A Method to Extract Social Representations from Linguistic Corpora

A new method id presented, which enables extracting the pattern of social representations of an object from corpora in natural language“about”this object. Interrogation of a source of common knowledge (a representative sample of a population, a dictionary, a set of articles or books), yields a corpus of linguistic statements concerning the object. The corpus is then processed with a software that breaks up the corpus into statements (e. g.: sentences), and then makes a classification of those statements, on the basis of co-occurrence of lexical traits. Each class is considered as a basic nucleus of the representation, characterised by typical traits. Demonstration of the method is presented on two corpuses about“eating”, (1) coming from a survey using the free association technique on a 2000 sample representative of the French population, (2) 544 definitions of synonyms and analogues of“to eat”from a large dictionary.