Despite the anecdotal evidence of social difficulties in children with Fetal Alcohol Spectrum Disorders (FASD), and the risk for secondary disabilities as a result of these social difficulties, very little research has examined social-emotional functioning in children with FASD. The majority of the research conducted thus far has relied on parent and teacher reports to document social impairments. These parent and teacher reports provide a broad measure of social functioning but are unable to elucidate the specific aspects of social functioning that this group of children might find difficult. As a result, it has been very difficult to develop effective social interventions for children with FASD because it is unclear what aspects of social functioning should be targeted. The current study aimed to examine emotion recognition abilities in children with FASD, as recognition of emotions is an important precursor for appropriate social interaction. The study included 22 participants with diagnosed FASD (ages 8-14), with age- and gender- matched typically developing controls. Participants were assessed using computerized measures of emotion recognition from three nonlinguistic modalities: facial expressions (static and dynamic, child and adult faces), emotional tone of voice (child and adult voices), and body positioning and movement (postures and point-light walkers). In addition, participants completed a task assessing emotion recognition in real-life scenarios. Finally, caregivers completed measures of behavioural functioning, adaptive functioning, FASD symptomatology, and a demographics questionnaire. Overall, findings suggest that children with FASD do have more difficulties than age-matched typically developing peers in aspects of emotion recognition, with particular difficulties in recognizing emotions from adult facial expressions and adult emotional prosody. In addition, children with FASD had more difficulty perceiving differences in facial expressions. When the effect of age was examined, it was found that some aspects of emotion recognition were more impaired in children with FASD between age eight and ten years compared to same-age typically developing peers and compared to children with FASD age 11-14. This finding suggests that younger children with FASD may demonstrate a delay in the acquisition of some aspects of emotion recognition or may be more vulnerable to the information processing demands of some tasks compared to older children with FASD. The types of emotion recognition difficulties found in the current study supported a pattern where children with FASD make more errors on emotion recognition tasks when the complexity of the task is increased and consequently demands greater information processing. As such, it is anticipated that children with FASD would be likely to have the most difficulty with emotion recognition abilities embedded within complex, rapidly changing, real-world social situations, and in recognizing more subtle emotional displays. Caregivers, teachers, and professionals living and working with children and youth with FASD should be aware of possible emotion recognition difficulties in complex social situations and should help foster stronger emotion recognition skills when difficulties are detected.

Submitted by THAIS HOLANDA DE ABREU-ZORZI null (thaishabreu@bol.com.br) on 2016-05-15T13:54:19Z No. of bitstreams: 1 Versão digital tese thais.pdf: 4840618 bytes, checksum: 83b58fcc566c791e3bb53c4516cb488d (MD5) / Approved for entry into archive by Juliano Benedito Ferreira (julianoferreira@reitoria.unesp.br) on 2016-05-17T14:10:07Z (GMT) No. of bitstreams: 1 abreu-zorzi_th_dr_arafc.pdf: 4840618 bytes, checksum: 83b58fcc566c791e3bb53c4516cb488d (MD5) / Made available in DSpace on 2016-05-17T14:10:07Z (GMT). No. of bitstreams: 1 abreu-zorzi_th_dr_arafc.pdf: 4840618 bytes, checksum: 83b58fcc566c791e3bb53c4516cb488d (MD5) Previous issue date: 2016-04-04 / Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) / Esta tese tem como objetivo realizar um estudo comparativo dos advérbios em –mente em duas sincronias da língua portuguesa - Português Arcaico (PA), século XIII, e Português Brasileiro (PB) atual, a fim de observar e descrever possíveis mudanças com relação ao estatuto prosódico dessas formas. Para a descrição do estatuto prosódico desses advérbios, sobretudo no PA, elegeram-se como corpus as 420 cantigas em louvor à Virgem Maria, conhecidas como Cantigas de Santa Maria (CSM), e as 1251 cantigas profanas (510 de amigo, 431 de escárnio e maldizer e 310 de amor). Por outro lado, elegeu-se como corpus de estudo do PB um recorte do banco de dados do “Corpus Online do Português”, elaborado em conjunto pelos pesquisadores Michael Ferreira, da Universidade de Georgetown, e Mark Davies, da Brigham Young University. A partir da coleta das formas adverbiais em -mente nos corpora, são investigadas algumas propriedades dos advérbios em -mente sob o viés da Fonologia Prosódica (SELKIRK, 1984; NESPOR; VOGEL, 1986) e da Fonologia Métrica (HAYES, 1995). Conclui-se que as formas adverbiais em –mente são, do ponto de vista prosódico, compostas (um acento lexical e um secundário), tanto em PA como em PB, visto que tais advérbios podem ser considerados elementos que são formados por partes independentes entre si, em que a Regra de Atribuição do Acento atua em domínios distintos: nas bases já flexionadas e no “sufixo” –mente. Por terem um domínio independente, cada uma das partes formadoras desses advérbios representa uma palavra fonológica (ω) distinta. / This thesis aims to conduct a comparative study of adverbs ending in -mente in two synchronies of the Portuguese language - Archaic Portuguese (AP), from the thirteenth century, and current Brazilian Portuguese (BP) - in order to observe and describe possible changes regarding the prosodic status of these forms. For the description of the prosodic status of these adverbs, especially in AP, a corpus composed of 420 cantigas [a genre of poetry] in praise of the Virgin Mary, known as Cantigas de Santa Maria (CSM), and 1251 profane cantigas (510 cantigas de amigo, 431 cantigas de escárnio e maldizer and 310 cantigas de amor). On the other hand, a part of the database from “Corpus Online do Português" [Online Corpus of the Portuguese Language], prepared jointly by researchers Michael Ferreira, from Georgetown University, and Mark Davies, from Brigham Young University, was elected as study corpus of the BP. From the collection of the adverbial forms ending in -mente in the corpora, some properties of adverbs ending in -mente are investigated under the bias of Prosodic Phonology (SELKIRK, 1984; NESPOR;VOGEL, 1986) and Metrical Phonology (HAYES, 1995) are investigated. It has been concluded that from the prosodic perspective the adverbial forms ending in -mente are compounds (one lexical stress and one secondary stress) in both AP and BP, since such adverbs may be considered elements formed of independent parts among themselves, where the Stress Assignment Rule operates in different domains: in bases already inflected and in the "suffix" ending in -mente. For having an independent domain, each forming part of these adverbs represents a distinct phonological word (ω). / FAPESP: 2011/18933-8

Ce travail a trois objectifs principaux. Tout d’abord, contribuer à l’étude du système rythmique des langues et des variétés dialectales à travers l’analyse du rythme de l’anglais parlé au pays de Galles et de la langue galloise en utilisant comme point de comparaison l’anglais dit standard. Puis, démontrer qu’il existe un allongement final bien plus marqué en gallois et en anglais gallois (bilingues & monolingues) qu’en anglais standard. Et enfin, prouver que les résultats obtenus lors de cette étude sont étroitement liés à l’influence majoritaire de la langue galloise sur celle des autres variétés d’anglais présentes dans ce pays. Dans cette optique, nous avons, dans un premier temps, élargi la base de données Eurom I et nous avons ensuite mené différents tests de perception et études statistiques. Il en ressort que la langue galloise est effectivement le substrat principal de l’anglais parlé au pays de Galles, que ce soit chez les gallois bilingues ou monolingues (personnes résidant au pays de Galles ne parlant qu’anglais). En effet, la longueur de l’allongement en syllabe finale inaccentuée va en décroissant du gallois à l’anglais standard. Cependant, étant donné l’impossibilité de catégoriser de façon précise le rythme du gallois et de l’anglais gallois en tenant compte de tous les paramètres métriques, il est plus difficile d’affirmer de façon catégorique que la langue galloise joue un rôle prépondérant sur ces derniers. / This thesis has three main objectives. First of all, to contribute to the study of the rhythmical system of languages and dialectal varieties through the analysis of rhythms in Welsh English and Welsh, using Standard English as a focal comparative element; then, to demonstrate that the final lengthening is much more notable in Welsh and in Welsh English (bilinguals and monolinguals) than in standard English; finally, to prove that this study’s results are tightly intertwined with the influence of the Welsh language, which is greater than that other English accents present in this country. For that purpose, we have first widened the Eurom 1 database and then carried out various perception tests and statistical analysis. The results show that the Welsh language is the substratum of the English spoken in Wales, as much for bilinguals as for monolinguals (residents of the country speaking only English). Indeed, the length of the final unstressed syllable decreases from Welsh to Standard English. However, as it is impossible to classify with precision the rhythm of Welsh and Welsh English taking into account all the metrics, it is more difficult to establish with certainty whether the Welsh language has a major influence on the latter.

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).

L’objectif de cette étude est d’apporter une contribution aux recherches portant sur les effets prosodiques du contact de langues à travers l’étude du système prosodique du français centrafricain, une variété de français ayant émergé suite au contact avec une langue africaine à densité tonale maximale, le sango. La République centrafricaine est un pays extrêmement multilingue. Entre 60 et 100 langues y sont parlées, outre les deux langues officielles que constituent le sango – lingua franca, et le français – langue principale de l’enseignement et de l’administration publique. Dans la capitale, Bangui, le sango est la langue parlée d’ordinaire dans la vie quotidienne, tandis que le français s’utilise essentiellement dans des contextes professionnels. Cette étude se base sur des enregistrements de parole spontanée de 12 locuteurs francophones de Bangui. Des analyses acoustiques montrent que l’intonation du français centrafricain partage des caractéristiques communes avec le sango. La majorité de mots ont des patrons tonals qui restent inchangés quel que soit leur place dans l’énoncé, et chaque syllabe porte un ton. Le système se distingue ainsi considérablement du système intonatif du français européen où la courbe mélodique est contrainte au niveau post-lexical et dépend entre autres de facteurs rythmiques, syntaxiques et pragmatiques. La conclusion principale de cette étude est que le français centrafricain se classifie d’un point de vue typologique comme une langue à tons lexicaux. Il ressort que le système prosodique du français centrafricain est plus proche de celui du sango et que de celui du français européen. Les faits mis au jour dans cette thèse montrent que la prosodie peut changer de façon fondamentale dans une situation de contact de langue. / This study is concerned with prosody and language contact. The fact that language contact induces change is well documented, but few studies focus on the prosodic effects of contact-induced change. The aim of this study is to provide a case study of the prosodic system of the contact variety Central African French, which has emerged from the contact between French and the African tone language, Sango.The Central African Republic is a multilingual country with between 60 to 100 different regional languages spoken within its borders in addition to two official languages, the lingua franca Sango and French. French has been the main language of education and of public administration since colonial times. In the capital Bangui, Sango is the most used language in everyday communication whereas French is spoken in professional contexts. This study is based on recordings of spontaneous speech of 12 French-speaking informants from Bangui. Acoustic analyses of the recordings show that the prosody of Central African French shares with Sango some fundamental characteristics: most words have fixed tonal patterns independently of their position in the sentence and every syllable carries a static tone. This system greatly differs from the system of European varieties of French, where the sentence melody is determined at the post-lexical level and depends on factors such as rhythm, syntax and pragmatics. The main conclusion of this study is that Central African French may be classified as a tone language and thus is endowed with a prosodic system that is closer to Sango than to European French. This finding suggests that intonation might change radically in contact situations ; the change is not only superficial but concerns the underlying system.

Linha de pesquisa: Curr?culos, avalia??o, pr?ticas pedag?gicas e forma??o de professores. / Submitted by Rodrigo Martins Cruz (rodrigo.cruz@ufvjm.edu.br) on 2018-05-03T14:25:28Z No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) maressa_carneiro_melo.pdf: 3083173 bytes, checksum: 1d3d359317036112df43c3c8e4876ea5 (MD5) / Approved for entry into archive by Rodrigo Martins Cruz (rodrigo.cruz@ufvjm.edu.br) on 2018-05-03T14:25:38Z (GMT) No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) maressa_carneiro_melo.pdf: 3083173 bytes, checksum: 1d3d359317036112df43c3c8e4876ea5 (MD5) / Made available in DSpace on 2018-05-03T14:25:38Z (GMT). No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) maressa_carneiro_melo.pdf: 3083173 bytes, checksum: 1d3d359317036112df43c3c8e4876ea5 (MD5) Previous issue date: 2017 / Universidade Federal dos Vales do Jequitinhonha e Mucuri (UFVJM) / Esta pesquisa tem como objetivo analisar a abordagem do ensino de pros?dia e, principalmente, dos atos diretivos de ordem e de pedido em Portugu?s Brasileiro nas atividades de quatro materiais did?ticos de Portugu?s Brasileiro como L?ngua Estrangeira utilizados em cursos de Portugu?s como L?ngua Estrangeira no N?cleo de L?nguas da Universidade Federal dos Vales do Jequitinhonha e Mucuri at? o ano de 2016. O ensino da pros?dia e dos atos de ordem e de pedido se fazem relevantes para falantes estrangeiros visto que s?o recursos que integram o sistema lingu?stico (SEARLE, 1981) e neles est?o impressos tra?os culturais do Portugu?s Brasileiro (BODOLAY, 2009). Para que o alocut?rio diferencie um ato diretivo de ordem do de pedido ? preciso que ele esteja atento ?s pistas pros?dicas utilizadas pelo locutor, aos pap?is sociais que ocupam no contexto de comunica??o e outros recursos extralingu?sticos que interferem na interpreta??o do enunciado. Assim sendo, tais recursos s?o totalmente exclusivos de uma l?ngua, segundo Marcuschi (2001) e Cruz-Ferreira (2003), tornando essenciais para a comunica??o de um falante estrangeiro. Para atender o objetivo central deste estudo, utilizamos como instrumento de coleta e an?lise a t?cnica de an?lise de conte?do de Bardin (2011). Como resultado principal, verificamos que tanto a pros?dia quanto os atos de ordem e de pedido s?o pouco explorados ou at? ignorados. Quando h? abordagem desses aspectos da l?ngua falada, os estilos de atividades dos materiais n?o utilizam os recursos orais, fazendo o uso da escrita para ensinar oralidade. / Disserta??o (Mestrado Profissional) ? Programa de P?s-Gradua??o em Educa??o, Universidade Federal dos Vales do Jequitinhonha e Mucuri, 2017. / The aim of this research is to analyse the approach of teaching prosody and, mainly, of order and request acts in Brazilian Portuguese in the activities of four didactic materials of Brazilian Portuguese as a foreign language used in courses of Portuguese as a foreign Language at N?cleo de L?nguas of Federal University of Jequitinhonha and Mucuri Valleys up to the year 2016. The teaching of prosody and the acts of order and request becomes relevant for foreign speakers since they are resources that integrate the linguistic system (SEARLE, 1981) and in them are printed cultural traits of Brazilian Portuguese (BODOLAY, 2009). In order for the speaker to differentiate a directive act from an order to a request, he/she must be aware of the prosodic clues used by the speaker, the social roles they occupy in the context of communication and other extralinguistic resources that interfere in the interpretation of the utterance. Thus, these resources are totally unique to a language, according to Marcuschi (2001) and Cruz-Ferreira (2003), making it essential for the communication of a foreign speaker. In order to achieve the central objective of this study, we used, as a collection of data and analysis instrument, the content an analysis technique of Bardin (2011). As a main result we find that both prosody and order and request acts are little explored or even ignored. When there is an approach with deals with these aspects of spoken language, the styles of the materials activities do not use oral resources, making use of writing to teach orality.

Syftet med studien är att undersöka erfarna lärare i svenska som andraspråks uppfattningar i arbetet med momentet uttal. Detta syfte avser jag uppfylla genom att besvara följande frågor: 1.      Vilka faktorer påverkar enligt lärarna elevers uttal i svenskan? 2.      Vilka arbetssätt väljer lärarna att använda? 3.      Vilka arbetsmetoder använder lärarna i sin undervisning? Det teoretiska ramverket i denna studie bygger dels på behaviorismen och dess kontrastiva hypotes dels på den sociokulturella teorin. I denna kvalitativa studie har fem behöriga lärare i svenska som andraspråk, som arbetar i årskurserna 4-6 och har lång erfarenhet intervjuats. Intervjuerna har kategoriserats med en kvalitativ innehållsanalys för att identifiera och ta fram teman för innehållet. Sammanfattningsvis kan konstateras att hos samtliga respondenter genomsyrar undervisningen i svenskt uttal de flesta momenten i ämnet svenska som andraspråk. Elevernas ankomstålder, modersmål och studieteknik är de faktorer som mest påverkar deras förmåga att tillägna sig ett korrekt svenskt uttal. För att nå detta mål söker lärarna att hitta elevernas proximala utvecklingszon, för att med den som utgångspunkt bestämma vilka arbetssätt (innehållet i undervisningen) som ska prioriteras. De vanligaste arbetsmetoderna är högläsning och imitation, där man tränar uttal av olika kontraster av fonem, kvantitet, betoning och prosodi. / The purpose of the study is to investigate experienced teachers´ perceptions about teaching pronunciation in swedish as a second language. This purpose I intend to fulfill by answering the following questions:   1. What factors do teachers consider important when working to develop pronunciation? 2. In which way do teachers choose to work? 3. What teaching methods do teachers use and how are they applied?   The theoretical framework in this study is based on behaviorism and its contrastive hypothesis and partly on socio-cultural theory. In this qualitative study, five qualified and experienced teachers in Swedish as second language, who work in grades 4-6 have been interviewed. The interviews have been categorized with a qualitative content analysis as a method to identify and develop themes for the content.   In conclusion, it can be noted that all interviewees consider that teaching of swedish pronunciation permeates most of the topics in the subject Swedish as other language. The students' age of arrival to Sweden, mother tongue and study technique are the factors that most affect their ability to acquire a correct Swedish pronunciation. To achieve this goal, the teachers seek to find the students' zone of proximal development as starting point in determining which way to work (the content of teaching) to be prioritized. The most common working methods are aloud reading and imitation, where pupils practice the pronunciation of different contrasts of phoneme, quantity, accent, and prosody.

[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.

The need for interacting with machines using spoken natural language is growing, along with the expectation that synthetic speech in this context sound natural. Such interaction includes answering questions, where prosody plays an important role in producing natural English synthetic speech by communicating the information structure of utterances. CCG is a theoretical framework that exploits the notion that, in English, information structure, prosodic structure and syntactic structure are isomorphic. This provides a way to convert a semantic representation of an utterance into a prosodically natural spoken utterance. GF is a framework for writing grammars, where abstract tree structures capture the semantic structure and concrete grammars render these structures in linearised strings. This research combines these frameworks to develop a system that converts semantic representations of utterances into linearised strings of natural language that are marked up to inform the prosody-generating component of a speech synthesis system. / Computing / M. Sc. (Computing)

De plus en plus, l'interaction entre personne et machine se rapproche du naturel afin de ressembler à l'interaction entre humains, incluant l'expressivité (en particulier les émotions et les attitudes). Dans la communication parlée, les attitudes, et plus généralement les affects sociaux, sont véhiculés principalement par la prosodie. Pour les langues tonales, la prosodie est utilisée aussi pour coder l'information sémantique dans les variations de tons. Ce travail de thèse présente une étude des affects sociaux du vietnamien, une langue à tons et une langue peu dotée, afin d'appliquer les résultats obtenus à un système de synthèse de haute qualité capable de produire la parole « expressive » pour le vietnamien. Le premier travail de cette thèse consiste en la construction du premier corpus audio-visuel des attitudes vietnamiennes, qui contient seize attitudes. Ce corpus est ensuite utilisé pour étudier la perception audio-visuelle et interculturelle des attitudes vietnamiennes. Pour cela, une série de tests perceptifs a été effectuée avec des auditeurs natifs et non-natifs (des auditeurs francophones pour les non-natifs). Les résultats de ces tests montrent que les facteurs influant sur la perception des attitudes sont l'expression de l'attitude elle-même et la modalité de présentation (audio, visuelle et audio-visuelle). Ces résultats nous ont ainsi permis de trouver des affects sociaux communs ou interculturels entre le vietnamien et le français. Puis, un autre test de perception a été réalisé sur des phrases avec tons afin d'explorer l'effet du système tonal du vietnamien sur la perception des attitudes. Les résultats montrent que les juges non-natifs peuvent traiter et séparer les indices tonals locaux et les traits saillants prosodiques de portée globale. Après une présentation de nos études sur les affects sociaux en vietnamien, nous décrivons notre modélisation de la prosodie des attitudes en vue de la synthèse de la parole expressive en vietnamien. En nous basant sur le modèle de superposition des contours fonctionnels, nous proposons une méthode pour modéliser et générer de la prosodie expressive en vietnamien. Cette méthode est ensuite appliquée pour générer de la parole expressive en vietnamien, puis évaluée par des tests de perception sur les énoncés synthétiques. Les résultats de perception valident bien la performance de notre modèle et confirment que l'approche de superposition de contours fonctionnels peut être utilisée pour modéliser une prosodie complexe comme dans le cas de la parole expressive d'une langue à tons. / Today, the human-computer interaction is reaching the naturalness and is increasingly similar to the human-human interaction, including the expressiveness (especially emotions and attitudes). In spoken communication, attitudes or social affects are mainly transferred through prosody. For tonal languages, prosody is also used to encode semantic information via tones. This thesis presents a study of social affects in Vietnamese, a tonal and under-resourced language, in order to apply the results to Vietnamese expressive speech synthesis task. The first task of this thesis concerns the construction of a first audio-visual corpus of Vietnamese attitudes which contains sixteen attitudes. This corpus is then used to study the audio-visual and intercultural perceptions of the Vietnamese attitudes. A series of perceptual tests was carried out with native and non-native listeners (French for non-native listeners). Experimental results reveal the fact that the influential factors on the perception of attitudes include the modality of presentation (audio, visual and audio-visual) and the attitudinal expression itself. These results also allow us to investigate the common specificities and cross-cultural specificities between Vietnamese and French attitudes. Another perception test was carried out using sentences with tonal variation to study the influence of Vietnamese tones on the perception of attitudes. The results show that non-native listeners can process the local prosodic cues of tones, together with the global cues of attitude patterns. After presenting our studies on Vietnamese social affects, we describe our work on attitude modelling to apply it to Vietnamese expressive speech synthesis. Based on the concept of prosodic contour superposition, a prosodic model was proposed to encode the attitudinal function of prosody for Vietnamese attitudes. This model was applied to generate the Vietnamese expressive speech and then evaluated in a perceptual experiment with synthetic utterances. The results validate the ability of applying our proposed model in generating the prosody of attitudes for a tonal language such as Vietnamese.