Encoder-Decoder Model for Semantic Role Labeling

Recently, advances in storage techniques have conducted to an increase in a lot amount of digital images all around the world. These developments have heightened the need for effective image retrieval techniques. Roughly speaking, image retrieval techniques can be divided into two areas: the traditional keyword-based approach and content-based image retrieval. But they still fail on low precision and lack semantic that will cause semantic gap between image features and the user. Previous studies have demonstrated that non-professional users prefer using event-based conceptual descriptions, such as “a woman wearing a hat”, to describe and search images. In many art image archives, these conceptual descriptions are manually annotated in free-text fields. We propose a novel approach for extracting event-based semantic knowledge automatically from free-text image descriptions. The semantic role labeling techniques appear to be a promising technology for transforming the unrestricted natural language texts into structured records that is easy to index/retrieve using relational database technologies. The precision of such SRL-assisted event-based image retrieval is much higher than that of the conventional keyword-based approach. This thesis proposes an ontology-based approach for integrating metadata-based image annotations (administrative knowledge) with event-based knowledge (administrative and conceptual knowledge), including subject, verb, object, location and temporal information from free-text image descriptions. The goal is to automatically derive certain interesting knowledge, such as, all the paintings with a same painting style and a same conceptual event in the image content that implicitly dispersed around different image archives. In practice, an answer to these questions requires a series of field-based queries, across different digital archives. This thesis investigated the possibility to utilize standardized ontology, CIDOC-CRM, and semantic web tools, such as Protege, SWRL and Jess inference engine in order to model an inference platform. We discuss the need for a robust inference platform for real-life knowledge discovery and integration among heterogeneous image archives. Experimental results indicate the approach can be implemented in real-life applications.

Word Sense Disambiguation (WSD) is the process of resolving the meaning of a word unambiguously in a given natural language context. Within the scope of this thesis, it is the process of marking text with explicit sense labels. What constitutes a sense is a subject of great debate. An appealing perspective, aims to define senses in terms of their multilingual correspondences, an idea explored by several researchers, Dyvik (1998), Ide (1999), Resnik & Yarowsky (1999), and Chugur, Gonzalo & Verdejo (2002) but to date it has not been given any practical demonstration. This thesis is an empirical validation of these ideas of characterizing word meaning using cross-linguistic correspondences. The idea is that word meaning or word sense is quantifiable as much as it is uniquely translated in some language or set of languages. Consequently, we address the problem of WSD from a multilingual perspective; we expand the notion of context to encompass multilingual evidence. We devise a new approach to resolve word sense ambiguity in natural language, using a source of information that was never exploited on a large scale for WSD before. The core of the work presented builds on exploiting word correspondences across languages for sense distinction. In essence, it is a practical and functional implementation of a basic idea common to research interest in defining word meanings in cross-linguistic terms. We devise an algorithm, SALAAM for Sense Assignment Leveraging Alignment And Multilinguality, that empirically investigates the feasibility and the validity of utilizing translations for WSD. SALAAM is an unsupervised approach for word sense tagging of large amounts of text given a parallel corpus—texts in translation—and a sense inventory for one of the languages in the corpus. Using SALAAM, we obtain large amounts of sense annotated data in both languages of the parallel corpus, simultaneously. The quality of the tagging is rigorously evaluated for both languages of the corpora. The automatic unsupervised tagged data produced by SALAAM is further utilized to bootstrap a supervised learning WSD system, in essence, combining supervised and unsupervised approaches in an intelligent way to alleviate the resources acquisition bottleneck for supervised methods. Essentially, SALAAM is extended as an unsupervised approach for WSD within a learning framework; in many of the cases of the words disambiguated, SALAAM coupled with the machine learning system rivals the performance of a canonical supervised WSD system that relies on human tagged data for training. Realizing the fundamental role of similarity for SALAAM, we investigate different dimensions of semantic similarity as it applies to verbs since they are relatively more complex than nouns, which are the focus of the previous evaluations. We design a human judgment experiment to obtain human ratings on verbs' semantic similarity. The obtained human ratings are cast as a reference point for comparing different automated similarity measures that crucially rely on various sources of information. Finally, a cognitively salient model integrating human judgments in SALAAM is proposed as a means of improving its performance on sense disambiguation for verbs in particular and other word types in general.

Bilinguals demonstrate benefits in cognition (involving regulation of mental activity to resolve information) during processing. “Bilingual advance” is attributed to consistent management of two languages. Since bilinguals have a lifelong experience in controlling their two languages, they should have received more practice than monolinguals in processes engaging executive functions. In this study, an investigation was conducted on bilingual and monolingual students in Azarbaijan Shahid Madani University, East Azerbaijan province, Iran, in 2018. 100 students were selected using purposive sampling method. It is assumed that balanced bilinguals; those who are equally proficient in two languages experience a host of cognitive advantages rather than monolinguals. In this study, it was attempted to investigate if advantage of bilingualism extends to social cognition and executive function (i.e. motivation, inhibitory control, and empathy). For this purpose, Executive Function Index and Reading the Mind in the Eyes Test were used in bilingual and monolingual individuals. Results showed that bilinguals possess better performance than monolinguals merely in social cognition (t (98) =-4.37, p < 0.05) and inhibitory control (F (1, 98) = 10.95, p < 0.05) but not in motivation and empathy.

In this paper, we consider the role of semantics in the spoken dialogue translation systems. We begin by looking at some of the key properties of an existing spoken dialogue system, namely the sundial system which provides flight and train information over the telephone, and how these properties affect the design methodology and functionality of spoken translation systems. These properties include the effects of speech processing, designing the system to meet the needs of users, and an analysis model which clearly separates the linguistic, conceptual, pragmatic and task levels. In this model many task functionalities are dependent upon, and sometimes realizable by, the semantic and pragmatic analysis components. Central to this approach, is the use of underspecified semantic representations which are further specified as and when required by domain and/or task analysis. This model can be applied in the development of spoken translation systems with two important effects: monolingual semantic and pragmatic analysis can be carried out by processes independent of, but correlated with, the (translation) function of the system; and the main functions of the transfer processes is to further specifying the representations for the target language and to deal with mismatches between source and target language representations. We illustrate this approach with semantic analyses of German utterances required for translation in the verbmobil spoken dialogue translation system.

This article considers the task of automatically inducing role-semantic annotations in the FrameNet paradigm for new languages. We propose a general framework that is based on annotation projection...

Text reuse is the act of borrowing text (either verbatim or paraphrased) from an earlier written text. It could occur within the same language (mono-lingual) or across languages (cross-lingual) where the reused text is in a different language than the original text. Text reuse and its related problem, plagiarism (the unacknowledged reuse of text), are becoming serious issues in many fields and research shows that paraphrased and especially the cross-lingual cases of reuse are much harder to detect. Moreover, the recent rise in readily available multi-lingual content on the Web and social media has increased the problem to an unprecedented scale. To develop, compare, and evaluate automatic methods for mono- and crosslingual text reuse and extrinsic (finding portion(s) of text that is reused from the original text) plagiarism detection, standard evaluation resources are of utmost importance. However, previous efforts on developing such resources have mostly focused on English and some other languages. On the other hand, the Urdu language, which is widely spoken and has a large digital footprint, lacks resources in terms of core language processing tools and corpora. With this consideration in mind, this PhD research focuses on developing standard evaluation corpora, methods, and supporting resources to automatically detect mono-lingual (Urdu) and cross-lingual (English-Urdu) cases of text reuse and extrinsic plagiarism This thesis contributes a mono-lingual (Urdu) text reuse corpus (COUNTER Corpus) that contains real cases of Urdu text reuse at document-level. Another contribution is the development of a mono-lingual (Urdu) extrinsic plagiarism corpus (UPPC Corpus) that contains simulated cases of Urdu paraphrase plagiarism. Evaluation results, by applying a wide range of state-of-the-art mono-lingual methods on both corpora, shows that it is easier to detect verbatim cases than paraphrased ones. Moreover, the performance of these methods decreases considerably on real cases of reuse. A couple of supporting resources are also created to assist methods used in the cross-lingual (English-Urdu) text reuse detection. A large-scale multi-domain English-Urdu parallel corpus (EUPC-20) that contains parallel sentences is mined from the Web and several bi-lingual (English-Urdu) dictionaries are compiled using multiple approaches from different sources. Another major contribution of this study is the development of a large benchmark cross-lingual (English-Urdu) text reuse corpus (TREU Corpus). It contains English to Urdu real cases of text reuse at the document-level. A diversified range of methods are applied on the TREU Corpus to evaluate its usefulness and to show how it can be utilised in the development of automatic methods for measuring cross-lingual (English-Urdu) text reuse. A new cross-lingual method is also proposed that uses bilingual word embeddings to estimate the degree of overlap amongst text documents by computing the maximum weighted cosine similarity between word pairs. The overall low evaluation results indicate that it is a challenging task to detect crosslingual real cases of text reuse, especially when the language pairs have unrelated scripts, i.e., English-Urdu. However, an improvement in the result is observed using a combination of methods used in the experiments. The research work undertaken in this PhD thesis contributes corpora, methods, and supporting resources for the mono- and cross-lingual text reuse and extrinsic plagiarism for a significantly under-resourced Urdu and English-Urdu language pair. It highlights that paraphrased and cross-lingual cross-script real cases of text reuse are harder to detect and are still an open issue. Moreover, it emphasises the need to develop standard evaluation and supporting resources for under-resourced languages to facilitate research in these languages. The resources that have been developed and methods proposed could serve as a framework for future research in other languages and language pairs.

The purpose of this study is to explore how English language learners' (ELLs) consultation of monolingual and bilingual dictionaries could influence their second language (L2) vocabulary acquisition. Although research has empirically indicated that dictionary use could possibly affect learners' second language acquisition (SLA) in various ways, inconclusive results can still be found pertaining to ELLs' L2 vocabulary acquisition based on their dictionary use. In order to bridge this gap, sixty-four ELLs were recruited and divided into two groups for this experimental study. Through both quantitative and qualitative analyses, the researcher examined how the consultation of monolingual and bilingual dictionaries could influence L2 lexical teaching and learning utilizing a 13,944-word lemmatized High Frequency Word List (HFWL) from the British National Corpus (BNC). The findings suggest that the consultation of monolingual dictionaries promoted students' L2 vocabulary acquisition, learning interest, confidence and positive attitude more effectively than that of bilingual dictionaries. Learners were also revealed to be cognizant of the salience of L2 exposure that monolingual dictionaries possess for reifying their L2 awareness and lexical development in an authentic fashion. Based on the outcomes yielded from this investigation, pedagogical and theoretical implications are discussed.

Only a handful of the world's languages are abundant with the resources that enable practical applications of speech processing technologies. One of the methods to overcome this problem is to use the resources existing in other languages to train a multilingual automatic speech recognition (ASR) model, which, intuitively, should learn some universal phonetic representations. In this work, we focus on gaining a deeper understanding of how general these representations might be, and how individual phones are getting improved in a multilingual setting. To that end, we select a phonetically diverse set of languages, and perform a series of monolingual, multilingual and crosslingual (zero-shot) experiments. The ASR is trained to recognize the International Phonetic Alphabet (IPA) token sequences. We observe significant improvements across all languages in the multilingual setting, and stark degradation in the crosslingual setting, where the model, among other errors, considers Javanese as a tone language. Notably, as little as 10 hours of the target language training data tremendously reduces ASR error rates. Our analysis uncovered that even the phones that are unique to a single language can benefit greatly from adding training data from other languages - an encouraging result for the low-resource speech community.

The globalization of higher education has reshaped language use and practice on campus. As increasing numbers of international students change the campus into a more multilingual space, a variety of multilingual signs posted on bulletin boards and streets on campus become commonplace. Against such a backdrop of sociolinguistic changes on the globalizing campus in higher education, this study explores the ways in which multiple languages are chosen and used in South Korean higher education. Drawing on the framework of linguistic landscape, the present project analyzes multilingual signs displayed in public spaces of Konkuk University Glocal Campus, a local university that has pursued internationalization policies over the past decade. For this purpose, 208 public signs were collected from thirteen locations and analyzed in terms of language, producers, and communicative functions. The analysis reveals that a typical multilingual sign is in English and Korean and is created by the university authorities for referential functions. The most multilingual spaces are the library, a building for international education, and the dormitory, as these locations display various languages and language combinations. English is the most dominant foreign language in multilingual signage. Chinese and Vietnamese signs have begun to be increasingly present at KU, but their communicative functions are relatively limited. Based on such findings, implications for the university’s language and internationalization policies are also discussed.

This dissertation tracks what I term the public dispensability of language: the apportioning of particular language varieties to contexts deemed public by their users, and the dialectical, productive potential for language forms to assemble and shape publics of varying durability and form. The work of linguistic representation, as phenomena described in this dissertation demonstrate acutely, produces not only the people involved as participants in communication, but also the groupings to which they belong, against which they mobilize, through which they live their relationships to the world. These representational processes are deeply internalized and reproduced at both the most unremarkably quotidian and unabashedly spectacular scales. Based on two years of ethnographic fieldwork in literacy classrooms, Kurdish language classes, and wider community spaces in Turkey, I examine how language is used in public and how language is used to assemble publics, and explore the fraught politics of a would-be public Kurdish. I investigate how ways with text – reading, writing, brandishing, and circulating text objects – become indices of, and means to, the transformations of selves, and explore some of the consequences – personal, political, linguistic, moral – that ensue. Although the official ban on Kurdish (the largest minority language in Turkey) was lifted in 1991, the use of Kurdish in public life in Turkey remains restricted and politically fraught. Decades of violent confrontation between state security forces and Kurdish militants continue to frame “the Kurdish question” in Turkey, defining national debates over the future of democratic pluralism and national security. In recent years, the Kurdish political project has been increasingly pitched in terms of expanded linguistic rights, particularly for “mother tongue education.” Yet the majority of able speakers – women and older generations, who grew up in monolingual rural settings – are also the least likely to have had formal schooling or to have developed comprehensive literacy skills. A paradox thus unfolds: a newly emergent Kurdish language press produces texts (grammars, novels, newspapers, etc.) that few can read, since those who wield literacy are unlikely to know Kurdish, and those most likely to be proficient in Kurdish are unlikely to know how to read. How do expectations around language forms both presuppose and produce contexts as variously public, and with what stakes for negotiations of political and personal sovereignty? I examine how language ideologies and linguistic practices mediate the everyday productions of self- and group-making, and highlight the importance of understanding the political economy of language in Turkey – the politics around which language is deployed in what contexts, to what purposes, and with what histories of association. Key words: language politics; publics; literacy; Kurdish; Turkey

Natural language texts frequently contain related information in different positions in discourse. As human readers, we can recognize such information across sentence boundaries and correctly infer relations between them. Given this inference capability, we understand texts that describe complex dependencies even if central aspects are not repeated in every sentence. In linguistics, certain omissions of redundant information are known under the term ellipsis and have been studied as cohesive devices in discourse (Halliday and Hasan, 1976). For computational approaches to semantic processing, such cohesive devices are problematic because methods are traditionally applied on the sentence level and barely take surrounding context into account. In this dissertation, we investigate omission phenomena on the level of predicate-argument structures. In particular, we examine instances of structures involving arguments that are not locally realized but inferable from context. The goal of this work is to automatically acquire and process such instances, which we also refer to as implicit arguments, to improve natural language processing applications. Our main contribution is a framework that identifies implicit arguments by aligning and comparing predicate-argument structures across pairs of comparable texts. As part of this framework, we develop a novel graph-based clustering approach, which detects corresponding predicate-argument structures using pairwise similarity metrics. To find discourse antecedents of implicit arguments, we further design a heuristic method that utilizes automatic annotations from various linguistic pre-processing tools. We empirically validate the utility of automatically induced instances of implicit arguments and discourse antecedents in three extrinsic evaluation scenarios. In the first scenario, we show that our induced pairs of arguments and antecedents can successfully be applied to improve a pre-existing model for linking implicit arguments in discourse. In two further evaluation settings, we show that induced instances of implicit arguments, together with their aligned explicit counterparts, can be used as training material for a novel model of local coherence. Given discourse-level and semantic features, this model can predict whether a specific argument should be explicitly realized to establish local coherence or whether it is inferable and hence redundant in context.

Hungary is a monolingual state in Central Eastern Europe, where the Hungarian language, as the official language, is spoken by the whole population, including persons belonging to national and linguistic minorities. On the territory of Hungary, in the course of history, there have always lived representatives of other cultures and speakers of other languages. Nevertheless, in terms of the ability of speaking more than one language, within the European Union, Hungary is left behind, according to the latest Eurobarometer survey. In this paper we will highlight some of the facts and problems undermining real multilingualism in Hungary.

This paper examines the patterns of language use among immigrant children from Anglophone African countries living in Botswana. The study reveals major differences in their patterns of language use when compared to that of their parents. Three distinct language use patterns were identified; namely, children who are monolingual in English; those who are bilingual in English and the mother tongue, English and Kiswahili, or English and Setswana; and a third group who are multilingual in the mother tongue, English, Kiswahili and/ or Setswana. The study also reveals that whereas majority of the immigrant parents do not speak Setswana, in spite of their length of residence in Botswana, a good percentage of their children speaks Setswana. Another interesting finding is that many of the immigrant children who cannot speak their mother tongues are fluent speakers of Setswana, the dominant language of the host community. Data for this study were collected using two instruments; namely, a questionnaire designed for parents and an interview schedule for children aged between 18- 25. The questionnaire was administered to 100 immigrant parents from Ghana, Nigeria, Kenya, Tanzania, and Zambia. Fifty immigrant children from these countries were interviewed to get more insightful data on the patterns of language use among this population. Convenience sampling was used to identify the participants. The theoretical framework of this study was based on the “opportunity versus motivation” argument on language maintenance and language shift. The researcher recommends that immigrant children should be given more exposure and opportunities to learn and speak their mother tongue languages, especially in the family domain.

The English teaching faculty at the authors' institution contains a wide range of teachers with varying ability and inclination to use Chinese in the classroom. Teachers and students alike differ on whether explanations should be offered in the mother tongue, and to what degree. The authors surveyed 708 students across 15 language lab sections in an effort to determine whether classroom language policy matters, and if so, what the direction of its influence should be. We found a preponderance of subjective student support for bilingual instruction, but very little evidence that it makes a positive contribution to graded student performance. In fact, the study found strong evidence that students' preference for mother-tongue explanation is associated with poor grade performance, but the students expressing this preference included both those not getting any L1 instruction and those at risk of becoming too dependent upon it. Examination on of individual teachers and classes did reveal a significant association between pure monolingual instruction and low grades, but also suggested that this could be due as much to correlated traits like overall pedagogic strictness as to language policy per se. Exactly what other teaching traits are most closely correlated with monolingual instruction, and how they influence student learning provides an interesting venue for further research.

Machine-readable semantic knowledge in the form of taxonomies (i.e., a collection of is-a edges) has proved to be beneficial in an array of NLP tasks including inference, textual entailment, question answering and information extraction. Such widespread utility of taxonomies has led to multiple large-scale manual efforts towards taxonomy induction such as WordNet and Cyc. However, manual construction of taxonomies is time-intensive, and usually, requires substantial annotation efforts by domain experts. Furthermore, the resulting taxonomies suffer from low coverage and are unavailable for specific domains or languages. Therefore, in recent years, there has been a growing body of work, which aims to induce taxonomies automatically, either from semi-structured knowledge resources or unstructured text. In this thesis, we focus on the task of automated taxonomy induction under a variety of different settings. We first focus on taxonomy induction from the largest semi-structured knowledge resource, i.e., Wikipedia. More specifically, we introduce a set of novel heuristics aimed towards inducing a large-scale taxonomy from the English Wikipedia categories network. We also propose a novel comprehensive path-based evaluation framework for taxonomies. Taxonomy induced using our approach significantly outperforms the state of the art across both edge-based as well as path-based evaluation metrics. Moreover, our experiments also demonstrate that high accuracy in edge-based evaluation metrics does not always translate to high-accuracy in path-based evaluation metrics. Subsequently, we propose a novel approach, which leverages the interlanguage links of Wikipedia to induce taxonomies in other Wikipedia languages. Compared to the state of the art, our approach is simpler, more principled, and results in taxonomies that are significantly more accurate across both edge-based and path-based evaluation metrics. In the second part of this thesis, we focus on taxonomy induction from unstructured text. We propose a novel approach towards taxonomy induction from an input vocabulary of seed terms. Unlike previous approaches, which extract singular hypernym edges for terms, our approach utilizes a novel probabilistic framework to extract long-range hypernym subsequences. Taxonomy induction from the extracted subsequences is cast as an instance of the minimum-cost flow optimization problem on a carefully-designed flow network. Through experiments, we demonstrate that our approach outperforms the state of the art across four languages. We also show that our approach is robust to the presence of noise in the input vocabulary. Our approach facilitates the relaxation of many simplifying assumptions, which were employed by previous taxonomy induction approaches. As a result, our work serves to automate the process of taxonomy induction from unstructured text in the true sense. Finally, we introduce a task of discovering and generalizing lexicalized templates from the titles of Wikipedia entities. The experimental results on this task demonstrate that taxonomies, which perform better on our proposed path-based metrics, result in a more accurate set of generalizations for a given set of entities. In summary, this thesis proposes new approaches towards automated taxonomy induction. It improves upon the state of the art in a variety of different settings. It also relaxes many simplifying assumptions that limited the applicability of prior approaches.