hide
Free keywords:
-
Abstract:
Fluency assessment is part of many official language tests (e.g., TOEFL iBT) which evaluate non-native speakers’ language proficiency. Operationalizing the notion of fluency involves disentangling the different factors that influence fluency judgments. One approach to this issue has been the correlational analysis of acoustic measures and subjective judgments: which disfluencies (pauses, fillers, corrections) play a large role in fluency assessment and which do not? This approach has mainly been concerned with non-native speech, since native speakers are commonly considered to be ‘fluent’ in their mother tongue. One example of this approach is a previous study by the authors in which they observed strong correlations between fluency ratings and the pause and speed characteristics of the L2 speech.
Disfluencies are, however, not limited to non-native speech: they also occur in native speech. It is as yet unclear whether there are any differences between the disfluencies of natives vs. non-natives, for instance in their relative contribution to fluency ratings. Therefore, the focus of this presentation lies on the perception of fluency in non-native and native speech. Crucially, instead of adopting a correlational approach, phonetic manipulations were applied to native and non-native speech such that causal relationships could be established between speech characteristics and fluency judgments.
In two experiments fluency judgments on native and non-native Dutch speech were collected. The stimuli consisted of phonetically manipulated speech: in Experiment 1 the number and duration of silent pauses were manipulated, in Experiment 2 the speed of the speech was altered. The manipulated speech samples were presented to native listeners, who rated them on fluency using a 9-point Likert scale. Linear Mixed Models revealed that (i) natives were rated higher than non-natives, (ii) increasing the number or the duration of silent pauses (Experiment 1) or slowing down the speech (Experiment 2) led to lower fluency judgments, and crucially, (iii) there was no difference in the effects of the manipulations across native and non-native speech. These results suggest that the contributions of pause and speed characteristics to fluency judgments are similar across native and non-native fluency perception. Therefore, human raters judge the fluency characteristics of native and non-native speakers according to the same principles.
The next step in our project is to study the online processing of disfluencies. Results from eye-tracking experiments using the Visual World Paradigm will be introduced that investigate the processing of native and non-native disfluencies.
