La tâche d'alignement d'un texte dans une langue source avec sa traduction en langue cible est souvent nommée alignement de bi-textes. Elle a pour but de faire émerger les relations de traduction qui peuvent s'exprimer à différents niveaux de granularité entre les deux faces du bi-texte. De nombreuses applications de traitement automatique des langues naturelles s'appuient sur cette étape afin d'accéder à des connaissances linguistiques de plus haut niveau.Parmi ces applications, nous pouvons citer bien sûr la traduction automatique, mais également l'extraction de lexiques et de terminologies bilingues, la désambigüisation sémantique ou l'apprentissage des langues assisté par ordinateur.La complexité de la tâche d'alignement de bi-textes s'explique par les différences linguistiques entre les langues. Ces différences peuvent être d'ordre sémantique, syntaxique, ou morphologique.Dans le cadre des approches probabilistes, l'alignement de bi-textes est modélisé par un ensemble de variables aléatoires cachés. Afin de réduire la complexité du problème, le processus aléatoire sous-jacent fait l'hypothèse simplificatrice qu'un mot en langue source est lié à au plus un mot en langue cible, ce qui induit une relation de traduction asymétrique. Néanmoins, cette hypothèse est simpliste, puisque les alignements peuvent de manière générale impliquer des groupes de mots dans chacune des langues. Afin de rétablir cette symétrie, chaque langue est considérée tour à tour comme la langue source et les deux alignements asymétriques résultants sont combinés à l'aide d'une heuristique. Cette étape de symétrisation revêt une importance particulière dans l’approche standard en traduction automatique, puisqu'elle précède l'extraction des unités de traduction, à savoir les paires de segments.L'objectif de cette thèse est de proposer de nouvelles approches pour d'une part l'alignement debi-texte, et d'autre part l'extraction des unités de traduction. La spécificité de notre approche consiste à remplacer les heuristiques utilisées par des modèles d'apprentissage discriminant.Nous présentons un modèle "Maximum d'entropie'' (ou MaxEnt) pour l'alignement de mot, pour lequel chaque lien d'alignement est prédit de manière indépendante. L'interaction entre les liens d'alignement est alors prise en compte par l'empilement ("stacking'') d'un second modèle prenant en compte la structure à prédire sans pour autant augmenter la complexité globale. Cette formulation peut être vue comme une manière d'apprendre la combinaison de différentes méthodes d'alignement: le modèle considère ainsi l'union des alignements d'entrées pour en sélectionner les liens jugés fiables. Le modèle MaxEnt proposé permet d'améliorer les performances d'un système état de l'art de traduction automatique en considérant le jeu de données de la tâche NIST'09, Arabe vers Anglais. Ces améliorations sont mesurées en terme de taux d'erreur sur les alignements et aussi en terme de qualité de traduction via la métrique automatique BLEU.Nous proposons également un modèle permettant à la fois de sélectionner et d'évaluer les unités de traduction extraites d'un bi texte aligné. Ces deux étapes sont reformulées dans le cadre de l'apprentissage supervisé, afin de modéliser la décision de garder ou pas une paire de segments comme une unité fiable de traduction. Ce cadre permet l'utilisation de caractéristiques riches et nombreuses favorisant ainsi une décision robuste. Nous proposons une méthode simple et efficace pour annoter les paires de segments utiles pour la traduction. Le problème d'apprentissage automatique qui se pose alors est particulier, puisque nous disposons que d'exemples positifs. Nous proposons donc d'utiliser l'approche SVM à une classe afin de modéliser la sélection des unités de traduction.Grâce à cette approche, nous obtenons des améliorations significatives en terme de score BLEU pour un système entrainé avec un petit ensemble de données. / Bitext alignment is the task of aligning a text in a source language and its translation in the target language. Aligning amounts to finding the translational correspondences between textual units at different levels of granularity. Many practical natural language processing applications rely on bitext alignments to access the rich linguistic knowledge present in a bitext. While the most predominant application for bitexts is statistical machine translation, they are also used in multilingual (and monolingual) lexicography, word sense disambiguation, terminology extraction, computer-aided language learning andtranslation studies, to name a few.Bitext alignment is an arduous task because meaning is not expressed seemingly across languages. It varies along linguistic properties and cultural backgrounds of different languages, and also depends on the translation strategy that have been used to produce the bitext.Current practices in bitext alignment model the alignment as a hidden variable in the translation process. In order to reduce the complexity of the task, such approaches suppose that a word in the source sentence is aligned to one word at most in the target sentence.However, this over-simplistic assumption results in asymmetric, one-to-many alignments, whereas alignments are typically symmetric and many-to-many.To achieve symmetry, two one-to-many alignments in opposite translation directions are built and combined using a heuristic.In order to use these word alignments in phrase-based translation systems which use phrases instead of words, a heuristic is used to extract phrase pairs that are consistent with the word alignment.In this dissertation we address both the problems of word alignment and phrase pairs extraction.We improve the state of the art in several ways using discriminative learning techniques.We present a maximum entropy (MaxEnt) framework for word alignment.In this framework, links are predicted independently from one another using a MaxEnt classifier.The interaction between alignment decisions is approximated using stackingtechniques, which allows us to account for a part of the structural dependencies without increasing the complexity. This formulation can be seen as an alignment combination method,in which the union of several input alignments is used to guide the output alignment. Additionally, input alignments are used to compute a rich set of feature functions.Our MaxEnt aligner obtains state of the art results in terms of alignment quality as measured by thealignment error rate, and translation quality as measured by BLEU on large-scale Arabic-English NIST'09 systems.We also present a translation quality informed procedure for both extraction and evaluation of phrase pairs. We reformulate the problem in the supervised framework in which we decide for each phrase pair whether we keep it or not in the translation model. This offers a principled way to combine several features to make the procedure more robust to alignment difficulties. We use a simple and effective method, based on oracle decoding,to annotate phrase pairs that are useful for translation. Using machine learning techniques based on positive examples only,these annotations can be used to learn phrase alignment decisions. Using this approach we obtain improvements in BLEU scores for recall-oriented translation models, which are suitable for small training corpora.

Submitted by Filipe dos Santos (fsantos@pucsp.br) on 2018-09-26T10:04:55Z No. of bitstreams: 1 Ulysses Camargo Corrêa Diegues.pdf: 1531825 bytes, checksum: 013effa494f4018a825471efea8d13a1 (MD5) / Made available in DSpace on 2018-09-26T10:04:55Z (GMT). No. of bitstreams: 1 Ulysses Camargo Corrêa Diegues.pdf: 1531825 bytes, checksum: 013effa494f4018a825471efea8d13a1 (MD5) Previous issue date: 2018-08-23 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / In a scenario which more and more the process of selecting candidates for job becomes more demanding (JOSEPH, 2013), the study of job interviews in English is of great importance. However, it has been receiving little attention in language studies. The purpose of this research is to compare the English job interview register with the other English language registers along the five dimensions of variation identified by Biber (1988 et seq.) through Multidimensional Analysis (MDA). To do so, this research is based on Corpus Linguistics (CL) that deals with the collection and exploitation of corpora with the purpose of helping to research a language or part of it. (BERBER SARDINHA, 2000; 2004). The corpus of this study was the Job Interview Corpus (JIC), composed of 40 real job interviews conducted in Germany with native speakers from Australia, Ireland, the United Kingdom and the United States, totaling approximately 50,000 words. In order to enable MDA, the study corpus, JIC, was grammatically tagged with the Biber Tagger and later processed by the Biber Tag Count, which calculated the frequency of 67 linguistic variables considered in this study. The MDA results showed how the English job interviews of the study corpus, JIC, resemble or differentiate from the other English language registers along the five dimensions of variation (BIBER, 1988 et seq.). Since there are no precedents of studies within the CL devoted to the investigation of English job interviews in a multidimensional analysis, this research intends to fill this gap in the academic field / Em um cenário em que cada vez mais o processo de seleção de candidatos a uma vaga de emprego se torna mais exigente (JOSEPH, 2013), o estudo das entrevistas de emprego em inglês é de grande importância. No entanto, o tema tem recebido pouca atenção nos estudos linguísticos. O objetivo desta pesquisa é comparar o registro entrevista de emprego em inglês com os outros registros da Língua Inglesa ao longo das cinco dimensões de variação identificadas por Biber (1988 et seq.) por meio da Análise Multidimensional (AMD). Para tanto, esta pesquisa se fundamenta na Linguística de Corpus (LC) que se ocupa da coleta e exploração de corpora com a finalidade de servir para uma pesquisa de uma língua (BERBER SARDINHA, 2000; 2004). O corpus de estudo utilizado nesta pesquisa foi o Job Interview Corpus (JIC), composto por 40 entrevistas de emprego reais realizadas na Alemanha com falantes nativos oriundos da Austrália, Estados Unidos, Irlanda e Reino Unido, totalizando aproximadamente 50 mil palavras. A fim de viabilizar a AMD, o corpus de estudo, JIC, foi etiquetado gramaticalmente com a ferramenta computacional Biber Tagger e posteriormente processado pelo Biber Tag Count que calculou a frequência das 67 variáveis linguísticas consideradas neste estudo. Os resultados da AMD demonstraram como as entrevistas de emprego em inglês presentes no corpus de estudo, JIC, se assemelham ou se diferencia dos demais registros da Língua Inglesa ao longo das cinco dimensões de variação (BIBER, 1988 et seq.). Uma vez que não há precedentes de estudos dentro da LC dedicados à investigação das entrevistas de emprego em inglês em uma análise multidimensional, a presente pesquisa pretende preencher esta lacuna na área acadêmica

華語三聲變調的運作範疇(domain)一直以來皆是漢語音韻學家激烈探討的主題，不同研究對於運作範疇的定義皆不同，而過去研究也留下了一些尚待解決的問題，例如分類詞(Classifier)以及介係詞(Preposition)如何與其他音節形成三聲變調的運作範疇。本篇論文藉由觀察語料庫中324句最短為6音節，最長為24音節且由數個三聲音節所組成的中文句子，來檢視台灣華語三聲變調若以Nespor & Vogel (1986)所提出之韻律階層(Prosodic hierarchy)為範疇來運作是否能得出自然且多樣的聲調組合。以下分析主要根據韻律階層的順序分成以下四部分：音韻詞(Phonological word)、附著詞組(Clitic group)、音韻詞組(Phonological phrase)、語調詞組(Intonational phrase)。而作者假設三聲變調有兩種運作方式，一種為循環(Cyclic)，一種為同時(Simultaneous)。循環運作時，三聲變調可先以音韻詞邊界為範疇，再以附著詞組邊界為範疇，再以音韻詞組邊界為範疇，最後再以語調詞組邊界為範疇。同時運作時，三聲變調則只以語調詞組邊界為範疇。語料庫分析結果顯示，中心語(Head)之非遞歸邊(Nonrecursive side)的分支性(Branchingness)需被納入音韻詞組的定義中以得出正確的聲調組合。再者，語調詞組的定義也需將語意(Semantic)以及韻律(Metric)的因素納入考量，以限制語調詞組邊界的形成。經過修正，韻律階層理論不僅能成功描述分類詞以及介係詞之三聲變調現象更可預測華語三聲變調的多樣性。 / Mandarin tone sandhi has been argued by different scholars to apply to different phonological domains such as the foot, the phonological phrase etc. This thesis tackles three main issues left by Shih (1986) and Hsiao (1991): the classifier, the preposition and the under-generation problem. In revising the framework of Nespor & Vogel (1986)’s Prosodic Hierarchy, this thesis proposes a possible explanation of the problems. This thesis constructed a corpus of 324 sentence tokens consisting of 6 to 24 third tone syllables. The tokens were recorded from four Taiwan Mandarin native speakers at a speech rate of at least 200 beats per minute. Through the observation of the tone patterns collected, it is found that the original definitions of the phonological phrase and the intonational phrase given by Nespor & Vogel (1986) are not able to describe the Taiwan Mandarin data. This thesis provides a revision of the prosodic hierarchy by taking into account the branchingness of the nonrecursive side on the phonological phrase level and by incorporating Selkirk (1984)’s Sense Unit Condition on the intonational phrase level. Finally, a foot formation condition is also proposed in order to restrict the minimal length of a phonological phrase or an intonational phrase.

This thesis is a textbook designed to assist LDS missionaries studying the Samoan language both in a two-month intensive language-learning school (the Missionary Training Center) and in individual study in the Samoan Islands. The thesis is prefaced with a rationale for and explanation of the Missionary Training Center language program and the design and purposes of the text itself.The textbook contains fifty Samoan language lessons, divided into ten units of five lessons each. At the end of each unit is a review module, or test. Except for two overview lessons (designed to introduce grammatical terms and pronunciation), each of the fifty lessons is either a grammar lesson (consisting of presentation and practice of grammar rules sequenced from simple to complex) or a Speak Your Language lesson (consisting of presentation and practice of phrases, patterns, and vocabulary required for competence in particular notions and situations, sequenced according to the communicative needs of the missionaries). Supplementary material included at the end of the book includes Extra Mile Lessons (additional information on grammar, phrases, patterns, and vocabulary) and two appendices (Samoan songs and a bibliography).

Cette thèse se centre sur les discordances dans l’ellipse périphérique (RNR) et propose une analyse basée sur l’identité de lexème entre le matériel manquant et le matériel périphérique. Les analyses précédentes accordent de l’importance au syncrétisme, ou identité phonologique (Pullum & Zwicky (1986)). Dans cette thèse, nous contestons cette hypothèse. Nous avons analysé 5 types de discordance dans l’ellipse périphérique: discordances de polarité, de possessifs, de prépositions, de voix et de formes verbales. Nos études de corpus sur Internet, sur Frtenten 2012 (1.6 milliard de mots) (Baroni et al. (2009)) et sur le English Web 2013 (19 milliard mots) montrent l’existence des discordances dans l’ellipse périphérique en français et en anglais. Les discordances sont assez nombreuses même dans des écrits soignés. Dans tous les cas, les discordances sont résolues par la forme qui correspond au second conjoint. Les résultats des expériences de jugements d’acceptabilité et de mouvements oculaires permettent d’intégrer les discordances dans la grammaire. Les résultats sont compatibles avec les analyses qui postulent l’identité sémantique entre le matériel manquant et l’antécédent pour l’ellipse. Nous proposons une analyse formelle en HPSG.Nous comparons les résultats obtenus avec les cas de coordination lexicale. Nous montrons que l’accord de proximité s’applique (Villavicencio et al. (2005)) et nous proposons une analyse HPSG pour la coordination de verbes et de prépositions. / This thesis focuses on mismatches in peripheral ellipsis (RNR) and proposes an analysis based on lexeme identity between the missing material and the peripheral material. Previous analyses put emphasis on syncretism, or phonological identity (Pullum & Zwicky (1986)). In this thesis, we challenge this hypothesis. We analyzed 5 types of mismatches in peripheral ellipsis: polarity mismatch, possessive mismatch, voice mismatch and verbal form mismatch. Our corpus studies on the Internet, on Frtenten 2012 (1.6 billion words) (Baroni et al. (2009)) and on English Web 2013 (19 billion words) shows the existence of mismatches in peripheral ellipsis in French and English. Mismatches are quite numerous even in careful writings. In all cases, the mismatches are resolved by the form that corresponds to the second conjunct. The results of acceptability judgment tests and eye tracking experiments allow the integration of these mismatches into the grammar. The results are compatible with analyses postulating semantic identity between the missing material and the antecedent for ellipsis. We formalize peripheral ellipsis with mismatch within HPSG.We finally compare our results with lexical coordination. We show that it obeys closest conjunct agreement (Villavicencio et al (2005)) and propose a HPSG analysis for coordination of verbs and prepositions.

This study is a detailed investigation of a number of issues in colloquial Sinhala morphosyntax. These issues primarily concern grammatical relations, argument structure, phrase structure and focus constructions. The theoretical framework of this study is Lexical Functional Grammar.¶Chapter 1 introduces the issues to be discussed, followed by a brief introduction of some essential aspects of colloquial Sinhala as background for the discussion in the following chapters. In Chapter 2 we present basic concepts of the theoretical framework of Lexical Functional Grammar.¶ The next three chapters mainly concern grammatical relations, argument structure and clause structure in colloquial Sinhala. Chapter 3 examines grammatical relations. The main focus lies in establishing the subject grammatical relation in terms of various subjecthood diagnostics. We show that only a very small number of diagnostics are reliable, and that the evidence for subject is weaker than assumed previously. All the subjecthood diagnostics that were examined select the most prominent argument in the argument structure as the subject, i.e. 'logical subject'. However, there appear to be no processes in the language that are sensitive to the subject in the grammatical relations structure, i.e. 'gr-subject'. Further, there is no evidence for other grammatical relations like objects. In Chapter 4 we discuss the agentless construction and related valency alternation phenomena. It was previously assumed that the agentless construction, valency alternation phenomena and the involitive construction are all related. We argue that the agentless construction should be treated as a different construction from the involitive construction. We also show that the agentless construction and the involitive construction have contrasting characteristics, and that treatment of them as separate constructions can account for some phenomena which did not receive an explanation previously. The valency alternation phenomena are related to the agentless construction, therefore there is no valency alternation in involitive constructions. It will be shown that verbs undergoing the valency alternation can be distinguished from the other verbs in terms of the lexical semantic properties of individual verbs. Chapter 5 examines the structure of non-verbal sentences in terms of a number of morphosyntactic phenomena. It was previously argued that verbal sentences and non-verbal sentences in colloquial Sinhala differ in terms of clause structure. However, the present study shows evidence to the contrary.¶ The next two chapters deal with modelling contrastive focus and the phrase structure of the language. Chapter 6 is a detailed analysis of the contrastive focus (cleft) construction in various clause types in the language, and proposes a unified syntactic treatment of contrastive focus. Contrastive focus is in some constructions morphologically encoded, while in others it involves both morphological and configurational assignment of focus. The complex interaction between focus markers and verb morphology in various focus constructions is accounted for by general well-formedness conditions applying to the f-structure, and the principles of Functional Uncertainty and Morphological Blocking. In Chapter 7, we discuss the phrase structure of the language, in particular such issues as its non-configurational nature and the lack of evidence for VP. We propose non-configurational S and some functional projections to account for word order freedom under S and to explain certain morphosyntactic phenomena, such as configurational focus assignment. Finally, Chapter 8 summarises the conclusions made in previous chapters.

本文主要是探討漢語短語及詞彙層次的輕聲現象。關於短語性的輕聲，鄭良偉先生(1989)是從句法的觀點來分析，本文則採用韻律体系理論來探討。我們提議了一組焦點詞語的原則及幾個相關的制約。此外，我們從蕭宇超先生(1991 & 1994)的韻板計數理論來觀察輕聲的節奏問題，因而修訂了虛詞音板指派及音板連接方向的原則，也建議了一條抵輔音板加入的規則。 　　關於詞彙性的輕聲，我們採用McCarthy及Prince(1986)的構詞規型理論及Kiparsky(1973 & 1982)，Mohanan(1982 & 1986)的字構音韻學理論來探討此一現象。我們建議了詞彙性的輕聲基本上運作於楊抑格音步，然而抵輔調卻運作於抑楊格音步。觀察了相關的音韻及構詞規則，我們也發現到文獻上輕聲的語料可由層級的觀念來分析。 / This thesis deals with neutral tone sandhi in Mandarin Chinese at the phrasal and lexical levels. Previous research of the phrasal neutral tone, e.g., R. Cheng (1989), has focused only at the syntactic level. In this research, we employ prosodic theories (Selkirk (1984 & 1986), Nespor and Vogel (1986), etc.) to condition phrasal neutral tone sandhi, and propose a set of focal phrasing principles and a couple of relevant constraints. In addition, we discuss the rhythm of neutral tone under the framework of the beat-counting theory (Hsiao, 1991 & 1994) to see in what ways the theory has to be modified. We propose a pair of revised principles of functor beat assignment and directional beat adjunction, along with a backup rule of the default beat addition. 　　As to neutral tone at the lexical level, we incorporate morphological template theory (McCarthy and Prince (1986)), and lexical phonology theory (Kiparsky (1973 & 1982), Mohanan (1982 & 1986), etc.). We propose that lexical neutral tone target a trochaic foot template, while a default tone target an iambic foot template. Observing the relevant phonological and morphological rules, we find that the neutral-toned data in the literature can be somehow generalized by way of level ordering.

<p>English noun phrases (NP) which include degree modified adjectives show some interesting variation of the position of the indefinite article. A particularly salient pattern is displayed in <em>This is anticipated to be <strong>more common a scenario</strong> than fleas spreading bubonic plague </em>(BoE, BU-NX022521)<em>.</em> The present paper is based on a study of utterances where this pattern was used even though a canonical word order would have been possible. Such constructs are referred to as the <em>Optional Postposed Indefinite Article Noun Phrase </em>(OPIANP) and have been collected from the British National Corpus (BNC) and <em>Collins Word Banks Online: English Corpus</em> (BoE). The central question is whether there is semantic motivation for this postposition of the indefinite article. The results suggest that there is such motivation, namely that the OPIANP could be an extension of a more frequent construction identified as the <em>Postposed Indefinite Article Noun Phrase</em> (PIANP). Furthermore, it is shown that the pattern’s semantics is unpredictable from the composition of its parts and that its primary function is that it positions already given arguments on an adjectival scale. That is, it foregrounds scalar qualities and backgrounds the noun. These conclusions stem from observations of patterns of unification with other constructions, illustrating how the OPIANP unifies best with the non-referential, descriptive PC-constructions and less well with referential constructions such as the subject and direct object constructions. These findings are remarkable as the idea of an adjective-scalar centred NP-construction challenges the idea of NPs being centred round their head, the noun.</p>

<p>In this thesis I have investigated verb frame divergences in a bilingual Head-driven Phrase Structure Grammar for machine translation. The purpose was threefold: (1) to describe and classify verb frame divergences (VFDs) between Swedish and English, (2) to practically implement a bilingual grammar that covered many of the identified VFDs and (3) to find out what cases of VFDs could be solved and implemented using a common semantic representation, or interlingua, for Swedish and English.</p><p>The implemented grammar, BiTSE, is a Head-driven Phrase Structure Grammar based on the LinGO Grammar Matrix, a language independent grammar base. BiTSE is a bilingual grammar containing both Swedish and English. The semantic representation used is Minimal Recursion Semantics (MRS). It is language independent, so generating from it gives all equivalent sentences in both Swedish and English. Both the core of the languages and a subset of the identified VFDs are successfully implemented in BiTSE. For other VFDs tentative solutions are discussed.</p><p>MRS have previously been proposed as suitable for semantic transfer machine translation. I have shown that VFDs can naturally be handled by an interlingual design in many cases, minimizing the need of transfer.</p><p>The main contributions of this thesis are: an inventory of English and Swedish verb frames and verb frame divergences; the bilingual grammar BiTSE and showing that it is possible in many cases to use MRS as an interlingua in machine translation.</p>

University of Macau / Faculty of Social Sciences and Humanities / Department of Portuguese