The relationship between acoustic features of second language speech and listener evaluation of speech quality

Mengxi Lin, Purdue University

Abstract

Second language (L2) speech is typically less fluent than native speech, and differs from it phonetically. While the speech of some L2 English speakers seems to be easily understood by native listeners despite the presence of a foreign accent, other L2 speech seems to be more demanding, such that listeners must expend considerable effort in order to understand it. One reason for this increased difficulty may simply be the speaker’s pronunciation accuracy or phonetic intelligibility. If a L2 speaker’s pronunciations of English sounds differ sufficiently from the sounds that native listeners expect, these differences may force native listeners to work much harder to understand the divergent speech patterns. However, L2 speakers also tend to differ from native ones in terms of fluency—the degree to which a speaker is able to produce appropriately structured phrases without unnecessary pauses, self-corrections or restarts. Previous studies have shown that measures of fluency are strongly predictive of listeners’ subjective ratings of the acceptability of L2 speech: Less fluent speech is consistently considered less acceptable (Ginther, Dimova, & Yang, 2010). However, since less fluent speakers tend also to have less accurate pronunciations, it is unclear whether or how these factors might interact to influence the amount of effort listeners exert to understand L2 speech, nor is it clear how listening effort might relate to perceived quality or acceptability of speech. In this dissertation, two experiments were designed to investigate these questions. The first experiment was designed to examine the specific acoustic features that have the greatest impact on listeners’ evaluations of L2 speech quality. The speech of twenty L2 speakers of English varying in proficiency (high and intermediate) and native language (Chinese and Korean) was evaluated by native listeners of American English. Subjective measures (listening effort, acceptability and intelligibility) were compared to the objective measure of word intelligibility, and to acoustic measures of fluency and pronunciation. Results showed that listening effort, acceptability and subjective intelligibility were highly related to one another and to word intelligibility, and were most strongly predicted by a set of fluency measures, including speech time ratio, speech rate, mean syllables per run, silent pause number, and silent pause time. Segmental and suprasegmental acoustic-phonetic properties did not predict subjective speech quality. These results suggested that fluency may effectively differentiate proficiency levels among relatively advanced L2 learners. The second experiment was designed to address the question of whether increasing fluency may reduce listening effort and improve the perceived intelligibility and acceptability of L2 speech when phonetic pronunciation remains constant. To this end, the fluency of the intermediate-proficiency L2 English speech samples used in the first experiment was increased by removing all non-juncture silent and filled pauses. The original and manipulated speech samples, as well as the high-proficiency L2 English speech samples, were evaluated by native American English listeners in terms of listening effort, intelligibility, and acceptability. These subjective measures were analyzed in relation to listeners’ individual working memory capacity. Results show that the manipulated speech received significantly higher ratings on all three measures compared to the original intermediate-proficiency speech, and was rated as similarly intelligible and acceptable as the high-proficiency speech samples. It was also demonstrated that listeners of relatively higher working memory capacity expended significantly less effort for processing all speech types and perceived them to be more intelligible than did listeners with lower working memory capacity. These results suggest substantial cognitive benefit of improved fluency on listeners’ perception of L2 speech. Overall, this study suggests that level of L2 fluency plays an important role in predicting listeners’ subjective ratings, possibly due to the manner in which fluency modulates listening effort through working memory capacity. These findings further enhance our understanding of the relationship between L2 speech fluency and intelligibility, and will have a direct impact on L2 instruction and assessment.

Degree

Ph.D.

Advisors

Francis, Purdue University.

Subject Area

Linguistics

COinS