Phonotactics in Inductive Logic Programming

Abstract

We examine the results of applying inductive logic programming (ILP) to a relatively simple linguistic task, that of recognizing monosyllables in one language. ILP is suited to linguistic problems given linguists’ preference for formulating their theories in discrete rules, and because of ILP’s ability to incorporate various background theories. But it turns out to be difficult to rival the performance that statistical theories achieve on the same task. Finally, we note that the theoretically preferred solutions are quite compact, but not optimally comprehensive. Perhaps this should better be interpreted as a reflection of what theoretical linguists prefer, rather than as a reflection of the learning technique.