Problems in the Difference-in-Distance measure of phonetic imitation

Abstract

Phonetic imitation elicited in studies where participants shadow model talkers is often quantified using the Difference-in-Distance (DID) metric, while a less commonly used method is linear combination. This study tests these two approaches for evidence of the Starting Distance Bias, where participants who are more distinct from the model talkers at baseline will be found to converge more than those with smaller starting distances. The two approaches are each applied to three datasets, one in which convergence was possible, an altered version in which it was not, and a control group in which no change at all is present. DID finds that greater starting distance gives rise to greater convergence in all cases, even in the datasets that lack convergence, suggesting that the starting distance effect is a bias, rather than a genuine behaviour of the participants. In contrast, linear combination found no such effect in any dataset. Graphical analysis illustrates that the Starting Distance Bias is introduced by DID because the range of possible DID values is restricted as a function of starting distance, whereas no such restriction applies in linear combination. This study suggests that future investigations of imitation should avoid using DID, especially when starting distance is likely to vary.