字典音總是完勝嗎？——從腦波實驗看台灣華語噝音變異之表徵

Abstract

本研究利用事件相關電位之實驗法，探討中文裡出現頻率不同之中噝音變異體之處理歷程，並研究變異體語音感知與產出之關係。實驗共招募42名母語人士參與跨感官促發實驗作業，聽覺促發詞為60個捲舌音起首詞、60個齒音起首詞，以及60個塞音起首詞，以三種不同口音呈現：即標準口音、變異口音及控制口音。此外，根據受試者對噝音的平翹舌合流程度，分為「非合流組」與「合流組」。前者可明顯區分平舌音及翹舌音，而後者傾向將翹舌音平舌化。結果顯示，在聽覺促發階段，與標準發音相比，變異發音之塞音所引發之N400效果最大，其次為齒音起首之變異發音，而捲舌音起首之變異發音之N400反應則與其標準發音無顯著差異。此即說明，假變異體與罕見變異體需耗費更多認知資源以完成詞彙觸接。後續之視覺詞階段顯示，相較於前面先聽到之標準口音，三組聲音的P2效果在前面接著變異口音時均有所增強，說明相較於標準發音，對變異發音的注意力皆較為上升。此外，此效果也表示標準口音其優勢所在；但N400的效果在平翹舌起首詞，並未發現標準與變異口音之差異，說明在短暫延遲後，所有形式的發音刺激皆具類似之語意促進作用。在假變異體之後的視覺詞中，LPC的效果比其標準口音更大，表示聽者有意識地重新分析聽覺促發詞與後續視覺詞之關係。因此，我們的研究結果顯示，不同出現頻率的變異體會影響詞彙觸接的歷程，但這個現象在視覺呈現後便很快消失。此外，在合流與非合流說話者之間並未發現腦波反應之顯著差異，即便合流組亦可區分出平翹舌之不同。

This study examined the processing of Mandarin sibilant variants of different frequencies using the ERP technique and investigated the relationship between variant perception and production. Forty-two native speakers were recruited to participate in a cross-modal priming task, in which they first heard an auditory prime, followed by a visual target. Auditory primes were 60 retroflex- and 60 dental-initial bisyllabic words, along with 120 frequency-matched non-sibilant-initial bisyllabic words. The sibilant-initial words were pronounced in two different pronunciation conditions, canonical and variant. The retroflex-initial words could surface as dental-initial variants through the frequent process of deretroflexion (/tʂ, tʂh, ʂ/ → [ts, tsh, s]) ) while the dental-initial words could surface as retroflex-initial variant through the infrequent process of hypercorrection (/ts, tsh, s/ → [tʂ, tʂh, ʂ]). Sixty unaspirated stop-initial words were also included and their variants were created through a nonexistent pseudo process of aspiration (/p, t, k/ → [pʰ, tʰ, kʰ]). Subjects were checked for their preferred pronunciation of sibilants and were subgrouped into “non-merged speakers” and “merged speakers”. The former included subjects who were inclined to pronounce their retroflexes as is, while the latter included subjects who tended to deretroflex their retroflexes. Results showed that compared to the canonical pronunciations, the largest N400 responses were elicited by the pseudo stop variants, followed by the infrequent dental variants, and the frequent retroflex variants elicited the smallest N400 response during the auditory prime, suggesting greater effort in lexical access for the pseudo and infrequent variants. Responses to the subsequent visual words showed enhanced P2 for variants of all three sound types relative to the canonical forms, but no difference was found for the N400 responses, indicating elevated attention following variants but similar semantic facilitation from the audio stimuli for all forms after a short delay. Furthermore, the P2 effects on the visual target preceded by the variant pronunciations implied a possible source of canonical form advantage. The late positivity component (LPC) was larger in the visual targets preceded by pseudo stop variants than their canonical forms, indicating conscious recollection of the prime-target relationship. Our findings thus suggested that the variant frequency effect emerged during lexical access prompted by variant pronunciation, but was quickly annihilated after visual presentation. Besides, no significant canonical-variant difference of neural responses was found between merged and non-merged speakers. Listeners could differentiate the contrast between retroflex and dental even for those who tended to merge retroflexes with dentals.