Novel-word pronunciation: A cross-language study

Abstract

In the case of a “novel word” absent from a text-to-speech system's pronouncing dictionary, traditional systems invoke context-dependent letter-to-phoneme rules to produce a pronunciation. A proposal in the psychological literature, however, is that human readers pronounce novel words not by using explicit rules, but by analogy with letter-to-phoneme patterns for words they already know. In this paper, a synthesos-by-analogy system is presented which is, accordingly, also a model of novel-word pronunciation by humans. It employs analogy in both orthographic and phonological domains and is applied here to the pronunciation of novel words in British (Received Pronunciation) English and German. In implementing the system, certain detailed questions were confronted which analogy theory is at present inadequately developed to answer. Thus, a major part of this work concerns the impact of implementational choices on performance, where this is defined as the ability of the system to produce pronunciations in line with those given by humans. The size and content of the lexical database on which any analogy system must be based are also considered. The better performing implementations produced useful results for both British English and German. However, best results for each of the two languages were obtained from rather different implementations.