從語音轉寫與聲學分析探討印尼學習者華語無聲擦音習得

Indonesian Learners' Acquisition of Mandarin Voiceless Fricatives: Perspectives of Phonemic Transcription and Acoustic Analysis

第二語言語音習得一直是學習者的難點之一，因此，對比分析 （Lado 1957）與語音習得模組（Flege 1987）皆試圖從不同的角度 預測並解釋學習者難點。華語有五個無聲擦音：唇齒擦音/f/、齒齦 擦音/s/、捲舌擦音/ʂ/、硬顎擦音/ɕ/與軟顎擦音/x/。而印尼語只有兩 個無聲擦音：齒齦擦音/s/與喉擦音/h/。因此，研究印尼學習者如何 習得華語無聲擦音除了可檢視上述兩個理論外也可瞭解語音習得 過程。本研究探討：（1）印尼學習者習得華語無聲擦音的難點， （2）印尼學習者使用哪些聲學特徵以分辨華語無聲擦音和（3）對 比分析和語音習得模組哪個理論能做較好的預測。為解答這些研 究問題，研究者製作華語與印尼語擦音詞表，詞表中所有的擦音都 出現在雙音節詞的第一個語詞，並考量搭配的元音。每組擦音與元 音的組合各有兩個語詞，因此，華語詞表共有 36 個語詞，印尼詞 表有 14 個語詞。二十位受試者參與詞表唸讀實驗，十位為華語母 語者，十位為華語程度中級的印尼學習者。分析方式有語音轉寫與 聲學分析，研究者先聽辨學習者音檔並將其擦音歸類為正確與不 正確，然後再將擦音噪音部分做頻譜動差分析並測量擦音與元音 交界處第二共振峰值。結果顯示捲舌擦音的正確率最低，這主要是 由於捲舌輔音為相對有標（marked）的語音。本研究同時發現印尼 學習者尚未學會顎化，致使[s]與[ɕ]難以有效地區分。另外，學習 者也與母語者使用不同的聲學特徵區別華語無聲擦音。最後，研究 結果顯示語音習得模組較能預測並解釋印尼學習者的難點。

Phonological acquisition of second language has been one of the difficulties for L2 learners. The CAH (Contrastive Analysis Hypothesis) claimed that inventories absent from L1 were difficult to learn, while SLM (Speech Learning Model) proposed that it was the similar sounds, but not the new sounds, that caused problems for L2 learners. To test which of these two can make a better prediction, this article attempted to analyze the data of Indonesian's acquisition of Mandarin voiceless fricatives. As contrasted in the two languages, there are in Mandarin five voiceless fricatives: labio-dental /f/, alveolar /s/, retroflex /ʂ/, palatal /ɕ/ and velar /x/. However, in Indonesian there are merely two: alveolar /s/ and glottal /h/. Three research questions are hinged here: (1) Which Mandarin fricative is the most challenging for Indonesian learners, (2) What acoustic cues these learners use to differentiate Chinese fricatives and (3) Which of CAH and. SLM can a better prediction. To answer these questions, the researcher designed two word lists, one for naïve speakers and the other for Indonesian learners. The data were then analyzed first by phonemic transcription and then acoustic analysis. Essentially, there are five phonetic cues adopted for the acoustic analysis, namely, (a) centroid/mean/M1, (b) standard deviation/M2, (c) skewness/M3, and (d) kurtosis/M4, plus (e) the onset of the second formant of the following vowels at the consonant-vowel boundary. The results show that the retroflex and palatal fricatives are more difficult for the learners. In addition, it is found that native speakers and Indonesian learners use different acoustic cues to differentiate those fricatives. In appearance, both CAH and SLM are equally successful in predicting that retroflex and palatal fricatives are challenging. Nevertheless, SLM is even better in claiming that similar inventories are the main causes for the problems.