How Can the Term Compositionality Be Useful for Acquiring Elementary Semantic Relations?
Résumé
Acquiring and enriching lexical resources is crucial for various areas of the computational linguistics applications, especially in specialized domains. In this paper, we propose a high-quality method exploiting the compositionality of complex terms issued from a structured terminology in order to infer three kinds of semantic relations (synonymy, hierarchical and meronymy) between words or terms. The approach has been applied and evaluated on the Gene Ontology biomedical terminology: 1,273 is-a, 178 part-of and 921 synonymy relations have been inferred and show a precision over 90%. We analyze these results and the possibility of their cross-validation through a graph representation.