Inductive and deductive inferences in a Crowdsourced Lexical-Semantic Network

Inductive and deductive inferences in a Crowdsourced Lexical-Semantic Network In Computational Linguistics, validated lexical-semantic networks are crucial resources. Regardless the construction strategies used, automatically inferring new relations from already existing ones may improve coverage and global quality of the resource. In this context, an inference engine aims at producing new conclusions (i.e. potential relations) from premises (pre-existing relations). The approach we propose is based on a triangulation method involving the semantic transitivity with a blocking mechanism to avoid proposing dubious relations. Inferred relations are then proposed to contributors to be validated or rejected. In cas of invalidation, a reconciliation strategy is implemented to identify the cause of the erroneous inference : an exception, an error in the premises, or a confusion caused by polysemy. MOTS-CLÉS : inférence de relations, réconciliation, enrichissement, réseau lexical, peuplonomie.