Aller à l’une de ces dates :

 20 mars 2008, Martin Forst, Parc, Palo Alto (USA)
 3 avril 2008, Thierry Declerck, DFKI (Allemagne)
 15 mai 2008, François-Régis Chaumartin, U .Paris 7 et Proxem (France)
 5 juin 2008,Temis, Paris
 5 juin 2008,Anoop Sarkar, Simon Fraser University, Burnaby (Canada)

20 mars 2008 ; 14:00-15:00, Salle A006

Martin FORST , Parc, Palo Alto (USA) parlera de Computing Linguistically-based Textual Inferences

Résumé : Joint work with Danny Bobrow, Cleo Condoravdi, Lauri Karttunen,Tracy Holloway King, Valeria de Paiva, and Annie Zaenen

In this talk I give an overview and a demo of PARC’s Bridge system. The particular task that I focus on is entailment and contradiction detection (ECD), a more refined variant of the PASCAL RTE (Recognizing Textual Entailment) challenge. Given a passage of text and a query, does the query sentence follow from the text in the passage, is it contradicted by it, or neither ?

The entailment and contradiction detection algorithm operates on the level of Abstract Knowledge Structure (AKR) without the need of disambiguation. An AKR representation, derived from the syntactic and semantic analyses of a sentence, is a flat set of facts that involves concepts, roles, and contexts. Texts are parsed to produce packed f-structures and these are rewritten and canonicalized, without unpacking, into AKR. Canonicalization is determined both by the structure of the representations and the lexical items involved. The system includes knowledge about words and their relations between them that are encoded in resources such as WordNet and VerbNet. It also includes knowledge about lexically or constructionally triggered presuppositions and entailments.

The ECD process first aligns context and concept terms and then computes specificity relations between the aligned concept terms. Some special case reasoners support identification of named objects, comparison of specificity of WordNet synsets, and compatibility of cardinality restrictions. All the query facts that are entailed by the corresponding passage facts get removed. If no query facts remain, the system responds YES. A conflict in the instantiation claims of two aligned terms marks a contradiction. In this case the system responds NO. If some query facts remain at the end, the response is UNKNOWN. The linguistic phenomena I illustrate in this presentation include lexical entailments (kill => die), relations between lexical predicates or phrasal constructions and their embedded complements (forget that A => A, forget to A => not A, take the trouble to S => S, waste an opportunity to S => not S), and inferring temporal relations from temporal modifiers.

3 avril 2008 ; 14:00-15:00, Salle A006

Thierry Declerck , DFKI (Allemagne) parlera de Language Technology for the Semantic Annotation of Multimedia

Résumé : This lecture will address the topic of the so-called "semantic gap" in the analysis and generation of multimedia (MM) content. NLP and Semantic Web (SW) technologies can help on this in providing for ontology-based semantic annotation of textual documents (including speech transcripts) that are associated with video and images. We will describe some work dedicated to the description of an ontological framework that combines annotations provided by the combination of NLP and SW on the one side, and the so-called low level annotation features provided by MM analysis on the other side. The course will rely on the most recent advances in this field, as proposed by the EU Network of Excellence "K-Space" (Knowledge Space of semantic inference for automatic annotation and retrieval of multimedia content, see www.k-space.eu).

15 mai 2008 ; 14:00-15:00, Salle A006

François-Régis Chaumartin , U .Paris 7 et PROXEM (France) parlera de Antelope : une plateforme de traitement linguistique permettant l’extraction de connaissances.

Résumé : La plateforme de traitement linguistique Antelope permet l’analyse syntaxique et sémantique de textes sur des corpus de volume important. Antelope intègre plusieurs composants préexistants (pour l’analyse syntaxique) ainsi que des données linguistiques de large couverture provenant de différentes sources. Un effort d’intégration permet néanmoins d’offrir une plateforme homogène. Notre contribution directe concerne l’ajout de composants d’analyse sémantique, et la formalisation d’un modèle complet d’analyse de documents. Nous présenterons Antelope en la positionnant par rapport à d’autres plateformes, et en en soulignant les précautions architecturales à prendre pour qu’un tel ensemble complexe reste maintenable. Enfin, nous présenterons un projet en cours utilisant Antelope pour effectuer l’extraction de connaissances encyclopédiques."

5 juin 2008 ; 10:30-12:00, Salle B13

Sylvie Guillemin-Lanne , Paris présentera la société TEMIS

Résumé :
 La société et les profils
 Présentation des outils Temis (Mathieu Plantefol)
 Démonstration Luxid et Extractions d’informations CI LSE
 Présentation des savoir-faire Temis (Mathilde Wiss-Thebaut )
 Extraction d’information et Skill Cartridge

15 mai 2008 ; 14:00-15:00, Salle A006

Anoop Sarkar , Simon Fraser University, Burnaby (Canada) parlera de Lexicalized Tree-adjoining Grammar applied to Semantic Role Labelling

Résumé : Semantic Role Labelling (SRL) is a "shallow" semantic parsing task : to identify the arguments for a given predicate in a sentence. SRL is a natural extension of the syntactic parsing task since SRL aims to identify and label particular syntactic constituents in a parse tree with their semantic roles (agent, patient, etc.). Most SRL systems exploit syntactic trees as the main source of features. We would like to take this one step further and show that using Lexicalized Tree-adjoining Grammar (LTAG) derivation trees as an additional source of features can improve both argument identification and classification accuracy in SRL. LTAG seems to be a natural fit for the SRL task since semantic roles are typically localized to a single LTAG elementary tree that is lexicalized by the predicate. However, there are still significant challenges in obtaining accuracy better than state-of-the-art using LTAG-based features. In this talk, I will discuss the various ways that LTAG can be used for the SRL task using the PropBank corpus as the source of semantic role annotations. I will also discuss the recently released spinal-LTAG TreeBank which combines the Penn TreeBank with the PropBank, and how we can use this resource for the SRL task. I will present experiments that show that using LTAG can significantly improve accuracy on the SRL task over the current best SRL systems.