The goal of this thesis is to advance the use of recurrent neural network language models (RNNLMs) for large vocabulary continuous speech recognition (LVCSR). RNNLMs are currently state-of-the-art and shown to consistently reduce the word error rates (WERs) of LVCSR tasks when compared to other language models. In this thesis we propose various advances to RNNLMs. The advances are: improved learning procedures for RNNLMs, enhancing the context, and adaptation of RNNLMs. We learned better parameters by a novel pre-training approach and enhanced the context using prosody and syntactic features. We present a pre-training method for RNNLMs, in which the output weights of a feed-forward neural network language model (NNLM) are shared with the RNNLM. This is accomplished by first fine-tuning the weights of the NNLM, which are then used to initialise the output weights of an RNNLM with the same number of hidden units. To investigate the effectiveness of the proposed pre-training method, we have carried out text-based experiments on the Penn Treebank Wall Street Journal data, and ASR experiments on the TED lectures data. Across the experiments, we observe small but significant improvements in perplexity (PPL) and ASR WER. Next, we present unsupervised adaptation of RNNLMs. We adapted the RNNLMs to a target domain (topic or genre or television programme (show)) at test time using ASR transcripts from first pass recognition. We investigated two approaches to adapt the RNNLMs. In the first approach the forward propagating hidden activations are scaled - learning hidden unit contributions (LHUC). In the second approach we adapt all parameters of RNNLM.We evaluated the adapted RNNLMs by showing the WERs on multi genre broadcast speech data. We observe small (on an average 0.1% absolute) but significant improvements in WER compared to a strong unadapted RNNLM model. Finally, we present the context-enhancement of RNNLMs using prosody and syntactic features. The prosody features were computed from the acoustics of the context words and the syntactic features were from the surface form of the words in the context. We trained the RNNLMs with word duration, pause duration, final phone duration, syllable duration, syllable F0, part-of-speech tag and Combinatory Categorial Grammar (CCG) supertag features. The proposed context-enhanced RNNLMs were evaluated by reporting PPL and WER on two speech recognition tasks, Switchboard and TED lectures. We observed substantial improvements in PPL (5% to 15% relative) and small but significant improvements in WER (0.1% to 0.5% absolute).

La mention de référents (personnes, objets et évènements) est au coeur du discours. Cette thèse porte sur l’étude de la variation d’emploi des marqueurs de référence, en combinant des indices syntaxiques et prosodiques, produits lors de narrations d’images séquentielles. L’objectif est de déterminer l’effet des paramètres situationnels sur la production des marqueurs syntaxiques et prosodiques. Deux autres objectifs sont également développés, l’un visant à mettre en évidence les compétences (socio)cognitives sous-jacentes et l’autre visant à étudier l’impact du vieillissement. Pour répondre à ces objectifs, une tâche de narration d’images séquentielles a été utilisée pour éliciter la production de marqueurs de référence. Les narrations d’images séquentielles ont été réalisées auprès de deux groupes de participants, l’un constitué de 30 jeunes adultes (19-39 ans) et l’autre de 30 séniors (59-79 ans). L’ensemble des participants a été soumis à des tests (socio)cognitifs afin de caractériser leurs profils.Nos résultats montrent que le marquage référentiel est un processus complexe et influencé par de multiples facteurs situationnels. En effet, les marqueurs syntaxiques et prosodiques varient en fonction des étapes de discours et du contexte référentiel. Selon l’accessibilité cognitive du référent – notamment l’accessibilité cognitive moindre, le marquage référentiel sollicite des compétences (socio)cognitives. Les résultats indiquent également des effets du vieillissement sur le marquage référentiel : l’emploi des marqueurs de référence chez les séniors est différent de l’emploi des jeunes adultes. Compte tenu de l’importance du marquage référentiel lors des conversations, nos résultats pourraient avoir des implications cliniques et théoriques. / Referring to people, objects and events is a cental piece of the Speech. This dissertation studies the variation in the use of referential markers, by combining syntactic and prosodic cues, produced during a storytelling task sequential pictures. The goal is to determine the effect of situational parameters on the production of the syntactic and prosodic markers. Two other goals are also pursued: (i) to highlight the underlying (socio)cognitive abilities and (ii) to study the impact of aging. To meet these objectives, a storytelling task with sequential pictures was used to obtain the production of referential markers. 30 younger (age span:19-39) and 30 older participants (age span : 59-79) undertook this storytelling task. All the participants were submitted to (socio)cognitive tests to characterize their profiles. Our results show that referential marking is a complex process influenced by multiple situational factors. Indeed, the syntactic and prosodic markers vary according to discourse stages and the referential context. Depending on the cognitive accessibility of the referent - in particular the lesser cognitive accessibility, referential marking requires (socio)cognitive abilities. The results also indicate effects of aging on referential marking : the use of referential markers is different in seniors and in younger participants. Given the importance of referential marking during conversations, our results may have some clinical implications.

Cette thèse s’intéresse aux fonctions des marqueurs discursifs ‘you know’, ‘then’ et ‘ya‘ni’ (I mean) en tant que structures autonomes du point de vue syntaxique et en tant qu’unités linguistiques opérationnelles et multifonctionnelles du point de vue conversationnel. Dans une perspective pragmatique, ce travail explore la corrélation entre la position et la fonction en maintenant que la valeur d’un marqueur en position initiale est différente de celle exprimée en position médiane ou en position finale. Dans le contexte des interactions verbales politiques au sein des émissions télévisées en arabe et en anglais, le but de cette thèse est de contribuer aux analyses conversationnelles pragmatiques. À travers une recherche empirique, l’analyse montre que la multifonctionnalité de ces items linguistiques est liée à leur flexibilité syntaxique. Ces marqueurs discursifs assument diverses fonctions contextuelles qui évoluent continuellement sur une échelle pragmatique. L’évolution pragmatique de ‘you know’, ‘then’ et ‘ya‘ni’ engendre des expressions complexes au niveau de leur sémantisme. Ces unités pragmatiques deviennent plus ambiguës car elles s’éloignent davantage de leur sens de base en acquérant des significations contextuelles. Ainsi, ces marqueurs évoquent d’autres interprétations et ne peuvent se limiter qu’à leur sens d’origine. En l’occurrence, on leur attribue des équivalents les plus proches selon le contexte de leur occurrence.Les résultats révèlent que ‘ya‘ni’ peut être substitué par d’autres marqueurs de différentes catégories grammaticales en anglais. Cette unité polysémique, polyfonctionnelle, poly-équivalente et poly-catégorielle partage des fonctions avec ‘you know’, ‘then’, ‘I think’, ‘so’, ‘in fact’ et d’autres marqueurs. La variation pragmatique suit le but illocutoire du locuteur, la prise en compte d’autrui et l’organisation de l’interaction verbale. Dans deux situations socioculturelles différentes, l’anglais et l’arabe confirment que ces items linguistiques sont des unités conversationnelles contextuelles et multifonctionnelles. Leur rôle est visiblement actif dans une situation sociale où l’interaction entre le locuteur et l’interlocuteur est un facteur saillant ; tel est le cas des échanges verbaux télévisés. / This PhD thesis deals with the pragmatic functions of ‘you know’, ‘then’ and ‘ya‘nĩ’ (I mean) as autonomous syntactic structures and as operational and multifunctional linguistic units in conversation. Within a pragmatic framework, the research discusses the correlation between position and function, in which the pragmatic value of a marker in initial position is different from that conveyed in medial or final positions. The goal of this study is to contribute to the pragmatic analysis of conversations by analyzing political verbal interactions in Arabic and English television broadcasts. The results of this empirical study show that the multi-functionality of these linguistic items is related to their syntactic flexibility. These discourse markers imply a variety of contextual functions which evolve gradually on a pragmatic scale.The pragmatic evolution of ‘you know’, ‘then’ and ‘ya‘nĩ’ generate complex semantic expressions. These pragmatic units in fact become increasingly complex; they go beyond their basic meaning to acquire progressively contextual implications. Thus, these markers refer to other interpretations and transcend their immediate semantic base. In this respect, they can be substituted according to the context of their occurrence. The results reveal that ‘ya‘nĩ’ can be substituted with other markers from different grammatical categories in English. This polysemous, multi-functional, poly-equivalent and poly-categorical conversational unit shares its pragmatic functions with ‘you know’, ‘then’, ‘I think’, ‘so’, ‘in fact’ and other discourse markers. Pragmatic variation depends on the illocutionary perspective of the speaker, the relation with the hearer and the organization of the verbal interaction. In two distinct sociocultural situations, English and Arabic, it is confirmed that these linguistic items are contextual and multifunctional conversational units. Their role is relevant in a social situation where the interaction between the speaker and the hearer is a salient factor, as in the case of political verbal exchanges in television broadcasts.

[pt] Essa dissertação investiga uma possível relação entre comprometimentos no domínio sintático e dificuldades de leitura. Mais especificamente, busca-se verificar se dificuldades na compreensão oral de estruturas altamente custosas – interrogativas QU mais N de objeto (OWH mais N) e relativas de objeto (ORCs) – preveem problemas na compreensão leitora, quando tais estruturas estão envolvidas. 78 alunos (idade média: 12) do sexto ano de três escolas públicas do Rio de Janeiro participaram desse estudo. Suas habilidades sintáticas foram inicialmente testadas. Dois grupos foram criados para participarem dos testes de leitura: com possível comprometimento sintático (SI) (n igual 25) e controle (CT) (n igual 53). Um teste de reconhecimento de palavras e outro de leitura de palavras/pseudopalavras isoladas foram elaborados e realizados, uma vez que a fluência em leitura requer que dificuldades nessas habilidades sejam superadas. Novos grupos (SI e CT), sem problemas nesse nível, foram definidos (com 12 participantes cada). Dois experimentos foram conduzidos, buscando testar suas habilidades de fluência (velocidade, precisão e prosódia) em leitura com OWH mais N e ORCs em sentenças isoladas (tarefa de identificação de imagens) e no discurso (tarefa de leitura automonitorada). No último caso, o sujeito interveniente foi manipulado quanto à complexidade estrutural (DP completo ou pronome). Um aspecto da prosódia (uso adequado de pitch) distinguiu os grupos, com menor desempenho no grupo SI em ambas as tarefas. Um aspecto da precisão (número de disfluências) também os distinguiu quando as sentenças investigadas foram apresentadas no discurso (mais disfluências no grupo SI). A compreensão foi particularmente afetada quando as sentenças foram lidas isoladamente. Fatores relacionados à escolha lexical, bem como fatores discursivos podem criar demandas diferenciadas para as sentenças no discurso e isso pode ter minimizado efeitos de grupo. O efeito de intervenção foi obtido na direção prevista nas interrogativas QU mais N (OWH mais N), em que uma demanda maior foi observada nas condições de DP completos. Já no caso das relativas de objeto (ORC), um efeito significativo não foi obtido, possivelmente devido a dificuldades na atribuição de força ilocucionária interrogativa às perguntas SIM /NÃO. Uma amostra de sentenças com diferentes estruturas e com pontuação variada foi examinada, a fim de se obter uma análise geral de fluência em leitura. Novamente o pitch é indicativo de comprometimento sintático. Comprometimentos na interface sintaxe-prosódia podem explicar esses resultados. / [en] This dissertation intends to verify whether syntactic impairment, as detected in in the oral comprehension of costly structures - object WH plus N questions (OWH plus N) and object relative clauses (ORCs) -, predicts difficulties in reading fluency and comprehension. The syntactic abilities of 6th graders from three public schools in Rio de Janeiro were evaluated, giving rise to syntactically impaired (SI) and control (CT) groups. Those who satisfied criteria at the word level in word-recognition/reading tasks proceeded. Two experiments assessed reading fluency (rate, accuracy, prosody) and the comprehension of each of the target structures in isolation (picture-identification task) and in discourse (self-paced reading task). In the latter, the intervening subject was manipulated for structural complexity (full nominal phrase and pronoun). One aspect of prosody (pitch contour) distinguished the groups, with lower scores in the SI group in both tasks. An aspect of accuracy (number of disfluencies) distinguished them in the discourse task (more disfluencies in the SI group). The comprehension of the target-sentences is isolation was harder for the SI group. Lexical/discourse factors can create differential demands for sentences in discourse, minimizing group effects. The effect of intervention was in the predicted direction for OWH plus N sentences (more demands for full nominal phase subjects). As for RC, this effect was not significant, possibly due to difficulties in the ascription of illocutionary force to YES/NO questions. In an overall analysis of reading fluency in discourse (different structures/punctuation marks), the pitch contour indicates syntactic impairment. Impairment at the syntax-prosody interface can account for these results.

In this thesis I do a comparative study between the Brazilian Portuguese Grammar (henceforth BP) and European Portuguese Grammar (henceforth EP) concerning to the word order, more specifically, with respect to the subject order in relation to the verb in finite declarative phrases. One defends that Verb-Subject order (VS) has been lost in the first language in (in)transitive contexts due a parametric change in the marking of Null Subject Parameter caused by weakness of AGR, what implies to consider that BP is losing the characterization of a prototypical null subject language as the EP for become a non-prodrop subject language as English. In this order, in inaccusative contexts, the visible morphologically agreement between a verbal inflection and the post-verbal DP is generally not observed and this DP can also be definite or indefinite, unlike what is proposed by Belletti s approach. Futhermore, researches have assumed that the subject position with respect to the verb comes from the fact of the syntax be conditioned by discoursive constraints. Therefore, with respect to these aspects, the goals of this research are: a) to develop an analysis about VS order in both grammars, considering the implications of the AGR richness for the position of the subjects ; b) to characterize this richness from the rediscussion of some theoretical proposals; c) to show counter-evidences to the proposal of that the word order codifies the informational structure; d) to explain the possibility of the non-visible morphologically agreement between the post-verbal DP and the verbal inflection in inaccusative contexts and its implications for the type of mechanism concerned to the nominative Case assignment and e) to argue against the universality of the Definiteness Efect on the post-verbal DP. To develope this study, the analysis bases itself on Pricinciples and Parameters, and the Distributed Morphology frameworks. The sentence structures are results from introspection data. During this research, it was possible to conclude that rich AGR that licences and identifies referential null subjects in null subject languages has no correlation with rich AGR that causes the verb movement in the syntax, and that preverbal subjects occupy an A-position in BP and in EP, unlike what is generally proposed. Futhermore, the different behaviour of VS order observed among these languages receives explanatory support from the type of syntactic output generated by each grammar in particular but not from discoursive constraints. With respect to the order in which there is no visibility of morphological agreement between the verbalinflection and the post-verbal DP in inaccusative contexts, I argue that it is a pseudo- VS order, in which Definiteness Efect does not act on. / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Nesta tese realizo um estudo comparativo entre a gramática do português brasileiro (doravante PB) e a do português europeu (doravante PE) no que diz respeito à ordem dos constituintes, mais especificamente, do sujeito em relação ao verbo em frases declarativas finitas. Tem-se defendido que a ordem Verbo-Sujeito (VS) tem sido perdida na primeira língua em contextos (in)transitivos em virtude de uma mudança paramétrica na marcação do Parâmetro do Sujeito Nulo ocasionada pelo enfraquecimento de AGR, o que implica considerar que o PB está deixando de ser uma língua de sujeito nulo prototípica como o PE para tornar-se uma língua de sujeito não-nulo como o inglês. Nessa ordem, em contextos inacusativos, a concordância morfologicamente visível entre a flexão verbal e o DP pós-verbal, em geral, não é observada e, ainda, esse DP pode ser definido ou indefinido, ao contrário do que é proposto pela análise belletiana. Ademais, estudos têm assumido que a posição do sujeito em relação ao verbo decorre do fato de a sintaxe estar a serviço de requerimentos de ordem discursiva. Portanto, face a esses aspectos, os objetivos desta pesquisa são: a) desenvolver uma análise sobre a ordem VS em ambas as gramáticas, levando em conta as implicações da riqueza de AGR para a posição dos sujeitos; b) caracterizar essa riqueza , a partir da rediscussão de algumas propostas teóricas; c) apresentar contra-evidências à proposta de que a ordem dos constituintes codifica a estrutura informacional; d) explicar a possibilidade de concordância morfologicamente não-vísível entre o DP pós-verbal e a flexão verbal em contextos inacusativos e suas implicações para o tipo de mecanismo relacionado à atribuição de Caso nominativo e e) argumentar contra a universalidade do Efeito de Definitude sobre o DP pós-verbal. Para a realização desse estudo, a análise fundamenta-se no arcabouço teórico do Modelo de Princípios e Parâmetros e da Morfologia Distribuída, sendo as estruturas frasais resultados de dados de introspecção. No decorrer desta pesquisa, é plausível concluir que AGR rico que licencia e idendifica sujeitos nulos referenciais em línguas de sujeito nulo não tem a ver com AGR rico que motiva o movimento do verbo na sintaxe, e que sujeitos pré-verbais ocupam uma posição-A no PB e no PE, ao contrário do que é comumente proposto na literatura. De mais a mais, o comportamento diferenciado da ordem VS atestado entre essas línguas ganha suporte explicativo no tipo de output sintático gerado por cada gramática em particular e não em requerimentos de ordem discursiva. No que concerne à ordem em que não há visibilidade da concordância morfológica entre a flexão-verbal e o DP pós-verbal em contextos inacusativos, argumento que seja uma pseudo-ordem VS , não sendo atuante sobre esse DP o Efeito de Definitude.

Summaries in Afrikaans and English / Ontluikende geletterdheid is waarskynlik die belangrik:ste tema wat in hierdie proefskrif behandel word. Die doelbewuste skepping van ouderdomsrelevante geleenthede vir die kleuter om op 'n informele wyse blootstelling en deelname aan luister- praat-, lees- en skryfvaardighede te ervaar, dien as vertrekpunt. Vroee ontwikkeling van geletterdheidsvaardigbede, as komponent van kognitiewe ontwikkeling by die kleuter, bet geblyk noodsaaklik te wees- nie net ten aansien van skoolgereedheid nie, maar ook ten opsigte van sy uiteindelike selfverwesenliking. Die oogmerk strek dus verder as die afiigting van die kleuter tot 'n toereikende vaardigbeidsvlak ten einde skoolgereed te kan wees. In teenstelling met die lank bestaande beleid in Suid-Afiika, wat bepaal bet dat voorskoolse onderwys byna uitsluitlik die onderrig van perseptuele vaardigbede moes bebels, word 'n alternatiewe benadering m hierdie proefskrif aangedui. Hulpverlening met geletterdheidsontwikkeling as doelwit behels waarskynlik 'n meer omvattende scenario ten einde die beboeftes van die affektief-, kognitief-, sosiaalwordende kind aan te spreek. Ontsluiting van 'n nuwe wereld van kennis waardeur nonne, boudings en vaardighede aangeleer word, bet 'n direkte invloed op die kwaliteit van sy totale lewe. By uitnemendbeid skep uitkomsgerigte onderwys die geleentbeid om reeds op voorskoolse vlak, geletterdbeidsvaardigbede op 'n informele wyse aan te leer. Die mite dat primere skole aileen die voorreg beskore is om leerders te leer lees en skryf, word krities in hierdie studie bevraagteken. Die klemverskuiwing is reeds sigbaar in die nuut-aangekondigde onderwysbeleid naamlik dat die leerprogram Geletterdheid minstens sowat 40% van die daaglikse onderrigprogram in die grondslagfase moet geniet. Die wyse waarop geletterdheidsvaardighede aangeleer kan word, word indiepte bespreek. Eweneens word belangrike prosesse waardeur die ontwikkeling van 'n fonologiese-, semantiese- en sintaktiese bewussyn, aan die orde gestel. Die aanleer van geletterdheidsvaardighede blyk nie alleen noodsaaklik te wees vir die ontwikkeling van lewensvaardighede nie dit bewerkstellig toegang tot en die begryping en belewing van die werklikheid wat die mens omring. / Emergent literacy is probably the most important theme addressed in this thesis. The purpose is to create age related opportunities for the pre-school child to experience exposure and participation in listening-, vocal-, reading- and writing abilities on an informal basis which serves as a point of departure. Early awareness of literacy abilities as a component of cognitive development of the pre-school child has proved to be essential - not only in respect of school readiness, but also with regard to ultimate selfactualisation. The aim therefore goes beyond the mere training of the pre-school child to an adequate level of competence to be well prepared for formal schooling. Contrary to the well established policy in South Africa which determined that pre-school education should be limited to the exclusive teaching of perceptual abilities, an alternative approach has been considered in this thesis. Assistance with a view to literacy development probably encompasses a more inclusive scenario in order to address the requirements of the emotional, cognitive and social developing pre-school child. Exposure to a new world of knowledge through which norms, attitudes and abilities are acquired, has a direct influence on the quality of his total life. Outcomes based education provides an excellent opportunity to acquire already at pre-school level, literacy skills on an informal basis. The myth that only primary schools have been allotted the privilege to teach learners to read and write, is being critically questioned in this study. The change in emphasis is already evident in the recently announced education policy, namely that the daily learning programme Literacy must include at least 400/o of the foundation programme phase. The manner in which literacy skills can be acquired are discussed in depth. Simultaneously Literacy must include at least 400/o of the foundation programme phase. The manner in which literacy skills can be acquired are discussed in depth. Simultaneously important processes affecting the development of a phonological, semantic and syntactic cognition are discussed. The acquisition of literacy skills appears not only to be important to the development of life skills - it accomplishes access to and the comprehension of as well as the perception of the reality surrounding man. / Psychology of Education / D. Ed. (Sielkundige Opvoedkunde)

A lack of research in adult literacy for both native speakers and speakers of English as a second language led to the development of the Test of Affixes in Syntactic Structures (TASS) for use in a pilot study (Dogger, January 1978) in which knowledge of derivational morphology was tested to determine its possible relationship with reading for English as a second language students. Test construction was followed by a thorough verification procedure which is the purpose of this study. In September 1978 the following measures of test strength were established: construct validity, content validity, item difficulty, item discrimination, internal consistency, rational equivalence, and concurrent validity. The degree of relationship between reading proficiency, as demonstrated by subject performance on the Iowa Silent Reading Test, Level III (ISRT,III), and knowledge of derivational affixes, as demonstrated by subject performance on TASS, was also established. Results show that successful performance on the ISRT, III includes reading strategies beyond those required for successful performance on TASS. In other words, mastery of language structures as represented by English orthography is necessary but not sufficient for college-level reading proficiency.

Submitted by Maike Costa (maiksebas@gmail.com) on 2016-01-12T11:38:01Z No. of bitstreams: 1 arquivototal.pdf: 2950569 bytes, checksum: fdcb47c13d14f4ee0af9b2223a6ea351 (MD5) / Made available in DSpace on 2016-01-12T11:38:01Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 2950569 bytes, checksum: fdcb47c13d14f4ee0af9b2223a6ea351 (MD5) Previous issue date: 2015-03-31 / This thesis analyzes the function of two grammatical forms which bring about the articulation of sentences through adverbial hypotaxis, a mechanism of textual connection characterized by the establishment of semantic-logical relationships, signaling the communicative intentions of the users of the language. These grammatical forms – sem(without) and the conjunctional periphrase semque (unless) - are placed in distinct grammatical classes: preposition and conjunction respectively, due to grammatical conventions. These conventions attribute to the first term, the function of transposing a nominal syntagm into an adnominal or adverbial adjunct within the sentence; and to the second, that of transposing an absolute sentence to a new structure, within which it occupies the position of a constituent, taking up, among other functions, that of the adverbial adjunct, under the form of a sentence. In this perspective, I address the syntactical and semantic re-categorization of the above mentioned preposition, defending the view that it figures as a conjunction not only when it is part of the conjunctional periphrase, but also in in the presence of the verb in the infinitive form, forming reduced infinitive sentences. Using Functional Theory, which analyzes the principles which govern the natural use of language, I discuss the cognitive and interactional motivations which cause the change.With specific reference to the treatment of the process of grammaticalization, I have used the North American branch of Functionalism to explain the fluctuation of category and semantic and discursive fluctuations experienced by this linguistic item. From the observation of data, I have deduced patterns of typical uses of reduced as well as highly evolved sentence structures, emphasizing that these two structures are not always interchangeable. The corpus from which I inventorythe syntactic, semantic and pragmatic properties of the transpositers under investigation, correlating textual and interpersonal functions, is constituted by texts from the argumentative sphere- articles of opinion, editorials and interviews from weekly magazines. I conclude that though the preposition sem does not introduce argumentative terms, it belongs to the listof prepositions which are moderately grammaticalized, and is susceptible to change, exhibiting a diversity of semantic nuances similar to conjunctions. / Esta tese analisa a função de duas formas gramaticais que promovem a articulação oracional via hipotaxe adverbial, mecanismo de conexão textual caracterizado por estabelecer relações lógico-semânticas, sinalizando as intenções comunicativas dos usuários da língua. Essas formas gramaticais – sem e a perífrase conjuntiva sem que – estão acomodadas em classes gramaticais distintas: preposição e conjunção respectivamente, devido a convenções da gramática, que atribui à primeira a função de transpor um sintagma nominal a adjunto adnominal ou adverbial, no nível da oração; e à segunda, de transpor uma oração absoluta para uma nova estrutura no interior da qual ocupa o lugar de constituinte, assumindo, dentre outras funções, a de adjunto adverbial, sob a forma de oração. Nessa perspectiva, abordo a recategorização sintática e semântica da preposição citada, defendendo que esta se configura como conjunção não só quando integra a perífrase conjuntiva, mas diante de verbo na forma infinitiva, formando orações reduzidas de infinitivo. Tomando como suporte a Teoria Funcionalista, que analisa os princípios que governam o uso natural da língua, discuto as motivações cognitivas e interacionais que pressionam a mudança; particularmente quando do tratamento do processo de gramaticalização, recorro à vertente do Funcionalismo Norte-americano, para explicar a flutuação categorial e semântico-discursiva experimentada por esse item linguístico. Da observação dos dados, depreendo padrões de uso típicos da estrutura oracional reduzida e da desenvolvida, ressalvando-se que nem sempre as duas estruturas são intercambiáveis. O corpus a partir do qual inventario as propriedades sintáticas e semântico-pragmáticas dos transpositores sob investigação, correlacionando às funções textual e interpessoal, constitui-se de textos da esfera argumentativa – artigos de opinião, editoriais e entrevistas de periódicos semanais. Concluo que, embora a preposição sem não introduza termos argumentais, daí integrar o rol das preposições medianamente gramaticalizadas, é suscetível à mudança, exibindo uma diversidade de matizes semânticos à semelhança das conjunções.

La présente thèse s’inscrit dans le projet FNS Sinergia 136230 “SMS communication in Switzerland: Facets of Linguistic Variation in a Multilingual Country”. Ella a pour but d’étudier la variabilité formelle des questions dans le Corpus suisse de SMS en français, en montrant comment l’analyse des données en provenance des interactions informelles autres que l’oral spontané – telles que messagerie instantanée, SMS, WhatsApp, etc. – peut nous permettre d’élargir les horizons et de porter un nouveau regard sur les phénomènes de la variation syntaxique. En effet, le français est réputé pour la variété importante de structures interrogatives : trois variantes pour la question totale (Tu vas au cours ?, Vas-tu au cours ?, Est-ce que tu vas au cours ?), et au moins six ou huit pour la question partielle (Tu vas où ?, Où tu vas ?, Où vas-tu ?, Où va ton frère ?, Où est-ce que tu vas ?, Qui va au cours ?, Où c’est que tu vas ?, C’est où que tu vas ?). Cette thèse vise, en conséquence, à comprendre comment les scripteurs de messages sont amenés à choisir entre les différentes variantes à disposition. L’hypothèse que nous faisons dans cette thèse est fonctionnaliste : elle postule que sous la pression de différents types de contraintes, aussi bien linguistiques que non linguistiques, le scripteur choisit telle variante qui lui permet de parvenir au mieux à ses fins communicatives. Afin d’identifier les différents types de contraintes ou de facteurs susceptibles d’avoir une incidence sur le choix des variantes, nous appliquons un modèle d’analyse multidimensionnel qui s’intéresse simultanément aux paramètres grammaticaux, interactionnels et sociolinguistiques. / This thesis is part of the project FNS Sinergia 136230 "SMS communication in Switzerland: Facets of Linguistic Variation in a Multilingual Country". The aim of the thesis is to provide analysis of the variety of French interrogative structures in the Swiss SMS Corpus. It also attempts to show how the data originating from spontaneous electronic interaction (instant messaging, texting, WhatsApp, etc.) allows us to broaden the scope and to take a new look at the phenomena of syntactic variation. The French language is known for the wide variety of interrogative structures: 3 variants for yes/no questions (Tu vas au cours ?, Vas-tu au cours ?, Est-ce que tu vas au cours ?), and at least six or eight variants for wh- questions (Tu vas où ?, Où tu vas ?, Où vas-tu ?, Où va ton frère ?, Où est-ce que tu vas ?, Qui va au cours ?, Où c’est que tu vas ?, C’est où que tu vas ?). Therefore, our goal is to understand how SMS writers make the choice between available French interrogative structures. The hypothesis of this study is functionalist: it postulates that under the pressure of various linguistic and non-linguistic constraints, the SMS writer chooses that variant which allows him to best achieve given communicative goals. In order to identify different types of constraints or factors that may influence the choice of variants, a multidimensional analysis model is applied which focuses simultaneously on grammatical, interactional and sociolinguistic parameters.

Les recherches en Traitement Automatique des Langues (TAL) ont identifié différents niveaux d'analyse lexicale, syntaxique et sémantique. Il en découle un découpage hiérarchique des différentes tâches à réaliser afin d'analyser un énoncé. Les systèmes classiques du TAL reposent sur des analyseurs indépendants disposés en cascade au sein de chaînes de traitement (pipelines). Cette approche présente un certain nombre de limitations : la dépendance des modèles à la sélection empirique des traits, le cumul des erreurs dans le pipeline et la sensibilité au changement de domaine. Ces limitations peuvent conduire à des pertes de performances particulièrement importantes lorsqu'il existe un décalage entre les conditions d'apprentissage des modèles et celles d'utilisation. Un tel décalage existe lors de l'analyse de transcriptions automatiques de parole spontanée comme par exemple les conversations téléphoniques enregistrées dans des centres d'appels. En effet l'analyse d'une langue non-canonique pour laquelle il existe peu de données d'apprentissage, la présence de disfluences et de constructions syntaxiques spécifiques à l'oral ainsi que la présence d'erreurs de reconnaissance dans les transcriptions automatiques mènent à une détérioration importante des performances des systèmes d'analyse. C'est dans ce cadre que se déroule cette thèse, en visant à mettre au point des systèmes d'analyse à la fois robustes et flexibles permettant de dépasser les limitations des systèmes actuels à l'aide de modèles issus de l'apprentissage par réseaux de neurones profonds. / NLP researchers has identified different levels of linguistic analysis. This lead to a hierarchical division of the various tasks performed in order to analyze a text statement. The traditional approach considers task-specific models which are subsequently arranged in cascade within processing chains (pipelines). This approach has a number of limitations: the empirical selection of models features, the errors accumulation in the pipeline and the lack of robusteness to domain changes. These limitations lead to particularly high performance losses in the case of non-canonical language with limited data available such as transcriptions of conversations over phone. Disfluencies and speech-specific syntactic schemes, as well as transcription errors in automatic speech recognition systems, lead to a significant drop of performances. It is therefore necessary to develop robust and flexible systems. We intend to perform a syntactic and semantic analysis using a deep neural network multitask model while taking into account the variations of domain and/or language registers within the data.