Final Report Summary - SCATS (Sound change and the acquisition of timing in speech)
The sounds of language are known to change over short and long time periods. An unresolved issue is how sound change emerges out of the way humans put their vocal apparatus and hearing systems to use for speech communication. The new approach in this project was to investigate this issue from the perspective of how children acquire speech in many different languages including Cantonese, German, Italian, and Polish. A general hypothesis was that sound change arises out of the sometimes perceptually ambiguous ways in which speech sounds overlap with each other in time and that such ambiguities are especially marked in children. An overall finding is that sound change may instead come about out because of both children's and adult's occasional difficulty in interpreting the way that speaking style varies depending on how predictable words are from context. A cognitively-inspired computational model of sound change was developed in which speakers, represented by computational agents, imitated each other's speech. The computational simulations showed that speech communication can occasionally bring about sound change depending on the degree and the direction in which speech sounds vary. There were various methodological innovations in the project. These included the first ever computational model in which the computational agents stored and exchanged dynamic speech signals from real speakers; new procedures for quantifying tongue movement by processing images of data captured from ultrasound; and a web-based system for creating, querying and analysing speech corpora. A current application of the project is the development of tools for the automatic diagnosis of speech impairments via the web.