Learning IS-A relations from specialized-domain texts with co-occurrence measures

Abstract

<p>Ontology  enrichment  is  a  classification  problem  in which  an  algorithm  categorizes  an  input conceptual unit  in the corresponding node  in a target ontology. Conceptual enrichment  is of great importance both to Knowledge Engineering and Natural Language Processing, because it helps maximize the efficacy of intelligent systems, making them more adaptable to scenarios where  information  is  produced  by  means  of  language.  Following  previous  research  on distributional  semantics,  this  paper  presents  a  case  study  of  ontology  enrichment  using  a feature-extraction  method  which  relies  on  collocational  information  from  corpora.  The  major advantage  of  this  method  is  that  it  can  help  locate  an  input  unit  within  its  corresponding superordinate node in a taxonomy using a relatively small number of lexical features. In order to  evaluate  the  proposed  framework,  this  paper  presents  an  experiment  consisting  of  the automatic classification of a chemical substance in a taxonomy of toxicology.</p>