台灣華語語音變異對詞彙辨識之影響

The effect of phonetic variation on word recognition in Taiwan Mandarin

莊育穎(Yu-Ying Chuang)

指導教授 ： 馮怡蓁

國立臺灣大學/文學院/語言學研究所/博士(2017年)

https://doi.org/10.6342/NTU201703463

摘要

本研究旨在探討台灣華語中語音變異如何影響母語人士詞彙辨識之過程。討論的語音變異現象共五個，包括去捲舌化、捲舌化、/in/→[iŋ]、/iŋ/→[in]以及/əŋ/→[ən]。前二者皆屬音節首子音變異，但由於去捲舌化較捲舌化更常發生於台灣華語，因此二者的出現頻率大相逕庭。後三者則為音節末鼻音變異，且三者間存在著社會意涵之差異：/in/→[iŋ]帶有正面社會意涵，/əŋ/→[ən]相比之下較為中性，而/iŋ/→[in]之社會意涵則最為負面。為了解這些語音變異對書寫詞彙辨識之影響，實驗一以跨感官形式促發實驗檢驗之。在此實驗中，標準體及變異體發音之促發詞以聽覺方式呈現，目標詞則以視覺方式呈現，而受試者須對目標詞做詞彙判斷，共計108位母語人士參與此實驗。結果顯示，變異體發音所造成之促發效果與標準體並無不同。實驗二進一步研究語音變異對口語詞彙辨識之影響。此實驗所使用之方法為聽覺形式促發實驗，因此促發詞及目標詞皆以聽覺方式呈現，共計240位母語人士參與實驗。實驗結果顯示，就促發效果而言，同實驗一，變異體與標準體發音亦產生相似的促發效果。然而在口語詞彙之立即辨識上，變異體及標準體之辨識結果卻有顯著的不同。整體而言，辨識效率與社會意涵息息相關，具正面意涵之發音方式比帶負面意涵之發音方式更易提高辨識效率。此外，變異出現頻率亦對詞彙辨識造成影響，但其效果卻較為薄弱，且影響力不及社會意涵。
基於前二實驗之結果，本研究第二部分探究提供說話者信息是否能影響或改變聽者對語音變異之處理。實驗三使用視覺刺激材料呈現說話者信息，在受試者對口語詞彙做詞彙判斷之前，先令其看到不同說話者之臉部照片。此實驗測試之變異包括去捲舌化及/in/→[iŋ]，共計136位母語人士參與此實驗。結果顯示，當說話者信息與常使用語音變異之族群相符時，變異體發音之辨識效率即有所提升；反之，當二者不相符時，變異體發音之辨識效率則降低。然而，此實驗中說話者信息之效果並不明顯，且並非適用於所有變異。實驗四則以聽覺刺激材料呈現說話者信息。其設計為在目標詞出現前，受試者會聽到不同腔調之句子發音。此實驗測試之語音變異為去捲舌化，共計64位母語人士參與實驗。實驗結果再一次證實，語音變異之處理的確涉及運用整合說話者信息，進而改變詞彙辨識之結果。此外，此實驗中說話者信息之效果，亦較實驗三更為顯著，且其影響力甚至更勝於實驗二所發現之社會意涵之效果。綜合以上結果，本研究顯示台灣華語語音變異之處理，不僅與變異本身之特質有密切關係，亦包含對語言及說話者等其他相關因素之考量，因而對詞彙辨識造成複雜且多層次之影響。

關鍵字

語音變異 ； 詞彙辨識 ； 變異頻率 ； 社會意涵 ； 說話者信息 ； 台灣華語

並列摘要

This study investigated how phonetic variation in Taiwan Mandarin affected word recognition of native listeners. Two onset sibilant and three final nasal variations were examined, including deretroflexion, retroflexion, /in/→[iŋ], /iŋ/→[in], and /əŋ/→[ən]. The discrepancy between the two sibilant variations lies in variant frequency, with deretroflexion occurring much more frequently than retroflexion in the language, whereas the three nasal variations differ in social connotation: /in/→[iŋ] is deemed fairly positive, /əŋ/→[ən] is relatively neutral, and /iŋ/→[in] is the most negative. Experiment 1 examined the effect of phonetic variation on written word recognition using a cross-modal form priming paradigm. Auditory primes with either canonical or variant realizations were followed by visual target presentation of the same word, on which listeners performed lexical decision. There were in total 108 native listeners tested. Results showed that for all variations, variant realizations induced comparable priming effects as canonical ones. Experiment 2 investigated how variant pronunciation impacted on spoken word recognition. An auditory form priming design was adopted, in which both primes and targets were presented auditorily. A total of 240 native listeners participated in this experiment. Similar to Experiment 1, the priming effects of both variant and canonical realizations were found comparable. However, the two forms were processed differently in immediate spoken word recognition. In particular, processing efficiency was determined by a given form’s social connotation to a large extent, as socially ideal forms were always better recognized than socially disfavored ones. In addition, variation processing was also subject to variant frequency, although its role seemed to be less decisive and more obscure than that of social connotation.
The second part of this study was devoted to exploring whether the integration of speaker information would alter the processing consequence of variant realizations. Speaker information was embedded in visual primes in Experiment 3, where face photographs of different speakers were presented to listeners before lexical decision on spoken words was performed. One sibilant onset variation (deretroflexion) and one final nasal variation (/in/→[iŋ]) were examined, and in total 136 native listeners were tested. Lexical recognition results showed that in general, congruent speaker information facilitated variation processing while incongruent information impeded it, but the effect was however moderate and variation-dependent. Experiment 4 used auditory primes to present speaker information, in which listeners heard sentence contexts uttered in different accents before spoken target words. This experiment focused on deretroflexion, and 64 native listeners were recruited for participation. Evidence of speaker information integration was again observed, as recognition results changed according to speaker accent primes. Nevertheless, dissimilar to Experiment 3, the magnitude of processing changes was much more prominent, an effect so robust that it even overrode the effect of social connotation revealed in Experiment 2. This study thus demonstrated that phonetic variation in Taiwan Mandarin has a multi-faceted effect on lexical processing, which not only bears an intimate relationship with the intrinsic characteristics of variations but also involves dynamic and complex interactions with a variety of linguistic and indexical factors.

並列關鍵字

phonetic variation ； word recognition ； variant frequency ； social connotation ； speaker information ； Taiwan Mandarin

參考文獻

Chang, Y. S. (2017). The influence of dialect information on the perception of the Mandarin alveolar-retroflex contrast. Concentric: Studies in Linguistics, 43(1), 1–23.

Su, H.-Y. (2012). The social implications of syllable-final nasal mergers in Taiwan Mandarin: A variation study. Language and Linguistics, 13(4), 767–802.

Yang, J. H. (2010). Phonetic evidence for the nasal coda shift in Mandarin. Taiwan Journal of Linguistics, 8(1), 29–58.

Brouwer, S., Mitterer, H., & Huettig, F. (2013). Discourse context and the recognition of reduced and canonical spoken words. Applied Psycholinguistics, 34(3), 519–539.

Brunellière, A., Dufour, S., & Nguyen, N. (2011). Regional differences in the listener’s phonemic inventory affect semantic processing: A mismatch negativity (MMN) study. Brain and Language, 117(1), 45–51.