The logic of the lexicon

Abstracts

• Introduction (Ch. Retoré) SLIDES
  
• Sylvain Salvati (INRIA & LABRI - Bordeaux)
  Simple account of basic lexical semantics
  We propose a model-based approach to lexical semantics that allows for a simple account of faithful and unfaithful co-predications. Back
  
• Tim van de Cruys (CNRS - IRIT - Toulouse)
  Word vectors, distributional similarity and applications
  In the course of the last two decades, significant progress has been made with regard to the automatic extraction of word meaning from large-scale text corpora. The most successful models of word meaning are based on distributional similarity, calculating the meaning of words according to the contexts in which those words appear. The first part of his talk provides a general overview of the algorithms and notions of context used to calculate semantic similarity. We will also look in some detail at dimensionality reduction, a technique that is able to reduce a large number of contexts to a limited number of latent semantic dimensions. The second part of this talk will then focus on a number of applications that make use of distributional similarity, such as the extraction of selectional preferences and word sense discrimination. Back
• Veronika Lux (CNRS - ATILF - Nancy)
  Introducing the French Lexical Network SLIDES
  We present the French Lexical Network or FLN [2, 1], a lexical resource that is currently being built at ATILF-CNRS by a team of lexicographers, along the principles of the Explanatory Combinatorial Lexicography [4, 3].
  We first present the macrostructure of the FLN: like WordNet and FrameNet, FLN is a network or, formally, a lexical graph but its structuring is mainly based on the system of Lexical Functions, which is part of the Meaning-Text theory [5].
  We then provide some information about our lexicographic methodology and tools : FLN builds on experience gathered along the Explanatory Combinatorial Dictionaries, the DiCo and Dicouebe databases, the Lexique Actif du Français dictionary and the DicoPop web pages [6].
  Last, we detail the microstructure of the FLN, ie. the structure of the de- scription attached to each linguistic unit which is a node in the network. There are six sections in this description. We focus on the section ”Lexical Functions”.
  [1] Nabil Gader, Veronika Lux-Pogodalla, and Alain Polguère. Hand-Crafting a Lexical Network With a Knowledge-Based Graph Editor. In Proceedings or the 3rd workshop on Cognitive Aspects of the Lexicon (CogALex-III) - COLING 2012, pages 109–126, Mumbai, 2012.
  [2] Veronika Lux-Pogodalla and Alain Polguère. Construction of a French Lexical Network : Methodological Issues. In Proceedings or the International Workshop on Lexical Resources (WoLeR 2011), ESSLLI, Ljubl- jana, 2011.
  [3] I. Mel’ˇcuk. Explanatory Combinatorial Dictionary. In Giandomenico Sica, editor, Open Problems in Linguistics and Lexicography, pages 225– 355. Polimetrica, Monza, 2006.
  [4] I. Mel’ˇcuk, A. Clas, and A. Polguère. Introduction `a la lexicologie explicative et combinatoire. Duculot, Paris/Louvain-la-Neuve, 1995.
  [5] Igor Mel’ˇcuk. Lexical Functions: A Tool for the Description of Lexical Relations in the Lexicon. In Leo Wanner, editor, Lexical Functions in Lexicography and Natural Language Processing, volume 31 of Language Companion Series, pages 37–102. Benjamins, Amsterdam/Philadelphia, 1996.
  [6] Alain Polguère. Like a Lexicographer Weaving Her Lexical Network. In
  Proceedings or the 3rd workshop on Cognitive Aspects of the Lexicon (CogALex-III) - COLING 2012, pages 1–4, Mumbai, 2012.Back
• Mathieu Lafourcade & Manel Zarrouk (U. Montpellier 2 - LIRMM)
  Deductive and inductive inferences from JeuxDeMots, a game-acquired lexicon  SLIDES
  (the talk will be in english) La construction et la validation des réseaux lexicaux sémantiques est un enjeu majeur en TAL. Indépendamment des stratégies de construction utilisées, inférer automatiquement de nouvelles relations à partir de celles déjà existantes est une approche possible pour améliorer la couverture et la qualité globale de la ressource. Dans ce contexte, un moteur d'inférences a pour objet de formuler de nouvelles conclusions (c'est-à-dire des relations entre les termes) à partir de prémisses (des relations préexistantes). Dans cette présentation, nous présentons un moteur d'inférences pour le réseau lexical JeuxDeMots qui contient des termes et des relations typées pondérées entre ces termes. Dans le projet JeuxDeMots, le réseau
  lexical est construit à l'aide d'un \emph{GWAP} (Game With A Purpose) et quelques milliers de joueurs. Les termes polysémiques peuvent être potentiellement raffinés en divers usages (le terme figure peut faire référence à figure>géométrie ou figure>visage). Mais étant donné que le réseau est indéfiniment en construction certains sens peuvent être manquants. L'approche que nous proposons est fondée sur une méthode de triangulation impliquant la transitivité sémantique avec un mécanisme de blocage pour éviter de proposer des relations douteuses. Les relations inférées par déduction (du général vers les spécifiques) et par indiction (des spécifiques vers le général) sont proposées aux contributeurs pour être validées. Dans le cas d'invalidation, une stratégie de réconciliation est engagée pour identifier la cause de l'inférence erronée : une exception, une erreur dans les prémisses, ou une confusion d'usage causée par la polysémie. Back
• Mickael Zock (CNRS - LIF - Marseille)
  How to help authors to overcome the Tip-Of-the-Tongue problem :
  lexical graphs (or associative networks) and some inherent problems SLIDES
  One of the most vexing problems in speaking or writing is failing to access a particular word eventhough it is clearly stored in the user's brain. This kind of search failure known as dysnomia or Tip of the Tongue-problem (TOT),1 occurs not only in language, but also in other activities of everyday life. It is basically a search- and index-problem which we are reminded of when we look for something that exists in real world or in our mind (keys, glasses, people's names), but which we are unable to locate, access or retrieve in time. [The TOT-problem is characterized by the fact that the author (speaker/writer) has only partial access to the word s/he is looking for. The typically lacking parts are phonological (syllables, phonemes). Since all information except this last one seems to be available, and since this is the one preceding articulation, we say: the word is stuck on the tip of the tongue. ]
  Word finding problems are generally dealt with via a lexicon. Our concern here is language production, a task still too much neglected in lexicographical work. When speaking or writing, authors typically start from meanings (set of concepts). This is the normal route and it works in most cases, yet sometimes it does not, in which case we may resort to alternative strategies: we start from a word in a foreign language (translation), we perform topic-based search (thesaurus), or we rely on associations, i.e. lexical (synonyms, antonyms, ...) or conceptual relations. These latter two strategies are illustrated via lexical graphs —(for example, WordNet [4], JeuxDeMots [2], BabelNet [5], etc.)— which are very precious resources. Yet, given their size and complexity (multi-dimensional space) they also pose certain problems. Hence, in order to be truly useful for authors (allowing them to stay on top of the data, i.e. words), care must be taken to minimize noise (drowning the user with
  irrelevant data) and to set sign-posts, i.e. lables or orientational clues, as otherwise the user does not know which direction to go: every node (word) having a great number of outgoing links and nodes (associated terms). These just mentionned probems arise whenever a word is connected to a large number of words (or associated terms), which is usually the case, and whenever the target word is not an immediate neighbour of the query- or source-word (a quite frequent case). Since this talk will be preceded and followed by two related presentations, I will sketch a framework allowing to integrate and situate all this work. In particular, I will try to answer the question whether the mental lexicon [1, 5] can be used as a blueprint for the dictionaries of tomorrow.
  1. Aitchison, J. (2003). (2003). Words in the Mind: an Introduction to the Mental Lexicon. Oxford, Blackwell
  2. Lafourcade, M. (2007). Making people play for Lexical Acquisition with the JeuxDeMots prototype. In 7th International Symposium on Natural Language Processing, Pattaya, Chonburi, Thailand.
  3. Levelt W., Roelofs A. et Meyer, A. (1999). A theory of lexical access in speech production. Behavioral and Brain Sciences, 22, 1-75
  4. Miller, G. (ed.) (1990). WordNet: An On-Line Lexical Data-base. International Journal of Lexicography, 3(4), 235- 312.
  5. Navigli R. and S. P Ponzetto. 2012. BabelNet: The Automatic Construction, Evaluation and Application of a Wide- Coverage Multilingual Semantic Network. Artificial Intelligence, 193, Elsevier, pp. 217-250.
  Back
• Bruno Gaume (CNRS - CLLE - Toulouse)
  A robust metrology of lexical networks based on random walks in lexical relations 
  (the talk will be in English) Les grands  réseaux de terrains sont les  réseaux que l'on trouve en pratique, ils sont construits à partir de données issues de différents domaines d'études : La sociologie comme le réseau d'amis de Facebook, la linguistique comme les réseaux de synonymie, la webologie comme le réseau des pages web. Plusieurs études montrent un fait remarquable qui est que tous ces  réseaux, pourtant d'origines si diverses, possèdent des propriétés identiques bien particulières et font partie de la classe des Réseaux Petits Mondes Hiérarchiques (RPMH). Un autre fait tout aussi remarquable est que cette classe des RPMH est très petite au regard de l'ensemble des  réseaux possibles : la probabilité de tirer au hasard parmi l'ensemble des  réseaux possibles un RPMH est très proche de zéro. C'est-à-dire que les  réseaux auxquels nous avons à faire dans la vraie vie se ressemblent tous par leurs structures communes, bien qu'intrinsèquement cette structure soit très rare du point de vue de la théorie de la mesure.
  L'étude des RPMH mobilisent un grand nombre de chercheurs dans le monde, en effet l'étude et la modélisation des RPMH restent un champ de recherche ouvert et très prometteur pour une meilleure compréhension des phénomènes sous-jacents et pour une meilleure exploitation des données dans de nombreux domaines.
  Dans cet exposé je commencerais par présenter les quatre propriétés fondamentales omniprésentes dans les réseaux lexicaux.
  Je présenterai ensuite la dynamique des trajets d'un marcheur qui se déplace aléatoirement sur les sommets d'un réseau lexical à travers les relations lexicales. Nous verrons que l'étude de ces dynamiques permet de définir des outils de métrologie lexicale.
  je présenterais ensuite cinq applications de ces métrologies :
      1) R2SW : Une modélisation réaliste des RPMH qui permet générer artificiellement à partir de réseaux aléatoires, des réseaux possédant les mêmes propriétés que les réseaux lexicaux [1]
      2) SMAC : Une méthode pour comparer la structure de deux réseaux lexicaux [6]
      3) WISIGOTH : une méthode d'enrichissement endogène des réseaux lexicaux [2]
      4) SLAM : un modèle de Solution Lexicale Automatique de Métaphores analogique [3] http://erss.irit.fr/demetweb/Metaphors/slam.php
      5) APPROX : Une modélisation de la dynamique d'acquisition du lexique par les jeunes enfants [4], [5]
  [1] Gaume B, Mathieu F, Navarro E (2010) Building Real-World Complex Networks by Wandering on Random Graphs. In I3 Information Interaction Intelligence, 2010, Vol 10, Num 1
  [2] Sajous F, Navarro E, Gaume B, Prévot L,Chudy Y (2010) Semi-automatic Endogenous Enrichment of Collaboratively Constructed Lexical Resources: Piggybacking onto Wiktionary. In Advances in Natural Language Processing, Lecture Notes in Computer Science vol. 6233, pp. 332--344
  [3] Desalle, Y , Gaume B, Duvignau K. (2009) SLAM : Solution Lexicale Automatique pour Métaphore. In revue TAL 2009 vol 50 num1.
  [4] Gaume B., Duvignau K., Prevot L., Desalle Y. (2008) Toward a cognitive organization for electronic dictionaries, the case for semantic proxemy. In GOGALEX Cognitive Aspects of the Lexicon COLLING, Manchester 2008
  [5] Desalle Y, Hsieh S-K, Gaume B, and Cheung H (2010) Towards an Automatic Measurement of Verbal Lexicon Acquisition: The Case for a Young Children-versus-Adults Classification in French and Mandarin. In Proceedings of the 2010 PACLIC 24 : Workshop on Model and Measurement of Meaning (M3) pp. 809-818, Tohoku University, Sendai, Japan
  [6] Emmanuel Navarro, Bruno Gaume, Henri Prade (2012) "Comparing and Fusing Terrain Network Information", In Scalable Uncertainty Management - 6th International Conference, SUM 2012, Marburg, Germany (Eyke Hüllermeier, Sebastian Link, Thomas Fober, Bernhard Seeger, eds.), Springer, vol. 7520, pp. 459-472, 2012.  Back
• Nathalie Aussenac-Gilles (CNRS - IRIT - Toulouse)
  Ontologies, texts and the lexicon SLIDES
  Back
• Alain Lecomte (U. Paris 8 & SFL)
  Ludics and the lexicon
  Ludics provides a game-theoretic approach to language, based on the interaction of processes (also called « designs »). In this talk, we are viewing lexical items as such processes (the actions of which are semes). Admissible combinations of items lead to convergent interactions between these processes. The framework allows to recast notions previously introduced in Structural Semantics (Pottier, Greimas, Rastier) and it provides flexible tools suitable for the description of metaphor and the influence of context. Back
• Christophe Fouqueré (U. Paris 13 & LIPN)
  About Coherence Use in Natural Language SLIDES
  What does a linguistic concept denote? What kind of coherence is there between such denotations? What are the consequences of using types/formulas? Lecomte and Quatrini's works on the use of Ludics in Natural Language shed new light on that subject. We reconsider the previous questions studying how and in which extent coherence defined in Linear Logic (coherence spaces, Ludics) may be helpful. Back
• Robin Cooper (U. Göteborg)
  Types, judgements and lexical meaning SLIDES
  The classical (model-theoretic) view of lexical meaning is based on truth and normally makes the simplifying assumption that the meaningof a word is fixed in the lexicon.  We will propose that lexical meaning is based on judgement (in something like the type-theoretical sense) and that word meaning is in a constant state of flux as we communicate with each other.
  We will start by considering predicates of personal taste like 'delicious' and argue that statements such as 'this soup is delicious' are based on a notion of judgement by an agent rather than on an objective fact such as 'this soup' is in the extension of 'delicious'.  This is not to rule out cases where judgements are related to objective facts such as 'the temperature of this soup is 40 degrees'.  Some judgements correspond to objective facts and some do
  not.
  What I call 'delicious' you may call 'mediocre' or even 'disgusting'. Does this mean that you and I have different meanings for the word 'delicious' or merely that we like different food? This is a variant of the classical question of where to draw the boundaries between lexical meaning and encyclopaedic or world knowledge.  I will argue that there is no way to draw a clear boundary and that a type theoretic approach combined with a theory of semantic coordination between speakers suggests that we should not try. Back
• Nicholas Asher (CNRS - IRIT - Toulouse)
  Word meaning in context: semantic approaches based on type theory Back
• Zhaohui Luo (Royal Holloway College, U. of London) 
  Modern Type Theories and Montague Semantics: Comparisons and Beyond  SLIDES
  Back
• Christian Retoré (U. Bordeaux 1 & IRIT - Toulouse)
  Quantifiers in a type theoretical model of compositional semantics and lexical pragmatics SLIDES
  We shall firstly present a type theoretical model for handling the adaptation of word meaning to the context. This model resembles the ones proposed by Asher and Luo, the main difference being that terms, rather than types, trigger meaning adaptation. Secondly, we shall present a treatment of quantifiers and generalised quantifiers in such a richly typed model. Our solution is an adaptation of Hilbert operators to type theory, which seem better suited than usual approaches to quantification, both from a syntactical and a lexical viewpoint. Back
• Tim van de Cruys (CNRS - IRIT - Toulouse)
  On the convergence between compositional semantics and distributional semantics:
  A Tensor-based Factorization Model of Semantic Compositionality
  In this presentation, a novel method for the computation of compositionality within a distributional framework is presented. The key idea is that compositionality is modeled as a multi-way interaction between latent factors, which are automatically constructed from corpus data. The method is used for the composition of subject-verb-object combinations. First, a latent factor model for both nouns and verbs is computed from standard co-occurrence data. Next, the latent factors are used to induce a latent model of three-way subject-verb-object interactions. The model has been evaluated on a similarity task for transitive phrases, in which it exceeds the state of the art.
  Back
• Laurence Danlos (U. Paris 7 & INRIA)
  Problems at the syntax-semantics interface with adverbial connectives SLIDES
  We will put forward some cases where the host syntactic sentence of an adverbial connective is not identical to its second semantic argument. Such cases raise problem for the syntax-semantics interface. The extension of such a problematic phenomenon will be discussed. Back
• Richard Moot  (CNRS - LaBRI - Bordeaux)
  Wide-coverage semantics applied to temporal ordering and presupposition  SLIDES
  Back
• Reinhard Muskens (U. Tilburg) 
Names SLIDES
In this talk I will propose a theory of ordinary names that does not run into substitutivity problems but does retain Saul Kripke's intuition that names denote rigidly, i.e. have the same extension in any world where they denote. Names in this theory are /predicates/ as in Quine (1948),
Quine (1960), Burge (1973), Muskens (1995), Matushansky (2006), and Fara (2011). Since the theory is based on a truly intensional logic, codesignating names like 'Cicero' and 'Tully' can have equal extensions in all possible worlds without having the same meaning. So, while the theory accepts that names denote rigidly, it rejects Millianism, the idea that the meaning of a name is just its bearer. Back