Automatic Alignment and Analysis of Linguistic Change.

University Of Pennsylvania, Philadelphia PA

Investigators

Abstract

The Atlas of North American English, based on the study of cities over 50,000 population, shows that American regional dialects are becoming increasingly different as a result of ongoing changes in their vowel systems, leading to considerable miscomprehension across dialects. An understanding of this process calls for studies of many more communities and speakers. However, currently used methods of analysis require several days or weeks to accurately define the vowel system of a single speaker. This project is based on an accelerated method that produces ten times as much data in one hundredth of the time required, using automatic alignment of the sound wave and transcriptions. Methods will be developed for automatically measuring each vowel in the transcript with an error rate equal to or lower than that of manual measurement. These techniques will be tested on a very large database on change in the speech of a metropolitan community, a series of interviews recorded annually in Philadelphia neighborhoods from 1973 to 2008. Dialect diversity is one of the major factors limiting the success of automatic speech recognition and provides a major challenge for our understanding of the causes of language change. The methods developed in this project will serve to accelerate sociolinguistic analysis across other communities and other languages. Findings from the analysis of the Philadelphia database will contribute to the understanding of gender differentiation in language change in progress, the reversal of the direction of change, and the diffusion of change across neighborhoods.