Learning Phonotactics Using ILP

This paper describes experiments on learning Dutch phonotactic rules using Inductive Logic Programming, a machine learning discipline based on inductive logical operators. Two different ways of approaching the problem are experimented with, and compared against each other as well as with related work on the task. The results show a direct correspondence between the quality and informedness of the background knowledge and the constructed theory, demonstrating the ability of ILP to take good advantage of the prior domain knowledge available. Further research is outlined.

💡 Research Summary

The paper investigates the use of Inductive Logic Programming (ILP) for learning phonotactic constraints of Dutch monosyllabic words. The authors employ the ILP system Aleph, which implements the Progol algorithm, to induce Horn‑clause theories that describe which consonant clusters may appear before and after a vowel nucleus. The learning task is framed as a single‑predicate problem: two predicates, prefix/3 and suffix/3, encode permissible pre‑vocalic and post‑vocalic affixes respectively. Positive examples are derived from the CELEX Dutch lexical database (5,095 monosyllabic words), each broken into a prevocalic cluster, a vowel (or diphthong), and a postvocalic cluster. These are turned into logical facts such as prefix(m,