Dissertations / Theses on the topic 'Semantic Error'

The present study replicates research by Tomoko Takahashi (1984) on lexico-semantic patterns used by students in an acquisition poor environment. The purpose of the current study was to determine how an acquisition rich environment affects learners' use of four lexico-semantic patterns: congruence occurs when the Ll definition of a lexical item forms a one-to-one correspondence with the L2 lexical item; convergence occurs when the Ll lexical item has broader applications than the L2 lexical item; divergence occurs when the L2 lexical item has broader applications than the Ll lexical item; and semantic gap occurs when the Ll lexical item has no appropriate corresponding L2 lexical item (Takahashi, 1984). The instrument, a lexico-semantics test, is the same instrument used in Takahashi's study. It was designed to measure which patterns are most frequently used by Japanese EFL students learning English. The results, unlike Takahashi's, suggest that beginning and advanced ESL students use the four patterns equally well. No significant difference was found between the two groups. These results are contrary to what had been expected. However, they show that the proposed hierarchical order of difficulty of congruence, convergence, divergence and semantic gap is the same in both studies. The results also indicate that the acquisition· rich environment seems to dramatically improve beginners' performance of the four patterns. Since the instrument was designed for EFL students (an acquisition poor environment) it may not have fully challenged the advanced ESL students (an acquisition rich environment) while challenging the beginning students. This may have been due to the fact that the students in the present study received a great deal of input from the acquisition rich environment, which could account for their increased ability to restructure hypotheses about L2 vocabulary items. In conclusion, more studies are needed to determine the complete role of the four lexico-semantic patterns in vocabulary acquisition. An expanded follow up study that fully tests the advanced and beginning ESL learners' ability could determine whether both groups progress along a language continuum with respect to the use of the four lexico-semantic patterns. Furthermore, although the patterns may serve, in a limited capacity, as indicators of a learner's difficulties in vocabulary acquisition, a wider body of research is needed before they can be applied in a language learning environment.

Type systems were invented in the early 1900s to provide foundations for Mathematics where types were used to avoid paradoxes. Type systems have then been developed and extended throughout the years to serve different purposes such as efficiency or expressiveness. The λ-calculus is used in programming languages, logic, mathematics, and linguistics. Intersection types are a kind of types used for building semantic models of the λ-calculus and for static analysis of computer programs. The confluence property was used to prove the λ-calculus’ consistency and the uniqueness of normal forms. Confluence is useful to show that logics are sensibly designed, and to make equality decision procedures for use in theorem provers. Some proofs of the λ-calculus’ confluence are based on syntactic concepts (reduction relations and λ-term sets) and some on semantic concepts (type interpretations). Part I of this thesis presents an original syntactic proof that is a simplification of a semantic proof based on a sound type interpretation w.r.t. an intersection type system. Our proof can be seen as bridging some semantic and syntactic proofs. Expansion is an operation on typings (pairs of type environments and result types) in type systems for the λ-calculus. It was introduced to prove that the principal typing property (i.e., that every typable term has a strongest typing) holds in intersection type systems. Expansion variables were introduced to simplify the expansion mechanism. Part II of this thesis presents a complete realisability semantics w.r.t. an intersection type system with infinitely many expansion variables. This represents the first study on semantics of expansion. Providing sound (and complete) realisability semantics allows one to study the algorithmic behaviour of typed λ-terms through their types w.r.t. a type system. We believe such semantics will cast some light on the not yet well understood expansion operation. Intersection types were used in a type error slicer for the SML programming language. Existing compilers for many languages have confusing type error messages. Type error slicing (TES) helps the programmer by isolating the part of a program contributing to a type error (a slice). TES was initially done for a tiny toy language (the λ-calculus with polymorphic let-expressions). Extending TES to a full language is extremely challenging, and for SML we needed a number of innovations. Some issues would be faced for any language, and some are SML-specific but representative of the complexity of language-specific issues likely to be faced for other languages. Part III of this thesis solves both kinds of issues and presents an original, simple, and general constraint system for providing type error slices for ill-typed programs. We believe TES helps demystify language features known to confuse users.

U radu se identifikuju unutarjezicke gre&scaron;ke,sistematizuju se prema jezickim nivoima itipovima, morfolo&scaron;kim i sintaksickosemantickimmodelima, zatim se opisuju igre&scaron;ke i uzroci njihovog nastanka (gde god je tomoguce). Izlažu se rezultati analize prema:- jezickim nivoima,- gramatickim kategorijama,- ucestalosti pojavljivanja gre&scaron;aka,- prvom jeziku studenata, tako ce se pokazatikoje gre&scaron;ke su odlika govornika samo jednogpolaznog jezika, a koje su zajednickegovornicima razlicitih ili možda cak svihpolaznih jezika.<br>In the paper, intralinguistic errors areidentified, systematized according to linguisticlevels and types, morphological and syntacticand semantic models. Errors are then describedincluding the causes of their occurrence (incases in which it is possible). The results of theanalysis are presented according to:- linguistic levels,- grammatical categories,- frequency of error occurrence,- students&#39; L1, which will indicate whicherrors are characteristic of speakers ofonly one L1 as well as which errors arecommon to speakers of different orpossibly all the native languages

La thèse concerne le rôle joué par le contexte dans l'accès au Web de données depuis les dispositifs mobiles. Le travail analyse ce problème de deux points de vue distincts: adapter au contexte la présentation de triplets, et protéger l'accès aux bases des données RDF depuis les dispositifs mobiles. La première contribution est PRISSMA, un moteur de rendu RDF qui étend Fresnel avec la sélection de la meilleure représentation pour le contexte physique où on se trouve. Cette opération est effectuée par un algorithme de recherche de sous-graphes tolérant aux erreurs basé sur la notion de distance d'édition sur les graphes. L'algorithme considère les différences entre les descriptions de contexte et le contexte détecté par les capteurs, supporte des dimensions de contexte hétérogènes et est exécuté sur le client pour ne pas révéler des informations privées. La deuxième contribution concerne le système de contrôle d'accès Shi3ld. Shi3ld supporte tous les triple stores et il ne nécessite pas de les modifier. Il utilise exclusivement les langages du Web sémantique, et il n'ajoute pas des nouveaux langages de définition de règles d'accès, y compris des analyseurs syntaxiques et des procédures de validation. Shi3ld offre une protection jusqu'au niveau des triplets. La thèse décrit les modèles, algorithmes et prototypes de PRISSMA et de Shi3ld. Des expériences montrent la validité des résultats de PRISSMA ainsi que les performances au niveau de mémoire et de temps de réponse. Le module de contrôle d'accès Shi3ld a été testé avec différents triple stores, avec et sans moteur SPARQL. Les résultats montrent l'impact sur le temps de réponse et démontrent la faisabilité de l'approche.

A nomeação constitui uma das tarefas mais importantes no processamento da linguagem. A nomeação de diferentes categorias semânticas e gramaticais diverge em suas propriedades lexicais e possui substratos neuroanatômicos distintos. Esta pesquisa tem como objetivos caracterizar as alterações de nomeação de sujeitos com afasia, comparando-os aos sujeitos sem alteração de linguagem e estabelecer relações da tipologia de resposta na tarefa de nomeação de verbos e de substantivos em afásicos. A amostra foi composta por indivíduos divididos em dois grupos: GC e GP. O GC foi constituído de 95 sujeitos sem alteração de linguagem, com média de idade de 56,7 anos (15,4) e escolaridade de 9,3 anos (4,4). O GP foi constituído por 33 afásicos, subdividido em não fluentes (GNF) e fluentes (GF). No GNF foram avaliados 17 sujeitos com média de idade de 65,5 anos (14) e escolaridade de 8,2 anos (6,4). No GF foram avaliados 16 sujeitos com média de idade de 60,5 anos (16,7) e escolaridade de 11,5 anos (7,6). O teste utilizado foi a bateria inglesa An Object and Action Naming Battery (Bateria de Nomeação de Objetos e Verbos - BNOV). Houve diferenças estatisticamente significativas entre os grupos de afásicos no número de acertos para substantivos e verbos (p<0,001). O número de acertos para a nomeação de substantivos é maior do que para verbos no GNF enquanto o número de acertos para a nomeação de verbos é maior no GF. Os sujeitos do GNF apresentaram como tipo de resposta desviante mais frequente a \"complementação por gesto\", seguido dos tipos xiii \"hiperônimo\" e \"componente\" na nomeação de substantivos e \"erros de associação\", \"co-hipônimo\", e \"substantivo relacionado\", na nomeação de verbos. Os sujeitos do GF apresentaram como tipo de resposta desviante mais frequente o \"neologismo\", seguido de \"parafasia fonêmica\" e \"circunlóquios\" para a nomeação de substantivos e \"neologismo\", \"erros fonológicos\" e \"circunlóquios\" para verbos.<br>Naming is one of the most important functions in language processing. The naming of different semantic and grammatical categories diverge in their lexical properties and possess distinct neuroanatomical substrates. This study aims at characterizing the naming alterations and types of errors of aphasic individuals in a verbs and nouns naming. The sample was composed of subjects divided into two groups: GC and GP. The GC consisted of 95 subjects with no language impairment, with the mean age of 56.7 years (15.4) and schooling of 9.3 years (4.4). The GP comprised 33 aphasic patients, subdivided into non fluent (GNF) and fluent (GF). In the GNF, 17 subjects with the mean age of 65.5 years (14) and schooling of 8.2 years (6.4) were evaluated. In the GF, 16 subjects with the mean age of 60.5 years (16.7) and 11.5 years of schooling (7.6) were evaluated. The test used was the English battery An Object and Action Naming Battery. There were statistically significant differences between the aphasic groups as to the number of correct answers for nouns and verbs (p <0.001). The number of correct answers in the naming task for nouns was higher than for verbs in the GNF, while the number of correct answers in the naming of verbs was higher in the GF. The most frequent type of error presented by the GNF presented \"complementation by gesture\", followed by \"hyperonym\" and \"component\" in the naming of nouns and \"association errors\", \"hyponym\", and \"related noun\" in the naming of verbs. The most frequent type of error presented by the GF was \"neologism\", followed by \"phonemic paraphasia\" and \"circumlocutions\" in the naming of nouns and \"neologism\", \"phonological errors\" and \"circumlocutions\" for verbs.

Spoken dialogue systems are application interfaces that enable humans to interact with computers using spoken natural language. A major challenge for these systems is dealing with the ubiquity of variability—in user behavior, in the performance of the various speech and language processing sub-components, and in the dynamics of the task domain. However, as the predominant methodology for dialogue system development is to handcraft the sub-components, these systems typically lack robustness in user interactions. Data-driven methods, on the other hand, have been shown to offer robustness to variability in various domains of computer science and are increasingly being used in dialogue systems research.     This thesis makes four novel contributions to the data-driven methods for spoken dialogue system development. First, a method for interpreting the meaning contained in spoken utterances is presented. Second, an approach for determining when in a user’s speech it is appropriate for the system to give a response is presented. Third, an approach for error detection and analysis in dialogue system interactions is reported. Finally, an implicitly supervised learning approach for knowledge acquisition through the interactive setting of spoken dialogue is presented.      The general approach taken in this thesis is to model dialogue system tasks as a classification problem and investigate features (e.g., lexical, syntactic, semantic, prosodic, and contextual) to train various classifiers on interaction data. The central hypothesis of this thesis is that the models for the aforementioned dialogue system tasks trained using the features proposed here perform better than their corresponding baseline models. The empirical validity of this claim has been assessed through both quantitative and qualitative evaluations, using both objective and subjective measures.<br>Den här avhandlingen utforskar datadrivna metoder för utveckling av talande dialogsystem. Motivet bakom sådana metoder är att dialogsystem måste kunna hantera en stor variation, i såväl användarnas beteende, som i prestandan hos olika tal- och språkteknologiska delkomponenter. Traditionella tillvägagångssätt, som baseras på handskrivna komponenter i dialogsystem, har ofta svårt att uppvisa robusthet i hanteringen av sådan variation. Datadrivna metoder har visat sig vara robusta mot variation i olika problem inom datavetenskap och artificiell intelligens, och har på senare tid blivit populära även inom forskning kring talande dialogsystem. Den här avhandlingen presenterar fyra nya bidrag till datadrivna metoder för utveckling av talande dialogsystem. Det första bidraget är en datadriven metod för semantisk tolkning av talspråk. Den föreslagna metoden har två viktiga egenskaper: robust hantering av ”ogrammatisk” indata (på grund av talets spontana natur samt fel i taligenkänning), samt bevarande av strukturella relationer mellan koncept i den semantiska representationen. Tidigare metoder för semantisk tolkning av talspråk har typiskt sett endast hanterat den ena av dessa två utmaningar. Det andra bidraget i avhandlingen är en datadriven metod för turtagning i dialogsystem. Den föreslagna modellen utnyttjar prosodi, syntax, semantik samt dialogkontext för att avgöra när i användarens tal som det är lämpligt för systemet att ge respons. Det tredje bidraget är en data-driven metod för detektering av fel och missförstånd i dialogsystem. Där tidigare arbeten har fokuserat på detektering av fel on-line och endast testats i enskilda domäner, presenterats här modeller för analys av fel såväl off-line som on-line, och som tränats samt utvärderats på tre skilda dialogsystemkorpusar. Slutligen presenteras en metod för hur dialogsystem ska kunna tillägna sig ny kunskap genom interaktion med användaren. Metoden är utvärderad i ett scenario där systemet ska bygga upp en kunskapsbas i en geografisk domän genom så kallad "crowdsourcing". Systemet börjar med minimal kunskap och använder den talade dialogen för att både samla ny information och verifiera den kunskap som inhämtats. Den generella ansatsen i den här avhandlingen är att modellera olika uppgifter för dialogsystem som  klassificeringsproblem, och undersöka särdrag i diskursens kontext som kan användas för att träna klassificerare. Under arbetets gång har olika slags lexikala, syntaktiska, prosodiska samt kontextuella särdrag undersökts. En allmän diskussion om dessa särdrags bidrag till modellering av ovannämnda uppgifter utgör ett av avhandlingens huvudsakliga bidrag. En annan central del i avhandlingen är att träna modeller som kan användas direkt i dialogsystem, varför endast automatiskt extraherbara särdrag (som inte kräver manuell uppmärkning) används för att träna modellerna. Vidare utvärderas modellernas prestanda på såväl taligenkänningsresultat som transkriptioner för att undersöka hur robusta de föreslagna metoderna är. Den centrala hypotesen i denna avhandling är att modeller som tränas med de föreslagna kontextuella särdragen presterar bättre än en referensmodell. Giltigheten hos denna hypotes har bedömts med såväl kvalitativa som kvantitativa utvärderingar, som nyttjar både objektiva och subjektiva mått.<br><p>QC 20160225</p>

Le principal objectif de cette thèse est d'étudier l'incidence des erreurs prépositionnelles sur l'intelligibilité de productions en L2 anglais par des apprenants francophones. Après un résumé des caractéristiques sémantiques et morpho-syntaxiques des prépositions en anglais et en français, nous abordons les questions liées à l'acquisition d'une langue seconde en général et à l'acquisition des prépositions en particulier, afin d' identifier les facteurs qui peuvent rendre leur acquisition problématique en L2. Nous proposons également des solutions pédagogiques pour améliorer l'apprentissage des prépositions en anglais. Afin de mesurer l'intelligibilité, nous analysons un corpus de productions orales et écrites contenant des emplois erronés, répartis entre erreurs lexciales et erreurs lexico- grammaticales (additions, omissions ou substitutions). Les résultats de cette analyse permettent de voir dans quelle mesure les erreurs prépositionnelles affectent l'intelligibilité du message.

A crucial component of child language acquisition is successful generalization. First, a speaker must acquire abstract knowledge of how a particular linguistic-structure conveys meaning, and use this knowledge to generalize the structure to new lexical-items. For example, a speaker can use abstract knowledge of a SUBJECT-VERB-OBJECT structure to produce a sentence such as The man rolled the ball, even if the verb roll has never been encountered in this structure before. Second, a learner must appropriately restrict ‘overgeneralizations’ whereby a structure is used with an unsuitable verb (e.g. *The man fell the boy). The most prominent theories regarding restriction of overgeneralizations are based on frequency of use and (semantic, phonological or pragmatic) compatibility between the item and construction. Since developmental evidence for these accounts is mostly limited to the judgment paradigm, which is unsuitable for testing children aged 5 and under, the aim of this thesis was to examine whether these restriction mechanisms are used by children as young as 3 or 4 – whose generalization mechanisms are at an earlier stage of development - and to develop new paradigms for doing so. Study 1 used a production priming paradigm to examine children’s (aged 3-4; 5-6) restrictions of verbal un- prefixation (e.g., *unbend). Children’s production probability of verbs in un- form (e.g., *unbend) was negatively predicted by the frequency of the target verb in bare form (i.e., bend/s/ed/ing) and by the frequency of synonyms to a verb’s un- form (e.g., straighten for *unbend). Additionally, grammaticality judgments from older children (aged 5-6) revealed that preferences for un- forms were positively related to the extent to which the verb’s semantics overlapped with a covert, probabilistic semantic “cryptotype” of meanings thought to be shared by verbs that are grammatical in un- form (e.g., tie, pack, twist, screw, cover). Study 2 investigated whether overgeneralization errors in the domain of English past-tense (i.e., when ‘regular’ inflections are applied to verbs that require ‘irregular’ inflection; e.g., *sleeped, *throwed) are best attributed to analogy across exemplars, or to a default, “add –ed” rule applied regardless of a verb’s memorized associations. Past-tense forms of novel verbs were elicited by showing children (aged 3-4; 5-6; 6-7; 9-10) animations of an animal performing a novel action described with a novel verb (e.g., gezz; chake) and asking what the animal ‘did yesterday’. A verb’s likelihood of receiving regular inflection (e.g., gezzed, chaked) was positively associated with its phonological similarity to existing regular verbs, consistent with the analogy-based approach. Study 3 investigated the suitability of an online measure of sentence processing, namely Event Related Potentials (ERP), to investigate the role of verb-frequency in restricting transitive overgeneralizations. In line with previous studies, ‘P600’ and ‘LAN’ components were evoked in response to overgeneralization errors. However, the magnitudes of these components were not sensitive to a manipulation of verb-frequency (e.g., *The clown laughed the boy vs. *The clown giggled the boy), raising doubt toward the suitability of ERP for examining the relative acceptability of overgeneralization errors. Overall, the research indicates that even young children’s generalizations are sensitive to the linguistic input (i.e., statistical regularities and generalized semantic or phonological patterns of use) and are not well explained by a system of abstract rules that act on discrete categories, whether this applies to syntactic categories (e.g., add -ed to any instance of the category “VERB”) or discrete verb classes (e.g., a narrow-range rule that acts invariably on any verb that is part of an ‘alternating’ verb-class).

Cette thèse s'inscrit dans le cadre de l'émergence de l'apprentissage profond et aborde la compréhension de la parole assimilée à l'extraction et à la représentation automatique du sens contenu dans les mots d'une phrase parlée. Nous étudions une tâche d'étiquetage en concepts sémantiques dans un contexte de dialogue oral évaluée sur le corpus français MEDIA. Depuis une dizaine d'années, les modèles neuronaux prennent l'ascendant dans de nombreuses tâches de traitement du langage naturel grâce à des avancées algorithmiques ou à la mise à disposition d'outils de calcul puissants comme les processeurs graphiques. De nombreux obstacles rendent la compréhension complexe, comme l'interprétation difficile des transcriptions automatiques de la parole étant donné que de nombreuses erreurs sont introduites par le processus de reconnaissance automatique en amont du module de compréhension. Nous présentons un état de l'art décrivant la compréhension de la parole puis les méthodes d'apprentissage automatique supervisé pour la résoudre en commençant par des systèmes classiques pour finir avec des techniques d'apprentissage profond. Les contributions sont ensuite exposées suivant trois axes. Premièrement, nous développons une architecture neuronale efficace consistant en un réseau récurent bidirectionnel encodeur-décodeur avec mécanisme d’attention. Puis nous abordons la gestion des erreurs de reconnaissance automatique et des solutions pour limiter leur impact sur nos performances. Enfin, nous envisageons une désambiguïsation de la tâche de compréhension permettant de rendre notre système plus performant<br>This thesis is a part of the emergence of deep learning and focuses on spoken language understanding assimilated to the automatic extraction and representation of the meaning supported by the words in a spoken utterance. We study a semantic concept tagging task used in a spoken dialogue system and evaluated with the French corpus MEDIA. For the past decade, neural models have emerged in many natural language processing tasks through algorithmic advances or powerful computing tools such as graphics processors. Many obstacles make the understanding task complex, such as the difficult interpretation of automatic speech transcriptions, as many errors are introduced by the automatic recognition process upstream of the comprehension module. We present a state of the art describing spoken language understanding and then supervised automatic learning methods to solve it, starting with classical systems and finishing with deep learning techniques. The contributions are then presented along three axes. First, we develop an efficient neural architecture consisting of a bidirectional recurrent network encoder-decoder with attention mechanism. Then we study the management of automatic recognition errors and solutions to limit their impact on our performances. Finally, we envisage a disambiguation of the comprehension task making the systems more efficient

This dissertation is about the nature of perspectival thoughts and the context-sensitivity of the language used to express them. It focuses on two kinds of perspectival thoughts: ‘subjective' evaluative thoughts about matters of personal taste, such as 'Beetroot is delicious' or 'Skydiving is fun', and first-personal or de se thoughts about oneself, such as 'I am hungry' or 'I have been fooled.' The dissertation defends of a novel form of relativism about truth - the idea that the truth of some (but not all) perspectival thought and talk is relative to the perspective of an evaluating subject or group. In Part I, I argue that the realm of ‘subjective' evaluative thought and talk whose truth is perspective-relative includes attributions of knowledge of the form 'S knows that p.' Following a brief introduction (chapter 1), chapter 2 presents a new, error-theoretic objection against relativism about knowledge attributions. The case for relativism regarding knowledge attributions rests on the claim that relativism is the only view that explains all of the empirical data from speakers' use of the word "know" without recourse to an error theory. In chapter 2, I show that the relativist can only account for sceptical paradoxes and ordinary epistemic closure puzzles if she attributes a problematic form of semantic blindness to speakers. However, in 3 I show that all major competitor theories - forms of invariantism and contextualism - are subject to equally serious error-theoretic objections. This raises the following fundamental question for empirical theorising about the meaning of natural language expressions: If error attributions are ubiquitous, by which criteria do we evaluate and compare the force of error-theoretic objections and the plausibility of error attributions? I provide a number of criteria and argue that they give us reason to think that relativism's error attributions are more plausible than those of its competitors. In Part II, I develop a novel unified account of the content and communication of perspectival thoughts. Many relativists regarding ‘subjective' thoughts and Lewisians about de se thoughts endorse a view of belief as self-location. In chapter 4, I argue that the self-location view of belief is in conflict with the received picture of linguistic communication, which understands communication as the transmission of information from speaker's head to hearer's head. I argue that understanding mental content and speech act content in terms of sequenced worlds allows a reconciliation of these views. On the view I advocate, content is modelled as a set of sequenced worlds - possible worlds ‘centred' on a group of individuals inhabiting the world at some time. Intuitively, a sequenced world is a way a group of people may be. I develop a Stalnakerian model of communication based on sequenced worlds content, and I provide a suitable semantics for personal pronouns and predicates of personal taste. In chapter 5, I show that one of the advantages of this model is its compatibility with both nonindexical contextualism and truth relativism about taste. I argue in chapters 5 and 6 that the empirical data from eavesdropping, retraction, and disagreement cases supports a relativist completion of the model, and I show in detail how to account for these phenomena on the sequenced worlds view.

Sammanfattning Den här uppsatsen undersöker verbanvändningen i skriftlig produktion hos en grupp avancerade svenska som andraspråkselever. Studien granskar såväl kvantitativa som kvalitativa aspekter i elevernas verbordförråd. Uppsatsens syfte är att kartlägga L2-elevernas verbvokabulär med avseende på frekvens, variation, fördelning över semantiska fält samt semantisk och konstruktionsmässig korrekthet. L2-gruppens resultat kontrasteras mot en kontrollgrupp bestående av L1-elever. Undersökningsmaterialet består av 20 argumenterande uppsatser skrivna av komvuxelever med svenska som andraspråk samt 20 argumenterande uppsatser skrivna av komvuxelever med svenska som förstaspråk. Samtliga verb har extraherats, räknats och, i de fall det varit tillämpbart, satts in i ett av de 13 semantiska fält som undersökts i uppsatsen. De olika typer av inkorrekt verbanvändning som påträffats i materialet har kategoriserats utifrån feltyp. De typer av fel som varit av intresse i denna studie är konstruktionsfel i prefixverb, reflexivverb, partikelverb samt fel i verb + substantiv- kollokationer. Förutom konstruktionsfel undersöker uppsatsen även primärt semantiska fel, dvs. fel som i första hand har att göra med verbets semantiska betydelse. Undersökningsresultatet visar att verbfrekvensen hos L2-talarna är lägre i jämförelse med L1-gruppen. Orsaken därtill är att L1-uppsatserna i snitt är längre än L2-uppsatserna och därmed innehåller många fler verb. Det verkar inte föreligga någon anmärkningsvärd skillnad i verbvariationen om man ser till verben i sin helhet. Vissa semantiska fält är dock mer utbyggda och mer varierade i L1-gruppen.         Verbfrekvenstabellerna visar att andelen partikelverb är hälften så stor i L2-gruppen jämfört med L1-gruppen. Listan över de 20 vanligaste verben visar att andraspråkseleverna, vid en jämförelse med L1-eleverna, underanvänder verben få och ska och överanvänder säga, vilja och finnas. Eftersom inga fler indikationer på överanvändning av kärnverb ges i materialet måste just detta interimspråkdrag ses som en mycket begränsad företeelse hos de avancerade L2-eleverna.         De allra vanligaste feltyperna hos såväl L1- som L2-talare är av primärt semantisk karaktär. Bland dessa fel är, i L2-gruppen, fel kopplade till det semantiska fältets betydelsesidoordning vanligast. Det innebär att studien pekar mot att det som ställer till mest problem för andraspråkseleverna, när det gäller verblexikonet, är sådant som har att göra med semantisk precisering och nyansering, dvs. att välja rätt synonym.<br>Abstract This essay deals with lexical competence in written language production among advanced learners of Swedish as a second language. The study focuses on the use of verbs with regard to both quantitative and qualitative aspects. The aim of this essay is to examine the L2-learners verb vocabulary, paying attention to frequency, variation, distribution in semantic fields as well as semantic and constructional correctness. The results of the investigations are compared to the results of a control group consisting of L1-speakers.         The material in the present investigation consists of 40 argumentative essays written by 20 adult students with Swedish as their second language and 20 adult students with Swedish as their first language. All students were enrolled in an adult education program, attending the courses Swedish as a second language B and Swedish B at senior high school level.         All verbs have been extracted, counted and, where applicable, inserted into the 13 semantic fields which are examined in this study. Errors in connection with the use of verb have been categorized according to two major error types, i.e. constructional errors (errors connected to prefixed verbs, reflexive verbs, particle verbs, and errors in verb + noun combinations) and primarily semantic errors.         The investigation demonstrates that the verb frequency is lower in the L2-group compared to the L1-group. The higher verb frequency in the L1-group is probably due to the, on average, longer essays in that particular group.            The verb frequency tables show that use of particle verbs is twice as high in the L1-essays compared to the L2-essays. The tables of the 20 most common verbs demonstrate that L2-students, compared to the L1-students, tend to underuse certain verbs and overuse others. Three verbs that are particularly overused by the L2-students, are säga, vilja and finnas, while the verbs få and ska underused to a great extent by the same group. The tree overused verbs are nuclear verbs in their specific semantic fields (VERBAL COMMUNICATION, WISH and EXISTENCE). Apart from the overuse of these nuclear verbs the investigation material does not give further indications of overuse of nuclear verbs. This shows that overuse of nuclear verbs as an interlanguage feature is not particularly prominent in the advanced L2-learners.              There does not seem to be any notable difference in verb variation when looking at the verbs in their entirety. However, certain semantic fields are more extended and more varied in the L1-group.         In the essays written by the L1-students there are very few occurrences of constructional errors. The most frequent error type, in both student groups, is of a semantic nature. The L2-material shows that the most common semantic error is related to the horizontal structure (and not the vertical structure) of the semantic fields. This means that L2-students find it hard to differ between synonyms in the same semantic field at the same hierarchical level. The current study thus indicates that, as far as the verb vocabulary goes, the L2-students main difficulty is semantic specification, i.e. choosing the right synonym. Among the combinatory errors, the errors connected to verb + noun combinations have the highest frequency.

Cette thèse s’inscrit dans le cadre de l’apprentissage profond appliqué à la compréhension de la parole. Jusqu'à présent, cette tâche était réalisée par l’intermédiaire d’une chaîne de composants mettant en oeuvre, par exemple, un système de reconnaissance de la parole, puis différents traitements du langage naturel, avant d’impliquer un système de compréhension du langage sur les transcriptions automatiques enrichies. Récemment, des travaux dans le domaine de la reconnaissance de la parole ont montré qu’il était possible de produire une séquence de mots directement à partir du signal acoustique. Dans le cadre de cette thèse, il est question d’exploiter ces avancées et de les étendre pour concevoir un système composé d’un seul modèle neuronal entièrement optimisé pour la tâche de compréhension de la parole, du signal au concept. Tout d’abord, nous présentons un état de l’art décrivant les principes de l’apprentissage neuronal profond, de la reconnaissance de la parole, et de la compréhension de la parole. Nous décrivons ensuite les contributions réalisées selon trois axes principaux. Nous proposons un premier système répondant à la problématique posée et l’appliquons à une tâche de reconnaissance des entités nommées. Puis, nous proposons une stratégie de transfert d’apprentissage guidée par une approche de type curriculum learning. Cette stratégie s’appuie sur les connaissances génériques apprises afin d’améliorer les performances d’un système neuronal sur une tâche d’extraction de concepts sémantiques. Ensuite, nous effectuons une analyse des erreurs produites par notre approche, tout en étudiant le fonctionnement de l’architecture neuronale proposée. Enfin, nous mettons en place une mesure de confiance permettant d’évaluer la fiabilité d’une hypothèse produite par notre système<br>This thesis is part of the deep learning applied to spoken language understanding. Until now, this task was performed through a pipeline of components implementing, for example, a speech recognition system, then different natural language processing, before involving a language understanding system on enriched automatic transcriptions. Recently, work in the field of speech recognition has shown that it is possible to produce a sequence of words directly from the acoustic signal. Within the framework of this thesis, the aim is to exploit these advances and extend them to design a system composed of a single neural model fully optimized for the spoken language understanding task, from signal to concept. First, we present a state of the art describing the principles of deep learning, speech recognition, and speech understanding. Then, we describe the contributions made along three main axes. We propose a first system answering the problematic posed and apply it to a task of named entities recognition. Then, we propose a transfer learning strategy guided by a curriculum learning approach. This strategy is based on the generic knowledge learned to improve the performance of a neural system on a semantic concept extraction task. Then, we perform an analysis of the errors produced by our approach, while studying the functioning of the proposed neural architecture. Finally, we set up a confidence measure to evaluate the reliability of a hypothesis produced by our system

Dissertation presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Knowledge Management and Business Intelligence<br>Neuroevolution is a field in which evolutionary algorithms are applied with the goal of evolving Artificial Neural Networks (ANNs). These evolutionary approaches can be used to evolve ANNs with fixed or dynamic topologies. This paper studies the Semantic Learning Machine (SLM) algorithm, a recently proposed neuroevolution method that searches over unimodal error landscapes in any supervised learning problem, where the error is measured as a distance to the known targets. SLM is compared with the topology-changing algorithm NeuroEvolution of Augmenting Topologies (NEAT) and with a fixed-topology neuroevolution approach. Experiments are performed on a total of 6 real-world datasets of classification and regression tasks. The results show that the best SLM variants outperform the other neuroevolution approaches in terms of generalization achieved, while also being more efficient in learning the training data. Further comparisons show that the best SLM variants also outperform the common ANN backpropagation-based approach under different topologies. A combination of the SLM with a recently proposed semantic stopping criterion also shows that it is possible to evolve competitive neural networks in a few seconds on the vast majority of the datasets considered.<br>Neuro evolução é uma área onde algoritmos evolucionários são aplicados com o objetivo de evoluir Artificial Neural Networks (ANN). Estas abordagens evolucionárias podem ser utilizadas para evoluir ANNs com topologias fixas ou dinâmicas. Este artigo estuda o algoritmo de Semantic Learning Machine (SLM), um método de neuro evolução proposto recentemente que percorre paisagens de erros unimodais em qualquer problema de aprendizagem supervisionada, onde o erro é medido como a distância com os alvos conhecidos previamente. SLM é comparado com o algoritmo de alteração de topologias NeuroEvolution of Augmenting Topologies (NEAT) e com uma abordagem neuro evolucionária de topologias fixas. Experiências são realizadas em 6 datasets reais de tarefas de regressão e classificação. Os resultados mostram que as melhores variantes de SLM são mais capazes de generalizar quando comparadas com outras abordagens de neuro evolução, ao mesmo tempo que são mais eficientes no processo de treino. Mais comparações mostram que as melhores variantes de SLM são mais eficazes que as abordagens mais comuns de treino de ANN usando diferentes topologias e retro propagação. A combinação de SLM com um critério semântico de paragem do processo de treino também mostra que é possível criar redes neuronais competitivas em poucos segundos, na maioria dos datasets considerados.

碩士<br>國立臺北教育大學<br>語文與創作學系華語文教學碩士班<br>103<br>This thesis studies the performance of Japanese CFL learners in Chinese kind measure word from semantic and syntactic aspects. In “Measure word +Noun” structure test, native speakers tend to use kind words “種(zhong)、類(lei)、樣(yang)”for concrete nouns than abstract nouns while the tendency is not obvious for Japanese CFL learner. In order to explain the different tendency, two more studies were included to see how the performance might change when the “Measure word + Noun” structures are used in sentences. The results from native speakers confirmed that different sentence structures played a role in which measure word is selected. The second study used the same six different sentence structures on Japanese CFL learners, and found that Japanese CFL learners did not have difficulties in the complete structures like “Numeral+ Measure word +Noun”,” Demonstrative pronouns+ Numeral+ Measure word +Noun” or ”Verb+Numeral+Measure word +Noun “structures. However, different patterns were found in the incomplete forms like “Ordinal Numbers+ Numeral+ Measure word +Noun”,” Demonstrative pronouns+ Measure word +Noun “and” Verb+ Measure word +Noun “structures, which highlighted the error performance of Japanese CFL learners. From these studies, we found unique error patterns in different sentence structures, which suggests that Mandarin teachers should teach all possible sentence structures when teaching Chinese measure words, before giving meanings or comparing synonyms, and that might reduce the error performances in the use of Chinese measure words.

Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics<br>Alignment in the error space is a recent idea to exploit semantic awareness in genetic programming. In a previous contribution, the concepts of optimally aligned and optimally coplanar individuals were introduced, and it was shown that given optimally aligned, or optimally coplanar, individuals, it is possible to construct a globally optimal solution analytically. Consequently, genetic programming methods, aimed at searching for optimally aligned, or optimally coplanar, individuals were introduced. This paper critically discusses those methods, analyzing their major limitations and introduces a new genetic programming system aimed at overcoming those limitations. The presented experimental results, conducted on five real-life symbolic regression problems, show that the proposed algorithms’ outperform not only the existing methods based on the concept of alignment in the error space, but also geometric semantic genetic programming and standard genetic programming.

碩士<br>淡江大學<br>資訊管理學系碩士班<br>95<br>Through the rapid development of internet, website contents also experiencing major changes in time and space. An important question is how to maintain the rationality, correctness and integrity of data. Most websites input data manually by the user, but this may usually result in incorrect data and thus the huge loss. Although it is not difficult to syntactically checking the correctness of the data on website, verifying the semantic meaning of the data involves complexity. Besides, it is also increasing the time and labor cost on maintaining website information. This research presents a semantic reasoning framework using expert system development program language Jess (Java Expert System Shell) for the reasoning under specific ontology. This allows the machine to understand the meaning of data and detect semantic errors of those data. This approach reduces unnecessary loss due to semantic error and decreasing the time and effort on checking data while promoting the work efficiency and business reputation.

This dissertation aims to identify and analyse the frequency distribution of mor- phological forms of nouns in contemporary Czech. The work is based on an error analysis of texts of non-native speakers revealing and describing frequent types of errors in noun inflection. Against the background of recent research in the field of second language acquisition on the one hand and the principles of communicative approaches to language teaching on the other hand nine textbooks of Czech as a foreign language are analysed for approaches to the introduction and presentation of grammatical cases, their functions and typical constructions. It is concluded that it is necessary to exemplify case functions by prototypical constructions and sufficient sample nouns for which these functions are typical. With regard to the often observed emphasis on the noun as a part of the inflectional paradigm and thus on the nominative singular as its basic form, it is eligible for textbooks and language classes to shift the focus from the dominant paradigmatic to the syntag- matic aspect of nouns and their word forms. One of the initial questions of these theses therefore is which forms of a noun are essential to the adequate (didactical- ly oriented) presentation of the noun and its use. A pilot study conducted on the data of...

碩士<br>長庚大學<br>職能治療學系<br>98<br>Most Chinese characters are phonograms and homophones resulting in the opportunity to increase the heterographic homophone errors while writing. However, the error type of heterographic homophone may also result from the lack of lexical and/or semantic knowledge. In this present study, we used the lexical-semantic processing strategy to enhance the lexical and semantic capacities and to decrease the heterographic homophone errors in children with writing difficulties. The writing theoretical models not only explain the overall and systematic writing processing, but also guide the assessment and intervention. However, this relationship between the theoretical models and clinical applications is usually ignored in clinics. We reviewed the alphabetic and Chinese writing models to form a dictation model used to explain the heterographic homophone errors. We referred to “treatment of the spelling impairment” proposed by Weekes & Coltheart (1996) and “copy and recall treatment” by Beeson (1999), and further developed “lexical-semantic processing strategy (semantic strategy)” and “lexical and logographeme output programming processing strategy (copying strategy)” respectively. In this present study, we compared the effectiveness of these two strategies on decreasing heterographic homophone errors. Because the semantic strategy could enhance lexical and semantic capacities directly, we expected the semantic strategy is better to decrease the heterographic homophone errors than the copying strategy. The vocabulary test with 145 items was developed by us to test 256 fourth- to sixth-grade students from different 9 classes in Taoyuan elementary schools. The test was performed in a group setting in each class, and the results of each grade were established as the norm. Besides, the writing problem screening questionnaire also developed by us was sent to the parents of fourth- to sixth-grade students and 565 questionnaires were returned. In order to recruit more students, we also put on posters at Chang Gung Memorial Hospital, Chang Gung community and website of Association of Learning Disabilities. The potential students were recruited from the aforementioned vocabulary test and the questionnaires. Forty-four potential students received the modified vocabulary test in an individual setting (by SuperLab Pro 2.0 software and the answer rate in subject’s pace). Finally, the 9 children who met the inclusion criteria of having both writing difficulties and heterographic homophone errors were recruited in this study. The other 9 children with normal writing abilities were also recruited to serve as the comparison basis while substantiating the characteristics of the aforementioned children. Most of these children (6/9) also met the criteria of having heterographic homophone errors. The data were collected in the end of the intervention week and on one day after the intervention week to study the learning and retention effects respectively. We combined ex post facto research and quasi-experimental research method and used the one-group pretest-posttest design and the concept of multiple treatment design to guide this study. All subjects received these two strategies which were arranged in a counterbalance order. The words used to train in two different strategic situations were different for the purpose to avoid the interference and carry-over effects between two strategic sessions. Moreover, for the purpose of considering the fairness, the words used in the two strategic situations were controlled at the same difficulty level. The intervention programs continued for 2 weeks totally and each strategy was given 3 times a week. We used Wilcoxon signed rank test to compare the effectiveness of these two strategies on heterographic homophone errors (α&amp;lt;.05). For children with writing difficulties, semantic strategy was more effective to decrease the heterographic homophone errors in the learning effect than the copying strategy, but not in the retention effect. The semantic strategy demonstrated the effect for children with writing difficulties to clarify meaning and orthographic structure even under the situation without repetitive practice. However, children with writing difficulties still could not master and discriminate the principles and differences clearly from these two strategies. Therefore, the aforementioned children were unable to formulate their self-strategies and retain the acquisition of words. Finally, this present study supports that the writing difficulties can be sourced and solved efficiently if the assessment and intervention are connected with contributed writing models.

The development of category knowledge is in part a function of one's experiences with the world. The types of errors produced during category generation tasks may reveal the boundaries of these experiences and the ways in which they are organized into lexical networks. Examining the errors made by bilingual children with and without language impairment (LI) and bilingual adults may help to distinguish the effects of ability versus experience on the development and organization of lexical-semantic categories. The purpose of this study was to examine the types of errors made by bilingual (Spanish-English) children with (n=37) and without (n=35) LI and bilingual adults (n=26) on category generation tasks in both their languages and at two category levels: taxonomic and slot-filler. Results revealed a main effect for level (taxonomic vs. slot-filler) and error type (semantic vs. other) and suggest that bilingual seven to nine-year-old children's and adults' proportions and types of errors produced on category generation tasks differ significantly based on ability (i.e., TD or LI) but not on experience (i.e., TD or Adults).<br>text

This diploma thesis is focused on collocations, i.e. repeatedly occurring combinations of lexical items. Main emphasis is put on lexical collocations produced incorrectly by Czech learners of English. The theoretical part of the thesis deals with various definitions describing this lexical phenomenon and with classification of collocations. A brief mention of the attitude to teaching collocations at schools is made, too. In the practical part, 200 collocations gathered from essays written by L2 learners of Czech secondary grammar schools are analysed as for a type of a restriction, semantic transparency and a part of speech. With erroneously produced collocation, an attempt to explain the cause leading to error is made and the correct combination or its closest equivalent is adduced. The results are presented in tables.

This report investigated the various types of memory intrusion errors of adults with language learning disability (LLD) in comparison to age- and gender-matched typically developing (TD) adults using lists that are specifically designed to induce memory intrusions adapted from Roediger and McDermott (1995) and modified by Watson et al. (2001; 2003). The 28 participants between the ages of 18:9 - 24:3 listened to pre-recorded lists of twelve words that converged on a critical lure either semantically, phonologically, or dually with a hybrid list. This report tested the hypotheses that 1) hybrid lists would be more likely to induce memory intrusions of the critical lure than either semantic or phonological lists for each group; 2) adults with LLD would demonstrate more intrusion errors than their TD counterparts; 3) the error profiles of the LLD and TD groups should be largely similar; however, the adults with LLD might show deficits in extracting the semantic gist of word lists in light of such patterns in children with specific language impairment (Sheng & McGregor, 2010a). Results showed that the hybrid lists induced the greatest number of critical lure intrusions producing a super-additive effect. Contrary to our hypothesis, the LLD group did not produce more memory intrusions than the TD group. The fact that the two groups performed similarly on all standardized measures suggests that the participants with LLD may have outgrown their disability. Results also revealed that interference and intrusions increased when there was an increase in phonological similarity among words for both groups. Lastly, our preliminary evidence suggests that adults with LLD are not as efficient as their TD counterparts at extracting the gist of semantically-related words. The inclusion of a greater number of participants may provide stronger support for the hypothesis that lexical-semantic organization is less efficient in young adults with LLD.<br>text

There exists a broad inclination among those who theorize about mental representation to assume that the meanings of linguistic units, like words, are going to be identical to, and work exactly like, mental representations, such as concepts. This has the effect of many theorists applying facts that seem to have been discovered about the meanings of linguistic units to mental representations. This is especially so for causal theories of content, which will be the primary exemplars here. It is the contention of this essay that this approach is mistaken. The influence of thinking about language and mental representation in this way has resulted in the adoption of certain positions by a broad swathe of theorists to the effect that the content of a concept is identical to the property in the world that the concept represents, and that because of this a concept only applies to an object in the world or it does not. The consequences of such commitments are what appear to be insoluble problems that arise when trying to account for, or explain, misrepresentation in cognitive systems. This essay presents the position that in order to actually account for misrepresentation, conceptual content must be understood as being very much like measurements, in that the application of a content to an object in the world is akin to measuring said object, and that conceptual content ought be understood as being graded in the same way that measurements are. On this view, then, concepts are the kinds of things that can be applied more, or less, accurately to particular objects in the world, and so are not identical to whatever it is that they represent.