Training Strategies for Improving Listeners' Comprehension of Foreign-Accented Speech

Holly Krech Thomas

PhD Dissertation Committee Members: Co-advisors - Lise Menn, Alice F. Healy (Psychology) Members - David Rood, Lyle E. Bourne, Jr. (Psychology), Dan Jurafsky

ABSTRACT. Understanding speech might appear to be a simple matter of bottom-up processing that involves identifying the sections of the speech signal which correspond to given phonemes and words, but variability makes the equation far from straightforward (e.g., Fowler, 1986; Liberman & Mattingly, 1985; Stevens & Blumstein, 1981). Due to coarticulatory effects, rate changes, accents, and differences in vocal tracts, phonemes and words are not uttered exactly the same way each time they are spoken. Work by Warren and his colleagues (e.g., Bashford & Warren, 1987; Warren & Obusek, 1971; Warren, 1999) suggests that listeners are usually able to comprehend the message of speech even when some of the speech segments are occluded or absent. Listeners restore fragments of the speech signal which are obliterated by noise, invoking higher level linguistic knowledge of syntax and semantics to predict the missing segments. Like bottom-up processing, this top-down processing is often unconscious and automatic.

Foreign-accented speech can be difficult for listeners to understand, but familiarity with the topic of speech or with accented speech in general can facilitate comprehension (Gass & Varonis, 1984). The present research examines training tasks which familiarized listeners to extemporaneous, foreign-accented speech. These tasks entailed active participation by the listener and implicit practice of either top-down processes or bottom-up processes in speech comprehension.

The first training task, paraphrasing the accented speech, encouraged a focus on meaning and led to practice of top-down processes, as evidenced by a preliminary focus-of-attention experiment. The second task, imitating the speech, promoted attention to the accented pronunciation and allowed practice of bottom-up processes, as supported by the focus-of-attention experiment and by increased ability to distinguish between accent types after training. The training experiments showed an improvement in comprehension after about an hour of training. The imitation task primarily improved understanding of relatively short, decontextualized utterances, which was measured by exact-word transcriptions of sentences 12-15 words long. The transcription sentences offered limited semantic and syntactic cues. This suggests that attention to accented pronunciation helps with perception of discrete speech segments. In contrast, the paraphrase task tended to improve comprehension of longer speech samples, measured by written recall and multiple-choice comprehension tests of 2-3 minute passages. This improvement implies that attention to meaning may be more beneficial for comprehension of conversational speech.

Therefore, by encouraging active, implicit learning to focus listeners' attention on different aspects of speech, training differentially influenced the automatic processes involved in speech comprehension. Training listeners to attend to the pronunciation of speech and practice bottom-up processes does not seem to enable comprehension of contextualized speech as effectively as practice of top-down processes. Because speech generally entails contextualized utterances, whether in casual conversations or classroom lectures, emphasizing top-down processes in training may be the best approach to enabling better comprehension of accented speech.

Holly Krech Thomas defended her dissertation in May and will be graduating with her PhD in August 2004 from the Department of Linguistics at the University of Colorado. She can be reached at Holly.Krech@Colorado.EDU.

Colorado Research in Linguistics - Volume 17, Issue 1 - June 2004