Spoken dialogue systems provide a natural conversational interface to computer applications. In recent years, the substantial improvements in the performance of speech recognition engines have helped shift the research focus to the next component of the dialogue system pipeline: the one in charge of language understanding. The role of this module is to translate user inputs into accurate representations of the user goal in the form that can be used by the system to interact with the underlying application. The challenges include the modelling of linguistic variation, speech recognition errors and the effects of dialogue context. Recently, the focus of language understanding research has moved to making use of word embeddings induced from large textual corpora using unsupervised methods. The work presented in this thesis demonstrates how these methods can be adapted to overcome the limitations of language understanding pipelines currently used in spoken dialogue systems. The thesis starts with a discussion of the pros and cons of language understanding models used in modern dialogue systems. Most models in use today are based on the delexicalisation paradigm, where exact string matching supplemented by a list of domain-specific rephrasings is used to recognise users' intents and update the system's internal belief state. This is followed by an attempt to use pretrained word vector collections to automatically induce domain-specific semantic lexicons, which are typically hand-crafted to handle lexical variation and account for a plethora of system failure modes. The results highlight the deficiencies of distributional word vectors which must be overcome to make them useful for downstream language understanding models. The thesis next shifts focus to overcoming the language understanding models' dependency on semantic lexicons. To achieve that, the proposed Neural Belief Tracking (NBT) model forsakes the use of standard one-hot n-gram representations used in Natural Language Processing in favour of distributed representations of user utterances, dialogue context and domain ontologies. The NBT model makes use of external lexical knowledge embedded in semantically specialised word vectors, obviating the need for domain-specific semantic lexicons. Subsequent work focuses on semantic specialisation, presenting an efficient method for injecting external lexical knowledge into word vector spaces. The proposed Attract-Repel algorithm boosts the semantic content of existing word vectors while simultaneously inducing high-quality cross-lingual word vector spaces. Finally, NBT models powered by specialised cross-lingual word vectors are used to train multilingual belief tracking models. These models operate across many languages at once, providing an efficient method for bootstrapping language understanding models for lower-resource languages with limited training data.

Le but principal de la thèse de recherche est celui d’analyser les mécanismes cognitifs et linguistiques à la base des métaphores « d’invention et défigées » en utilisant une approche qui vise à mélanger la théorie de la métaphore conceptuelle de Lakoff et Johnson (1980 et 1999), la théorie de l’intégration conceptuelle de Turner et Fauconnier (2002) et la théorie de l’analogie de Monneret (2004 et 2014). Pour mettre à l’épreuve le fonctionnement de la complémentarité de ces trois modèles, nous avons choisi d’utiliser deux textes littéraires de langue allemande des années 1980-1990 qui sont «Kassandra» (1983) de Christa Wolf et «Minotaurus. Eine Ballade» (1985) de Friedrich Dürrenmatt. Pour identifier le degré de figurativité des métaphores à l’intérieur des textes trois méthodologies ont été mise en place : les interviews des locuteurs allemands ; la consultation des dictionnaires en langue allemande et une observation des éléments présents à l’intérieur de la métaphore à travers la méthodologie MIPVU (Steen et al., 2010). L’analyse nous permet de comprendre le fonctionnement des métaphores dans les textes et les symboles culturels utilisés par les écrivains. De plus, cette recherche vise à montrer l’usage des « frames » et des « blends » conceptuels qui structurent notre cognition. / The main aim of this PhD thesis is the analysis of the cognitive and linguistic mechanisms within the creative and delexicalised metaphors using a mixed approach of Lakoff and Johnson’s (1980 and 1999) theory of conceptual metaphor; the Fauconnier and Turner’s (2002) theory of blending and the Monneret’s (2004 and 2014) theory of analogy. To test the complementariness of these three models, we have chosen two German literary texts from 1980 to 1990 that are «Kassandra» (1983) by Christa Wolf and «Minotaurus: eine Ballade» (1985) by Friedrich Dürrenmatt. To identify the degree of figurativeness of the metaphors within these texts, we have used three methodologies: the interviews to German speakers; the consultation of the German language dictionaries and the observation of the elements within metaphors through the MIPVU methodology (Steen et al., 2010). The analysis allows us to understand how metaphors work in the different texts and how cultural symbols are used by authors. Additionally, this research aims to show how the use of conceptual «frames» and conceptual «blends» that structure our cognition. / Lo scopo centrale della tesi di ricerca è quello di analizzare i meccanismi cognitivi e linguistici alla base delle metafore «d’invenzione e delessicalizzate» utilizzando un approccio che miri a fondere la teoria della metafora concettuale di Lakoff e Johnson (1980 e 1999), la teoria dell’integrazione concettuale di Fauconnier e Turner (2002) e la teoria dell’analogia di Monneret (2004 e 2014). Per mettere alla prova il funzionamento della complementarietà di questi tre modelli, è stato scelto di utilizzare due testi letterari di lingua tedesca degli anni 1980-1990 che sono «Kassandra» (1983) di Christa Wolf e «Minotaurus. Eine Ballade» (1985) di Friedrich Dürrenmatt. Per identificare il grado di figuratività delle metafore all’interno dei testi sono state messe in atto tre metodologie: le interviste ai parlanti tedeschi; la consultazione dei dizionari in lingua tedesca e un’osservazione degli elementi presenti all’interno della metafora attraverso la metodologia MIPVU (Steen et al., 2010). L’analisi ci permette di comprendere il funzionamento delle metafore all’interno dei testi e come sono utilizzati i simboli culturali da parte dei singoli autori. In più, questa ricerca rappresenta, quindi, un tentativo di comprendere l’uso dei «frame» e dei «blend» concettuali che strutturano la nostra cognizione.