Semi-automatic Learning of Two-Level Phonological Rules for Agentive Nouns

The subject of the research presented in this paper is to build a system that will automatically obtain the two-level rules for agentive nouns in two languages: English and Macedonian. A set of pairs base_word - derived_agentive noun is constructed and for all these pairs a string edit sequence is built using the insert, delete and replace operations. Using the resulting sequences, a minimal acyclic finite state automaton is constructed, which has only one start and one final state. From this automaton, the two-level phonological rules are obtained. These rules are extended to improve some of their limitations. In total, seventeen two-level rules are implemented to the English agentive nouns, as well as nine two-level rules for the Macedonian agentive nouns.