Syntactic and Semantic Type and Selection. SPARKLE Deliverable 5.2

This document describes research on techniques for semantic annotation of subcategorization frames involving semantic classification of heads of phrases. In keeping with the Sparkle policy of exploring alternative approaches, the methods vary widely in statistical models using (i) the minimum description principle, (ii) analogy-based semantic similary, (iii) EM-based clustering, and (iv) Markov models induced by semantic class hierarchies