Strand 5

Strand 5

Computational semantic analysis (8 WP) (resp. A. Nazarenko, LIPN, Paris 13)

In the field of computational linguistics, Strand 5 focuses on semantic analysis and its application on various content access tools.

The goal is to address various tasks (e.g. information extraction, indexation, summarizing), which may or may not require the construction of logical formulae, in the spirit of Montague. Models, computation methods and – in some cases – tools have been proposed to account for various semantic phenomena but those phenomena have often been considered in isolation and the computation models are heterogeneous. The challenge today consists in evaluating these semantic models, to make them operational and, above all, to integrate them so as to produce rich, covering and consistent semantic analyses, in the same way as past works have lead to robust and high-quality syntactic parsers.

The goal is to boost research on the integration of semantic models and methods, through the collaboration of three computational linguistic teams (Alpage, Lattice, LIPN, LPP-P3) and the confrontation of the heterogeneous results of the participants in parsing, textual analysis, corpus analysis and processing, semantic models, text-based inferential reasoning, knowledge acquisition, machine learning and content access tools.

Strand 5 mainly focuses on French, but specific attention will be paid to the identification of language independent methods, which can be used to process a wide diversity of languages as soon as resources are available. This is actually a promising approach to equip resource-scarce languages studied in Strands 2 and 3.

Strand 5 considers both written and spoken language. The applicative needs for speech processing are increasing and it is a big challenge to develop new methods taking into account oral features such as dysfluencies, in order to compensate for the relative low quality of semantic analysis of oral transcriptions.

Strand 5 comprises 6 sub-stands:

·       Historical and reflexive perspective (HP)

·       Specific studies of semantic processing (SSP)

·       Knowledge acquisition methods (KA)

·       Semi-supervised machine learning (SSL)

·       Integration approaches (IA)

·       Applications: Towards richer access to text content (APP)

List of ongoing research operations :

Historical and reflexive perspective (HP)

·       Meaning theories and Natural Language Processing (HP1, resp.: J. Léon)

Specific studies of semantic processing (SSP)

·       Traits spécifiques à l’oral pour le traitement sémantique (SSP2, resp. M. Adda-Decker)

·       Filtering of non discursive occurrences of connectives (SSP3, resp. L. Danlos)

·       Deep syntactic analysis (SSP4, resp. M. Candito)

Knowledge acquisition methods (KA)

·       Induction automatique de patrons lexico-grammaticaux représentatifs à partir de textes (KA2, resp. T. Poibeau)

·       Text-based knowledge acquisition (KA3, resp.: A. Nazarenko)

Semi-supervised machine learning (SSL)

·       Apport de l’analyse syntaxique à l’extraction d’information : adaptation au domaine et apprentissage multi-objectif (SSL3, resp. J. Le Roux)

Integration approaches (IA)

·       Annotation platform (IA1)

Applications: Towards richer access to text content (APP)

·       Design and development of new methods for accessing textual content (APP2, resp. H. Zargayouna)