Representational Momentum May Explain Aspects of Vowel Shifts

We present a computational model of vowel shifts, applied in particular to the Northern Cities Vowel Shift. Our model in- corporates several empirically-derived rules of vowel change. The key aspect of this model is the use of representational momentum, which, we argue, explains multiple observed fea- tures of the shift. We compare our model with data on the Northern Cities Shift spanning more than a century and show that, when representational momentum is included, the re- sults of the model match the data well. In the present work, we develop an agent-based model of the mechanism of vowel change based on empirical observa- tions by sociolinguists over many years. These observations are incorporated into our model as a system of constraints and update rules that try to preserve these constraints. One key addition we make to the model is to include representa- tional momentum, which is a psychologically documented phenomenon wherein people's memories of sequences of tones ordered by pitch exhibit overshoot. We demonstrate that if we include momentum in the mechanism of individual vowel change, it can account for the long-term population- wide shift in vowels. The mechanism of vowel change in our model is a phe- nomenon known as accommodation, which is an attempt on the part of a hearer to adjust his vowel system to match the speaker's. We apply our model to data on the Northern Cities Vowel Shift (NCVS), and show that when representational momen- tum is included, the results of the model match the data well, while they do not when representational momentum is not included. The rest of this article is organized as follows. We first present a brief overview of the Northern Cities Shift and the data. Then we discuss prior efforts to model vowel shifts. After that we present our model and experiments with and without representational momentum. We do a statistical analysis of our results to show how they compare with the data, and we end with a discussion of aspects of the shift that are and aren't explained by our model.