In this experiment, learners A and B showed many
pronunciation errors in several phonemes. In general, it is well known
that the non-native learners from Korea, China, Taiwan, have
pronunciation problem on voiced-voiceless distinction: e.g.,
/d/ /t/, /g/ /k/, etc. The
following tendencies were observed by human judges.
- (1)
- Learner H tends to pronounce /u/ instead of /o/.
- (2)
- Learners A and B were turned out to make an
error between voiced and unvoiced sound. In this case, they
mispronounced /t/ for /d/.
- (3)
- Learner B did the same error as the above. More
specifically, he did /t/ for /d/ sound; e.g., [kyoodo] not [kyooto], [wadashi] not [watashi].
- (4)
- Learner A and B frequently mispronounced /sh/ for /s/ sound; e.g., [mosimosi] not [moshimoshi], [watasi] not
[watashi]. It is generally known that for /sh/ sound,
non-native speakers tend to pronounce /s/ sound[26].
- (5)
- Learner B made an error between /z/ and /j/.
The use of relative thresholds correlated well with human judgement,
along with effectively detecting pronunciation errors. For some
phonemes such as /m/, /n/, /r/, rather, the native model speaker
showed score degradations. In a sense, the use of multiple native
speakers is thought to be more reasonable as a criterion of threshold
function.
Next: 3.6.4 Correlation between Human
Up: 3.6.3 Task 3 :
Previous: Automatic Scoring
Jo Chul-Ho
Wed Oct 13 17:59:27 JST 1999