學習障礙兒童對華語捲舌音特徵的聽知覺辨識

The Auditory Identification of Mandarin Retroflex Features for Children with Learning Disabilities

華語中有三對捲舌音對比。在聲學方面，捲舌音對比的不同主要在於噪音頻譜型態的差異，學習障礙兒童對於捲舌音對比聽覺區辨是否異於一般兒童？本研究目的在比較學習障礙兒童和一般學童對於華語捲舌音特徵的聽知覺辨識，並探討和捲舌音判斷有關的聲學頻譜參數。共計有六十位兒童參與聽覺實驗，兒童的年齡範圍分佈由九歲到十二歲，其中三十位兒童具有學習障礙，三十位無學習障礙兒童為控制組。在第一項作業中兒童需判斷所聽到的華語單音節是否為捲舌音，在第二項作業中判斷雙音節詞中的第一音節是否為捲舌音。結果顯示無論是對單音節詞或是雙音節詞語音刺激，學習障礙兒童的捲舌判斷正確率皆顯著低於一般兒童，尤其對不捲舌語音刺激的誤判率較高。根據兒童捲舌判斷的反應分析刺激項噪音之聲學頻譜動差參數，發現學障組在四個動差參數上皆有對比性減弱的趨勢，顯示學障兒童對於語音噪音段能量的頻譜型態辨識能力較弱。多元迴歸分析的結果顯示控制組的反應在頻譜重心頻率（M1）變項上具有最大的解釋量，而在學障組也是以M1 有最大的解釋量，但解釋量的大小卻遠不及控制組。由於捲舌音對比在聲學上主要的差異是噪音位在頻譜頻帶的分布型態，推論學障兒童在聽知覺上較無法掌握語音中噪音頻譜的頻率重心特徵。

In Mandarin Chinese, there are three retroflex contrasts of fricatives and affricates. The major acoustical difference between these retroflex contrasts is the spectral pattern of frication noise, but the perceptual distinction between these contrasts is subtle in nature. It is of interest to discover if children with learning disabilities (LD) have difficulty differentiating the retroflex contrasts perceptually, as well as whether children with LD have different response patterns for the task of retroflex detection. Purpose: The purpose of this study was to compare differences in the auditory identification of retroflexes between children with LD and children without LD, and to investigate the acoustical spectral parameters related to the retroflex perceptual judgment of the two groups of children. Methods: There were a total of 60 children participating in the experiment, 30 of which had LD and 30 that did not. The children that did not have LD served as the control group. The ages of the subjects ranged from 9 to 12 years old. The perceptual task was to listen to monosyllabic words one by one and judge if the stimuli were retroflexes or not as quickly as possible. In the second task, the stimuli were bisyllabic words, and the retroflex judgment was constrained to the first syllable of each word. Results: The results showed that the percentage of correct retroflex identifications of monosyllabic words for children with LD was 52%, which was significantly lower than that for the control group, at 66%. For the bisyllabic word stimuli, the trend of the response patterns was similar to that for the monosyllabic words. The percentage of correct retroflex identifications for children with LD was 51%, which was significantly lower than that for the control group, at 66%. For nonretroflexed stimuli, the LD group had more ‘false alarm’ responses. Moreover, the LD group had more errors for words with nonretroflexed aspirated consonants. This suggested that for listeners with LD, the detection of retroflex features might interfere with the aspirated features. Using the retroflexion response of the two groups, the spectral moments of the frication noise in the words were acoustically analyzed. The results showed that the level of contrast between two categories (retroflexed vs. nonretroflexed) on the four moment values for the LD group was greatly reduced, especially in the first moment, M1, which implied the spectral frequency gravity of frication noise. In fact, M1 for retroflexed consonants should be lower than for their nonretroflexed counterparts; however, the responses of children with LD showed no such trends. Stepwise multi-regression also showed M1 to be the only robust variable entering the model. This held true for both groups, but the multi-regression correlation coefficient for the control group (R = 0.66) was much higher than that for the LD group (R = 0.23). Conclusions/Implications: The conclusions and implications of this study were as follows. The children with LD had difficulties differentiating the retroflex contrasts perceptually. The results of the spectral moment analysis suggested that the children with LD did not differentiate retroflexion according to M1, which is an important cue for normal listeners. The perceptual processing of the frequency gravity for the frication noise in speech may be defective for children with LD. This suggested that they may have auditory frequency resolution deficits during speech noise processing.