Partial Disambiguation of Very Ambiguous Grammatical Words
Marc Hug · Journal of Quantitative Linguistics · 2000
Disambiguation of grammatical words is one of the problems that automated parsing of huge textual databases should deal with. But it is sometimes neglected because semantic content analysis may be performed - as it seems - without detailed knowledge of the grammatical relations within the texts. The French word que has as much as 10 different functional uses; the problem dealt with here is what part of the occurrences of que encountered in texts could be disambiguated by means of a micro-syntax - or local grammar. A first corpus of 2000 occurrences of que found in texts from the newspaper Le Monde was used do define and refine the micro-syntax designed to distribute occurrences of que between the 10 classes of use. Then another corpus of 1000 occurrences from Le Monde and the corpus of all occurrences found in a literary text (La Porte étroite, by A. Gide) were used to test the efficiency of the ‘final’ state of the micro-syntax. As expected, about 60 % of all occurrences can clearly be attributed to one of the 10 classes of use, but when one tryes to improve the micro-syntactic system, the number and complexity of the new rules will be greater and greater, and the whole system ends up not to be a micro-syntax any more, but a more or less complete surface grammar of French.