Automated syntactic text description enhancement : determination analysis

This paper seeks to make a pragmatic contribution to computer-assisted discourse analysis by showing how a syntactical description could be used to gain a deeper understanding of the texts. During the '80s, a large corpus (5,000 pages) of varied political discourses covering twenty five years (1934-1960) was collected and a double description of it was carried out. On the one hand, to deal with the occurrence of terms referring to the same notions in the lexicon, a set of 144 sociological categories was assigned to relevant expressions. For example, the following terms were tagged as 'financial notions' : bank, credit, savings, dividend, etc. On the other hand, a surface syntactical description of each sentence of the corpus was produced by means of a parser (Plante 1979) controlled by an heuristic strategy and programmed in Déredec (a LISP sub-language). The use of a syntactical description was founded on the hypothesis that it would reduce the random nature of the lexical distribution of the words in the texts. A lexicon of the clause subjects is more 'qualified' than a lexicon with no criteria. This particular description identifies three contextual dependency relations for every clause, namely: theme/rheme, determination and verbal arguments.