This paper describes a procedure based on attachment resolution for evaluating an unsupervised strategy to acquire both nominal and verbal subcategorisation information. The notion of subcategorisation is based on two specific linguistic assumptions. First, it is assumed that two interdependent words impose requirements on each other. Second, it is also claimed that a linguistic requirement may be extensionally defined as the set of words that may occur in similar syntactic contexts. The main aim of the learning strategy is to cluster similar syntactic contexts by identifying the words that define extensionally their linguistic requirements. The subcategorisation information acquired is used to constrain attachments in a parsing task. Evaluation method of this task is described in the article.
Language Technology for Portuguese: shallow processing tools and resources
