In this paper the adequacy of the SETimes corpus as a basis for the comparison of closely related languages that are used in countries that emerged after the breakup of Yugoslavia is discussed by comparing it with other corpora. It is shown that the phenomena observed in this corpus and used to illustrate differences most specifically between Serbian and Croatian are consistent neither with their standards nor with other sources. Thus, results obtained on the basis of the SETimes corpus are corpus-biased and have to be reconsidered. This proves that the size of a corpus and its composition used in a linguistic research are crucial for assessing the obtained results.

The combination of the meanings ‘while’ and ‘until’ in a single lexeme and the use of expletive negation with the latter meaning are widespread phenomena that are a rich source of research problems. In this paper we present a comparative bilingual Bulgarian and Ukrainian corpus-based study of several conjunctions that share these two meanings. We discuss the difference in the frequency of expletive negation in the two languages, the use of až ‘even, all the way’ in Ukrainian and the impact of the original language in translated texts.

This paper describes the Romanian treebank annotated according to the Universal Dependency principles. We present the types of texts included in the treebank, their processing phases and the tools used for doing it, as well as the levels of annotation, with a focus on the syntactic level. We briefly present the syntactic formalism used, the principles followed and the set of relations. The perspective we adopted is the linguist’s who searches the treebank for information with relevance for the study of Romanian. (S)He can interpret the statistics based on the corpus and can also query the treebank for finding examples to support a theory, for testing hypothesis or for discovering new tendencies. We use here the passive constructions in Romanian as a case study for showing how statistical data help understanding this linguistic phenomenon. We also discuss the kinds of linguistic information retrievable and non-retrievable form the treebank, based on the annotation principles.

This paper presents work in progress focused on developing a method for automatic identification of light verb constructions (LVCs) as a subclass of Bulgarian verbal MWEs. The method is based on machine learning and is trained on a set of LVCs extracted from the Bulgarian WordNet (BulNet) and the Bulgarian National Corpus (BulNC). The machine learning uses lexical, morphosyntactic, syntactic and semantic features of LVCs. We trained and tested two separate classifiers using the Java package Weka and two learning decision tree algorithms – J48 and RandomTree. The evaluation of the method includes 10-fold cross-validation on the training data from BulNet (F1 = 0.766 obtained by the J48 decision tree algorithm and F1 = 0.725 by the RandomTree algorithm), as well as evaluation of the performance on new instances from the BulNC (F1 = 0.802 by J48 and F1 = 0.607 by the RandomTree algorithm). Preliminary filtering of the candidates gives a slight improvement (F1 = 0.802 by J48 and F1 = 0.737 by RandomTree).

In this paper we present the HR4EU – web portal for e-learning of Croatian language. The web portal offers a new method of computer aided language learning (CALL) by encouraging language learners to use different language resources available for Croatian: corpora, inflectional and derivational morphological lexicons, treebank, Wordnet, etc. Apart from the previously developed language resources, the new ones are created in order to further facilitate the learning of Croatian language. We will focus on the usage of the treebank annotated at syntactic and semantic level in the CALL and describe the new HR4EU sub-corpus of the Croatian Dependency Treebank (HOBS). The HR4EU sub-corpus consists of approx. 550 sentences, which are manually annotated on syntactic and semantic role level according to the specifications used for the HOBS. The syntactic and the semantic structure of the sentence can be visualized as a dependency tree via the SynSem Visualizer. The visualization of the syntactic and the semantic structure of sentences will help users to produce syntactically and semantically correct sentences on their own.

This paper presents a web tool for syntactic and semantic annotation and two of its applications. It gives the linguists the possibility to work with corpora and syntactic and semantic frames in XML format without having computer skills. The system is OS and platform independent and could be used both online and offline.

Community Question Answering (CQA) is a form of question answering that is getting increasingly popular as a research direction recently. Given a question posted in an online community forum and the thread of answers to it, a common formulation of the task is to rank automatically the answers, so that the good ones are ranked higher than the bad ones. Despite the vast research in CQA for English, very little attention has been paid to other languages. To bridge this gap, here we present our method for Community Question Answering in Bulgarian. We create annotated training and testing datasets for Bulgarian, and we further explore the applicability of machine translation for reusing English CQA data for building a Bulgarian system. The evaluation results show improvement over the baseline and can serve as a basis for further research.

This paper presents a method for automatic retrieval and attribution of quotations from media texts in Bulgarian. It involves recognition of report verbs (including their analytical forms) and syntactic patterns introducing quotations, as well as source attribution of the quote by identification of personal names, descriptors, and anaphora. The method is implemented in a fully-functional online system which offers a live service processing media content and extracting quotations on a daily basis. The system collects and processes written news texts from six Bulgarian media websites. The results are presented in a structured way with description, as well as sorting and filtering functionalities which facilitate the monitoring and analysis of media content. The method has been applied to extract quotations from English texts as well and can be adapted to work with other languages, provided that the respective language specific resources are supplied.

The paper presents an ongoing research on the stress patterns of compounds and MWEs of the type ADJ+N and their corresponding free NPs in English and Bulgarian. The research focuses on the identification and the formal representation of the possible stress patterns of compounds and MWEs and free NPs. During our research so far, we have compiled a corpus of over 2000 compounds and MWEs, approx. 1000 for each language – English and Bulgarian. Our theoretical framework includes elements from different theories, i.e. the Generative Phonology Theory, the Metrical Theory, and the Theory of Primary accent first which all define the stress as a prosodic element. Our main goals are to specify the prosodic region where the stress is defined in English and Bulgarian MWEs and noun phrases and to define the main features of the stress in MWEs and free NPs in English and Bulgarian. The results of our research can serve for implementation into NLP modules for spoken language processing and generation.

The paper deals with verbal multiword expressions in Croatian. We focus on four types of verbal constructions: light verb constructions, i.e. constructions consisting of a light verb and a noun or prepositional phrase, complex predicate constructions, i.e. constructions consisting of a finite and infinitive verb, prepositional verb constructions, i.e. constructions consisting of a verb and a typical preposition, and, finally, verbal idioms, i.e. constructions with completely idiosyncratic meanings. All the constructions are annotated in the Universal Dependency treebank for Croatian. The identification of verbal multiword expressions is an important task in numerous NLP tasks. It is also important to define and delimitate this concept in linguistic theory.

In this study we outline a potential problem in the normalisation stage of processing texts that are based on a modified version of the Arabic alphabet. The main source of resources available for processing resource-scarce languages is raw text. We have identified an interesting challenge that must be addressed when normalising certain natural language texts. Many less-resourced languages, such as Kurdish, Farsi, Urdu, Pashtu, etc., use a modified version of the Arabic writing system. Many characters in harvested data from the Internet may have exactly the same form but encoded with different Unicode values (ambiguous characters). It is important to identify ambiguous characters during the normalisation stage of most text processing tasks. We will demonstrate cases related to ambiguous Kurdish and Farsi characters and propose a semi-automatic approach to identifying and unifying ambiguously encoded characters.

The paper‘s main subject is concerned with the problems related to machine translation of verb forms from Bulgarian to English. In separate sections of this article we discuss the problems related to differences between word formation in both languages and differences in the information that the verb forms grammaticalize. We also introduce the idea of implementing the statistical method of machine translation altogether with the rule-based method as a proposal for future research and the possible practical and theoretical outcomes.