Catalogue of Artificial Intelligence Techniques
Constituent Likelihood Grammar
Keywords: Probabilistic techniques
Categories: Natural Language
Author(s): Geoffrey Sampson, Karen Sp\"arck Jones
Natural language analysis techniques based on generative grammars normally make a sharp distinction between grammatical and ungrammatical sequences and thus fail when exposed to authentic discourse. Probabilistic techniques for (i) syntactic category assignment and (ii) syntactic structure determination use only the relative frequency of constructions and so do not rule anything out as ungrammatical. The CLAWS system developed at Lancaster University for word tagging using tag pair transition frequencies derived from a large, heterogeneous text corpus achieves 95-96
- Atwell, E.S., Constituent-likelihood grammar., The Computational Analysis of English: a corpus-based approach
(Garside, R., Sampson, G. and Leech, G.
, eds.), Longman, London, 1987, pp.57--65.