Dissertations / Theses on the topic 'Domain translation'

Statistical Machine Translation (SMT) is currently the most promising and widely studied paradigm in the broader field of Machine Translation, continuously explored in order to improve its performance and to find solutions to its current shortcomings, in particular the sparsity of big bilingual corpora in a variety of domains or genres to be used as training data. However, while one the main trends is still to rely as much as possible on already available large collections of data, even when they do not fit quite well specific translation tasks in terms of relatedness of content, the possibility of using less but appropriately selected training sets - depending on the textual variety of the documents that need to be translated case by case - has not been extensively explored as much so far. The goal of this research is to investigate whether this latter possibility, i.e. the lack of availability of large quantities of assorted data, can have a possible solution in the application of strategies commonly used in genre and domain classification (including unsupervised topic modeling and document dissimilarity techniques), in particular performing subsampling experiments on bilingual corpora in order to obtain a good fit between training data and the texts that need to be translated with SMT. For the purposes of this study, already existing freely available large corpora were found to be unsuitable for the selection of domain/document specifc subsamples, so two new parallel corpora - English-Italian and English-German - were compiled employing the \web as corpus" approach on websites containing translated content. Then some tests were made on documents belonging to different varieties, translated with SMT systems built using subsamples of training data selected using document dissimilarity measures in order to pick up the most suitable documents as training data. Such method has shown how the choice of subsampling strategy heavily depends on the text variety of each considered document, but it has also proven that better translation results can be obtained from small samples of training sets rather than using all the available data, which brings benefits also in terms of quicker training times and use of fewer computational resources.

Cette thèse explore la génération multi-modale dans un contexte d'apprentissage semi-supervisé, en abordant deux défis cruciaux : la prise en charge de configurations flexibles d'entrées et de sorties à travers plusieurs domaines, et le développement d'une stratégie d'entraînement efficace des données semi-supervisées. Alors que les systèmes d'intelligence artificielle progressent, il existe un besoin croissant de modèles capables d'intégrer et de générer de manière flexible plusieurs modalités, reflétant les capacités cognitives humaines. Les systèmes d'apprentissage profond conventionnels peinent souvent lorsqu'ils s'écartent de leur configuration d'entraînement, notamment lorsque certaines modalités sont indisponibles dans les applications réelles. Par exemple, dans le domaine médical, les patients pourraient ne pas faire tous les examens possibles pour un système d'analyse complet. Obtenir un contrôle plus fin sur les modalités générées est crucial pour améliorer les capacités de génération et fournir des informations contextuelles plus riches. De plus, l'augmentation du nombre de domaines rend plus difficile l'obtention d'une supervision simultanée. Nous nous concentrons sur la translation multi-domaine dans un contexte semi-supervisé, étendant le paradigme classique de translation de domaine. Plutôt que de considérer une direction de translation spécifique ou de les limiter entre paires de domaines, nous développons des méthodes facilitant les translations entre toutes les configurations possibles de domaines. L'aspect semi-supervisé reflète des scénarios réels où une annotation complète des données est souvent infaisable ou prohibitivement coûteuse. Notre travail présente trois contributions : (1) l'étude des fonctions de régularisation pour l'espace latent avec une supervision limitée, (2) l'étude de la mise à l'échelle et de la flexibilité des modèles de translation basés sur les modèles de diffusion, et (3) l'amélioration de la vitesse de génération des modèles d'inpainting par diffusion. Premièrement, nous proposons LSM, un modèle de translation semi-supervisé exploitant des données d'entrée supplémentaires et des données de sortie structurées pour régulariser les dépendances inter-domaines et intra-domaines. Deuxièmement, nous développons MDD, un modèle semi-supervisé de translation multi-domaine basé sur la diffusion. MDD transforme la fonction de perte classique des modèles de diffusion d'une fonction de reconstruction vers une fonction de translations en modélisant différents niveaux de bruit par domaine. Le modèle exploite les domaines moins bruités pour reconstruire les domaines plus bruités, permettant de modéliser les données manquantes comme du bruit pur et d'obtenir une configuration flexible des domaines condition et cible. Enfin, nous introduisons TD-Paint, un modèle d'inpainting basé sur la diffusion améliorant la vitesse de génération et à réduire la charge de calcul associée à la génération. Notre étude révèle que les modèles d'inpainting par diffusion souffrent d'une désynchronisation entre génération et conditionnement. Les solutions existantes, reposant sur le rééchantillonnage ou des régularisations supplémentaires, augmentent la complexité computationnelle. TD-Paint résout ce problème en modélisant des niveaux de bruit variables au niveau des pixels, permettant une utilisation efficace de la condition dès le début du processus
This thesis explores multi-modal generation and semi-supervised learning, addressing two critical challenges: supporting flexible configurations of input and output across multiple domains, and developing efficient training strategies for semi-supervised data settings. As artificial intelligence systems advance, there is growing need for models that can flexibly integrate and generate multiple modalities, mirroring human cognitive abilities. Conventional deep learning systems often struggle when deviating from their training configuration, which occurs when certain modalities are unavailable in real-world applications. For instance, in medical settings, patients might not undergo all possible scans for a comprehensive analysis system. Additionally, obtaining finer control over generated modalities is crucial for enhancing generation capabilities and providing richer contextual information. As the number of domains increases, obtaining simultaneous supervision across all domains becomes increasingly challenging. We focus on multi-domain translation in a semi-supervised setting, extending the classical domain translation paradigm. Rather than addressing specific translation directions or limiting translations to domain pairs, we develop methods facilitating translations between any possible domain configurations, determined at test time. The semi-supervised aspect reflects real-world scenarios where complete data annotation is often infeasible or prohibitively expensive. Our work explores three main areas: (1) studying latent space regularization functions to enhance domain translation learning with limited supervision, (2) examining the scalability and flexibility of diffusion-based translation models, and (3) improving the generation speed of diffusion-based inpainting models. First, we propose LSM, a semi-supervised translation framework leveraging additional input and structured output data to regularize inter-domain and intra-domain dependencies. Second, we develop MDD, a novel diffusion-based multi-domain translation semi-supervised framework. MDD shifts the classical reconstruction loss of diffusion models to a translation loss by modeling different noise levels per domain. The model leverages less noisy domains to reconstruct noisier ones, modeling missing data from the semi-supervised setting as pure noise and enabling flexible configuration of condition and target domains. Finally, we introduce TD-Paint, a novel diffusion-based inpainting model improving generation speed and reducing computational burden. Through investigation of the generation sampling process, we observe that diffusion-based inpainting models suffer from unsynchronized generation and conditioning. Existing models often rely on resampling steps or additional regularization losses to realign condition and generation, increasing time and computational complexity. TD-Paint addresses this by modeling variable noise levels at the pixel level, enabling efficient use of the condition from the generation onset

[ES] La Traducción Automática Estadística es un sup-campo de la lingüística computacional que investiga como emplear los ordenadores en el proceso de traducción de un texto de un lenguaje humano a otro. La traducción automática estadística es el enfoque más popular que se emplea para construir estos sistemas de traducción automáticos. La calidad de dichos sistemas depende en gran medida de los ejemplos de traducción que se emplean durante los procesos de entrenamiento y adaptación de los modelos. Los conjuntos de datos empleados son obtenidos a partir de una gran variedad de fuentes y en muchos casos puede que no tengamos a mano los datos más adecuados para un dominio específico. Dado este problema de carencia de datos, la idea principal para solucionarlo es encontrar aquellos conjuntos de datos más adecuados para entrenar o adaptar un sistema de traducción. En este sentido, esta tesis propone un conjunto de técnicas de selección de datos que identifican los datos bilingües más relevantes para una tarea extraídos de un gran conjunto de datos. Como primer paso en esta tesis, las técnicas de selección de datos son aplicadas para mejorar la calidad de la traducción de los sistemas de traducción bajo el paradigma basado en frases. Estas técnicas se basan en el concepto de representación continua de las palabras o las oraciones en un espacio vectorial. Los resultados experimentales demuestran que las técnicas utilizadas son efectivas para diferentes lenguajes y dominios. El paradigma de Traducción Automática Neuronal también fue aplicado en esta tesis. Dentro de este paradigma, investigamos la aplicación que pueden tener las técnicas de selección de datos anteriormente validadas en el paradigma basado en frases. El trabajo realizado se centró en la utilización de dos tareas diferentes de adaptación del sistema. Por un lado, investigamos cómo aumentar la calidad de traducción del sistema, aumentando el tamaño del conjunto de entrenamiento. Por otro lado, el método de selección de datos se empleó para crear un conjunto de datos sintéticos. Los experimentos se realizaron para diferentes dominios y los resultados de traducción obtenidos son convincentes para ambas tareas. Finalmente, cabe señalar que las técnicas desarrolladas y presentadas a lo largo de esta tesis pueden implementarse fácilmente dentro de un escenario de traducción real.
[CAT] La Traducció Automàtica Estadística és un sup-camp de la lingüística computacional que investiga com emprar els ordinadors en el procés de traducció d'un text d'un llenguatge humà a un altre. La traducció automàtica estadística és l'enfocament més popular que s'empra per a construir aquests sistemes de traducció automàtics. La qualitat d'aquests sistemes depèn en gran mesura dels exemples de traducció que s'empren durant els processos d'entrenament i adaptació dels models. Els conjunts de dades emprades són obtinguts a partir d'una gran varietat de fonts i en molts casos pot ser que no tinguem a mà les dades més adequades per a un domini específic. Donat aquest problema de manca de dades, la idea principal per a solucionar-ho és trobar aquells conjunts de dades més adequades per a entrenar o adaptar un sistema de traducció. En aquest sentit, aquesta tesi proposa un conjunt de tècniques de selecció de dades que identifiquen les dades bilingües més rellevants per a una tasca extrets d'un gran conjunt de dades. Com a primer pas en aquesta tesi, les tècniques de selecció de dades són aplicades per a millorar la qualitat de la traducció dels sistemes de traducció sota el paradigma basat en frases. Aquestes tècniques es basen en el concepte de representació contínua de les paraules o les oracions en un espai vectorial. Els resultats experimentals demostren que les tècniques utilitzades són efectives per a diferents llenguatges i dominis. El paradigma de Traducció Automàtica Neuronal també va ser aplicat en aquesta tesi. Dins d'aquest paradigma, investiguem l'aplicació que poden tenir les tècniques de selecció de dades anteriorment validades en el paradigma basat en frases. El treball realitzat es va centrar en la utilització de dues tasques diferents. D'una banda, investiguem com augmentar la qualitat de traducció del sistema, augmentant la grandària del conjunt d'entrenament. D'altra banda, el mètode de selecció de dades es va emprar per a crear un conjunt de dades sintètiques. Els experiments es van realitzar per a diferents dominis i els resultats de traducció obtinguts són convincents per a ambdues tasques. Finalment, cal assenyalar que les tècniques desenvolupades i presentades al llarg d'aquesta tesi poden implementar-se fàcilment dins d'un escenari de traducció real.
[EN] La Traducció Automàtica Estadística és un sup-camp de la lingüística computacional que investiga com emprar els ordinadors en el procés de traducció d'un text d'un llenguatge humà a un altre. La traducció automàtica estadística és l'enfocament més popular que s'empra per a construir aquests sistemes de traducció automàtics. La qualitat d'aquests sistemes depèn en gran mesura dels exemples de traducció que s'empren durant els processos d'entrenament i adaptació dels models. Els conjunts de dades emprades són obtinguts a partir d'una gran varietat de fonts i en molts casos pot ser que no tinguem a mà les dades més adequades per a un domini específic. Donat aquest problema de manca de dades, la idea principal per a solucionar-ho és trobar aquells conjunts de dades més adequades per a entrenar o adaptar un sistema de traducció. En aquest sentit, aquesta tesi proposa un conjunt de tècniques de selecció de dades que identifiquen les dades bilingües més rellevants per a una tasca extrets d'un gran conjunt de dades. Com a primer pas en aquesta tesi, les tècniques de selecció de dades són aplicades per a millorar la qualitat de la traducció dels sistemes de traducció sota el paradigma basat en frases. Aquestes tècniques es basen en el concepte de representació contínua de les paraules o les oracions en un espai vectorial. Els resultats experimentals demostren que les tècniques utilitzades són efectives per a diferents llenguatges i dominis. El paradigma de Traducció Automàtica Neuronal també va ser aplicat en aquesta tesi. Dins d'aquest paradigma, investiguem l'aplicació que poden tenir les tècniques de selecció de dades anteriorment validades en el paradigma basat en frases. El treball realitzat es va centrar en la utilització de dues tasques diferents d'adaptació del sistema. D'una banda, investiguem com augmentar la qualitat de traducció del sistema, augmentant la grandària del conjunt d'entrenament. D'altra banda, el mètode de selecció de dades es va emprar per a crear un conjunt de dades sintètiques. Els experiments es van realitzar per a diferents dominis i els resultats de traducció obtinguts són convincents per a ambdues tasques. Finalment, cal assenyalar que les tècniques desenvolupades i presentades al llarg d'aquesta tesi poden implementar-se fàcilment dins d'un escenari de traducció real.
Chinea Ríos, M. (2019). Advanced techniques for domain adaptation in Statistical Machine Translation [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/117611
TESIS

In this thesis we investigate methods for deploying machine translation (MT) in real-world application scenarios related to the use of MT in computer assisted translation (CAT), where human translators post-edit MT outputs. In particular, we investigate (in chronological order) MT adaptation under two working conditions: single-domain and multi-domain. In the former, we assume that MT receives requests by a single user working on a single domain, while in the latter we assume the MT system to receive requests i) from multiple users working on different domains, ii) with no predefined order, and iii) without domain information. In the single-domain case, we first focus on word alignment, a core component of online adaptive phrase-based MT (PBMT) that is crucial for extracting features from a post-edited segment. In particular, we concentrate on improving word alignment in presence of out-of-vocabulary words observed in the source sentences or introduced by the post-editor. In the multi-domain scenario we turned our focus to the neural MT (NMT) paradigm. In particular, we introduce a scalable solution that adapts on-the-fly a generic NMT model to each incoming translation request. It relies on a procedure that locally fine-tunes the model to each input sentence using samples retrieved from a pool of parallel data. Our instance-based adaptation uses a more general formulation of the log-likelihood approach to control the contribution of relevant and irrelevant words during model update. Finally, we test our approach on a simulated continuous learning setting, where the system receives user feedback under form of post-editing.

In this thesis we investigate methods for deploying machine translation (MT) in real-world application scenarios related to the use of MT in computer assisted translation (CAT), where human translators post-edit MT outputs. In particular, we investigate (in chronological order) MT adaptation under two working conditions: single-domain and multi-domain. In the former, we assume that MT receives requests by a single user working on a single domain, while in the latter we assume the MT system to receive requests i) from multiple users working on different domains, ii) with no predefined order, and iii) without domain information. In the single-domain case, we first focus on word alignment, a core component of online adaptive phrase-based MT (PBMT) that is crucial for extracting features from a post-edited segment. In particular, we concentrate on improving word alignment in presence of out-of-vocabulary words observed in the source sentences or introduced by the post-editor. In the multi-domain scenario we turned our focus to the neural MT (NMT) paradigm. In particular, we introduce a scalable solution that adapts on-the-fly a generic NMT model to each incoming translation request. It relies on a procedure that locally fine-tunes the model to each input sentence using samples retrieved from a pool of parallel data. Our instance-based adaptation uses a more general formulation of the log-likelihood approach to control the contribution of relevant and irrelevant words during model update. Finally, we test our approach on a simulated continuous learning setting, where the system receives user feedback under form of post-editing.

O treinamento de sistemas de Tradução de Máquina baseada em Estatística (TME) é bastante dependente da disponibilidade de corpora paralelos. Entretanto, este tipo de recurso costuma ser difícil de ser encontrado, especialmente quando lida com idiomas com poucos recursos ou com tópicos muito específicos, como, por exemplo, dermatologia. Para contornar esta situação, uma possibilidade é utilizar corpora comparáveis, que são recursos muito mais abundantes. Um modo de adquirir corpora comparáveis é a aplicação de algoritmos de Coleta Focada (CF). Neste trabalho, são propostas novas abordagens para CF, algumas baseadas em n-gramas e outras no poder expressivo das expressões multipalavra. Também são avaliadas a viabilidade do uso de CF para realização de adaptação de domínio para sistemas genéricos de TME e se há alguma correlação entre a qualidade dos algoritmos de CF e dos sistemas de TME que podem ser construídos a partir dos respectivos dados coletados. Os resultados indicam que algoritmos de CF podem ser bons meios para adquirir corpora comparáveis para realizar adaptação de domínio para TME e que há uma correlação entre a qualidade dos dois processos.
Statistical Machine Translation (SMT) is highly dependent on the availability of parallel corpora for training. However, these kinds of resource may be hard to be found, especially when dealing with under-resourced languages or very specific domains, like the dermatology. For working this situation around, one possibility is the use of comparable corpora, which are much more abundant resources. One way of acquiring comparable corpora is to apply Focused Crawling (FC) algorithms. In this work we propose novel approach for FC algorithms, some based on n-grams and other on the expressive power of multiword expressions. We also assess the viability of using FC for performing domain adaptations for generic SMT systems and whether there is a correlation between the quality of the FC algorithms and of the SMT systems that can be built with its collected data. Results indicate that the use of FCs is, indeed, a good way for acquiring comparable corpora for SMT domain adaptation and that there is a correlation between the qualities of both processes.

Image-to-image (i2i) translation networks can generate fake images beneficial for many applications in augmented reality, computer graphics, and robotics. However, they require large scale datasets and high contextual understanding to be trained correctly. In this thesis, we propose strategies for solving these problems, improving performances of i2i translation networks by using domain- or physics-related priors. The thesis is divided into two parts. In Part I, we exploit human abstraction capabilities to identify existing relationships in images, thus defining domains that can be leveraged to improve data usage efficiency. We use additional domain-related information to train networks on web-crawled data, hallucinate scenarios unseen during training, and perform few-shot learning. In Part II, we instead rely on physics priors. First, we combine realistic physics-based rendering with generative networks to boost outputs realism and controllability. Then, we exploit naive physical guidance to drive a manifold reorganization, which allowed generating continuous conditions such as timelapses.

Unsupervised image-to-image translation (UNIT) is a challenging task in the image manipulation field, where input images in a visual domain are mapped into another domain with desired visual patterns (also called styles). An ideal direction in this field is to build a model that can map an input image in a domain to multiple target domains and generate diverse outputs in each target domain, which is termed as multi-domain and multi-modal unsupervised image-to-image translation (MMUIT). Recent studies have shown remarkable results in UNIT but they suffer from four main limitations: (1) State-of-the-art UNIT methods are either built from several two-domain mappings that are required to be learned independently or they generate low-diversity results, a phenomenon also known as model collapse. (2) Most of the manipulation is with the assistance of visual maps or digital labels without exploring natural languages, which could be more scalable and flexible in practice. (3) In an MMUIT system, the style latent space is usually disentangled between every two image domains. While interpolations within domains are smooth, interpolations between two different domains often result in unrealistic images with artifacts when interpolating between two randomly sampled style representations from two different domains. Improving the smoothness of the style latent space can lead to gradual interpolations between any two style latent representations even between any two domains. (4) It is expensive to train MMUIT models from scratch at high resolution. Interpreting the latent space of pre-trained unconditional GANs can achieve pretty good image translations, especially high-quality synthesized images (e.g., 1024x1024 resolution). However, few works explore building an MMUIT system with such pre-trained GANs. In this thesis, we focus on these vital issues and propose several techniques for building better MMUIT systems. First, we base on the content-style disentangled framework and propose to fit the style latent space with Gaussian Mixture Models (GMMs). It allows a well-trained network using a shared disentangled style latent space to model multi-domain translations. Meanwhile, we can randomly sample different style representations from a Gaussian component or use a reference image for style transfer. Second, we show how the GMM-modeled latent style space can be combined with a language model (e.g., a simple LSTM network) to manipulate multiple styles by using textual commands. Then, we not only propose easy-to-use constraints to improve the smoothness of the style latent space in MMUIT models, but also design a novel metric to quantitatively evaluate the smoothness of the style latent space. Finally, we build a new model to use pretrained unconditional GANs to do MMUIT tasks.

Thesis (S.M.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2008.
Includes bibliographical references (p. 86-87).
Second language learning is a compelling activity in today's global markets. This thesis focuses on critical technology necessary to produce a computer spoken translation game for learning Mandarin Chinese in a relatively broad travel domain. Three main aspects are addressed: efficient Chinese parsing, high-quality English-Chinese machine translation, and how these technologies can be integrated into a translation game system. In the language understanding component, the TINA parser is enhanced with bottom-up and long distance constraint features. The results showed that with these features, the Chinese grammar ran ten times faster and covered 15% more of the test set. In the machine translation component, a combined method of linguistic and statistical system is introduced. The English-Chinese translation is done via an intermediate language "Zhonglish", where the English-Zhonglish translation is accomplished by a parse-and-paraphrase paradigm using hand-coded rules, mainly for structural reconstruction. Zhonglish-Chinese translation is accomplished by a standard phrase based statistical machine translation system, mostly accomplishing word sense disambiguation and lexicon mapping. We evaluated in an independent test set in IWSLT travel domain spoken language corpus. Substantial improvements were achieved for GIZA alignment crossover: we obtained a 45% decrease in crossovers compared to a traditional phrase-based statistical MT system. Furthermore, the BLEU score improved by 2 points. Finally, a framework of the translation game system is described, and the feasibility of integrating the components to produce reference translation and to automatically assess student's translation is verified.
by Yushi Xu.
S.M.

Els aspectes generals de qualitat de la traducció automàtica (TA), com l’adequació i la fluïdesa, s’han estudiat àmpliament, però els aspectes més detallats, com la qualitat de la traducció de la terminologia, s’han subestimat, especialment en el context dels estudis de traducció. L’objectiu d’aquest estudi és analitzar els tipus i freqüències d’errors terminològics en la traducció automàtica estadística (TAE) i la traducció automàtica neuronal (TAN) personalitzades amb l’objectiu final de comprendre com el tipus de sistema de TA, el tipus de corpus i la mida del corpus afecten la qualitat de la traducció de terminologia. Un corpus paral·lel turc-anglès obtingut a partir de resums de revistes de cardiologia va ser creat des de zero per entrenar motors de TAE i TAN de dominis específics. Després, aquest corpus es va combinar amb un corpus de domini mixt i es van entrenar dos motors més. Després de realitzar una avaluació automàtica i una avaluació humana en aquests 4 motors, els errors de terminologia es van anotar segons una tipologia d’errors de terminologia personalitzada. S’ha trobat que els tipus i freqüències dels errors terminològics són significativament diferents en els sistemes TAE i TAN, i que els canvis en la mida i tipus de corpus han tingut un impacte més dràstic en el TAN en comparació amb la TAE. Una contribució clau de la dissertació a la investigació sobre TA és la tipologia d’error de terminologia independent del llenguatge per avaluar les fortaleses i debilitats relatives de diferents sistemes de TA en termes de terminologia. A més, la troballa que els sistemes TAN exhibeixen diferents tipus d’errors de terme amb diferents freqüències implica que les guies de postedició concebudes específicament per a sistemes TAE podrien requerir canvis per tal d’adaptar-se al nou patró de comportament de TAN.
Los aspectos generales de calidad de la traducción automática (TA), como la adecuación y la fluidez, se han estudiado ampliamente, pero los aspectos más detallados, como la calidad de la traducción de la terminología, se han subestimado, especialmente en el contexto de los estudios de traducción. El objetivo de este estudio es analizar los tipos y frecuencias de errores terminológicos en la traducción automática estadística (TAE) y la traducción automática neuronal (TAN) con el objetivo final de comprender cómo el tipo de sistema de TA, el tipo de corpus y el tamaño del corpus afectan la calidad de la traducción de terminología. Un corpus paralelo turco-inglés obtenido a partir de resúmenes de revistas de cardiología se creó desde cero para entrenar motores de TAE y TAN de dominios específicos. Luego, este corpus se combina con un corpus de dominio mixto y se entrenaron dos motores más. Después de realizar una evaluación automática y una evaluación humana en estos 4 motores, los errores de terminología se anotaron según una tipología de errores de terminología personalizada. Se ha encontrado que los tipos y frecuencias de los errores terminológicos son significativamente diferentes en los sistemas TAE y TAN, y que los cambios en el tamaño y tipo de corpus tienen un impacto más drástico en el TAN en comparación con el TAE. Una contribución clave de la disertación es la tipología de error de terminología que se puede utilizar para evaluar las fortalezas y debilidades relativas de diferentes sistemas de TA en términos de terminología. Además, el hallazgo de que los sistemas TAN exhiben diferentes tipos de errores en los términos con diferentes frecuencias implica que las pautas de posedición que se prepararon para los textos resultantes de TAE deben actualizarse para adaptarse al nuevo patrón de comportamiento de TAN.
General quality aspects of machine translation (MT) such as adequacy and fluency are studied extensively, more fine-grained aspects such as the terminology translation quality have not received much attention especially in the context of translation studies. The objective of this study is to analyze the types and frequencies of terminology errors in custom statistical machine translation (SMT) and neural machine translation (NMT) with the goal of understanding how MT system type, corpus type and corpus size affect the terminology translation quality. A Turkish – English parallel corpus obtained from cardiology journal abstracts was built from scratch for training domain-specific SMT and NMT engines. Then, this domain-specific corpus is combined with a mixed domain corpus and two more engines were trained. After conducting automatic evaluation and human evaluation on these 4 engines, terminology errors were annotated based on a custom terminology error typology. It was found that the types and frequencies of terminology errors are significantly different in SMT and NMT systems, and that changes in corpus size and corpus type had more drastic impact on NMT compared to SMT. A key contribution of the dissertation to the MT research is the crafted language-agnostic terminology error typology which can be used for evaluating the relative strengths and weakness of different MT systems in terms of terminology. Besides, the finding that NMT systems exhibit different types of term errors with different frequencies implies that postediting guidelines conceived specifically for SMT systems could require changes to accommodate the behavior pattern of NMT.
Universitat Autònoma de Barcelona. Programa de Doctorat en Traducció i Estudis Interculturals

This thesis addresses a critical problem in computer vision of dealing with dataset bias between source and target environments. Variations in image data can arise from multiple factors including contrasts in picture quality (shading, brightness, colour, resolution, and occlusion), diverse backgrounds, distinct circumstances, changes in camera viewpoint, and implicit heterogeneity of the samples themselves. This research developed strategies to address this domain shift problem for the object recognition task. Several domain adaptation and generalization approaches based on deep neural networks were introduced to improve poor performance due to domain shift or domain bias.

In this thesis I have combined cognitive metaphor theory, based mostly on the theories of the book Metaphors we live by by Lakoff & Johnson, with translation theory in order to try and show how the former can contribute with useful analytical tools within me- taphor translation. More specifically, my objective is to try and show how the knowled- ge of how metaphor works according to the cognitive perspective can help translating metaphors in a way that corresponds to the recommendations established by Peter Newmark (1992), i.e. try to maintain, in as much as posible, both grammatical structure and semantics. Through a qualitative analysis I considered gramatical structure and se- mantics of three original literary metaphors taken from the swedish book ”Gösta Ber- lings saga” by Selma Lagerlöf, and their translations to spanish made by Slaby (1955). I identified how well the translations corresponded to the recommendations of Newmark, as well as suggested new translations based on the results of the cognitive analysis. As a result we could see the importance, not only of maintaining both source domain and tar- get domain of the original metaphor, or at least the epistemical correspondences in case the source domain changed, but also the ontological correspondences, the specific con- tact, that the original metaphor establishes between these two domains, for a translation that better follows the norms defined by Newmark. In other words, limited to these th- ree examples, I managed to confirm that a cognitive approach to metaphor can be very useful within translation.

Popular science articles are nowadays a key component of the scientific writing landscape: science is popularised through a recontextualization of a primary scientific discourse to fit the knowledge and expectations of a broader audience, but then disseminated further through translation. This recontextualization is often achieved using metaphors to help the non-expert reader to access complex and abstract scientific concepts based on shared author-reader experiences, which are, however, not necessarily shared with the new target-culture audience, potentially endangering cross-linguistic communication of the scientific content. This thesis aims to investigate metaphors in American popular science articles dealing with astronomy and astrophysics published in Scientific American and their Arabic translations published in Majallat-Al-Oloom. The thesis focuses more particularly on metaphors fulfilling a pedagogical role which are embedded in culture specific domains. Although English is argued to be the global lingual franca of the sciences nowadays, the use of culture specific metaphors might raise difficulties in disseminating the scientific content in English and in its translations. The field of astronomy and astrophysics has been chosen because of the complex and often abstract nature of its concepts that requires an appropriate discourse strategy to bring abstract concepts closer to the general reader’s understanding. It is also a domain that is visible in the public understanding of the sciences through its large diffusion. In this scientific communication metaphor fulfils not only a terminological function but is also used as a pedagogical tool to achieve popularisation. Despite its role in disseminating scientific content, metaphor in the discourse of astronomy and astrophysics has remained so far unexplored from both metaphor studies and translation perspectives. To achieve these aims, a multidimensional framework combining a conceptual approach with linguistic and functional elements was devised to capture the complexity of metaphor from a translation perspective, especially between languages of differing diffusion where English is a global lingua franca. A bilingual corpus was compiled (circa 150,000 words) and analysed quantitatively and qualitatively. The updated version of the metaphor identification procedure (MIPVU) was further adapted to allow the identification of the linguistic metaphors and their functions. The methodology also accounted for how the conceptual metaphors are implied from the linguistic data, a step that is often unaccounted for in the literature. The study shows that linguistic metaphors used in the source texts fulfil mainly a pedagogical function and are often embedded in culture-specific domains, presenting challenges for translation. A wide range of strategies was identified in the translation of these metaphors, where the same conceptual metaphor is often reproduced in the target text by combining many strategies (couplets). This results in new metaphors in the target system that are argued to achieve a dual purpose: they facilitate access to scientific concepts communicated in the source text by unpacking the metaphorical images for a new audience; and they contribute to the enrichment of the target-language system.

The aim of this dissertation is the creation of a terminology project of 50 entries in Slovak language. The domain of the project is green cars. The terminology resource was created as part of a project with the Terminology Coordination unit of the European Parliament (TermCoord), which is responsible for the managing of the IATE (InterActive Terminology for Europe) terminology database. The final result of the terminology project was presented to the TermCoord to be implemented into IATE. Apart from creating the resource, the results of the work conducted for this dissertation has lead to a deep analysis of the IATE database, putting the focus on the managing strategy of the resource and highlighting strong points and criticisms for the users of the database. The dissertation is composed of four chapters. The first chapter is dedicated to a theoretical overview of special languages and terminology, with a brief section dedicated to the building of corpora for terminology purposes. The second chapter presents the translation and terminology services at the EU institutions, which it was possible to analyse during the internship at the DGTranslation of the European Commission in Brussels. In the third chapter, the process for the extraction of the Slovak terms to input into the termbase is described, starting from the retrievement of terms in English and then analysisng the process to extract the Slovak equivalent to build the termbase. The fourth chapter analyses the results of the terminology research and illustrates the structure of the final monolingual termbase.

This essay deals with the translation of certain metaphors in a journalistic political essay. It focuses on metaphor as a conceptual and rhetorical device, and on the translation of metaphors that make use of the source domains NATURAL FORCE and ILLNESS. The aim of the essay is to investigate to what extent the two source domains are transferred to the target text in translation, and to show how the fact that metaphors are both conceptual and rhetorical is reason for why they should be transferred when possible. The study is quantitative in that it looks at frequency of source domain transfer into the target text and at the frequency of a few metaphor translation methods, and it is qualitative in that it explains how the metaphors are both conceptual and rhetorical choices on behalf of the author, and in that it explains and analyses the processes leading to certain metaphor translation strategies. The study finds that four different methods are used when translating the metaphors in questions and that it is by far most common to directly translate them. The second most common strategy allows for the source domain to be transferred to the target text but for the actual lexemes or phrases belonging to it to be exchanged for other lexemes or phrases within the same source domain.

Nous avons observé depuis plusieurs années l'émergence des approches statistiques pour la traduction automatique. Cependant, l'efficacité des modèles construits est soumise aux variabilités inhérentes au langage naturel. Des études ont montré la présence de vocabulaires spécifique et général composant les corpus de textes de domaines spécialisés. Cette particularité peut être prise en charge par des ressources terminologiques comme les lexiques bilingues.Toutefois, nous pensons que si le vocabulaire est différent entre des textes spécialisés ou génériques, le contenu sémantique et la structure syntaxique peuvent aussi varier. Dans nos travaux,nous considérons la tâche d'adaptation aux domaines spécialisés pour la traduction automatique statistique selon deux axes majeurs : l'acquisition de lexiques bilingues et l'édition a posteriori de traductions issues de systèmes automatiques. Nous évaluons l'efficacité des approches proposées dans un contexte spécialisé : le domaine médical. Nos résultats sont comparés aux travaux précédents concernant cette tâche. De manière générale, la qualité des traductions issues de systèmes automatiques pour le domaine médical est améliorée par nos propositions. Des évaluations en oracle tendent à montrer qu'il existe une marge de progression importante

This essay deals with the translation methods used for a German-Swedish translation of metaphors in two books about climate change and environmental policy. These books contain a large number of metaphors, and this essay raises the question of how those are translated. The theory used for defining and classifying the metaphors found in the material is the conceptual metaphor theory, as defined by Lakoff and Johnson (1980), and their notion of metaphor is also fundamental in how it is later being analysed. The study finds that of the four different  translation strategies used, direct translation is by far the most common one. It also shows that a few metaphors were altered in terms of the lexicalisation of the metaphorical expression, however with the underlying conceptual metaphor still intact. Only one metaphor was translated with a completely different conceptual metaphor than in the original, and in two other cases the metaphor was translated with non-figurative language.

Le développent technologique des salles d’opérations a accéléré de manière spectaculaire ces dernières années. Cependant, la capacité des praticiens à différencier les tissus sains des tissus malsains à travers le champ chirurgical est principalement basée sur leur propre perception et expérience. Ceci est pourtant d’une importance majeure en chirurgie oncologique, tant pour la résection de tumeurs que pour les actes de reconstruction. C’est pourquoi la capacité d’évaluer le statut des tissus biologiques à travers des zones étendues en temps réel est cruciale. Le manque d’outils permettant l’évaluation de la viabilité des tissus biologique dans un contexte intra opératoire a été la motivation principale de ce projet. Un prototype d’imageur multimodal clinique a été développé pour l’imagerie d’oxygénation et de fluorescence en temps-réel. La capacité de la plateforme à quantifier l’ischémie a été démontrée lors de tests précliniques, par comparaison avec les méthodes standards. Le caractère multimodal de la plateforme d’imagerie a été exploité pour combiner l’imagerie endogène mesurant les propriétés optiques des tissus et l’imagerie exogène par fluorescence, dans le cadre de la chirurgie du cancer. Une méthode de quantification a été employée lors d’essais précliniques sur des modèles de cancers colorectaux et pancréatiques, mettant en évidence les défaillances de l’imagerie de fluorescence conventionnelle
The deployment of technology in operating rooms dramatically accelerated over the last decades. More precisely, the surgeons’ ability to distinguish healthy from diseased tissues is still mostly based on their own subjective perception. As tissue status assessment is of upmost importance in oncologic surgery, both for tumor resection and reconstruction procedures, the ability to assess the tissues intraoperatively and in real-time over a large field is crucial for surgical act guidance. The lack of tools for biological intraoperative tissue status assessment has been the main source of motivation for this thesis work. A clinically-compatible imaging platform has been developed for oxygenation and fluorescence imaging in real-time. The capability of the platform to detect and quantify ischemia has been demonstrated through preclinical trials, by comparison with standard of care methods. Furthermore, the multimodal nature of the developed imaging device has been exploited by combining endogenous imaging of optical properties with exogenous fluorescence imaging, in the context of oncologic surgery. A fluorescence quantification technique was validated in preclinical trials with colorectal and pancreatic cancer models, highlighting the limitations of conventional fluorescence imaging

In the recent years, remote sensing has faced a huge evolution. The constantly growing availability of remote sensing data has opened up new opportunities and laid the foundations for many new challenges. The continuous space missions and new constellations of satellites allow in fact more and more frequent acquisitions, at increasingly higher spatial resolutions, and at an almost total coverage of the globe. The availability of such an huge amount data has highlighted the need for automatic techniques capable of processing the data and exploiting all the available information. Meanwhile, the almost unlimited potential of machine learning has changed the world we live in. Artificial neural Networks have break trough everyday life, with applications that include computer vision, speech processing, autonomous driving but which are also the basis of commonly used tools such as online search engines. However, the vast majority of such models are of the supervised type and therefore their applicability rely on the availability of an enormous quantity of labeled data available to train the models themselves. Unfortunately, this is not the case with remote sensing, in which the enormous amounts of data are opposed to the almost total absence of ground truth. The purpose of this thesis is to find the way to exploit the most recent deep learning techniques, defining a common thread between two worlds, those of remote sensing and deep learning, which is often missing. In particular, this thesis proposes three novel contributions which face current issues in remote sensing. The first one is related to multisensor image registration and combines generative adversarial networks and non-linear optimization of crosscorrelation-like functionals to deal with the complexity of the setting. The proposed method was proved able to outperform state of the art approaches. The second novel contribution faces one of the main issues in deep learning for remote sensing: the scarcity of ground truth data for semantic segmentation. The proposed solution combines convolutional neural networks and probabilistic graphical models, two very active areas in machine learning for remote sensing, and approximate a fully connected conditional random field. The proposed method is capable of filling part of the gap which separate a densely trained model from a weakly trained one. Then, the third approach is aimed at the classification of high resolution satellite images for climate change purposes. It consist of a specific formulation of an energy minimization which allows to fuse multisensor information and the application a markov random field in a fast and efficient way for global scale applications. The results obtained in this thesis shows how deep learning methods based on artificial neural networks can be combined with statistical analysis to overcome their limitations, going beyond the classic benchmark environments and addressing practical, real and large-scale application cases.

Submitted by Ricardo Cedraz Duque Moliterno (ricardo.moliterno@uefs.br) on 2018-07-31T22:54:08Z No. of bitstreams: 1 O SERT?O TRADUZIDO ESTUDO DOS MARCADORES CULTURAIS DO DOM?NIO ECOL?GICO NA TRADU??O DE OS SERT?ES PARA A L?NGUA ESPANHOLA GEOVANIO SILVA DO NASIMENTO.pdf: 14426323 bytes, checksum: aa31d92272372d79c0c599f15ca2738c (MD5)
Made available in DSpace on 2018-07-31T22:54:08Z (GMT). No. of bitstreams: 1 O SERT?O TRADUZIDO ESTUDO DOS MARCADORES CULTURAIS DO DOM?NIO ECOL?GICO NA TRADU??O DE OS SERT?ES PARA A L?NGUA ESPANHOLA GEOVANIO SILVA DO NASIMENTO.pdf: 14426323 bytes, checksum: aa31d92272372d79c0c599f15ca2738c (MD5) Previous issue date: 2018-03-08
Os Sert?es, from Euclides da Cunha, enjoys wide recognition by part of the literary criticism and it is considered one of the most representative books of Brazilian literature. That occurs, mostly, due to the its subject matter, the War of Canudos, and the way the author has built his narrative, allowing the book to be studied through various fields of knowledge. The prestige of ?Os Sert?es? reveals itself in the interest it has aroused in readers outside Brazil, resulting the translation to various languages, among them Spanish, which was the first foreign language to which the book was translated. This dissertation aimed at performing a study on the Cultural Markers present in Da Cunha?s book Os Sert?es (1984[1902]) as well as investigating how those CMTs were translated to Spanish (SANTOS, 1980). The research starts with a discussion on the writing process of the book, highlighting the contacts and the research sources of the author, setting up his works and his critical fortune, such as Santana (1995), Valente (2007; 1996), Galv?o (1977; 1981; 1984), Medeiros (2009), among others. Then, it was discussed the vocabulary of Euclides Os Sert?es, starting from some studies already done (GALV?O, 1985; SANTOS, 1980; ZACHARIAS, 2001). The theoretical- methodological cutout was done from notions, definitions and concepts of Corpus Linguistics and Translation (BAKER, 1993; SARDINHA, 2002; 2004; CAMARGO 2007; 2008), Cultural Domains, Cultural Markers (NIDA, 1945; AUBERT, 2006a; 2006b; 2008; CAMARGO, 2008; e REICHMAN; ZAVAGLIA, 2014) and Translation Modalities (AUBERT, 1998; 2006a; 2006b). Based on the precepts of Corpus Linguistics, it was compiled a paralleled corpus formed by the Portuguese version of Os Sert?es and its translation to Spanish in order to search for the CMTs candidates and their respective translation components. For the corpus treatment and analysis, it was used, in this investigative work, some computer softwares, mainly the WordSmith Tools 7.0, their tools and utilities like Concord, Keyword, Wordlist, Text convert and Aligner. The results were the compilation of 95 Cultural Markers from the Portuguese version of the book, their Spanish translation correspondents and the identification of the Translation Modalities. Those data were organized in lexicographic files, created to this end, and which present definition and subscriptions. Those data served as a parameter for the analysis, which turned out to be a trend in the use of one single type of Translation Modality for the Cultural Markers with sporadic possibilities of variation, causing the use of only one variable in the translation of those lexical unities, in general, with a little possibility of variation.
Os Sert?es, de Euclides da Cunha, disfruta de amplo reconhecimento por parte da cr?tica e ? considerado um dos livros mais representativos da literatura brasileira. Isso ocorre, principalmente, por conta do tema tratado na obra, a Guerra de Canudos, e o modo como o autor construiu a sua narrativa, permitindo que o livro possa ser estudado por v?rias ?reas do conhecimento. O prest?gio d?Os Sert?es revela-se tamb?m no interesse que despertou em leitores fora do Brasil, resultando na tradu??o para diversos idiomas, dentre eles, o espanhol, que foi a primeira l?ngua estrangeira para a qual foi traduzido. O trabalho em quest?o teve como objetivo principal realizar um estudo dos Marcadores Culturais (MCs) no livro Os Sert?es de Euclides da Cunha (1984 [1902]) e investigar como esses MCs foram traduzidos para a l?ngua espanhola (SANTOS, 1980). Iniciamos a pesquisa com uma discuss?o sobre a escrita da obra, destacando os contatos e as fontes de pesquisa do autor, situando os trabalhos de sua fortuna cr?tica, tais como: Santana (1995), Valente (2007; 1996), Galv?o (1977; 1981; 1984), Medeiros (2009) dentre outros. Em seguida, tra?amos uma discuss?o sobre o vocabul?rio de Euclides da Cunha e d?Os Sert?es, partindo de alguns estudos j? feitos (GALV?O, 1985; SANTOS, 1980; ZACHARIAS, 2001). O recorte te?rico-metodol?gico foi feito a partir das no??es, defini??es e conceitos de Corpus Lingu?stico e Tradu??o (BAKER, 1993; SARDINHA, 2002; 2004; CAMARGO 2007; 2008), Dom?nios Culturais, Marcas e Marcadores Culturais (NIDA, 1945; AUBERT, 2006a; 2006b; 2008; CAMARGO, 2008; e REICHMAN; ZAVAGLIA, 2014) e Modalidades de Tradu??o (AUBERT, 1998; 2006a; 2006b). Pautado nos preceitos da Lingu?stica de Corpus, compilou-se um corpus paralelo formado pela obra Os Sert?es em portugu?s e sua tradu??o para o espanhol para buscar os candidatos a MCs e seus respectivos correspondentes de tradu??o. Al?m do corpus paralelo, foi elaborado um corpus de refer?ncia para verificar a chavicidade destes candidatos. Para an?lise e tratamento do corpus, utilizou-se, nesta investiga??o, alguns programas computacionais, principalmente, o WordSmith Tools 7.0, suas ferramentas e seus utilit?rios como o Concord, Keyword, Wordlist, Text convert e Aligner. O resultado foi a compila??o de 95 Marcadores Culturais da obra em portugu?s, seus correspondentes de tradu??o em espanhol e a identifica??o das Modalidades de Tradu??o. Estes dados foram organizados em fichas lexicogr?ficas, criadas para este fim, as quais apresentam tamb?m as defini??es e as abona??es. Esses dados serviram de par?metro para a an?lise, que demonstrou uma tend?ncia no uso de um ?nico tipo de Modalidade de Tradu??o para os Marcadores Culturais com possibilidades espor?dicas de varia??o. Ocasionando o uso de apenas uma vari?vel na tradu??o destas unidades l?xicas, de uma maneira geral, com pouca possibilidade de varia??o.

Nous nous sommes donné pour objectif d'étudier les questions posées par la traduction littéraire en B à partir du coréen effectuée en binôme par un traducteur coréen et un cotraducteur spécialiste de la langue d'arrivée, et d'en préciser les conditions de réussite.En matière littéraire du coréen vers le français, ce couplage vise à pallier l'absence de traducteurs français capables de travailler seuls.L'opération traduisante étant effectuée par des traducteurs coréens travaillant en B, l'exigence formelle de la traduction littéraire implique l'intervention d'un réviseur français.De plus en plus d'œuvres coréennes sont traduites selon cette modalité et publiées en France. Nous avons analysé les différentes étapes du processus, de la compréhension à la réexpression, à la lumière e la théorie interprétative de la traduction en tentant de montrer pourquoi le concept clé de la déverbalisation légitime la cotraduction, l'activité traduisante étant une opération de texte à texte, non de langue à langue. La première partie de notre recherche est consacrée à l'étude du cadre théorique (définition des notions auxquelles nous recourons, description de la théorie interprétative) et pratique (aperçu de l'état présent de la traduction française des textes littéraires coréens). Dans la deuxième partie, nous avons analysé des exemples de traduction en montrant le cheminement des échanges entre le traducteur et le cotraducteur, et en précisant la nature du guidage donné par le premier au second ainsi que la nature de la contribution de ce dernier. Nous aboutissons à la conclusion qu'une pratique souvent considérée seulement comme un pis-aller peut prétendre à produire des textes d'une réelle qualité littéraire dès lors que la technique mise en œuvre respecte les conditions que nous tentons de définir ici
Our aim was to study the issue of literary translation when it is carried into the B language of the translator (here from Korean to French) with the assistance of a co-translator native speaker of the target language and to point out the conditions of success. In the field of literary translation from Korean into French, this kind of team-work aims at overcoming the lack of French translators able to produce literary standard translations by themselves. When the translation is done by a Korean translator working into his B language, a thorough work of editing and rewriting is required due to the formal requirements of literary writing. An increasing number of Korean novels and short stories are translated by such a dual team and published in France.We analyze the different steps of the process of translation, from understanding to reformulation in the light of the Interpretive Theory of Translation, trying to show why the key concept of “de-verbalization” makes co-translation justifiable, translation being defined as an operation from text to text not from language to language.The first part of our research is devoted to the theoretical aspects of our study (definition of some key notions, mainly of the Interpretive Theory of Translation) and its practical aspects (general survey of translated Korean literary works into French). In the second part, we analyze a large number of our translation samples with the objective of showing the sort of dialog the translator and her co-translator entertain, and underlining the nature of guidance proposed by the first one to second one and the contribution of the last one.Our conclusion is that, thanks to this dual-team process, often wrongly considered as a lesser evil, we are able to produce quality literary translation provided that the method implemented takes into account the conditions we describe here

We expand the scope of image-to-image translation to include more distinct image domains, where the image sets have analogous structures, but may not share object types between them. Semantic-Driven Unsupervised Image-to-Image Translation for Distinct Image Domains (SUNIT) is built to more successfully translate images in this setting, where content from one domain is not found in the other. Our method trains an image translation model by learning encodings for semantic segmentations of images. These segmentations are translated between image domains to learn meaningful mappings between the structures in the two domains. The translated segmentations are then used as the basis for image generation. Beginning image generation with encoded segmentation information helps maintain the original structure of the image. We qualitatively and quantitatively show that SUNIT improves image translation outcomes, especially for image translation tasks where the image domains are very distinct.

The biosynthesis of proteins occurs at the ribosomes, where amino acids are linked together into linear chains. Nascent protein chains may undergo several different processes during their synthesis. Some proteins begin to fold, while others interact with chaperones, targeting factors or processing enzymes. Nascent membrane proteins are targeted to the cell membrane for integration, which involves the translocation of periplasmic domains and the insertion of membrane-embedded parts. The aim of this thesis was to gain insights about the dynamics of nascent peptide chains undergoing folding, membrane translocation and integration. To this end, we explored the use of arrest peptides (APs) as force sensors. APs stall ribosomes when translated unless there is tension in the nascent peptide chain: the higher the tension, the more full-length protein can be detected. By using APs, we could show that a transmembrane helix is strongly ‘pulled’ twice on its way into the membrane and that strong electric forces act on negatively charged peptide segments translocating through the membrane. Furthermore, we discovered that APs could be used to detect protein folding and made the surprising discovery that a small protein domain folded well inside the ribosomal tunnel. Finally, we explored the arrest-stability of a large set of AP variants and found two extremely stable APs.

Cytoplasmic polyadenylation-element binding protein (CPEB) is required for the translational regulation in multiple cell types. CPEB is known to play important roles in early germ cell development, in neuronal synaptic plasticity, and in the process of cellular senescence. CPEB is able to control translation by first interacting with a specific sequence of mRNA known as the CPE site. CPEB recognizes a specific sequence of mRNA, called the cytoplasmic polyadenylation element. This is a uracil rich sequence that is located on the 3' UTR of mRNA. Once CPEB is bound to the CPE site, CPEB can interact with other proteins. CPEB is most notably known for interacting with a cleavage and polyadenylation specificity factor (CPSF), with a poly(A)-specific ribonuclease, and with a poly(A) polymerase in the Gld2 family. This complex of proteins controls polyadenylation on the 3' end of mRNA. By controlling the lengthening of the poly(A) tail, translation can be regulated. CPEB is believed to contain two RNA recognition motifs and a zinc binding region on the N-terminus. The zinc binding region contains six cysteine and two histidine residues that bind to two zinc atoms in a tetrahedral geometry. Using NMR spectroscopy, the structure of zinc binding region of CPEB1 was determined. This protein was shown to bind to two zinc ions in a cross-braced topology. The zinc binding region of CPEB was also determined that the correct classification for this zinc finger is a ZZ domain.

Several Saccharomyces cerevisiae mutant tRNA- ^Q2 species (glutamine isoacceptor, CUG anticodon) were synthesized and assayed for aminoacylation activity with Saccharomyces cerevisiae glutaminyl-tRNA synthetase. The derived steady state parameters were compared to similar datasets from the literature. The mutants behaved analogously to similar mutant species based on tRNA from Escherichia coli, but with slightly relaxed specificity as revealed by comparison of k_cat/K_M values relative to wild type in vitro transcribed tRNA. Additionally the eukaryotic N-terminal domain appendage, as found in Sce glutaminyl-tRNA synthetase, is considered in light of the discovery of non-canonical aminoacyl-tRNA synthetase functions, including its role in the assembly of the multiple aminoacyl-tRNA synthetase complex.

Proteins and RNAs are two of the most versatile macromolecules that carry out almost all functions within living organisms. In this thesis I have explored evolutionary and regulatory aspects of proteins and RNAs by studying their structures, modifications and interactions. In the first chapter of my thesis I investigate domain atrophy, a term I coined to describe large-scale deletions of core structural elements within protein domains. By looking into truncated domain boundaries across several domain families using Pfam, I was able to identify rare cases of domains that showed atrophy. Given that even point mutations can be deleterious, it is surprising that proteins can tolerate such large-scale deletions. Some of the structures of atrophied domains show novel protein-protein interaction interfaces that appear to compensate and stabilise their folds. Protein-protein interactions are largely influenced by the surface and charge complementarity, while RNA-RNA interactions are governed by base-pair complementarity; both interaction types are inherently different and these differences might be observed in their interaction networks. Based on this hypothesis I have explored the protein-protein, RNA-protein and the RNA-RNA interaction networks of yeast in the second chapter. By analysing the three networks I found no major differences in their network properties, which indicates an underlying uniformity in their interactomes despite their individual differences. In the third chapter I focus on RNA-protein interactions by investigating post-translational modifications (PTMs) in RNA-binding proteins (RBPs). By comparing occurrences of PTMs, I observe that RBPs significantly undergo more PTMs than non-RBPs. I also found that within RBPs, PTMs are more frequently targeted at regions that directly interact with RNA compared to regions that do not. Moreover disorderedness and amino acid composition were not observed to significantly influence the differential PTMs observed between RBPs and nonRBPs. The results point to a direct regulatory role of PTMs in RNA-protein interactions of RBPs. In the last chapter, I explore regulatory RNA-RNA interactions. Using differential expression data of mRNAs and lncRNAs from mouse models of hereditary hemochromatosis, I investigated competing regulatory interactions between mRNA, lncRNA and miRNA. A mutual interaction network was created from the predicted miRNA interaction sites on mRNAs and lncRNAs to identify regulatory RNAs in the disease. I also observed interesting relations between the sense-antisense mRNA-lncRNA pairs that indicate mutual regulation of expression levels through a yet unknown mechanism.

We use simultaneous observation of translational and rotational Brownian motion of domains in lipid membranes to test the hydrodynamics-based theory for the viscous drag on the membrane inclusion. We find that translational and rotational diffusion coefficients of micrometer-sized solid (gel-phase) domains in giant unilamellar vesicles showing fluid–gel phase coexistence are in excellent agreement with the theoretical predictions
Dieser Beitrag ist mit Zustimmung des Rechteinhabers aufgrund einer (DFG-geförderten) Allianz- bzw. Nationallizenz frei zugänglich

We use simultaneous observation of translational and rotational Brownian motion of domains in lipid membranes to test the hydrodynamics-based theory for the viscous drag on the membrane inclusion. We find that translational and rotational diffusion coefficients of micrometer-sized solid (gel-phase) domains in giant unilamellar vesicles showing fluid–gel phase coexistence are in excellent agreement with the theoretical predictions.
Dieser Beitrag ist mit Zustimmung des Rechteinhabers aufgrund einer (DFG-geförderten) Allianz- bzw. Nationallizenz frei zugänglich.

Executable and Translatable UML (xtUML) is the next abstraction level in software development, where both programming language and software architecture have been abstracted away. xtUML is a well defined UML profile, extended with precise action semantics. This allows the developers to define a problem area, domain, in such a detail that it can be executed. By defining the system with xtUML-models, domains, the system functionality can be verified early in the development process. Translation to code can be done in different ways and this work will be performed in an environment where code is automatically generated with a model compiler.

The goal with a domain is that it should be independent of other domains, reused without modification and exchanged with another domain solving the same problem. However a domain can make assumptions that certain functionality is available and these assumptions are requirements for another domain.

To fulfil these goals there must be a minimal coupling between the domains. This can be solved with the technique Implicit Bridging, where the bridge dependency between domains is defined in a bridge. The dependency is in the form of mappings/coupling between elements in both domains. By defining a bridge interface for a server domain a client domain can use the resources offered by the server domain.

The work performed shows how an implementation of Implicit Bridging could be realized by applying the technique in a microwave oven system. From the system design five different mapping types have been implemented. The applicability and the quality of the implementation have been verified by testing the generated system functionality and also verifying the goals, exchangeability and reuse of domains, of the system.

Setdb1 est une «histone» lysine méthyltransférase (KMT) appartenant à la famille SUV39, l’une des principales machineries épigénétiques de répression des gènes. Setdb1 établit notamment la mono-, la di- et la tri- méthylation sur la lysine 9 de l'histone H3 (H3K9).Setdb1 est essentiel pour la survie, la pluripotence et l'auto-renouvellement des cellules souches embryonnaires de souris (mESCs); son knock-out est mortel au stade de la péri-implantation à 3,5 dpc chez la souris. Setdb1 est également nécessaire pour la différenciation de nombreux types de cellules progénitrices : spermatogenèse, neurogenèse, différenciation des chondrocytes et différenciation des muscles squelettiques. De plus, Setdb1 a été associé à plusieurs maladies: il est amplifié dans le mélanome et le cancer du poumon et il est dérégulé dans les cancers du foie, de la prostate, colorectal et du sein, dans la maladie de Huntington et la schizophrénie. Remarquablement, au-delà des histones, Setdb1 méthyle nombreux substrats non-histones, y compris UBF, p53, Akt, Tat et Ing2.Bien que Setdb1 ait toujours été associé à son rôle nucléaire, il s'avère que Setdb1 est la seule KMT de la famille SUV39 à avoir également une localisation cytoplasmique, dans plusieurs types de cellules, y compris les mESCs, les fibroblastes embryonnaires de souris (MEFs) et les cellules HeLa. Cependant, la fonction de Setdb1 dans le cytoplasme reste totalement inconnue. Pour étudier le rôle cytoplasmique de Setdb1, nous avons utilisé des cellules souches embryonnaires de souris (mESCs), dans lesquelles Setdb1 est essentiel. Nos résultats montrent que Setdb1 cytoplasmique est crucial pour la survie des mESCs: en effet, le nombre de cellules apoptotiques augmente après la perte de Setdb1 cytoplasmique. Nous avons constaté que le Setdb1 cytoplasmique affecte la synthèse de protéines dans les mESCs. Nous montrons en outre que le Setdb1 cytoplasmique interagit avec la protéine Trim71 spécifique de mESC (également appelée Lin41) et avec le facteur de traduction d'initiation eIF3c dans les mESC. Enfin, nous avons démontré que Setdb1 et Trim71 co-régulent ensemble la stabilité et la traduction des mARNs. Nos données actuelles mettent au jour la fonction cytoplasmique essentielle d'une lysine méthyltransférase appelée Setdb1, au début considérée comme étant spécifique uniquement des histones et apportent de nouvelles informations sur la régulation post-transcriptionnelle de l'expression génique médiée par un régulateur épigénétique fondamental
Setdb1 is a “histone” lysine methyltransferase (KMT) belonging to the SUV39 family that methylates lysine 9 of histone H3 (H3K9), one of the major epigenetic machineries mainly involved in gene repression. Notably, Setdb1 establishes mono-, di- and tri-methylation of H3K9. Setdb1, or Eset in mice, is essential for the survival, the pluripotency and the self-renewal of mouse embryonic stem cells (mESCs); Eset knockout is lethal at the peri-implantation stage at 3.5 dpc in mice. Setdb1 is also required for the differentiation of many progenitor cell types: spermatogenesis, neurogenesis, chondrocyte differentiation and skeletal muscle differentiation. Moreover, Setdb1 has been associated with several diseases: it is amplified in melanoma and lung cancer and it is dysregulated in liver, prostate, colorectal and breast cancers, Huntington disease and schizophrenia.Remarkably, beyond histones, Setdb1 methylates many non-histone substrates, such as UBF, p53, AKT, Tat and ING2 proteins. Although Setdb1 has been always associated with its nuclear role, it turns out that Setdb1 is the only H3K9 KMT to have also a cytoplasmic localization, in several cell types, including mESCs, mouse embryonic fibroblasts (MEFs) and HeLa cells. However, the function of Setdb1 in the cytoplasm remains totally unknown. To investigate Setdb1 cytoplasmic role, we have used mouse embryonic stem cells (mESCs), in which Setdb1 is essential. Our results show that cytoplasmic Setdb1 is crucial for the survival of mESCs: indeed, the number of apoptotic cells increases after the loss of cytoplasmic Setdb1. We found that cytoplasmic Setdb1 affects newly protein synthesis in mESCs. We further show that cytoplasmic Setdb1 interacts with mESCs-specific protein Trim71 (also called Lin41) and with the initiation translation factor eIF3c in mESCs. Finally, we reported that Setdb1 and Trim71 together co-regulate mRNA stability and translation. Our current data unravel the essential cytoplasmic function of Setdb1, for long time considered exclusively an “histone” lysine methyltransferase, and provide new insights into the post-transcriptional regulation of gene expression mediated by a fundamental epigenetic regulator

The importance of the T-matrix method is well-known when frequency domain scattering problems are of interest. With the relatively recent and wide-spread interest in time domain scattering problems, similar applications of the T-matrix method are expected to be useful in the time domain. In this thesis, the time domain spherical scalar wave functions are introduced, translational addition theorems for the time domain spherical scalar wave functions necessary for the solution of multiple scattering problems are given, and the formulation of time domain scattering of scalar waves by two spheres and by two scatterers of arbitrary shape is presented. The whole analysis is performed in the time domain requiring no inverse Fourier integrals to be evaluated. Scattering examples are studied in order to check the numerical accuracy, and demonstrate the utility of the expressions.

La réponse immunitaire innée est la première ligne de défense contre les infections et les signaux de danger. Elle est déclenchée par l’activation de récepteurs via la reconnaissance de motifs conservés spécifiques des pathogènes appelés PAMP (Pathogen-Associated Molecular Patterns) et/ou de signaux de danger endogènes appelés DAMP (Damage-Associated Molecular Patterns). Parmi eux, NLRP3 (NACHT, leucine-rich repeat (LRR), and pyrin domain (PYD) -containing 3) assemble un complexe de signalisation appelé inflammasome, qui contrôle la sécrétion d’IL-1β, d’IL-18 et la mort cellulaire par pyroptose. La dérégulation de l’inflammasome NLRP3 est impliquée dans plusieurs maladies. Les modifications post-traductionnelles (PTM) sont essentielles à l’activation de cet inflammasome, en particulier la déubiquitination du domaine LRR de NLRP3 par BRCC3. Notre but est de mieux comprendre les mécanismes de régulation des PTM au niveau du domaine LRR de NLRP3. En utilisant la spectrométrie de masse, nous avons identifié 3 lysines ubiquitinées et 3 sérines phosphorylées. La sécrétion d’IL-1β et d’IL-18 est significativement diminuée dans les lignées cellulaires humaines et murines reconstituées exprimant les mutants de substitution des lysines ubiquitinées de NLRP3. Cela supporte que ces sites sont importants pour l’activation de l’inflammasome NLRP3. L’impact de ces ubiquitinations devra être confirmé en cellules primaires et in vivo. En parallèle, une des sérines phosphorylées est aussi importante pour l’assemblage de l’inflammasome dans les lignées cellulaires reconstituées ainsi que dans les BMDM issus des souris knock-in portant une substitution phosphomimétique de cette sérine. Concernant le mécanisme, la substitution S/D conduit à l’hyper-ubiquitination de NLRP3 et empêche son activation via l’altération du recrutement de BRCC3. Par conséquent, nous avons découvert un nouveau point de contrôle en amont du recrutement de BRCC3. De plus, nous avons identifié 4 kinases comme candidats pouvant cibler cette sérine et nous avons aussi mis en évidence par un screen siRNA des phosphatases régulant l’inflammasome NLRP3. Dans le futur, cibler cette phosphorylation pourrait être utilisé comme une nouvelle voie thérapeutique pour les traitements anti-inflammatoires
Innate immune response is the first line of defense against infections and damages and is triggered by dedicated receptors activated by pathogen-associated molecular patterns (PAMPs) and/or damage-associated molecular patterns (DAMPs). Among them, NACHT, leucine-rich repeat (LRR), and pyrin domain (PYD) -containing 3 (NLRP3) assembled the inflammasome signaling complex that controls the secretion of IL-1β and IL-18 and cell death by pyroptosis. Dysregulation of NLRP3 inflammasome is involved in many diseases. The post-translational modifications (PTMs) are essential for NLRP3 inflammasome activation, in particular deubiquitination of NLRP3 LRR by BRCC3. Our goal is to better understand the regulation mechanism of PTMs in NLRP3 LRR domain. Using mass spectrometry, we identified 3 ubiquitinated lysines and 3 phosphorylated serines. The IL-1β and IL-18 secretion were significantly decreased in reconstituted human and mouse cell lines expressing NLRP3 mutants bearing substitutions of the ubiquitinated lysines, supporting that these sites are important for NLRP3 inflammasome activation. The impact of these ubiquitinations will need to be confirmed in primary cells and in vivo. Similarly, one of the phosphorylated serines is critical for inflammasome assembly in reconstituted cell lines, as well as in primary bone marrow derived macrophages from in-house generated knock-in mice bearing a phosphomimetic substitution of this serine. Mechanistically, S/D substitution leads to NLRP3 hyper-ubiquitination and suppresses NLRP3 inflammasome activity via impairing BRCC3 recruitment. Therefore, we discovered an additional checkpoint upstream of BRCC3 recruitment. Besides, we identified 4 kinases as candidates to target this serine and identified by siRNA screen phosphatases regulating the NLRP3 inflammasome pathway. In the future, targeting this phosphorylation could be used as a novel therapeutical target for anti-inflammatory drugs

Cette thèse a pour but l'analyse du discours de l'essai poétique et des moyens qui permettent sa traduction du français à l'italien. Elle s'articule en trois phases: - étude de l'essai poétique dans une perspective diachronique (son histoire et sa constitution en genre littéraire) et synchronique (sa forme) ;- analyse de ce qui rend l'essai poétique, tant du point de vue de l'auteur que du point de vue du lecteur ;- examen de la notion de traduction pragmatique et de traduction littéraire telles qu'elles sont proposées par la Théorie Interprétative de la traduction et l'application de cette dernière aux traductions de deux essais de Predrag Matvejevitch. La démarche est moins prescriptive que descriptive : c'est la compréhension des mécanismes qui président à la construction et à la communication du sens qui est privilégiée. C'est de cette manière que l'on pourra percevoir les possibilités expressives du discours de l'essai et pénétrer un peu dans ce prodige qu'est la communication interlinguistique réalisée par le traducteur.

High quality reagents are preconditions for high performance protein analyses. But despite progress in some techniques, e.g. mass spectrometry, there is still a lack of affinity-based detection techniques with enhanced precision, specificity, and sensitivity. Building on the concept of multiple affinity recognition reactions and signal amplification, a proximity ligation assay (PLA) was developed as a molecular tool for analyzing proteins and their post-translational modification and interactions. PLA enhanced the analysis of protein expression levels and post-translational modifications in western blotting (Paper I), which had elevated sensitivity and specificity, and an ability to investigate protein phosphorylation. A general and straightforward method was established for the functionalization of affinity reagents through adding DNA strands to protein domains for protein analysis in medicine (Paper II). A method for protein domain-mediated conjugation was developed to simplify the use of recombinant affinity reagents, such as designed ankyrin repeat protein (DARPin), in DNA-mediated protein analyses. Alzheimer’s disease (AD) is characterized by progressive cognitive decline and memory impairment, and amyloid-beta plaques and neurofibrillary tangles (NFT) in the brain are clinical hallmarks of the disease. In order to understand the mechanisms underlying the formation of NFT, in situ PLA was used to explore the role of microtubule affinity related kinase 2 (MARK2) in phosphorylating tau protein during the pathological progress of AD (Paper III). The analyses of roles of MARK proteins 1-4 in phosphorylating tau protein in cells and in post-mortem human brains were performed in Paper IV. The focus of this thesis was the study of post-translational modifications and interactions of proteins in medicine. Procedures for high performance protein analysis in western blotting via proximity ligation were developed, and a functionalization method for recombinant affinity reagents in DNA-mediated protein analysis was established. These and other techniques were used to investigate the roles of tau-phosphorylating MARK family proteins in AD.

Dissipative spots are found in physical experiments of many branches of natural science. In this thesis we use three-component reaction-diffusion systems on two-dimensional domains in order to generate these patterns. Using a dynamical system approach we proceed with a Fourier analysis on a linearized reaction-diffusion system in order to provide the bifurcation conditions for a given homogeneous state. We validate our results and establish it's limitations through numerical experiments. We report very interesting behavior during these simulations, notably hysteresis and multi-stability. We will then turn our attention to the relatively unexplored phenomenon of rotating spots. Based on previous work done for spiral waves, we investigate the effect of translational symmetry-breaking on a rotating spot mainly through careful numerical analysis.

Cancer involves the dysregulated proliferation and growth of cells throughout the body. C-terminal binding proteins (CtBP) 1 and 2 are transcriptional co-regulators upregulated in several cancers, including breast, colorectal, and ovarian tumors. CtBPs drive oncogenic properties, including migration, invasion, proliferation, and survival, in part through repression of tumor suppressor genes. CtBPs encode an intrinsic dehydrogenase activity, utilizing intracellular NADH concentrations and the substrate 4-methylthio-2-oxobutyric acid (MTOB), to regulate the recruitment of transcriptional regulatory complexes. High levels of MTOB inhibit CtBP dehydrogenase function and induce cytotoxicity among cancer cells in a CtBP-dependent manner. While encouraging, a good therapeutic would utilize >100-fold lower concentrations. Therefore, we endeavored to design better CtBP-specific therapeutics. The best of these drugs, 3-Cl and 4-Cl HIPP, exhibit nanomolar enzymatic inhibition and micromolar cytotoxicity and showed that CtBP enzymatic function is subject to allosteric interactions. Additionally, the function of the substrate-binding domain has yet to be examined in context of CtBP’s oncogenic activity. To this end, we created several point mutations in the CtBP substrate-binding pocket and determined key residues for CtBP’s enzymatic activity. We found that a conserved tryptophan in the catalytic domain is imperative for function and unique to CtBPs among dehydrogenases. Knowledge of this and other residues allows the directed synthesis of drugs with increased potency and higher CtBP specificity. Early work interrogated the importance of these residues in cell migration. Taken together, this work addresses the utility of the CtBP substrate-binding domain as a target for cancer therapeutics.

Ce travail a porté sur les domaines PDZ, une famille de domaines globulaires reconnaissant des motifs de liaison aux PDZ (appelés PBM, pour ‘PDZ-Binding Motifs’) généralement situés à l'extrémité C-terminale de leurs protéines partenaires. Les réseaux domaines-motifs sont souvent modulés par des modifications post-traductionnelles réversibles (PTM). Nous avons utilisé des PBM synthétiques simulant différentes conditions: motifs sauvages de diverses longueurs, acétylés, phosphorylés ou portant des mutations ‘imitant’ les PTM. Ces peptides ont été utilisés pour des études d'interaction à l'aide du test ‘Hold-Up’, un test développé à l'origine dans notre laboratoire. Nous avons évalué l'impact de diverses modifications des complexes PBM/PDZ, qui conduisent à un changement global de leur capacité de liaison du PDZ. Ces résultats fournissent des informations quantitatives sur l'effet biologique que de telles modifications pourraient avoir dans le contexte des protéines entières
This thesis focuses on PDZ domains, a family of globular domains that bind to conserved PDZ-Binding Motifs (called henceforth PBMs) generally situated at the extreme C-terminus of their partner proteins. Domain-motif networks are often modulated by reversible post-translational modifications (PTMs). We used synthetized PBMs to reproduce different conditions, such as a wild-type, acetylation or phosphorylation, addition of extra exosites or residue mimication of PTM in the literature. These peptides were used for interaction studies using the holdup assay, an assay originally developed in our laboratory. We evaluated the impact of diverse modifications of the PBM/PDZ interactions, which led to a global change of the PDZ-binding capability. These results provided quantitative information on the biological effects that such modifications may have in the context of full-length proteins

Dans ce mémoire, nous présentons outre les résultats de recherche en vue d’un système de traduction automatique français–chinois, les apports théoriques à partir de la théorie SyGULAC et de la théorie micro-systémique avec ses calculs ainsi que les méthodologies élaborées tendant à une application sure et fiable dans le cadre de la traduction automatique. L’application porte sur des domaines de sécurité critique tels que l’aéronautique, la médecine, la sécurité civile. Tout d’abord un état de l’art du domaine de la traduction automatique, en Chine et en France, est utile pour commencer la lecture. Les faiblesses des systèmes actuels à travers des tests que nous réalisons prouvent l’intérêt de cette recherche. Nous donnons les raisons pour lesquelles nous avons choisi la théorie micro-systémique et la théorie SyGULAC. Nous expliquons ensuite les problématiques rencontrées au cours de notre recherche. L’ambigüité, obstacle majeur pour la compréhensibilité et la traductibilité d’un texte, se situe à tous les niveaux de la langue : syntaxique, morphologique, lexical, nominal ou encore verbal. L’identification des unités d’une phrase est aussi une étape préalable à la compréhension globale, que cela soit pour un être humain ou un système de traduction. Nous dressons un état des lieux de la divergence entre la langue française et la langue chinoise en vue de réaliser un système de traduction automatique. Nous essayons d’observer la structure aux niveaux verbal, nominal et lexical, de comprendre leurs liens et leurs interactions. Egalement nous définissons les obstacles sources d’entrave à la réalisation de cette recherche, avec un point de vue théorique mais aussi en étudiant notre corpus concret. Le formalisme pour lequel nous avons opté part d’une étude approfondie de la langue utilisée dans les protocoles de sécurité. Une langue ne se prête au traitement automatique que si elle est formalisée. De ce fait, nous avons procédé à l’analyse de plusieurs corpus bilingues français/chinois mais aussi monolingues émanant d’organismes de sécurité civile. Le but est de dégager les particularités linguistiques (lexicales, syntaxiques, …) qui caractérisent la langue de la sécurité en général et de recenser toutes les structures syntaxiques qu’utilise cette langue. Après avoir présenté la formalisation de notre système, nous montrons les processus de reconnaissance, de transfert et de génération
In this paper, in addition to our research results for a French-Chinese machine translation system, we present the theoretical contributions from the SyGULAC theory and from the micro-systemic theory with its calculations as well as the methodologies developed aimed at a secure and reliable application in the context of machine translation. The application covers critical safety areas such as aerospace, medicine and civil security.After presenting the state of the art in the field of machine translation in China and France, the reasons of the choice of the micro-systemic theory and SyGULAC theory are explained. Then, we explain the problems encountered during our research. The ambiguity, which is the major obstacle to the understandability and to the translatability of a text, is present at all language levels: syntactic, morphological, lexical, nominal and verbal. The identification of the units of a sentence is also a preliminary step for global understanding, whether for human beings or for a translation system. We present an inventory of the divergences between the french and the chinese language in order to achieve an machine translation system. We try to observe the verbal, nominal and vocabulary structure levels, in order to understand their interconnections and their interactions. We also define the obstacles to this research, with a theoretical point of view but also by studying our corpus.The chosen formalism starts from a thorough study of the language used in security protocols. A language is suitable for automatic processing only if this language is formalized. Therefore, An analysis of several French/Chinese bilingual corpora, but also monolingual, from civil security agencies, was conducted. The goal is to find out and present the linguistic characteristics (lexical, syntactic ...) which characterize the language of security in general, and to identify all the syntactic structures used by this language. After presenting the formalization of our system, we show the recognition, transfer and generation processes

Tout d'abord, nous présentons des méthodes permettant d'inférer un graphe à partir de signaux, afin de modéliser le support des données à classifier. Ensuite, des translations préservant les voisinages des sommets sont identifiées sur le graphe inféré. Enfin, ces translations sont utilisées pour déplacer un noyau convolutif sur le graphe, afin dedéfinir un réseau de neurones convolutif adapté aux données d'entrée.Nous avons illustré notre méthodologie sur une base de données d'images. Sans utiliser de connaissances sur les signaux, nous avons pu inférer un graphe proche d'une grille. Les translations sur ce graphe sont proches des translations Euclidiennes, ce qui nous a permis de définir un réseau de neurones convolutif très similaire à ce que l'on aurait pu obtenir en utilisant l'information que les signaux sont des images. Ce réseau, entraîné sur les données initiales, a dépassé lesperformances des méthodes de l'état de l'art de plus de 13 points, tout en étant simple et facilement améliorable.La méthode que nous avons introduite est une généralisation des réseaux de neurones convolutifs, car ceux-ci sont des cas particuliers de notre approche quand le graphe est une grille. Nos travaux ouvrent donc de nombreuses perspectives, car ils fournissent une méthode efficace pour construire des réseaux adaptés aux données
This manuscript sums up our work on extending convolutional neuralnetworks to irregular domains through graph inference. It consists of three main chapters, each giving the details of a part of a methodology allowing the definition of such networks to process signals evolving on graphs with unknown structures.First, graph inference from data is explored, in order to provide a graph modeling the support of the signals to classify. Second, translation operators that preserve neighborhood properties of the vertices are identified on the inferred graph. Third, these translations are used to shift a convolutional kernel on the graph in order to define a convolutional neural network that is adapted to the input data.We have illustrated our methodology on a dataset of images. While not using any particular knowledge on the signals, we have been able to infer a graph that is close to a grid. Translations on this graph resemble Euclidean translations. Therefore, this has allowed us to define an adapted convolutional neural network that is very close what one would obtain when using the information that signals are images. This network, trained on the initial data, has out performed state of the art methods by more than 13 points, while using a very simple and easily improvable architecture.The method we have introduced is a generalization of convolutional neural networks. As a matter of fact, they can be seen as aparticularization of our approach in the case where the graph is a grid. Our work thus opens the way to numerous perspectives, as it provides an efficient way to build networks that are adapted to the data

[ES] La fibrilación auricular es la arritmia cardíaca más común. Durante la fibrilación auricular, el sustrato auricular sufre una serie de cambios o remodelados a nivel eléctrico y estructural. La remodelación eléctrica se caracteriza por la alteración de una serie de canales iónicos, lo que cambia la morfología del potential de transmembrana conocido como potencial de acción. La remodelación estructural es un proceso complejo que involucra la interacción de varios procesos de señalización, interacción celular y cambios en la matriz extracelular. Durante la remodelación estructural, los fibroblastos que abundan en el tejido cardíaco, comienzan a diferenciarse en miofibroblastos que son los encargados de mantener la estructura de la matriz extracelular depositando colágeno. Además, la señalización paracrina de los miofibroblastos afecta a los canales iónicos de los miocitos circundantes. Se utilizaron modelos computacionales muy detallados a diferentes escalas para estudiar la remodelación estructural inducida a nivel celular y tisular. Se realizó una adaptación de un modelo de fibroblastos humanos a nivel celular para reproducir la electrofisiología de los miofibroblastos durante la fibrilación auricular. Además, se evaluó la exploración de la interacción del calcio en la electrofisiología de los miofibroblastos ajustando el canal de calcio a los datos experimentales. A nivel tisular, se estudió la infiltración de miofibroblastos para cuantificar el aumento de vulnerabilidad a una arritmia cardíaca. Los miofibroblastos cambian la dinámica de la reentrada. Una baja densidad de miofibroblastos permite la propagación a través del área fibrótica y crea puntos de salida de actividad focal y roturas de ondas dentro de esta área. Además, las composiciones de fibrosis juegan un papel clave en la alteración del patrón de propagación. La alteración del patrón de propagación afecta a los electrogramas recogidos en la superficie del tejido. La morfología del electrograma se alteró dependiendo de la disposición y composición del tejido fibrótico. Se combinaron modelos detallados de tejido cardíaco con modelos realistas de los catéteres de mapeo disponibles comercialmente para comprender las señales registradas clínicamente. Se generó un modelo de ruido a partir de señales clínicas para reproducir los artefactos de señal en el modelo. Se utilizaron electrogramas de modelos de dos dominios altamente detallados para entrenar un algoritmo de aprendizaje automático para caracterizar el sustrato fibrótico auricular. Las características que cuantifican la complejidad de las señales fueron extraídas para identificar la densidad fibrótica y la transmuralidad fibrótica. Posteriormente, se generaron mapas de fibrosis utilizando el registro del paciente como prueba de concepto. El mapa de fibrosis proporciona información sobre el sustrato fibrótico sin utilizar un valor único de corte de 0,5 milivoltios. Además, utilizando la medición del flujo de información como la entropía de transferencia combinada con gráficos dirigidos, en este estudio, se siguió la dirección de propagación del frente de onda. La transferencia de entropía con gráficos dirigidos proporciona información crucial durante la electrofisiología para comprender la dinámica de propagación de ondas durante la fibrilación auricular. En conclusión, esta tesis presenta un estudio in silico multiescala que proporciona información sobre los mediadores celulares responsables de la remodelación de la matriz extracelular y su electrofisiología. Además, proporciona una configuración realista para crear datos in silico que pueden ser usados para aplicaciones clínicas y servir de soporte al tratamiento de ablación.
[CA] La fibril·lació auricular és l'arrítmia cardíaca més freqüent, en la qual el substrat auricular patix una sèrie de remodelacions elèctriques i estructurals. La remodelació de tipus elèctric es caracteritza per l'alteració d'un conjunt de canals iònics que modifica la morfologia del voltatge transmembrana, conegut com a potencial d'acció. La remodelació estructural és un fenomen complex que implica la relació entre diversos processos de senyalització, interaccions cel·lulars i canvis en la matriu extracel·lular. Durant la remodelació estructural, els abundants fibroblasts presents en el teixit cardíac comencen a diferenciar-se en miofibroblasts, els quals s'encarreguen de mantenir l'estructura de la matriu extracel·lular dipositant-hi col·lagen. A més, la senyalització paracrina dels miofibroblasts amb els miòcits circumdants també afectarà els canals iònics. Es van utilitzar models computacionals molt detallats a diferents escales per estudiar la remodelació estructural induïda a nivell tissular i cel·lular. Es va fer una adaptació a nivell cel·lular d'un model de fibroblasts humans per reproduir-hi l'electrofisiologia dels miofibroblasts durant la fibril·lació auricular. A més, l'exploració de la interacció del calci amb l'electrofisiologia dels miofibroblasts va ser avaluada mitjançant l'adequació del canal de calci a les dades experimentals. A nivell tissular es va estudiar la infiltració de miofibroblasts per tal de quantificar l'augment de vulnerabilitat que això conferia per patir una arrítmia cardíaca. Els miofibroblasts canvien la dinàmica de la reentrada, i presentar-ne una baixa densitat permet la propagació a través de la zona fibròtica, tot creant punts de sortida d'activitat focal i trencaments d'ones dins d'aquesta àrea. A més, les composicions de fibrosi tenen un paper clau en l'alteració del patró de propagació, afectant els electrogrames recollits en la superfície del teixit. La morfologia dels electrogrames es va veure alterada en funció de la disposició i la composició del teixit fibròtic. Per comprendre els senyals clínicament registrats es van combinar models detallats de teixits cardíacs amb models realistes dels catèters de cartografia disponibles comercialment. Es va generar un model de soroll a partir de senyals clínics per reproduir-hi els artefactes de senyal. Es van utilitzar electrogrames de models de bidominis molt detallats per entrenar un algoritme d'aprenentatge automàtic destinat a caracteritzar el substrat fibròtic auricular. Les característiques que quantifiquen la complexitat dels senyals van ser extretes per identificar la densitat i transmuralitat fibròtica. Posteriorment, es van generar mapes de fibrosi mitjançant la gravació del pacient com a prova de concepte. El mapa de fibrosi proporciona informació sobre el substrat fibròtic sense utilitzar un sol valor de tensió de tall de 0,5 mV. A més, utilitzant la mesura del flux d'informació com l'entropia de transferència combinada amb gràfics dirigits, en aquest estudi es va fer un seguiment de la direcció de propagació de l'ona. L'entropia de transferència amb gràfics dirigits proporciona informació crucial durant l'electrofisiologia per entendre la dinàmica de propagació d'ones durant la fibril·lació auricular. En conclusió, aquesta tesi presenta un estudi multi-escala in silico que proporciona informació sobre els mediadors cel·lulars responsables de la remodelació de la matriu extracel·lular i la seva electrofisiologia. A més, proporciona una configuració realista per crear dades in silico que es poden traduir a aplicacions clíniques que puguen donar suport al tractament de l'ablació.
[EN] Atrial fibrillation is the most common cardiac arrhythmia. During atrial fibrillation, the atrial substrate undergoes a series of electrical and structural remodeling. The electrical remodeling is characterized by the alteration of specific ionic channels, which changes the morphology of the transmembrane voltage known as action potential. Structural remodeling is a complex process involving the interaction of several signalling pathways, cellular interaction, and changes in the extracellular matrix. During structural remodeling, fibroblasts, abundant in the cardiac tissue, start to differentiate into myofibroblasts, which are responsible for maintaining the extracellular matrix structure by depositing collagen. Additionally, myofibroblasts paracrine signalling with surrounding myocytes will also affect ionic channels. Highly detailed computational models at different scales were used to study the effect of structural remodeling induced at the cellular and tissue levels.At the cellular level, a human fibroblast model was adapted to reproduce the myofibroblast electrophsyiology during atrial fibrillation. Additionally, the calcium handling in myofibroblast electrophysiology was assessed by fitting calcium ion channel to experimental data. At the tissue level, myofibroblasts infiltration was studied to quantify the increase of vulnerability to cardiac arrhythmia. Myofibroblasts alter the dynamics of reentry. A low density of myofibroblasts allows the propagation through the fibrotic area and creates focal activity exit points and wave breaks inside this area. Moreover, fibrosis composition plays a key role in the alteration of the propagation pattern. The alteration of the propagation pattern affects the electrograms computed at the surface of the tissue. Electrogram morphology was altered depending on the arrangement and composition of the fibrotic tissue. Detailed cardiac tissue models were combined with realistic models of the commercially available mapping catheters to understand the clinically recorded signals. A noise model from clinical signals was generated to reproduce the signal artifacts in the model. Electrograms from highly detailed bidomain models were used to train a machine learning algorithm to characterize the atrial fibrotic substrate. Features that quantify the complexity of the signals were extracted to identify fibrotic density and fibrotic transmurality. Subsequently, fibrosis maps were generated using patient recordings as a proof of concept. Fibrosis map provides information about the fibrotic substrate without using a single cut-off voltage value of 0.5 mV. Furthermore, in this study, using information theory measurements such as transfer entropy combined with directed graphs, the wave propagation direction was tracked. Transfer entropy with directed graphs provides crucial information during electrophysiology to understand wave propagation dynamics during atrial fibrillation. In conclusion, this thesis presents a multiscale in silico study atrial fibrillation mechanisms providing insight into the cellular mediators responsible for the extracellular matrix remodeling and its electrophysiology. Additionally, it provides a realistic setup to create in silico data that can be translated to clinical applications that could support ablation treatment.
Sánchez Arciniegas, JP. (2021). A Multiscale in Silico Study to Characterize the Atrial Electrical Activity of Patients With Atrial Fibrillation. A Translational Study to Guide Ablation Therapy [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/171456
TESIS

Un des problèmes majeurs rencontrés en traduction automatique (et plus généralement en traitement automatique multi-domaine du langage naturel) est celui de la surgénération d'interprétations profondes des énoncés lors de l'analyse. Le système dont nous exposons la conception et l'implémentation, et que nous avons nomme syvac (système de validation de la cohérence), est destiné à réduire la surgénération d'interprétations sémantiques des énoncés générés par des analyseurs utilisés en traduction automatique, en éliminant celles qui sont incohérentes par rapport aux connaissances de base, générales et sur le domaine traité par le texte, dont on dispose. A cet effet, nous munissons les mots et les concepts d'un certain nombre de relations conceptuelles potentielles qui décrivent la faculté qu'ont les concepts mis en jeu à s'associer au sein d'une interprétation donnée. Nous montrons enfin qu'en vérifiant la cohérence de plusieurs interprétations d'un énoncé analysé, avec les informations disponibles sur le domaine (relations potentielles et quelques règles de raisonnements simples), syvac arrive à déterminer celles qui ont le plus de chances d'avoir un sens, sans obligatoirement donner une description fine de ce sens. L'implémentation est réalisée en nexpertobjet, ce qui permet une représentation objet des connaissances avec héritage multiple des propriétés, d'une part, et l'emploi d'un moteur d'inférences à base de règles, de l'autre