Final Report Summary - WORDS (WORDS: Asymmetry, change and processing in phonological mental representation)
This project has investigated the representation of sounds to further the development of the Featurally Underspecified Lexicon (FUL) model of sound representation which aims to explain how we deal with the significant amount of variability in the speech sounds we hear. Since no speaker ever pronounces the same word identically, the decoding of speech and its mapping onto representations stored in the lexicon is a controversial area of research. The data collected throughout the project has substantiated the claims made by the FUL model that certain aspects of certain sounds are not stored fully which allows our speech recognition system to be more flexible when we hear sounds that might not match their representations completely. This asymmetrical storage also makes our system more efficient since there is less information to consider. The historical research which was conducted alongside the experimental part provided further insight into which sounds would be worth investigating by the way these were treated in early manuscripts and how they were affected by sound change. One very concrete result of this research is the implementation of a speech recognition engine which is based on the principles of FUL (and thus the research conducted throughout the project) and which is able to detect mispronunciations of, for example, language learners and provide feedback on the nature of their mistake.