Les corpus bilingues sont des ressources essentielles pour s'affranchir de la barrière de la langue en traitement automatique des langues (TAL) dans un contexte multilingue. La plupart des travaux actuels utilisent des corpus parallèles qui sont surtout disponibles pour des langues majeurs et pour des domaines spécifiques. Les corpus comparables, qui rassemblent des textes comportant des informations corrélées, sont cependant moins coûteux à obtenir en grande quantité. Plusieurs travaux antérieurs ont montré que l'utilisation des corpus comparables est bénéfique à différentes taches en TAL. En parallèle à ces travaux, nous proposons dans cette thèse d'améliorer la qualité des corpus comparables dans le but d'améliorer les performances des applications qui les exploitent. L'idée est avantageuse puisqu'elle peut être utilisée avec n'importe quelle méthode existante reposant sur des corpus comparables. Nous discuterons en premier la notion de comparabilité inspirée des expériences d'utilisation des corpus bilingues. Cette notion motive plusieurs implémentations de la mesure de comparabilité dans un cadre probabiliste, ainsi qu'une méthodologie pour évaluer la capacité des mesures de comparabilité à capturer un haut niveau de comparabilité. Les mesures de comparabilité sont aussi examinées en termes de robustesse aux changements des entrées du dictionnaire. Les expériences montrent qu'une mesure symétrique s'appuyant sur l'entrelacement du vocabulaire peut être corrélée avec un haut niveau de comparabilité et est robuste aux changements des entrées du dictionnaire. En s'appuyant sur cette mesure de comparabilité, deux méthodes nommées: greedy approach et clustering approach, sont alors développées afin d'améliorer la qualité d'un corpus comparable donnée. L'idée générale de ces deux méthodes est de choisir une sous partie du corpus original qui soit de haute qualité, et d'enrichir la sous-partie de qualité moindre avec des ressources externes. Les expériences montrent que l'on peut améliorer avec ces deux méthodes la qualité en termes de score de comparabilité d'un corpus comparable donnée, avec la méthode clustering approach qui est plus efficace que la method greedy approach. Le corpus comparable ainsi obtenu, permet d'augmenter la qualité des lexiques bilingues en utilisant l'algorithme d'extraction standard. Enfin, nous nous penchons sur la tâche d'extraction d'information interlingue (Cross-Language Information Retrieval, CLIR) et l'application des corpus comparables à cette tâche. Nous développons de nouveaux modèles CLIR en étendant les récents modèles proposés en recherche d'information monolingue. Le modèle CLIR montre de meilleurs performances globales. Les lexiques bilingues extraits à partir des corpus comparables sont alors combinés avec le dictionnaire bilingue existant, est utilisé dans les expériences CLIR, ce qui induit une amélioration significative des systèmes CLIR. / Bilingual corpora are an essential resource used to cross the language barrier in multilingual Natural Language Processing (NLP) tasks. Most of the current work makes use of parallel corpora that are mainly available for major languages and constrained areas. Comparable corpora, text collections comprised of documents covering overlapping information, are however less expensive to obtain in high volume. Previous work has shown that using comparable corpora is beneficent for several NLP tasks. Apart from those studies, we will try in this thesis to improve the quality of comparable corpora so as to improve the performance of applications exploiting them. The idea is advantageous since it can work with any existing method making use of comparable corpora. We first discuss in the thesis the notion of comparability inspired from the usage experience of bilingual corpora. The notion motivates several implementations of the comparability measure under the probabilistic framework, as well as a methodology to evaluate the ability of comparability measures to capture gold-standard comparability levels. The comparability measures are also examined in terms of robustness to dictionary changes. The experiments show that a symmetric measure relying on vocabulary overlapping can correlate very well with gold-standard comparability levels and is robust to dictionary changes. Based on the comparability measure, two methods, namely the greedy approach and the clustering approach, are then developed to improve the quality of any given comparable corpus. The general idea of these two methods is to choose the highquality subpart from the original corpus and to enrich the low-quality subpart with external resources. The experiments show that one can improve the quality, in terms of comparability scores, of the given comparable corpus by these two methods, with the clustering approach being more efficient than the greedy approach. The enhanced comparable corpus further results in better bilingual lexicons extracted with the standard extraction algorithm. Lastly, we investigate the task of Cross-Language Information Retrieval (CLIR) and the application of comparable corpora in CLIR. We develop novel CLIR models extending the recently proposed information-based models in monolingual IR. The information-based CLIR model is shown to give the best performance overall. Bilingual lexicons extracted from comparable corpora are then combined with the existing bilingual dictionary and used in CLIR experiments, which results in significant improvement of the CLIR system.

This dissertation explores a unique automated approach to map one phoneme set to another, based on the acoustic distances between the individual phonemes. Although the focus of this investigation is on cross-language applications, this automated approach can be extended to same-language but different-database applications as well. The main goal of this investigation is to be able to use the data of a source language, to train the initial acoustic models of a target language for which very little speech data may be available. To do this, an automatic technique for mapping the phonemes of the two data sets must be found. Using this technique, it would be possible to accelerate the development of a speech recognition system for a new language. The current research in the cross-language speech recognition field has focused on manual methods to map phonemes. This investigation has considered an English-to-Afrikaans phoneme mapping, as well as an Afrikaans-to-English phoneme mapping. This has been previously applied to these language instances, but utilising manual phoneme mapping methods. To determine the best phoneme mapping, different acoustic distance measures are compared. The distance measures that are considered are the Kullback-Leibler measure, the Bhattacharyya distance metric, the Mahalanobis measure, the Euclidean measure, the L2 metric and the Jeffreys-Matusita distance. The distance measures are tested by comparing the cross-database recognition results obtained on phoneme models created from the TIMIT speech corpus and a locally-compiled South African SUN Speech database. By selecting the most appropriate distance measure, an automated procedure to map phonemes from the source language to the target language can be done. The best distance measure for the mapping gives recognition rates comparable to a manual mapping process undertaken by a phonetic expert. This study also investigates the effect of the number of Gaussian mixture components on the mapping and on the speech recognition system’s performance. The results indicate that the recogniser’s performance increases up to a limit as the number of mixtures increase. In addition, this study has explored the effect of excluding the Mel Frequency delta and acceleration cepstral coefficients. It is found that the inclusion of these temporal features help improve the mapping and the recognition system’s phoneme recognition rate. Experiments are also carried out to determine the impact of the number of HMM recogniser states. It is found that single-state HMMs deliver the optimum cross-language phoneme recognition results. After having done the mapping, speaker adaptation strategies are applied on the recognisers to improve their target-language performance. The models of a fully trained speech recogniser in a source language are adapted to target-language models using Maximum Likelihood Linear Regression (MLLR) followed by Maximum A Posteriori (MAP) techniques. Embedded Baum-Welch re-estimation is used to further adapt the models to the target language. These techniques result in a considerable improvement in the phoneme recognition rate. Although a combination of MLLR and MAP techniques have been used previously in speech adaptation studies, the combination of MLLR, MAP and EBWR in cross-language speech recognition is a unique contribution of this study. Finally, a data pooling technique is applied to build a new recogniser using the automatically mapped phonemes from the target language as well as the source language phonemes. This new recogniser demonstrates moderate bilingual phoneme recognition capabilities. The bilingual recogniser is then further adapted to the target language using MAP and embedded Baum-Welch re-estimation techniques. This combination of adaptation techniques together with the data pooling strategy is uniquely applied in the field of cross-language recognition. The results obtained using this technique outperform all other techniques tested in terms of phoneme recognition rates, although it requires a considerably more time consuming training process. It displays only slightly poorer phoneme recognition than the recognisers trained and tested on the same language database. / Dissertation (MEng (Computer Engineering))--University of Pretoria, 2006. / Electrical, Electronic and Computer Engineering / unrestricted

Speech recognition systems have been developed for the major languages of the world, yet for the majority of languages there are currently no large vocabulary continuous speech recognition (LVCSR) systems. The development of an LVCSR system for a new language is very costly, mainly because a large speech database has to be compiled to robustly capture the acoustic characteristics of the new language. This thesis investigates techniques that enable the re-use of acoustic information from a source language, in which a large amount of data is available, in implementing a system for a new target language. The assumption is that too little data is available in the target language to train a robust speech recognition system on that data alone, and that use of acoustic information from a source language can improve the performance of a target language recognition system. Strategies for cross-language use of acoustic information are proposed, including training on pooled source and target language data, adaptation of source language models using target language data, adapting multilingual models using target language data and transforming source language data to augment target language data for model training. These strategies are allied with Bayesian and transformation-based techniques, usually used for speaker adaptation, as well as with discriminative learning techniques, to present a framework for cross-language re-use of acoustic information. Extensions to current adaptation techniques are proposed to improve the performance of these techniques specifically for cross-language adaptation. A new technique for transformation-based adaptation of variance parameters and a cost-based extension of the minimum classification error (MCE) approach are proposed. Experiments are performed for a large number of approaches from the proposed framework for cross-language re-use of acoustic information. Relatively large amounts of English speech data are used in conjunction with smaller amounts of Afrikaans speech data to improve the performance of an Afrikaans speech recogniser. Results indicate that a significant reduction in word error rate (between 26% and 50%, depending on the amount of Afrikaans data available) is possible when English acoustic data is used in addition to Afrikaans speech data from the same database (i.e both sets of data were recorded under the same c`12onditions and the same labelling process was used). For same-database experiments, best results are achieved for approaches that train models on pooled source and target language data and then perform further adaptation of the models using Bayesian or discriminative techniques on target language data only. Experiments are also performed to evaluate the use of English data from a different database than the Afrikaans data. Peak reductions in word error rate of between 16% and 35% are delivered, depending on the amount of Afrikaans data available. Best results are achieved for an approach that performs a simple transformation of source model parameters using target language data, and then performs Bayesian adaptation of the transformed model on target language data. / Thesis (PhD (Electrical, Electronic and Computer Engineering))--University of Pretoria, 2006. / Electrical, Electronic and Computer Engineering / unrestricted

In this dissertation, we deal with the cross-view tasks related to information retrieval using embedding methods. We study existing methodologies and propose new methods to overcome their limitations. We formally introduce the concept of mixed-script IR, which deals with the challenges faced by an IR system when a language is written in different scripts because of various technological and sociological factors. Mixed-script terms are represented by a small and finite feature space comprised of character n-grams. We propose the cross-view autoencoder (CAE) to model such terms in an abstract space and CAE provides the state-of-the-art performance. We study a wide variety of models for cross-language information retrieval (CLIR) and propose a model based on compositional neural networks (XCNN) which overcomes the limitations of the existing methods and achieves the best results for many CLIR tasks such as ad-hoc retrieval, parallel sentence retrieval and cross-language plagiarism detection. We empirically test the proposed models for these tasks on publicly available datasets and present the results with analyses. In this dissertation, we also explore an effective method to incorporate contextual similarity for lexical selection in machine translation. Concretely, we investigate a feature based on context available in source sentence calculated using deep autoencoders. The proposed feature exhibits statistically significant improvements over the strong baselines for English-to-Spanish and English-to-Hindi translation tasks. Finally, we explore the the methods to evaluate the quality of autoencoder generated representations of text data and analyse its architectural properties. For this, we propose two metrics based on reconstruction capabilities of the autoencoders: structure preservation index (SPI) and similarity accumulation index (SAI). We also introduce a concept of critical bottleneck dimensionality (CBD) below which the structural information is lost and present analyses linking CBD and language perplexity. / En esta disertación estudiamos problemas de vistas-múltiples relacionados con la recuperación de información utilizando técnicas de representación en espacios de baja dimensionalidad. Estudiamos las técnicas existentes y proponemos nuevas técnicas para solventar algunas de las limitaciones existentes. Presentamos formalmente el concepto de recuperación de información con escritura mixta, el cual trata las dificultades de los sistemas de recuperación de información cuando los textos contienen escrituras en distintos alfabetos debido a razones tecnológicas y socioculturales. Las palabras en escritura mixta son representadas en un espacio de características finito y reducido, compuesto por n-gramas de caracteres. Proponemos los auto-codificadores de vistas-múltiples (CAE, por sus siglas en inglés) para modelar dichas palabras en un espacio abstracto, y esta técnica produce resultados de vanguardia. En este sentido, estudiamos varios modelos para la recuperación de información entre lenguas diferentes (CLIR, por sus siglas en inglés) y proponemos un modelo basado en redes neuronales composicionales (XCNN, por sus siglas en inglés), el cual supera las limitaciones de los métodos existentes. El método de XCNN propuesto produce mejores resultados en diferentes tareas de CLIR tales como la recuperación de información ad-hoc, la identificación de oraciones equivalentes en lenguas distintas y la detección de plagio entre lenguas diferentes. Para tal efecto, realizamos pruebas experimentales para dichas tareas sobre conjuntos de datos disponibles públicamente, presentando los resultados y análisis correspondientes. En esta disertación, también exploramos un método eficiente para utilizar similitud semántica de contextos en el proceso de selección léxica en traducción automática. Específicamente, proponemos características extraídas de los contextos disponibles en las oraciones fuentes mediante el uso de auto-codificadores. El uso de las características propuestas demuestra mejoras estadísticamente significativas sobre sistemas de traducción robustos para las tareas de traducción entre inglés y español, e inglés e hindú. Finalmente, exploramos métodos para evaluar la calidad de las representaciones de datos de texto generadas por los auto-codificadores, a la vez que analizamos las propiedades de sus arquitecturas. Como resultado, proponemos dos nuevas métricas para cuantificar la calidad de las reconstrucciones generadas por los auto-codificadores: el índice de preservación de estructura (SPI, por sus siglas en inglés) y el índice de acumulación de similitud (SAI, por sus siglas en inglés). También presentamos el concepto de dimensión crítica de cuello de botella (CBD, por sus siglas en inglés), por debajo de la cual la información estructural se deteriora. Mostramos que, interesantemente, la CBD está relacionada con la perplejidad de la lengua. / En aquesta dissertació estudiem els problemes de vistes-múltiples relacionats amb la recuperació d'informació utilitzant tècniques de representació en espais de baixa dimensionalitat. Estudiem les tècniques existents i en proposem unes de noves per solucionar algunes de les limitacions existents. Presentem formalment el concepte de recuperació d'informació amb escriptura mixta, el qual tracta les dificultats dels sistemes de recuperació d'informació quan els textos contenen escriptures en diferents alfabets per motius tecnològics i socioculturals. Les paraules en escriptura mixta són representades en un espai de característiques finit i reduït, composat per n-grames de caràcters. Proposem els auto-codificadors de vistes-múltiples (CAE, per les seves sigles en anglès) per modelar aquestes paraules en un espai abstracte, i aquesta tècnica produeix resultats d'avantguarda. En aquest sentit, estudiem diversos models per a la recuperació d'informació entre llengües diferents (CLIR , per les sevas sigles en anglès) i proposem un model basat en xarxes neuronals composicionals (XCNN, per les sevas sigles en anglès), el qual supera les limitacions dels mètodes existents. El mètode de XCNN proposat produeix millors resultats en diferents tasques de CLIR com ara la recuperació d'informació ad-hoc, la identificació d'oracions equivalents en llengües diferents, i la detecció de plagi entre llengües diferents. Per a tal efecte, realitzem proves experimentals per aquestes tasques sobre conjunts de dades disponibles públicament, presentant els resultats i anàlisis corresponents. En aquesta dissertació, també explorem un mètode eficient per utilitzar similitud semàntica de contextos en el procés de selecció lèxica en traducció automàtica. Específicament, proposem característiques extretes dels contextos disponibles a les oracions fonts mitjançant l'ús d'auto-codificadors. L'ús de les característiques proposades demostra millores estadísticament significatives sobre sistemes de traducció robustos per a les tasques de traducció entre anglès i espanyol, i anglès i hindú. Finalment, explorem mètodes per avaluar la qualitat de les representacions de dades de text generades pels auto-codificadors, alhora que analitzem les propietats de les seves arquitectures. Com a resultat, proposem dues noves mètriques per quantificar la qualitat de les reconstruccions generades pels auto-codificadors: l'índex de preservació d'estructura (SCI, per les seves sigles en anglès) i l'índex d'acumulació de similitud (SAI, per les seves sigles en anglès). També presentem el concepte de dimensió crítica de coll d'ampolla (CBD, per les seves sigles en anglès), per sota de la qual la informació estructural es deteriora. Mostrem que, de manera interessant, la CBD està relacionada amb la perplexitat de la llengua. / Gupta, PA. (2017). Cross-view Embeddings for Information Retrieval [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/78457 / TESIS

La mise à disposition massive de documents via Internet (pages Web, entrepôts de données,documents numériques, numérisés ou retranscrits, etc.) rend de plus en plus aisée la récupération d’idées. Malheureusement, ce phénomène s’accompagne d’une augmentation des cas de plagiat.En effet, s’approprier du contenu, peu importe sa forme, sans le consentement de son auteur (ou de ses ayants droit) et sans citer ses sources, dans le but de le présenter comme sa propre œuvre ou création est considéré comme plagiat. De plus, ces dernières années, l’expansion d’Internet a également facilité l’accès à des documents du monde entier (écrits dans des langues étrangères)et à des outils de traduction automatique de plus en plus performants, accentuant ainsi la progression d’un nouveau type de plagiat : le plagiat translingue. Ce plagiat implique l’emprunt d’un texte tout en le traduisant (manuellement ou automatiquement) de sa langue originale vers la langue du document dans lequel le plagiaire veut l’inclure. De nos jours, la prévention du plagiat commence à porter ses fruits, grâce notamment à des logiciels anti-plagiat performants qui reposent sur des techniques de comparaison monolingue déjà bien éprouvées. Néanmoins, ces derniers ne traitent pas encore de manière efficace les cas translingues. Cette thèse est née du besoin de Compilatio, une société d’édition de l’un de ces logiciels anti-plagiat, de mesurer des similarités textuelles sémantiques translingues (sous-tâche de la détection du plagiat). Après avoir défini le plagiat et les différents concepts abordés au cours de cette thèse, nous établissons un état de l’art des différentes approches de détection du plagiat translingue. Nousprésentons également les différents corpus déjà existants pour la détection du plagiat translingue et exposons les limites qu’ils peuvent rencontrer lors d’une évaluation de méthodes de détection du plagiat translingue. Nous présentons ensuite le corpus que nous avons constitué et qui ne possède pas la plupart des limites rencontrées par les différents corpus déjà existants. Nous menons,à l’aide de ce nouveau corpus, une évaluation de plusieurs méthodes de l’état de l’art et découvrons que ces dernières se comportent différemment en fonction de certaines caractéristiques des textes sur lesquelles elles opèrent. Ensuite, nous présentons des nouvelles méthodes de mesure de similarités textuelles sémantiques translingues basées sur des représentations continues de mots(word embeddings). Nous proposons également une notion de pondération morphosyntaxique et fréquentielle de mots, qui peut aussi bien être utilisée au sein d’un vecteur qu’au sein d’un sac de mots, et nous montrons que son introduction dans ces nouvelles méthodes augmente leurs performances respectives. Nous testons ensuite différents systèmes de fusion et combinaison entre différentes méthodes et étudions les performances, sur notre corpus, de ces méthodes et fusions en les comparant à celles des méthodes de l’état de l’art. Nous obtenons ainsi de meilleurs résultats que l’état de l’art dans la totalité des sous-corpus étudiés. Nous terminons en présentant et discutant les résultats de ces méthodes lors de notre participation à la tâche de similarité textuelle sémantique (STS) translingue de la campagne d’évaluation SemEval 2017, où nous nous sommes classés 1er à la sous-tâche correspondant le plus au scénario industriel de Compilatio. / The massive amount of documents through the Internet (e.g. web pages, data warehouses anddigital or transcribed texts) makes easier the recycling of ideas. Unfortunately, this phenomenonis accompanied by an increase of plagiarism cases. Indeed, claim ownership of content, withoutthe consent of its author and without crediting its source, and present it as new and original, isconsidered as plagiarism. In addition, the expansion of the Internet, which facilitates access todocuments throughout the world (written in foreign languages) as well as increasingly efficient(and freely available) machine translation tools, contribute to spread a new kind of plagiarism:cross-language plagiarism. Cross-language plagiarism means plagiarism by translation, i.e. a texthas been plagiarized while being translated (manually or automatically) from its original languageinto the language of the document in which the plagiarist wishes to include it. While prevention ofplagiarism is an active field of research and development, it covers mostly monolingual comparisontechniques. This thesis is a joint work between an academic laboratory (LIG) and Compilatio (asoftware publishing company of solutions for plagiarism detection), and proposes cross-lingualsemantic textual similarity measures, which is an important sub-task of cross-language plagiarismdetection.After defining the plagiarism and the different concepts discussed during this thesis, wepresent a state-of-the-art of the different cross-language plagiarism detection approaches. Wealso present the preexisting corpora for cross-language plagiarism detection and show their limits.Then we describe how we have gathered and built a new dataset, which does not contain mostof the limits encountered by the preexisting corpora. Using this new dataset, we conduct arigorous evaluation of several state-of-the-art methods and discover that they behave differentlyaccording to certain characteristics of the texts on which they operate. We next present newmethods for measuring cross-lingual semantic textual similarities based on word embeddings.We also propose a notion of morphosyntactic and frequency weighting of words, which can beused both within a vector and within a bag-of-words, and we show that its introduction inthe new methods increases their respective performance. Then we test different fusion systems(mostly based on linear regression). Our experiments show that we obtain better results thanthe state-of-the-art in all the sub-corpora studied. We conclude by presenting and discussingthe results of these methods obtained during our participation to the cross-lingual SemanticTextual Similarity (STS) task of SemEval-2017, where we ranked 1st on the sub-task that bestcorresponds to Compilatio’s use-case scenario.

La popularisation des réseaux sociaux et des documents numériques a rapidement accru l'information disponible sur Internet. Cependant, cette quantité massive de données ne peut pas être analysée manuellement. Parmi les applications existantes du Traitement Automatique du Langage Naturel (TALN), nous nous intéressons dans cette thèse au résumé cross-lingue de texte, autrement dit à la production de résumés dans une langue différente de celle des documents sources. Nous analysons également d'autres tâches du TALN (la représentation des mots, la similarité sémantique ou encore la compression de phrases et de groupes de phrases) pour générer des résumés cross-lingues plus stables et informatifs. La plupart des applications du TALN, celle du résumé automatique y compris, utilisent une mesure de similarité pour analyser et comparer le sens des mots, des séquences de mots, des phrases et des textes. L’une des façons d'analyser cette similarité est de générer une représentation de ces phrases tenant compte de leur contenu. Le sens des phrases est défini par plusieurs éléments, tels que le contexte des mots et des expressions, l'ordre des mots et les informations précédentes. Des mesures simples, comme la mesure cosinus et la distance euclidienne, fournissent une mesure de similarité entre deux phrases. Néanmoins, elles n'analysent pas l'ordre des mots ou les séquences de mots. En analysant ces problèmes, nous proposons un modèle de réseau de neurones combinant des réseaux de neurones récurrents et convolutifs pour estimer la similarité sémantique d'une paire de phrases (ou de textes) en fonction des contextes locaux et généraux des mots. Sur le jeu de données analysé, notre modèle a prédit de meilleurs scores de similarité que les systèmes de base en analysant mieux le sens local et général des mots mais aussi des expressions multimots. Afin d'éliminer les redondances et les informations non pertinentes de phrases similaires, nous proposons de plus une nouvelle méthode de compression multiphrase, fusionnant des phrases au contenu similaire en compressions courtes. Pour ce faire, nous modélisons des groupes de phrases semblables par des graphes de mots. Ensuite, nous appliquons un modèle de programmation linéaire en nombres entiers qui guide la compression de ces groupes à partir d'une liste de mots-clés ; nous cherchons ainsi un chemin dans le graphe de mots qui a une bonne cohésion et qui contient le maximum de mots-clés. Notre approche surpasse les systèmes de base en générant des compressions plus informatives et plus correctes pour les langues française, portugaise et espagnole. Enfin, nous combinons les méthodes précédentes pour construire un système de résumé de texte cross-lingue. Notre système génère des résumés cross-lingue de texte en analysant l'information à la fois dans les langues source et cible, afin d’identifier les phrases les plus pertinentes. Inspirés par les méthodes de résumé de texte par compression en analyse monolingue, nous adaptons notre méthode de compression multiphrase pour ce problème afin de ne conserver que l'information principale. Notre système s'avère être performant pour compresser l'information redondante et pour préserver l'information pertinente, en améliorant les scores d'informativité sans perdre la qualité grammaticale des résumés cross-lingues du français vers l'anglais. En analysant les résumés cross-lingues depuis l’anglais, le français, le portugais ou l’espagnol, vers l’anglais ou le français, notre système améliore les systèmes par extraction de l'état de l'art pour toutes ces langues. En outre, une expérience complémentaire menée sur des transcriptions automatiques de vidéo montre que notre approche permet là encore d'obtenir des scores ROUGE meilleurs et plus stables, même pour ces documents qui présentent des erreurs grammaticales et des informations inexactes ou manquantes. / The popularization of social networks and digital documents increased quickly the informationavailable on the Internet. However, this huge amount of data cannot be analyzedmanually. Natural Language Processing (NLP) analyzes the interactions betweencomputers and human languages in order to process and to analyze natural languagedata. NLP techniques incorporate a variety of methods, including linguistics, semanticsand statistics to extract entities, relationships and understand a document. Amongseveral NLP applications, we are interested, in this thesis, in the cross-language textsummarization which produces a summary in a language different from the languageof the source documents. We also analyzed other NLP tasks (word encoding representation,semantic similarity, sentence and multi-sentence compression) to generate morestable and informative cross-lingual summaries.Most of NLP applications (including all types of text summarization) use a kind ofsimilarity measure to analyze and to compare the meaning of words, chunks, sentencesand texts in their approaches. A way to analyze this similarity is to generate a representationfor these sentences that contains the meaning of them. The meaning of sentencesis defined by several elements, such as the context of words and expressions, the orderof words and the previous information. Simple metrics, such as cosine metric andEuclidean distance, provide a measure of similarity between two sentences; however,they do not analyze the order of words or multi-words. Analyzing these problems,we propose a neural network model that combines recurrent and convolutional neuralnetworks to estimate the semantic similarity of a pair of sentences (or texts) based onthe local and general contexts of words. Our model predicted better similarity scoresthan baselines by analyzing better the local and the general meanings of words andmulti-word expressions.In order to remove redundancies and non-relevant information of similar sentences,we propose a multi-sentence compression method that compresses similar sentencesby fusing them in correct and short compressions that contain the main information ofthese similar sentences. We model clusters of similar sentences as word graphs. Then,we apply an integer linear programming model that guides the compression of theseclusters based on a list of keywords. We look for a path in the word graph that has goodcohesion and contains the maximum of keywords. Our approach outperformed baselinesby generating more informative and correct compressions for French, Portugueseand Spanish languages. Finally, we combine these previous methods to build a cross-language text summarizationsystem. Our system is an {English, French, Portuguese, Spanish}-to-{English,French} cross-language text summarization framework that analyzes the informationin both languages to identify the most relevant sentences. Inspired by the compressivetext summarization methods in monolingual analysis, we adapt our multi-sentencecompression method for this problem to just keep the main information. Our systemproves to be a good alternative to compress redundant information and to preserve relevantinformation. Our system improves informativeness scores without losing grammaticalquality for French-to-English cross-lingual summaries. Analyzing {English,French, Portuguese, Spanish}-to-{English, French} cross-lingual summaries, our systemsignificantly outperforms extractive baselines in the state of the art for all these languages.In addition, we analyze the cross-language text summarization of transcriptdocuments. Our approach achieved better and more stable scores even for these documentsthat have grammatical errors and missing information.

This study focuses on how the cross-language perception of German vowels by native speakers of North American English differs based on various levels of classroom instruction and experience in a German-speaking country. Of special interest is whether more advanced students and those with target country experience have a different cross-language perception of German vowels from naive or less-experienced listeners. It further examines how English-speaking learners perceive German sounds that are not found in English, namely the front-rounded vowels. Study participants were students at Brigham Young University, divided into four groups: those 1) without knowledge of German; 2) in their 3rd semester of German without stay abroad; 3) in their 5th semester of German or above without stay abroad; and 4) in their 5th semester or above with at least 12 months in a German-speaking country. The subjects performed two tasks. While listening to German words, they first selected the English word with the vowel that most closely matched the German vowel heard from a list of English words on the computer screen; and secondly, they rated how much alike the German vowel sounded like the English vowel they chose. The results indicate that level of instruction does indeed affect how subjects perceive German vowels. Moreover, perception of the vowels was to some degree affected by the consonant environment. Finally, it was found that all groups rated the similarity of vowels in a similar manner regardless of experience.

The Social Web, or Web 2.0, has recently gained popularity because of its low cost and ease of use. Social tagging sites (e.g. Flickr and YouTube) offer new principles for end-users to publish and classify their content (data). Tagging systems contain free-keywords (tags) generated by end-users to annotate and categorise data. Lack of semantics is the main drawback in social tagging due to the use of unstructured vocabulary. Therefore, tagging systems suffer from shortcomings such as low precision, lack of collocation, synonymy, multilinguality, and use of shorthands. Consequently, relevant contents are not visible, and thus not retrievable while searching in tag-based systems. On the other hand, the Semantic Web, so-called Web 3.0, provides a rich semantic infrastructure. Ontologies are the key enabling technology for the Semantic Web. Ontologies can be integrated with the Social Web to overcome the lack of semantics in tagging systems. In the work presented in this thesis, we build an architecture to address a number of tagging systems drawbacks. In particular, we make use of the controlled vocabularies presented by ontologies to improve the information retrieval in tag-based systems. Based on the tags provided by the end-users, we introduce the idea of adding “system tags” from semantic, as well as social, resources. The “system tags” are comprehensive and wide-ranging in comparison with the limited “user tags”. The system tags are used to fill the gap between the user tags and the search terms used for searching in the tag-based systems. We restricted the scope of our work to tackle the following tagging systems shortcomings: - The lack of semantic relations between user tags and search terms (e.g. synonymy, hypernymy), - The lack of translation mediums between user tags and search terms (multilinguality), - The lack of context to define the emergent shorthand writing user tags. To address the first shortcoming, we use the WordNet ontology as a semantic lingual resource from where system tags are extracted. For the second shortcoming, we use the MultiWordNet ontology to recognise the cross-languages linkages between different languages. Finally, to address the third shortcoming, we use tag clusters that are obtained from the Social Web to create a context for defining the meaning of shorthand writing tags. A prototype for our architecture was implemented. In the prototype system, we built our own database to host videos that we imported from real tag-based system (YouTube). The user tags associated with these videos were also imported and stored in the database. For each user tag, our algorithm adds a number of system tags that came from either semantic ontologies (WordNet or MultiWordNet), or from tag clusters that are imported from the Flickr website. Therefore, each system tag added to annotate the imported videos has a relationship with one of the user tags on that video. The relationship might be one of the following: synonymy, hypernymy, similar term, related term, translation, or clustering relation. To evaluate the suitability of our proposed system tags, we developed an online environment where participants submit search terms and retrieve two groups of videos to be evaluated. Each group is produced from one distinct type of tags; user tags or system tags. The videos in the two groups are produced from the same database and are evaluated by the same participants in order to have a consistent and reliable evaluation. Since the user tags are used nowadays for searching the real tag-based systems, we consider its efficiency as a criterion (reference) to which we compare the efficiency of the new system tags. In order to compare the relevancy between the search terms and each group of retrieved videos, we carried out a statistical approach. According to Wilcoxon Signed-Rank test, there was no significant difference between using either system tags or user tags. The findings revealed that the use of the system tags in the search is as efficient as the use of the user tags; both types of tags produce different results, but at the same level of relevance to the submitted search terms.

Research on clear speech has shown that the type of clear speech produced can vary depending on the speaker, the listener and the medium. Although prior research has suggested that clear speech is more intelligible than conversational speech for normal-hearing listeners in noisy environments, it is not known which acoustic features of clear speech are the most responsible for enhanced intelligibility and comprehension. This thesis focused on investigating the acoustic characteristics that are produced in clear speech to foreigners and infants. Its aim was to assess the utility of these features in enhancing speech intelligibility and comprehension. The results of Experiment 1 showed that native speakers produced exaggerated vowel space in natural interactions with foreign-accented listeners compared to native-accented listeners. Results of Experiment 2 indicated that native speakers exaggerated vowel space and pitch to infants compared to clear read speech. Experiments 3 and 4 focused on speech perception and used transcription and clarity rating tasks. Experiment 3 contained speech directed at foreigners and showed that speech to foreign-accented speakers was rated clearer than speech to native-accented speakers. Experiment 4 contained speech directed at infants and showed that native speakers rated infant-directed speech as clearer than clear read speech. In the fifth and final experiment, naturally elicited clear speech towards foreign-accented interlocutors was used in speech comprehension tasks for native and non-native listeners with varying proficiency of English. It was revealed that speech with expanded vowel space improved listeners’ comprehension of speech in quiet and noise conditions. Results are discussed in terms of the Lindblom’s (1990) theory of Hyper and Hypoarticulation, an influential framework of speech production and perception.

Cette thèse porte sur la construction automatique d’outils et de ressources pour l’analyse linguistique de textes des langues peu dotées. Nous proposons une approche utilisant des réseaux de neurones récurrents (RNN - Recurrent Neural Networks) et n'ayant besoin que d'un corpus parallèle ou mutli-parallele entre une langue source bien dotée et une ou plusieurs langues cibles moins bien ou peu dotées. Ce corpus parallèle ou mutli-parallele est utilisé pour la construction d'une représentation multilingue des mots des langues source et cible. Nous avons utilisé cette représentation multilingue pour l’apprentissage de nos modèles neuronaux et nous avons exploré deux architectures neuronales : les RNN simples et les RNN bidirectionnels. Nous avons aussi proposé plusieurs variantes des RNN pour la prise en compte d'informations linguistiques de bas niveau (informations morpho-syntaxiques) durant le processus de construction d'annotateurs linguistiques de niveau supérieur (SuperSenses et dépendances syntaxiques). Nous avons démontré la généricité de notre approche sur plusieurs langues ainsi que sur plusieurs tâches d'annotation linguistique. Nous avons construit trois types d'annotateurs linguistiques multilingues: annotateurs morpho-syntaxiques, annotateurs en SuperSenses et annotateurs en dépendances syntaxiques, avec des performances très satisfaisantes. Notre approche a les avantages suivants : (a) elle n'utilise aucune information d'alignement des mots, (b) aucune connaissance concernant les langues cibles traitées n'est requise au préalable (notre seule supposition est que, les langues source et cible n'ont pas une grande divergence syntaxique), ce qui rend notre approche applicable pour le traitement d'un très grand éventail de langues peu dotées, (c) elle permet la construction d'annotateurs multilingues authentiques (un annotateur pour N langages). / This thesis focuses on the automatic construction of linguistic tools and resources for analyzing texts of low-resource languages. We propose an approach using Recurrent Neural Networks (RNN) and requiring only a parallel or multi-parallel corpus between a well-resourced language and one or more low-resource languages. This parallel or multi-parallel corpus is used to construct a multilingual representation of words of the source and target languages. We used this multilingual representation to train our neural models and we investigated both uni and bidirectional RNN models. We also proposed a method to include external information (for instance, low-level information from Part-Of-Speech tags) in the RNN to train higher level taggers (for instance, SuperSenses taggers and Syntactic dependency parsers). We demonstrated the validity and genericity of our approach on several languages and we conducted experiments on various NLP tasks: Part-Of-Speech tagging, SuperSenses tagging and Dependency parsing. The obtained results are very satisfactory. Our approach has the following characteristics and advantages: (a) it does not use word alignment information, (b) it does not assume any knowledge about target languages (one requirement is that the two languages (source and target) are not too syntactically divergent), which makes it applicable to a wide range of low-resource languages, (c) it provides authentic multilingual taggers (one tagger for N languages).