The Asian tiger mosquito, Aedes albopictus (Skuse), is a worldwide nuisance pest that is capable of vectoring several viruses of public health concern. This invasive mosquito has recently expanded its habitable range through its utilization of artificial breeding sites, often due to the activity of humans. These factors, combined with additional expansion due to global changes in climate, have led to invigorated efforts to mitigate the impact of Ae. albopictus. Because it is a diurnal species, standard mosquito control efforts utilizing spray trucks or planes to administer insecticides offer little control, as these methods are directed towards crepuscular species. Barrier spray applications, however, have been shown to achieve a significant reduction in local mosquito pressure while requiring less insecticide application. The design behind barrier sprays is to apply insecticide treatments only around areas of interest, instead of trying to eradicate the local population of mosquitoes. These studies were conducted to evaluate the efficacy of different pyrethroid barrier treatments against Ae. albopictus mosquitoes, and to examine the impact of the most effective treatment on local mosquito populations when applied to suburban residences. Three pyrethroids were examined in these studies: Demand® CS (lambda-cyhalothrin), Talstar® Professional (bifenthrin), and Suspend® Polyzone® (deltamethrin). The following factors affecting pyrethroid barrier treatments showed significant impacts on the knockdown and mortality rates of Ae. albopictus mosquitoes: the plant species, the label rate at which treatments were applied, the active ingredient used in the treatment applications, the time of exposure to the treated foliage, the presence/absence of a blood meal in the mosquito, and the time after treatment. Demand CS treatments showed the highest proportions of knockdown and mortality in adult female Ae. albopictus mosquitoes and did so for the longest amount of time, regardless of the length of the exposure time. Because the Demand CS formulation of lambda-cyhalothrin was shown to be the most effective treatment in the previous studies, it was applied as a barrier treatment to suburban residences in Roanoke, Virginia, in a field trial. Applications of Demand CS as a barrier spray were shown to significantly reduce mosquito catch numbers inside the treated barrier throughout the 8 week study, as compared to the control properties. The findings of these studies indicate that many factors, pertaining to both the insecticides used and to the environment in which they are applied, play a role in influencing the efficacy of a pyrethroid barrier treatment for the control of Aedes mosquitoes. Thus, it is important to gather relevant information before the application of a barrier spray treatment to design the most appropriate program for the situation. / Doctor of Philosophy / Mosquitoes in Virginia are capable of transmitting many different diseases to humans and livestock. Many different treatment options are available to protect humans from these populations of mosquitoes. Some of these options can be performed for a whole community, such as area-wide fogging or treatments from truck-mounted sprayers, while others are applied to properties individually, like mosquito misting systems or barrier sprays. Applying long-lasting insecticides to the edge of an area can help to protect the inside of the area from mosquitoes, and this is called barrier spraying. Barrier sprays, in particular, have become a popular choice for homeowners, and they are successful at limiting human exposure to local mosquito populations. The experiments conducted here looked at three different insecticides used in these treatments and compared them for their ability to inhibit and kill Asian tiger mosquito. Other factors that potentially influenced effectiveness were examined, such as the plant type, the length of time since the treatment was applied, and whether or not the mosquito had taken a blood meal. These experiments examined three commonly used pyrethroid insecticides to determine their efficacy against the local Asian tiger mosquito. Suspend Polyzone lasted a long time but did not produce sufficient mortality, Talstar Professional killed large numbers of mosquitoes, but for a short period, and Demand CS lasted a long time and showed a high mortality rate. These studies showed that the three different insecticides lasted on the plants for different amounts of time, but also that the insecticides needed different amounts of contact time to kill mosquitoes. This information is essential because understanding how a treatment loses effectiveness over time can help with deciding when retreatment is needed. It was also found that the different insecticides were affected by the plants that they were sprayed onto, meaning that certain insecticides did better when sprayed on specific plants. Results from these studies revealed that Asian tiger mosquitoes that had just fed on human blood were more likely to die from insecticide treatments. Thus, even if a mosquito should bite an infected person, barrier spray treatments of their property will lower the chances that the mosquito can spread the disease to other people.

A growing use for serverless computing is large parallel data processing applications that take advantage of its on-demand scalability. Because individual serverless compute nodes, which are called functions, run in isolated containers, a major challenge with this paradigm is transferring temporary computation data between functions. Previous works have performed inter-function communication using object storage, which is slow, or in-memory databases, which are expensive. We evaluate the use of direct network connections between functions to overcome these limitations. Although function containers block incoming connections, we are able to bypass this restriction using standard NAT traversal techniques. By using an external server, we implement TCP hole punching to establish direct TCP connections between functions. In addition, we develop a communications framework to manage NAT traversal and data flow for applications using direct network connections. We evaluate this framework with a reduce-by-key application compared to an equivalent version that uses object storage for communication. For a job with 100+ functions, our TCP implementation runs 4.7 times faster at almost half the cost. / Master of Science / Serverless computing is a branch of cloud computing where users can remotely run small programs, called "functions," and pay only based on how long they run. A growing use for serverless computing is running large data processing applications that use many of these serverless functions at once, taking advantage of the fact that serverless programs can be started quickly and on-demand. Because serverless functions run on isolated networks from each other and can only make outbound connections to the public internet, a major challenge with this paradigm is transferring temporary computation data between functions. Previous works have used separate types of cloud storage services in combination with serverless computing to allow functions to exchange data. However, hard-drive--based storage is slow and memory-based storage is expensive. We evaluate the use of direct network connections between functions to overcome these limitations. Although functions cannot receive incoming network connections, we are able to bypass this restriction by using a standard networking technique called Network Address Translation (NAT) traversal. We use an external server as an initial relay to setup a network connection between two functions such that once the connection is established, the functions can communicate directly with each other without using the server anymore. In addition, we develop a communications framework to manage NAT traversal and data flow for applications using direct network connections. We evaluate this framework with an application for combining matching data entries and compare it to an equivalent version that uses storage based on hard drives for communication. For a job with over 100 functions, our implementation using direct network connections runs 4.7 times faster at almost half the cost.

Yes / Interferon lambda (IFNλ) is important for epidermal defence against viruses. It is produced by, and acts on, keratinocytes, whereas fibroblasts were previously considered to be unresponsive to this type III IFN. Herein we report findings revealing cell type-specific differences in IFNλ signalling and function in skin resident cells. In dermal fibroblasts, IFNλ induced the expression of MxA, a potent antiviral factor, but not other IFN signature genes as it does in primary keratinocytes. In contrast to its effect on keratinocytes, IFNλ did not phosphorylate STAT1 in fibroblasts, but instead activated MAPKs. Accordingly, inhibition of MAPK activation (p38 and p42/44) blocked the expression of MxA protein in fibroblasts but not in keratinocytes. Functionally, IFNλ inhibited proliferation in keratinocytes but not in fibroblasts. Moreover, IFNλ upregulated the expression of TGFβ1-induced collagens in fibroblasts. Taken together, our findings identify primary human dermal fibroblasts as responder cells to IFNλ. Our study shows cutaneous cell type-specific IFN signalling and suggests that IFNλ, whilst important for epidermal anti-viral competence, may also have a regulatory role in the dermal compartment balancing type I IFN-induced inhibition of tissue repair processes.

Cette thèse concerne la combinaison du lambda-calcul et de la réécriture, dont nous étudions principalement deux propriétés : la confluence et la normalisation forte. Nous commençons par étudier sous quelles conditions la combinaison d'une relation de réécriture conditionnelle confluente au lambda-calcul donne une relation de réécriture confluente. Ensuite nous nous intéressons aux preuves de normalisation forte de lambda-calculs typés utilisant la technique de réductibilité. Notre contribution la plus importante est une comparaison de diverses variantes de cette technique, utilisant comme outil de comparaison la manière dont ces variantes s'étendent à la réécriture et dont elles prennent en compte les types unions et les types existentiels implicites. Enfin, nous présentons un critère, basé sur un système de types contraints, pour la normalisation forte de la réécriture conditionnelle combinée au lambda-calcul. Notre approche étend des critères de terminaison existants qui utilisent des annotations de taille. C'est à notre connaissance le premier critère de terminaison pour la réécriture conditionnelle avec membres droits d'ordre supérieur qui prenne en compte, dans l'argument de terminaison, de l'information issue de la satisfaction des conditions des règles de réécriture / This thesis is about the combination of lambda-calculus with rewriting. We mainly study two properties: confluence and strong normalization. We begin by studying under which conditions the combination of a confluent conditional rewrite relation to the lambda-calculus leads to a confluent relation. Next, we study strong normalization proofs of typed lambda-calculi that use the reducibility technique. Our main contribution is a comparison of variants of this technique, with respect to how they extend to rewriting and how they handle union and implicit existential types. Finally, we present a termination criterion for the combination of conditional rewriting and lambda-calculus based on a constrained type system. Our approach, which extends known criteria that use sized types, is to our knowledge the first termination criterion for conditional rewriting with higher-order right-hand sides that takes into account in the termination argument some information generated by the satisfaction of the conditions of the rewrite rules

En la presente tesis se desarrolla un sistema de control de la mezcla de aire y gasolina en un motor alternativo de cuatro tiempos, basado en redes neuronales. Para que el catalizador logre un grado de depuración aceptable con todos los gases contaminantes simultáneamente, debe mantenerse la proporción de aire y combustible, dentro de una banda muy estrecha. En estado estacionario, este objetivo se cumple sin demasiados problemas, pero el funcionamiento habitual de un vehículo es en régimen muy transitorio, donde los sistemas convencionales no logran evitar desviaciones importantes del punto de consigna. Estos se basan además en una gran cantidad de tablas estáticas, que deben calibrarse de forma experimental, lo cual es bastante costoso tanto en tiempo como en dinero.Para evitar estos problemas se ha diseñado un modelo matemático de un motor, que comprende todo el proceso de formación de la mezcla, sensores, la generación del par motor y la dinámica del vehículo. La realización final se ha realizado en el lenguaje de simulación MatLab/Simulink®. Los datos requeridos son fácilmente obtenibles bien por metrología, bien de forma experimental. Se ha validado con un motor SEAT de 1,6 l y 74 kW.Como primer paso se ha aplicado una estrategia de control convencional bastante simple, consistente en un controlador feedforward estático, más un controlador feedback de tipo PI ó PID. Esto ha permitido el estudio de las principales características del motor desde el punto de vista de control. Con los resultados obtenidos se ha diseñado un observador basado en una red neuronal, que elimine los retardos puros del sistema y que pueda ser utilizado para cerrar el lazo de control. Primero se ha usado una red feedforward, pero vistos los malos resultados, se ha desarrollado una red neuronal recurrente a partir de la red de Elman, que se ha modificado convenientemente para adaptarla a las dificultades propias del problema. El algoritmo de entrenamiento utilizado se basa en el de retropropagación clásico, y modifica no sólo los pesos entre capas, sino también los correspondientes a las neuronas de contexto, las cuales permiten memorizar estados internos. La principal mejora consiste en separar las neuronas de contexto en tantos grupos como entradas tiene la red, y entrenarlos por separado, de modo que cada grupo se adapte a la dinámica particular de la entrada a la que va asociado. Se muestra mediante simulación el comportamiento del conjunto motor más observador en lazo cerrado, y se compara con el esquema convencional. Se prueba asimismo la robustez del sistema frente a distintas consignas, ruido en la planta y defectos de sintonía. / In the present thesis a control system for the air-fuel mixture in a reciprocating four-stroke engine is developed, based on neural networks. The air-fuel ratio has to be kept within a very narrow window so that the catalyst achieves an acceptable degree of purification simultaneously with all the polluting gases. In steady state, this goal can be fulfilled without difficulties, but the usual operation of a vehicle is in a very transient state, where the conventional systems are not able to avoid important excursions from the set point. They also rely on a great number of look-up tables, which have to be tuned experimentally, thus with an enormous investment of money and time.To avoid those problems a mathematical model of an engine has been designed, in such a way that it comprises of the whole mixture formation process, sensors, the torque generation and the vehicle dynamics. Finally it has been implemented in the simulation language MatLab/Simulink®. The required data is easily available both from metrology, and experimental work. It has been validated with a 1,6 litre 74 kW SEAT engine.As a first step, a quite simple conventional control strategy has been applied, consisting of a static feedforward controller, and a PI or PID feedback controller. This has permitted studying the main features of the engine from the control point of view.With the obtained results an observer based on a neural network has been designed, which eliminates the delays of the system and that can be used to close the control loop. First a feedforward network has been used, but due to the bad results, a recurrent neural network has been developed starting from the Elman network, which has been properly modified in order to adapt it to the characteristic difficulties of the problem. The training algorithm used is based on that of classical backpropagation, and it modifies not only the weights interconnecting different layers, but also those corresponding to the context neurons, which allow the memorising of internal states. The main improvement consists in separating the context neurons in as many groups as the network has inputs, and to train them separately, so that each group adapts to the particular dynamics of the input with which it is associated. The behaviour of the engine plus the observer in closed loop is shown by means of simulation, and is compared with the conventional scheme. It is proven the robustness of system response to different set points, noise in the plant and tuning defects.

Cette thèse étudie deux modèles de réalisabilité pour la logique classique construits sur la sémantique des jeux HO, interprétant la logique, l'arithmétique et l'analyse classiques directement par des programmes manipulant un espace de stockage d'ordre supérieur.La non-innocence en jeux HO autorise les références d'ordre supérieur, et le non parenthésage révèle la CPS des jeux HO et fournit une catégorie de continuations dans laquelle interpréter le lambda-mu calcul de Parigot. Deux modèles de réalisabilité sont construits sur cette interprétation calculatoire directe des preuves classiques.Le premier repose sur l'orthogonalité, comme celui de Krivine, mais il est simplement typé et au premier ordre. En l'absence de codage de l'absurdité au second ordre, une mu-variable libre dans les réaliseurs permet l'extraction. Nous définissons un bar-récurseur et prouvons qu'il réalise l'axiome du choix dépendant, utilisant deux conséquences de la structure de CPO du modèle de jeux: toute fonction sur les entiers (même non calculable) existe dans le modèle, et toute fonctionnelle sur des séquences est Scott-continue. La bar-récursion est habituellement utilisée pour réaliser intuitionnistiquement le « double negation shift » et en déduire la traduction négative de l'axiome du choix. Ici, nous réalisons directement l'axiome du choix dans un cadre classique.Le second, très spécifique au modèle de jeux, repose sur des conditions de gain: des ensembles de positions d'un jeu munis de propriétés de cohérence. Un réaliseur est alors une stratégie dont les positions sont toutes gagnantes. / This thesis investigates two realizability models for classical logic built on HO game semantics. The main motivation is to have a direct computational interpretation of classical logic, arithmetic and analysis with programs manipulating a higher-order store.Relaxing the innocence condition in HO games provides higher-order references, and dropping the well-bracketing of strategies reveals the CPS of HO games and gives a category of continuations in which we can interpret Parigot's lambda-mu calculus. This permits a direct computational interpretation of classical proofs from which we build two realizability models.The first model is orthogonality-based, as the one of Krivine. However, it is simply-typed and first-order. This means that we do not use a second-order coding of falsity, and extraction is handled by considering realizers with a free mu-variable. We provide a bar-recursor in this model and prove that it realizes the axiom of dependent choice, relying on two consequences of the CPO structure of the games model: every function on natural numbers (possibly non computable) exists in the model, and every functional on sequences is Scott-continuous. Usually, bar-recursion is used to intuitionistically realize the double negation shift and consequently the negative translation of the axiom of choice. Here, we directly realize the axiom of choice in a classical setting.The second model relies on winning conditions and is very specific to the games model. A winning condition is a set of positions in a game which satisfies some coherence properties, and a realizer of a formula is then a strategy which positions are all winning.

Grâce au développement des technologies de métagénomique au cours des dix dernières années, il a été constaté que les micro-organismes représentent la plus grande ressource de diversité métabolique et génétique sur Terre. En effet, un gramme de sol contient 109 cellules bactériennes et 103-104 différentes espèces bactériennes. Certaines sont en mesure de réaliser des réactions enzymatiques conduisant à la dégradation complète de certains polluants toxiques pour l’environnement comme les composés organiques tels que la quinoléine. Cependant, l'immense réservoir de molécules et enzymes microbiennes n'a pas encore été exploité, car plus de 99% d'entre elles ne sont, pour l’instant, pas cultivables in vitro. Mon travail s’inscrit dans le cadre d’une collaboration entre l’Université SJTU (Shanghai Jiao Tong Université en Chine) et le groupe de G. M.E (Génomique Microbienne Environmentale) du laboratoire Ampère à l’Ecole Centrale de Lyon. Nos partenaires à l’Université SJTU ont construit un réacteur de dénitrification à l'échelle du laboratoire capable de dégrader la quinoléine en retirant la demande chimique en oxygène. Un nouvel outil appelé "Genefish" a été developpé dans notre laboratoire comme une méthode alternative de la métagénomique pour aider à la découverte de nouveaux gènes d’intérêt industriel ou environnemental. A la suite des premiers travaux réalisés dans notre laboratoire, ma thèse présentée ici comporte deux parties.Dans la première partie de ce travail, nous avons étudié le potentiel de dégradation de la quinoléine présente dans les bactéries d’un sol de référence largement étudié au laboratoire. Pour cela nous avons mis en place des expériences de microcosme qui visent à révéler la diversité potentielle des bactéries responsables de la dégradation de la quinoléine. Des analyses comparatives des profils RISA (Ribosomal Intergenic Spacer analysis) nous ont permis de mettre en évidence des changements dans la structure de la communauté des bactéries du sol incubé en conditions aérobie et anaérobie en présence de quinoléine. La dégradation de la quinoléine a été confirmée par technique de GC/MS (Gas Chromatography-Mass Spectrometry). Les travaux futurs seront de vérifier la communauté de bactéries responsables de la dégradation de quinoléine en utilisant la technique de NGS (Next Generation Sequencing).Le deuxième objectif de ma thèse a été d'utiliser Genefish dont la finalité est de capturer des gènes ciblés (le gène bcr qui serait responsable de la degradation de quinoléine dans le réacteur de nos partenaires) dans l'ADN métagénomique extrait du sol. Genefish consiste à élaborer une souche d’E.coli incluant un plasmide de capture permettant de pêcher les gènes recherchés dans un échantillon d’ADN metagénomique par recombinaison homologue. Le plasmide de capture comprend une cassette de deux gènes toxiques pour la souche qui activés par induction chimique vont permettre la sélection positive directe des clones recombinants, et deux sites multiples de clonage dans lesquels sont insérées les zones de recombinaison qui vont jouer le rôle d’hameçons. Nous avons testé la capacité de Genefish à capturer des produits PCR du gène bcr, l'efficacité de recombinaison reste faible à cause de la persistance de plusieurs copies du plasmide suicide dans la cellule après l’ évenement de recombinaison. Par conséquent, trois stratégies ont été essayées pour améliorer l’efficacité: la co-électroporation, la ségrégation de plasmide et la construction de plasmide suicide en mono-copie. Finalement, la stratégie de la ségrégation plasmidique fonctionne mais l'efficacité de recombinaison est encore trop faible peut-être due à l’incertitude des modèles de recombinaison homologue. Les travaux futurs se concentreront sur l'amélioration des fréquences de recombinaison par transfert de fragments du plasmide de capture dans le chromosome de la souche Genefish. / As the development of metagenomic technologies in the past ten years, it is unquestionned that microorganisms encompass the largest resource of metabolic and genetic diversity in the world. Actually, one gramme of soil contains more than 109 bacteria and 103-104 species. Some of their members are able to carry out enzymatic reactions leading to the complete degradation of pollutants (such as quinoline). So, the biodegradation of some highly toxic or organic compounds by microorganisms will be a general trend for pollutant treatment. However, the huge reservoir of molecules and enzymes from microorganisms still need to be explored because more than 99% of microorganisms cannot be cultivated in vitro.My work was based on collaboration between the University SJTU and Ecole Centrale de Lyon. Our partners at the University SJTU have built a laboratory scale denitrification reactor which was capable of degrading quinoline by removing the chemical oxygen demand. A new tool called "Genefish" has been developed in our laboratory as an alternative method for metagenomics which aims to discover novel industrial or environmental genes of interest. Following the early work in our laboratory, my thesis is presented here in two parts.In the first part, we set up a quinoline microcosm experiment both under aerobic and anaerobic condition using reference soil extensively studied in the laboratory at Ecole Centrale de LYON. This work aimed to reveal the potential bacterial diversity and even genes responsible for quinoline degradation. We used RISA(Ribosomal intergenic Spacer analysis) to analyze the bacteria community structure changes and GC/MS (Gas Chromatography-Mass Spectrometry) was also used to detect the quinoline degradation and reveal potential quinoline metabolic pathways under aerobic and anaerobic condition. Results showed great bacteria community structure changes and high quinoline degradation activity after the quinoline addition under aerobic condition. The future work is to investigate the bacteria community which may be responsible for quinoline degradation using the technique of NGS (Next Generation Sequencing).The second object of my thesis was to use the Genefish tool to capture targeted genes (the bcr gene responsible for the quinoline degradation in the wastewater treatment bioreactor) from the soil metagenome. The aim was to construct an E.coli strain containing a capture plasmid and Red system for capturing targeted genes from metagenomic DNA by homologous recombination. The capture plasmid includes a toxic cassette consisting of two suicide genes which can be activated by chemical induction, finally support the positive recombinants selection. It also contains two multiply cloning sites in which highly conserved sequences were inserted and works as the bait during recombination. We have tested the capacity of Genefish to capture the PCR products of bcr gene; the efficiency was low because of the persistence of several copies of the capture plasmid into the Genefish strain after recombination events. So, three strategies were tried to improve the recombination efficiency: co-electroporation, plasmid segregation and mono-copy capture plasmid construction. Finally, the strategy of plasmid segregation works but the recombination efficiency was still low maybe caused by the uncertain model of homologous recombination. The further research will focus on the transfer of the toxic cassette and homologous arms into the host strain chromosome, this new strategy will exclude the bad effect of low copy number capture plasmid, uncertain model of λ Red induced homologous recombination and the homologous arms site in the capture plasmid which are the most important factors influencing the homologous recombination efficiency in Genefish.

Cette thèse se compose de trois parties. La première partie est consacrée aux partitions planes cylindriques, la deuxième aux lambda-déterminants et enfin la troisième aux commutateurs dans les systèmes semi-circulaires. La classe des partitions planes cylindriques est une généralisation naturelle de celle des partitions planes inverses. Borodin a donnée récemment une série génératrice pour les partitions planes cylindriques. Notre premier résultat est une preuve bijective de cette identité utilisant les diagrammes de croissance de Fomin for la correspondance RSK généralisée. Le deuxième résultat est un (q, t)-analogue de la formule de Borodin, qui généralise un résultat d'Okada. Enfin le troisième résultat de la première partie est une description combinatoire explicite du poids de Macdonald intervenant dans cette formule, qui utilise un modèle de chemins non-intersectant pour les partitions planes cylindriques. Les matrices à signes alternants ont ́été découvertes par Robbins et Rumsey alors qu’ils étudiaient les λ-déterminants. Dans la deuxième partie de cette thèse nous démontrons une généralisation à plusieurs paramètres de ce λ-déterminant, généralisant un résultat récent de di Francesco. Comme le λ-déterminant, notre formule est un exemple du phénomène de Laurent. Les systèmes semi-circulaires ont ́été introduits par Voiculescu afin d' ́étudier les algèbres de von Neumann des groupes libres. Dans la troisième partie de la thèse, nous étudions les commutateurs dans l'algèbre engendré par un système semi-circulaire. Nous avons mis en ́évidence une matrice possédant une structure auto-similaire intéressante, qui nous permet de donner une formule explicite pour la projection sur l'espace des commutateurs de degré donnée. En utilisant cette expression, nous donnons une preuve simple du fait que les systèmes semi-circulaires engendrent des facteurs / This thesis is divided into three parts. The first part deals with cylindric plane partitions. The second with lambda-determinants and the third with commutators in semi-circular systems. Cylindric plane partitions may be thought of as a natural generalization of reverse plane partitions. A generating series for the enumeration of cylindric plane partitions was recently given by Borodin. The first result of section one is a new bijective proof of Borodin's identity which makes use of Fomin's growth diagram framework for generalized RSK correspondences. The second result is a (q, t)-analog of Borodin's identity which extends previous work by Okada in the reverse plane partition case. The third result is an explicit combinatorial interpretation of the Macdonald weight occuring in the(q, t)-analog using the non-intersecting lattice path model for cylindric plane partitions. Alternating sign matrices were discovered by Robbins and Rumsey whilst studying λ-determinants. In the second part of this thesis we prove a multi-parameter generalization of the λ-determinant, generalizing a recent result by di Francesco. Like the original λ-determinant, our formula exhibits the Laurent phenomenon. Semicircular systems were first introduced by Voiculescu as a part of his study of von Neumann algebras. In the third part of this thesis we study certain commutator sub algebras of the semicircular system. We find a projection matrix with an interesting self-similar structure. Making use of our projection formula we given an alternative, elementary proof that the semicircular system is a factor

La thèse présentée ici a pour but de proposer des systèmes de détection, de composition de constituants sémantiques et d’interprétation dans la compréhension de la langue naturelle parlée. Cette compréhension se base sur un système de reconnaissance automatique de la parole qui traduit les signaux oraux en énoncés utilisables par la machine. Le signal de la parole, ainsi transcrit, comporte un ensemble d’erreurs liées aux erreurs de reconnaissance (bruits, parasites, mauvaise prononciation...). L’interprétation de cet énoncé est d’autant plus difficile qu’il est issu d’un discours parlé, soumis à la disfluence du discours, aux auto-corrections... L’énoncé est de plus agrammatical, car le discours parlé lui-même est agrammatical. L’application de méthodes d’analyses grammaticales ne produit pas de bons résultats d’interprétation, sur des textes issus de transcriptions de la parole. L’utilisation de méthodes d’analyses syntaxiques profondes est à éviter. De ce fait, une analyse superficielle est envisagée. Un des premiers objectifs est de proposer une représentation du sens. Il s’agit de considérer des ontologies afin de conceptualiser le monde que l’on décrit. On peut exprimer les composants sémantiques en logique du premier ordre avec des prédicats. Dans les travaux décrits ici, nous représentons les éléments sémantiques par des frames (FrameNet ). Les structures de frames sont hiérarchisées, et sont des fragments de connaissances auxquels on peut insérer, fusionner ou inférer d’autres fragments de connaissances. Les structures de frames sont dérivables en formules logiques. Nous proposons un système de compréhension de la parole à partir de règles logiques avec le support d’une ontologie, afin de pouvoir créer des liens à partir de composants sémantiques. Puis, nous avons mené une étude sur la découverte des supports syntaxiques des relations sémantiques. Nous proposons une expérience de composition sémantique afin d’enrichir les composants sémantiques de base. Enfin, nous présentons un système de détection de lambda-expression pour mettre en hypothèse les relations à trouver à travers le discours / The thesis presented here is intended to provide detection systems, composition of components and semantic interpretation in the natural spoken language understanding. This understanding is based on an automatic speech recognition system that translates the signals into oral statements used by the machine. The transcribed speech signal, contains a series of errors related to recognition errors (noise, poor pronunciation...). The interpretation of this statement is difficult because it is derived from a spoken discourse, subject to the disfluency of speech, forself-correction... The statement is more ungrammatical, because the spoken discourse itself is ungrammatical. The application of grammatical analysis methods do not produce good results interpretation, on the outcome of speech transcription. The use of deep syntactic analysis methods should be avoided. Thus, a superficial analysis is considered. A primary objective is to provide a representation of meaning. It is considered ontologies to conceptualize the world we describe. We can express the semantic components in first order logic with predicates. In the work described here, we represent the semantic elements by frames (FrameNet ). The frames are hierarchical structures, and are fragments of knowledge which can be inserted, merge or infer other fragments of knowledge. The frames are differentiable structures in logical formulas. We propose a system for speech understanding from logical rules with the support of an ontology in order to create links from semantic components. Then, we conducted a study on the discovery supports syntactic semantic relationships. We propose a compositional semantics experience to enrich the basic semantic components. Finally, we present a detection system for lambda-expression hypothesis to find the relationship through discourse

No description available.