Constrained Lexical Attraction Models

All Papers

Lexical Attraction Models (LAMs) were first introduced by Deniz Yuret to exemplify how an algorithm can learn word dependencies from raw text. His general thesis is that lexical attraction is the likelihood of a syntactic relation. However, Yuret's lexical attraction acquisition algorithm does not take into account the morpho-syntactical information provided by a part-of-speech (POS) tagger and, thus, is unable to impose certain linguistically motivated restrictions on the creation of the links. Furthermore, it does not behave well when encountering unknown words. The present article presents a new link discovery algorithm using the annotation provided by a POS-tagger. The results show an F-measure of approximately 70% when comparing the links produced by this algorithm with those produced by a fully-fledged parser.