The Modern pharmaceutical industry depends on the iterative design of novel synthetic routes for drugs while not infringing on existing intellectual property rights. Such a design process calls for analyzing many existing synthetic chemical reactions and planning the synthesis of novel chemicals. These procedures have been historically available in unstructured raw text form in publications and patents. To facilitate automated synthetic chemical reactions analysis and design the novel synthetic reactions using Natural Language Processing (NLP) methods, we introduce a Named Entity Recognition (NER) dataset of the Examples section in 180 full-text patent documents with 5188 synthetic procedures annotated by domain experts. All the chemical entities which are part of the synthetic discourse were annotated with suitable class labels. We present the second-largest chemical NER corpus with 100,129 annotations and the highest IAA value of 98.73% (F-measure) on a 45 document subset. We discuss this new resource in detail and highlight some specific challenges in annotating synthetic chemical procedures with chemical named entities. We make the corpus available to the community to promote further research and development of downstream NLP systems applications. We also provide baseline results for the NER model to the community to improve on.
Semantic annotation has become an important piece of information within corpus linguistics. This information is usually included for every lexical unit of the corpus providing a more exhaustive analysis of language. There are some resources such as lexicons or ontologies that allow this type of annotation. However, expanding these resources is a time-consuming task. This paper describes a simple NLP baseline for increasing accuracy of the existing semantic resources of the UCREL Semantic Analysis System (USAS). In our experiments, Spanish token accuracy is improved by up to 30% using this method.
Pāṇini in his Aṣṭādhyāyī has written the grammar of Sanskrit in an extremely concise manner in the form of about 4000 sūtras. We have attempted to mathematically remodel the data produced by these sūtras. The mathematical modelling is a way to show that the Pāṇinian approach is a minimal method of capturing the grammatical data for Sanskrit which is a natural language. The sūtras written by Pāṇini can be written as functions, that is for a single input the function produces a single output of the form y=f(x), where x and y is the input and output respectively. However, we observe that for some input dhātus, we get multiple outputs. For such cases, we have written multivalued functions that is the functions which give two or more outputs for a single input. In other words, multivalued function is a way to represent optional output forms which are expressed in Pāṇinian grammar with the help of 3 terms i.e. vā, vibhaṣā, and anyatarasyam. Comparison between the techniques employed by Pāṇini and our notation of functions helps us understand how Pāṇinian techniques ensure brevity and terseness, hence illustrating that Pāṇinian grammar is minimal.
In this paper we explore the use of Prediction by partial matching (PPM) compression based to segment Hadith into its two main components (Isnad and Matan). The experiments utilized the PPMD variant of the PPM, showing that PPMD is effective in Hadith segmentation. It was also tested on Hadith corpora of different structures. In the first experiment we used the non- authentic Hadith (NAH) corpus for train- ing models and testing, and in the second experiment we used the NAH corpus for training models and the Leeds University and King Saud University (LK) Hadith cor- pus for testing PPMD segmenter. PPMD of order 7 achieved an accuracy of 92.76% and 90.10% in the first and second experiments, respectively.
Current voice assistants typically use the best hypothesis yielded by their Automatic Speech Recognition (ASR) module as input to their Natural Language Understanding (NLU) module, thereby losing helpful information that might be stored in lower-ranked ASR hypotheses. We explore the change in performance of NLU associated tasks when utilizing five-best ASR hypotheses when compared to status quo for two language datasets, German and Portuguese. To harvest information from the ASR five-best, we leverage extractive summarization and joint extractive-abstractive summarization models for Domain Classification (DC) experiments while using a sequence-to-sequence model with a pointer generator network for Intent Classification (IC) and Named Entity Recognition (NER) multi-task experiments. For the DC full test set, we observe significant improvements of up to 7.2% and 15.5% in micro-averaged F1 scores, for German and Portuguese, respectively. In cases where the best ASR hypothesis was not an exact match to the transcribed utterance (mismatched test set), we see improvements of up to 6.7% and 8.8% micro-averaged F1 scores, for German and Portuguese, respectively. For IC and NER multi-task experiments, when evaluating on the mismatched test set, we see improvements across all domains in German and in 17 out of 19 domains in Portuguese (improvements based on change in SeMER scores). Our results suggest that the use of multiple ASR hypotheses, as opposed to one, can lead to significant performance improvements in the DC task for these non-English datasets. In addition, it could lead to significant improvement in the performance of IC and NER tasks in cases where the ASR model makes mistakes.
Asur belongs to North Munda sub-branch of Austro-Asiatic languages which now has less than 10,000 speakers. This is a very first attempt at describing and documenting Asur language, therefore the approach of this paper is descriptive rather than that of answering research questions. The paper attempts to describe the grammatical features such as gender, number, case, pronouns, tense-aspect-mood, negation, question formation, etc. of Asur language. It briefly touches upon the morphosyntactic and typological features of Asur, with the intent to present a concise overview of the language, which has so far remained almost untouched by documentary linguistics.
We present the first study on the post-editing (PE) effort required to build a parallel dataset for English-Manipuri and English-Mizo, in the context of a project on creating data for machine translation (MT). English source text from a local daily newspaper are machine translated into Manipuri and Mizo using PBSMT systems built in-house. A Computer Assisted Translation (CAT) tool is used to record the time, keystroke and other indicators to measure PE effort in terms of temporal and technical effort. A positive correlation between the technical effort and the number of function words is seen for English-Manipuri and English-Mizo but a negative correlation between the technical effort and the number of noun words for English-Mizo. However, average time spent per token in PE English-Mizo text is negatively correlated with the temporal effort. The main reason for these results are due to (i) English and Mizo using the same script, while Manipuri uses a different script and (ii) the agglutinative nature of Manipuri. Further, we check the impact of training a MT system in an incremental approach, by including the post-edited dataset as additional training data. The result shows an increase in HBLEU of up to 4.6 for English-Manipuri.
Knowledge Graph (KG) Embedding methods have been widely studied in the past few years and many methods have been proposed. These methods represent entities and relations in the KG as vectors in a vector space, trained to distinguish correct edges from the incorrect ones. For this distinction, simple functions of vectors’ dimensions, called interactions, are used. These interactions are used to calculate the candidate tail entity vector which is matched against all entities in the KG. However, for most of the existing methods, these interactions are fixed and manually specified. In this work, we propose an automated framework for discovering the interactions while training the KG Embeddings. The proposed method learns relevant interactions along with other parameters during training, allowing it to adapt to different datasets. Many of the existing methods can be seen as special cases of the proposed framework. We demonstrate the effectiveness of the proposed method on link prediction task by extensive experiments on multiple benchmark datasets.
We study the problem of inducing interpretability in Knowledge Graph (KG) embeddings. Learning KG embeddings has been an active area of research in the past few years, resulting in many different models. However, most of these methods do not address the interpretability (semantics) of individual dimensions of the learned embeddings. In this work, we study this problem and propose a method for inducing interpretability in KG embeddings using entity co-occurrence statistics. The proposed method significantly improves the interpretability, while maintaining comparable performance in other KG tasks.
This paper outlines the use of Transformer networks trained to translate math word problems to equivalent arithmetic expressions in infix, prefix, and postfix notations. We compare results produced by a large number of neural configurations and find that most configurations outperform previously reported approaches on three of four datasets with significant increases in accuracy of over 20 percentage points. The best neural approaches boost accuracy by 30% on average when compared to the previous state-of-the-art.
A corpus of Hindi news headlines shared on Twitter was created by collecting tweets of 5 mainstream Hindi news sources for a period of 4 months. 7 independent annotators were recruited to mark the 20 most retweeted news posts by each of the 5 news sources on its clickbait nature. The clickbait score hence generated was assessed for its correlation with interactions on the platform (retweets, favorites, reader replies), tweet word count, and normalized POS (part-of-speech) tag counts in tweets. A positive correlation was observed between readers’ interactions with tweets and tweets’ clickbait score. Significant correlations were also observed for POS tag counts and clickbait score. The prevalence of clickbait in mainstream Hindi news media was found to be similar to its prevalence in English news media. We hope that our observations would provide a platform for discussions on clickbait in mainstream Hindi news media.
We propose a novel deep neural architecture for dependency parsing, which is built upon a Transformer Encoder (Vaswani et al. 2017) and a Stack Pointer Network (Ma et al. 2018). We first encode each sentence using a Transformer Network and then the dependency graph is generated by a Stack Pointer Network by selecting the head of each word in the sentence through a head selection process. We evaluate our model on Turkish and English treebanks. The results show that our trasformer-based model learns long term dependencies efficiently compared to sequential models such as recurrent neural networks. Our self attended stack pointer network improves UAS score around 6% upon the LSTM based stack pointer (Ma et al. 2018) for Turkish sentences with a length of more than 20 words.
Part-of-Speech (POS) is one of the essential tasks for many Natural Language Processing (NLP) applications. There has been a significant amount of work done in POS tagging for resource-rich languages. POS tagging is an essential phase of text analysis in understanding the semantics and context of language. These tags are useful for higher-level tasks such as building parse trees, which can be used for Named Entity Recognition, Coreference resolution, Sentiment Analysis, and Question Answering. There has been work done on code-mixed social media corpus but not on POS tagging of Kannada-English code-mixed data. Here, we present Kannada-English code- mixed social media corpus annotated with corresponding POS tags. We also experimented with machine learning classification models CRF, Bi-LSTM, and Bi-LSTM-CRF models on our corpus.
Automatic recognition of customer complaints on products or services that they purchase can be crucial for the organisations, multinationals and online retailers since they can exploit this information to fulfil their customers’ expectations including managing and resolving the complaints. Recently, researchers have applied supervised learning strategies to automatically identify users’ complaints expressed in English on Twitter. The downside of these approaches is that they require labeled training data for learning, which is expensive to create. This poses a barrier for them being applied to low-resource languages and domains for which task-specific data is not available. Machine translation (MT) can be used as an alternative to the tools that require such task-specific data. In this work, we use state-of-the-art neural MT (NMT) models for translating Hindi reviews into English and investigate performance of the downstream classification task (complaints identification) on their English translations.
Data sparsity is one of the key challenges associated with model development in Natural Language Understanding (NLU) for conversational agents. The challenge is made more complex by the demand for high quality annotated utterances commonly required for supervised learning, usually resulting in weeks of manual labor and high cost. In this paper, we present our results on boosting NLU model performance through training data augmentation using a sequential generative adversarial network (GAN). We explore data generation in the context of two tasks, the bootstrapping of a new language and the handling of low resource features. For both tasks we explore three sequential GAN architectures, one with a token-level reward function, another with our own implementation of a token-level Monte Carlo rollout reward, and a third with sentence-level reward. We evaluate the performance of these feedback models across several sampling methodologies and compare our results to upsampling the original data to the same scale. We further improve the GAN model performance through the transfer learning of the pre-trained embeddings. Our experiments reveal synthetic data generated using the sequential generative adversarial network provides significant performance boosts across multiple metrics and can be a major benefit to the NLU tasks.
Identifying the author of a given text can be useful in historical literature, plagiarism detection, or police investigations. Authorship Attribution (AA) has been well studied and mostly relies on a large feature engineering work. More recently, deep learning-based approaches have been explored for Authorship Attribution (AA). In this paper, we introduce BertAA, a fine-tuning of a pre-trained BERT language model with an additional dense layer and a softmax activation to perform authorship classification. This approach reaches competitive performances on Enron Email, Blog Authorship, and IMDb (and IMDb62) datasets, up to 5.3% (relative) above current state-of-the-art approaches. We performed an exhaustive analysis allowing to identify the strengths and weaknesses of the proposed method. In addition, we evaluate the impact of including additional features (e.g. stylometric and hybrid features) in an ensemble approach, improving the macro-averaged F1-Score by 2.7% (relative) on average.
This paper proposes language independent natural language generator for Tree Adjoining Grammar (TAG)[8] based Machine Translation System. In this model, the TAG based parsing and generation approach considered for the syntactic and semantic analysis of a source language. This model provides an efficient and a systematic way of encapsulating language resources with engineering solution to develop the machine translation System. A TAG based Generator is developed with existing resources using TAG formalism to generate the target language from TAG based parser derivation. The process allows syntactic feature-marking, the Subject-Predicate Agreement marking and multiple synthesized generated outputs in complex and morphological rich language. The challenge in applying such approach is to handle the linguistically diversified features. It is achieved using rule-based translation grammar model to align the source language to corresponding target languages. The computational experiments demonstrate that substantial performance in terms of time and memory could also be obtained by using this approach. Nevertheless, this paper also describes the process of lexicalization and explain the state charts, TAG based adjunction and substitution function and the complexity and challenges beneath parsing-generation process.
Cross-lingual summarization(CLS) is the process of generating a summary in one particular language for a source document in a different language. Low resource languages like Kannada greatly benefit from such systems because they help in delivering a concise representation of the same information in a different popular language. We propose a novel dataset generation pipeline and a first of its kind dataset that will aid in CLS for Kannada-English language pair. This work is also an attempt to inspect the existing systems and extend them to the Kannada-English language pair using our dataset.
Contending hate speech in social media is one of the most challenging social problems of our time. There are various types of anti-social behavior in social media. Foremost of them is aggressive behavior, which is causing many social issues such as affecting the social lives and mental health of social media users. In this paper, we propose an end-to-end ensemble-based architecture to automatically identify and classify aggressive tweets. Tweets are classified into three categories - Covertly Aggressive, Overtly Aggressive, and Non-Aggressive. The proposed architecture is an ensemble of smaller subnetworks that are able to characterize the feature embeddings effectively. We demonstrate qualitatively that each of the smaller subnetworks is able to learn unique features. Our best model is an ensemble of Capsule Networks and results in a 65.2% F1 score on the Facebook test set, which results in a performance gain of 0.95% over the TRAC-2018 winners. The code and the model weights are publicly available at https://github.com/parthpatwa/Hater-O-Genius-Aggression-Classification-using-Capsule-Networks.
The more we are advancing towards a modern world, the more it opens the path to falsification in every aspect of life. Even in case of knowing the surrounding, common people can not judge the actual scenario as the promises, comments and opinions of the influential people at power keep changing every day. Therefore computationally determining the truthfulness of such claims and comments has a very important societal impact. This paper describes a unique method to extract check-worthy claims from the 2016 US presidential debates and verify the truthfulness of the check-worthy claims. We classify the claims for check-worthiness with our modified Tf-Idf model which is used in background training on fact-checking news articles (NBC News and Washington Post). We check the truthfulness of the claims by using POS, sentiment score and cosine similarity features.
In text matching applications, coattentions have proved to be highly effective attention mechanisms. Coattention enables the learning to attend based on computing word level affinity scores between two texts. In this paper, we propose two improvements to coattention mechanism in the context of passage ranking (re-ranking). First, we extend the coattention mechanism by applying it across all word n-grams of query and passage. We show that these word n-gram coattentions can capture local context in query and passage to better judge the relevance between them. Second, we further improve the model performance by proposing a query based attention pooling on passage encodings. We evaluate these two methods on MSMARCO passage re-ranking task. The experiment results shows that these two methods resulted in a relative increase of 8.04% in Mean Reciprocal Rank @10 (MRR@10) compared to the naive coattention mechanism. At the time of writing this paper, our methods are the best non transformer model on MS MARCO passage re-ranking task and are competitive to BERT base while only having less than 10% of the parameters.
Artificial Neural networks are mathematical models at their core. This truism presents some fundamental difficulty when networks are tasked with Natural Language Processing. A key problem lies in measuring the similarity or distance among vectors in NLP embedding space, since the mathematical concept of distance does not always agree with the linguistic concept. We suggest that the best way to measure linguistic distance among vectors is by employing the Language Model (LM) that created them. We introduce Language Model Distance (LMD) for measuring accuracy of vector transformations based on the Distributional Hypothesis ( LMD Accuracy ). We show the efficacy of this metric by applying it to a simple neural network learning the Procrustes algorithm for bilingual word mapping.
Automatic essay grading (AEG) is a process in which machines assign a grade to an essay written in response to a topic, called the prompt. Zero-shot AEG is when we train a system to grade essays written to a new prompt which was not present in our training data. In this paper, we describe a solution to the problem of zero-shot automatic essay grading, using cognitive information, in the form of gaze behaviour. Our experiments show that using gaze behaviour helps in improving the performance of AEG systems, especially when we provide a new essay written in response to a new prompt for scoring, by an average of almost 5 percentage points of QWK.
Although the manual evaluation of essays is a time-consuming process, writing essays has a significant role in assessing learning outcomes. Therefore, automated essay evaluation represents a solution, especially for schools, universities, and testing companies. Moreover, the existence of such systems overcomes some factors that influence manual evaluation such as the evaluator’s mental state, the disparity between evaluators, and others. In this paper, we propose an Arabic essay evaluation system based on a support vector regression (SVR) model along with a wide range of features including morphological, syntactic, semantic, and discourse features. The system evaluates essays according to five criteria: spelling, essay structure, coherence level, style, and punctuation marks, without the need for domain-representative essays (a model essay). A specific model is developed for each criterion; thus, the overall evaluation of the essay is a combination of the previous criteria results. We develop our dataset based on essays written by university students and journalists whose native language is Arabic. The dataset is then evaluated by experts. The experimental results show that 96% of our dataset is correctly evaluated in the overall score and the correlation between the system and the experts’ evaluation is 0.87. Additionally, the system shows variant results in evaluating criteria separately.
The anthology of spoken languages today is inundated with textual information, necessitating the development of automatic summarization models. In this manuscript, we propose an extractor-paraphraser based abstractive summarization system that exploits semantic overlap as opposed to its predecessors that focus more on syntactic information overlap. Our model outperforms the state-of-the-art baselines in terms of ROUGE, METEOR and word mover similarity (WMS), establishing the superiority of the proposed system via extensive ablation experiments. We have also challenged the summarization capabilities of the state of the art Pointer Generator Network (PGN), and through thorough experimentation, shown that PGN is more of a paraphraser, contrary to the prevailing notion of a summarizer; illustrating it’s incapability to accumulate information across multiple sentences.
This paper describes how we developed a neural-based dependency parser, namely ThamizhiUDp, which provides a complete pipeline for the dependency parsing of the Tamil language text using Universal Dependency formalism. We have considered the phases of the dependency parsing pipeline and identified tools and resources in each of these phases to improve the accuracy and to tackle data scarcity. ThamizhiUDp uses Stanza for tokenisation and lemmatisation, ThamizhiPOSt and ThamizhiMorph for generating Part of Speech (POS) and Morphological annotations, and uuparser with multilingual training for dependency parsing. ThamizhiPOSt is our POS tagger, which is based on the Stanza, trained with Amrita POS-tagged corpus. It is the current state-of-the-art in Tamil POS tagging with an F1 score of 93.27. Our morphological analyzer, ThamizhiMorph is a rule-based system with a very good coverage of Tamil. Our dependency parser ThamizhiUDp was trained using multilingual data. It shows a Labelled Assigned Score (LAS) of 62.39, 4 points higher than the current best achieved for Tamil dependency parsing. Therefore, we show that breaking up the dependency parsing pipeline to accommodate existing tools and resources is a viable approach for low-resource languages.
The performance of deep learning models depends on the quality and quantity of data. Data construction, however, is time- consuming and costly. In addition, when expert domain data are constructed, the availability of experts is limited. In such cases, active learning can efficiently increase the performance of the learning models with minimal data construction. Although various datasets have been constructed using active learning techniques, vigorous studies on the construction of Korean data on expert domains are yet to be conducted. In this study, a corpus for named entity recognition was constructed for the financial domain using the active learning technique. The contributions of the study are as follows. (1) It was verified that the active learning technique could effectively construct the named entity recognition corpus for the financial domain, and (2) a named entity recognizer for the financial domain was developed. Data of 8,043 sentences were constructed using the proposed method, and the performance of the named entity recognizer reached 80.84%. Moreover, the proposed method reduced data construction costs by 12–25%
We propose a novel, attention-based self-supervised approach to identify “claim-worthy” sentences in a fake news article, an important first step in automated fact-checking. We leverage aboutness of headline and content using attention mechanism for this task. The identified claims can be used for downstream task of claim verification for which we are releasing a benchmark dataset of manually selected compelling articles with veracity labels and associated evidence. This work goes beyond stylistic analysis to identifying content that influences reader belief. Experiments with three datasets show the strength of our model.
Shayari is a form of poetry mainly popular in the Indian subcontinent, in which the poet expresses his emotions and feelings in a very poetic manner. It is one of the best ways to express our thoughts and opinions. Therefore, it is of prime importance to have an annotated corpus of Hindi shayaris for the task of sentiment analysis. In this paper, we introduce SUKHAN, a dataset consisting of Hindi shayaris along with sentiment polarity labels. To the best of our knowledge, this is the first corpus of Hindi shayaris annotated with sentiment polarity information. This corpus contains a total of 733 Hindi shayaris of various genres. Also, this dataset is of utmost value as all the annotation is done manually by five annotators and this makes it a very rich dataset for training purposes. This annotated corpus is also used to build baseline sentiment classification models using machine learning techniques.
Sanskrit is one of the oldest languages of the Asian Subcontinent that fell out of common usage around 600 B.C. In this paper, we attempt to translate Sanskrit to English using Neural Machine Translation approaches based on Reinforcement Learning and Transfer learning that were never tried and tested on Sanskrit. Along with the paper, we also release monolingual Sanskrit and parallel aligned Sanskrit-English corpora for the research community. Our methodologies outperform the previous approaches applied to Sanskrit by various re- searchers and will further help the linguistic community to accelerate the costly and time consuming manual translation process.
Parsing news Headlines is one of the difficult tasks of Natural Language Processing. It is mostly because news Headlines (NHs) are not complete grammatical sentences. News editors use all sorts of tricks to grab readers’ attention, for instance, unusual capitalization as in the headline’ Ear SHOT ashok rajagopalan’; some are world knowledge demanding like ‘Church reformation celebrated’ where the ‘Church reformation’ refers to a historical event and not a piece of news about an ordinary church. The lack of transparency in NHs can be linguistic, cultural, social, or contextual. The lack of space provided for a news headline has led to creative liberty. Though many works like news value extraction, summary generation, emotion classification of NHs have been going on, parsing them had been a tough challenge. Linguists have also been interested in NHs for creativity in the language used by bending traditional grammar rules. Researchers have conducted studies on news reportage, discourse analysis of NHs, and many more. While the creativity seen in NHs is fascinating for language researchers, it poses a computational challenge for Natural Language Processing researchers. This paper presents an outline of the ongoing doctoral research on the parsing of Indian English NHs. The ultimate aim of this research is to provide a module that will generate correctly parsed NHs. The intention is to enhance the broad applicability of newspaper corpus for future Natural Language Processing applications.
Every language used in this word has ambiguous words. The process of analyzing the word tokens and assigning the correct meanings to the ambiguous words according the context in which they are used is called word sense disambiguation(WSD). WSD is a very hot research topic in Natural Language Processing. The main purpose of my research work is to tackle the WSD problem for Kashmiri language using Supervised Machine Learning Approaches
The paper investigates how well poetry can be generated to contain a specific sentiment, and whether readers of the poetry experience the intended sentiment. The poetry generator consists of a bi-directional Long Short-Term Memory (LSTM) model, combined with rhyme pair generation, rule-based word prediction methods, and tree search for extending generation possibilities. The LSTM network was trained on a set of English poetry written and published by users on a public website. Human judges evaluated poems generated by the system, both with a positive and negative sentiment. The results indicate that while there are some weaknesses in the system compared to other state-of-the-art solutions, it is fully capable of generating poetry with an inherent sentiment that is perceived by readers.
This is a pilot study that aims to explore the potential of using WEKA in forensic authorship analysis. It is a corpus-based research using data from Twitter collected from thirteen authors from Riyadh, Saudi Arabia. It examines the performance of unbalanced and balanced data sets using different classifiers and parameters of word grams. The attributes are dialect-specific linguistic features categorized as word grams. The findings further support previous studies in computational authorship identification.
The paper explores how an attention-based approach can increase performance on the task of native-language identification (NLI), i.e., to identify an author’s first language given information expressed in a second language. Previously, Support Vector Machines have consistently outperformed deep learning-based methods on the TOEFL11 data set, the de facto standard for evaluating NLI systems. The attention-based system BERT (Bidirectional Encoder Representations from Transformers) was first tested in isolation on the TOEFL11 data set, then used in a meta-classifier stack in combination with traditional techniques to produce an accuracy of 0.853. However, more labelled NLI data is now available, so BERT was also trained on the much larger Reddit-L2 data set, containing 50 times as many examples as previously used for English NLI, giving an accuracy of 0.902 on the Reddit-L2 in-domain test scenario, improving the state-of-the-art by 21.2 percentage points.
Text classification is a fundamental problem in the field of natural language processing. Text classification mainly focuses on giving more importance to all the relevant features that help classify the textual data. Apart from these, the text can have redundant or highly correlated features. These features increase the complexity of the classification algorithm. Thus, many dimensionality reduction methods were proposed with the traditional machine learning classifiers. The use of dimensionality reduction methods with machine learning classifiers has achieved good results. In this paper, we propose a hybrid feature selection method for obtaining relevant features by combining various filter-based feature selection methods and fastText classifier. We then present three ways of implementing a feature selection and neural network pipeline. We observed a reduction in training time when feature selection methods are used along with neural networks. We also observed a slight increase in accuracy on some datasets.
The Tamang language is spoken mainly in Nepal, Sikkim, West Bengal, some parts of Assam, and the North East region of India. As per the 2011 census conducted by the Nepal Government, there are about 1.35 million Tamang speakers in Nepal itself. In this regard, a Machine Translation System for Tamang-Nepali language pair is significant both from research and practical outcomes in terms of enabling communication between the Tamang and the Nepali communities. In this work, we train the Transformer Neural Machine Translation (NMT) architecture with attention using a small hand-labeled or aligned Tamang-Nepali corpus (15K sentence pairs). Our preliminary results show BLEU scores of 27.74 for the Nepali→Tamang direction and 23.74 in the Tamang→Nepali direction. We are currently working on increasing the datasets as well as improving the model to obtain better BLEU scores. We also plan to extend the work to add the English language to the model, thus making it a trilingual Machine Translation System for Tamang-Nepali-English languages.
Event Argument extraction refers to the task of extracting structured information from unstructured text for a particular event of interest. The existing works exhibit poor capabilities to extract causal event arguments like Reason and After Effects. Futhermore, most of the existing works model this task at a sentence level, restricting the context to a local scope. While it may be effective for short spans of text, for longer bodies of text such as news articles, it has often been observed that the arguments for an event do not necessarily occur in the same sentence as that containing an event trigger. To tackle the issue of argument scattering across sentences, the use of global context becomes imperative in this task. In our work, we propose an external knowledge aided approach to infuse document level event information to aid the extraction of complex event arguments. We develop a causal network for our event-annotated dataset by extracting relevant event causal structures from ConceptNet and phrases from Wikipedia. We use the extracted event causal features in a bi-directional transformer encoder to effectively capture long-range inter-sentence dependencies. We report the effectiveness of our proposed approach through both qualitative and quantitative analysis. In this task, we establish our findings on an event annotated dataset in 5 Indian languages. This dataset adds further complexity to the task by labeling arguments of entity type (like Time, Place) as well as more complex argument types (like Reason, After-Effect). Our approach achieves state-of-the-art performance across all the five languages. Since our work does not rely on any language specific features, it can be easily extended to other languages as well.
Building an end to end fake news detection system consists of detecting claims in text and later verifying them for their authenticity. Although most of the recent works have focused on political claims, fake news can also be propagated in the form of religious intolerance, conspiracy theories etc. Since there is a lack of training data specific to all these scenarios, we compiled a homogeneous and balanced dataset by combining some of the currently available data. Moreover, it is shown in the paper that how recent advancements in transfer learning can be leveraged to detect claims, in general. The obtained result shows that the recently developed transformers can transfer the tendency of research from claim detection to the problem of check worthiness of claims in domains of interest.
Word sense disambiguation (WSD) is a problem to determine a word according to a context in which it occurs. There are plenty amount of works done in WSD for some languages such as English, but research work on Assamese WSD remains limited. It is a more exigent task because Assamese has an intrinsic complexity in its writing structure and ambiguity, such as syntactic, semantic, and anaphoric ambiguity levels.A novel unsupervised genetic word sense disambiguation algorithm is proposed in this paper. The algorithm first uses WordNet to extract all possible senses for a given ambiguous word, then a genetic algorithm is used taking Wu-Palmer’s similarity measure as the fitness function and calculating the similarity measure for all extracted senses. The winner sense which will have the highest score declared as he winner sense.
The common wisdom about Sanskrit is that it is free word order language. This word order poses challenges such as handling non-projectivity in parsing. The earlier works on the word order of Sanskrit have shown that there are syntactic structures in Sanskrit which cannot be covered under even the non-planarity. In this paper, we study these structures further to investigate if they can fall under well-nestedness or not. A small manually tagged corpus of the verses of Śrīmad-Bhagavad-Gītā was considered for this study. It was noticed that there are as many well-nested trees as there are ill-nested ones. From the linguistic point of view, we could get a list of relations that are involved in the planarity violations. All these relations had one thing in common - that they have unilateral expectancy. It was this loose binding, as against the mutual expectancy with certain other relations, that allowed them to cross the phrasal boundaries.
Automatic prediction of personality traits has many real-life applications, e.g., in forensics, recommender systems, personalized services etc.. In this work, we have proposed a solution framework for solving the problem of predicting the personality traits of a user from videos. Ambient, facial and the audio features are extracted from the video of the user. These features are used for the final output prediction. The visual and audio modalities are combined in two different ways: averaging of predictions obtained from the individual modalities, and concatenation of features in multi-modal setting. The dataset released in Chalearn-16 is used for evaluating the performance of the system. Experimental results illustrate that it is possible to obtain better performance with a hand full of images, rather than using all the images present in the video
Smart devices are often deployed in some edge-devices, which require quality solutions in limited amount of memory usage. In most of the user-interaction based smart devices, coreference resolution is often required. Keeping this in view, we have developed a fast and lightweight coreference resolution model which meets the minimum memory requirement and converges faster. In order to generate the embeddings for solving the task of coreference resolution, DistilBERT, a light weight BERT module is utilized. DistilBERT consumes less memory (only 60% of memory in comparison to BERT-based heavy model) and it is suitable for deployment in edge devices. DistilBERT embedding helps in 60% faster convergence with an accuracy compromise of 2.59%, and 6.49% with respect to its base model and current state-of-the-art, respectively.
Semantic slot prediction is one of the important task for natural language understanding (NLU). They depend on the quality and quantity of the human crafted training data, which affects model generalization. With the advent of voice assistants exposing AI platforms to third party developers, training data quality and quantity matters for any machine learning algorithm to learn and generalize properly.AI platforms provides provision to add custom external plist defined by the developers for the training data. Hence we are exploring dataset, called LowCorpusSlotData, containing low corpus training data with larger number of slots and significant test data. We also use external plist for the above dataset to aid in slot identification. We experimented using state of the art architectures like Bi-directional Encoder Representations from Transformers (BERT) with variants and Bi-directional Encoder with Custom Decoder. To address the low corpus problem, we propose a pipeline approach where we extract candidate slot information using the external plist extractor module and feed as input along with utterance.
Identification of the language spoken from speech utterances is an interesting task because of the diversity associated with different languages and human voices. Indian languages have diverse origins and identifying them from speech utterances would help several language recognition, translation and relationship mining tasks. The current approaches for tackling the problem of languages identification in the Indian context heavily use feature engineering and classical speech processing techniques. This is a bottleneck for language identification systems, as we require to exploit necessary features in speech, required for machine identification, which are learnt by a probabilistic framework, rather than handcrafted feature engineering. In this paper, we tackle the problem of language identification using latent representations learnt from speech using Variational Autoencoders (VAEs) and leverage the representations learnt to train sequence models. Our framework attains an accuracy of 89% in the identification of 8 well known Indian languages (namely Tamil, Telugu, Punjabi, Marathi, Gujarati, Hindi, Kannada and Bengali) from the CMU Indic Speech Database. The presented approach can be applied to several scenarios for speech processing by employing representation learning and leveraging them for sequence models.
Acoustically, English lexical stress is multidimensional and involving manipulation of duration, intensity, fundamental frequency (F0) and vowel quality. The current study investigates the acquisition of English lexical stress by L1 Bengali speakers at the phonological level in terms of the properties of acoustic cues. For this purpose, this study compares 20 L1 Bengali speakers’ use of acoustic correlates for the production of English lexical stress in context sentence and neutral frame sentence. The result of this study showed that L1 Bengali speakers were not able to achieve neutral frame sentence like control over duration, intensity, F0 and to a limited extent vowel quality in context sentence. As a result, unlike neutral frame sentence, L1 Bengali speakers were not sensitive to English lexical stress contrast in context sentence. This analysis reveals that, the difference between the neutral frame and context sentences in terms of L1 Bengali speakers’ realization of phonology of English lexical stress contrast was probably due to the influence of Bengali phonology of lexical stress placement (restricted to the initial syllable of a word) on L1 Bengali speakers’ English speech.
Sign language is a complete natural language used by deaf and dumb people. It has its own grammar and it differs with spoken language to a great extent. Since people without hearing and speech impairment lack the knowledge of the sign language, the deaf and dumb people find it difficult to communicate with them. The conception of system that would be able to translate the sign language into text would facilitate understanding of sign language without human interpreter. This paper describes a systematic approach that takes Indian Sign Language (ISL) video as input and converts it into text using frame sequence generator and image augmentation techniques. By incorporating these two concepts, we have increased dataset size and reduced overfitting. It is demonstrated that using simple image manipulation techniques and batch of shifted frames of videos, performance of sign language recognition can be significantly improved. Approach described in this paper achieves 99.57% accuracy on the dynamic gesture dataset of ISL.
Question Answer pair generation is a task that has been worked upon by multiple researchers in many languages. It has been a topic of interest due to its extensive uses in different fields like self assessment, academics, business website FAQs etc. Many experiments were conducted on Question Answering pair generation in English, concentrating on basic Wh-questions with a rule-based approach. We have built the first hybrid machine learning and rule-based solution in Telugu which is efficient for short stories or short passages in children’s books. Our work covers the fundamental question forms with the question types: adjective, yes/no, adverb, verb, when, where, whose, quotative, and quantitative(how many/ how much). We constructed rules for question generation using POS tags and UD tags along with linguistic information of the surrounding context of the word.
Language detection is considered a difficult task especially for similar languages, varieties, and dialects. With the growing number of online content in different languages, the need for reliable and robust language detection tools also increased. In this work, we use supervised autoencoders with a bayesian optimizer for language detection and highlights its efficiency in detecting similar languages with dialect variance in comparison to other state-of-the-art techniques. We evaluated our approach on multiple datasets (Ling10, Discriminating between Similar Language (DSL), and Indo-Aryan Language Identification (ILI)). Obtained results demonstrate that SAE are higly effective in detecting languages, up to a 100% accuracy in the Ling10. Similarly, we obtain a competitive performance in identifying similar languages, and dialects, 92% and 85% for DSL ans ILI datasets respectively.
In this paper, we propose to use linguistic knowledge to automatically augment a small manually annotated corpus to obtain a large annotated corpus for training Information Extraction models. We propose a powerful patterns specification language for specifying linguistic rules for entity extraction. We define an Enriched Text Format (ETF) to represent rich linguistic information about a text in the form of XML-like tags. The patterns in our patterns specification language are then matched on the ETF text rather than raw text to extract various entity mentions. We demonstrate how an entity extraction system can be quickly built for a domain-specific entity type for which there are no readily available annotated datasets.
In this paper we present a novel transliteration technique based on Orthographic Syllable(OS) segmentation for low-resource Indian languages (ILs). Given that alignment has produced promising results in Statistical Machine Transliteration systems and phonology plays an important role in transliteration, we introduce a new model which uses alignment representation similar to that of IBM model 3 to pre-process the tokenized input sequence and then use pre-trained source and target OS-embeddings for training. We apply our model for transliteration from ILs to English and report our accuracy based on Top-1 Exact Match. We also compare our accuracy with a previously proposed Phrase-Based model and report improvements.
We investigate the problem of extracting Indian-locations from a given crowd-sourced textual dataset. The problem of extracting fine-grained Indian-locations has many challenges. One challenge in the task is to collect relevant dataset from the crowd-sourced platforms that contain locations. The second challenge lies in extracting the location entities from the collected data. We provide an in-depth review of the information collection process and our annotation guidelines such that a reliable dataset annotation is guaranteed. We evaluate many recent algorithms and models, including Conditional Random fields (CRF), Bi-LSTM-CNN and BERT (Bidirectional Encoder Representations from Transformers), on our developed dataset named . The study shows the best F1-score of 72.49% for BERT, followed by Bi-LSTM-CNN and CRF. As a result of our work, we prepare a publicly-available annotated dataset of Indian geolocations that can be used by the research community. Code and dataset are available at https://github.com/vkartik2k/STHAL.
Sentence completion detection (SCD) is an important task for various downstream Natural Language Processing (NLP) based applications. For NLP based applications, which use the Automatic Speech Recognition (ASR) from third parties as a service, SCD is essential to prevent unnecessary processing. Conventional approaches for SCD operate within the confines of sentence boundary detection using language models or sentence end detection using speech and text features. These have limitations in terms of relevant available data for training, performance within the memory and latency constraints, and the generalizability across voice assistant domains. In this paper, we propose a novel sentence completion detection method with low memory footprint for On-Device applications. We explore various sequence-level and sentence-level experiments using state-of-the-art Bi-LSTM and BERT based models for English language.
This paper is an attempt to study polarisation on social media data. We focus on two hugely controversial and talked about events in the Indian diaspora, namely 1) the Sabarimala Temple (located in Kerala, India) incident which became a nationwide controversy when two women under the age of 50 secretly entered the temple breaking a long standing temple rule that disallowed women of menstruating age (10-50) to enter the temple and 2) the Indian government’s move to demonetise all existing 500 and 1000 denomination banknotes, comprising of 86% of the currency in circulation, in November 2016. We gather tweets around these two events in various time periods, preprocess and annotate them with their sentiment polarity and emotional category, and analyse trends to help us understand changing polarity over time around controversial events. The tweets collected are in English, Hindi and code-mixed Hindi-English. Apart from the analysis on the annotated data, we also present the twitter data comprising a total of around 1.5 million tweets.
In the field of Natural Language Processing (NLP) the process of stemming plays a significant role. Stemmer transforms an inflected word to its root form. Stemmer significantly increases the efficiency of Information Retrieval (IR) systems. It is a very basic yet fundamental text pre-processing task widely used in many NLP tasks. Several important works on stemming have been carried out by researchers in English and other major languages. In this paper, we study and review existing works on stemming in Bengali and other Indian languages. Finally, we propose a rule based approach that explores Bengali morphology and leverages WordNet to achieve better accuracy. Our algorithm produced stemming accuracy of 98.86% for Nouns and 99.75% for Verbs.
We present a novel approach for improving the performance of an End-to-End speech recognition system for the Gujarati language. We follow a deep learning based approach which includes Convolutional Neural Network (CNN), Bi-directional Long Short Term Memory (BiLSTM) layers, Dense layers, and Connectionist Temporal Classification (CTC) as a loss function. In order to improve the performance of the system with the limited size of the dataset, we present a combined language model (WLM and CLM) based prefix decoding technique and Bidirectional Encoder Representations from Transformers (BERT) based post-processing technique. To gain key insights from our Automatic Speech Recognition (ASR) system, we proposed different analysis methods. These insights help to understand our ASR system based on a particular language (Gujarati) as well as can govern ASR systems’ to improve the performance for low resource languages. We have trained the model on the Microsoft Speech Corpus, and we observe a 5.11% decrease in Word Error Rate (WER) with respect to base-model WER.
The recognition task of Multi-Word Named Entities (MNEs) in itself is a challenging task when the language is inflectional and agglutinative. Having breakthrough NLP researches with deep neural network and language modelling techniques, the applicability of such techniques/algorithms for Indian language like Manipuri remains unanswered. In this paper an attempt to recognize Manipuri MNE is performed using a Long Short Term Memory (LSTM) recurrent neural network model in conjunction with Part Of Speech (POS) embeddings. To further improve the classification accuracy, word cluster information using K-means clustering approach is added as a feature embedding. The cluster information is generated using a Skip-gram based words vector that contains the semantic and syntactic information of each word. The model so proposed does not use extensive language morphological features to elevate its accuracy. Finally the model’s performance is compared with the other machine learning based Manipuri MNE models.
Automatic short answer grading (ASAG) techniques are designed to automatically assess short answers written in natural language. Apart from MCQs, evaluating free text answer is essential to assess the knowledge and understanding of children in the subject. But assessing descriptive answers in low resource languages in a linguistically diverse country like India poses significant hurdles. To solve this assessment problem and advance NLP research in regional Indian languages, we present the Science Answer Assessment (ScAA) dataset of children’s answers in the age group of 8-14. ScAA dataset is a 2-way (correct/incorrect) labeled dataset and contains 10,988 and 1,955 pairs of natural answers along with model answers for Hindi and Marathi respectively for 32 questions. We benchmark various state-of-the-art ASAG methods, and show the data presents a strong challenge for future research.
In this paper, we address the task of improving pair-wise machine translation for specific low resource Indian languages. Multilingual NMT models have demonstrated a reasonable amount of effectiveness on resource-poor languages. In this work, we show that the performance of these models can be significantly improved upon by using back-translation through a filtered back-translation process and subsequent fine-tuning on the limited pair-wise language corpora. The analysis in this paper suggests that this method can significantly improve multilingual models’ performance over its baseline, yielding state-of-the-art results for various Indian languages.
Nowadays, the spread of Internet memes on online social media platforms such as Instagram, Facebook, Reddit, and Twitter is very fast. Analyzing the sentiment of memes can provide various useful insights. Meme sentiment classification is a new area of research that is not explored yet. Recently SemEval provides a dataset for meme sentiment classification. As this dataset is highly imbalanced, we extend this dataset by annotating new instances and use a sampling strategy to build a meme sentiment classifier. We propose a multi-modal framework for meme sentiment classification by utilizing textual and visual features of the meme. We found that for meme sentiment classification, only textual or only visual features are not sufficient. Our proposed framework utilizes textual as well as visual features together. We propose to use the attention mechanism to improve meme classification performance. Our proposed framework achieves macro F1 and accuracy of 34.23 and 50.02, respectively. It increases the accuracy by 6.77 and 7.86 compared to only textual and visual features, respectively.
Distributional word vector representation or word embedding has become an essential ingredient in many natural language processing (NLP) tasks such as machine translation, document classification, information retrieval and question answering. Investigation of embedding model helps to reduce the feature space and improves textual semantic as well as syntactic relations. This paper presents three embedding techniques (such as Word2Vec, GloVe, and FastText) with different hyperparameters implemented on a Bengali corpus consists of 180 million words. The performance of the embedding techniques is evaluated with extrinsic and intrinsic ways. Extrinsic performance evaluated by text classification, which achieved a maximum of 96.48% accuracy. Intrinsic performance evaluated by word similarity (e.g., semantic, syntactic and relatedness) and analogy tasks. The maximum Pearson (rˆ) correlation accuracy of 60.66% (Ssrˆ) achieved for semantic similarities and 71.64% (Syrˆ) for syntactic similarities whereas the relatedness obtained 79.80% (Rsrˆ). The semantic word analogy tasks achieved 44.00% of accuracy while syntactic word analogy tasks obtained 36.00%.
Emotion recognition is a very well-attended problem in Natural Language Processing (NLP). Most of the existing works on emotion recognition focus on the general domain and in some cases to specific domains like fairy tales, blogs, weather, Twitter etc. But emotion analysis systems in the domains of security, social issues, technology, politics, sports, etc. are very rare. In this paper, we create a benchmark setup for emotion recognition in these specialised domains. First, we construct a corpus of 18,921 tweets in English annotated with Paul Ekman’s six basic emotions (Anger, Disgust, Fear, Happiness, Sadness, Surprise) and a non-emotive class Others. Thereafter, we propose a deep neural framework to perform emotion recognition in an end-to-end setting. We build various models based on Convolutional Neural Network (CNN), Bi-directional Long Short Term Memory (Bi-LSTM), Bi-directional Gated Recurrent Unit (Bi-GRU). We propose a Hierarchical Attention-based deep neural network for Emotion Detection (HAtED). We also develop multiple systems by considering different sets of emotion classes for each system and report the detailed comparative analysis of the results. Experiments show the hierarchical attention-based model achieves best results among the considered baselines with accuracy of 69%.
Data Augmentation methods for Neural Machine Translation (NMT) such as back- translation (BT) and self-training (ST) are quite popular. In a multilingual NMT system, simply copying monolingual source sentences to the target (Copying) is an effective data augmentation method. Back-translation aug- ments parallel data by translating monolingual sentences in the target side to source language. In this work we propose to use a partial back- translation method in a multilingual setting. Instead of translating the entire monolingual target sentence back into the source language, we replace selected high confidence phrases only and keep the rest of the words in the target language itself. (We call this method PhraseOut). Our experiments on low resource multilingual translation models show that PhraseOut gives reasonable improvements over the existing data augmentation methods.
In this paper, we have designed a character-level pre-trained language model for extracting support phrases from tweets based on the sentiment label. We also propose a character-level ensemble model designed by properly blending Pre-trained Contextual Embeddings (PCE) models- RoBERTa, BERT, and ALBERT along with Neural network models- RNN, CNN and WaveNet at different stages of the model. For a given tweet and associated sentiment label, our model predicts the span of phrases in a tweet that prompts the particular sentiment in the tweet. In our experiments, we have explored various model architectures and configuration for both single as well as ensemble models. We performed a systematic comparative analysis of all the model’s performance based on the Jaccard score obtained. The best performing ensemble model obtained the highest Jaccard scores of 73.5, giving it a relative improvement of 2.4% over the best performing single RoBERTa based character-level model, at 71.5(Jaccard score).
We describe the development of a dedicated, high-accuracy part-of-speech (PoS) tagging solution for Faroese, a North Germanic language with about 50,000 speakers. To achieve this, a state-of-the-art neural PoS tagger for Icelandic, ABLTagger, was trained on a 100,000 word PoS-tagged corpus for Faroese, standardised with methods previously applied to Icelandic corpora. This tagger was supplemented with a novel Experimental Database of Faroese Inflection (EDFM), which contains morphological information on 67,488 Faroese words with about one million inflectional forms. This approach produced a PoS-tagging model for Faroese which achieves a 91.40% overall accuracy when evaluated with 10-fold cross validation, which is currently the highest reported accuracy for a dedicated Faroese PoS-tagger. The tagging model, morphological database, proposed revised PoS tagset for Faroese as well as a revised and standardised PoS tagged corpus are all presented as products of this project and are made available for use in further research in Faroese language technology
With the exponential rise in user-generated web content on social media, the proliferation of abusive languages towards an individual or a group across the different sections of the internet is also rapidly increasing. It is very challenging for human moderators to identify the offensive contents and filter those out. Deep neural networks have shown promise with reasonable accuracy for hate speech detection and allied applications. However, the classifiers are heavily dependent on the size and quality of the training data. Such a high-quality large data set is not easy to obtain. Moreover, the existing data sets that have emerged in recent times are not created following the same annotation guidelines and are often concerned with different types and sub-types related to hate. To solve this data sparsity problem, and to obtain more global representative features, we propose a Convolution Neural Network (CNN) based multi-task learning models (MTLs) to leverage information from multiple sources. Empirical analysis performed on three benchmark datasets shows the efficacy of the proposed approach with the significant improvement in accuracy and F-score to obtain state-of-the-art performance with respect to the existing systems.