Automatic Grapheme-to-Phoneme Translation of English

Abstract

In the design of a machine that scans printed-English text and generates the corresponding spoken output with virtually no restriction on the vocabulary size, a procedure for the translation of the graphemes to the phonemes is necessary. Translation technique using the letter context alone, such as is used in the phonic method of teaching English to children, is shown to be entirely unsatisfactory. Instead, by the use of morphographemic rules, written words are decomposed into their constituent morphs and the corresponding morphemes are looked up in a morph-morpheme lexicon. Morphophonemic rules are used to produce the output-phoneme string for the given word. Syntactic markers, stored as part of the lexicon entries, can be used to resolve certain ambiguities and to perform sentence parsing, which is necessary in the assignment of suprasegmental phoneme values. Engineering considerations in the implementation are discussed. [This work was supported by a grant from the National Institutes of Health and by a Joint Services Electronics Program Contract.]