Inductive Learning of Pronunciation Rules by hypothesis Testing and Correction

This paper describes a system that learns the ru les o f pronunciat ion i nduc t i ve l y . I t begins wi th a set of 26 ru les for s i n g l e l e t t e r pronunc ia t ion. Ind iv idua l words are presented to i t , and the system uses i t s ru le set to hypothesise a pronunc ia t ion . This is compared wi th a d ic t ionary pronunciat ion, and if any part of the pronunciat ion is incor rec t new ru les are created to handle the word as an exception cond i t ion . These ru les are checked for s i m i l a r i t y wi th others already produced, and where su i tab le a "genera l " rule is produced to deal wi th two or more created r u l e s . The e f fec t is to produce ru les that are more and more general , and these approach the general pronunciat ion ru le sets that have been produced manually by other workers.