Alignment of phonemes with their corresponding orthography

Abstract

A technique has been developed for aligning the phonemes in a phonemic transcription of a word with the graphemes in its orthographic representation. For example, creationism /kri:eızam/ can be alighed as c r e a t ion i s m /k r i: ei ∫ n i z am/ Tables of phonemic-to-orthographic correspondences are given together with the frequency of occurrence of each correspondence in the texts constituting the Lancaster-Oslo/Bergen Corpus (LOB) (1978). The initial motivation for this algorithm was as an aid to the production of a computerized dictionary of English pronunciation. It has been used subsequently to extract from the dictionary all possible pronunciations of English prefixes, and is being used as part of the inferential process in the building of rules for the automated phonemic transcription of English. Uses are also seen in automatic speech recognition as part of the process of decoding the phonetic information extracted from the analysis of the acoustic signal. The algorithm has been checked using a phonemically tagged version of the LOB Corpus. The accent of English used in this work is Received Pronunciation (RP), see Gimson (1980).