A Supervised learning for the identification of semantic relations in parallel enumerative structures (Apprentissage supervisé pour l’identification de relations sémantiques au sein de structures énumératives parallèles) [in French]

1938

Authors

nan

Abstract

A Supervised learning for the identification of semantic relations in parallel enumerative structures This work falls within the framework of ontology engineering and learning from encyclopedic or scientific texts. Our original contribution lies within the extraction of semantic relations expressed beyond the text linearity. To this end, we relied on the semantics behind the typo-dispositional characters whose function is to supplement the strictly linguistic formulations that could be more difficult to exploit. The work reported here is dealing with the semantic relations carried by the parallel enumerative structures. Although they display discontinuities between their various components, these enumerative structures form a whole at the semantic level. They are textual structures that are prone to hierarchic relations. After defining a typology of the relationships carried by this type of structure, we are proposing a learning approach aimed at their identification. Based on features including lexico-syntactic and typo-dispositional informations, the first results led an accuracy of 61.1%. MOTS-CLÉS : extraction de relations, structures énumératives parallèles, mise en forme maté- rielle, apprentissage supervisé, construction d’ontologies. KEYWORDS: relationship extraction, parallel enumerative structures, material shaping, supervi- sed learning, ontology learning. TALN-RÉCITAL 2013, 17-21 Juin, Les Sables d’Olonne 132 c￿ ATALA 1

Keywords

nan

Full Content

Apprentissage supervisé pour l’identification de relations sémantiques au sein de structures énumératives parallèles Jean-Philippe Fauconnier1 Mouna Kamel1 Bernard Rothenburger1 Nathalie Aussenac-Gilles1 (1) IRIT, 118 route de Narbonne 31060 Toulouse Cedex 5 {prénom}.{nom}@irit.fr RÉSUMÉ Ce travail s’inscrit dans le cadre de la construction et l’enrichissement d’ontologies à partir de textes de type encyclopédique ou scientifique. L’originalité de notre travail réside dans l’extraction de relations sémantiques exprimées au-delà de la linéarité du texte. Pour cela, nous nous appuyons sur la sémantique véhiculée par les caractères typo-dispositionels qui ont pour fonction de suppléer des formulations strictement linguistiques qui seraient plus difficilement exploitables. L’étude que nous proposons concerne les relations sémantiques portées par les structures énumératives parallèles qui, bien qu’affichant des discontinuités entre ses différents composants, présentent un tout sur le plan sémantique. Ce sont des structures textuelles qui sont propices aux relations hiérarchiques. Après avoir défini une typologie des relations portées par ce type de structure, nous proposons une approche par apprentissage visant à leur identification. Sur la base de traits incorporant informations lexico-syntaxiques et typo-dispositionnelles, les premiers résultats aboutissent à une exactitude de 61,1%. ABSTRACT A Supervised learning for the identification of semantic relations in parallel enumerative structures This work falls within the framework of ontology engineering and learning from encyclopedic or scientific texts. Our original contribution lies within the extraction of semantic relations expressed beyond the text linearity. To this end, we relied on the semantics behind the typo-dispositional characters whose function is to supplement the strictly linguistic formulations that could be more difficult to exploit. The work reported here is dealing with the semantic relations carried by the parallel enumerative structures. Although they display discontinuities between their various components, these enumerative structures form a whole at the semantic level. They are textual structures that are prone to hierarchic relations. After defining a typology of the relationships carried by this type of structure, we are proposing a learning approach aimed at their identification. Based on features including lexico-syntactic and typo-dispositional informations, the first results led an accuracy of 61.1%. MOTS-CLÉS : extraction de relations, structures énumératives parallèles, mise en forme maté- rielle, apprentissage supervisé, construction d’ontologies. KEYWORDS: relationship extraction, parallel enumerative structures, material shaping, supervi- sed learning, ontology learning. TALN-RÉCITAL 2013, 17-21 Juin, Les Sables d’Olonne 132 c￿ ATALA 1 Introduction La construction d’ontologies est un processus fastidieux qui nécessite la contribution d’experts d’un domaine. Une manière de rendre ce processus moins coûteux consiste à exploiter automati- quement certains types de textes, comme les textes de nature encyclopédique ou scientifique, afin d’en extraire les connaissances. Généralement, cette exploitation de textes s’appuie sur des analyses statistiques et/ou des analyses linguistiques essentiellement focalisées sur les niveaux lexicaux et syntaxiques. Citons notamment l’approche par apprentissage automatique (Nédellec et al., 2009), l’utilisation de patrons (Giuliano et al., 2006) ou encore une approche hybride combinant les deux (Giovannetti et al., 2008). Cependant, ces approches souffrent de deux limites : (1) l’analyse se situe en général à un niveau intraphrastique ou, du moins, textuellement linéaire et (2) …

Content truncated to 500 words

Labeled Text

A Supervised learning for the identification of semantic relations in parallel enumerative structures This work falls within the framework of ontology engineering and learning from encyclopedic or scientific texts. Our original contribution lies within the extraction of semantic relations expressed beyond the text linearity. To this end, we relied on the semantics behind the typo-dispositional characters whose function is to supplement the strictly linguistic formulations that could be more difficult to exploit. The work reported here is dealing with the semantic relations carried by the parallel enumerative structures. Although they display discontinuities between their various components, these enumerative structures form a whole at the semantic level. They are textual structures that are prone to hierarchic relations. After defining a typology of the relationships carried by this type of structure, we are proposing a learning approach aimed at their identification. Based on features including lexico-syntactic and typo-dispositional informations, the first results led an accuracy of 61.1%. MOTS-CLÉS : extraction de relations, structures énumératives parallèles, mise en forme matérielle, apprentissage supervisé, construction d’ontologies. KEYWORDS: relationship extraction, parallel enumerative structures, material shaping, supervised learning, ontology learning. TALN-RÉCITAL 2013, 17-21 Juin, Les Sables d’Olonne 132 c￿ ATALA 1

Publication Details

Paper ID:
1938
Year:
2013
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan