the results are usable in practice: precision of the tool ranges between 30 and 75 percent (depending on the major corpus text types) and we were able to extract nearly 2 million definition candidates from an English corpus with 1.4 billion words. The feature is embedded into the interface as a concordance filter, so that users can search for definitions of any query to the corpus, including very specific multi-word queries. The results also indicate that ordinary texts (unlike explanatory texts) contain rather low number of definitions, which is perhaps the most important problem with automatic definition finding in general.
The aim of this paper is to study the effect that the use of Basic English versus common English has on information extraction from online resources. The amount of online information available to the public grows exponentially, and is potentially an excellent resource for information extraction. The problem is that this information often comes in an unstructured format, such as plain text. In order to retrieve knowledge from this type of text, it must first be analysed to find the relevant details, and the nature of the language used can greatly impact the quality of the extracted information. In this paper, we compare triplets that represent definitions or properties of concepts obtained from three online collaborative resources (English Wikipedia, Simple English Wikipedia and Simple English Wiktionary) and study the differences in the results when Basic English is used instead of common English. The results show that resources written in Basic English produce less quantity of triplets, but with higher quality.
In this paper we present PIERINO (PIattaforma per l’Estrazione e il Recupero di INformazione Online), a system that was implemented in collaboration with the Italian Ministry of Education, University and Research to analyse the citizens’ comments given in #labuonascuola survey. The platform includes various levels of automatic analysis such as key-concept extraction and word co-occurrences. Each analysis is displayed through an intuitive view using different types of visualizations, for example radar charts and sunburst. PIERINO was effectively used to support shaping the last Italian school reform, proving the potential of NLP in the context of policy making.
This paper presents the evaluation of the translation quality and Cross-Lingual Information Retrieval (CLIR) performance when using session information as the context of queries. The hypothesis is that previous queries provide context that helps to solve ambiguous translations in the current query. We tested several strategies on the TREC 2010 Session track dataset, which includes query reformulations grouped by generalization, specification, and drifting types. We study the Basque to English direction, evaluating both the translation quality and CLIR performance, with positive results in both cases. The results show that the quality of translation improved, reducing error rate by 12% (HTER) when using session information, which improved CLIR results 5% (nDCG). We also provide an analysis of the improvements across the three kinds of sessions: generalization, specification, and drifting. Translation quality improved in all three types (generalization, specification, and drifting), and CLIR improved for generalization and specification sessions, preserving the performance in drifting sessions.
This paper presents a new Vietnamese text corpus which contains around 4.05 billion words. It is a collection of Wikipedia texts, newspaper articles and random web texts. The paper describes the process of collecting, cleaning and creating the corpus. Processing Vietnamese texts faced several challenges, for example, different from many Latin languages, Vietnamese language does not use blanks for separating words, hence using common tokenizers such as replacing blanks with word boundary does not work. A short review about different approaches of Vietnamese tokenization is presented together with how the corpus has been processed and created. After that, some statistical analysis on this data is reported including the number of syllable, average word length, sentence length and topic analysis. The corpus is integrated into a framework which allows searching and browsing. Using this web interface, users can find out how many times a particular word appears in the corpus, sample sentences where this word occurs, its left and right neighbors.
Text analysis methods for the automatic identification of emerging technologies by analyzing the scientific publications, are gaining attention because of their socio-economic impact. The approaches so far have been mainly focused on retrospective analysis by mapping scientific topic evolution over time. We propose regression based approaches to predict future keyword distribution. The prediction is based on historical data of the keywords, which in our case, are LREC conference proceedings. Considering the insufficient number of data points available from LREC proceedings, we do not employ standard time series forecasting methods. We form a dataset by extracting the keywords from previous year proceedings and quantify their yearly relevance using tf-idf scores. This dataset additionally contains ranked lists of related keywords and experts for each keyword.
Understanding the experimental results of a scientific paper is crucial to understanding its contribution and to comparing it with related work. We introduce a structured, queryable representation for experimental results and a baseline system that automatically populates this representation. The representation can answer compositional questions such as: “Which are the best published results reported on the NIST 09 Chinese to English dataset?” and “What are the most important methods for speeding up phrase-based decoding?” Answering such questions usually involves lengthy literature surveys. Current machine reading for academic papers does not usually consider the actual experiments, but mostly focuses on understanding abstracts. We describe annotation work to create an initial hscientific paper; experimental results representationi corpus. The corpus is composed of 67 papers which were manually annotated with a structured representation of experimental results by domain experts. Additionally, we present a baseline algorithm that characterizes the difficulty of the inference task.
Domain-specific annotations for NLP are often centered on real-world applications of text, and incorrect annotations may be particularly unacceptable. In medical text, the process of manual chart review (of a patient’s medical record) is error-prone due to its complexity. We propose a staggered NLP-assisted approach to the refinement of clinical annotations, an interactive process that allows initial human judgments to be verified or falsified by means of comparison with an improving NLP system. We show on our internal Asthma Timelines dataset that this approach improves the quality of the human-produced clinical annotations.
This paper presents QUANDHO (QUestion ANswering Data for italian HistOry), an Italian question answering dataset created to cover a specific domain, i.e. the history of Italy in the first half of the XX century. The dataset includes questions manually classified and annotated with Lexical Answer Types, and a set of question-answer pairs. This resource, freely available for research purposes, has been used to retrain a domain independent question answering system so to improve its performances in the domain of interest. Ongoing experiments on the development of a question classifier and an automatic tagger of Lexical Answer Types are also presented.
We present a successfully implemented document repository REST service for flexible SCRUD (search, crate, read, update, delete) storage of social media conversations, using a GATE/TIPSTER-like document object model and providing a query language for document features. This software is currently being used in the SENSEI research project and will be published as open-source software before the project ends. It is, to the best of our knowledge, the first freely available, general purpose data repository to support large-scale multimodal (i.e., speech or text) conversation analytics.
The Dictionaries division at Oxford University Press (OUP) is aiming to model, integrate, and publish lexical content for 100 languages focussing on digitally under-represented languages. While there are multiple ontologies designed for linguistic resources, none had adequate features for meeting our requirements, chief of which was the capability to losslessly capture diverse features of many different languages in a dictionary format, while supplying a framework for inferring relations like translation, derivation, etc., between the data. Building on valuable features of existing models, and working with OUP monolingual and bilingual dictionary datasets, we have designed and implemented a new linguistic ontology. The ontology has been reviewed by a number of computational linguists, and we are working to move more dictionary data into it. We have also developed APIs to surface the linked data to dictionary websites.
Language resources (LR) are indispensable for the development of tools for machine translation (MT) or various kinds of computer-assisted translation (CAT). In particular language corpora, both parallel and monolingual are considered most important for instance for MT, not only SMT but also hybrid MT. The Language Technology Observatory will provide easy access to information about LRs deemed to be useful for MT and other translation tools through its LR Catalogue. In order to determine what aspects of an LR are useful for MT practitioners, a user study was made, providing a guide to the most relevant metadata and the most relevant quality criteria. We have seen that many resources exist which are useful for MT and similar work, but the majority are for (academic) research or educational use only, and as such not available for commercial use. Our work has revealed a list of gaps: coverage gap, awareness gap, quality gap, quantity gap. The paper ends with recommendations for a forward-looking strategy.
The NSF-SI2-funded LAPPS Grid project is a collaborative effort among Brandeis University, Vassar College, Carnegie-Mellon University (CMU), and the Linguistic Data Consortium (LDC), which has developed an open, web-based infrastructure through which resources can be easily accessed and within which tailored language services can be efficiently composed, evaluated, disseminated and consumed by researchers, developers, and students across a wide variety of disciplines. The LAPPS Grid project recently adopted Galaxy (Giardine et al., 2005), a robust, well-developed, and well-supported front end for workflow configuration, management, and persistence. Galaxy allows data inputs and processing steps to be selected from graphical menus, and results are displayed in intuitive plots and summaries that encourage interactive workflows and the exploration of hypotheses. The Galaxy workflow engine provides significant advantages for deploying pipelines of LAPPS Grid web services, including not only means to create and deploy locally-run and even customized versions of the LAPPS Grid as well as running the LAPPS Grid in the cloud, but also access to a huge array of statistical and visualization tools that have been developed for use in genomics research.
After celebrating its 20th anniversary in 2015, ELRA is carrying on its strong involvement in the HLT field. To share ELRA’s expertise of those 21 past years, this article begins with a presentation of ELRA’s strategic Data and LR Management Plan for a wide use by the language communities. Then, we further report on ELRA’s activities and services provided since LREC 2014. When looking at the cataloguing and licensing activities, we can see that ELRA has been active at making the Meta-Share repository move toward new developments steps, supporting Europe to obtain accurate LRs within the Connecting Europe Facility programme, promoting the use of LR citation, creating the ELRA License Wizard web portal.The article further elaborates on the recent LR production activities of various written, speech and video resources, commissioned by public and private customers. In parallel, ELDA has also worked on several EU-funded projects centred on strategic issues related to the European Digital Single Market. The last part gives an overview of the latest dissemination activities, with a special focus on the celebration of its 20th anniversary organised in Dubrovnik (Croatia) and the following up of LREC, as well as the launching of the new ELRA portal.
This paper presents an investigation of mirroring facial expressions and the emotions which they convey in dyadic naturally occurring first encounters. Mirroring facial expressions are a common phenomenon in face-to-face interactions, and they are due to the mirror neuron system which has been found in both animals and humans. Researchers have proposed that the mirror neuron system is an important component behind many cognitive processes such as action learning and understanding the emotions of others. Preceding studies of the first encounters have shown that overlapping speech and overlapping facial expressions are very frequent. In this study, we want to determine whether the overlapping facial expressions are mirrored or are otherwise correlated in the encounters, and to what extent mirroring facial expressions convey the same emotion. The results of our study show that the majority of smiles and laughs, and one fifth of the occurrences of raised eyebrows are mirrored in the data. Moreover some facial traits in co-occurring expressions co-occur more often than it would be expected by chance. Finally, amusement, and to a lesser extent friendliness, are often emotions shared by both participants, while other emotions indicating individual affective states such as uncertainty and hesitancy are never showed by both participants, but co-occur with complementary emotions such as friendliness and support. Whether these tendencies are specific to this type of conversations or are more common should be investigated further.
The New Yorker publishes a weekly captionless cartoon. More than 5,000 readers submit captions for it. The editors select three of them and ask the readers to pick the funniest one. We describe an experiment that compares a dozen automatic methods for selecting the funniest caption. We show that negative sentiment, human-centeredness, and lexical centrality most strongly match the funniest captions, followed by positive sentiment. These results are useful for understanding humor and also in the design of more engaging conversational agents in text and multimodal (vision+text) systems. As part of this work, a large set of cartoons and captions is being made available to the community.
The paper presents a corpus of text data and its corresponding gaze fixations obtained from autistic and non-autistic readers. The data was elicited through reading comprehension testing combined with eye-tracking recording. The corpus consists of 1034 content words tagged with their POS, syntactic role and three gaze-based measures corresponding to the autistic and control participants. The reading skills of the participants were measured through multiple-choice questions and, based on the answers given, they were divided into groups of skillful and less-skillful readers. This division of the groups informs researchers on whether particular fixations were elicited from skillful or less-skillful readers and allows a fair between-group comparison for two levels of reading ability. In addition to describing the process of data collection and corpus development, we present a study on the effect that word length has on reading in autism. The corpus is intended as a resource for investigating the particular linguistic constructions which pose reading difficulties for people with autism and hopefully, as a way to inform future text simplification research intended for this population.
The ability to efficiently speak in public is an essential asset for many professions and is used in everyday life. As such, tools enabling the improvement of public speaking performance and the assessment and mitigation of anxiety related to public speaking would be very useful. Multimodal interaction technologies, such as computer vision and embodied conversational agents, have recently been investigated for the training and assessment of interpersonal skills. Once central requirement for these technologies is multimodal corpora for training machine learning models. This paper addresses the need of these technologies by presenting and sharing a multimodal corpus of public speaking presentations. These presentations were collected in an experimental study investigating the potential of interactive virtual audiences for public speaking training. This corpus includes audio-visual data and automatically extracted features, measures of public speaking anxiety and personality, annotations of participants’ behaviors and expert ratings of behavioral aspects and overall performance of the presenters. We hope this corpus will help other research teams in developing tools for supporting public speaking training.
We propose a comparison between various supervised machine learning methods to predict and detect humor in dialogues. We retrieve our humorous dialogues from a very popular TV sitcom: “The Big Bang Theory”. We build a corpus where punchlines are annotated using the canned laughter embedded in the audio track. Our comparative study involves a linear-chain Conditional Random Field over a Recurrent Neural Network and a Convolutional Neural Network. Using a combination of word-level and audio frame-level features, the CNN outperforms the other methods, obtaining the best F-score of 68.5% over 66.5% by CRF and 52.9% by RNN. Our work is a starting point to developing more effective machine learning and neural network models on the humor prediction task, as well as developing machines capable in understanding humor in general.
This paper aims to implement what is referred to as the collocation of the Arabic keywords approach for extracting formulaic sequences (FSs) in the form of high frequency but semantically regular formulas that are not restricted to any syntactic construction or semantic domain. The study applies several distributional semantic models in order to automatically extract relevant FSs related to Arabic keywords. The data sets used in this experiment are rendered from a new developed corpus-based Arabic wordlist consisting of 5,189 lexical items which represent a variety of modern standard Arabic (MSA) genres and regions, the new wordlist being based on an overlapping frequency based on a comprehensive comparison of four large Arabic corpora with a total size of over 8 billion running words. Empirical n-best precision evaluation methods are used to determine the best association measures (AMs) for extracting high frequency and meaningful FSs. The gold standard reference FSs list was developed in previous studies and manually evaluated against well-established quantitative and qualitative criteria. The results demonstrate that the MI.log_f AM achieved the highest results in extracting significant FSs from the large MSA corpus, while the T-score association measure achieved the worst results.
In this paper we present a rule-based method for multi-word term extraction that relies on extensive lexical resources in the form of electronic dictionaries and finite-state transducers for modelling various syntactic structures of multi-word terms. The same technology is used for lemmatization of extracted multi-word terms, which is unavoidable for highly inflected languages in order to pass extracted data to evaluators and subsequently to terminological e-dictionaries and databases. The approach is illustrated on a corpus of Serbian texts from the mining domain containing more than 600,000 simple word forms. Extracted and lemmatized multi-word terms are filtered in order to reject falsely offered lemmas and then ranked by introducing measures that combine linguistic and statistical information (C-Value, T-Score, LLR, and Keyness). Mean average precision for retrieval of MWU forms ranges from 0.789 to 0.804, while mean average precision of lemma production ranges from 0.956 to 0.960. The evaluation showed that 94% of distinct multi-word forms were evaluated as proper multi-word units, and among them 97% were associated with correct lemmas.
In this paper, we focus on Czech complex predicates formed by a light verb and a predicative noun expressed as the direct object. Although Czech ― as an inflectional language encoding syntactic relations via morphological cases ― provides an excellent opportunity to study the distribution of valency complements in the syntactic structure with complex predicates, this distribution has not been described so far. On the basis of a manual analysis of the richly annotated data from the Prague Dependency Treebank, we thus formulate principles governing this distribution. In an automatic experiment, we verify these principles on well-formed syntactic structures from the Prague Dependency Treebank and the Prague Czech-English Dependency Treebank with very satisfactory results: the distribution of 97% of valency complements in the surface structure is governed by the proposed principles. These results corroborate that the surface structure formation of complex predicates is a regular process.
A verb-noun Multi-Word Expression (MWE) is a combination of a verb and a noun with or without other words, in which the combination has a meaning different from the meaning of the words considered separately. In this paper, we present a new lexical resource of Hebrew Verb-Noun MWEs (VN-MWEs). The VN-MWEs of this resource were manually collected and annotated from five different web resources. In addition, we analyze the lexical properties of Hebrew VN-MWEs by classifying them to three types: morphological, syntactic, and semantic. These two contributions are essential for designing algorithms for automatic VN-MWEs extraction. The analysis suggests some interesting features of VN-MWEs for exploration. The lexical resource enables to sample a set of positive examples for Hebrew VN-MWEs. This set of examples can either be used for training supervised algorithms or as seeds in unsupervised bootstrapping algorithms. Thus, this resource is a first step towards automatic identification of Hebrew VN-MWEs, which is important for natural language understanding, generation and translation systems.
This paper reports on an approach and experiments to automatically build a cross-lingual multi-word entity resource. Starting from a collection of millions of acronym/expansion pairs for 22 languages where expansion variants were grouped into monolingual clusters, we experiment with several aggregation strategies to link these clusters across languages. Aggregation strategies make use of string similarity distances and translation probabilities and they are based on vector space and graph representations. The accuracy of the approach is evaluated against Wikipedia’s redirection and cross-lingual linking tables. The resulting multi-word entity resource contains 64,000 multi-word entities with unique identifiers and their 600,000 multilingual lexical variants. We intend to make this new resource publicly available.
This paper presents SemLinker, an open source system that discovers named entities, connects them to a reference knowledge base, and clusters them semantically. SemLinker relies on several modules that perform surface form generation, mutual disambiguation, entity clustering, and make use of two annotation engines. SemLinker was evaluated in the English Entity Discovery and Linking track of the Text Analysis Conference on Knowledge Base Population, organized by the US National Institute of Standards and Technology. Along with the SemLinker source code, we release our annotation files containing the discovered named entities, their types, and position across processed documents.
More and more knowledge bases are publicly available as linked data. Since these knowledge bases contain structured descriptions of real-world entities, they can be exploited by entity linking systems that anchor entity mentions from text to the most relevant resources describing those entities. In this paper, we investigate adaptation of the entity linking task using contextual knowledge. The key intuition is that entity linking can be customized depending on the textual content, as well as on the application that would make use of the extracted information. We present an adaptive approach that relies on contextual knowledge from text to enhance the performance of ADEL, a hybrid linguistic and graph-based entity linking system. We evaluate our approach on a domain-specific corpus consisting of annotated WikiNews articles.
Recently, due to the increasing popularity of social media, the necessity for extracting information from informal text types, such as microblog texts, has gained significant attention. In this study, we focused on the Named Entity Recognition (NER) problem on informal text types for Turkish. We utilized a semi-supervised learning approach based on neural networks. We applied a fast unsupervised method for learning continuous representations of words in vector space. We made use of these obtained word embeddings, together with language independent features that are engineered to work better on informal text types, for generating a Turkish NER system on microblog texts. We evaluated our Turkish NER system on Twitter messages and achieved better F-score performances than the published results of previously proposed NER systems on Turkish tweets. Since we did not employ any language dependent features, we believe that our method can be easily adapted to microblog texts in other morphologically rich languages.
The task of Named Entity Linking is to link entity mentions in the document to their correct entries in a knowledge base and to cluster NIL mentions. Ambiguous, misspelled, and incomplete entity mention names are the main challenges in the linking process. We propose a novel approach that combines two state-of-the-art models ― for entity disambiguation and for paraphrase detection ― to overcome these challenges. We consider name variations as paraphrases of the same entity mention and adopt a paraphrase model for this task. Our approach utilizes a graph-based disambiguation model based on Personalized Page Rank, and then refines and clusters its output using the paraphrase similarity between entity mention strings. It achieves a competitive performance of 80.5% in B3+F clustering score on diagnostic TAC EDL 2014 data.
In this paper we explain how we created a labelled corpus in English for a Named Entity Recognition (NER) task from multi-source and multi-domain data, for an industrial partner. We explain the specificities of this corpus with examples and describe some baseline experiments. We present some results of domain adaptation on this corpus using a labelled Twitter corpus (Ritter et al., 2011). We tested a semi-supervised method from (Garcia-Fernandez et al., 2014) combined with a supervised domain adaptation approach proposed in (Raymond and Fayolle, 2010) for machine learning experiments with CRFs (Conditional Random Fields). We use the same technique to improve the NER results on the Twitter corpus (Ritter et al., 2011). Our contributions thus consist in an industrial corpus creation and NER performance improvements.
We describe IRIS, a statistical machine translation (SMT) system for translating from English into Irish and vice versa. Since Irish is considered an under-resourced language with a limited amount of machine-readable text, building a machine translation system that produces reasonable translations is rather challenging. As translation is a difficult task, current research in SMT focuses on obtaining statistics either from a large amount of parallel, monolingual or other multilingual resources. Nevertheless, we collected available English-Irish data and developed an SMT system aimed at supporting human translators and enabling cross-lingual language technology tasks.
The present article reports on efforts to improve the translation accuracy of a corpus―based Machine Translation (MT) system. In order to achieve that, an error analysis performed on past translation outputs has indicated the likelihood of improving the translation accuracy by augmenting the coverage of the Target-Language (TL) side language model. The method adopted for improving the language model is initially presented, based on the concatenation of consecutive phrases. The algorithmic steps are then described that form the process for augmenting the language model. The key idea is to only augment the language model to cover the most frequent cases of phrase sequences, as counted over a TL-side corpus, in order to maximize the cases covered by the new language model entries. Experiments presented in the article show that substantial improvements in translation accuracy are achieved via the proposed method, when integrating the grown language model to the corpus-based MT system.
An open-source rule-based machine translation system is developed for Scots, a low-resourced minor language closely related to English and spoken in Scotland and Ireland. By concentrating on translation for assimilation (gist comprehension) from Scots to English, it is proposed that the development of dictionaries designed to be used with in the Apertium platform will be sufficient to produce translations that improve non-Scots speakers understanding of the language. Mono- and bilingual Scots dictionaries are constructed using lexical items gathered from a variety of resources across several domains. Although the primary goal of this project is translation for gisting, the system is evaluated for both assimilation and dissemination (publication-ready translations). A variety of evaluation methods are used, including a cloze test undertaken by human volunteers. While evaluation results are comparable to, and in some cases superior to, those of other language pairs within the Apertium platform, room for improvement is identified in several areas of the system.
This paper describes a hybrid machine translation system that explores a parser to acquire syntactic chunks of a source sentence, translates the chunks with multiple online machine translation (MT) system application program interfaces (APIs) and creates output by combining translated chunks to obtain the best possible translation. The selection of the best translation hypothesis is performed by calculating the perplexity for each translated chunk. The goal of this approach is to enhance the baseline multi-system hybrid translation (MHyT) system that uses only a language model to select best translation from translations obtained with different APIs and to improve overall English ― Latvian machine translation quality over each of the individual MT APIs. The presented syntax-based multi-system translation (SyMHyT) system demonstrates an improvement in terms of BLEU and NIST scores compared to the baseline system. Improvements reach from 1.74 up to 2.54 BLEU points.
In this paper, we address the problem of Machine Translation (MT) for a specialised domain in a language pair for which only a very small domain-specific parallel corpus is available. We conduct a series of experiments using a purely phrase-based SMT (PBSMT) system and a hybrid MT system (TectoMT), testing three different strategies to overcome the problem of the small amount of in-domain training data. Our results show that adding a small size in-domain bilingual terminology to the small in-domain training corpus leads to the best improvements of a hybrid MT system, while the PBSMT system achieves the best results by adding a combination of in-domain bilingual terminology and a larger out-of-domain corpus. We focus on qualitative human evaluation of the output of two best systems (one for each approach) and perform a systematic in-depth error analysis which revealed advantages of the hybrid MT system over the pure PBSMT system for this specific task.
This paper presents CATaLog online, a new web-based MT and TM post-editing tool. CATaLog online is a freeware software that can be used through a web browser and it requires only a simple registration. The tool features a number of editing and log functions similar to the desktop version of CATaLog enhanced with several new features that we describe in detail in this paper. CATaLog online is designed to allow users to post-edit both translation memory segments as well as machine translation output. The tool provides a complete set of log information currently not available in most commercial CAT tools. Log information can be used both for project management purposes as well as for the study of the translation process and translator’s productivity.
This paper presents the multimodal Interlingual Map Task Corpus (ILMT-s2s corpus) collected at Trinity College Dublin, and discuss some of the issues related to the collection and analysis of the data. The corpus design is inspired by the HCRC Map Task Corpus which was initially designed to support the investigation of linguistic phenomena, and has been the focus of a variety of studies of communicative behaviour. The simplicity of the task, and the complexity of phenomena it can elicit, make the map task an ideal object of study. Although there are studies that used replications of the map task to investigate communication in computer mediated tasks, this ILMT-s2s corpus is, to the best of our knowledge, the first investigation of communicative behaviour in the presence of three additional “filters”: Automatic Speech Recognition (ASR), Machine Translation (MT) and Text To Speech (TTS) synthesis, where the instruction giver and the instruction follower speak different languages. This paper details the data collection setup and completed annotation of the ILMT-s2s corpus, and outlines preliminary results obtained from the data.
We propose named entity abstraction methods with fine-grained named entity labels for improving statistical machine translation (SMT). The methods are based on a bilingual named entity recognizer that uses a monolingual named entity recognizer with transliteration. Through experiments, we demonstrate that incorporating fine-grained named entities into statistical machine translation improves the accuracy of SMT with more adequate granularity compared with the standard SMT, which is a non-named entity abstraction method.
In this paper we present our work on the usage of lexical resources for the Machine Translation English and Malayalam. We describe a comparative performance between different Statistical Machine Translation (SMT) systems on top of phrase based SMT system as baseline. We explore different ways of utilizing lexical resources to improve the quality of English Malayalam statistical machine translation. In order to enrich the training corpus we have augmented the lexical resources in two ways (a) additional vocabulary and (b) inflected verbal forms. Lexical resources include IndoWordnet semantic relation set, lexical words and verb phrases etc. We have described case studies, evaluations and have given detailed error analysis for both Malayalam to English and English to Malayalam machine translation systems. We observed significant improvement in evaluations of translation quality. Lexical resources do help uplift performance when parallel corpora are scanty.
Resources for evaluating sentence-level and word-level alignment algorithms are unsatisfactory. Regarding sentence alignments, the existing data is too scarce, especially when it comes to difficult bitexts, containing instances of non-literal translations. Regarding word-level alignments, most available hand-aligned data provide a complete annotation at the level of words that is difficult to exploit, for lack of a clear semantics for alignment links. In this study, we propose new methodologies for collecting human judgements on alignment links, which have been used to annotate 4 new data sets, at the sentence and at the word level. These will be released online, with the hope that they will prove useful to evaluate alignment software and quality estimation tools for automatic alignment. Keywords: Parallel corpora, Sentence Alignments, Word Alignments, Confidence Estimation
We present PROTEST, a test suite for the evaluation of pronoun translation by MT systems. The test suite comprises 250 hand-selected pronoun tokens and an automatic evaluation method which compares the translations of pronouns in MT output with those in the reference translation. Pronoun translations that do not match the reference are referred for manual evaluation. PROTEST is designed to support analysis of system performance at the level of individual pronoun groups, rather than to provide a single aggregate measure over all pronouns. We wish to encourage detailed analyses to highlight issues in the handling of specific linguistic mechanisms by MT systems, thereby contributing to a better understanding of those problems involved in translating pronouns. We present two use cases for PROTEST: a) for measuring improvement/degradation of an incremental system change, and b) for comparing the performance of a group of systems whose design may be largely unrelated. Following the latter use case, we demonstrate the application of PROTEST to the evaluation of the systems submitted to the DiscoMT 2015 shared task on pronoun translation.
Out-of-vocabulary (OOV) word is a crucial problem in statistical machine translation (SMT) with low resources. OOV paraphrasing that augments the translation model for the OOV words by using the translation knowledge of their paraphrases has been proposed to address the OOV problem. In this paper, we propose using word embeddings and semantic lexicons for OOV paraphrasing. Experiments conducted on a low resource setting of the OLYMPICS task of IWSLT 2012 verify the effectiveness of our proposed method.
We introduce the Denoised Web Treebank: a treebank including a normalization layer and a corresponding evaluation metric for dependency parsing of noisy text, such as Tweets. This benchmark enables the evaluation of parser robustness as well as text normalization methods, including normalization as machine translation and unsupervised lexical normalization, directly on syntactic trees. Experiments show that text normalization together with a combination of domain-specific and generic part-of-speech taggers can lead to a significant improvement in parsing accuracy on this test set.
In this paper we propose an approach to predict punctuation marks for unsegmented speech transcript. The approach is purely lexical, with pre-trained Word Vectors as the only input. A training model of Deep Neural Network (DNN) or Convolutional Neural Network (CNN) is applied to classify whether a punctuation mark should be inserted after the third word of a 5-words sequence and which kind of punctuation mark the inserted one should be. TED talks within IWSLT dataset are used in both training and evaluation phases. The proposed approach shows its effectiveness by achieving better result than the state-of-the-art lexical solution which works with same type of data, especially when predicting puncuation position only.
Greedy transition-based parsers are appealing for their very fast speed, with reasonably high accuracies. In this paper, we build a fast shift-reduce neural constituent parser by using a neural network to make local decisions. One challenge to the parsing speed is the large hidden and output layer sizes caused by the number of constituent labels and branching options. We speed up the parser by using a hierarchical output layer, inspired by the hierarchical log-bilinear neural language model. In standard WSJ experiments, the neural parser achieves an almost 2.4 time speed up (320 sen/sec) compared to a non-hierarchical baseline without significant accuracy loss (89.06 vs 89.13 F-score).
We focus on the improvement of accuracy of raw text parsing, from the viewpoint of language resource addition. In Japanese, the raw text parsing is divided into three steps: word segmentation, part-of-speech tagging, and dependency parsing. We investigate the contribution of language resource addition in each of three steps to the improvement in accuracy for two domain corpora. The experimental results show that this improvement depends on the target domain. For example, when we handle well-written texts of limited vocabulary, white paper, an effective language resource is a word-POS pair sequence corpus for the parsing accuracy. So we conclude that it is important to check out the characteristics of the target domain and to choose a suitable language resource addition strategy for the parsing accuracy improvement.
We present E-TIPSY, a search query corpus annotated with named Entities, Term Importance, POS tags, and SYntactic parses. This corpus contains crowdsourced (gold) annotations of the three most important terms in each query. In addition, it contains automatically produced annotations of named entities, part-of-speech tags, and syntactic parses for the same queries. This corpus comes in two formats: (1) Sober Subset: annotations that two or more crowd workers agreed upon, and (2) Full Glass: all annotations. We analyze the strikingly low correlation between term importance and syntactic headedness, which invites research into effective ways of combining these different signals. Our corpus can serve as a benchmark for term importance methods aimed at improving search engine quality and as an initial step toward developing a dataset of gold linguistic analysis of web search queries. In addition, it can be used as a basis for linguistic inquiries into the kind of expressions used in search.
Compared to well-resourced languages such as English and Dutch, natural language processing (NLP) tools for Afrikaans are still not abundant. In the context of the AfriBooms project, KU Leuven and the North-West University collaborated to develop a first, small treebank, a dependency parser, and an easy to use online linguistic search engine for Afrikaans for use by researchers and students in the humanities and social sciences. The search tool is based on a similar development for Dutch, i.e. GrETEL, a user-friendly search engine which allows users to query a treebank by means of a natural language example instead of a formal search instruction.
The Index Thomisticus Treebank is the largest available treebank for Latin; it contains Medieval Latin texts by Thomas Aquinas. After experimenting on its data with a number of dependency parsers based on different supervised machine learning techniques, we found that DeSR with a multilayer perceptron algorithm, a right-to-left transition, and a tailor-made feature model is the parser providing the highest accuracy rates. We improved the results further by using a technique that combines the output parses of DeSR with those provided by other parsers, outperforming the previous state of the art in parsing the Index Thomisticus Treebank. The key idea behind such improvement is to ensure a sufficient diversity and accuracy of the outputs to be combined; for this reason, we performed an in-depth evaluation of the results provided by the different parsers that we combined. Finally, we assessed that, although the general architecture of the parser is portable to Classical Latin, yet the model trained on Medieval Latin is inadequate for such purpose.
Phrase chunking remains an important natural language processing (NLP) technique for intermediate syntactic processing. This paper describes the development of protocols, annotated phrase chunking data sets and automatic phrase chunkers for ten South African languages. Various problems with adapting the existing annotation protocols of English are discussed as well as an overview of the annotated data sets. Based on the annotated sets, CRF-based phrase chunkers are created and tested with a combination of different features, including part of speech tags and character n-grams. The results of the phrase chunking evaluation show that disjunctively written languages can achieve notably better results for phrase chunking with a limited data set than conjunctive languages, but that the addition of character n-grams improve the results for conjunctive languages.
Continuous word representations appeared to be a useful feature in many natural language processing tasks. Using fixed-dimension pre-trained word embeddings allows avoiding sparse bag-of-words representation and to train models with fewer parameters. In this paper, we use fixed pre-trained word embeddings as additional features for a neural scoring function in the MST parser. With the multi-layer architecture of the scoring function we can avoid handcrafting feature conjunctions. The continuous word representations on the input also allow us to reduce the number of lexical features, make the parser more robust to out-of-vocabulary words, and reduce the total number of parameters of the model. Although its accuracy stays below the state of the art, the model size is substantially smaller than with the standard features set. Moreover, it performs well for languages where only a smaller treebank is available and the results promise to be useful in cross-lingual parsing.
We describe a method for analysing Constraint Grammars (CG) that can detect internal conflicts and redundancies in a given grammar, without the need for a corpus. The aim is for grammar writers to be able to automatically diagnose, and then manually improve their grammars. Our method works by translating the given grammar into logical constraints that are analysed by a SAT-solver. We have evaluated our analysis on a number of non-trivial grammars and found inconsistencies.
The treatment of medieval texts is a particular challenge for parsers. I compare how two dependency parsers, one graph-based, the other transition-based, perform on Old French, facing some typical problems of medieval texts: graphical variation, relatively free word order, and syntactic variation of several parameters over a diachronic period of about 300 years. Both parsers were trained and evaluated on the “Syntactic Reference Corpus of Medieval French” (SRCMF), a manually annotated dependency treebank. I discuss the relation between types of parsers and types of language, as well as the differences of the analyses from a linguistic point of view.
Parsing Web information, namely parsing content to find relevant documents on the basis of a user’s query, represents a crucial step to guarantee fast and accurate Information Retrieval (IR). Generally, an automated approach to such task is considered faster and cheaper than manual systems. Nevertheless, results do not seem have a high level of accuracy, indeed, as also Hjorland (2007) states, using stochastic algorithms entails: • Low precision due to the indexing of common Atomic Linguistic Units (ALUs) or sentences. • Low recall caused by the presence of synonyms. • Generic results arising from the use of too broad or too narrow terms. Usually IR systems are based on invert text index, namely an index data structure storing a mapping from content to its locations in a database file, or in a document or a set of documents. In this paper we propose a system, by means of which we will develop a search engine able to process online documents, starting from a natural language query, and to return information to users. The proposed approach, based on the Lexicon-Grammar (LG) framework and its language formalization methodologies, aims at integrating a semantic annotation process for both query analysis and document retrieval.
The Interdisciplinary Longitudinal Study on Adult Development and Aging (ILSE) was created to facilitate the study of challenges posed by rapidly aging societies in developed countries such as Germany. ILSE contains over 8,000 hours of biographic interviews recorded from more than 1,000 participants over the course of 20 years. Investigations on various aspects of aging, such as cognitive decline, often rely on the analysis of linguistic features which can be derived from spoken content like these interviews. However, transcribing speech is a time and cost consuming manual process and so far only 380 hours of ILSE interviews have been transcribed. Thus, it is the aim of our work to establish technical systems to fully automatically transcribe the ILSE interview data. The joint occurrence of poor recording quality, long audio segments, erroneous transcriptions, varying speaking styles & crosstalk, and emotional & dialectal speech in these interviews presents challenges for automatic speech recognition (ASR). We describe our ongoing work towards the fully automatic transcription of all ILSE interviews and the steps we implemented in preparing the transcriptions to meet the interviews’ challenges. Using a recursive long audio alignment procedure 96 hours of the transcribed data have been made accessible for ASR training.
A speech database has been collected for use to highlight the importance of “speaker factor” in forensic voice comparison. FABIOLE has been created during the FABIOLE project funded by the French Research Agency (ANR) from 2013 to 2016. This corpus consists in more than 3 thousands excerpts spoken by 130 French native male speakers. The speakers are divided into two categories: 30 target speakers who everyone has 100 excerpts and 100 “impostors” who everyone has only one excerpt. The data were collected from 10 different French radio and television shows where each utterance turns with a minimum duration of 30s and has a good speech quality. The data set is mainly used for investigating speaker factor in forensic voice comparison and interpreting some unsolved issue such as the relationship between speaker characteristics and system behavior. In this paper, we present FABIOLE database. Then, preliminary experiments are performed to evaluate the effect of the “speaker factor” and the show on a voice comparison system behavior.
Corpus design for speech synthesis is a well-researched topic in languages such as English compared to Modern Standard Arabic, and there is a tendency to focus on methods to automatically generate the orthographic transcript to be recorded (usually greedy methods). In this work, a study of Modern Standard Arabic (MSA) phonetics and phonology is conducted in order to create criteria for a greedy method to create a speech corpus transcript for recording. The size of the dataset is reduced a number of times using these optimisation methods with different parameters to yield a much smaller dataset with identical phonetic coverage than before the reduction, and this output transcript is chosen for recording. This is part of a larger work to create a completely annotated and segmented speech corpus for MSA.
With emerging conversational data, automated content analysis is needed for better data interpretation, so that it is accurately understood and can be effectively integrated and utilized in various applications. ICSI meeting corpus is a publicly released data set of multi-party meetings in an organization that has been released over a decade ago, and has been fostering meeting understanding research since then. The original data collection includes transcription of participant turns as well as meta-data annotations, such as disfluencies and dialog act tags. This paper presents an extended set of annotations for the ICSI meeting corpus with a goal of deeply understanding meeting conversations, where participant turns are annotated by actionable items that could be performed by an automated meeting assistant. In addition to the user utterances that contain an actionable item, annotations also include the arguments associated with the actionable item. The set of actionable items are determined by aligning human-human interactions to human-machine interactions, where a data annotation schema designed for a virtual personal assistant (human-machine genre) is adapted to the meetings domain (human-human genre). The data set is formed by annotating participants’ utterances in meetings with potential intents/actions considering their contexts. The set of actions target what could be accomplished by an automated meeting assistant, such as taking a note of action items that a participant commits to, or finding emails or topic related documents that were mentioned during the meeting. A total of 10 defined intents/actions are considered as actionable items in meetings. Turns that include actionable intents were annotated for 22 public ICSI meetings, that include a total of 21K utterances, segmented by speaker turns. Participants’ spoken turns, possible actions along with associated arguments and their vector representations as computed by convolutional deep structured semantic models are included in the data set for future research. We present a detailed statistical analysis of the data set and analyze the performance of applying convolutional deep structured semantic models for an actionable item detection task. The data is available at http://research.microsoft.com/ projects/meetingunderstanding/.
Current state-of-the-art speech synthesizers for domain-independent systems still struggle with the challenge of generating understandable and natural-sounding speech. This is mainly because the pronunciation of words of foreign origin, inflections and compound words often cannot be handled by rules. Furthermore there are too many of these for inclusion in exception dictionaries. We describe an approach to evaluating text-to-speech synthesizers with a subjective listening experiment. The focus is to differentiate between known problem classes for speech synthesizers. The target language is German but we believe that many of the described phenomena are not language specific. We distinguish the following problem categories: Normalization, Foreign linguistics, Natural writing, Language specific and General. Each of them is divided into five to three problem classes. Word lists for each of the above mentioned categories were compiled and synthesized by both a commercial and an open source synthesizer, both being based on the non-uniform unit-selection approach. The synthesized speech was evaluated by human judges using the Speechalyzer toolkit and the results are discussed. It shows that, as expected, the commercial synthesizer performs much better than the open-source one, and especially words of foreign origin were pronounced badly by both systems.
Recent spoken dialog systems have been able to recognize freely spoken user input in restricted domains thanks to statistical methods in the automatic speech recognition. These methods require a high number of natural language utterances to train the speech recognition engine and to assess the quality of the system. Since human speech offers many variants associated with a single intent, a high number of user utterances have to be elicited. Developers are therefore turning to crowdsourcing to collect this data. This paper compares three different methods to elicit multiple utterances for given semantics via crowd sourcing, namely with pictures, with text and with semantic entities. Specifically, we compare the methods with regard to the number of valid data and linguistic variance, whereby a quantitative and qualitative approach is proposed. In our study, the method with text led to a high variance in the utterances and a relatively low rate of invalid data.
This paper describes the characteristics and structure of a Basque singing voice database of bertsolaritza. Bertsolaritza is a popular singing style from Basque Country sung exclusively in Basque that is improvised and a capella. The database is designed to be used in statistical singing voice synthesis for bertsolaritza style. Starting from the recordings and transcriptions of numerous singers, diarization and phoneme alignment experiments have been made to extract the singing voice from the recordings and create phoneme alignments. This labelling processes have been performed applying standard speech processing techniques and the results prove that these techniques can be used in this specific singing style.
We introduce a unique, comprehensive Austrian German multi-sensor corpus with moving and non-moving speakers to facilitate the evaluation of estimators and detectors that jointly detect a speaker’s spatial and temporal parameters. The corpus is suitable for various machine learning and signal processing tasks, linguistic studies, and studies related to a speaker’s fundamental frequency (due to recorded glottograms). Available corpora are limited to (synthetically generated/spatialized) speech data or recordings of musical instruments that lack moving speakers, glottograms, and/or multi-channel distant speech recordings. That is why we recorded 24 spatially non-moving and moving speakers, balanced male and female, to set up a two-room and 43-channel Austrian German multi-sensor speech corpus. It contains 8.2 hours of read speech based on phonetically balanced sentences, commands, and digits. The orthographic transcriptions include around 53,000 word tokens and 2,070 word types. Special features of this corpus are the laryngograph recordings (representing glottograms required to detect a speaker’s instantaneous fundamental frequency and pitch), corresponding clean-speech recordings, and spatial information and video data provided by four Kinects and a camera.
Auditory voice quality judgements are used intensively for the clinical assessment of pathological voice. Voice quality concepts are fuzzily defined and poorly standardized however, which hinders scientific and clinical communication. The described database documents a wide variety of pathologies and is used to investigate auditory voice quality concepts with regard to phonation mechanisms. The database contains 375 laryngeal high-speed videos and simultaneous high-quality audio recordings of sustained phonations of 80 pathological and 40 non-pathological subjects. Interval wise annotations regarding video and audio quality, as well as voice quality ratings are provided. Video quality is annotated for the visibility of anatomical structures and artefacts such as blurring or reduced contrast. Voice quality annotations include ratings on the presence of dysphonia and diplophonia. The purpose of the database is to aid the formulation of observationally well-founded models of phonation and the development of model-based automatic detectors for distinct types of phonation, especially for clinically relevant nonmodal voice phenomena. Another application is the training of audio-based fundamental frequency extractors on video-based reference fundamental frequencies.
In this paper we introduce a Bulgarian speech database, which was created for the purpose of ASR technology development. The paper describes the design and the content of the speech database. We present also an empirical evaluation of the performance of a LVCSR system for Bulgarian trained on the BulPhonC data. The resource is available free for scientific usage.
In this paper the authors present a speech corpus designed and created for the development and evaluation of dictation systems in Latvian. The corpus consists of over nine hours of orthographically annotated speech from 30 different speakers. The corpus features spoken commands that are common for dictation systems for text editors. The corpus is evaluated in an automatic speech recognition scenario. Evaluation results in an ASR dictation scenario show that the addition of the corpus to the acoustic model training data in combination with language model adaptation allows to decrease the WER by up to relative 41.36% (or 16.83% in absolute numbers) compared to a baseline system without language model adaptation. Contribution of acoustic data augmentation is at relative 12.57% (or 3.43% absolute).
This paper introduces the LetsRead Corpus of European Portuguese read speech from 6 to 10 years old children. The motivation for the creation of this corpus stems from the inexistence of databases with recordings of reading tasks of Portuguese children with different performance levels and including all the common reading aloud disfluencies. It is also essential to develop techniques to fulfill the main objective of the LetsRead project: to automatically evaluate the reading performance of children through the analysis of reading tasks. The collected data amounts to 20 hours of speech from 284 children from private and public Portuguese schools, with each child carrying out two tasks: reading sentences and reading a list of pseudowords, both with varying levels of difficulty throughout the school grades. In this paper, the design of the reading tasks presented to children is described, as well as the collection procedure. Manually annotated data is analyzed according to disfluencies and reading performance. The considered word difficulty parameter is also confirmed to be suitable for the pseudoword reading tasks.
The BAS CLARIN speech data repository is introduced. At the current state it comprises 31 pre-dominantly German corpora of spoken language. It is compliant to the CLARIN-D as well as the OLAC requirements. This enables its embedding into several infrastructures. We give an overview over its structure, its implementation as well as the corpora it contains.
We present a new Dutch dysarthric speech database containing utterances of neurological patients with Parkinson’s disease, traumatic brain injury and cerebrovascular accident. The speech content is phonetically and linguistically diversified by using numerous structured sentence and word lists. Containing more than 6 hours of mildly to moderately dysarthric speech, this database can be used for research on dysarthria and for developing and testing speech-to-text systems designed for medical applications. Current activities aimed at extending this database are also discussed.
Language resources, such as corpora, are important for various natural language processing tasks. Urdu has millions of speakers around the world but it is under-resourced in terms of standard evaluation resources. This paper reports the construction of a benchmark corpus for Urdu summaries (abstracts) to facilitate the development and evaluation of single document summarization systems for Urdu language. In Urdu, space does not always mark word boundary. Therefore, we created two versions of the same corpus. In the first version, words are separated by space. In contrast, proper word boundaries are manually tagged in the second version. We further apply normalization, part-of-speech tagging, morphological analysis, lemmatization, and stemming for the articles and their summaries in both versions. In order to apply these annotations, we re-implemented some NLP tools for Urdu. We provide Urdu Summary Corpus, all these annotations and the needed software tools (as open-source) for researchers to run experiments and to evaluate their work including but not limited to single-document summarization task.
In this paper we report our effort to construct the first ever Indonesian corpora for chat summarization. Specifically, we utilized documents of multi-participant chat from a well known online instant messaging application, WhatsApp. We construct the gold standard by asking three native speakers to manually summarize 300 chat sections (152 of them contain images). As result, three reference summaries in extractive and either abstractive form are produced for each chat sections. The corpus is still in its early stage of investigation, yielding exciting possibilities of future works.
Evaluation of text summarization approaches have been mostly based on metrics that measure similarities of system generated summaries with a set of human written gold-standard summaries. The most widely used metric in summarization evaluation has been the ROUGE family. ROUGE solely relies on lexical overlaps between the terms and phrases in the sentences; therefore, in cases of terminology variations and paraphrasing, ROUGE is not as effective. Scientific article summarization is one such case that is different from general domain summarization (e.g. newswire data). We provide an extensive analysis of ROUGE’s effectiveness as an evaluation metric for scientific summarization; we show that, contrary to the common belief, ROUGE is not much reliable in evaluating scientific summaries. We furthermore show how different variants of ROUGE result in very different correlations with the manual Pyramid scores. Finally, we propose an alternative metric for summarization evaluation which is based on the content relevance between a system generated summary and the corresponding human written summaries. We call our metric SERA (Summarization Evaluation by Relevance Analysis). Unlike ROUGE, SERA consistently achieves high correlations with manual scores which shows its effectiveness in evaluation of scientific article summarization.
In this paper we present the OnForumS corpus developed for the shared task of the same name on Online Forum Summarisation (OnForumS at MultiLing’15). The corpus consists of a set of news articles with associated readers’ comments from The Guardian (English) and La Repubblica (Italian). It comes with four levels of annotation: argument structure, comment-article linking, sentiment and coreference. The former three were produced through crowdsourcing, whereas the latter, by an experienced annotator using a mature annotation scheme. Given its annotation breadth, we believe the corpus will prove a useful resource in stimulating and furthering research in the areas of Argumentation Mining, Summarisation, Sentiment, Coreference and the interlinks therein.
WordNet represents a cornerstone in the Computational Linguistics field, linking words to meanings (or senses) through a taxonomical representation of synsets, i.e., clusters of words with an equivalent meaning in a specific context often described by few definitions (or glosses) and examples. Most of the approaches to the Word Sense Disambiguation task fully rely on these short texts as a source of contextual information to match with the input text to disambiguate. This paper presents the first attempt to enrich synsets data with common-sense definitions, automatically retrieved from ConceptNet 5, and disambiguated accordingly to WordNet. The aim was to exploit the shared- and immediate-thinking nature of common-sense knowledge to extend the short but incredibly useful contextual information of the synsets. A manual evaluation on a subset of the entire result (which counts a total of almost 600K synset enrichments) shows a very high precision with an estimated good recall.
We present VPS-GradeUp ― a set of 11,400 graded human decisions on usage patterns of 29 English lexical verbs from the Pattern Dictionary of English Verbs by Patrick Hanks. The annotation contains, for each verb lemma, a batch of 50 concordances with the given lemma as KWIC, and for each of these concordances we provide a graded human decision on how well the individual PDEV patterns for this particular lemma illustrate the given concordance, indicated on a 7-point Likert scale for each PDEV pattern. With our annotation, we were pursuing a pilot investigation of the foundations of human clustering and disambiguation decisions with respect to usage patterns of verbs in context. The data set is publicly available at http://hdl.handle.net/11234/1-1585.
We describe the construction of GLASS, a newly sense-annotated version of the German lexical substitution data set used at the GermEval 2015: LexSub shared task. Using the two annotation layers, we conduct the first known empirical study of the relationship between manually applied word senses and lexical substitutions. We find that synonymy and hypernymy/hyponymy are the only semantic relations directly linking targets to their substitutes, and that substitutes in the target’s hypernymy/hyponymy taxonomy closely align with the synonyms of a single GermaNet synset. Despite this, these substitutes account for a minority of those provided by the annotators. The results of our analysis accord with those of a previous study on English-language data (albeit with automatically induced word senses), leading us to suspect that the sense―substitution relations we discovered may be of a universal nature. We also tentatively conclude that relatively cheap lexical substitution annotations can be used as a knowledge source for automatic WSD. Also introduced in this paper is Ubyline, the web application used to produce the sense annotations. Ubyline presents an intuitive user interface optimized for annotating lexical sample data, and is readily adaptable to sense inventories other than GermaNet.
We present an annotation study on a representative dataset of literal and idiomatic uses of German infinitive-verb compounds in newspaper and journal texts. Infinitive-verb compounds form a challenge for writers of German, because spelling regulations are different for literal and idiomatic uses. Through the participation of expert lexicographers we were able to obtain a high-quality corpus resource which offers itself as a testbed for automatic idiomaticity detection and coarse-grained word-sense disambiguation. We trained a classifier on the corpus which was able to distinguish literal and idiomatic uses with an accuracy of 85 %.
We launch the SemDaX corpus which is a recently completed Danish human-annotated corpus available through a CLARIN academic license. The corpus includes approx. 90,000 words, comprises six textual domains, and is annotated with sense inventories of different granularity. The aim of the developed corpus is twofold: i) to assess the reliability of the different sense annotation schemes for Danish measured by qualitative analyses and annotation agreement scores, and ii) to serve as training and test data for machine learning algorithms with the practical purpose of developing sense taggers for Danish. To these aims, we take a new approach to human-annotated corpus resources by double annotating a much larger part of the corpus than what is normally seen: for the all-words task we double annotated 60% of the material and for the lexical sample task 100%. We include in the corpus not only the adjucated files, but also the diverging annotations. In other words, we consider not all disagreement to be noise, but rather to contain valuable linguistic information that can help us improve our annotation schemes and our learning algorithms.
We present a pilot analysis of a new linguistic resource, VPS-GradeUp (available at http://hdl.handle.net/11234/1-1585). The resource contains 11,400 graded human decisions on usage patterns of 29 English lexical verbs, randomly selected from the Pattern Dictionary of English Verbs (Hanks, 2000 2014) based on their frequency and the number of senses their lemmas have in PDEV. This data set has been created to observe the interannotator agreement on PDEV patterns produced using the Corpus Pattern Analysis (Hanks, 2013). Apart from the graded decisions, the data set also contains traditional Word-Sense-Disambiguation (WSD) labels. We analyze the associations between the graded annotation and WSD annotation. The results of the respective annotations do not correlate with the size of the usage pattern inventory for the respective verbs lemmas, which makes the data set worth further linguistic analysis.
Chinese sentences are written as sequences of characters, which are elementary units of syntax and semantics. Characters are highly polysemous in forming words. We present a position-sensitive skip-gram model to learn multi-prototype Chinese character embeddings, and explore the usefulness of such character embeddings to Chinese NLP tasks. Evaluation on character similarity shows that multi-prototype embeddings are significantly better than a single-prototype baseline. In addition, used as features in the Chinese NER task, the embeddings result in a 1.74% F-score improvement over a state-of-the-art baseline.
Named Entity Disambiguation (NED) is the task of linking a named-entity mention to an instance in a knowledge-base, typically Wikipedia-derived resources like DBpedia. This task is closely related to word-sense disambiguation (WSD), where the mention of an open-class word is linked to a concept in a knowledge-base, typically WordNet. This paper analyzes the relation between two annotated datasets on NED and WSD, highlighting the commonalities and differences. We detail the methods to construct a NED system following the WSD word-expert approach, where we need a dictionary and one classifier is built for each target entity mention string. Constructing a dictionary for NED proved challenging, and although similarity and ambiguity are higher for NED, the results are also higher due to the larger number of training data, and the more crisp and skewed meaning differences.
The experiments presented here exploit the properties of the Apertium RDF Graph, principally cycle density and nodes’ degree, to automatically generate new translation relations between words, and therefore to enrich existing bilingual dictionaries with new entries. Currently, the Apertium RDF Graph includes data from 22 Apertium bilingual dictionaries and constitutes a large unified array of linked lexical entries and translations that are available and accessible on the Web (http://linguistic.linkeddata.es/apertium/). In particular, its graph structure allows for interesting exploitation opportunities, some of which are addressed in this paper. Two ‘massive’ experiments are reported: in the first one, the original EN-ES translation set was removed from the Apertium RDF Graph and a new EN-ES version was generated. The results were compared against the previously removed EN-ES data and against the Concise Oxford Spanish Dictionary. In the second experiment, a new non-existent EN-FR translation set was generated. In this case the results were compared against a converted wiktionary English-French file. The results we got are really good and perform well for the extreme case of correlated polysemy. This lead us to address the possibility to use cycles and nodes degree to identify potential oddities in the source data. If cycle density proves efficient when considering potential targets, we can assume that in dense graphs nodes with low degree may indicate potential errors.
We introduce PreMOn (predicate model for ontologies), a linguistic resource for exposing predicate models (PropBank, NomBank, VerbNet, and FrameNet) and mappings between them (e.g, SemLink) as Linked Open Data. It consists of two components: (i) the PreMOn Ontology, an extension of the lemon model by the W3C Ontology-Lexica Community Group, that enables to homogeneously represent data from the various predicate models; and, (ii) the PreMOn Dataset, a collection of RDF datasets integrating various versions of the aforementioned predicate models and mapping resources. PreMOn is freely available and accessible online in different ways, including through a dedicated SPARQL endpoint.
This paper describes work on incorporating Princenton’s WordNet morphosemantics links to the fabric of the Portuguese OpenWordNet-PT. Morphosemantic links are relations between verbs and derivationally related nouns that are semantically typed (such as for tune-tuner ― in Portuguese “afinar-afinador” – linked through an “agent” link). Morphosemantic links have been discussed for Princeton’s WordNet for a while, but have not been added to the official database. These links are very useful, they help us to improve our Portuguese WordNet. Thus we discuss the integration of these links in our base and the issues we encountered with the integration.
The development of standard models for describing general lexical resources has led to the emergence of numerous lexical datasets of various languages in the Semantic Web. However, equivalent models covering the linguistic domain of morphology do not exist. As a result, there are hardly any language resources of morphemic data available in RDF to date. This paper presents the creation of the Hebrew Morpheme Inventory from a manually compiled tabular dataset comprising around 52.000 entries. It is an ongoing effort of representing the lexemes, word-forms and morphologigal patterns together with their underlying relations based on the newly created Multilingual Morpheme Ontology (MMoOn). It will be shown how segmented Hebrew language data can be granularly described in a Linked Data format, thus, serving as an exemplary case for creating morpheme inventories of any inflectional language with MMoOn. The resulting dataset is described a) according to the structure of the underlying data format, b) with respect to the Hebrew language characteristic of building word-forms directly from roots, c) by exemplifying how inflectional information is realized and d) with regard to its enrichment with external links to sense resources.
We present a new collection of multilingual corpora automatically created from the content available in the Global Voices websites, where volunteers have been posting and translating citizen media stories since 2004. We describe how we crawled and processed this content to generate parallel resources comprising 302.6K document pairs and 8.36M segment alignments in 756 language pairs. For some language pairs, the segment alignments in this resource are the first open examples of their kind. In an initial use of this resource, we discuss how a set of document pair detection algorithms performs on the Greek-English corpus.
High accuracy for automated translation and information retrieval calls for linguistic annotations at various language levels. The plethora of informal internet content sparked the demand for porting state-of-art natural language processing (NLP) applications to new social media as well as diverse language adaptation. Effort launched by the BOLT (Broad Operational Language Translation) program at DARPA (Defense Advanced Research Projects Agency) successfully addressed the internet information with enhanced NLP systems. BOLT aims for automated translation and linguistic analysis for informal genres of text and speech in online and in-person communication. As a part of this program, the Linguistic Data Consortium (LDC) developed valuable linguistic resources in support of the training and evaluation of such new technologies. This paper focuses on methodologies, infrastructure, and procedure for developing linguistic annotation at various language levels, including Treebank (TB), word alignment (WA), PropBank (PB), and co-reference (CoRef). Inspired by the OntoNotes approach with adaptations to the tasks to reflect the goals and scope of the BOLT project, this effort has introduced more annotation types of informal and free-style genres in English, Chinese and Egyptian Arabic. The corpus produced is by far the largest multi-lingual, multi-level and multi-genre annotation corpus of informal text and speech.
Large Web corpora containing full documents with permissive licenses are crucial for many NLP tasks. In this article we present the construction of 12 million-pages Web corpus (over 10 billion tokens) licensed under CreativeCommons license family in 50+ languages that has been extracted from CommonCrawl, the largest publicly available general Web crawl to date with about 2 billion crawled URLs. Our highly-scalable Hadoop-based framework is able to process the full CommonCrawl corpus on 2000+ CPU cluster on the Amazon Elastic Map/Reduce infrastructure. The processing pipeline includes license identification, state-of-the-art boilerplate removal, exact duplicate and near-duplicate document removal, and language detection. The construction of the corpus is highly configurable and fully reproducible, and we provide both the framework (DKPro C4CorpusTools) and the resulting data (C4Corpus) to the research community.
We present a new major release of the OpenSubtitles collection of parallel corpora. The release is compiled from a large database of movie and TV subtitles and includes a total of 1689 bitexts spanning 2.6 billion sentences across 60 languages. The release also incorporates a number of enhancements in the preprocessing and alignment of the subtitles, such as the automatic correction of OCR errors and the use of meta-data to estimate the quality of each subtitle and score subtitle pairs.
This paper introduces LexFr, a corpus-based French lexical resource built by adapting the framework LexIt, originally developed to describe the combinatorial potential of Italian predicates. As in the original framework, the behavior of a group of target predicates is characterized by a series of syntactic (i.e., subcategorization frames) and semantic (i.e., selectional preferences) statistical information (a.k.a. distributional profiles) whose extraction process is mostly unsupervised. The first release of LexFr includes information for 2,493 verbs, 7,939 nouns and 2,628 adjectives. In these pages we describe the adaptation process and evaluated the final resource by comparing the information collected for 20 test verbs against the information available in a gold standard dictionary. In the best performing setting, we obtained 0.74 precision, 0.66 recall and 0.70 F-measure.
Polarity lexicons are a basic resource for analyzing the sentiments and opinions expressed in texts in an automated way. This paper explores three methods to construct polarity lexicons: translating existing lexicons from other languages, extracting polarity lexicons from corpora, and annotating sentiments Lexical Knowledge Bases. Each of these methods require a different degree of human effort. We evaluate how much manual effort is needed and to what extent that effort pays in terms of performance improvement. Experiment setup includes generating lexicons for Basque, and evaluating them against gold standard datasets in different domains. Results show that extracting polarity lexicons from corpora is the best solution for achieving a good performance with reasonable human effort.
This paper describes the conversion into LMF, a standard lexicographic digital format of ‘al-qāmūs al-muḥīṭ, a Medieval Arabic lexicon. The lexicon is first described, then all the steps required for the conversion are illustrated. The work is will produce a useful lexicographic resource for Arabic NLP, but is also interesting per se, to study the implications of adapting the LMF model to the Arabic language. Some reflections are offered as to the status of roots with respect to previously suggested representations. In particular, roots are, in our opinion are to be not treated as lexical entries, but modeled as lexical metadata for classifying and identifying lexical entries. In this manner, each root connects all entries that are derived from it.
In this work we introduce and describe a language resource composed of lists of simpler synonyms for Spanish. The synonyms are divided in different senses taken from the Spanish OpenThesaurus, where context disambiguation was performed by using statistical information from the Web and Google Books Ngrams. This resource is freely available online and can be used for different NLP tasks such as lexical simplification. Indeed, so far it has been already integrated into four tools.
The National Library of Finland has digitized a large proportion of the historical newspapers published in Finland between 1771 and 1910 (Bremer-Laamanen 2001). This collection contains approximately 1.95 million pages in Finnish and Swedish. Finnish part of the collection consists of about 2.39 billion words. The National Library’s Digital Collections are offered via the digi.kansalliskirjasto.fi web service, also known as Digi. Part of this material is also available freely downloadable in The Language Bank of Finland provided by the Fin-CLARIN consortium . The collection can also be accessed through the Korp environment that has been developed by Spräkbanken at the University of Gothenburg and extended by FIN-CLARIN team at the University of Helsinki to provide concordances of text resources. A Cranfield-style information retrieval test collection has been produced out of a small part of the Digi newspaper material at the University of Tampere (Järvelin et al., 2015). The quality of the OCRed collections is an important topic in digital humanities, as it affects general usability and searchability of collections. There is no single available method to assess the quality of large collections, but different methods can be used to approximate the quality. This paper discusses different corpus analysis style ways to approximate the overall lexical quality of the Finnish part of the Digi collection.
A trend to digitize historical paper-based archives has emerged in recent years, with the advent of digital optical scanners. A lot of paper-based books, textbooks, magazines, articles, and documents are being transformed into electronic versions that can be manipulated by a computer. For this purpose, Optical Character Recognition (OCR) systems have been developed to transform scanned digital text into editable computer text. However, different kinds of errors in the OCR system output text can be found, but Automatic Error Correction tools can help in performing the quality of electronic texts by cleaning and removing noises. In this paper, we perform a qualitative and quantitative comparison of several error-correction techniques for historical French documents. Experimentation shows that our Machine Translation for Error Correction method is superior to other Language Modelling correction techniques, with nearly 13% relative improvement compared to the initial baseline.
We present further work on evaluation of the fully automatic post-correction of Early Dutch Books Online, a collection of 10,333 18th century books. In prior work we evaluated the new implementation of Text-Induced Corpus Clean-up (TICCL) on the basis of a single book Gold Standard derived from this collection. In the current paper we revisit the same collection on the basis of a sizeable 1020 item random sample of OCR post-corrected strings from the full collection. Both evaluations have their own stories to tell and lessons to teach.
Crowdsourcing approaches for post-correction of OCR output (Optical Character Recognition) have been successfully applied to several historic text collections. We report on our crowd-correction platform Kokos, which we built to improve the OCR quality of the digitized yearbooks of the Swiss Alpine Club (SAC) from the 19th century. This multilingual heritage corpus consists of Alpine texts mainly written in German and French, all typeset in Antiqua font. Finding and engaging volunteers for correcting large amounts of pages into high quality text requires a carefully designed user interface, an easy-to-use workflow, and continuous efforts for keeping the participants motivated. More than 180,000 characters on about 21,000 pages were corrected by volunteers in about 7 month, achieving an OCR gold standard with a systematically evaluated accuracy of 99.7% on the word level. The crowdsourced OCR gold standard and the corresponding original OCR recognition results from Abby FineReader 7 for each page are available as a resource. Additionally, the scanned images (300dpi) of all pages are included in order to facilitate tests with other OCR software.
Given a controversial issue, argument mining from natural language texts (news papers, and any form of text on the Internet) is extremely challenging: domain knowledge is often required together with appropriate forms of inferences to identify arguments. This contribution explores the types of knowledge that are required and how they can be paired with reasoning schemes, language processing and language resources to accurately mine arguments. We show via corpus analysis that the Generative Lexicon, enhanced in different manners and viewed as both a lexicon and a domain knowledge representation, is a relevant approach. In this paper, corpus annotation for argument mining is first developed, then we show how the generative lexicon approach must be adapted and how it can be paired with language processing patterns to extract and specify the nature of arguments. Our approach to argument mining is thus knowledge driven.
In the present paper, we analyse variation of discourse phenomena in two typologically different languages, i.e. in German and Czech. The novelty of our approach lies in the nature of the resources we are using. Advantage is taken of existing resources, which are, however, annotated on the basis of two different frameworks. We use an interoperable scheme unifying discourse phenomena in both frameworks into more abstract categories and considering only those phenomena that have a direct match in German and Czech. The discourse properties we focus on are relations of identity, semantic similarity, ellipsis and discourse relations. Our study shows that the application of interoperable schemes allows an exploitation of discourse-related phenomena analysed in different projects and on the basis of different frameworks. As corpus compilation and annotation is a time-consuming task, positive results of this experiment open up new paths for contrastive linguistics, translation studies and NLP, including machine translation.
In sources used in oral history research (such as interviews with eye witnesses), passages where the degree of personal emotional involvement is found to be high can be of particular interest, as these may give insight into how historical events were experienced, and what moral dilemmas and psychological or religious struggles were encountered. In a pilot study involving a large corpus of interview recordings with Dutch war veterans, we have investigated if it is possible to develop a method for automatically identifying those passages where the degree of personal emotional involvement is high. The method is based on the automatic detection of exceptionally large silences and filled pause segments (using Automatic Speech Recognition), and cues taken from specific n-grams. The first results appear to be encouraging enough for further elaboration of the method.
Existing discourse research only focuses on the monolingual languages and the inconsistency between languages limits the power of the discourse theory in multilingual applications such as machine translation. To address this issue, we design and build a bilingual discource corpus in which we are currently defining and annotating the bilingual elementary discourse units (BEDUs). The BEDUs are then organized into hierarchical structures. Using this discourse style, we have annotated nearly 20K LDC sentences. Finally, we design a bilingual discourse based method for machine translation evaluation and show the effectiveness of our bilingual discourse annotations.
DiMLex is a lexicon of German connectives that can be used for various language understanding purposes. We enhanced the coverage to 275 connectives, which we regard as covering all known German discourse connectives in current use. In this paper, we consider the task of adding the semantic relations that can be expressed by each connective. After discussing different approaches to retrieving semantic information, we settle on annotating each connective with senses from the new PDTB 3.0 sense hierarchy. We describe our new implementation in the extended DiMLex, which will be available for research purposes.
Dispute mediation is a growing activity in the resolution of conflicts, and more and more research emerge to enhance and better understand this (until recently) understudied practice. Corpus analyses are necessary to study discourse in this context; yet, little data is available, mainly because of its confidentiality principle. After proposing hints and avenues to acquire transcripts of mediation sessions, this paper presents the Dispute Mediation Corpus, which gathers annotated excerpts of mediation dialogues. Although developed as part of a project on argumentation, it is freely available and the text data can be used by anyone. This first-ever open corpus of mediation interactions can be of interest to scholars studying discourse, but also conflict resolution, argumentation, linguistics, communication, etc. We advocate for using and extending this resource that may be valuable to a large variety of domains of research, particularly those striving to enhance the study of the rapidly growing activity of dispute mediation.
This paper presents the creation of a corpus of labeled disabilities in scientific papers. The identification of medical concepts in documents and, especially, the identification of disabilities, is a complex task mainly due to the variety of expressions that can make reference to the same problem. Currently there is not a set of documents manually annotated with disabilities with which to evaluate an automatic detection system of such concepts. This is the reason why this corpus arises, aiming to facilitate the evaluation of systems that implement an automatic annotation tool for extracting biomedical concepts such as disabilities. The result is a set of scientific papers manually annotated. For the selection of these scientific papers has been conducted a search using a list of rare diseases, since they generally have associated several disabilities of different kinds.
We present a new corpus, PersonaBank, consisting of 108 personal stories from weblogs that have been annotated with their Story Intention Graphs, a deep representation of the content of a story. We describe the topics of the stories and the basis of the Story Intention Graph representation, as well as the process of annotating the stories to produce the Story Intention Graphs and the challenges of adapting the tool to this new personal narrative domain. We also discuss how the corpus can be used in applications that retell the story using different styles of tellings, co-tellings, or as a content planner.
This paper explores several aspects together for a fine-grained Chinese discourse analysis. We deal with the issues of ambiguous discourse markers, ambiguous marker linkings, and more than one discourse marker. A universal feature representation is proposed. The pair-once postulation, cross-discourse-unit-first rule and word-pair-marker-first rule select a set of discourse markers from ambiguous linkings. Marker-Sum feature considers total contribution of markers and Marker-Preference feature captures the probability distribution of discourse functions of a representative marker by using preference rule. The HIT Chinese discourse relation treebank (HIT-CDTB) is used to evaluate the proposed models. The 25-way classifier achieves 0.57 micro-averaged F-score.
In discourse relation annotation, there is currently a variety of different frameworks being used, and most of them have been developed and employed mostly on written data. This raises a number of questions regarding interoperability of discourse relation annotation schemes, as well as regarding differences in discourse annotation for written vs. spoken domains. In this paper, we describe ouron annotating two spoken domains from the SPICE Ireland corpus (telephone conversations and broadcast interviews) according todifferent discourse annotation schemes, PDTB 3.0 and CCR. We show that annotations in the two schemes can largely be mappedone another, and discuss differences in operationalisations of discourse relation schemes which present a challenge to automatic mapping. We also observe systematic differences in the prevalence of implicit discourse relations in spoken data compared to written texts,find that there are also differences in the types of causal relations between the domains. Finally, we find that PDTB 3.0 addresses many shortcomings of PDTB 2.0 wrt. the annotation of spoken discourse, and suggest further extensions. The new corpus has roughly theof the CoNLL 2015 Shared Task test set, and we hence hope that it will be a valuable resource for the evaluation of automatic discourse relation labellers.
In this article, we present the RATP-DECODA Corpus which is composed by a set of 67 hours of speech from telephone conversations of a Customer Care Service (CCS). This corpus is already available on line at http://sldr.org/sldr000847/fr in its first version. However, many enhancements have been made in order to allow the development of automatic techniques to transcript conversations and to capture their meaning. These enhancements fall into two categories: firstly, we have increased the size of the corpus with manual transcriptions from a new operational day; secondly we have added new linguistic annotations to the whole corpus (either manually or through an automatic processing) in order to perform various linguistic tasks from syntactic and semantic parsing to dialog act tagging and dialog summarization.
We present the first corpus of texts annotated with two alternative approaches to discourse structure, Rhetorical Structure Theory (Mann and Thompson, 1988) and Segmented Discourse Representation Theory (Asher and Lascarides, 2003). 112 short argumentative texts have been analyzed according to these two theories. Furthermore, in previous work, the same texts have already been annotated for their argumentation structure, according to the scheme of Peldszus and Stede (2013). This corpus therefore enables studies of correlations between the two accounts of discourse structure, and between discourse and argumentation. We converted the three annotation formats to a common dependency tree format that enables to compare the structures, and we describe some initial findings.
We propose a scheme for annotating direct speech in literary texts, based on the Text Encoding Initiative (TEI) and the coreference annotation guidelines from the Message Understanding Conference (MUC). The scheme encodes the speakers and listeners of utterances in a text, as well as the quotative verbs that reports the utterances. We measure inter-annotator agreement on this annotation task. We then present statistics on a manually annotated corpus that consists of books from the New Testament. Finally, we visualize the corpus as a conversational network.
This paper presents a method for the normalization of historical texts using a combination of weighted finite-state transducers and language models. We have extended our previous work on the normalization of dialectal texts and tested the method against a 17th century literary work in Basque. This preprocessed corpus is made available in the LREC repository. The performance of this method for learning relations between historical and contemporary word forms is evaluated against resources in three languages. The method we present learns to map phonological changes using a noisy channel model. The model is based on techniques commonly used for phonological inference and producing Grapheme-to-Grapheme conversion systems encoded as weighted transducers and produces F-scores above 80% in the task for Basque. A wider evaluation shows that the approach performs equally well with all the languages in our evaluation suite: Basque, Spanish and Slovene. A comparison against other methods that address the same task is also provided.
In this paper, we present Farasa (meaning insight in Arabic), which is a fast and accurate Arabic segmenter. Segmentation involves breaking Arabic words into their constituent clitics. Our approach is based on SVMrank using linear kernels. The features that we utilized account for: likelihood of stems, prefixes, suffixes, and their combination; presence in lexicons containing valid stems and named entities; and underlying stem templates. Farasa outperforms or equalizes state-of-the-art Arabic segmenters, namely QATARA and MADAMIRA. Meanwhile, Farasa is nearly one order of magnitude faster than QATARA and two orders of magnitude faster than MADAMIRA. The segmenter should be able to process one billion words in less than 5 hours. Farasa is written entirely in native Java, with no external dependencies, and is open-source.
This paper discusses the internal structure of complex Esperanto words (CWs). Using a morphological analyzer, possible affixation and compounding is checked for over 50,000 Esperanto lexemes against a list of 17,000 root words. Morpheme boundaries in the resulting analyses were then checked manually, creating a CW dictionary of 28,000 words, representing 56.4% of the lexicon, or 19.4% of corpus tokens. The error percentage of the EspGram morphological analyzer for new corpus CWs was 4.3% for types and 6.4% for tokens, with a recall of almost 100%, and wrong/spurious boundaries being more common than missing ones. For pedagogical purposes a morpheme frequency dictionary was constructed for a 16 million word corpus, confirming the importance of agglutinative derivational morphemes in the Esperanto lexicon. Finally, as a means to reduce the morphological ambiguity of CWs, we provide POS likelihoods for Esperanto suffixes.
Vietnamese word segmentation (VWS) is a challenging basic issue for natural language processing. This paper addresses the problem of how does dictionary size influence VWS performance, proposes two novel measures: square overlap ratio (SOR) and relaxed square overlap ratio (RSOR), and validates their effectiveness. The SOR measure is the product of dictionary overlap ratio and corpus overlap ratio, and the RSOR measure is the relaxed version of SOR measure under an unsupervised condition. The two measures both indicate the suitable degree between segmentation dictionary and object corpus waiting for segmentation. The experimental results show that the more suitable, neither smaller nor larger, dictionary size is better to achieve the state-of-the-art performance for dictionary-based Vietnamese word segmenters.
Démonette is a derivational morphological network designed for the description of French. Its original architecture enables its use as a formal framework for the description of morphological analyses and as a repository for existing lexicons. It is fed with a variety of resources, which all are already validated. The harmonization of their content into a unified format provides them a second life, in which they are enriched with new properties, provided these are deductible from their contents. Démonette is released under a Creative Commons license. It is usable for theoretical and descriptive research in morphology, as a source of experimental material for psycholinguistics, natural language processing (NLP) and information retrieval (IR), where it fills a gap, since French lacks a large-coverage derivational resources database. The article presents the integration of two existing lexicons into Démonette. The first is Verbaction, a lexicon of deverbal action nouns. The second is Lexeur, a database of agent nouns in -eur derived from verbs or from nouns.
The paper introduces a “train once, use many” approach for the syntactic analysis of phrasal compounds (PC) of the type XP+N like “Would you like to sit on my knee?” nonsense. PCs are a challenge for NLP tools since they require the identification of a syntactic phrase within a morphological complex. We propose a method which uses a state-of-the-art dependency parser not only to analyse sentences (the environment of PCs) but also to compound the non-head of PCs in a well-defined particular condition which is the analysis of the non-head spanning from the left boundary (mostly marked by a determiner) to the nominal head of the PC. This method contains the following steps: (a) the use an English state-of-the-art dependency parser with data comprising sentences with PCs from the British National Corpus (BNC), (b) the detection of parsing errors of PCs, (c) the separate treatment of the non-head structure using the same model, and (d) the attachment of the non-head to the compound head. The evaluation of the method showed that the accuracy of 76% could be improved by adding a step in the PC compounder module which specified user-defined contexts being sensitive to the part of speech of the non-head parts and by using TreeTagger, in line with our approach.
Dialectal Arabic (DA) poses serious challenges for Natural Language Processing (NLP). The number and sophistication of tools and datasets in DA are very limited in comparison to Modern Standard Arabic (MSA) and other languages. MSA tools do not effectively model DA which makes the direct use of MSA NLP tools for handling dialects impractical. This is particularly a challenge for the creation of tools to support learning Arabic as a living language on the web, where authentic material can be found in both MSA and DA. In this paper, we present the Dialectal Arabic Linguistic Learning Assistant (DALILA), a Chrome extension that utilizes cutting-edge Arabic dialect NLP research to assist learners and non-native speakers in understanding text written in either MSA or DA. DALILA provides dialectal word analysis and English gloss corresponding to each word.
The CELEX database is one of the standard lexical resources for German. It yields a wealth of data especially for phonological and morphological applications. The morphological part comprises deep-structure morphological analyses of German. However, as it was developed in the Nineties, both encoding and spelling are outdated. About one fifth of over 50,000 datasets contain umlauts and signs such as ß. Changes to a modern version cannot be obtained by simple substitution. In this paper, we shortly describe the original content and form of the orthographic and morphological database for German in CELEX. Then we present our work on modernizing the linguistic data. Lemmas and morphological analyses are transferred to a modern standard of encoding by first merging orthographic and morphological information of the lemmas and their entries and then performing a second substitution for the morphs within their morphological analyses. Changes to modern German spelling are performed by substitution rules according to orthographical standards. We show an example of the use of the data for the disambiguation of morphological structures. The discussion describes prospects of future work on this or similar lexicons. The Perl script is publicly available on our website.
We propose an automatic approach towards determining the relative location of adjectives on a common scale based on their strength. We focus on adjectives expressing different degrees of goodness occurring in French product (perfumes) reviews. Using morphosyntactic patterns, we extract from the reviews short phrases consisting of a noun that encodes a particular aspect of the perfume and an adjective modifying that noun. We then associate each such n-gram with the corresponding product aspect and its related star rating. Next, based on the star scores, we generate adjective scales reflecting the relative strength of specific adjectives associated with a shared attribute of the product. An automatic ordering of the adjectives “correct” (correct), “sympa” (nice), “bon” (good) and “excellent” (excellent) according to their score in our resource is consistent with an intuitive scale based on human judgments. Our long-term objective is to generate different adjective scales in an empirical manner, which could allow the enrichment of lexical resources.
The automatic analysis of texts containing opinions of users about, e.g., products or political views has gained attention within the last decades. However, previous work on the task of analyzing user reviews about mobile applications in app stores is limited. Publicly available corpora do not exist, such that a comparison of different methods and models is difficult. We fill this gap by contributing the Sentiment Corpus of App Reviews (SCARE), which contains fine-grained annotations of application aspects, subjective (evaluative) phrases and relations between both. This corpus consists of 1,760 annotated application reviews from the Google Play Store with 2,487 aspects and 3,959 subjective phrases. We describe the process and methodology how the corpus was created. The Fleiss Kappa between four annotators reveals an agreement of 0.72. We provide a strong baseline with a linear-chain conditional random field and word-embedding features with a performance of 0.62 for aspect detection and 0.63 for the extraction of subjective phrases. The corpus is available to the research community to support the development of sentiment analysis methods on mobile application reviews.
Aspect Based Sentiment Analysis (ABSA) is the task of mining and summarizing opinions from text about specific entities and their aspects. This article describes two datasets for the development and testing of ABSA systems for French which comprise user reviews annotated with relevant entities, aspects and polarity values. The first dataset contains 457 restaurant reviews (2365 sentences) for training and testing ABSA systems, while the second contains 162 museum reviews (655 sentences) dedicated to out-of-domain evaluation. Both datasets were built as part of SemEval-2016 Task 5 “Aspect-Based Sentiment Analysis” where seven different languages were represented, and are publicly available for research purposes.
In this article we describe our method of automatically expanding an existing lexicon of words with affective valence scores. The automatic expansion process was done in English. In addition, we describe our procedure for automatically creating lexicons in languages where such resources may not previously exist. The foreign languages we discuss in this paper are Spanish, Russian and Farsi. We also describe the procedures to systematically validate our newly created resources. The main contributions of this work are: 1) A general method for expansion and creation of lexicons with scores of words on psychological constructs such as valence, arousal or dominance; and 2) a procedure for ensuring validity of the newly constructed resources.
In this paper, we introduce a novel comprehensive dataset of 7,992 German tweets, which were manually annotated by two human experts with fine-grained opinion relations. A rich annotation scheme used for this corpus includes such sentiment-relevant elements as opinion spans, their respective sources and targets, emotionally laden terms with their possible contextual negations and modifiers. Various inter-annotator agreement studies, which were carried out at different stages of work on these data (at the initial training phase, upon an adjudication step, and after the final annotation run), reveal that labeling evaluative judgements in microblogs is an inherently difficult task even for professional coders. These difficulties, however, can be alleviated by letting the annotators revise each other’s decisions. Once rechecked, the experts can proceed with the annotation of further messages, staying at a fairly high level of agreement.
This paper discusses the challenges in carrying out fair comparative evaluations of sentiment analysis systems. Firstly, these are due to differences in corpus annotation guidelines and sentiment class distribution. Secondly, different systems often make different assumptions about how to interpret certain statements, e.g. tweets with URLs. In order to study the impact of these on evaluation results, this paper focuses on tweet sentiment analysis in particular. One existing and two newly created corpora are used, and the performance of four different sentiment analysis systems is reported; we make our annotated datasets and sentiment analysis applications publicly available. We see considerable variations in results across the different corpora, which calls into question the validity of many existing annotated datasets and evaluations, and we make some observations about both the systems and the datasets as a result.
This paper describes EmoTweet-28, a carefully curated corpus of 15,553 tweets annotated with 28 emotion categories for the purpose of training and evaluating machine learning models for emotion classification. EmoTweet-28 is, to date, the largest tweet corpus annotated with fine-grained emotion categories. The corpus contains annotations for four facets of emotion: valence, arousal, emotion category and emotion cues. We first used small-scale content analysis to inductively identify a set of emotion categories that characterize the emotions expressed in microblog text. We then expanded the size of the corpus using crowdsourcing. The corpus encompasses a variety of examples including explicit and implicit expressions of emotions as well as tweets containing multiple emotions. EmoTweet-28 represents an important resource to advance the development and evaluation of more emotion-sensitive systems.
Sentiment composition is the determining of sentiment of a multi-word linguistic unit, such as a phrase or a sentence, based on its constituents. We focus on sentiment composition in phrases formed by at least one positive and at least one negative word ― phrases like ‘happy accident’ and ‘best winter break’. We refer to such phrases as opposing polarity phrases. We manually annotate a collection of opposing polarity phrases and their constituent single words with real-valued sentiment intensity scores using a method known as Best―Worst Scaling. We show that the obtained annotations are consistent. We explore the entries in the lexicon for linguistic regularities that govern sentiment composition in opposing polarity phrases. Finally, we list the current and possible future applications of the lexicon.
Emotions are an important part of the human experience. They are responsible for the adaptation and integration in the environment, offering, most of the time together with the cognitive system, the appropriate responses to stimuli in the environment. As such, they are an important component in decision-making processes. In today’s society, the avalanche of stimuli present in the environment (physical or virtual) makes people more prone to respond to stronger affective stimuli (i.e., those that are related to their basic needs and motivations ― survival, food, shelter, etc.). In media reporting, this is translated in the use of arguments (factual data) that are known to trigger specific (strong, affective) behavioural reactions from the readers. This paper describes initial efforts to detect such arguments from text, based on the properties of concepts. The final system able to retrieve and label this type of data from the news in traditional and social platforms is intended to be integrated Europe Media Monitor family of applications to detect texts that trigger certain (especially negative) reactions from the public, with consequences on citizen safety and security.
The paper describes the new Russian sentiment lexicon - RuSentiLex. The lexicon was gathered from several sources: opinionated words from domain-oriented Russian sentiment vocabularies, slang and curse words extracted from Twitter, objective words with positive or negative connotations from a news collection. The words in the lexicon having different sentiment orientations in specific senses are linked to appropriate concepts of the thesaurus of Russian language RuThes. All lexicon entries are classified according to four sentiment categories and three sources of sentiment (opinion, emotion, or fact). The lexicon can serve as the first version for the construction of domain-specific sentiment lexicons or can be used for feature generation in machine-learning approaches. In this role, the RuSentiLex lexicon was utilized by the participants of the SentiRuEval-2016 Twitter reputation monitoring shared task and allowed them to achieve high results.
This paper presents a framework and methodology for the annotation of perspectives in text. In the last decade, different aspects of linguistic encoding of perspectives have been targeted as separated phenomena through different annotation initiatives. We propose an annotation scheme that integrates these different phenomena. We use a multilayered annotation approach, splitting the annotation of different aspects of perspectives into small subsequent subtasks in order to reduce the complexity of the task and to better monitor interactions between layers. Currently, we have included four layers of perspective annotation: events, attribution, factuality and opinion. The annotations are integrated in a formal model called GRaSP, which provides the means to represent instances (e.g. events, entities) and propositions in the (real or assumed) world in relation to their mentions in text. Then, the relation between the source and target of a perspective is characterized by means of perspective annotations. This enables us to place alternative perspectives on the same entity, event or proposition next to each other.
With the explosive growth of online social media (forums, blogs, and social networks), exploitation of these new information sources has become essential. Our work is based on the sud4science project. The goal of this project is to perform multidisciplinary work on a corpus of authentic SMS, in French, collected in 2011 and anonymised (88milSMS corpus: http://88milsms.huma-num.fr). This paper highlights a new method to integrate opinion detection knowledge from an SMS corpus by combining lexical and semantic information. More precisely, our approach gives more weight to words with a sentiment (i.e. presence of words in a dedicated dictionary) for a classification task based on three classes: positive, negative, and neutral. The experiments were conducted on two corpora: an elongated SMS corpus (i.e. repetitions of characters in messages) and a non-elongated SMS corpus. We noted that non-elongated SMS were much better classified than elongated SMS. Overall, this study highlighted that the integration of semantic knowledge always improves classification.
This paper presents some experiments for specialising Paragraph Vectors, a new technique for creating text fragment (phrase, sentence, paragraph, text, ...) embedding vectors, for text polarity detection. The first extension regards the injection of polarity information extracted from a polarity lexicon into embeddings and the second extension aimed at inserting word order information into Paragraph Vectors. These two extensions, when training a logistic-regression classifier on the combined embeddings, were able to produce a relevant gain in performance when compared to the standard Paragraph Vector methods proposed by Le and Mikolov (2014).
In this article, we propose to evaluate the lexical similarity information provided by word representations against several opinion resources using traditional Information Retrieval tools. Word representation have been used to build and to extend opinion resources such as lexicon, and ontology and their performance have been evaluated on sentiment analysis tasks. We question this method by measuring the correlation between the sentiment proximity provided by opinion resources and the semantic similarity provided by word representations using different correlation coefficients. We also compare the neighbors found in word representations and list of similar opinion words. Our results show that the proximity of words in state-of-the-art word representations is not very effective to build sentiment similarity.
Vector space models and distributional information are widely used in NLP. The models typically rely on complex, high-dimensional objects. We present an interactive visualisation tool to explore salient lexical-semantic features of high-dimensional word objects and word similarities. Most visualisation tools provide only one low-dimensional map of the underlying data, so they are not capable of retaining the local and the global structure. We overcome this limitation by providing an additional trust-view to obtain a more realistic picture of the actual object distances. Additional tool options include the reference to a gold standard classification, the reference to a cluster analysis as well as listing the most salient (common) features for a selected subset of the words.
This paper introduces a ruled-based method and software tool, called SemAligner, for aligning chunks across texts in a given pair of short English texts. The tool, based on the top performing method at the Interpretable Short Text Similarity shared task at SemEval 2015, where it was used with human annotated (gold) chunks, can now additionally process plain text-pairs using two powerful chunkers we developed, e.g. using Conditional Random Fields. Besides aligning chunks, the tool automatically assigns semantic relations to the aligned chunks (such as EQUI for equivalent and OPPO for opposite) and semantic similarity scores that measure the strength of the semantic relation between the aligned chunks. Experiments show that SemAligner performs competitively for system generated chunks and that these results are also comparable to results obtained on gold chunks. SemAligner has other capabilities such as handling various input formats and chunkers as well as extending lookup resources.
We present an experimental study making use of a machine learning approach to identify the factors that affect the aspectual value that characterizes verbs under each of their readings. The study is based on various morpho-syntactic and semantic features collected from a French lexical resource and on a gold standard aspectual classification of verb readings designed by an expert. Our results support the tested hypothesis, namely that agentivity and abstractness influence lexical aspect.
This paper presents mwetoolkit+sem: an extension of the mwetoolkit that estimates semantic compositionality scores for multiword expressions (MWEs) based on word embeddings. First, we describe our implementation of vector-space operations working on distributional vectors. The compositionality score is based on the cosine distance between the MWE vector and the composition of the vectors of its member words. Our generic system can handle several types of word embeddings and MWE lists, and may combine individual word representations using several composition techniques. We evaluate our implementation on a dataset of 1042 English noun compounds, comparing different configurations of the underlying word embeddings and word-composition models. We show that our vector-based scores model non-compositionality better than standard association measures such as log-likelihood.
Although meaning is at the core of human cognition, state-of-the-art distributional semantic models (DSMs) are often agnostic to the findings in the area of semantic cognition. In this work, we present a novel type of DSMs motivated by the dual-processing cognitive perspective that is triggered by lexico-semantic activations in the short-term human memory. The proposed model is shown to perform better than state-of-the-art models for computing semantic similarity between words. The fusion of different types of DSMs is also investigated achieving results that are comparable or better than the state-of-the-art. The used corpora along with a set of tools, as well as large repositories of vectorial word representations are made publicly available for four languages (English, German, Italian, and Greek).
This paper describes our independent effort for extending the monolingual semantic textual similarity (STS) task setting to multiple cross-lingual settings involving English, Japanese, and Chinese. So far, we have adopted a “monolingual similarity after translation” strategy to predict the semantic similarity between a pair of sentences in different languages. With this strategy, a monolingual similarity method is applied after having (one of) the target sentences translated into a pivot language. Therefore, this paper specifically details the required and developed resources to implement this framework, while presenting our current results for English-Japanese-Chinese cross-lingual STS tasks that may exemplify the validity of the framework.
We describe resources aimed at increasing the usability of the semantic representations utilized within the DELPH-IN (Deep Linguistic Processing with HPSG) consortium. We concentrate in particular on the Dependency Minimal Recursion Semantics (DMRS) formalism, a graph-based representation designed for compositional semantic representation with deep grammars. Our main focus is on English, and specifically English Resource Semantics (ERS) as used in the English Resource Grammar. We first give an introduction to ERS and DMRS and a brief overview of some existing resources and then describe in detail a new repository which has been developed to simplify the use of ERS/DMRS. We explain a number of operations on DMRS graphs which our repository supports, with sketches of the algorithms, and illustrate how these operations can be exploited in application building. We believe that this work will aid researchers to exploit the rich and effective but complex DELPH-IN resources.
How-knowledge is indispensable in daily life, but has relatively less quantity and poorer quality than what-knowledge in publicly available knowledge bases. This paper first extracts task-subtask pairs from wikiHow, then mines linguistic patterns from search query logs, and finally applies the mined patterns to extract subtasks to complete given how-to tasks. To evaluate the proposed methodology, we group tasks and the corresponding recommended subtasks into pairs, and evaluate the results automatically and manually. The automatic evaluation shows the accuracy of 0.4494. We also classify the mined patterns based on prepositions and find that the prepositions like “on”, “to”, and “with” have the better performance. The results can be used to accelerate how-knowledge base construction.
In this paper we present a novel application of compositional distributional semantic models (CDSMs): prediction of lexical typology. The paper introduces the notion of typological closeness, which is a novel rigorous formalization of semantic similarity based on comparison of multilingual data. Starting from the Moscow Database of Qualitative Features for adjective typology, we create four datasets of typological closeness, on which we test a range of distributional semantic models. We show that, on the one hand, vector representations of phrases based on data from one language can be used to predict how words within the phrase translate into different languages, and, on the other hand, that typological data can serve as a semantic benchmark for distributional models. We find that compositional distributional models, especially parametric ones, perform way above non-compositional alternatives on the task.
The problem of understanding the stream of messages exchanged on social media such as Facebook and Twitter is becoming a major challenge for automated systems. The tremendous amount of data exchanged on these platforms as well as the specific form of language adopted by social media users constitute a new challenging context for existing argument mining techniques. In this paper, we describe a resource of natural language arguments called DART (Dataset of Arguments and their Relations on Twitter) where the complete argument mining pipeline over Twitter messages is considered: (i) we identify which tweets can be considered as arguments and which cannot, and (ii) we identify what is the relation, i.e., support or attack, linking such tweets to each other.
This paper introduces Port4NooJ v3.0, the latest version of the Portuguese module for NooJ, highlights its main features, and details its three main new components: (i) a lexicon-grammar based dictionary of 5,177 human intransitive adjectives, and a set of local grammars that use the distributional properties of those adjectives for paraphrasing (ii) a polarity dictionary with 9,031 entries for sentiment analysis, and (iii) a set of priority dictionaries and local grammars for named entity recognition. These new components were derived and/or adapted from publicly available resources. The Port4NooJ v3.0 resource is innovative in terms of the specificity of the linguistic knowledge it incorporates. The dictionary is bilingual Portuguese-English, and the semantico-syntactic information assigned to each entry validates the linguistic relation between the terms in both languages. These characteristics, which cannot be found in any other public resource for Portuguese, make it a valuable resource for translation and paraphrasing. The paper presents the current statistics and describes the different complementary and synergic components and integration efforts.
This paper describes corpora collection activity for building large machine translation systems for Latvian e-Government platform. We describe requirements for corpora, selection and assessment of data sources, collection of the public corpora and creation of new corpora from miscellaneous sources. Methodology, tools and assessment methods are also presented along with the results achieved, challenges faced and conclusions made. Several approaches to address the data scarceness are discussed. We summarize the volume of obtained corpora and provide quality metrics of MT systems trained on this data. Resulting MT systems for English-Latvian, Latvian English and Latvian Russian are integrated in the Latvian e-service portal and are freely available on website HUGO.LV. This paper can serve as a guidance for similar activities initiated in other countries, particularly in the context of European Language Resource Coordination action.
The Nederlab project aims to bring together all digitized texts relevant to the Dutch national heritage, the history of the Dutch language and culture (circa 800 – present) in one user friendly and tool enriched open access web interface. This paper describes Nederlab halfway through the project period and discusses the collections incorporated, back-office processes, system back-end as well as the Nederlab Research Portal end-user web application.
In most of the research studies on Author Profiling, large quantities of correctly labeled data are used to train the models. However, this does not reflect the reality in forensic scenarios: in practical linguistic forensic investigations, the resources that are available to profile the author of a text are usually scarce. To pay tribute to this fact, we implemented a Semi-Supervised Learning variant of the k nearest neighbors algorithm that uses small sets of labeled data and a larger amount of unlabeled data to classify the authors of texts by gender (man vs woman). We describe the enriched KNN algorithm and show that the use of unlabeled instances improves the accuracy of our gender identification model. We also present a feature set that facilitates the use of a very small number of instances, reaching accuracies higher than 70% with only 113 instances to train the model. It is also shown that the algorithm also performs well using publicly available data.
Classifying research grants into useful categories is a vital task for a funding body to give structure to the portfolio for analysis, informing strategic planning and decision-making. Automating this classification process would save time and effort, providing the accuracy of the classifications is maintained. We employ five classification models to classify a set of BBSRC-funded research grants in 21 research topics based on unigrams, technical terms and Latent Dirichlet Allocation models. To boost precision, we investigate methods for combining their predictions into five aggregate classifiers. Evaluation confirmed that ensemble classification models lead to higher precision.It was observed that there is not a single best-performing aggregate method for all research topics. Instead, the best-performing method for a research topic depends on the number of positive training instances available for this topic. Subject matter experts considered the predictions of aggregate models to correct erroneous or incomplete manual assignments.
In this work, we introduced a corpus for categorizing edit types in Wikipedia. This fine-grained taxonomy of edit types enables us to differentiate editing actions and find editor roles in Wikipedia based on their low-level edit types. To do this, we first created an annotated corpus based on 1,996 edits obtained from 953 article revisions and built machine-learning models to automatically identify the edit categories associated with edits. Building on this automated measurement of edit types, we then applied a graphical model analogous to Latent Dirichlet Allocation to uncover the latent roles in editors’ edit histories. Applying this technique revealed eight different roles editors play, such as Social Networker, Substantive Expert, etc.
We present new language resources for Moroccan and Sanaani Yemeni Arabic. The resources include corpora for each dialect which have been morphologically annotated, and morphological analyzers for each dialect which are derived from these corpora. These are the first sets of resources for Moroccan and Yemeni Arabic. The resources will be made available to the public.
The paper deals with merging two complementary resources of morphological data previously existing for Czech, namely the inflectional dictionary MorfFlex CZ and the recently developed lexical network DeriNet. The MorfFlex CZ dictionary has been used by a morphological analyzer capable of analyzing/generating several million Czech word forms according to the rules of Czech inflection. The DeriNet network contains several hundred thousand Czech lemmas interconnected with links corresponding to derivational relations (relations between base words and words derived from them). After summarizing basic characteristics of both resources, the process of merging is described, focusing on both rather technical aspects (growth of the data, measuring the quality of newly added derivational relations) and linguistic issues (treating lexical homonymy and vowel/consonant alternations). The resulting resource contains 970 thousand lemmas connected with 715 thousand derivational relations and is publicly available on the web under the CC-BY-NC-SA license. The data were incorporated in the MorphoDiTa library version 2.0 (which provides morphological analysis, generation, tagging and lemmatization for Czech) and can be browsed and searched by two web tools (DeriNet Viewer and DeriNet Search tool).
The goal of a Hungarian research project has been to create an integrated Hungarian natural language processing framework. This infrastructure includes tools for analyzing Hungarian texts, integrated into a standardized environment. The morphological analyzer is one of the core components of the framework. The goal of this paper is to describe a fast and customizable morphological analyzer and its development framework, which synthesizes and further enriches the morphological knowledge implemented in previous tools existing for Hungarian. In addition, we present the method we applied to add semantic knowledge to the lexical database of the morphology. The method utilizes neural word embedding models and morphological and shallow syntactic knowledge.
The Mixer series of speech corpora were collected over several years, principally to support annual NIST evaluations of speaker recognition (SR) technologies. These evaluations focused on conversational speech over a variety of channels and recording conditions. One of the series, Mixer-6, added a new condition, read speech, to support basic scientific research on speaker characteristics, as well as technology evaluation. With read speech it is possible to make relatively precise measurements of phonetic events and features, which can be correlated with the performance of speaker recognition algorithms, or directly used in phonetic analysis of speaker variability. The read speech, as originally recorded, was adequate for large-scale evaluations (e.g., fixed-text speaker ID algorithms) but only marginally suitable for acoustic-phonetic studies. Numerous errors due largely to speaker behavior remained in the corpus, with no record of their locations or rate of occurrence. We undertook the effort to correct this situation with automatic methods supplemented by human listening and annotation. The present paper describes the tools and methods, resulting corrections, and some examples of the kinds of research studies enabled by these enhancements.
In this paper, we describe a new database with audio recordings of non-native (L2) speakers of English, and the perceptual evaluation experiment conducted with native English speakers for assessing the prosody of each recording. These annotations are then used to compute the gold standard using different methods, and a series of regression experiments is conducted to evaluate their impact on the performance of a regression model predicting the degree of naturalness of L2 speech. Further, we compare the relevance of different feature groups modelling prosody in general (without speech tempo), speech rate and pauses modelling speech tempo (fluency), voice quality, and a variety of spectral features. We also discuss the impact of various fusion strategies on performance.Overall, our results demonstrate that the prosody of non-native speakers of English as L2 can be reliably assessed using supra-segmental audio features; prosodic features seem to be the most important ones.
The IFCASL corpus is a French-German bilingual phonetic learner corpus designed, recorded and annotated in a project on individualized feedback in computer-assisted spoken language learning. The motivation for setting up this corpus was that there is no phonetically annotated and segmented corpus for this language pair of comparable of size and coverage. In contrast to most learner corpora, the IFCASL corpus incorporate data for a language pair in both directions, i.e. in our case French learners of German, and German learners of French. In addition, the corpus is complemented by two sub-corpora of native speech by the same speakers. The corpus provides spoken data by about 100 speakers with comparable productions, annotated and segmented on the word and the phone level, with more than 50% manually corrected data. The paper reports on inter-annotator agreement and the optimization of the acoustic models for forced speech-text alignment in exercises for computer-assisted pronunciation training. Example studies based on the corpus data with a phonetic focus include topics such as the realization of /h/ and glottal stop, final devoicing of obstruents, vowel quantity and quality, pitch range, and tempo.
In this paper, we investigate some language acquisition facets of an auto-adaptative system that can automatically acquire most of the relevant lexical knowledge and authoring practices for an application in a given domain. This is the LELIO project: producing customized LELIE solutions. Our goal, within the framework of LELIE (a system that tags language uses that do not follow the Constrained Natural Language principles), is to automate the long, costly and error prone lexical customization of LELIE to a given application domain. Technical texts being relatively restricted in terms of syntax and lexicon, results obtained show that this approach is feasible and relatively reliable. By auto-adaptative, we mean that the system learns from a sample of the application corpus the various lexical terms and uses crucial for LELIE to work properly (e.g. verb uses, fuzzy terms, business terms, stylistic patterns). A technical writer validation method is developed at each step of the acquisition.
The fact that Japanese employs scriptio continua, or a writing system without spaces, complicates the first step of an NLP pipeline. Word segmentation is widely used in Japanese language processing, and lexical knowledge is crucial for reliable identification of words in text. Although external lexical resources like Wikipedia are potentially useful, segmentation mismatch prevents them from being straightforwardly incorporated into the word segmentation task. If we intentionally violate segmentation standards with the direct incorporation, quantitative evaluation will be no longer feasible. To address this problem, we propose to define a separate task that directly links given texts to an external resource, that is, wikification in the case of Wikipedia. By doing so, we can circumvent segmentation mismatch that may not necessarily be important for downstream applications. As the first step to realize the idea, we design the task of Japanese wikification and construct wikification corpora. We annotated subsets of the Balanced Corpus of Contemporary Written Japanese plus Twitter short messages. We also implement a simple wikifier and investigate its performance on these corpora.
This article presents Walenty - a new valence dictionary of Polish predicates, concentrating on its creation process and access via Internet browser. The dictionary contains two layers, syntactic and semantic. The syntactic layer describes syntactic and morphosyntactic constraints predicates put on their dependants. The semantic layer shows how predicates and their arguments are involved in a situation described in an utterance. These two layers are connected, representing how semantic arguments can be realised on the surface. Walenty also contains a powerful phraseological (idiomatic) component. Walenty has been created and can be accessed remotely with a dedicated tool called Slowal. In this article, we focus on most important functionalities of this system. First, we will depict how to access the dictionary and how built-in filtering system (covering both syntactic and semantic phenomena) works. Later, we will describe the process of creating dictionary by Slowal tool that both supports and controls the work of lexicographers.
CEPLEXicon (version 1.1) is a child lexicon resulting from the automatic tagging of two child corpora: the corpus Santos (Santos, 2006; Santos et al. 2014) and the corpus Child ― Adult Interaction (Freitas et al. 2012), which integrates information from the corpus Freitas (Freitas, 1997). This lexicon includes spontaneous speech produced by seven children (1;02.00 to 3;11.12) during approximately 86h of child-adult interaction. The automatic tagging comprised the lemmatization and morphosyntactic classification of the speech produced by the seven children included in the two child corpora; the lexicon contains information pertaining to lemmas and syntactic categories as well as absolute number of occurrences and frequencies in three age intervals: < 2 years; ≥ 2 years and < 3 years; ≥ 3 years. The information included in this lexicon and the format in which it is presented enables research in different areas and allows researchers to obtain measures of lexical growth. CEPLEXicon is available through the ELRA catalogue.
Language models are used in applications as diverse as speech recognition, optical character recognition and information retrieval. They are used to predict word appearance, and to weight the importance of words in these applications. One basic element of language models is the list of words in a language. Another is the unigram frequency of each word. But this basic information is not available for most languages in the world. Since the multilingual Wikipedia project encourages the production of encyclopedic-like articles in many world languages, we can find there an ever-growing source of text from which to extract these two language modelling elements: word list and frequency. Here we present a simple technique for converting this Wikipedia text into lexicons of weighted unigrams for the more than 280 languages present currently present in Wikipedia. The lexicons produced, and the source code for producing them in a Linux-based system are here made available for free on the Web.
GLAWI is a free, large-scale and versatile Machine-Readable Dictionary (MRD) that has been extracted from the French language edition of Wiktionary, called Wiktionnaire. In (Sajous and Hathout, 2015), we introduced GLAWI, gave the rationale behind the creation of this lexicographic resource and described the extraction process, focusing on the conversion and standardization of the heterogeneous data provided by this collaborative dictionary. In the current article, we describe the content of GLAWI and illustrate how it is structured. We also suggest various applications, ranging from linguistic studies, NLP applications to psycholinguistic experimentation. They all can take advantage of the diversity of the lexical knowledge available in GLAWI. Besides this diversity and extensive lexical coverage, GLAWI is also remarkable because it is the only free lexical resource of contemporary French that contains definitions. This unique material opens way to the renewal of MRD-based methods, notably the automated extraction and acquisition of semantic relations.
In recent years, several datasets have been released that include images and text, giving impulse to new methods that combine natural language processing and computer vision. However, there is a need for datasets of images in their natural textual context. The ION corpus contains 300K news articles published between August 2014 - 2015 in five online newspapers from two countries. The 1-year coverage over multiple publishers ensures a broad scope in terms of topics, image quality and editorial viewpoints. The corpus consists of JSON-LD files with the following data about each article: the original URL of the article on the news publisher’s website, the date of publication, the headline of the article, the URL of the image displayed with the article (if any), and the caption of that image. Neither the article text nor the images themselves are included in the corpus. Instead, the images are distributed as high-dimensional feature vectors extracted from a Convolutional Neural Network, anticipating their use in computer vision tasks. The article text is represented as a list of automatically generated entity and topic annotations in the form of Wikipedia/DBpedia pages. This facilitates the selection of subsets of the corpus for separate analysis or evaluation.
There are as many sign languages as there are deaf communities in the world. Linguists have been collecting corpora of different sign languages and annotating them extensively in order to study and understand their properties. On the other hand, the field of computer vision has approached the sign language recognition problem as a grand challenge and research efforts have intensified in the last 20 years. However, corpora collected for studying linguistic properties are often not suitable for sign language recognition as the statistical methods used in the field require large amounts of data. Recently, with the availability of inexpensive depth cameras, groups from the computer vision community have started collecting corpora with large number of repetitions for sign language recognition research. In this paper, we present the BosphorusSign Turkish Sign Language corpus, which consists of 855 sign and phrase samples from the health, finance and everyday life domains. The corpus is collected using the state-of-the-art Microsoft Kinect v2 depth sensor, and will be the first in this sign language research field. Furthermore, there will be annotations rendered by linguists so that the corpus will appeal both to the linguistic and sign language recognition research communities.
Ambient Assisted Living aims at enhancing the quality of life of older and disabled people at home thanks to Smart Homes. In particular, regarding elderly living alone at home, the detection of distress situation after a fall is very important to reassure this kind of population. However, many studies do not include tests in real settings, because data collection in this domain is very expensive and challenging and because of the few available data sets. The C IRDO corpus is a dataset recorded in realistic conditions in D OMUS , a fully equipped Smart Home with microphones and home automation sensors, in which participants performed scenarios including real falls on a carpet and calls for help. These scenarios were elaborated thanks to a field study involving elderly persons. Experiments related in a first part to distress detection in real-time using audio and speech analysis and in a second part to fall detection using video analysis are presented. Results show the difficulty of the task. The database can be used as standardized database by researchers to evaluate and compare their systems for elderly person’s assistance.
Speech data currently receives a growing attention and is an important source of information. We still lack suitable corpora of transcribed speech annotated with semantic roles that can be used for semantic role labeling (SRL), which is not the case for written data. Semantic role labeling in speech data is a challenging and complex task due to the lack of sentence boundaries and the many transcription errors such as insertion, deletion and misspellings of words. In written data, SRL evaluation is performed at the sentence level, but in speech data sentence boundaries identification is still a bottleneck which makes evaluation more complex. In this work, we semi-automatically align the predicates found in transcribed speech obtained with an automatic speech recognizer (ASR) with the predicates found in the corresponding written documents of the OntoNotes corpus and manually align the semantic roles of these predicates thus obtaining annotated semantic frames in the speech data. This data can serve as gold standard alignments for future research in semantic role labeling of speech data.
CORILSE is a computerized corpus of Spanish Sign Language (Lengua de Signos Española, LSE). It consists of a set of recordings from different discourse genres by Galician signers living in the city of Vigo. In this paper we describe its annotation system, developed on the basis of pre-existing ones (mostly the model of Auslan corpus). This includes primary annotation of id-glosses for manual signs, annotation of non-manual component, and secondary annotation of grammatical categories and relations, because this corpus is been built for grammatical analysis, in particular argument structures in LSE. Up until this moment the annotation has been basically made by hand, which is a slow and time-consuming task. The need to facilitate this process leads us to engage in the development of automatic or semi-automatic tools for manual and facial recognition. Finally, we also present the web repository that will make the corpus available to different types of users, and will allow its exploitation for research purposes and other applications (e.g. teaching of LSE or design of tasks for signed language assessment).
The OFAI Multimodal Task Description Corpus (OFAI-MMTD Corpus) is a collection of dyadic teacher-learner (human-human and human-robot) interactions. The corpus is multimodal and tracks the communication signals exchanged between interlocutors in task-oriented scenarios including speech, gaze and gestures. The focus of interest lies on the communicative signals conveyed by the teacher and which objects are salient at which time. Data are collected from four different task description setups which involve spatial utterances, navigation instructions and more complex descriptions of joint tasks.
In recent years there has been a surge of interest in the natural language prosessing related to the real world, such as symbol grounding, language generation, and nonlinguistic data search by natural language queries. In order to concentrate on language ambiguities, we propose to use a well-defined “real world,” that is game states. We built a corpus consisting of pairs of sentences and a game state. The game we focus on is shogi (Japanese chess). We collected 742,286 commentary sentences in Japanese. They are spontaneously generated contrary to natural language annotations in many image datasets provided by human workers on Amazon Mechanical Turk. We defined domain specific named entities and we segmented 2,508 sentences into words manually and annotated each word with a named entity tag. We describe a detailed definition of named entities and show some statistics of our game commentary corpus. We also show the results of the experiments of word segmentation and named entity recognition. The accuracies are as high as those on general domain texts indicating that we are ready to tackle various new problems related to the real world.
In this paper, we describe the organization and the implementation of the CAMOMILE collaborative annotation framework for multimodal, multimedia, multilingual (3M) data. Given the versatile nature of the analysis which can be performed on 3M data, the structure of the server was kept intentionally simple in order to preserve its genericity, relying on standard Web technologies. Layers of annotations, defined as data associated to a media fragment from the corpus, are stored in a database and can be managed through standard interfaces with authentication. Interfaces tailored specifically to the needed task can then be developed in an agile way, relying on simple but reliable services for the management of the centralized annotations. We then present our implementation of an active learning scenario for person annotation in video, relying on the CAMOMILE server; during a dry run experiment, the manual annotation of 716 speech segments was thus propagated to 3504 labeled tracks. The code of the CAMOMILE framework is distributed in open source.
The Frankfurt Image GestURE corpus (FIGURE) is introduced. The corpus data is collected in an experimental setting where 50 naive participants spontaneously produced gestures in response to five to six terms from a total of 27 stimulus terms. The stimulus terms have been compiled mainly from image schemata from psycholinguistics, since such schemata provide a panoply of abstract contents derived from natural language use. The gestures have been annotated for kinetic features. FIGURE aims at finding (sets of) stable kinetic feature configurations associated with the stimulus terms. Given such configurations, they can be used for designing HCI gestures that go beyond pre-defined gesture vocabularies or touchpad gestures. It is found, for instance, that movement trajectories are far more informative than handshapes, speaking against purely handshape-based HCI vocabularies. Furthermore, the mean temporal duration of hand and arm movements associated vary with the stimulus terms, indicating a dynamic dimension not covered by vocabulary-based approaches. Descriptive results are presented and related to findings from gesture studies and natural language dialogue.
The term smart home refers to a living environment that by its connected sensors and actuators is capable of providing intelligent and contextualised support to its user. This may result in automated behaviors that blends into the user’s daily life. However, currently most smart homes do not provide such intelligent support. A first step towards such intelligent capabilities lies in learning automation rules by observing the user’s behavior. We present a new type of corpus for learning such rules from user behavior as observed from the events in a smart homes sensor and actuator network. The data contains information about intended tasks by the users and synchronized events from this network. It is derived from interactions of 59 users with the smart home in order to solve five tasks. The corpus contains recordings of more than 40 different types of data streams and has been segmented and pre-processed to increase signal quality. Overall, the data shows a high noise level on specific data types that can be filtered out by a simple smoothing approach. The resulting data provides insights into event patterns resulting from task specific user behavior and thus constitutes a basis for machine learning approaches to learn automation rules.
In this paper we describe our work in building an online tool for manually annotating texts in any spoken language with SignWriting in any sign language. The existence of such tool will allow the creation of parallel corpora between spoken and sign languages that can be used to bootstrap the creation of efficient tools for the Deaf community. As an example, a parallel corpus between English and American Sign Language could be used for training Machine Learning models for automatic translation between the two languages. Clearly, this kind of tool must be designed in a way that it eases the task of human annotators, not only by being easy to use, but also by giving smart suggestions as the annotation progresses, in order to save time and effort. By building a collaborative, online, easy to use annotation tool for building parallel corpora between spoken and sign languages we aim at helping the development of proper resources for sign languages that can then be used in state-of-the-art models currently used in tools for spoken languages. There are several issues and difficulties in creating this kind of resource, and our presented tool already deals with some of them, like adequate text representation of a sign and many to many alignments between words and signs.
In South-Asian languages such as Hindi and Urdu, action verbs having compound constructions and serial verbs constructions pose serious problems for natural language processing and other linguistic tasks. Urdu is an Indo-Aryan language spoken by 51, 500, 0001 speakers in India. Action verbs that occur spontaneously in day-to-day communication are highly ambiguous in nature semantically and as a consequence cause disambiguation issues that are relevant and applicable to Language Technologies (LT) like Machine Translation (MT) and Natural Language Processing (NLP). IMAGACT4ALL is an ontology-driven web-based platform developed by the University of Florence for storing action verbs and their inter-relations. This group is currently collaborating with Jawaharlal Nehru University (JNU) in India to connect Indian languages on this platform. Action verbs are frequently used in both written and spoken discourses and refer to various meanings because of their polysemic nature. The IMAGACT4ALL platform stores each 3d animation image, each one of them referring to a variety of possible ontological types, which in turn makes the annotation task for the annotator quite challenging with regard to selecting verb argument structure having a range of probability distribution. The authors, in this paper, discuss the issues and challenges such as complex predicates (compound and conjunct verbs), ambiguously animated video illustrations, semantic discrepancies, and the factors of verb-selection preferences that have produced significant problems in annotating Urdu verbs on the IMAGACT ontology.
Ontologies are powerful to support semantic based applications and intelligent systems. While ontology learning are challenging due to its bottleneck in handcrafting structured knowledge sources and training data. To address this difficulty, many researchers turn to ontology enrichment and population using external knowledge sources such as DBpedia. In this paper, we propose a method using DBpedia in a different manner. We utilize relation instances in DBpedia to supervise the ontology learning procedure from unstructured text, rather than populate the ontology structure as a post-processing step. We construct three language resources in areas of computer science: enriched Wikipedia concept tree, domain ontology, and gold standard from NSFC taxonomy. Experiment shows that the result of ontology learning from corpus of computer science can be improved via the relation instances extracted from DBpedia in the same field. Furthermore, making distinction between the relation instances and applying a proper weighting scheme in the learning procedure lead to even better result.
We present the development of a Norwegian Academic Wordlist (AKA list) for the Norwegian Bokmäl variety. To identify specific academic vocabulary we developed a 100-million-word academic corpus based on the University of Oslo archive of digital publications. Other corpora were used for testing and developing general word lists. We tried two different methods, those of Carlund et al. (2012) and Gardner & Davies (2013), and compared them. The resulting list is presented on a web site, where the words can be inspected in different ways, and freely downloaded.
This paper presents the Event and Implied Situation Ontology (ESO), a manually constructed resource which formalizes the pre and post situations of events and the roles of the entities affected by an event. The ontology is built on top of existing resources such as WordNet, SUMO and FrameNet. The ontology is injected to the Predicate Matrix, a resource that integrates predicate and role information from amongst others FrameNet, VerbNet, PropBank, NomBank and WordNet. We illustrate how these resources are used on large document collections to detect information that otherwise would have remained implicit. The ontology is evaluated on two aspects: recall and precision based on a manually annotated corpus and secondly, on the quality of the knowledge inferred by the situation assertions in the ontology. Evaluation results on the quality of the system show that 50% of the events typed and enriched with ESO assertions are correct.
In this paper, we describe experiments on the morphosyntactic annotation of historical language varieties for the example of Middle Low German (MLG), the official language of the German Hanse during the Middle Ages and a dominant language around the Baltic Sea by the time. To our best knowledge, this is the first experiment in automatically producing morphosyntactic annotations for Middle Low German, and accordingly, no part-of-speech (POS) tagset is currently agreed upon. In our experiment, we illustrate how ontology-based specifications of projected annotations can be employed to circumvent this issue: Instead of training and evaluating against a given tagset, we decomponse it into independent features which are predicted independently by a neural network. Using consistency constraints (axioms) from an ontology, then, the predicted feature probabilities are decoded into a sound ontological representation. Using these representations, we can finally bootstrap a POS tagset capturing only morphosyntactic features which could be reliably predicted. In this way, our approach is capable to optimize precision and recall of morphosyntactic annotations simultaneously with bootstrapping a tagset rather than performing iterative cycles.
As part of a human-robot interaction project, we are interested by gestural modality as one of many ways to communicate. In order to develop a relevant gesture recognition system associated to a smart home butler robot. Our methodology is based on an IQ game-like Wizard of Oz experiment to collect spontaneous and implicitly produced gestures in an ecological context. During the experiment, the subject has to use non-verbal cues (i.e. gestures) to interact with a robot that is the referee. The subject is unaware that his gestures will be the focus of our study. In the second part of the experiment, we asked the subjects to do the gestures he had produced in the experiment, those are the explicit gestures. The implicit gestures are compared with explicitly produced ones to determine a relevant ontology. This preliminary qualitative analysis will be the base to build a big data corpus in order to optimize acceptance of the gesture dictionary in coherence with the “socio-affective glue” dynamics.
In this paper we describe our work in progress in the automatic development of a taxonomy of Spanish nouns, we offer the Perl implementation we have so far, and we discuss the different problems that still need to be addressed. We designed a statistically-based taxonomy induction algorithm consisting of a combination of different strategies not involving explicit linguistic knowledge. Being all quantitative, the strategies we present are however of different nature. Some of them are based on the computation of distributional similarity coefficients which identify pairs of sibling words or co-hyponyms, while others are based on asymmetric co-occurrence and identify pairs of parent-child words or hypernym-hyponym relations. A decision making process is then applied to combine the results of the previous steps, and finally connect lexical units to a basic structure containing the most general categories of the language. We evaluate the quality of the taxonomy both manually and also using Spanish Wordnet as a gold-standard. We estimate an average of 89.07% precision and 25.49% recall considering only the results which the algorithm presents with high degree of certainty, or 77.86% precision and 33.72% recall considering all results.
In this paper, we present a GOLD standard of part-of-speech tagged transcripts of spoken German. The GOLD standard data consists of four annotation layers ― transcription (modified orthography), normalization (standard orthography), lemmatization and POS tags ― all of which have undergone careful manual quality control. It comes with guidelines for the manual POS annotation of transcripts of German spoken data and an extended version of the STTS (Stuttgart Tübingen Tagset) which accounts for phenomena typically found in spontaneous spoken German. The GOLD standard was developed on the basis of the Research and Teaching Corpus of Spoken German, FOLK, and is, to our knowledge, the first such dataset based on a wide variety of spontaneous and authentic interaction types. It can be used as a basis for further development of language technology and corpus linguistic applications for German spoken language.
Part-of-Speech(POS) tagging is a key step in many NLP algorithms. However, tweets are difficult to POS tag because they are short, are not always written maintaining formal grammar and proper spelling, and abbreviations are often used to overcome their restricted lengths. Arabic tweets also show a further range of linguistic phenomena such as usage of different dialects, romanised Arabic and borrowing foreign words. In this paper, we present an evaluation and a detailed error analysis of state-of-the-art POS taggers for Arabic when applied to Arabic tweets. On the basis of this analysis, we combine normalisation and external knowledge to handle the domain noisiness and exploit bootstrapping to construct extra training data in order to improve POS tagging for Arabic tweets. Our results show significant improvements over the performance of a number of well-known taggers for Arabic.
This paper relates to the challenge of morphological tagging and lemmatization in morphologically rich languages by example of German and Latin. We focus on the question what a practitioner can expect when using state-of-the-art solutions out of the box. Moreover, we contrast these with old(er) methods and implementations for POS tagging. We examine to what degree recent efforts in tagger development are reflected by improved accuracies ― and at what cost, in terms of training and processing time. We also conduct in-domain vs. out-domain evaluation. Out-domain evaluations are particularly insightful because the distribution of the data which is being tagged by a user will typically differ from the distribution on which the tagger has been trained. Furthermore, two lemmatization techniques are evaluated. Finally, we compare pipeline tagging vs. a tagging approach that acknowledges dependencies between inflectional categories.
We present a morphological tagger for Latin, called TTLab Latin Tagger based on Conditional Random Fields (TLT-CRF) which uses a large Latin lexicon. Beyond Part of Speech (PoS), TLT-CRF tags eight inflectional categories of verbs, adjectives or nouns. It utilizes a statistical model based on CRFs together with a rule interpreter that addresses scenarios of sparse training data. We present results of evaluating TLT-CRF to answer the question what can be learnt following the paradigm of 1st order CRFs in conjunction with a large lexical resource and a rule interpreter. Furthermore, we investigate the contigency of representational features and targeted parts of speech to learn about selective features.
Because of the small size of Romanian corpora, the performance of a PoS tagger or a dependency parser trained with the standard supervised methods fall far short from the performance achieved in most languages. That is why, we apply state-of-the-art methods for cross-lingual transfer on Romanian tagging and parsing, from English and several Romance languages. We compare the performance with monolingual systems trained with sets of different sizes and establish that training on a few sentences in target language yields better results than transferring from large datasets in other languages.
In this paper we present a tagger developed for inflectionally rich languages for which both a training corpus and a lexicon are available. We do not constrain the tagger by the lexicon entries, allowing both for lexicon incompleteness and noisiness. By using the lexicon indirectly through features we allow for known and unknown words to be tagged in the same manner. We test our tagger on Slovene data, obtaining a 25% error reduction of the best previous results both on known and unknown words. Given that Slovene is, in comparison to some other Slavic languages, a well-resourced language, we perform experiments on the impact of token (corpus) vs. type (lexicon) supervision, obtaining useful insights in how to balance the effort of extending resources to yield better tagging results.
Treebanks are important resources for researchers in natural language processing, speech recognition, theoretical linguistics, etc. To strengthen the automatic processing of the Vietnamese language, a Vietnamese treebank has been built. However, the quality of this treebank is not satisfactory and is a possible source for the low performance of Vietnamese language processing. We have been building a new treebank for Vietnamese with about 40,000 sentences annotated with three layers: word segmentation, part-of-speech tagging, and bracketing. In this paper, we describe several challenges of Vietnamese language and how we solve them in developing annotation guidelines. We also present our methods to improve the quality of the annotation guidelines and ensure annotation accuracy and consistency. Experiment results show that inter-annotator agreement ratios and accuracy are higher than 90% which is satisfactory.
This paper provides a new method to correct annotation errors in a treebank. The previous error correction method constructs a pseudo parallel corpus where incorrect partial parse trees are paired with correct ones, and extracts error correction rules from the parallel corpus. By applying these rules to a treebank, the method corrects errors. However, this method does not achieve wide coverage of error correction. To achieve wide coverage, our method adopts a different approach. In our method, we consider that an infrequent pattern which can be transformed to a frequent one is an annotation error pattern. Based on a tree mining technique, our method seeks such infrequent tree patterns, and constructs error correction rules each of which consists of an infrequent pattern and a corresponding frequent pattern. We conducted an experiment using the Penn Treebank. We obtained 1,987 rules which are not constructed by the previous method, and the rules achieved good precision.
The question of the type of text used as primary data in treebanks is of certain importance. First, it has an influence at the discourse level: an article is not organized in the same way as a novel or a technical document. Moreover, it also has consequences in terms of semantic interpretation: some types of texts can be easier to interpret than others. We present in this paper a new type of treebank which presents the particularity to answer to specific needs of experimental linguistic. It is made of short texts (book backcovers) that presents a strong coherence in their organization and can be rapidly interpreted. This type of text is adapted to short reading sessions, making it easy to acquire physiological data (e.g. eye movement, electroencepholagraphy). Such a resource offers reliable data when looking for correlations between computational models and human language processing.
This paper presents a new linguistic resource for the study and computational processing of Portuguese. CINTIL DependencyBank PREMIUM is a corpus of Portuguese news text, accurately manually annotated with a wide range of linguistic information (morpho-syntax, named-entities, syntactic function and semantic roles), making it an invaluable resource specially for the development and evaluation of data-driven natural language processing tools. The corpus is under active development, reaching 4,000 sentences in its current version. The paper also reports on the training and evaluation of a dependency parser over this corpus. CINTIL DependencyBank PREMIUM is freely-available for research purposes through META-SHARE.
This paper presents the first version of Estonian Universal Dependencies Treebank which has been semi-automatically acquired from Estonian Dependency Treebank and comprises ca 400,000 words (ca 30,000 sentences) representing the genres of fiction, newspapers and scientific writing. Article analyses the differences between two annotation schemes and the conversion procedure to Universal Dependencies format. The conversion has been conducted by manually created Constraint Grammar transfer rules. As the rules enable to consider unbounded context, include lexical information and both flat and tree structure features at the same time, the method has proved to be reliable and flexible enough to handle most of transformations. The automatic conversion procedure achieved LAS 95.2%, UAS 96.3% and LA 98.4%. If punctuation marks were excluded from the calculations, we observed LAS 96.4%, UAS 97.7% and LA 98.2%. Still the refinement of the guidelines and methodology is needed in order to re-annotate some syntactic phenomena, e.g. inter-clausal relations. Although automatic rules usually make quite a good guess even in obscure conditions, some relations should be checked and annotated manually after the main conversion.
This paper presents the construction of an open-source dependency treebank of spoken Slovenian, the first syntactically annotated collection of spontaneous speech in Slovenian. The treebank has been manually annotated using the Universal Dependencies annotation scheme, a one-layer syntactic annotation scheme with a high degree of cross-modality, cross-framework and cross-language interoperability. In this original application of the scheme to spoken language transcripts, we address a wide spectrum of syntactic particularities in speech, either by extending the scope of application of existing universal labels or by proposing new speech-specific extensions. The initial analysis of the resulting treebank and its comparison with the written Slovenian UD treebank confirms significant syntactic differences between the two language modalities, with spoken data consisting of shorter and more elliptic sentences, less and simpler nominal phrases, and more relations marking disfluencies, interaction, deixis and modality.
This paper introduces the ALT project initiated by the Advanced Speech Translation Research and Development Promotion Center (ASTREC), NICT, Kyoto, Japan. The aim of this project is to accelerate NLP research for Asian languages such as Indonesian, Japanese, Khmer, Laos, Malay, Myanmar, Philippine, Thai and Vietnamese. The original resource for this project was English articles that were randomly selected from Wikinews. The project has so far created a corpus for Myanmar and will extend in scope to include other languages in the near future. A 20000-sentence corpus of Myanmar that has been manually translated from an English corpus has been word segmented, word aligned, part-of-speech tagged and constituency parsed by human annotators. In this paper, we present the implementation steps for creating the treebank in detail, including a description of the ALT web-based treebanking tool. Moreover, we report statistics on the annotation quality of the Myanmar treebank created so far.
This article describes the conversion of the Norwegian Dependency Treebank to the Universal Dependencies scheme. This paper details the mapping of PoS tags, morphological features and dependency relations and provides a description of the structural changes made to NDT analyses in order to make it compliant with the UD guidelines. We further present PoS tagging and dependency parsing experiments which report first results for the processing of the converted treebank. The full converted treebank was made available with the 1.2 release of the UD treebanks.
META-NET is a European network of excellence, founded in 2010, that consists of 60 research centres in 34 European countries. One of the key visions and goals of META-NET is a truly multilingual Europe, which is substantially supported and realised through language technologies. In this article we provide an overview of recent developments around the multilingual Europe topic, we also describe recent and upcoming events as well as recent and upcoming strategy papers. Furthermore, we provide overviews of two new emerging initiatives, the CEF.AT and ELRC activity on the one hand and the Cracking the Language Barrier federation on the other. The paper closes with several suggested next steps in order to address the current challenges and to open up new opportunities.
We present here the context and results of two surveys (a French one and an international one) concerning Ethics and NLP, which we designed and conducted between June and September 2015. These surveys follow other actions related to raising concern for ethics in our community, including a Journée d’études, a workshop and the Ethics and Big Data Charter. The concern for ethics shows to be quite similar in both surveys, despite a few differences which we present and discuss. The surveys also lead to think there is a growing awareness in the field concerning ethical issues, which translates into a willingness to get involved in ethics-related actions, to debate about the topic and to see ethics be included in major conferences themes. We finally discuss the limits of the surveys and the means of action we consider for the future. The raw data from the two surveys are freely available online.
An assessment of the intellectual property requirements for data used in machine-aided translation is provided based on a recent EC-funded legal review. This is compared against the capabilities offered by current linked open data standards from the W3C for publishing and sharing translation memories from translation projects, and proposals for adequately addressing the intellectual property needs of stakeholders in translation projects using open data vocabularies are suggested.
This article presents the latest dissemination activities and technical developments that were carried out for the International Standard Language Resource Number (ISLRN) service. It also recalls the main principle and submission process for providers to obtain their 13-digit ISLRN identifier. Up to March 2016, 2100 Language Resources were allocated an ISLRN number, not only ELRA’s and LDC’s catalogued Language Resources, but also the ones from other important organisations like the Joint Research Centre (JRC) and the Resource Management Agency (RMA) who expressed their strong support to this initiative.In the research field, not only assigning a unique identification number is important, but also referring to a Language Resource as an object per se (like publications) has now become an obvious requirement. The ISLRN could also become an important parameter to be considered to compute a Language Resource Impact Factor (LRIF) in order to recognize the merits of the producers of Language Resources. Integrating the ISLRN number into a LR-oriented bibliographical reference is thus part of the objective. The idea is to make use of a BibTeX entry that would take into account Language Resources items, including ISLRN.The ISLRN being a requested field within the LREC 2016 submission, we expect that several other LRs will be allocated an ISLRN number by the conference date. With this expansion, this number aims to be a spreadly-used LR citation instrument within works referring to LRs.
Since its inception in 2010, the Linguistic Data Consortium’s data scholarship program has awarded no cost grants in data to 64 recipients from 26 countries. A survey of the twelve cycles to date ― two awards each in the Fall and Spring semesters from Fall 2010 through Spring 2016 ― yields an interesting view into graduate program research trends in human language technology and related fields and the particular data sets deemed important to support that research. The survey also reveals regions in which such activity appears to be on a rise, including in Arabic-speaking regions and portions of the Americas and Asia.
The paper introduces a new annotated French data set for Sentiment Analysis, which is a currently missing resource. It focuses on the collection from Twitter of data related to the socio-political debate about the reform of the bill for wedding in France. The design of the annotation scheme is described, which extends a polarity label set by making available tags for marking target semantic areas and figurative language devices. The annotation process is presented and the disagreement discussed, in particular, in the perspective of figurative language use and in that of the semantic oriented annotation, which are open challenges for NLP systems.
In this paper we present a new corpus of Arabic tweets that mention some form of violent event, developed to support the automatic identification of Human Rights Abuse. The dataset was manually labelled for seven classes of violence using crowdsourcing.
Personality profiling is the task of detecting personality traits of authors based on writing style. Several personality typologies exist, however, the Briggs-Myer Type Indicator (MBTI) is particularly popular in the non-scientific community, and many people use it to analyse their own personality and talk about the results online. Therefore, large amounts of self-assessed data on MBTI are readily available on social-media platforms such as Twitter. We present a novel corpus of tweets annotated with the MBTI personality type and gender of their author for six Western European languages (Dutch, German, French, Italian, Portuguese and Spanish). We outline the corpus creation and annotation, show statistics of the obtained data distributions and present first baselines on Myers-Briggs personality profiling and gender prediction for all six languages.
Microblogging platforms such as Twitter provide active communication channels during mass convergence and emergency events such as earthquakes, typhoons. During the sudden onset of a crisis situation, affected people post useful information on Twitter that can be used for situational awareness and other humanitarian disaster response efforts, if processed timely and effectively. Processing social media information pose multiple challenges such as parsing noisy, brief and informal messages, learning information categories from the incoming stream of messages and classifying them into different classes among others. One of the basic necessities of many of these tasks is the availability of data, in particular human-annotated data. In this paper, we present human-annotated Twitter corpora collected during 19 different crises that took place between 2013 and 2015. To demonstrate the utility of the annotations, we train machine learning classifiers. Moreover, we publish first largest word2vec word embeddings trained on 52 million crisis-related tweets. To deal with tweets language issues, we present human-annotated normalized lexical resources for different lexical variations.
Code-Switching (CS) between two languages is extremely common in communities with societal multilingualism where speakers switch between two or more languages when interacting with each other. CS has been extensively studied in spoken language by linguists for several decades but with the popularity of social-media and less formal Computer Mediated Communication, we now see a big rise in the use of CS in the text form. This poses interesting challenges and a need for computational processing of such code-switched data. As with any Computational Linguistic analysis and Natural Language Processing tools and applications, we need annotated data for understanding, processing, and generation of code-switched language. In this study, we focus on CS between English and Hindi Tweets extracted from the Twitter stream of Hindi-English bilinguals. We present an annotation scheme for annotating the pragmatic functions of CS in Hindi-English (Hi-En) code-switched tweets based on a linguistic analysis and some initial experiments.
We present an attempt to port the international syntactic annotation scheme, Universal Dependencies, to the Japanese language in this paper. Since the Japanese syntactic structure is usually annotated on the basis of unique chunk-based dependencies, we first introduce word-based dependencies by using a word unit called the Short Unit Word, which usually corresponds to an entry in the lexicon UniDic. Porting is done by mapping the part-of-speech tagset in UniDic to the universal part-of-speech tagset, and converting a constituent-based treebank to a typed dependency tree. The conversion is not straightforward, and we discuss the problems that arose in the conversion and the current solutions. A treebank consisting of 10,000 sentences was built by converting the existent resources and currently released to the public.
Cross-linguistically consistent annotation is necessary for sound comparative evaluation and cross-lingual learning experiments. It is also useful for multilingual system development and comparative linguistic studies. Universal Dependencies is an open community effort to create cross-linguistically consistent treebank annotation for many languages within a dependency-based lexicalist framework. In this paper, we describe v1 of the universal guidelines, the underlying design principles, and the currently available treebanks for 33 languages.
The recognition of multiword expressions (MWEs) in a sentence is important for such linguistic analyses as syntactic and semantic parsing, because it is known that combining an MWE into a single token improves accuracy for various NLP tasks, such as dependency parsing and constituency parsing. However, MWEs are not annotated in Penn Treebank. Furthermore, when converting word-based dependency to MWE-aware dependency directly, one could combine nodes in an MWE into a single node. Nevertheless, this method often leads to the following problem: A node derived from an MWE could have multiple heads and the whole dependency structure including MWE might be cyclic. Therefore we converted a phrase structure to a dependency structure after establishing an MWE as a single subtree. This approach can avoid an occurrence of multiple heads and/or cycles. In this way, we constructed an English dependency corpus taking into account compound function words, which are one type of MWEs that serve as functional expressions. In addition, we report experimental results of dependency parsing using a constructed corpus.
TANL is a suite of tools for text analytics based on the software architecture paradigm of data driven pipelines. The strategies for upgrading TANL to the use of Universal Dependencies range from a minimalistic approach consisting of introducing pre/post-processing steps into the native pipeline to revising the whole pipeline. We explore the issue in the context of the Italian Treebank, considering both the efforts involved, how to avoid losing linguistically relevant information and the loss of accuracy in the process. In particular we compare different strategies for parsing and discuss the implications of simplifying the pipeline when detailed part-of-speech and morphological annotations are not available, as it is the case for less resourceful languages. The experiments are relative to the Italian linguistic pipeline, but the use of different parsers in our evaluations and the avoidance of language specific tagging make the results general enough to be useful in helping the transition to UD for other languages.
We present a dependency treebank of the Chinese Buddhist Canon, which contains 1,514 texts with about 50 million Chinese characters. The treebank was created by an automatic parser trained on a smaller treebank, containing four manually annotated sutras (Lee and Kong, 2014). We report results on word segmentation, part-of-speech tagging and dependency parsing, and discuss challenges posed by the processing of medieval Chinese. In a case study, we exploit the treebank to examine verbs frequently associated with Buddha, and to analyze usage patterns of quotative verbs in direct speech. Our results suggest that certain quotative verbs imply status differences between the speaker and the listener.
Polysemy is the capacity for a word to have multiple meanings. Polysemy detection is a first step for Word Sense Induction (WSI), which allows to find different meanings for a term. The polysemy detection is also important for information extraction (IE) systems. In addition, the polysemy detection is important for building/enriching terminologies and ontologies. In this paper, we present a novel approach to detect if a biomedical term is polysemic, with the long term goal of enriching biomedical ontologies. This approach is based on the extraction of new features. In this context we propose to extract features following two manners: (i) extracted directly from the text dataset, and (ii) from an induced graph. Our method obtains an Accuracy and F-Measure of 0.978.
We introduce Cro36WSD, a freely-available medium-sized lexical sample for Croatian word sense disambiguation (WSD).Cro36WSD comprises 36 words: 12 adjectives, 12 nouns, and 12 verbs, balanced across both frequency bands and polysemy levels. We adopt the multi-label annotation scheme in the hope of lessening the drawbacks of discrete sense inventories and obtaining more realistic annotations from human experts. Sense-annotated data is collected through multiple annotation rounds to ensure high-quality annotations: with a 115 person-hours effort we reached an inter-annotator agreement score of 0.877. We analyze the obtained data and perform a correlation analysis between several relevant variables, including word frequency, number of senses, sense distribution skewness, average annotation time, and the observed inter-annotator agreement (IAA). Using the obtained data, we compile multi- and single-labeled dataset variants using different label aggregation schemes. Finally, we evaluate three different baseline WSD models on both dataset variants and report on the insights gained. We make both dataset variants freely available.
Word Sense Disambiguation (WSD) systems tend to have a strong bias towards assigning the Most Frequent Sense (MFS), which results in high performance on the MFS but in a very low performance on the less frequent senses. We addressed the MFS bias in WSD systems by combining the output from a WSD system with a set of mostly static features to create a MFS classifier to decide when to and not to choose the MFS. The output from this MFS classifier, which is based on the Random Forest algorithm, is then used to modify the output from the original WSD system. We applied our classifier to one of the state-of-the-art supervised WSD systems, i.e. IMS, and to of the best state-of-the-art unsupervised WSD systems, i.e. UKB. Our main finding is that we are able to improve the system output in terms of choosing between the MFS and the less frequent senses. When we apply the MFS classifier to fine-grained WSD, we observe an improvement on the less frequent sense cases, whereas we maintain the overall recall.
Linking concepts and named entities to knowledge bases has become a crucial Natural Language Understanding task. In this respect, recent works have shown the key advantage of exploiting textual definitions in various Natural Language Processing applications. However, to date there are no reliable large-scale corpora of sense-annotated textual definitions available to the research community. In this paper we present a large-scale high-quality corpus of disambiguated glosses in multiple languages, comprising sense annotations of both concepts and named entities from a unified sense inventory. Our approach for the construction and disambiguation of the corpus builds upon the structure of a large multilingual semantic network and a state-of-the-art disambiguation system; first, we gather complementary information of equivalent definitions across different languages to provide context for disambiguation, and then we combine it with a semantic similarity-based refinement. As a result we obtain a multilingual corpus of textual definitions featuring over 38 million definitions in 263 languages, and we make it freely available at http://lcl.uniroma1.it/disambiguated-glosses. Experiments on Open Information Extraction and Sense Clustering show how two state-of-the-art approaches improve their performance by integrating our disambiguated corpus into their pipeline.
We propose an unsupervised system for a variant of cross-lingual lexical substitution (CLLS) to be used in a reading scenario in computer-assisted language learning (CALL), in which single-word translations provided by a dictionary are ranked according to their appropriateness in context. In contrast to most alternative systems, ours does not rely on either parallel corpora or machine translation systems, making it suitable for low-resource languages as the language to be learned. This is achieved by a graph-based scoring mechanism which can deal with ambiguous translations of context words provided by a dictionary. Due to this decoupling from the source language, we need monolingual corpus resources only for the target language, i.e. the language of the translation candidates. We evaluate our approach for the language pair Norwegian Nynorsk-English on an exploratory manually annotated gold standard and report promising results. When running our system on the original SemEval CLLS task, we rank 6th out of 18 (including 2 baselines and our 2 system variants) in the best evaluation.
The Potsdam Commentary Corpus is a collection of 175 German newspaper commentaries annotated on a variety of different layers. This paper introduces a new layer that covers the linguistic notion of information-structural topic (not to be confused with ‘topic’ as applied to documents in information retrieval). To our knowledge, this is the first larger topic-annotated resource for German (and one of the first for any language). We describe the annotation guidelines and the annotation process, and the results of an inter-annotator agreement study, which compare favourably to the related work. The annotated corpus is freely available for research.
In this paper we present the Corpus of REcommendation STrength (CREST), a collection of HTML-formatted clinical guidelines annotated with the location of recommendations. Recommendations are labelled with an author-provided indicator of their strength of importance. As data was drawn from many disparate authors, we define a unified scheme of importance labels, and provide a mapping for each guideline. We demonstrate the utility of the corpus and its annotations in some initial measurements investigating the type of language constructions associated with strong and weak recommendations, and experiments into promising features for recommendation classification, both with respect to strong and weak labels, and to all labels of the unified scheme. An error analysis indicates that, while there is a strong relationship between lexical choices and strength labels, there can be substantial variance in the choices made by different authors.
Using the Methodius Natural Language Generation (NLG) System, we have created a corpus which consists of a collection of generated texts which describe ancient Greek artefacts. Each text is linked to two representations created as part of the NLG process. The first is a content plan, which uses rhetorical relations to describe the high-level discourse structure of the text, and the second is a logical form describing the syntactic structure, which is sent to the OpenCCG surface realization module to produce the final text output. In recent work, White and Howcroft (2015) have used the SPaRKy restaurant corpus, which contains similar combination of texts and representations, for their research on the induction of rules for the combination of clauses. In the first instance this corpus will be used to test their algorithms on an additional domain, and extend their work to include the learning of referring expression generation rules. As far as we know, the SPaRKy restaurant corpus is the only existing corpus of this type, and we hope that the creation of this new corpus in a different domain will provide a useful resource to the Natural Language Generation community.
The task of recommending relevant scientific literature for a draft academic paper has recently received significant interest. In our effort to ease the discovery of scientific literature and augment scientific writing, we aim to improve the relevance of results based on a shallow semantic analysis of the source document and the potential documents to recommend. We investigate the utility of automatic argumentative and rhetorical annotation of documents for this purpose. Specifically, we integrate automatic Core Scientific Concepts (CoreSC) classification into a prototype context-based citation recommendation system and investigate its usefulness to the task. We frame citation recommendation as an information retrieval task and we use the categories of the annotation schemes to apply different weights to the similarity formula. Our results show interesting and consistent correlations between the type of citation and the type of sentence containing the relevant information.
This paper presents SciCorp, a corpus of full-text English scientific papers of two disciplines, genetics and computational linguistics. The corpus comprises co-reference and bridging information as well as information status labels. Since SciCorp is annotated with both labels and the respective co-referent and bridging links, we believe it is a valuable resource for NLP researchers working on scientific articles or on applications such as co-reference resolution, bridging resolution or information status classification. The corpus has been reliably annotated by independent human coders with moderate inter-annotator agreement (average kappa = 0.71). In total, we have annotated 14 full papers containing 61,045 tokens and marked 8,708 definite noun phrases. The paper describes in detail the annotation scheme as well as the resulting corpus. The corpus is available for download in two different formats: in an offset-based format and for the co-reference annotations in the widely-used, tabular CoNLL-2012 format.
Discourse parsing is a challenging task in NLP and plays a crucial role in discourse analysis. To enable discourse analysis for Hindi, Hindi Discourse Relations Bank was created on a subset of Hindi TreeBank. The benefits of a discourse analyzer in automated discourse analysis, question summarization and question answering domains has motivated us to begin work on a discourse analyzer for Hindi. In this paper, we focus on discourse connective identification for Hindi. We explore various available syntactic features for this task. We also explore the use of dependency tree parses present in the Hindi TreeBank and study the impact of the same on the performance of the system. We report that the novel dependency features introduced have a higher impact on precision, in comparison to the syntactic features previously used for this task. In addition, we report a high accuracy of 96% for this task.
This paper contributes to the limited body of empirical research in the domain of discourse structure of information seeking queries. We describe the development of an annotation schema for coding topic development in information seeking queries and the initial observations from a pilot sample of query sessions. The main idea that we explore is the relationship between constant and variable discourse entities and their role in tracking changes in the topic progression. We argue that the topicalized entities remain stable across development of the discourse and can be identified by a simple mechanism where anaphora resolution is a precursor. We also claim that a corpus annotated in this framework can be used as training data for dialogue management and computational semantics systems.
The PML-Tree Query is a general, powerful and user-friendly system for querying richly linguistically annotated treebanks. The paper shows how the PML-Tree Query can be used for searching for discourse relations in the Penn Discourse Treebank 2.0 mapped onto the syntactic annotation of the Penn Treebank.
This study describes a new corpus of over 60,000 hand-annotated metadiscourse acts from 106 OpenCourseWare lectures, from two different disciplines: Physics and Economics. Metadiscourse is a set of linguistic expressions that signal different functions in the discourse. This type of language is hypothesised to be helpful in finding a structure in unstructured text, such as lectures discourse. A brief summary is provided about the annotation scheme and labelling procedures, inter-annotator reliability statistics, overall distributional statistics, a description of auxiliary data that will be distributed with the corpus, and information relating to how to obtain the data. The results provide a deeper understanding of lecture structure and confirm the reliable coding of metadiscursive acts in academic lectures across different disciplines. The next stage of our research will be to build a classification model to automate the tagging process, instead of manual annotation, which take time and efforts. This is in addition to the use of these tags as indicators of the higher level structure of lecture discourse.
We present a large, free, French corpus of online written conversations extracted from the Ubuntu platform’s forums, mailing lists and IRC channels. The corpus is meant to support multi-modality and diachronic studies of online written conversations. We choose to build the corpus around a robust metadata model based upon strong principles, such as the “stand off” annotation principle. We detail the model, we explain how the data was collected and processed - in terms of meta-data, text and conversation - and we detail the corpus’contents through a series of meaningful statistics. A portion of the corpus - about 4,700 sentences from emails, forum posts and chat messages sent in November 2014 - is annotated in terms of dialogue acts and sentiment. We discuss how we adapted our dialogue act taxonomy from the DIT++ annotation scheme and how the data was annotated, before presenting our results as well as a brief qualitative analysis of the annotated data.
We present the DUEL corpus, consisting of 24 hours of natural, face-to-face, loosely task-directed dialogue in German, French and Mandarin Chinese. The corpus is uniquely positioned as a cross-linguistic, multimodal dialogue resource controlled for domain. DUEL includes audio, video and body tracking data and is transcribed and annotated for disfluency, laughter and exclamations.
The paper steps outside the comfort-zone of the traditional NLP tasks like automatic speech recognition (ASR) and machine translation (MT) to addresses two novel problems arising in the automated multilingual news monitoring: segmentation of the TV and radio program ASR transcripts into individual stories, and clustering of the individual stories coming from various sources and languages into storylines. Storyline clustering of stories covering the same events is an essential task for inquisitorial media monitoring. We address these two problems jointly by engaging the low-dimensional semantic representation capabilities of the sequence to sequence neural translation models. To enable joint multi-task learning for multilingual neural translation of morphologically rich languages we replace the attention mechanism with the sliding-window mechanism and operate the sequence to sequence neural translation model on the character-level rather than on the word-level. The story segmentation and storyline clustering problem is tackled by examining the low-dimensional vectors produced as a side-product of the neural translation process. The results of this paper describe a novel approach to the automatic story segmentation and storyline clustering problem.
Handling figurative language like irony is currently a challenging task in natural language processing. Since irony is commonly used in user-generated content, its presence can significantly undermine accurate analysis of opinions and sentiment in such texts. Understanding irony is therefore important if we want to push the state-of-the-art in tasks such as sentiment analysis. In this research, we present the construction of a Twitter dataset for two languages, being English and Dutch, and the development of new guidelines for the annotation of verbal irony in social media texts. Furthermore, we present some statistics on the annotated corpora, from which we can conclude that the detection of contrasting evaluations might be a good indicator for recognizing irony.
We present an analysis of the performance of machine learning classifiers on discriminating between similar languages and language varieties. We carried out a number of experiments using the results of the two editions of the Discriminating between Similar Languages (DSL) shared task. We investigate the progress made between the two tasks, estimate an upper bound on possible performance using ensemble and oracle combination, and provide learning curves to help us understand which languages are more challenging. A number of difficult sentences are identified and investigated further with human annotation
Inspired by the Oxford Children’s Corpus, we have developed a prototype corpus of Arabic texts written and/or selected for children. Our Arabic Children’s Corpus of 2950 documents and nearly 2 million words has been collected manually from the web during a 3-month project. It is of high quality, and contains a range of different children’s genres based on sources located, including classic tales from The Arabian Nights, and popular fictional characters such as Goha. We anticipate that the current and subsequent versions of our corpus will lead to interesting studies in text classification, language use, and ideology in children’s texts.
We introduce a framework for quality assurance of corpora, and apply it to the Reuters Multilingual Corpus (RCV2). The results of this quality assessment of this standard newsprint corpus reveal a significant duplication problem and, to a lesser extent, a problem with corrupted articles. From the raw collection of some 487,000 articles, almost one tenth are trivial duplicates. A smaller fraction of articles appear to be corrupted and should be excluded for that reason. The detailed results are being made available as on-line appendices to this article. This effort also demonstrates the beginnings of a constraint-based methodological framework for quality assessment and quality assurance for corpora. As a first implementation of this framework, we have investigated constraints to verify sample integrity, and to diagnose sample duplication, entropy aberrations, and tagging inconsistencies. To help identify near-duplicates in the corpus, we have employed both entropy measurements and a simple byte bigram incidence digest.
Attribution bias refers to the tendency of people to attribute successes to their own abilities but failures to external factors. In a business context an internal factor might be the restructuring of the firm and an external factor might be an unfavourable change in exchange or interest rates. In accounting research, the presence of an attribution bias has been demonstrated for the narrative sections of the annual financial reports. Previous studies have applied manual content analysis to this problem but in this paper we present novel work to automate the analysis of attribution bias through using machine learning algorithms. Previous studies have only applied manual content analysis on a small scale to reveal such a bias in the narrative section of annual financial reports. In our work a group of experts in accounting and finance labelled and annotated a list of 32,449 sentences from a random sample of UK Preliminary Earning Announcements (PEAs) to allow us to examine whether sentences in PEAs contain internal or external attribution and which kinds of attributions are linked to positive or negative performance. We wished to examine whether human annotators could agree on coding this difficult task and whether Machine Learning (ML) could be applied reliably to replicate the coding process on a much larger scale. Our best machine learning algorithm correctly classified performance sentences with 70% accuracy and detected tone and attribution in financial PEAs with accuracy of 79%.
The paper describes a comparative study of existing and novel text preprocessing and classification techniques for domain detection of user utterances. Two corpora are considered. The first one contains customer calls to a call centre for further call routing; the second one contains answers of call centre employees with different kinds of customer orientation behaviour. Seven different unsupervised and supervised term weighting methods were applied. The collective use of term weighting methods is proposed for classification effectiveness improvement. Four different dimensionality reduction methods were applied: stop-words filtering with stemming, feature selection based on term weights, feature transformation based on term clustering, and a novel feature transformation method based on terms belonging to classes. As classification algorithms we used k-NN and a SVM-based algorithm. The numerical experiments have shown that the simultaneous use of the novel proposed approaches (collectives of term weighting methods and the novel feature transformation method) allows reaching the high classification results with very small number of features.
Paraphrase plagiarism is a significant and widespread problem and research shows that it is hard to detect. Several methods and automatic systems have been proposed to deal with it. However, evaluation and comparison of such solutions is not possible because of the unavailability of benchmark corpora with manual examples of paraphrase plagiarism. To deal with this issue, we present the novel development of a paraphrase plagiarism corpus containing simulated (manually created) examples in the Urdu language - a language widely spoken around the world. This resource is the first of its kind developed for the Urdu language and we believe that it will be a valuable contribution to the evaluation of paraphrase plagiarism detection systems.
Assessing the suitability of an Open Source Software project for adoption requires not only an analysis of aspects related to the code, such as code quality, frequency of updates and new version releases, but also an evaluation of the quality of support offered in related online forums and issue trackers. Understanding the content types of forum messages and issue trackers can provide information about the extent to which requests are being addressed and issues are being resolved, the percentage of issues that are not being fixed, the cases where the user acknowledged that the issue was successfully resolved, etc. These indicators can provide potential adopters of the OSS with estimates about the level of available support. We present a detailed hierarchy of content types of online forum messages and issue tracker comments and a corpus of messages annotated accordingly. We discuss our experiments to classify forum messages and issue tracker comments into content-related classes, i.e.~to assign them to nodes of the hierarchy. The results are very encouraging.
The availability of labelled corpus is of great importance for supervised learning in emotion classification tasks. Because it is time-consuming to manually label text, hashtags have been used as naturally annotated labels to obtain a large amount of labelled training data from microblog. However, natural hashtags contain too much noise for it to be used directly in learning algorithms. In this paper, we design a three-stage semi-automatic method to construct an emotion corpus from microblogs. Firstly, a lexicon based voting approach is used to verify the hashtag automatically. Secondly, a SVM based classifier is used to select the data whose natural labels are consistent with the predicted labels. Finally, the remaining data will be manually examined to filter out the noisy data. Out of about 48K filtered Chinese microblogs, 39k microblogs are selected to form the final corpus with the Kappa value reaching over 0.92 for the automatic parts and over 0.81 for the manual part. The proportion of automatic selection reaches 54.1%. Thus, the method can reduce about 44.5% of manual workload for acquiring quality data. Experiment on a classifier trained on this corpus shows that it achieves comparable results compared to the manually annotated NLP&CC2013 corpus.
Social media texts are often fairly informal and conversational, and when produced by bilinguals tend to be written in several different languages simultaneously, in the same way as conversational speech. The recent availability of large social media corpora has thus also made large-scale code-switched resources available for research. The paper addresses the issues of evaluation and comparison these new corpora entail, by defining an objective measure of corpus level complexity of code-switched texts. It is also shown how this formal measure can be used in practice, by applying it to several code-switched corpora.
To attract foreign students is among the goals of the Karlsruhe Institute of Technology (KIT). One obstacle to achieving this goal is that lectures at KIT are usually held in German which many foreign students are not sufficiently proficient in, as, e.g., opposed to English. While the students from abroad are learning German during their stay at KIT, it is challenging to become proficient enough in it in order to follow a lecture. As a solution to this problem we offer our automatic simultaneous lecture translation. It translates German lectures into English in real time. While not as good as human interpreters, the system is available at a price that KIT can afford in order to offer it in potentially all lectures. In order to assess whether the quality of the system we have conducted a user study. In this paper we present this study, the way it was conducted and its results. The results indicate that the quality of the system has passed a threshold as to be able to support students in their studies. The study has helped to identify the most crucial weaknesses of the systems and has guided us which steps to take next.
This paper introduces the ACL Reference Dataset for Terminology Extraction and Classification, version 2.0 (ACL RD-TEC 2.0). The ACL RD-TEC 2.0 has been developed with the aim of providing a benchmark for the evaluation of term and entity recognition tasks based on specialised text from the computational linguistics domain. This release of the corpus consists of 300 abstracts from articles in the ACL Anthology Reference Corpus, published between 1978–2006. In these abstracts, terms (i.e., single or multi-word lexical units with a specialised meaning) are manually annotated. In addition to their boundaries in running text, annotated terms are classified into one of the seven categories method, tool, language resource (LR), LR product, model, measures and measurements, and other. To assess the quality of the annotations and to determine the difficulty of this annotation task, more than 171 of the abstracts are annotated twice, independently, by each of the two annotators. In total, 6,818 terms are identified and annotated in more than 1300 sentences, resulting in a specialised vocabulary made of 3,318 lexical forms, mapped to 3,471 concepts. We explain the development of the annotation guidelines and discuss some of the challenges we encountered in this annotation task.
We present our guidelines and annotation procedure to create a human corrected machine translated post-edited corpus for the Modern Standard Arabic. Our overarching goal is to use the annotated corpus to develop automatic machine translation post-editing systems for Arabic that can be used to help accelerate the human revision process of translated texts. The creation of any manually annotated corpus usually presents many challenges. In order to address these challenges, we created comprehensive and simplified annotation guidelines which were used by a team of five annotators and one lead annotator. In order to ensure a high annotation agreement between the annotators, multiple training sessions were held and regular inter-annotator agreement measures were performed to check the annotation quality. The created corpus of manual post-edited translations of English to Arabic articles is the largest to date for this language pair.
This work addresses the need to aid Machine Translation (MT) development cycles with a complete workflow of MT evaluation methods. Our aim is to assess, compare and improve MT system variants. We hereby report on novel tools and practices that support various measures, developed in order to support a principled and informed approach of MT development. Our toolkit for automatic evaluation showcases quick and detailed comparison of MT system variants through automatic metrics and n-gram feedback, along with manual evaluation via edit-distance, error annotation and task-based feedback.
Automatic Speech recognition (ASR) is one of the most widely used components in spoken language processing applications. ASR errors are of varying importance with respect to the application, making error analysis keys to improving speech processing applications. Knowing the most serious errors for the applicative case is critical to build better systems. In the context of Automatic Speech Recognition (ASR) used as a first step towards Named Entity Recognition (NER) in speech, error seriousness is usually determined by their frequency, due to the use of the WER as metric to evaluate the ASR output, despite the emergence of more relevant measures in the literature. We propose to use a different evaluation metric form the literature in order to classify ASR errors according to their seriousness for NER. Our results show that the ASR errors importance is ranked differently depending on the used evaluation metric. A more detailed analysis shows that the estimation of the error impact given by the ATENE metric is more adapted to the NER task than the estimation based only on the most used frequency metric WER.
The aim of this experiment is to present an easy way to compare fragments of texts in order to detect (supposed) results of copy & paste operations between articles in the domain of Natural Language Processing (NLP). The search space of the comparisons is a corpus labeled as NLP4NLP gathering a large part of the NLP field. The study is centered on LREC papers in both directions, first with an LREC paper borrowing a fragment of text from the collection, and secondly in the reverse direction with fragments of LREC documents borrowed and inserted in the collection.
Motivated by the adage that a “picture is worth a thousand words” it can be reasoned that automatically enriching the textual content of a document with relevant images can increase the readability of a document. Moreover, features extracted from the additional image data inserted into the textual content of a document may, in principle, be also be used by a retrieval engine to better match the topic of a document with that of a given query. In this paper, we describe our approach of building a ground truth dataset to enable further research into automatic addition of relevant images to text documents. The dataset is comprised of the official ImageCLEF 2010 collection (a collection of images with textual metadata) to serve as the images available for automatic enrichment of text, a set of 25 benchmark documents that are to be enriched, which in this case are children’s short stories, and a set of manually judged relevant images for each query story obtained by the standard procedure of depth pooling. We use this benchmark dataset to evaluate the effectiveness of standard information retrieval methods as simple baselines for this task. The results indicate that using the whole story as a weighted query, where the weight of each query term is its tf-idf value, achieves an precision of 0:1714 within the top 5 retrieved images on an average.
Text analysis methods widely used in digital humanities often involve word co-occurrence, e.g. concept co-occurrence networks. These methods provide a useful corpus overview, but cannot determine the predicates that relate co-occurring concepts. Our goal was identifying propositions expressing the points supported or opposed by participants in international climate negotiations. Word co-occurrence methods were not sufficient, and an analysis based on open relation extraction had limited coverage for nominal predicates. We present a pipeline which identifies the points that different actors support and oppose, via a domain model with support/opposition predicates, and analysis rules that exploit the output of semantic role labelling, syntactic dependencies and anaphora resolution. Entity linking and keyphrase extraction are also performed on the propositions related to each actor. A user interface allows examining the main concepts in points supported or opposed by each participant, which participants agree or disagree with each other, and about which issues. The system is an example of tools that digital humanities scholars are asking for, to render rich textual information (beyond word co-occurrence) more amenable to quantitative treatment. An evaluation of the tool was satisfactory.
The task of Relation Extraction from texts is one of the main challenges in the area of Information Extraction, considering the required linguistic knowledge and the sophistication of the language processing techniques employed. This task aims at identifying and classifying semantic relations that occur between entities recognized in a given text. In this paper, we evaluated a Conditional Random Fields classifier for the extraction of any relation descriptor occurring between named entities (Organisation, Person and Place categories), as well as pre-defined relation types between these entities in Portuguese texts.
This work proposes an information retrieval evaluation set for the Slovak language. A set of 80 queries written in the natural language is given together with the set of relevant documents. The document set contains 3980 newspaper articles sorted into 6 categories. Each document in the result set is manually annotated for relevancy with its corresponding query. The evaluation set is mostly compatible with the Cranfield test collection using the same methodology for queries and annotation of relevancy. In addition to that it provides annotation for document title, author, publication date and category that can be used for evaluation of automatic document clustering and categorization.
This paper proposes how to utilize a search engine in order to predict market shares. We propose to compare rates of concerns of those who search for Web pages among several companies which supply products, given a specific products domain. We measure concerns of those who search for Web pages through search engine suggests. Then, we analyze whether rates of concerns of those who search for Web pages have certain correlation with actual market share. We show that those statistics have certain correlations. We finally propose how to predict the market share of a specific product genre based on the rates of concerns of those who search for Web pages.
Keyphrase extraction is the task of finding phrases that represent the important content of a document. The main aim of keyphrase extraction is to propose textual units that represent the most important topics developed in a document. The output keyphrases of automatic keyphrase extraction methods for test documents are typically evaluated by comparing them to manually assigned reference keyphrases. Each output keyphrase is considered correct if it matches one of the reference keyphrases. However, the choice of the appropriate textual unit (keyphrase) for a topic is sometimes subjective and evaluating by exact matching underestimates the performance. This paper presents a dataset of evaluation scores assigned to automatically extracted keyphrases by human evaluators. Along with the reference keyphrases, the manual evaluations can be used to validate new evaluation measures. Indeed, an evaluation measure that is highly correlated to the manual evaluation is appropriate for the evaluation of automatic keyphrase extraction methods.
We present the Royal Society Corpus (RSC) built from the Philosophical Transactions and Proceedings of the Royal Society of London. At present, the corpus contains articles from the first two centuries of the journal (1665―1869) and amounts to around 35 million tokens. The motivation for building the RSC is to investigate the diachronic linguistic development of scientific English. Specifically, we assume that due to specialization, linguistic encodings become more compact over time (Halliday, 1988; Halliday and Martin, 1993), thus creating a specific discourse type characterized by high information density that is functional for expert communication. When building corpora from uncharted material, typically not all relevant meta-data (e.g. author, time, genre) or linguistic data (e.g. sentence/word boundaries, words, parts of speech) is readily available. We present an approach to obtain good quality meta-data and base text data adopting the concept of Agile Software Development.
Common people often experience difficulties in accessing relevant, correct, accurate and understandable health information online. Developing search techniques that aid these information needs is challenging. In this paper we present the datasets created by CLEF eHealth Lab from 2013-2015 for evaluation of search solutions to support common people finding health information online. Specifically, the CLEF eHealth information retrieval (IR) task of this Lab has provided the research community with benchmarks for evaluating consumer-centered health information retrieval, thus fostering research and development aimed to address this challenging problem. Given consumer queries, the goal of the task is to retrieve relevant documents from the provided collection of web pages. The shared datasets provide a large health web crawl, queries representing people’s real world information needs, and relevance assessment judgements for the queries.
This article presents a corpus for development and testing of event schema induction systems in English. Schema induction is the task of learning templates with no supervision from unlabeled texts, and to group together entities corresponding to the same role in a template. Most of the previous work on this subject relies on the MUC-4 corpus. We describe the limits of using this corpus (size, non-representativeness, similarity of roles across templates) and propose a new, partially-annotated corpus in English which remedies some of these shortcomings. We make use of Wikinews to select the data inside the category Laws & Justice, and query Google search engine to retrieve different documents on the same events. Only Wikinews documents are manually annotated and can be used for evaluation, while the others can be used for unsupervised learning. We detail the methodology used for building the corpus and evaluate some existing systems on this new data.
The current study focuses on optimization of Levenshtein algorithm for the purpose of computing the optimal alignment between two phoneme transcriptions of spoken utterance containing sequences of phonetic symbols. The alignment is computed with the help of a confusion matrix in which costs for phonetic symbol deletion, insertion and substitution are defined taking into account various phonological processes that occur in fluent speech, such as anticipatory assimilation, phone elision and epenthesis. The corpus containing about 30 hours of Russian read speech was used to evaluate the presented algorithms. The experimental results have shown significant reduction of misalignment rate in comparison with the baseline Levenshtein algorithm: the number of errors has been reduced from 1.1 % to 0.28 %
This paper describes speech data recording, processing and annotation of a new speech corpus CoRuSS (Corpus of Russian Spontaneous Speech), which is based on connected communicative speech recorded from 60 native Russian male and female speakers of different age groups (from 16 to 77). Some Russian speech corpora available at the moment contain plain orthographic texts and provide some kind of limited annotation, but there are no corpora providing detailed prosodic annotation of spontaneous conversational speech. This corpus contains 30 hours of high quality recorded spontaneous Russian speech, half of it has been transcribed and prosodically labeled. The recordings consist of dialogues between two speakers, monologues (speakers’ self-presentations) and reading of a short phonetically balanced text. Since the corpus is labeled for a wide range of linguistic - phonetic and prosodic - information, it provides basis for empirical studies of various spontaneous speech phenomena as well as for comparison with those we observe in prepared read speech. Since the corpus is designed as a open-access resource of speech data, it will also make possible to advance corpus-based analysis of spontaneous speech data across languages and speech technology development as well.
Recently, there has been an explosion in the availability of large, good-quality cross-linguistic databases such as WALS (Dryer & Haspelmath, 2013), Glottolog (Hammarstrom et al., 2015) and Phoible (Moran & McCloy, 2014). Databases such as Phoible contain the actual segments used by various languages as they are given in the primary language descriptions. However, this segment-level representation cannot be used directly for analyses that require generalizations over classes of segments that share theoretically interesting features. Here we present a method and the associated R (R Core Team, 2014) code that allows the flexible definition of such meaningful classes and that can identify the sets of segments falling into such a class for any language inventory. The method and its results are important for those interested in exploring cross-linguistic patterns of phonetic and phonological diversity and their relationship to extra-linguistic factors and processes such as climate, economics, history or human genetics.
SEA_AP (Segmentador e Etiquetador Automático para Análise Prosódica, Automatic Segmentation and Labelling for Prosodic Analysis) toolkit is an application that performs audio segmentation and labelling to create a TextGrid file which will be used to launch a prosodic analysis using Praat. In this paper, we want to describe the improved functionality of the tool achieved by adding a dialectometric analysis module using R scripts. The dialectometric analysis includes computing correlations among F0 curves and it obtains prosodic distances among the different variables of interest (location, speaker, structure, etc.). The dialectometric analysis requires large databases in order to be adequately computed, and automatic segmentation and labelling can create them thanks to a procedure less costly than the manual alternative. Thus, the integration of these tools into the SEA_AP allows to propose a distribution of geoprosodic areas by means of a quantitative method, which completes the traditional dialectological point of view. The current version of the SEA_AP toolkit is capable of analysing Galician, Spanish and Brazilian Portuguese data, and hence the distances between several prosodic linguistic varieties can be measured at present.
In this paper, we present a music retrieval and recommendation system using machine learning techniques. We propose a query by humming system for music retrieval that uses deep neural networks for note transcription and a note-based retrieval system for retrieving the correct song from the database. We evaluate our query by humming system using the standard MIREX QBSH dataset. We also propose a similar artist recommendation system which recommends similar artists based on acoustic features of the artists’ music, online text descriptions of the artists and social media data. We use supervised machine learning techniques over all our features and compare our recommendation results to those produced by a popular similar artist recommendation website.
Vocal User Interfaces in domestic environments recently gained interest in the speech processing community. This interest is due to the opportunity of using it in the framework of Ambient Assisted Living both for home automation (vocal command) and for call for help in case of distress situations, i.e. after a fall. C IRDO X, which is a modular software, is able to analyse online the audio environment in a home, to extract the uttered sentences and then to process them thanks to an ASR module. Moreover, this system perfoms non-speech audio event classification; in this case, specific models must be trained. The software is designed to be modular and to process on-line the audio multichannel stream. Some exemples of studies in which C IRDO X was involved are described. They were operated in real environment, namely a Living lab environment.
Computer-assisted transcription promises high-quality speech transcription at reduced costs. This is achieved by limiting human effort to transcribing parts for which automatic transcription quality is insufficient. Our goal is to improve the human transcription quality via appropriate user interface design. We focus on iterative interfaces that allow humans to solve tasks based on an initially given suggestion, in this case an automatic transcription. We conduct a user study that reveals considerable quality gains for three variations of iterative interfaces over a non-iterative from-scratch transcription interface. Our iterative interfaces included post-editing, confidence-enhanced post-editing, and a novel retyping interface. All three yielded similar quality on average, but we found that the proposed retyping interface was less sensitive to the difficulty of the segment, and superior when the automatic transcription of the segment contained relatively many errors. An analysis using mixed-effects models allows us to quantify these and other factors and draw conclusions over which interface design should be chosen in which circumstance.
This paper introduces a new British English speech database, named the homeService corpus, which has been gathered as part of the homeService project. This project aims to help users with speech and motor disabilities to operate their home appliances using voice commands. The audio recorded during such interactions consists of realistic data of speakers with severe dysarthria. The majority of the homeService corpus is recorded in real home environments where voice control is often the normal means by which users interact with their devices. The collection of the corpus is motivated by the shortage of realistic dysarthric speech corpora available to the scientific community. Along with the details on how the data is organised and how it can be accessed, a brief description of the framework used to make the recordings is provided. Finally, the performance of the homeService automatic recogniser for dysarthric speech trained with single-speaker data from the corpus is provided as an initial baseline. Access to the homeService corpus is provided through the dedicated web page at http://mini.dcs.shef.ac.uk/resources/homeservice-corpus/. This will also have the most updated description of the data. At the time of writing the collection process is still ongoing.
Perceptive evaluation of speech disorders is still the standard method in clinical practice for the diagnosing and the following of the condition progression of patients. Such methods include different tasks such as read speech, spontaneous speech, isolated words, sustained vowels, etc. In this context, automatic speech processing tools have proven pertinence in speech quality evaluation and assistive technology-based applications. Though, a very few studies have investigated the use of automatic tools on spontaneous speech. This paper investigates the behavior of an automatic phone-based anomaly detection system when applied on read and spontaneous French dysarthric speech. The behavior of the automatic tool reveals interesting inter-pathology differences across speech styles.
We present a text-to-speech (TTS) system designed for the dialect of Bengali spoken in Bangladesh. This work is part of an ongoing effort to address the needs of under-resourced languages. We propose a process for streamlining the bootstrapping of TTS systems for under-resourced languages. First, we use crowdsourcing to collect the data from multiple ordinary speakers, each speaker recording small amount of sentences. Second, we leverage an existing text normalization system for a related language (Hindi) to bootstrap a linguistic front-end for Bangla. Third, we employ statistical techniques to construct multi-speaker acoustic models using Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) and Hidden Markov Model (HMM) approaches. We then describe our experiments that show that the resulting TTS voices score well in terms of their perceived quality as measured by Mean Opinion Score (MOS) evaluations.
With the increasing amount of audiovisual and digital data deriving from televisual and radiophonic sources, professional archives such as INA, France’s national audiovisual institute, acknowledge a growing need for efficient indexing tools. In this paper, we describe the Speech Trax system that aims at analyzing the audio content of TV and radio documents. In particular, we focus on the speaker tracking task that is very valuable for indexing purposes. First, we detail the overall architecture of the system and show the results obtained on a large-scale experiment, the largest to our knowledge for this type of content (about 1,300 speakers). Then, we present the Speech Trax demonstrator that gathers the results of various automatic speech processing techniques on top of our speaker tracking system (speaker diarization, speech transcription, etc.). Finally, we provide insight on the obtained performances and suggest hints for future improvements.
In this article we propose a descriptive study of a chat conversations corpus from an assistance contact center. Conversations are described from several view points, including interaction analysis, language deviation analysis and typographic expressivity marks analysis. We provide in particular a detailed analysis of language deviations that are encountered in our corpus of 230 conversations, corresponding to 6879 messages and 76839 words. These deviations may be challenging for further syntactic and semantic parsing. Analysis is performed with a distinction between Customer messages and Agent messages. On the overall only 4% of the observed words are misspelled but 26% of the messages contain at least one erroneous word (rising to 40% when focused on Customer messages). Transcriptions of telephone conversations from an assistance call center are also studied, allowing comparisons between these two interaction modes to be drawn. The study reveals significant differences in terms of conversation flow, with an increased efficiency for chat conversations in spite of longer temporal span.
Monitoring social media has been shown to be an interesting approach for the early detection of drug adverse effects. In this paper, we describe a system which extracts medical entities in French drug reviews written by users. We focus on the identification of medical conditions, which is based on the concept of post-coordination: we first extract minimal medical-related entities (pain, stomach) then we combine them to identify complex ones (It was the worst [pain I ever felt in my stomach]). These two steps are respectively performed by two classifiers, the first being based on Conditional Random Fields and the second one on Support Vector Machines. The overall results of the minimal entity classifier are the following: P=0.926; R=0.849; F1=0.886. A thourough analysis of the feature set shows that, when combined with word lemmas, clusters generated by word2vec are the most valuable features. When trained on the output of the first classifier, the second classifier’s performances are the following: p=0.683;r=0.956;f1=0.797. The addition of post-processing rules did not add any significant global improvement but was found to modify the precision/recall ratio.
Data acquisition in dialectology is typically a tedious task, as dialect samples of spoken language have to be collected via questionnaires or interviews. In this article, we suggest to use the “web as a corpus” approach for dialectology. We present a case study that demonstrates how authentic language data for the Bavarian dialect (ISO 639-3:bar) can be collected automatically from the social network Facebook. We also show that Facebook can be used effectively as a crowdsourcing platform, where users are willing to translate dialect words collaboratively in order to create a common lexicon of their Bavarian dialect. Key insights from the case study are summarized as “lessons learned”, together with suggestions for future enhancements of the lexicon creation approach.
In order to gain a deep understanding of how social context manifests in interactions, we need data that represents interactions from a large community of people over a long period of time, capturing different aspects of social context. In this paper, we present a large corpus of Wikipedia Talk page discussions that are collected from a broad range of topics, containing discussions that happened over a period of 15 years. The dataset contains 166,322 discussion threads, across 1236 articles/topics that span 15 different topic categories or domains. The dataset also captures whether the post is made by an registered user or not, and whether he/she was an administrator at the time of making the post. It also captures the Wikipedia age of editors in terms of number of months spent as an editor, as well as their gender. This corpus will be a valuable resource to investigate a variety of computational sociolinguistics research questions regarding online social interactions.
Natural Language Engineering tasks require large and complex annotated datasets to build more advanced models of language. Corpora are typically annotated by several experts to create a gold standard; however, there are now compelling reasons to use a non-expert crowd to annotate text, driven by cost, speed and scalability. Phrase Detectives Corpus 1.0 is an anaphorically-annotated corpus of encyclopedic and narrative text that contains a gold standard created by multiple experts, as well as a set of annotations created by a large non-expert crowd. Analysis shows very good inter-expert agreement (kappa=.88-.93) but a more variable baseline crowd agreement (kappa=.52-.96). Encyclopedic texts show less agreement (and by implication are harder to annotate) than narrative texts. The release of this corpus is intended to encourage research into the use of crowds for text annotation and the development of more advanced, probabilistic language models, in particular for anaphoric coreference.
This paper presents Summ-it++, an enriched version the Summ-it corpus. In this new version, the corpus has received new semantic layers, named entity categories and relations between named entities, adding to the previous coreference annotation. In addition, we change the original Summ-it format to SemEval
Despite the popularity of coreference resolution as a research topic, the overwhelming majority of the work in this area focused so far on single antecedence coreference only. Multiple antecedent coreference (MAC) has been largely neglected. This can be explained by the scarcity of the phenomenon of MAC in generic discourse. However, in specialized discourse such as patents, MAC is very dominant. It seems thus unavoidable to address the problem of MAC resolution in the context of tasks related to automatic patent material processing, among them abstractive summarization, deep parsing of patents, construction of concept maps of the inventions, etc. We present the first version of an operational rule-based MAC resolution strategy for patent material that covers the three major types of MAC: (i) nominal MAC, (ii) MAC with personal / relative pronouns, and MAC with reflexive / reciprocal pronouns. The evaluation shows that our strategy performs well in terms of precision and recall.
This paper presents a second release of the ARRAU dataset: a multi-domain corpus with thorough linguistically motivated annotation of anaphora and related phenomena. Building upon the first release almost a decade ago, a considerable effort had been invested in improving the data both quantitatively and qualitatively. Thus, we have doubled the corpus size, expanded the selection of covered phenomena to include referentiality and genericity and designed and implemented a methodology for enforcing the consistency of the manual annotation. We believe that the new release of ARRAU provides a valuable material for ongoing research in complex cases of coreference as well as for a variety of related tasks. The corpus is publicly available through LDC.
We describe a method for identifying and performing functional analysis of structured regions that are embedded in natural language documents, such as tables or key-value lists. Such regions often encode information according to ad hoc schemas and avail themselves of visual cues in place of natural language grammar, presenting problems for standard information extraction algorithms. Unlike previous work in table extraction, which assumes a relatively noiseless two-dimensional layout, our aim is to accommodate a wide variety of naturally occurring structure types. Our approach has three main parts. First, we collect and annotate a a diverse sample of “naturally” occurring structures from several sources. Second, we use probabilistic text segmentation techniques, featurized by skip bigrams over spatial and token category cues, to automatically identify contiguous regions of structured text that share a common schema. Finally, we identify the records and fields within each structured region using a combination of distributional similarity and sequence alignment methods, guided by minimal supervision in the form of a single annotated record. We evaluate the last two components individually, and conclude with a discussion of further work.
In distributional semantics words are represented by aggregated context features. The similarity of words can be computed by comparing their feature vectors. Thus, we can predict whether two words are synonymous or similar with respect to some other semantic relation. We will show on six different datasets of pairs of similar and non-similar words that a supervised learning algorithm on feature vectors representing pairs of words outperforms cosine similarity between vectors representing single words. We compared different methods to construct a feature vector representing a pair of words. We show that simple methods like pairwise addition or multiplication give better results than a recently proposed method that combines different types of features. The semantic relation we consider is relatedness of terms in thesauri for intellectual document classification. Thus our findings can directly be applied for the maintenance and extension of such thesauri. To the best of our knowledge this relation was not considered before in the field of distributional semantics.
We present a proposal for the annotation of factuality of event mentions in Spanish texts and a free available annotated corpus. Our factuality model aims to capture a pragmatic notion of factuality, trying to reflect a casual reader judgements about the realis / irrealis status of mentioned events. Also, some learning experiments (SVM and CRF) have been held, showing encouraging results.
Lately, with the success of Deep Learning techniques in some computational linguistics tasks, many researchers want to explore new models for their linguistics applications. These models tend to be very different from what standard Neural Networks look like, limiting the possibility to use standard Neural Networks frameworks. This work presents NNBlocks, a new framework written in Python to build and train Neural Networks that are not constrained by a specific kind of architecture, making it possible to use it in computational linguistics.
This paper presents some preliminary results of the OPLON project. It aimed at identifying early linguistic symptoms of cognitive decline in the elderly. This pilot study was conducted on a corpus composed of spontaneous speech sample collected from 39 subjects, who underwent a neuropsychological screening for visuo-spatial abilities, memory, language, executive functions and attention. A rich set of linguistic features was extracted from the digitalised utterances (at phonetic, suprasegmental, lexical, morphological and syntactic levels) and the statistical significance in pinpointing the pathological process was measured. Our results show remarkable trends for what concerns both the linguistic traits selection and the automatic classifiers building.
This paper describes the recording of a speech corpus focused on prosody of people with intellectual disabilities. To do this, a video game is used with the aim of improving the user’s motivation. Moreover, the player’s profiles and the sentences recorded during the game sessions are described. With the purpose of identifying the main prosodic troubles of people with intellectual disabilities, some prosodic features are extracted from recordings, like fundamental frequency, energy and pauses. After that, a comparison is made between the recordings of people with intellectual disabilities and people without intellectual disabilities. This comparison shows that pauses are the best discriminative feature between these groups. To check this, a study has been done using machine learning techniques, with a classification rate superior to 80%.
Autism spectrum disorder (ASD) is a complex neurodevelopmental condition that would benefit from low-cost and reliable improvements to screening and diagnosis. Human language technologies (HLTs) provide one possible route to automating a series of subjective decisions that currently inform “Gold Standard” diagnosis based on clinical judgment. In this paper, we describe a new resource to support this goal, comprised of 100 20-minute semi-structured English language samples labeled with child age, sex, IQ, autism symptom severity, and diagnostic classification. We assess the feasibility of digitizing and processing sensitive clinical samples for data sharing, and identify areas of difficulty. Using the methods described here, we propose to join forces with researchers and clinicians throughout the world to establish an international repository of annotated language samples from individuals with ASD and related disorders. This project has the potential to improve the lives of individuals with ASD and their families by identifying linguistic features that could improve remote screening, inform personalized intervention, and promote advancements in clinically-oriented HLTs.
We propose in this work a novel acoustic phonetic study for Arabic people suffering from language disabilities and non-native learners of Arabic language to classify Arabic continuous speech to pathological or healthy and to identify phonemes that pose pronunciation problems (case of pathological speeches). The main idea can be summarized in comparing between the phonetic model reference to Arabic spoken language and that proper to concerned speaker. For this task, we use techniques of automatic speech processing like forced alignment and artificial neural network (ANN) (Basheer, 2000). Based on a test corpus containing 100 speech sequences, recorded by different speakers (healthy/pathological speeches and native/foreign speakers), we attain 97% as classification rate. Algorithms used in identifying phonemes that pose pronunciation problems show high efficiency: we attain an identification rate of 100%.
In this paper, we investigate a covert labeling cue, namely the probability that a title (by example of the Wikipedia titles) is a noun. If this probability is very large, any list such as or comparable to the Wikipedia titles can be used as a reliable word-class (or part-of-speech tag) predictor or noun lexicon. This may be especially useful in the case of Low Resource Languages (LRL) where labeled data is lacking and putatively for Natural Language Processing (NLP) tasks such as Word Sense Disambiguation, Sentiment Analysis and Machine Translation. Profitting from the ease of digital publication on the web as opposed to print, LRL speaker communities produce resources such as Wikipedia and Wiktionary, which can be used for an assessment. We provide statistical evidence for a strong noun bias for the Wikipedia titles from 2 corpora (English, Persian) and a dictionary (Japanese) and for a typologically balanced set of 17 languages including LRLs. Additionally, we conduct a small experiment on predicting noun tags for out-of-vocabulary items in part-of-speech tagging for English.
Although some words carry strong associations with specific colors (e.g., the word danger is associated with the color red), few studies have investigated these relationships. This may be due to the relative rarity of databases that contain large quantities of such information. Additionally, these resources are often limited to particular languages, such as English. Moreover, the existing resources often do not consider the possible contexts of words in assessing the associations between a word and a color. As a result, the influence of context on word―color associations is not fully understood. In this study, we constructed a novel language resource for word―color associations. The resource has two characteristics: First, our resource is the first to include Japanese word―color associations, which were collected via crowdsourcing. Second, the word―color associations in the resource are linked to contexts. We show that word―color associations depend on language and that associations with certain colors are affected by context information.
This paper presents a collection of annotations (tags or keywords) for a set of 2,133 environmental sounds taken from the Freesound database (www.freesound.org). The annotations are acquired through an open-ended crowd-labeling task, in which participants were asked to provide keywords for each of three sounds. The main goal of this study is to find out (i) whether it is feasible to collect keywords for a large collection of sounds through crowdsourcing, and (ii) how people talk about sounds, and what information they can infer from hearing a sound in isolation. Our main finding is that it is not only feasible to perform crowd-labeling for a large collection of sounds, it is also very useful to highlight different aspects of the sounds that authors may fail to mention. Our data is freely available, and can be used to ground semantic models, improve search in audio databases, and to study the language of sound.
We present a new large dataset of 12403 context-sensitive verb relations manually annotated via crowdsourcing. These relations capture fine-grained semantic information between verb-centric propositions, such as temporal or entailment relations. We propose a novel semantic verb relation scheme and design a multi-step annotation approach for scaling-up the annotations using crowdsourcing. We employ several quality measures and report on agreement scores. The resulting dataset is available under a permissive CreativeCommons license at www.ukp.tu-darmstadt.de/data/verb-relations/. It represents a valuable resource for various applications, such as automatic information consolidation or automatic summarization.
We describe an experiment for the acquisition of opposition relations among Italian verb senses, based on a crowdsourcing methodology. The goal of the experiment is to discuss whether the types of opposition we distinguish (i.e. complementarity, antonymy, converseness and reversiveness) are actually perceived by the crowd. In particular, we collect data for Italian by using the crowdsourcing platform CrowdFlower. We ask annotators to judge the type of opposition existing among pairs of sentences -previously judged as opposite- that differ only for a verb: the verb in the first sentence is opposite of the verb in second sentence. Data corroborate the hypothesis that some opposition relations exclude each other, while others interact, being recognized as compatible by the contributors.
We announce the release of the CROWDED CORPUS: a pair of speech corpora collected via crowdsourcing, containing a native speaker corpus of English (CROWDED_ENGLISH), and a corpus of German/English bilinguals (CROWDED_BILINGUAL). Release 1 of the CROWDED CORPUS contains 1000 recordings amounting to 33,400 tokens collected from 80 speakers and is freely available to other researchers. We recruited participants via the Crowdee application for Android. Recruits were prompted to respond to business-topic questions of the type found in language learning oral tests. We then used the CrowdFlower web application to pass these recordings to crowdworkers for transcription and annotation of errors and sentence boundaries. Finally, the sentences were tagged and parsed using standard natural language processing tools. We propose that crowdsourcing is a valid and economical method for corpus collection, and discuss the advantages and disadvantages of this approach.
Our interest is in people’s capacity to efficiently and effectively describe geographic objects in urban scenes. The broader ambition is to develop spatial models capable of equivalent functionality able to construct such referring expressions. To that end we present a newly crowd-sourced data set of natural language references to objects anchored in complex urban scenes (In short: The REAL Corpus ― Referring Expressions Anchored Language). The REAL corpus contains a collection of images of real-world urban scenes together with verbal descriptions of target objects generated by humans, paired with data on how successful other people were able to identify the same object based on these descriptions. In total, the corpus contains 32 images with on average 27 descriptions per image and 3 verifications for each description. In addition, the corpus is annotated with a variety of linguistically motivated features. The paper highlights issues posed by collecting data using crowd-sourcing with an unrestricted input format, as well as using real-world urban scenes.
Crowdsourcing is an arising collaborative approach applicable among many other applications to the area of language and speech processing. In fact, the use of crowdsourcing was already applied in the field of speech processing with promising results. However, only few studies investigated the use of crowdsourcing in computational paralinguistics. In this contribution, we propose a novel evaluator for crowdsourced-based ratings termed Weighted Trustability Evaluator (WTE) which is computed from the rater-dependent consistency over the test questions. We further investigate the reliability of crowdsourced annotations as compared to the ones obtained with traditional labelling procedures, such as constrained listening experiments in laboratories or in controlled environments. This comparison includes an in-depth analysis of obtainable classification performances. The experiments were conducted on the Speaker Likability Database (SLD) already used in the INTERSPEECH Challenge 2012, and the results lend further weight to the assumption that crowdsourcing can be applied as a reliable annotation source for computational paralinguistics given a sufficient number of raters and suited measurements of their reliability.
In online computer-mediated communication, speakers were considered to have experienced difficulties in catching their partner’s emotions and in conveying their own emotions. To explain why online emotional communication is so difficult and to investigate how this problem should be solved, multimodal online emotional communication corpus was constructed by recording approximately 100 speakers’ emotional expressions and reactions in a modality-controlled environment. Speakers communicated over the Internet using video chat, voice chat or text chat; their face-to-face conversations were used for comparison purposes. The corpora incorporated emotional labels by evaluating the speaker’s dynamic emotional states and the measurements of the speaker’s facial expression, vocal expression and autonomic nervous system activity. For the initial study of this project, which used a large-scale emotional communication corpus, the accuracy of online emotional understanding was assessed to demonstrate the emotional labels evaluated by the speakers and to summarize the speaker’s answers on the questionnaire regarding the difference between an online chat and face-to-face conversations in which they actually participated. The results revealed that speakers have difficulty communicating their emotions in online communication environments, regardless of the type of communication modality and that inaccurate emotional understanding occurs more frequently in online computer-mediated communication than in face-to-face communication.
This paper presents a quantitative description of laughter in height 1-hour French spontaneous conversations. The paper includes the raw figures for laughter as well as more details concerning inter-individual variability. It firstly describes to what extent the amount of laughter and their durations varies from speaker to speaker in all dialogs. In a second suite of analyses, this paper compares our corpus with previous analyzed corpora. In a final set of experiments, it presents some facts about overlapping laughs. This paper have quantified these all effects in free-style conversations, for the first time.
It has been shown that adding expressivity and emotional expressions to an agent’s communication systems would improve the interaction quality between this agent and a human user. In this paper we present a multimodal database of affect bursts, which are very short non-verbal expressions with facial, vocal, and gestural components that are highly synchronized and triggered by an identifiable event. This database contains motion capture and audio data of affect bursts representing disgust, startle and surprise recorded at three different levels of arousal each. This database is to be used for synthesis purposes in order to generate affect bursts of these emotions on a continuous arousal level scale.
Emotional aspects play a vital role in making human communication a rich and dynamic experience. As we introduce more automated system in our daily lives, it becomes increasingly important to incorporate emotion to provide as natural an interaction as possible. To achieve said incorporation, rich sets of labeled emotional data is prerequisite. However, in Japanese, existing emotion database is still limited to unimodal and bimodal corpora. Since emotion is not only expressed through speech, but also visually at the same time, it is essential to include multiple modalities in an observation. In this paper, we present the first audio-visual emotion corpora in Japanese, collected from 14 native speakers. The corpus contains 100 minutes of annotated and transcribed material. We performed preliminary emotion recognition experiments on the corpus and achieved an accuracy of 61.42% for five classes of emotion.
In this paper we compare different context selection approaches to improve the creation of Emotive Vector Space Models (VSMs). The system is based on the results of an existing approach that showed the possibility to create and update VSMs by exploiting crowdsourcing and human annotation. Here, we introduce a method to manipulate the contexts of the VSMs under the assumption that the emotive connotation of a target word is a function of both its syntagmatic and paradigmatic association with the various emotions. To study the differences among the proposed spaces and to confirm the reliability of the system, we report on two experiments: in the first one we validated the best candidates extracted from each model, and in the second one we compared the models’ performance on a random sample of target words. Both experiments have been implemented as crowdsourcing tasks.
Sentence alignment is a task that consists in aligning the parallel sentences in a translated article pair. This paper describes a method to perform sentence boundary detection and alignment simultaneously, which significantly improves the alignment accuracy on languages like Chinese with uncertain sentence boundaries. It relies on the definition of hard (certain) and soft (uncertain) punctuation delimiters, the latter being possibly ignored to optimize the alignment result. The alignment method is used in combination with lexicons automatically generated from the input article pairs using pivot-based MT, achieving better coverage of the input words with fewer entries than pre-existing dictionaries. Pivot-based MT makes it possible to build dictionaries for language pairs that have scarce parallel data. The alignment method is implemented in a tool that will be freely available in the near future.
Parallel corpora are often injected with bilingual lexical resources for improved Indian language machine translation (MT). In absence of such lexical resources, multilingual topic models have been used to create coarse lexical resources in the past, using a Cartesian product approach. Our results show that for morphologically rich languages like Hindi, the Cartesian product approach is detrimental for MT. We then present a novel ‘sentential’ approach to use this coarse lexical resource from a multilingual topic model. Our coarse lexical resource when injected with a parallel corpus outperforms a system trained using parallel corpus and a good quality lexical resource. As demonstrated by the quality of our coarse lexical resource and its benefit to MT, we believe that our sentential approach to create such a resource will help MT for resource-constrained languages.
In this paper, we describe the details of the ASPEC (Asian Scientific Paper Excerpt Corpus), which is the first large-size parallel corpus of scientific paper domain. ASPEC was constructed in the Japanese-Chinese machine translation project conducted between 2006 and 2010 using the Special Coordination Funds for Promoting Science and Technology. It consists of a Japanese-English scientific paper abstract corpus of approximately 3 million parallel sentences (ASPEC-JE) and a Chinese-Japanese scientific paper excerpt corpus of approximately 0.68 million parallel sentences (ASPEC-JC). ASPEC is used as the official dataset for the machine translation evaluation workshop WAT (Workshop on Asian Translation).
This paper presents how an state-of-the-art SMT system is enriched by using an extra in-domain parallel corpora extracted from Wikipedia. We collect corpora from parallel titles and from parallel fragments in comparable articles from Wikipedia. We carried out an evaluation with a double objective: evaluating the quality of the extracted data and evaluating the improvement due to the domain-adaptation. We think this can be very useful for languages with limited amount of parallel corpora, where in-domain data is crucial to improve the performance of MT sytems. The experiments on the Spanish-English language pair improve a baseline trained with the Europarl corpus in more than 2 points of BLEU when translating in the Computer Science domain.
This paper presents ProphetMT, a tree-based SMT-driven Controlled Language (CL) authoring and post-editing tool. ProphetMT employs the source-side rules in a translation model and provides them as auto-suggestions to users. Accordingly, one might say that users are writing in a Controlled Language that is understood by the computer. ProphetMT also allows users to easily attach structural information as they compose content. When a specific rule is selected, a partial translation is promptly generated on-the-fly with the help of the structural information. Our experiments conducted on English-to-Chinese show that our proposed ProphetMT system can not only better regularise an author’s writing behaviour, but also significantly improve translation fluency which is vital to reduce the post-editing time. Additionally, when the writing and translation process is over, ProphetMT can provide an effective colour scheme to further improve the productivity of post-editors by explicitly featuring the relations between the source and target rules.
Bilingual lexica are the basis for many cross-lingual natural language processing tasks. Recent works have shown success in learning bilingual dictionary by taking advantages of comparable corpora and a diverse set of signals derived from monolingual corpora. In the present work, we describe an approach to automatically learn bilingual lexica by training a supervised classifier using word embedding-based vectors of only a few hundred translation equivalent word pairs. The word embedding representations of translation pairs were obtained from source and target monolingual corpora, which are not necessarily related. Our classifier is able to predict whether a new word pair is under a translation relation or not. We tested it on two quite distinct language pairs Chinese-Spanish and English-Spanish. The classifiers achieved more than 0.90 precision and recall for both language pairs in different evaluation scenarios. These results show a high potential for this method to be used in bilingual lexica production for language pairs with reduced amount of parallel or comparable corpora, in particular for phrase table expansion in Statistical Machine Translation systems.
In this paper, we introduce a coverage-based scoring function that discriminates between parallel and non-parallel sentences. When plugged into Bleualign, a state-of-the-art sentence aligner, our function improves both precision and recall of alignments over the originally proposed BLEU score. Furthermore, since our scoring function uses Moses phrase tables directly we avoid the need to translate the texts to be aligned, which is time-consuming and a potential source of alignment errors.
This paper presents a solution to evaluate spoken post-editing of imperfect machine translation output by a human translator. We compare two approaches to the combination of machine translation (MT) and automatic speech recognition (ASR): a heuristic algorithm and a machine learning method. To obtain a data set with spoken post-editing information, we use the French version of TED talks as the source texts submitted to MT, and the spoken English counterparts as their corrections, which are submitted to an ASR system. We experiment with various levels of artificial ASR noise and also with a state-of-the-art ASR system. The results show that the combination of MT with ASR improves over both individual outputs of MT and ASR in terms of BLEU scores, especially when ASR performance is low.
This paper presents our work towards a novel approach for Quality Estimation (QE) of machine translation based on sequences of adjacent words, the so-called phrases. This new level of QE aims to provide a natural balance between QE at word and sentence-level, which are either too fine grained or too coarse levels for some applications. However, phrase-level QE implies an intrinsic challenge: how to segment a machine translation into sequence of words (contiguous or not) that represent an error. We discuss three possible segmentation strategies to automatically extract erroneous phrases. We evaluate these strategies against annotations at phrase-level produced by humans, using a new dataset collected for this purpose.
In this paper, we present a freely available corpus of human and automatic translations of subtitles. The corpus comprises, the original English subtitles (SRC), both human (HT) and machine translations (MT) into German, as well as post-editions (PE) of the MT output. HT and MT are annotated with errors. Moreover, human evaluation is included in HT, MT, and PE. Such a corpus is a valuable resource for both human and machine translation communities, enabling the direct comparison – in terms of errors and evaluation – between human and machine translations and post-edited machine translations.
This paper describes a pilot study in lexical encoding of multi-word expressions (MWEs) in 4 Latin American dialects of Spanish: Costa Rican, Colombian, Mexican and Peruvian. We describe the variability of MWE usage across dialects. We adapt an existing data model to a dialect-aware encoding, so as to represent dialect-related specificities, while avoiding redundancy of the data common for all dialects. A dozen of linguistic properties of MWEs can be expressed in this model, both on the level of a whole MWE and of its individual components. We describe the resulting lexical resource containing several dozens of MWEs in four dialects and we propose a method for constructing a web corpus as a support for crowdsourcing examples of MWE occurrences. The resource is available under an open license and paves the way towards a large-scale dialect-aware language resource construction, which should prove useful in both traditional and novel NLP applications.
Automatic Term Extraction (ATE) or Recognition (ATR) is a fundamental processing step preceding many complex knowledge engineering tasks. However, few methods have been implemented as public tools and in particular, available as open-source freeware. Further, little effort is made to develop an adaptable and scalable framework that enables customization, development, and comparison of algorithms under a uniform environment. This paper introduces JATE 2.0, a complete remake of the free Java Automatic Term Extraction Toolkit (Zhang et al., 2008) delivering new features including: (1) highly modular, adaptable and scalable ATE thanks to integration with Apache Solr, the open source free-text indexing and search platform; (2) an extended collection of state-of-the-art algorithms. We carry out experiments on two well-known benchmarking datasets and compare the algorithms along the dimensions of effectiveness (precision) and efficiency (speed and memory consumption). To the best of our knowledge, this is by far the only free ATE library offering a flexible architecture and the most comprehensive collection of algorithms.
Synaesthesia is a type of metaphor associating linguistic expressions that refer to two different sensory modalities. Previous studies, based on the analysis of poetic texts, have shown that synaesthetic transfers tend to go from the lower toward the higher senses (e.g., sweet music vs. musical sweetness). In non-literary language synaesthesia is rare, and finding a sufficient number of examples manually would be too time-consuming. In order to verify whether the directionality also holds for conventional synaesthesia found in non-literary texts, an automatic procedure for the identification of instances of synaesthesia is therefore highly desirable. In this paper, we first focus on the preliminary step of this procedure, that is, the creation of a controlled lexicon of perception. Next, we present the results of a small pilot study that applies the extraction procedure to English and Italian corpus data.
The purpose of this paper is to introduce the TermoPL tool created to extract terminology from domain corpora in Polish. The program extracts noun phrases, term candidates, with the help of a simple grammar that can be adapted for user’s needs. It applies the C-value method to rank term candidates being either the longest identified nominal phrases or their nested subphrases. The method operates on simplified base forms in order to unify morphological variants of terms and to recognize their contexts. We support the recognition of nested terms by word connection strength which allows us to eliminate truncated phrases from the top part of the term list. The program has an option to convert simplified forms of phrases into correct phrases in the nominal case. TermoPL accepts as input morphologically annotated and disambiguated domain texts and creates a list of terms, the top part of which comprises domain terminology. It can also compare two candidate term lists using three different coefficients showing asymmetry of term occurrences in this data.
This paper presents a novel gold standard of German noun-noun compounds (Ghost-NN) including 868 compounds annotated with corpus frequencies of the compounds and their constituents, productivity and ambiguity of the constituents, semantic relations between the constituents, and compositionality ratings of compound-constituent pairs. Moreover, a subset of the compounds containing 180 compounds is balanced for the productivity of the modifiers (distinguishing low/mid/high productivity) and the ambiguity of the heads (distinguishing between heads with 1, 2 and >2 senses
We introduce DeQue, a lexicon covering French complex prepositions (CPRE) like “à partir de” (from) and complex conjunctions (CCONJ) like “bien que” (although). The lexicon includes fine-grained linguistic description based on empirical evidence. We describe the general characteristics of CPRE and CCONJ in French, with special focus on syntactic ambiguity. Then, we list the selection criteria used to build the lexicon and the corpus-based methodology employed to collect entries. Finally, we quantify the ambiguity of each construction by annotating around 100 sentences randomly taken from the FRWaC. In addition to its theoretical value, the resource has many potential practical applications. We intend to employ DeQue for treebank annotation and to train a dependency parser that can takes complex constructions into account.
This paper summarizes the preliminary results of an ongoing survey on multiword resources carried out within the IC1207 Cost Action PARSEME (PARSing and Multi-word Expressions). Despite the availability of language resource catalogs and the inventory of multiword datasets on the SIGLEX-MWE website, multiword resources are scattered and difficult to find. In many cases, language resources such as corpora, treebanks, or lexical databases include multiwords as part of their data or take them into account in their annotations. However, these resources need to be centralized to make them accessible. The aim of this survey is to create a portal where researchers can easily find multiword(-aware) language resources for their research. We report on the design of the survey and analyze the data gathered so far. We also discuss the problems we have detected upon examination of the data as well as possible ways of enhancing the survey.
The goal of this work is to introduce CHILDES-MWE, which contains English CHILDES corpora automatically annotated with Multiword Expressions (MWEs) information. The result is a resource with almost 350,000 sentences annotated with more than 70,000 distinct MWEs of various types from both longitudinal and latitudinal corpora. This resource can be used for large scale language acquisition studies of how MWEs feature in child language. Focusing on compound nouns (CN), we then verify in a longitudinal study if there are differences in the distribution and compositionality of CNs in child-directed and child-produced sentences across ages. Moreover, using additional latitudinal data, we investigate if there are further differences in CN usage and in compositionality preferences. The results obtained for the child-produced sentences reflect CN distribution and compositionality in child-directed sentences.
While measuring the readability of texts has been a long-standing research topic, assessing the technicality of terms has only been addressed more recently and mostly for the English language. In this paper, we train a learning-to-rank model to determine a specialization degree for each term found in a given list. Since no training data for this task exist for French, we train our system with non-lexical features on English data, namely, the Consumer Health Vocabulary, then apply it to French. The features include the likelihood ratio of the term based on specialized and lay language models, and tests for containing morphologically complex words. The evaluation of this approach is conducted on 134 terms from the UMLS Metathesaurus and 868 terms from the Eugloss thesaurus. The Normalized Discounted Cumulative Gain obtained by our system is over 0.8 on both test sets. Besides, thanks to the learning-to-rank approach, adding morphological features to the language model features improves the results on the Eugloss thesaurus.
Collocations such as “heavy rain” or “make [a] decision”, are combinations of two elements where one (the base) is freely chosen, while the choice of the other (collocate) is restricted, depending on the base. Collocations present difficulties even to advanced language learners, who usually struggle to find the right collocate to express a particular meaning, e.g., both “heavy” and “strong” express the meaning ‘intense’, but while “rain” selects “heavy”, “wind” selects “strong”. Lexical Functions (LFs) describe the meanings that hold between the elements of collocations, such as ‘intense’, ‘perform’, ‘create’, ‘increase’, etc. Language resources with semantically classified collocations would be of great help for students, however they are expensive to build, since they are manually constructed, and scarce. We present an unsupervised approach to the acquisition and semantic classification of collocations according to LFs, based on word embeddings in which, given an example of a collocation for each of the target LFs and a set of bases, the system retrieves a list of collocates for each base and LF.
By means of an online survey, we have investigated ways in which various types of multiword expressions are annotated in existing treebanks. The results indicate that there is considerable variation in treatments across treebanks and thereby also, to some extent, across languages and across theoretical frameworks. The comparison is focused on the annotation of light verb constructions and verbal idioms. The survey shows that the light verb constructions either get special annotations as such, or are treated as ordinary verbs, while VP idioms are handled through different strategies. Based on insights from our investigation, we propose some general guidelines for annotating multiword expressions in treebanks. The recommendations address the following application-based needs: distinguishing MWEs from similar but compositional constructions; searching distinct types of MWEs in treebanks; awareness of literal and nonliteral meanings; and normalization of the MWE representation. The cross-lingually and cross-theoretically focused survey is intended as an aid to accessing treebanks and an aid for further work on treebank annotation.
Multiword Expressions (MWEs) are used frequently in natural languages, but understanding the diversity in MWEs is one of the open problem in the area of Natural Language Processing. In the context of Indian languages, MWEs play an important role. In this paper, we present MWEs annotation dataset created for Indian languages viz., Hindi and Marathi. We extract possible MWE candidates using two repositories: 1) the POS-tagged corpus and 2) the IndoWordNet synsets. Annotation is done for two types of MWEs: compound nouns and light verb constructions. In the process of annotation, human annotators tag valid MWEs from these candidates based on the standard guidelines provided to them. We obtained 3178 compound nouns and 2556 light verb constructions in Hindi and 1003 compound nouns and 2416 light verb constructions in Marathi using two repositories mentioned before. This created resource is made available publicly and can be used as a gold standard for Hindi and Marathi MWE systems.
The question of how to compare languages and more generally the domain of linguistic typology, relies on the study of different linguistic properties or phenomena. Classically, such a comparison is done semi-manually, for example by extracting information from databases such as the WALS. However, it remains difficult to identify precisely regular parameters, available for different languages, that can be used as a basis towards modeling. We propose in this paper, focusing on the question of syntactic typology, a method for automatically extracting such parameters from treebanks, bringing them into a typology perspective. We present the method and the tools for inferring such information and navigating through the treebanks. The approach has been applied to 10 languages of the Universal Dependencies Treebank. We approach is evaluated by showing how automatic classification correlates with language families.
Wikipedia, the well known internet encyclopedia, is nowadays a widely used source of information. To leverage its rich information, already parsed versions of Wikipedia have been proposed. We present an annotated dump of the English Wikipedia. This dump draws upon previously released Wikipedia parsed dumps. Still, we head in a different direction. In this parse we focus more into the syntactical characteristics of words: aside from the classical Part-of-Speech (PoS) tags and dependency parsing relations, we provide the full constituent parse branch for each word in a succinct way. Additionally, we propose a hypergraph network representation of the extracted linguistic information. The proposed modelization aims to take advantage of the information stocked within our parsed Wikipedia dump. We hope that by releasing these resources, researchers from the concerned communities will have a ready-to-experiment Wikipedia corpus to compare and distribute their work. We render public our parsed Wikipedia dump as well as the tool (and its source code) used to perform the parse. The hypergraph network and its related metadata is also distributed.
To ensure portability of NLP systems across multiple domains, existing treebanks are often extended by adding trees from interesting domains that were not part of the initial annotation effort. In this paper, we will argue that it is both useful from an application viewpoint and enlightening from a linguistic viewpoint to detect and reduce divergence in annotation schemes between extant and new parts in a set of treebanks that is to be used in evaluation experiments. The results of our correction and harmonization efforts will be made available to the public as a test suite for the evaluation of constituent parsing.
The Persian Universal Dependency Treebank (Persian UD) is a recent effort of treebanking Persian with Universal Dependencies (UD), an ongoing project that designs unified and cross-linguistically valid grammatical representations including part-of-speech tags, morphological features, and dependency relations. The Persian UD is the converted version of the Uppsala Persian Dependency Treebank (UPDT) to the universal dependencies framework and consists of nearly 6,000 sentences and 152,871 word tokens with an average sentence length of 25 words. In addition to the universal dependencies syntactic annotation guidelines, the two treebanks differ in tokenization. All words containing unsegmented clitics (pronominal and copula clitics) annotated with complex labels in the UPDT have been separated from the clitics and appear with distinct labels in the Persian UD. The treebank has its original syntactic annotation scheme based on Stanford Typed Dependencies. In this paper, we present the approaches taken in the development of the Persian UD.
We present the French Question Bank, a treebank of 2600 questions. We show that classical parsing model performance drop while the inclusion of this data set is highly beneficial without harming the parsing of non-question data. when facing out-of- domain data with strong structural diver- gences. Two thirds being aligned with the QB (Judge et al., 2006) and being freely available, this treebank will prove useful to build robust NLP systems.
Many shallow natural language understanding tasks use dependency trees to extract relations between content words. However, strict surface-structure dependency trees tend to follow the linguistic structure of sentences too closely and frequently fail to provide direct relations between content words. To mitigate this problem, the original Stanford Dependencies representation also defines two dependency graph representations which contain additional and augmented relations that explicitly capture otherwise implicit relations between content words. In this paper, we revisit and extend these dependency graph representations in light of the recent Universal Dependencies (UD) initiative and provide a detailed account of an enhanced and an enhanced++ English UD representation. We further present a converter from constituency to basic, i.e., strict surface structure, UD trees, and a converter from basic UD trees to enhanced and enhanced++ English UD graphs. We release both converters as part of Stanford CoreNLP and the Stanford Parser.
This paper describes our efforts for the development of a Proposition Bank for Urdu, an Indo-Aryan language. Our primary goal is the labeling of syntactic nodes in the existing Urdu dependency Treebank with specific argument labels. In essence, it involves annotation of predicate argument structures of both simple and complex predicates in the Treebank corpus. We describe the overall process of building the PropBank of Urdu. We discuss various statistics pertaining to the Urdu PropBank and the issues which the annotators encountered while developing the PropBank. We also discuss how these challenges were addressed to successfully expand the PropBank corpus. While reporting the Inter-annotator agreement between the two annotators, we show that the annotators share similar understanding of the annotation guidelines and of the linguistic phenomena present in the language. The present size of this Propbank is around 180,000 tokens which is double-propbanked by the two annotators for simple predicates. Another 100,000 tokens have been annotated for complex predicates of Urdu.
We introduce a new member of the family of Prague dependency treebanks. The Czech Legal Text Treebank 1.0 is a morphologically and syntactically annotated corpus of 1,128 sentences. The treebank contains texts from the legal domain, namely the documents from the Collection of Laws of the Czech Republic. Legal texts differ from other domains in several language phenomena influenced by rather high frequency of very long sentences. A manual annotation of such sentences presents a new challenge. We describe a strategy and tools for this task. The resulting treebank can be explored in various ways. It can be downloaded from the LINDAT/CLARIN repository and viewed locally using the TrEd editor or it can be accessed on-line using the KonText and TreeQuery tools.
We present an attempt to link the large amount of different concept lists which are used in the linguistic literature, ranging from Swadesh lists in historical linguistics to naming tests in clinical studies and psycholinguistics. This resource, our Concepticon, links 30 222 concept labels from 160 conceptlists to 2495 concept sets. Each concept set is given a unique identifier, a unique label, and a human-readable definition. Concept sets are further structured by defining different relations between the concepts. The resource can be used for various purposes. Serving as a rich reference for new and existing databases in diachronic and synchronic linguistics, it allows researchers a quick access to studies on semantic change, cross-linguistic polysemies, and semantic associations.
In this study we elaborate a road map for the conversion of a traditional lexical syntactico-semantic resource for French into a linguistic linked open data (LLOD) model. Our approach uses current best-practices and the analyses of earlier similar undertakings (lemonUBY and PDEV-lemon) to tease out the most appropriate representation for our resource.
In knowledge bases where concepts have associated properties, there is a large amount of comparative information that is implicitly encoded in the values of the properties these concepts share. Although there have been previous approaches to generating riddles, none of them seem to take advantage of structured information stored in knowledge bases such as Thesaurus Rex, which organizes concepts according to the fine grained ad-hoc categories they are placed into by speakers in everyday language, along with associated properties or modifiers. Taking advantage of these shared properties, we have developed a riddle generator that creates riddles about concepts represented as common nouns. The base of these riddles are comparisons between the target concept and other entities that share some of its properties. In this paper, we describe the process we have followed to generate the riddles starting from the target concept and we show the results of the first evaluation we have carried out to test the quality of the resulting riddles.
The paper presents the strategy and results of mapping adjective synsets between plWordNet (the wordnet of Polish, cf. Piasecki et al. 2009, Maziarz et al. 2013) and Princeton WordNet (cf. Fellbaum 1998). The main challenge of this enterprise has been very different synset relation structures in the two networks: horizontal, dumbbell-model based in PWN and vertical, hyponymy-based in plWN. Moreover, the two wordnets display differences in the grouping of adjectives into semantic domains and in the size of the adjective category. The handle the above contrasts, a series of automatic prompt algorithms and a manual mapping procedure relying on corresponding synset and lexical unit relations as well as on inter-lingual relations between noun synsets were proposed in the pilot stage of mapping (Rudnicka et al. 2015). In the paper we discuss the final results of the mapping process as well as explain example mapping choices. Suggestions for further development of mapping are also given.
Recent research shows the importance of linking linguistic knowledge resources for the creation of large-scale linguistic data. We describe our approach for combining two English resources, FrameNet and sar-graphs, and illustrate the benefits of the linked data in a relation extraction setting. While FrameNet consists of schematic representations of situations, linked to lexemes and their valency patterns, sar-graphs are knowledge resources that connect semantic relations from factual knowledge graphs to the linguistic phrases used to express instances of these relations. We analyze the conceptual similarities and differences of both resources and propose to link sar-graphs and FrameNet on the levels of relations/frames as well as phrases. The former alignment involves a manual ontology mapping step, which allows us to extend sar-graphs with new phrase patterns from FrameNet. The phrase-level linking, on the other hand, is fully automatic. We investigate the quality of the automatically constructed links and identify two main classes of errors.
This paper presents the process of enriching the verb frame database of a Hungarian natural language parser to enable the assignment of semantic roles. We accomplished this by linking the parser’s verb frame database to existing linguistic resources such as VerbNet and WordNet, and automatically transferring back semantic knowledge. We developed OWL ontologies that map the various constraint description formalisms of the linked resources and employed a logical reasoning device to facilitate the linking procedure. We present results and discuss the challenges and pitfalls that arose from this undertaking.
Wikipedia has been increasingly used as a knowledge base for open-domain Named Entity Linking and Disambiguation. In this task, a dictionary with entity surface forms plays an important role in finding a set of candidate entities for the mentions in text. Existing dictionaries mostly rely on the Wikipedia link structure, like anchor texts, redirect links and disambiguation links. In this paper, we introduce a dictionary for Entity Linking that includes name variations extracted from Wikipedia article text, in addition to name variations derived from the Wikipedia link structure. With this approach, we show an increase in the coverage of entities and their mentions in the dictionary in comparison to other Wikipedia based dictionaries.
The Open Linguistics Working Group (OWLG) brings together researchers from various fields of linguistics, natural language processing, and information technology to present and discuss principles, case studies, and best practices for representing, publishing and linking linguistic data collections. A major outcome of our work is the Linguistic Linked Open Data (LLOD) cloud, an LOD (sub-)cloud of linguistic resources, which covers various linguistic databases, lexicons, corpora, terminologies, and metadata repositories. We present and summarize five years of progress on the development of the cloud and of advancements in open data in linguistics, and we describe recent community activities. The paper aims to serve as a guideline to orient and involve researchers with the community and/or Linguistic Linked Open Data.
Various lexical resources are being published in RDF. To enhance the usability of these resources, identical resources in different data sets should be linked. If lexical resources are described in different natural languages, then techniques to deal with multilinguality are required for interlinking. In this paper, we evaluate machine translation for interlinking concepts, i.e., generic entities named with a common noun or term. In our previous work, the evaluated method has been applied on named entities. We conduct two experiments involving different thesauri in different languages. The first experiment involves concepts from the TheSoz multilingual thesaurus in three languages: English, French and German. The second experiment involves concepts from the EuroVoc and AGROVOC thesauri in English and Chinese respectively. Our results demonstrate that machine translation can be beneficial for cross-lingual thesauri interlinking independently of a dataset structure.
Language Resources (LRs) are an essential ingredient of current approaches in Linguistics, Computational Linguistics, Language Technology and related fields. LRs are collections of spoken or written language data, typically annotated with linguistic analysis information. Different types of LRs exist, for example, corpora, ontologies, lexicons, collections of spoken language data (audio), or collections that also include video (multimedia, multimodal). Often, LRs are distributed with specific tools, documentation, manuals or research publications. The different phases that involve creating and distributing an LR can be conceptualised as a life cycle. While the idea of handling the LR production and maintenance process in terms of a life cycle has been brought up quite some time ago, a best practice model or common approach can still be considered a research gap. This article wants to help fill this gap by proposing an initial version of a generic Language Resource Life Cycle that can be used to inform, direct, control and evaluate LR research and development activities (including description, management, production, validation and evaluation workflows).
Everyday meals are an important part of our daily lives and, currently, there are many Internet sites that help us plan these meals. Allied to the growth in the amount of food data such as recipes available on the Internet is an increase in the number of studies on these data, such as recipe analysis and recipe search. However, there are few publicly available resources for food research; those that do exist do not include a wide range of food data or any meal data (that is, likely combinations of recipes). In this study, we construct a large-scale recipe and meal data collection as the underlying infrastructure to promote food research. Our corpus consists of approximately 1.7 million recipes and 36000 meals in cookpad, one of the largest recipe sites in the world. We made the corpus available to researchers in February 2015 and as of February 2016, 82 research groups at 56 universities have made use of it to enhance their studies.
Although there are many tools for natural language processing tasks in Estonian, these tools are very loosely interoperable, and it is not easy to build practical applications on top of them. In this paper, we introduce a new Python library for natural language processing in Estonian, which provides unified programming interface for various NLP components. The EstNLTK toolkit provides utilities for basic NLP tasks including tokenization, morphological analysis, lemmatisation and named entity recognition as well as offers more advanced features such as a clause segmentation, temporal expression extraction and normalization, verb chain detection, Estonian Wordnet integration and rule-based information extraction. Accompanied by a detailed API documentation and comprehensive tutorials, EstNLTK is suitable for a wide range of audience. We believe EstNLTK is mature enough to be used for developing NLP-backed systems both in industry and research. EstNLTK is freely available under the GNU GPL version 2+ license, which is standard for academic software.
This presentation introduces the imminent establishment of a new language resource infrastructure focusing on languages spoken in Southern Africa, with an eventual aim to become a hub for digital language resources within Sub-Saharan Africa. The Constitution of South Africa makes provision for 11 official languages all with equal status. The current language Resource Management Agency will be merged with the new Centre, which will have a wider focus than that of data acquisition, management and distribution. The Centre will entertain two main programs: Digitisation and Digital Humanities. The digitisation program will focus on the systematic digitisation of relevant text, speech and multi-modal data across the official languages. Relevancy will be determined by a Scientific Advisory Board. This will take place on a continuous basis through specified projects allocated to national members of the Centre, as well as through open-calls aimed at the academic as well as local communities. The digital resources will be managed and distributed through a dedicated web-based portal. The development of the Digital Humanities program will entail extensive academic support for projects implementing digital language based data. The Centre will function as an enabling research infrastructure primarily supported by national government and hosted by the North-West University.
In this paper, we report on the design and development of an online search platform for the MERLIN corpus of learner texts in Czech, German and Italian. It was created in the context of the MERLIN project, which aims at empirically illustrating features of the Common European Framework of Reference (CEFR) for evaluating language competences based on authentic learner text productions compiled into a learner corpus. Furthermore, the project aims at providing access to the corpus through a search interface adapted to the needs of multifaceted target groups involved with language learning and teaching. This article starts by providing a brief overview on the project ambition, the data resource and its intended target groups. Subsequently, the main focus of the article is on the design and development process of the platform, which is carried out in a user-centred fashion. The paper presents the user studies carried out to collect requirements, details the resulting decisions concerning the platform design and its implementation, and reports on the evaluation of the platform prototype and final adjustments.
Language resources are valuable assets, both for institutions and researchers. To safeguard these resources requirements for repository systems and data management have been specified by various branch organizations, e.g., CLARIN and the Data Seal of Approval. This paper describes these and some additional ones posed by the authors’ home institutions. And it shows how they are met by FLAT, to provide a new home for language resources. The basis of FLAT is formed by the Fedora Commons repository system. This repository system can meet many of the requirements out-of-the box, but still additional configuration and some development work is needed to meet the remaining ones, e.g., to add support for Handles and Component Metadata. This paper describes design decisions taken in the construction of FLAT’s system architecture via a mix-and-match strategy, with a preference for the reuse of existing solutions. FLAT is developed and used by the Meertens Institute and The Language Archive, but is also freely available for anyone in need of a CLARIN-compliant repository for their language resources.
I introduce CLARIAH in the Netherlands, which aims to contribute the Netherlands part of a Europe-wide humanities research infrastructure. I describe the digital turn in the humanities, the background and context of CLARIAH, both nationally and internationally, its relation to the CLARIN and DARIAH infrastructures, and the rationale for joining forces between CLARIN and DARIAH in the Netherlands. I also describe the first results of joining forces as achieved in the CLARIAH-SEED project, and the plans of the CLARIAH-CORE project, which is currently running
The Component MetaData Infrastructure (CMDI) is a framework for the creation and usage of metadata formats to describe all kinds of resources in the CLARIN world. To better connect to the library world, and to allow librarians to enter metadata for linguistic resources into their catalogues, a crosswalk from CMDI-based formats to bibliographic standards is required. The general and rather fluid nature of CMDI, however, makes it hard to map arbitrary CMDI schemas to metadata standards such as Dublin Core (DC) or MARC 21, which have a mature, well-defined and fixed set of field descriptors. In this paper, we address the issue and propose crosswalks between CMDI-based profiles originating from the NaLiDa project and DC and MARC 21, respectively.
Data management plans, data sharing plans and the like are now required by funders worldwide as part of research proposals. Concerned with promoting the notion of open scientific data, funders view such plans as the framework for satisfying the generally accepted requirements for data generated in funded research projects, among them that it be accessible, usable, standardized to the degree possible, secure and stable. This paper examines the origins of data management plans, their requirements and issues they raise for data centers and HLT resource development in general.
We introduce JCoRe 2.0, the relaunch of a UIMA-based open software repository for full-scale natural language processing originating from the Jena University Language & Information Engineering (JULIE) Lab. In an attempt to put the new release of JCoRe on firm software engineering ground, we uploaded it to GitHub, a social coding platform, with an underlying source code versioning system and various means to support collaboration for software development and code modification management. In order to automate the builds of complex NLP pipelines and properly represent and track dependencies of the underlying Java code, we incorporated Maven as part of our software configuration management efforts. In the meantime, we have deployed our artifacts on Maven Central, as well. JCoRe 2.0 offers a broad range of text analytics functionality (mostly) for English-language scientific abstracts and full-text articles, especially from the life sciences domain.
The issue for CLARIN archives at the metadata level is to facilitate the user’s possibility to describe their data, even with their own standard, and at the same time make these metadata meaningful for a variety of users with a variety of resource types, and ensure that the metadata are useful for search across all resources both at the national and at the European level. We see that different people from different research communities fill in the metadata in different ways even though the metadata was defined and documented. This has impacted when the metadata are harvested and displayed in different environments. A loss of information is at stake. In this paper we view the challenges of ensuring metadata interoperability through examples of propagation of metadata values from the CLARIN-DK archive to the VLO. We see that the CLARIN Community in many ways support interoperability, but argue that agreeing upon standards, making clear definitions of the semantics of the metadata and their content is inevitable for the interoperability to work successfully. The key points are clear and freely available definitions, accessible documentation and easily usable facilities and guidelines for the metadata creators.
The article describes the current status of a large national project, CoRoLa, aiming at building a reference corpus for the contemporary Romanian language. Unlike many other national corpora, CoRoLa contains only - IPR cleared texts and speech data, obtained from some of the country’s most representative publishing houses, broadcasting agencies, editorial offices, newspapers and popular bloggers. For the written component 500 million tokens are targeted and for the oral one 300 hours of recordings. The choice of texts is done according to their functional style, domain and subdomain, also with an eye to the international practice. A metadata file (following the CMDI model) is associated to each text file. Collected texts are cleaned and transformed in a format compatible with the tools for automatic processing (segmentation, tokenization, lemmatization, part-of-speech tagging). The paper also presents up-to-date statistics about the structure of the corpus almost two years before its official launching. The corpus will be freely available for searching. Users will be able to download the results of their searches and those original files when not against stipulations in the protocols we have with text providers.
The paper concentrates on the design, composition and annotation of SYN2015, a new 100-million representative corpus of contemporary written Czech. SYN2015 is a sequel of the representative corpora of the SYN series that can be described as traditional (as opposed to the web-crawled corpora), featuring cleared copyright issues, well-defined composition, reliability of annotation and high-quality text processing. At the same time, SYN2015 is designed as a reflection of the variety of written Czech text production with necessary methodological and technological enhancements that include a detailed bibliographic annotation and text classification based on an updated scheme. The corpus has been produced using a completely rebuilt text processing toolchain called SynKorp. SYN2015 is lemmatized, morphologically and syntactically annotated with state-of-the-art tools. It has been published within the framework of the Czech National Corpus and it is available via the standard corpus query interface KonText at http://kontext.korpus.cz as well as a dataset in shuffled format.
This proposal describes a new way to visualise resources in the LREMap, a community-built repository of language resource descriptions and uses. The LREMap is represented as a force-directed graph, where resources, papers and authors are nodes. The analysis of the visual representation of the underlying graph is used to study how the community gathers around LRs and how LRs are used in research.
Researchers in Natural Language Processing rely on availability of data and software, ideally under open licenses, but little is done to actively encourage it. In fact, the current Copyright framework grants exclusive rights to authors to copy their works, make them available to the public and make derivative works (such as annotated language corpora). Moreover, in the EU databases are protected against unauthorized extraction and re-utilization of their contents. Therefore, proper public licensing plays a crucial role in providing access to research data. A public license is a license that grants certain rights not to one particular user, but to the general public (everybody). Our article presents a tool that we developed and whose purpose is to assist the user in the licensing process. As software and data should be licensed under different licenses, the tool is composed of two separate parts: Data and Software. The underlying logic as well as elements of the graphic interface are presented below.
The Information System for Syntactic and Semantic Analysis of the Lithuanian language (lith. Lietuvių kalbos sintaksinės ir semantinės analizės informacinė sistema, LKSSAIS) is the first infrastructure for the Lithuanian language combining Lithuanian language tools and resources for diverse linguistic research and applications tasks. It provides access to the basic as well as advanced natural language processing tools and resources, including tools for corpus creation and management, text preprocessing and annotation, ontology building, named entity recognition, morphosyntactic and semantic analysis, sentiment analysis, etc. It is an important platform for researchers and developers in the field of natural language technology.
We introduce CODE ALLTAG, a text corpus composed of German-language e-mails. It is divided into two partitions: the first of these portions, CODE ALLTAG_XL, consists of a bulk-size collection drawn from an openly accessible e-mail archive (roughly 1.5M e-mails), whereas the second portion, CODE ALLTAG_S+d, is much smaller in size (less than thousand e-mails), yet excels with demographic data from each author of an e-mail. CODE ALLTAG, thus, currently constitutes the largest E-Mail corpus ever built. In this paper, we describe, for both parts, the solicitation process for gathering e-mails, present descriptive statistical properties of the corpus, and, for CODE ALLTAG_S+d, reveal a compilation of demographic features of the donors of e-mails.
The Low Resource Language research conducted under DARPA’s Broad Operational Language Translation (BOLT) program required the rapid creation of text corpora of typologically diverse languages (Turkish, Hausa, and Uzbek) which were annotated with morphological information, along with other types of annotation. Since the output of morphological analyzers is a significant aid to morphological annotation, we developed a morphological analyzer for each language in order to support the annotation task, and also as a deliverable by itself. Our framework for analyzer creation results in tables similar to those used in the successful SAMA analyzer for Arabic, but with a more abstract linguistic level, from which the tables are derived. A lexicon was developed from available resources for integration with the analyzer, and given the speed of development and uncertain coverage of the lexicon, we assumed that the analyzer would necessarily be lacking in some coverage for the project annotation. Our analyzer framework was therefore focused on rapid implementation of the key structures of the language, together with accepting “wildcard” solutions as possible analyses for a word with an unknown stem, building upon our similar experiences with morphological annotation with Modern Standard Arabic and Egyptian Arabic.
We propose a novel neural lemmatization model which is language independent and supervised in nature. To handle the words in a neural framework, word embedding technique is used to represent words as vectors. The proposed lemmatizer makes use of contextual information of the surface word to be lemmatized. Given a word along with its contextual neighbours as input, the model is designed to produce the lemma of the concerned word as output. We introduce a new network architecture that permits only dimension specific connections between the input and the output layer of the model. For the present work, Bengali is taken as the reference language. Two datasets are prepared for training and testing purpose consisting of 19,159 and 2,126 instances respectively. As Bengali is a resource scarce language, these datasets would be beneficial for the respective research community. Evaluation method shows that the neural lemmatizer achieves 69.57% accuracy on the test dataset and outperforms the simple cosine similarity based baseline strategy by a margin of 1.37%.
~This paper describes the development of free/open-source finite-state morphological transducers for Tuvan, a Turkic language spoken in and around the Tuvan Republic in Russia. The finite-state toolkit used for the work is the Helsinki Finite-State Toolkit (HFST), we use the lexc formalism for modelling the morphotactics and twol formalism for modelling morphophonological alternations. We present a novel description of the morphological combinatorics of pseudo-derivational morphemes in Tuvan. An evaluation is presented which shows that the transducer has a reasonable coverage―around 93%―on freely-available corpora of the languages, and high precision―over 99%―on a manually verified test set.
We describe an extensive and versatile lexical resource for Latvian, an under-resourced Indo-European language, which we call Tezaurs (Latvian for ‘thesaurus’). It comprises a large explanatory dictionary of more than 250,000 entries that are derived from more than 280 external sources. The dictionary is enriched with phonetic, morphological, semantic and other annotations, as well as augmented by various language processing tools allowing for the generation of inflectional forms and pronunciation, for on-the-fly selection of corpus examples, for suggesting synonyms, etc. Tezaurs is available as a public and widely used web application for end-users, as an open data set for the use in language technology (LT), and as an API ― a set of web services for the integration into third-party applications. The ultimate goal of Tezaurs is to be the central computational lexicon for Latvian, bringing together all Latvian words and frequently used multi-word units and allowing for the integration of other LT resources and tools.
Morphological analysis is a fundamental task in natural-language processing, which is used in other NLP applications such as part-of-speech tagging, syntactic parsing, information retrieval, machine translation, etc. In this paper, we present our work on the development of free/open-source finite-state morphological analyser for Sindhi. We have used Apertium’s lttoolbox as our finite-state toolkit to implement the transducer. The system is developed using a paradigm-based approach, wherein a paradigm defines all the word forms and their morphological features for a given stem (lemma). We have evaluated our system on the Sindhi Wikipedia corpus and achieved a reasonable coverage of 81% and a precision of over 97%.
This paper presents a semi-automatic method to derive morphological analyzers from a limited number of example inflections suitable for languages with alphabetic writing systems. The system we present learns the inflectional behavior of morphological paradigms from examples and converts the learned paradigms into a finite-state transducer that is able to map inflected forms of previously unseen words into lemmas and corresponding morphosyntactic descriptions. We evaluate the system when provided with inflection tables for several languages collected from the Wiktionary.
We report on the implementation of a morphological analyzer for the Sahidic dialect of Coptic, a now extinct Afro-Asiatic language. The system is developed in the finite-state paradigm. The main purpose of the project is provide a method by which scholars and linguists can semi-automatically gloss extant texts written in Sahidic. Since a complete lexicon containing all attested forms in different manuscripts requires significant expertise in Coptic spanning almost 1,000 years, we have equipped the analyzer with a core lexicon and extended it with a “guesser” ability to capture out-of-vocabulary items in any inflection. We also suggest an ASCII transliteration for the language. A brief evaluation is provided.
We present the new online edition of a dictionary of Polish inflection ― the Grammatical Dictionary of Polish (http://sgjp.pl). The dictionary is interesting for several reasons: it is comprehensive (over 330,000 lexemes corresponding to almost 4,300,000 different textual words; 1116 handcrafted inflectional patterns), the inflection is presented in an explicit manner in the form of carefully designed tables, the user interface facilitates advanced queries by several features (lemmas, forms, applicable grammatical categories, types of inflection). Moreover, the data of the dictionary is used in morphological analysers, including our product Morfeusz (http://sgjp. pl/morfeusz). From the start, the dictionary was meant to be comfortable for the human reader as well as to be ready for use in NLP applications. In the paper we briefly discuss both aspects of the resource.
This paper presents a collection of 350,000 German lemmatised words, rated on four psycholinguistic affective attributes. All ratings were obtained via a supervised learning algorithm that can automatically calculate a numerical rating of a word. We applied this algorithm to abstractness, arousal, imageability and valence. Comparison with human ratings reveals high correlation across all rating types. The full resource is publically available at: http://www.ims.uni-stuttgart.de/data/affective_norms/
Despite a centuries-long tradition in lexicography, Latin lacks state-of-the-art computational lexical resources. This situation is strictly related to the still quite limited amount of linguistically annotated textual data for Latin, which can help the building of new lexical resources by supporting them with empirical evidence. However, projects for creating new language resources for Latin have been launched over the last decade to fill this gap. In this paper, we present Latin Vallex, a valency lexicon for Latin built in mutual connection with the semantic and pragmatic annotation of two Latin treebanks featuring texts of different eras. On the one hand, such a connection between the empirical evidence provided by the treebanks and the lexicon allows to enhance each frame entry in the lexicon with its frequency in real data. On the other hand, each valency-capable word in the treebanks is linked to a frame entry in the lexicon.
Given lexical-semantic resources in different languages, it is useful to establish cross-lingual correspondences, preferably with semantic relation labels, between the concept nodes in these resources. This paper presents a framework for enabling a cross-lingual/node-wise alignment of lexical-semantic resources, where cross-lingual correspondence candidates are first discovered and ranked, and then classified by a succeeding module. Indeed, we propose that a two-tier classifier configuration is feasible for the second module: the first classifier filters out possibly irrelevant correspondence candidates and the second classifier assigns a relatively fine-grained semantic relation label to each of the surviving candidates. The results of Japanese-to-English alignment experiments using EDR Electronic Dictionary and Princeton WordNet are described to exemplify the validity of the proposal.
The last two decades have seen the development of various semantic lexical resources such as WordNet (Miller, 1995) and the USAS semantic lexicon (Rayson et al., 2004), which have played an important role in the areas of natural language processing and corpus-based studies. Recently, increasing efforts have been devoted to extending the semantic frameworks of existing lexical knowledge resources to cover more languages, such as EuroWordNet and Global WordNet. In this paper, we report on the construction of large-scale multilingual semantic lexicons for twelve languages, which employ the unified Lancaster semantic taxonomy and provide a multilingual lexical knowledge base for the automatic UCREL semantic annotation system (USAS). Our work contributes towards the goal of constructing larger-scale and higher-quality multilingual semantic lexical resources and developing corpus annotation tools based on them. Lexical coverage is an important factor concerning the quality of the lexicons and the performance of the corpus annotation tools, and in this experiment we focus on evaluating the lexical coverage achieved by the multilingual lexicons and semantic annotation tools based on them. Our evaluation shows that some semantic lexicons such as those for Finnish and Italian have achieved lexical coverage of over 90% while others need further expansion.
We present the dict_to_4lang tool for processing entries of three monolingual dictionaries of English and mapping definitions to concept graphs following the 4lang principles of semantic representation introduced by (Kornai, 2010). 4lang representations are domain- and language-independent, and make use of only a very limited set of primitives to encode the meaning of all utterances. Our pipeline relies on the Stanford Dependency Parser for syntactic analysis, the dep to 4lang module then builds directed graphs of concepts based on dependency relations between words in each definition. Several issues are handled by construction-specific rules that are applied to the output of dep_to_4lang. Manual evaluation suggests that ca. 75% of graphs built from the Longman Dictionary are either entirely correct or contain only minor errors. dict_to_4lang is available under an MIT license as part of the 4lang library and has been used successfully in measuring Semantic Textual Similarity (Recski and Ács, 2015). An interactive demo of core 4lang functionalities is available at http://4lang.hlt.bme.hu.
This article presents the semantic layer of Walenty―a new valence dictionary of Polish predicates, with a number of novel features, as compared to other such dictionaries. The dictionary contains two layers, syntactic and semantic. The syntactic layer describes syntactic and morphosyntactic constraints predicates put on their dependants. In particular, it includes a comprehensive and powerful phraseological component. The semantic layer shows how predicates and their arguments are involved in a described situation in an utterance. These two layers are connected, representing how semantic arguments can be realised on the surface. Each syntactic schema and each semantic frame are illustrated by at least one exemplary sentence attested in linguistic reality. The semantic layer consists of semantic frames represented as lists of pairs and connected with PlWordNet lexical units. Semantic roles have a two-level representation (basic roles are provided with an attribute) enabling representation of arguments in a flexible way. Selectional preferences are based on PlWordNet structure as well.
This paper proposes a new method for Italian verb classification -and a preliminary example of resulting classes- inspired by Levin (1993) and VerbNet (Kipper-Schuler, 2005), yet partially independent from these resources; we achieved such a result by integrating Levin and VerbNet’s models of classification with other theoretic frameworks and resources. The classification is rooted in the constructionist framework (Goldberg, 1995; 2006) and is distribution-based. It is also semantically characterized by a link to FrameNet’ssemanticframesto represent the event expressed by a class. However, the new Italian classes maintain the hierarchic “tree” structure and monotonic nature of VerbNet’s classes, and, where possible, the original names (e.g.: Verbs of Killing, Verbs of Putting, etc.). We therefore propose here a taxonomy compatible with VerbNet but at the same time adapted to Italian syntax and semantics. It also addresses a number of problems intrinsic to the original classifications, such as the role of argument alternations, here regarded simply as epiphenomena, consistently with the constructionist approach.
Patients are often exposed to medical terms, such as anosognosia, myelodysplastic, or hepatojejunostomy, that can be semantically complex and hardly understandable by non-experts in medicine. Hence, it is important to assess which words are potentially non-understandable and require further explanations. The purpose of our work is to build specific lexicon in which the words are rated according to whether they are understandable or non-understandable. We propose to work with medical words in French such as provided by an international medical terminology. The terms are segmented in single words and then each word is manually processed by three annotators. The objective is to assign each word into one of the three categories: I can understand, I am not sure, I cannot understand. The annotators do not have medical training nor they present specific medical problems. They are supposed to represent an average patient. The inter-annotator agreement is then computed. The content of the categories is analyzed. Possible applications in which this lexicon can be helpful are proposed and discussed. The rated lexicon is freely available for the research purposes. It is accessible online at http://natalia.grabar.perso.sfr.fr/rated-lexicon.html
Word sense embeddings represent a word sense as a low-dimensional numeric vector. While this representation is potentially useful for NLP applications, its interpretability is inherently limited. We propose a simple technique that improves interpretability of sense vectors by mapping them to synsets of a lexical resource. Our experiments with AdaGram sense embeddings and BabelNet synsets show that it is possible to retrieve synsets that correspond to automatically learned sense vectors with Precision of 0.87, Recall of 0.42 and AUC of 0.78.
This paper presents a lexical resource developed for Portuguese. The resource contains sentences annotated with semantic roles. The sentences were extracted from two domains: Cardiology research papers and newspaper articles. Both corpora were analyzed with the PALAVRAS parser and subsequently processed with a subcategorization frames extractor, so that each sentence that contained at least one main verb was stored in a database together with its syntactic organization. The annotation was manually carried out by a linguist using an annotation interface. Both the annotated and non-annotated data were exported to an XML format, which is readily available for download. The reason behind exporting non-annotated data is that there is syntactic information collected from the parser annotation in the non-annotated data, and this could be useful for other researchers. The sentences from both corpora were annotated separately, so that it is possible to access sentences either from the Cardiology or from the newspaper corpus. The full resource presents more than seven thousand semantically annotated sentences, containing 192 different verbs and more than 15 thousand individual arguments and adjuncts.
This paper presents the Predicate Matrix 1.3, a lexical resource resulting from the integration of multiple sources of predicate information including FrameNet, VerbNet, PropBank and WordNet. This new version of the Predicate Matrix has been extended to cover nominal predicates by adding mappings to NomBank. Similarly, we have integrated resources in Spanish, Catalan and Basque. As a result, the Predicate Matrix 1.3 provides a multilingual lexicon to allow interoperable semantic analysis in multiple languages.
We present a gold standard for evaluating scale membership and the order of scalar adjectives. In addition to evaluating existing methods of ordering adjectives, this knowledge will aid in studying the organization of adjectives in the lexicon. This resource is the result of two elicitation tasks conducted with informants from Amazon Mechanical Turk. The first task is notable for gathering open-ended lexical data from informants. The data is analyzed using Cultural Consensus Theory, a framework from anthropology, to not only determine scale membership but also the level of consensus among the informants (Romney et al., 1986). The second task gathers a culturally salient ordering of the words determined to be members. We use this method to produce 12 scales of adjectives for use in evaluation.
In this paper we describe VerbCROcean, a broad-coverage repository of fine-grained semantic relations between Croatian verbs. Adopting the methodology of Chklovski and Pantel (2004) used for acquiring the English VerbOcean, we first acquire semantically related verb pairs from a web corpus hrWaC by relying on distributional similarity of subject-verb-object paths in the dependency trees. We then classify the semantic relations between each pair of verbs as similarity, intensity, antonymy, or happens-before, using a number of manually-constructed lexico-syntatic patterns. We evaluate the quality of the resulting resource on a manually annotated sample of 1000 semantic verb relations. The evaluation revealed that the predictions are most accurate for the similarity relation, and least accurate for the intensity relation. We make available two variants of VerbCROcean: a coverage-oriented version, containing about 36k verb pairs at a precision of 41%, and a precision-oriented version containing about 5k verb pairs, at a precision of 56%.
In this article we present an exploratory approach to enrich a WordNet-like lexical ontology with the synonyms present in a standard monolingual Portuguese dictionary. The dictionary was converted from PDF into XML and senses were automatically identified and annotated. This allowed us to extract them, independently of definitions, and to create sets of synonyms (synsets). These synsets were then aligned with WordNet synsets, both in the same language (Portuguese) and projecting the Portuguese terms into English, Spanish and Galician. This process allowed both the addition of new term variants to existing synsets, as to create new synsets for Portuguese.
Collecting data for sentiment analysis in resource-limited languages carries a significant risk of sample selection bias, since the small quantities of available data are most likely not representative of the whole population. Ignoring this bias leads to less robust machine learning classifiers and less reliable evaluation results. In this paper we present a dataset balancing algorithm that minimizes the sample selection bias by eliminating irrelevant systematic differences between the sentiment classes. We prove its superiority over the random sampling method and we use it to create the Serbian movie review dataset ― SerbMR ― the first balanced and topically uniform sentiment analysis dataset in Serbian. In addition, we propose an incremental way of finding the optimal combination of simple text processing options and machine learning features for sentiment classification. Several popular classifiers are used in conjunction with this evaluation approach in order to establish strong but reliable baselines for sentiment analysis in Serbian.
This paper introduces the augmented NTU sentiment dictionary, abbreviated as ANTUSD, which is constructed by collecting sentiment stats of words in several sentiment annotation work. A total of 26,021 words were collected in ANTUSD. For each word, the CopeOpi numerical sentiment score and the number of positive annotation, neutral annotation, negative annotation, non-opinionated annotation, and not-a-word annotation are provided. Words and their sentiment information in ANTUSD have been linked to the Chinese ontology E-HowNet to provide rich semantic information. We demonstrate the usage of ANTUSD in polarity classification of words, and the results show that a superior f-score 98.21 is achieved, which supports the usefulness of the ANTUSD. ANTUSD can be freely obtained through application from NLPSA lab, Academia Sinica: http://academiasinicanlplab.github.io/
Due to the phenomenal growth of online product reviews, sentiment analysis (SA) has gained huge attention, for example, by online service providers. A number of benchmark datasets for a wide range of domains have been made available for sentiment analysis, especially in resource-rich languages. In this paper we assess the challenges of SA in Hindi by providing a benchmark setup, where we create an annotated dataset of high quality, build machine learning models for sentiment analysis in order to show the effective usage of the dataset, and finally make the resource available to the community for further advancement of research. The dataset comprises of Hindi product reviews crawled from various online sources. Each sentence of the review is annotated with aspect term and its associated sentiment. As classification algorithms we use Conditional Random Filed (CRF) and Support Vector Machine (SVM) for aspect term extraction and sentiment analysis, respectively. Evaluation results show the average F-measure of 41.07% for aspect term extraction and accuracy of 54.05% for sentiment classification.
This paper deals with building linguistic resources for Gulf Arabic, one of the Arabic variations, for sentiment analysis task using machine learning. To our knowledge, no previous works were done for Gulf Arabic sentiment analysis despite the fact that it is present in different online platforms. Hence, the first challenge is the absence of annotated data and sentiment lexicons. To fill this gap, we created these two main linguistic resources. Then we conducted different experiments: use Naive Bayes classifier without any lexicon; add a sentiment lexicon designed basically for MSA; use only the compiled Gulf Arabic sentiment lexicon and finally use both MSA and Gulf Arabic sentiment lexicons. The Gulf Arabic lexicon gives a good improvement of the classifier accuracy (90.54 %) over a baseline that does not use the lexicon (82.81%), while the MSA lexicon causes the accuracy to drop to (76.83%). Moreover, mixing MSA and Gulf Arabic lexicons causes the accuracy to drop to (84.94%) compared to using only Gulf Arabic lexicon. This indicates that it is useless to use MSA resources to deal with Gulf Arabic due to the considerable differences and conflicting structures between these two languages.
Sentiment shifters, i.e., words and expressions that can affect text polarity, play an important role in opinion mining. However, the limited ability of current automated opinion mining systems to handle shifters represents a major challenge. The majority of existing approaches rely on a manual list of shifters; few attempts have been made to automatically identify shifters in text. Most of them just focus on negating shifters. This paper presents a novel and efficient semi-automatic method for identifying sentiment shifters in drug reviews, aiming at improving the overall accuracy of opinion mining systems. To this end, we use weighted association rule mining (WARM), a well-known data mining technique, for finding frequent dependency patterns representing sentiment shifters from a domain-specific corpus. These patterns that include different kinds of shifter words such as shifter verbs and quantifiers are able to handle both local and long-distance shifters. We also combine these patterns with a lexicon-based approach for the polarity classification task. Experiments on drug reviews demonstrate that extracted shifters can improve the precision of the lexicon-based approach for polarity classification 9.25 percent.
This paper describes the STAC resource, a corpus of multi-party chats annotated for discourse structure in the style of SDRT (Asher and Lascarides, 2003; Lascarides and Asher, 2009). The main goal of the STAC project is to study the discourse structure of multi-party dialogues in order to understand the linguistic strategies adopted by interlocutors to achieve their conversational goals, especially when these goals are opposed. The STAC corpus is not only a rich source of data on strategic conversation, but also the first corpus that we are aware of that provides full discourse structures for multi-party dialogues. It has other remarkable features that make it an interesting resource for other topics: interleaved threads, creative language, and interactions between linguistic and extra-linguistic contexts.
This paper presents an automatic corpus-based process to author an open-domain conversational strategy usable both in chatterbot systems and as a fallback strategy for out-of-domain human utterances. Our approach is implemented on a corpus of television drama subtitles. This system is used as a chatterbot system to collect a corpus of 41 open-domain textual dialogues with 27 human participants. The general capabilities of the system are studied through objective measures and subjective self-reports in terms of understandability, repetition and coherence of the system responses selected in reaction to human utterances. Subjective evaluations of the collected dialogues are presented with respect to amusement, engagement and enjoyability. The main factors influencing those dimensions in our chatterbot experiment are discussed.
In this study, we describe the use of back-channelling patterns extracted from a dialogue corpus as a mean to characterising text-based dialogue systems. Our goal was to provide system users with the feeling that they are interacting with distinct individuals rather than artificially created characters. An analysis of the corpus revealed that substantial difference exists among speakers regarding the usage patterns of back-channelling. The patterns consist of back-channelling frequency, types, and expressions. They were used for system characterisation. Implemented system characters were tested by asking users of the dialogue system to identify the source speakers in the corpus. Experimental results suggest that possibility of using back-channelling patterns alone to characterize the dialogue system in some cases even among the same age and gender groups.
Team word-guessing games where one player, the clue-giver, gives clues attempting to elicit a target-word from another player, the receiver, are a popular form of entertainment and also used for educational purposes. Creating an engaging computational agent capable of emulating a talented human clue-giver in a timed word-guessing game depends on the ability to provide effective clues (clues able to elicit a correct guess from a human receiver). There are many available web resources and databases that can be mined for the raw material for clues for target-words; however, a large number of those clues are unlikely to be able to elicit a correct guess from a human guesser. In this paper, we propose a method for automatically filtering a clue corpus for effective clues for an arbitrary target-word from a larger set of potential clues, using machine learning on a set of features of the clues, including point-wise mutual information between a clue’s constituent words and a clue’s target-word. The results of the experiments significantly improve the average clue quality over previous approaches, and bring quality rates in-line with measures of human clue quality derived from a corpus of human-human interactions. The paper also introduces the data used to develop this method; audio recordings of people making guesses after having heard the clues being spoken by a synthesized voice.
In this paper, a novel approach is proposed to automatically construct parallel discourse corpus for dialogue machine translation. Firstly, the parallel subtitle data and its corresponding monolingual movie script data are crawled and collected from Internet. Then tags such as speaker and discourse boundary from the script data are projected to its subtitle data via an information retrieval approach in order to map monolingual discourse to bilingual texts. We not only evaluate the mapping results, but also integrate speaker information into the translation. Experiments show our proposed method can achieve 81.79% and 98.64% accuracy on speaker and dialogue boundary annotation, and speaker-based language model adaptation can obtain around 0.5 BLEU points improvement in translation qualities. Finally, we publicly release around 100K parallel discourse data with manual speaker and dialogue boundary annotation.
Automatic evaluation of Machine Translation (MT) is typically approached by measuring similarity between the candidate MT and a human reference translation. An important limitation of existing evaluation systems is that they are unable to distinguish candidate-reference differences that arise due to acceptable linguistic variation from the differences induced by MT errors. In this paper we present a new metric, UPF-Cobalt, that addresses this issue by taking into consideration the syntactic contexts of candidate and reference words. The metric applies a penalty when the words are similar but the contexts in which they occur are not equivalent. In this way, Machine Translations (MTs) that are different from the human translation but still essentially correct are distinguished from those that share high number of words with the reference but alter the meaning of the sentence due to translation errors. The results show that the method proposed is indeed beneficial for automatic MT evaluation. We report experiments based on two different evaluation tasks with various types of manual quality assessment. The metric significantly outperforms state-of-the-art evaluation systems in varying evaluation settings.
The usual concern when opting for a rule-based or a hybrid machine translation (MT) system is how much effort is required to adapt the system to a different language pair or a new domain. In this paper, we describe a way of adapting an existing hybrid MT system to a new language pair, and show that such a system can outperform a standard phrase-based statistical machine translation system with an average of 10 persons/month of work. This is specifically important in the case of domain-specific MT for which there is not enough parallel data for training a statistical machine translation system.
Word translations arise in dictionary-like organization as well as via machine learning from corpora. The former is exemplified by Wiktionary, a crowd-sourced dictionary with editions in many languages. Ács et al. (2013) obtain word translations from Wiktionary with the pivot-based method, also called triangulation, that infers word translations in a pair of languages based on translations to other, typically better resourced ones called pivots. Triangulation may introduce noise if words in the pivot are polysemous. The reliability of each triangulated translation is basically estimated by the number of pivot languages (Tanaka et al 1994). Mikolov et al (2013) introduce a method for generating or scoring word translations. Translation is formalized as a linear mapping between distributed vector space models (VSM) of the two languages. VSMs are trained on monolingual data, while the mapping is learned in a supervised fashion, using a seed dictionary of some thousand word pairs. The mapping can be used to associate existing translations with a real-valued similarity score. This paper exploits human labor in Wiktionary combined with distributional information in VSMs. We train VSMs on gigaword corpora, and the linear translation mapping on direct (non-triangulated) Wiktionary pairs. This mapping is used to filter triangulated translations based on scores. The motivation is that scores by the mapping may be a smoother measure of merit than considering only the number of pivot for the triangle. We evaluate the scores against dictionaries extracted from parallel corpora (Tiedemann 2012). We show that linear translation really provides a more reliable method for triangle scoring than pivot count. The methods we use are language-independent, and the training data is easy to obtain for many languages. We chose the German-Hungarian pair for evaluation, in which the filtered triangles resulting from our experiments are the greatest freely available list of word translations we are aware of.
This paper reports on an experiment where 795 human participants answered to the questions taken from second language proficiency tests that were translated to their native language. The output of three machine translation systems and two different human translations were used as the test material. We classified the translation errors in the questions according to an error taxonomy and analyzed the participants’ response on the basis of the type and frequency of the translation errors. Through the analysis, we identified several types of errors that deteriorated most the accuracy of the participants’ answers, their confidence on the answers, and their overall evaluation of the translation quality.
Although it is commonly assumed that word sense disambiguation (WSD) should help to improve lexical choice and improve the quality of machine translation systems, how to successfully integrate word senses into such systems remains an unanswered question. Some successful approaches have involved reformulating either WSD or the word senses it produces, but work on using traditional word senses to improve machine translation have met with limited success. In this paper, we build upon previous work that experimented on including word senses as contextual features in maxent-based translation models. Training on a large, open-domain corpus (Europarl), we demonstrate that this aproach yields significant improvements in machine translation from English to Portuguese.
This paper reports SuperCAT, a corpus analysis toolkit. It is a radical extension of SubCAT, the Sublanguage Corpus Analysis Toolkit, from sublanguage analysis to corpus analysis in general. The idea behind SuperCAT is that representative corpora have no tendency towards closure―that is, they tend towards infinity. In contrast, non-representative corpora have a tendency towards closure―roughly, finiteness. SuperCAT focuses on general techniques for the quantitative description of the characteristics of any corpus (or other language sample), particularly concerning the characteristics of lexical distributions. Additionally, SuperCAT features a complete re-engineering of the previous SubCAT architecture.
The web data contains immense amount of data, hundreds of billion words are waiting to be extracted and used for language research. In this work we introduce our tool LanguageCrawl which allows NLP researchers to easily construct web-scale corpus from Common Crawl Archive: a petabyte scale, open repository of web crawl information. Three use-cases are presented: filtering Polish websites, building an N-gram corpora and training continuous skip-gram language model with hierarchical softmax. Each of them has been implemented within the LanguageCrawl toolkit, with the possibility to adjust specified language and N-gram ranks. Special effort has been put on high computing efficiency, by applying highly concurrent multitasking. We make our tool publicly available to enrich NLP resources. We strongly believe that our work will help to facilitate NLP research, especially in under-resourced languages, where the lack of appropriately sized corpora is a serious hindrance to applying data-intensive methods, such as deep neural networks.
The availability of large corpora for more and more languages enforces generic querying and standard interfaces. This development is especially relevant in the context of integrated research environments like CLARIN or DARIAH. The paper focuses on several applications and implementation details on the basis of a unified corpus format, a unique POS tag set, and prepared data for word similarities. All described data or applications are already or will be in the near future accessible via well-documented RESTful Web services. The target group are all kinds of interested persons with varying level of experience in programming or corpus query languages.
Several parallel corpora built from European Union language resources are presented here. They were processed by state-of-the-art tools and made available for researchers in the corpus manager Sketch Engine. A completely new resource is introduced: EUR-Lex Corpus, being one of the largest parallel corpus available at the moment, containing 840 million English tokens and the largest language pair English-French has more than 25 million aligned segments (paragraphs).
The present paper describes Corpus Query Lingua Franca (ISO CQLF), a specification designed at ISO Technical Committee 37 Subcommittee 4 “Language resource management” for the purpose of facilitating the comparison of properties of corpus query languages. We overview the motivation for this endeavour and present its aims and its general architecture. CQLF is intended as a multi-part specification; here, we concentrate on the basic metamodel that provides a frame that the other parts fit in.
The present paper describes the current release of the Bochum English Countability Lexicon (BECL 2.1), a large empirical database consisting of lemmata from Open ANC (http://www.anc.org) with added senses from WordNet (Fellbaum 1998). BECL 2.1 contains ≈ 11,800 annotated noun-sense pairs, divided in four major countability classes and 18 fine-grained subclasses. In the current version, BECL also provides information on nouns whose senses occur in more than one class allowing a closer look on polysemy and homonymy with regard to countability. Further included are sets of similar senses using the Leacock and Chodorow (LCH) score for semantic similarity (Leacock & Chodorow 1998), information on orthographic variation, on the completeness of all WordNet senses in the database and an annotated representation of different types of proper names. The further development of BECL will investigate the different countability classes of proper names and the general relation between semantic similarity and countability as well as recurring syntactic patterns for noun-sense pairs. The BECL 2.1 database is also publicly available via http://count-and-mass.org.
We propose a novel method for detecting optional arguments of Hungarian verbs using only positive data. We introduce a custom variant of collexeme analysis that explicitly models the noise in verb frames. Our method is, for the most part, unsupervised: we use the spectral clustering algorithm described in Brew and Schulte in Walde (2002) to build a noise model from a short, manually verified seed list of verbs. We experimented with both raw count- and context-based clusterings and found their performance almost identical. The code for our algorithm and the frame list are freely available at http://hlt.bme.hu/en/resources/tade.
We address the task of automatically correcting preposition errors in learners’ Dutch by modelling preposition usage in native language. Specifically, we build two models exploiting a large corpus of Dutch. The first is a binary model for detecting whether a preposition should be used at all in a given position or not. The second is a multiclass model for selecting the appropriate preposition in case one should be used. The models are tested on native as well as learners data. For the latter we exploit a crowdsourcing strategy to elicit native judgements. On native test data the models perform very well, showing that we can model preposition usage appropriately. However, the evaluation on learners’ data shows that while detecting that a given preposition is wrong is doable reasonably well, detecting the absence of a preposition is a lot more difficult. Observing such results and the data we deal with, we envisage various ways of improving performance, and report them in the final section of this article.
In this paper we describe 1) the process of converting a corpus of Dante Alighieri from a TEI XML format in to a pseudo-CoNLL format; 2) how a pos-tagger trained on modern Italian performs on Dante’s Italian 3) the performances of two different pos-taggers trained on the given corpus. We are making our conversion scripts and models available to the community. The two other models trained on the corpus performs reasonably well. The tool used for the conversion process might turn useful for bridging the gap between traditional digital humanities and modern NLP applications since the TEI original format is not usually suitable for being processed with standard NLP tools. We believe our work will serve both communities: the DH community will be able to tag new documents and the NLP world will have an easier way in converting existing documents to a standardized machine-readable format.
In this paper, we present a corpus of news blog conversations in Italian annotated with gold standard agreement/disagreement relations at message and sentence levels. This is the first resource of this kind in Italian. From the analysis of ADRs at the two levels emerged that agreement annotated at message level is consistent and generally reflected at sentence level, moreover, the argumentation structure of disagreement is more complex than agreement. The manual error analysis revealed that this resource is useful not only for the analysis of argumentation, but also for the detection of irony/sarcasm in online debates. The corpus and annotation tool are available for research purposes on request.
We introduce an approach based on using the dependency grammar representations of sentences to compute sentence similarity for extractive multi-document summarization. We adapt and investigate the effects of two untyped dependency tree kernels, which have originally been proposed for relation extraction, to the multi-document summarization problem. In addition, we propose a series of novel dependency grammar based kernels to better represent the syntactic and semantic similarities among the sentences. The proposed methods incorporate the type information of the dependency relations for sentence similarity calculation. To our knowledge, this is the first study that investigates using dependency tree based sentence similarity for multi-document summarization.
In this paper, we focus on the verb-particle (V-Prt) split construction in English and German and its difficulty for parsing and Machine Translation (MT). For German, we use an existing test suite of V-Prt split constructions, while for English, we build a new and comparable test suite from raw data. These two data sets are then used to perform an analysis of errors in dependency parsing, word-level alignment and MT, which arise from the discontinuous order in V-Prt split constructions. In the automatic alignments of parallel corpora, most of the particles align to NULL. These mis-alignments and the inability of phrase-based MT system to recover discontinuous phrases result in low quality translations of V-Prt split constructions both in English and German. However, our results show that the V-Prt split phrases are correctly parsed in 90% of cases, suggesting that syntactic-based MT should perform better on these constructions. We evaluate a syntactic-based MT system on German and compare its performance to the phrase-based system.
We report on the creation of a lexical resource for the identification of potentially unspecific or imprecise constructions in German requirements documentation from the car manufacturing industry. In requirements engineering, such expressions are called “weak words”: they are not sufficiently precise to ensure an unambiguous interpretation by the contractual partners, who for the definition of their cooperation, typically rely on specification documents (Melchisedech, 2000); an example are dimension adjectives, such as kurz or lang (‘short’, ‘long’) which need to be modified by adverbials indicating the exact duration, size etc. Contrary to standard practice in requirements engineering, where the identification of such weak words is merely based on stopword lists, we identify weak uses in context, by querying annotated text. The queries are part of the resource, as they define the conditions when a word use is weak. We evaluate the recognition of weak uses on our development corpus and on an unseen evaluation corpus, reaching stable F1-scores above 0.95.
The granularity of PolNet (Polish Wordnet) is the main theoretical issue discussed in the paper. We describe the latest extension of PolNet including valency information of simple verbs and noun-verb collocations using manual and machine-assisted methods. Valency is defined to include both semantic and syntactic selectional restrictions. We assume the valency structure of a verb to be an index of meaning. Consistently we consider it an attribute of a synset. Strict application of this principle results in fine granularity of the verb section of the wordnet. Considering valency as a distinctive feature of synsets was an essential step to transform the initial PolNet (first intended as a lexical ontology) into a lexicon-grammar. For the present refinement of PolNet we assume that the category of language register is a part of meaning. The totality of PolNet 2.0 synsets is being revised in order to split the PolNet 2.0 synsets that contain different register words into register-uniform sub-synsets. We completed this operation for synsets that were used as values of semantic roles. The operation augmented the number of considered synsets by 29%. In the paper we report an extension of the class of collocation-based verb synsets.
This paper presents C-WEP, the Collection of Writing Errors by Professionals Writers of German. It currently consists of 245 sentences with grammatical errors. All sentences are taken from published texts. All authors are professional writers with high skill levels with respect to German, the genres, and the topics. The purpose of this collection is to provide seeds for more sophisticated writing support tools as only a very small proportion of those errors can be detected by state-of-the-art checkers. C-WEP is annotated on various levels and freely available.
In this paper, we describe an addition to the corpus query system Kontext that enables to enhance the search using syntactic attributes in addition to the existing features, mainly lemmas and morphological categories. We present the enhancements of the corpus query system itself, the attributes we use to represent syntactic structures in data, and some examples of querying the syntactically annotated corpora, such as treebanks in various languages as well as an automatically parsed large corpus.
Starting from the English affective lexicon ANEW (Bradley and Lang, 1999a) we have created the first Greek affective lexicon. It contains human ratings for the three continuous affective dimensions of valence, arousal and dominance for 1034 words. The Greek affective lexicon is compared with affective lexica in English, Spanish and Portuguese. The lexicon is automatically expanded by selecting a small number of manually annotated words to bootstrap the process of estimating affective ratings of unknown words. We experimented with the parameters of the semantic-affective model in order to investigate their impact to its performance, which reaches 85% binary classification accuracy (positive vs. negative ratings). We share the Greek affective lexicon that consists of 1034 words and the automatically expanded Greek affective lexicon that contains 407K words.
In this paper we present a Hungarian sentiment corpus manually annotated at aspect level. Our corpus consists of Hungarian opinion texts written about different types of products. The main aim of creating the corpus was to produce an appropriate database providing possibilities for developing text mining software tools. The corpus is a unique Hungarian database: to the best of our knowledge, no digitized Hungarian sentiment corpus that is annotated on the level of fragments and targets has been made so far. In addition, many language elements of the corpus, relevant from the point of view of sentiment analysis, got distinct types of tags in the annotation. In this paper, on the one hand, we present the method of annotation, and we discuss the difficulties concerning text annotation process. On the other hand, we provide some quantitative and qualitative data on the corpus. We conclude with a description of the applicability of the corpus.
Sentiment analysis has so far focused on the detection of explicit opinions. However, of late implicit opinions have received broader attention, the key idea being that the evaluation of an event type by a speaker depends on how the participants in the event are valued and how the event itself affects the participants. We present an annotation scheme for adding relevant information, couched in terms of so-called effect functors, to German lexical items. Our scheme synthesizes and extends previous proposals. We report on an inter-annotator agreement study. We also present results of a crowdsourcing experiment to test the utility of some known and some new functors for opinion inference where, unlike in previous work, subjects are asked to reason from event evaluation to participant evaluation.
In this paper, a German verb resource for verb-centered sentiment inference is introduced and evaluated. Our model specifies verb polarity frames that capture the polarity effects on the fillers of the verb’s arguments given a sentence with that verb frame. Verb signatures and selectional restrictions are also part of the model. An algorithm to apply the verb resource to treebank sentences and the results of our first evaluation are discussed.
In this paper we present the TWitterBuonaScuola corpus (TW-BS), a novel Italian linguistic resource for Sentiment Analysis, developed with the main aim of analyzing the online debate on the controversial Italian political reform “Buona Scuola” (Good school), aimed at reorganizing the national educational and training systems. We describe the methodologies applied in the collection and annotation of data. The collection has been driven by the detection of the hashtags mainly used by the participants to the debate, while the annotation has been focused on sentiment polarity and irony, but also extended to mark the aspects of the reform that were mainly discussed in the debate. An in-depth study of the disagreement among annotators is included. We describe the collection and annotation stages, and the in-depth analysis of disagreement made with Crowdflower, a crowdsourcing annotation platform.
This paper presents NileULex, which is an Arabic sentiment lexicon containing close to six thousands Arabic words and compound phrases. Forty five percent of the terms and expressions in the lexicon are Egyptian or colloquial while fifty five percent are Modern Standard Arabic. While the collection of many of the terms included in the lexicon was done automatically, the actual addition of any term was done manually. One of the important criterions for adding terms to the lexicon, was that they be as unambiguous as possible. The result is a lexicon with a much higher quality than any translated variant or automatically constructed one. To demonstrate that a lexicon such as this can directly impact the task of sentiment analysis, a very basic machine learning based sentiment analyser that uses unigrams, bigrams, and lexicon based features was applied on two different Twitter datasets. The obtained results were compared to a baseline system that only uses unigrams and bigrams. The same lexicon based features were also generated using a publicly available translation of a popular sentiment lexicon. The experiments show that usage of the developed lexicon improves the results over both the baseline and the publicly available lexicon.
The paper contains a description of OPFI: Opinion Finder for the Polish Language, a freely available tool for opinion target extraction. The goal of the tool is opinion finding: a task of identifying tuples composed of sentiment (positive or negative) and its target (about what or whom is the sentiment expressed). OPFI is not dependent on any particular method of sentiment identification and provides a built-in sentiment dictionary as a convenient option. Technically, it contains implementations of three different modes of opinion tuple generation: one hybrid based on dependency parsing and CRF, the second based on shallow parsing and the third on deep learning, namely GRU neural network. The paper also contains a description of related language resources: two annotated treebanks and one set of tweets.
The fine-grained task of automatically detecting all sentiment expressions within a given document and the aspects to which they refer is known as aspect-based sentiment analysis. In this paper we present the first full aspect-based sentiment analysis pipeline for Dutch and apply it to customer reviews. To this purpose, we collected reviews from two different domains, i.e. restaurant and smartphone reviews. Both corpora have been manually annotated using newly developed guidelines that comply to standard practices in the field. For our experimental pipeline we perceive aspect-based sentiment analysis as a task consisting of three main subtasks which have to be tackled incrementally: aspect term extraction, aspect category classification and polarity classification. First experiments on our Dutch restaurant corpus reveal that this is indeed a feasible approach that yields promising results.
We construct a case-based English-to-Chinese semantic constituent parallel Treebank for a Statistical Machine Translation (SMT) task by labelling each node of the Deep Syntactic Tree (DST) with our refined semantic cases. Since subtree span-crossing is harmful in tree-based SMT, DST is adopted to alleviate this problem. At the same time, we tailor an existing case set to represent bilingual shallow semantic relations more precisely. This Treebank is a part of a semantic corpus building project, which aims to build a semantic bilingual corpus annotated with syntactic, semantic cases and word senses. Data in our Treebank is from the news domain of Datum corpus. 4,000 sentence pairs are selected to cover various lexicons and part-of-speech (POS) n-gram patterns as much as possible. This paper presents the construction of this case Treebank. Also, we have tested the effect of adopting DST structure in alleviating subtree span-crossing. Our preliminary analysis shows that the compatibility between Chinese and English trees can be significantly increased by transforming the parse-tree into the DST. Furthermore, the human agreement rate in annotation is found to be acceptable (90% in English nodes, 75% in Chinese nodes).
Morphologically-rich languages pose problems for machine translation (MT) systems, including word-alignment errors, data sparsity and multiple affixes. Current alignment models at word-level do not distinguish words and morphemes, thus yielding low-quality alignment and subsequently affecting end translation quality. Models using morpheme-level alignment can reduce the vocabulary size of morphologically-rich languages and overcomes data sparsity. The alignment data based on smallest units reveals subtle language features and enhances translation quality. Recent research proves such morpheme-level alignment (MA) data to be valuable linguistic resources for SMT, particularly for languages with rich morphology. In support of this research trend, the Linguistic Data Consortium (LDC) created Uzbek-English and Turkish-English alignment data which are manually aligned at the morpheme level. This paper describes the creation of MA corpora, including alignment and tagging process and approaches, highlighting annotation challenges and specific features of languages with rich morphology. The light tagging annotation on the alignment layer adds extra value to the MA data, facilitating users in flexibly tailoring the data for various MT model training.
Parallel corpora are crucial for machine translation (MT), however they are quite scarce for most language pairs and domains. As comparable corpora are far more available, many studies have been conducted to extract parallel sentences from them for MT. In this paper, we exploit the neural network features acquired from neural MT for parallel sentence extraction. We observe significant improvements for both accuracy in sentence extraction and MT performance.
We introduce TweetMT, a parallel corpus of tweets in four language pairs that combine five languages (Spanish from/to Basque, Catalan, Galician and Portuguese), all of which have an official status in the Iberian Peninsula. The corpus has been created by combining automatic collection and crowdsourcing approaches, and it is publicly available. It is intended for the development and testing of microtext machine translation systems. In this paper we describe the methodology followed to build the corpus, and present the results of the shared task in which it was tested.
The biomedical scientific literature is a rich source of information not only in the English language, for which it is more abundant, but also in other languages, such as Portuguese, Spanish and French. We present the first freely available parallel corpus of scientific publications for the biomedical domain. Documents from the ”Biological Sciences” and ”Health Sciences” categories were retrieved from the Scielo database and parallel titles and abstracts are available for the following language pairs: Portuguese/English (about 86,000 documents in total), Spanish/English (about 95,000 documents) and French/English (about 2,000 documents). Additionally, monolingual data was also collected for all four languages. Sentences in the parallel corpus were automatically aligned and a manual analysis of 200 documents by native experts found that a minimum of 79% of sentences were correctly aligned in all language pairs. We demonstrate the utility of the corpus by running baseline machine translation experiments. We show that for all language pairs, a statistical machine translation system trained on the parallel corpora achieves performance that rivals or exceeds the state of the art in the biomedical domain. Furthermore, the corpora are currently being used in the biomedical task in the First Conference on Machine Translation (WMT’16).
This paper presents an approach for building large monolingual corpora and, at the same time, extracting parallel data by crawling the top-level domain of a given language of interest. For gathering linguistically relevant data from top-level domains we use the SpiderLing crawler, modified to crawl data written in multiple languages. The output of this process is then fed to Bitextor, a tool for harvesting parallel data from a collection of documents. We call the system combining these two tools Spidextor, a blend of the names of its two crucial parts. We evaluate the described approach intrinsically by measuring the accuracy of the extracted bitexts from the Croatian top-level domain “.hr” and the Slovene top-level domain “.si”, and extrinsically on the English-Croatian language pair by comparing an SMT system built from the crawled data with third-party systems. We finally present parallel datasets collected with our approach for the English-Croatian, English-Finnish, English-Serbian and English-Slovene language pairs.
We describe a strategy for the acquisition of training data necessary to build a social-media-driven early detection system for individuals at risk for (preventable) type 2 diabetes mellitus (T2DM). The strategy uses a game-like quiz with data and questions acquired semi-automatically from Twitter. The questions are designed to inspire participant engagement and collect relevant data to train a public-health model applied to individuals. Prior systems designed to use social media such as Twitter to predict obesity (a risk factor for T2DM) operate on entire communities such as states, counties, or cities, based on statistics gathered by government agencies. Because there is considerable variation among individuals within these groups, training data on the individual level would be more effective, but this data is difficult to acquire. The approach proposed here aims to address this issue. Our strategy has two steps. First, we trained a random forest classifier on data gathered from (public) Twitter statuses and state-level statistics with state-of-the-art accuracy. We then converted this classifier into a 20-questions-style quiz and made it available online. In doing so, we achieved high engagement with individuals that took the quiz, while also building a training set of voluntarily supplied individual-level data for future classification.
We set out to investigate whether TV ratings and mentions of TV programmes on the Twitter social media platform are correlated. If such a correlation exists, Twitter may be used as an alternative source for estimating viewer popularity. Moreover, the Twitter-based rating estimates may be generated during the programme, or even before. We count the occurrences of programme-specific hashtags in an archive of Dutch tweets of eleven popular TV shows broadcast in the Netherlands in one season, and perform correlation tests. Overall we find a strong correlation of 0.82; the correlation remains strong, 0.79, if tweets are counted a half hour before broadcast time. However, the two most popular TV shows account for most of the positive effect; if we leave out the single and second most popular TV shows, the correlation drops to being moderate to weak. Also, within a TV show, correlations between ratings and tweet counts are mostly weak, while correlations between TV ratings of the previous and next shows are strong. In absence of information on previous shows, Twitter-based counts may be a viable alternative to classic estimation methods for TV ratings. Estimates are more reliable with more popular TV shows.
In this paper we consider the problem of out-of-vocabulary term classification in web forum text from the automotive domain. We develop a set of nine domain- and application-specific categories for out-of-vocabulary terms. We then propose a supervised approach to classify out-of-vocabulary terms according to these categories, drawing on features based on word embeddings, and linguistic knowledge of common properties of out-of-vocabulary terms. We show that the features based on word embeddings are particularly informative for this task. The categories that we predict could serve as a preliminary, automatically-generated source of lexical knowledge about out-of-vocabulary terms. Furthermore, we show that this approach can be adapted to give a semi-automated method for identifying out-of-vocabulary terms of a particular category, automotive named entities, that is of particular interest to us.
Many people post about their daily life on social media. These posts may include information about the purchase activity of people, and insights useful to companies can be derived from them: e.g. profile information of a user who mentioned something about their product. As a further advanced analysis, we consider extracting users who are likely to buy a product from the set of users who mentioned that the product is attractive. In this paper, we report our methodology for building a corpus for Twitter user purchase behavior prediction. First, we collected Twitter users who posted a want phrase + product name: e.g. “want a Xperia” as candidate want users, and also candidate bought users in the same way. Then, we asked an annotator to judge whether a candidate user actually bought a product. We also annotated whether tweets randomly sampled from want/bought user timelines are relevant or not to purchase. In this annotation, 58% of want user tweets and 35% of bought user tweets were annotated as relevant. Our data indicate that information embedded in timeline tweets can be used to predict purchase behavior of tweeted products.
Hashtags, which are commonly composed of multiple words, are increasingly used to convey the actual messages in tweets. Understanding what tweets are saying is getting more dependent on understanding hashtags. Therefore, identifying the individual words that constitute a hashtag is an important, yet a challenging task due to the abrupt nature of the language used in tweets. In this study, we introduce a feature-rich approach based on using supervised machine learning methods to segment hashtags. Our approach is unsupervised in the sense that instead of using manually segmented hashtags for training the machine learning classifiers, we automatically create our training data by using tweets as well as by automatically extracting hashtag segmentations from a large corpus. We achieve promising results with such automatically created noisy training data.
Language varies not only between countries, but also along regional and socio-demographic lines. This variation is one of the driving factors behind language change. However, investigating language variation is a complex undertaking: the more factors we want to consider, the more data we need. Traditional qualitative methods are not well-suited to do this, an therefore restricted to isolated factors. This reduction limits the potential insights, and risks attributing undue importance to easily observed factors. While there is a large interest in linguistics to increase the quantitative aspect of such studies, it requires training in both variational linguistics and computational methods, a combination that is still not common. We take a first step here to alleviating the problem by providing an interface, www.languagevariation.com, to explore large-scale language variation along multiple socio-demographic factors – without programming knowledge. It makes use of large amounts of data and provides statistical analyses, maps, and interactive features that will enable scholars to explore language variation in a data-driven way.
Much research has focused on detecting trends on Twitter, including health-related trends such as mentions of Influenza-like illnesses or their symptoms. The majority of this research has been conducted using Twitter’s public feed, which includes only about 1% of all public tweets. It is unclear if, when, and how using Twitter’s 1% feed has affected the evaluation of trend detection methods. In this work we use a larger feed to investigate the effects of sampling on Twitter trend detection. We focus on using health-related trends to estimate the prevalence of Influenza-like illnesses based on tweets. We use ground truth obtained from the CDC and Google Flu Trends to explore how the prevalence estimates degrade when moving from a 100% to a 1% sample. We find that using the 1% sample is unlikely to substantially harm ILI estimates made at the national level, but can cause poor performance when estimates are made at the city level.
In this paper, we present a study on tweet classification which aims to define the communication behavior of the 103 French museums that participated in 2014 in the Twitter operation: MuseumWeek. The tweets were automatically classified in four communication categories: sharing experience, promoting participation, interacting with the community, and promoting-informing about the institution. Our classification is multi-class. It combines Support Vector Machines and Naive Bayes methods and is supported by a selection of eighteen subtypes of features of four different kinds: metadata information, punctuation marks, tweet-specific and lexical features. It was tested against a corpus of 1,095 tweets manually annotated by two experts in Natural Language Processing and Information Communication and twelve Community Managers of French museums. We obtained an state-of-the-art result of F1-score of 72% by 10-fold cross-validation. This result is very encouraging since is even better than some state-of-the-art results found in the tweet classification literature.
Word sense induction (WSI) seeks to induce senses of words from unannotated corpora. In this paper, we address the WSI task for the Croatian language. We adopt the word clustering approach based on co-occurrence graphs, in which senses are taken to correspond to strongly inter-connected components of co-occurring words. We experiment with a number of graph construction techniques and clustering algorithms, and evaluate the sense inventories both as a clustering problem and extrinsically on a word sense disambiguation (WSD) task. In the cluster-based evaluation, Chinese Whispers algorithm outperformed Markov Clustering, yielding a normalized mutual information score of 64.3. In contrast, in WSD evaluation Markov Clustering performed better, yielding an accuracy of about 75%. We are making available two induced sense inventories of 10,000 most frequent Croatian words: one coarse-grained and one fine-grained inventory, both obtained using the Markov Clustering algorithm.
We describe the word sense annotation layer in Eukalyptus, a freely available five-domain corpus of contemporary Swedish with several annotation layers. The annotation uses the SALDO lexicon to define the sense inventory, and allows word sense annotation of compound segments and multiword units. We give an overview of the new annotation tool developed for this project, and finally present an analysis of the inter-annotator agreement between two annotators.
This work presents parallel corpora automatically annotated with several NLP tools, including lemma and part-of-speech tagging, named-entity recognition and classification, named-entity disambiguation, word-sense disambiguation, and coreference. The corpora comprise both the well-known Europarl corpus and a domain-specific question-answer troubleshooting corpus on the IT domain. English is common in all parallel corpora, with translations in five languages, namely, Basque, Bulgarian, Czech, Portuguese and Spanish. We describe the annotated corpora and the tools used for annotation, as well as annotation statistics for each language. These new resources are freely available and will help research on semantic processing for machine translation and cross-lingual transfer.
We present a VerbNet-based annotation scheme for semantic roles that we explore in an annotation study on German language data that combines word sense and semantic role annotation. We reannotate a substantial portion of the SALSA corpus with GermaNet senses and a revised scheme of VerbNet roles. We provide a detailed evaluation of the interaction between sense and role annotation. The resulting corpus will allow us to compare VerbNet role annotation for German to FrameNet and PropBank annotation by mapping to existing role annotations on the SALSA corpus. We publish the annotated corpus and detailed guidelines for the new role annotation scheme.
Word Sense Disambiguation (WSD) is one of the open problems in the area of natural language processing. Various supervised, unsupervised and knowledge based approaches have been proposed for automatically determining the sense of a word in a particular context. It has been observed that such approaches often find it difficult to beat the WordNet First Sense (WFS) baseline which assigns the sense irrespective of context. In this paper, we present our work on creating the WFS baseline for Hindi language by manually ranking the synsets of Hindi WordNet. A ranking tool is developed where human experts can see the frequency of the word senses in the sense-tagged corpora and have been asked to rank the senses of a word by using this information and also his/her intuition. The accuracy of WFS baseline is tested on several standard datasets. F-score is found to be 60%, 65% and 55% on Health, Tourism and News datasets respectively. The created rankings can also be used in other NLP applications viz., Machine Translation, Information Retrieval, Text Summarization, etc.
In this paper, a new approach towards semantic clustering of the results of ambiguous search queries is presented. We propose using distributed vector representations of words trained with the help of prediction-based neural embedding models to detect senses of search queries and to cluster search engine results page according to these senses. The words from titles and snippets together with semantic relationships between them form a graph, which is further partitioned into components related to different query senses. This approach to search engine results clustering is evaluated against a new manually annotated evaluation data set of Russian search queries. We show that in the task of semantically clustering search results, prediction-based models slightly but stably outperform traditional count-based ones, with the same training corpora.
This paper describes the evaluation methodology followed to measure the impact of using a machine learning algorithm to automatically segment intralingual subtitles. The segmentation quality, productivity and self-reported post-editing effort achieved with such approach are shown to improve those obtained by the technique based in counting characters, mainly employed for automatic subtitle segmentation currently. The corpus used to train and test the proposed automated segmentation method is also described and shared with the community, in order to foster further research in this area.
Images naturally appear alongside text in a wide variety of media, such as books, magazines, newspapers, and in online articles. This type of multi-modal data offers an interesting basis for vision and language research but most existing datasets use crowdsourced text, which removes the images from their original context. In this paper, we introduce the KBK-1M dataset of 1.6 million images in their original context, with co-occurring texts found in Dutch newspapers from 1922 - 1994. The images are digitally scanned photographs, cartoons, sketches, and weather forecasts; the text is generated from OCR scanned blocks. The dataset is suitable for experiments in automatic image captioning, image―article matching, object recognition, and data-to-text generation for weather forecasting. It can also be used by humanities scholars to analyse photographic style changes, the representation of people and societal issues, and new tools for exploring photograph reuse via image-similarity-based search.
The task of automatically generating sentential descriptions of image content has become increasingly popular in recent years, resulting in the development of large-scale image description datasets and the proposal of various metrics for evaluating image description generation systems. However, not much work has been done to analyse and understand both datasets and the metrics. In this paper, we propose using a leave-one-out cross validation (LOOCV) process as a means to analyse multiply annotated, human-authored image description datasets and the various evaluation metrics, i.e. evaluating one image description against other human-authored descriptions of the same image. Such an evaluation process affords various insights into the image description datasets and evaluation metrics, such as the variations of image descriptions within and across datasets and also what the metrics capture. We compute and analyse (i) human upper-bound performance; (ii) ranked correlation between metric pairs across datasets; (iii) lower-bound performance by comparing a set of descriptions describing one image to another sentence not describing that image. Interesting observations are made about the evaluation metrics and image description datasets, and we conclude that such cross-validation methods are extremely useful for assessing and gaining insights into image description datasets and evaluation metrics for image descriptions.
In American Sign Language (ASL) as well as other signed languages, different classes of signs (e.g., lexical signs, fingerspelled signs, and classifier constructions) have different internal structural properties. Continuous sign recognition accuracy can be improved through use of distinct recognition strategies, as well as different training datasets, for each class of signs. For these strategies to be applied, continuous signing video needs to be segmented into parts corresponding to particular classes of signs. In this paper we present a multiple instance learning-based segmentation system that accurately labels 91.27% of the video frames of 500 continuous utterances (including 7 different subjects) from the publicly accessible NCSLGR corpus (Neidle and Vogler, 2012). The system uses novel feature descriptors derived from both motion and shape statistics of the regions of high local motion. The system does not require a hand tracker.
Lexical Simplification is the task of replacing complex words in a text with simpler alternatives. A variety of strategies have been devised for this challenge, yet there has been little effort in comparing their performance. In this contribution, we present a benchmarking of several Lexical Simplification systems. By combining resources created in previous work with automatic spelling and inflection correction techniques, we introduce BenchLS: a new evaluation dataset for the task. Using BenchLS, we evaluate the performance of solutions for various steps in the typical Lexical Simplification pipeline, both individually and jointly. This is the first time Lexical Simplification systems are compared in such fashion on the same data, and the findings introduce many contributions to the field, revealing several interesting properties of the systems evaluated.
Scientific literature records the research process with a standardized structure and provides the clues to track the progress in a scientific field. Understanding its internal structure and content is of paramount importance for natural language processing (NLP) technologies. To meet this requirement, we have developed a multi-layered annotated corpus of scientific papers in the domain of Computer Graphics. Sentences are annotated with respect to their role in the argumentative structure of the discourse. The purpose of each citation is specified. Special features of the scientific discourse such as advantages and disadvantages are identified. In addition, a grade is allocated to each sentence according to its relevance for being included in a summary.To the best of our knowledge, this complex, multi-layered collection of annotations and metadata characterizing a set of research papers had never been grouped together before in one corpus and therefore constitutes a newer, richer resource with respect to those currently available in the field.
In this paper we report a comparison of various techniques for single-document extractive summarization under strict length budgets, which is a common commercial use case (e.g. summarization of news articles by news aggregators). We show that, evaluated using ROUGE, numerous algorithms from the literature fail to beat a simple lead-based baseline for this task. However, a supervised approach with lightweight and efficient features improves over the lead-based baseline. Additional human evaluation demonstrates that the supervised approach also performs competitively with a commercial system that uses more sophisticated features.
Automatic summarization of reader comments in on-line news is an extremely challenging task and a capability for which there is a clear need. Work to date has focussed on producing extractive summaries using well-known techniques imported from other areas of language processing. But are extractive summaries of comments what users really want? Do they support users in performing the sorts of tasks they are likely to want to perform with reader comments? In this paper we address these questions by doing three things. First, we offer a specification of one possible summary type for reader comment, based on an analysis of reader comment in terms of issues and viewpoints. Second, we define a task-based evaluation framework for reader comment summarization that allows summarization systems to be assessed in terms of how well they support users in a time-limited task of identifying issues and characterising opinion on issues in comments. Third, we describe a pilot evaluation in which we used the task-based evaluation framework to evaluate a prototype reader comment clustering and summarization system, demonstrating the viability of the evaluation framework and illustrating the sorts of insight such an evaluation affords.
Unsupervised learning of morphological segmentation of words in a language, based only on a large corpus of words, is a challenging task. Evaluation of the learned segmentations is a challenge in itself, due to the inherent ambiguity of the segmentation task. There is no way to posit unique “correct” segmentation for a set of data in an objective way. Two models may arrive at different ways of segmenting the data, which may nonetheless both be valid. Several evaluation methods have been proposed to date, but they do not insist on consistency of the evaluated model. We introduce a new evaluation methodology, which enforces correctness of segmentation boundaries while also assuring consistency of segmentation decisions across the corpus.
Bilingual lexicon extraction from comparable corpora is usually based on distributional methods when dealing with single word terms (SWT). These methods often treat SWT as single tokens without considering their compositional property. However, many SWT are compositional (composed of roots and affixes) and this information, if taken into account can be very useful to match translational pairs, especially for infrequent terms where distributional methods often fail. For instance, the English compound xenograft which is composed of the root xeno and the lexeme graft can be translated into French compositionally by aligning each of its elements (xeno with xéno and graft with greffe) resulting in the translation: xénogreffe. In this paper, we experiment several distributional modellings at the morpheme level that we apply to perform compositional translation to a subset of French and English compounds. We show promising results using distributional analysis at the root and affix levels. We also show that the adapted approach significantly improve bilingual lexicon extraction from comparable corpora compared to the approach at the word level.
Structured, complete inflectional paradigm data exists for very few of the world’s languages, but is crucial to training morphological analysis tools. We present methods inspired by linguistic fieldwork for gathering inflectional paradigm data in a machine