表題番号：2011B-152
日付：2012/04/12
研究課題第二言語習得における躊躇現象
| 研究者所属（当時） | 資格 | 氏名 | |
|---|---|---|---|
| （代表者） | 理工学術院 | 准教授 | Ｒｏｓｅ Ｒａｌｐｈ |
- 研究成果概要
- One of the many linguistic needs of second language learners is to learn the temporal patterns of the target language. That is, in addition to developing grammatical accuracy and broad lexical knowledge, learners must learn how and when to pause appropriately and in a native-like manner in order to fully master their second language productive skills. Hesitation phenomena --- including silent pauses, filled pauses (uh, um), repairs, repeats, lengthenings, and so on --- are a key overt feature of speech and can provide cues to a learner's mastery of the temporal patterns of a target language. Thus, a close examination of HP is useful to understand how learners develop and where they need to develop.
This research project has therefore focused on the creation, annotation, and analysis of a corpus of first and second language speech called the Crosslinguistic Corpus of Hesitation Phenomena (CCHP) as part of a larger effort to understand the second language learner's developmental process. In principle, the project was performed as a pilot project in order to test the elicitation methods, annotation procedures, and overall design of the corpus in preparation for a larger scale corpus-building effort. However, even with its small-scale nature, the CCHP has yielded some interesting and novel findings.
The CCHP currently consists of elicited recordings from ten native speakers of Japanese. Recordings were elicited in their native language (L1) as well as in a second language (L2), English. Participants were asked to perform parallel elicitation tasks in each language, varying from read speech to spontanous speech. Samples from these recordings were transcribed and annotated with respect to word and pause intervals and hesitation phenomena. Acoustic measurements of duration and first and second formant (F1,F2) values of filled pauses were also taken. Finally, native English speakers (experienced English as a foreign language teachers) rated the English speech samples for degree of native/nonnative accent and fluency.
Analysis of the corpus data shows that the corpus is largely consistent with previous research of a similar nature. Results show that several features correlate well with second language proficiency. These are described one by one in greater detail in the following paragraphs.
Speech rate refers to the number of tokens (i.e., words) per minute. Previous research shows that as higher level L2 speakers have a higher speech rate in their L2. The results from the CCHP are consistent with this, yet suggest that this is only part of the story: Higher level L2 speakers also have a higher speech rate in their L1. This result further suggests that speech rate ― taken by many as an indicator of L2 proficiency ― is not, in fact, a reliable indicator. Or, if it is, it is a predictor in a more determinative sense: people who naturally speak faster are more likely to achieve a higher L2 proficiency level.
Token duration refers to the time duration of speakers' utterance of individual tokens in the corpus. The mean token duration shows a pattern that is consistent with previous work: Higher proficiency L2 speakers have a shorter mean token duration that lower proficiency L2 speakers. Furthermore, results show that the mean token duration does not vary when speaking in L1. Thus, unlike speech rate above, token duration can be seen as a reliable indicator of L2 proficiency., as distinct from L1 performance characteristics.
Silent pause rate refers to the number of silent pauses which occur per 100 tokens. In previous research, silent pause rate has been observed to correlate well with L2 proficiency: Higher-level L2 speakers have a lower silent pause rate than lower-level speakers. The corpus in the present project confirms this result, and further shows that it is independent of L1 performance characteristics.
Silent pause duration refers to the time duration of individual silent pauses. The corpus results are consistent with previous research showing that higher-level L2 speakers have a shorter mean silent pause duration, but the results further suggest that this not independent of L1 performance characteristics. That is, just like for speech rate above, results suggest that those who pause shorter in L1 are also those who pause shorter in L2.
Filled pause rate refers to number of filled pauses ('uh'/'um' in English, 'e-(to)'/'ano'/'n' in Japanese) per 100 tokens. The results of the corpus are consistent with most previous research showing no difference in mean filled pause rate across L1 or L2. Filled pause rate is not a reliable indicator of L2 proficiency at all.
Filled pause duration refers to the time duration of filled pauses. Previous work has not looked at this feature of speech closely. In the current project, results show that higher-level L2 speakers use shorter filled pauses than lower-level speakers. Furthermore, this trend is independent of L1 speaker performance characteristics. Hence, the results pattern like those of token duration and silent pause rate, above.
Formant measurements of the filled pause vowels showed that low-level learners have an L2 vowel height that is close to the typical L1 vowel (/E/) and that as L2 proficiency gets higher, the L2 vowel height gets lower and closer to that of the typical L2 vowel (/@/). However, this same trend was also observed in L1, suggesting a backward influence of L2 filled pause vowels on L1 filled pause production. In contrast, vowel backness did not show the same trend: Higher-level L2 speakers use a more centralized vowel than lower-level speakers ― more back than the typical L1 vowel. Thus, filled pause vowel backness is a better predictor of L2 proficiency than vowel height. These results parallel other results from the corpus that relate speakers' production to accent ratings. Speakers who consistently lowered their vowels were judged more native-like (i.e., closer to L2 target pronunciation) than those who consistently backed their vowels.
Repair rate refers to the number of repairs (repeats and self-corrections) that occur per 100 tokens. The corpus results here show merely that speakers have a higher repair rate when speaking in their L2 than in their L1. This is consistent across L2 proficiency levels.
Repair duration refers the the length in tokens of repair sequences. Similar to the results for repair rate, speakers use longer repair sequences in their L2 than in their L1. However, this is consistent across L2 proficiency levels.
These various findings are consistent with previous work showing the validity of the CCHP and the method used to obtain and annotate it. Furthermore, the findings that speech rate and silent pause duration are not strong predictors of L2 proficiency, as well as the findings that filled pause duration and filled pause vowel height are predictors of L2 proficiency are novel findings that have implications for how such things as fluency and accent are measured.