北四縣客語單字調與連讀調聲學研究

本研究探討北四縣客語的母語人士在單字詞、協同發音以及變調情境中如何產出該語言的字調。研究動機有二，首先是有鑑於過去之研究在北四縣客語字調系統的描述方面十分不一致，這是因為前人究取徑不同、或是處理聲學資料的程序不同所造成；其次是，鮮少研究探討北四縣客語鄰近字調的互動以及變調的特質。利用六位男性與六位女性北四縣客語母語人士所產出的單字調與二字調做為聲學分析以及統計分析的語料、同時比較多人次語料正規化分析方法的適切性。本研究發現對數z-分數正規化方法最能有效降低原始基頻資料中跨語者之間的基頻變化，於是我們將此正規化方法應用到本研究所有的基頻資料，以求最佳的正規化結果、並正確了解北四縣客語的字調聲學特質。同時我們也提出多項式模型與統計模型混合的研究方法，提供多項式係數做為標示字調的平均基頻音高、線性斜率、以及字調曲率的指標，以期更清楚區辨連讀變異動因。在字調呈現的議題方面，為利於比較，本研究利用石峰（1990）的T-值正規化方式計算北四縣客語六個單字調的調值，所得的結果和張月琴（1995）的聲學研究結果除了陽入調之外，其餘調值均一致。本研究的陰入對比陽入的調值是52對比55、比起張月琴（1995）的52對比53更具有差異識別性。在協同發音的相關議題發現：（1）本研究語料顯示北四縣客語的協同發音有順向及逆向兩個方向，而且正如大部份前人的研究所示，順向協同發音是同化作用多於異化作用、而逆向協同發音是異化作用多於同化作用。（2）然而，有別於其他語言的地方是，北四縣客語位於後字的降調，如T2 [51]、T4 [52]、T5 [31]，傾向受到前字調T1 [35]高調尾的同化影響而使整體調高提高，但後字的T1 [35]卻是受到前字的高平調T3 [55]及T6 [55]異化影響而調高變低。基於（2）的發現，因此，（3）曲折調受到的影響大於平聲調，因此曲折/平聲調的不對稱，而非像其他語言是高調/低調上的不對稱；（4）在逆向協同發音方面，北四縣客語的模式和其他聲調語言的發現大致一致，在較佳目標調與較佳促發調的字調上也是呈現高調/低調的不對稱性，例如，具有低調特質的中降調T5 [31]位於後字、是較佳促發調、能將前字調的高調部份之音高、透過逆向異化作用而將之提高、因此前字調的高調部份是較佳目標調；（5）至於字調受到協同發音影響的程度，不必然是順向高於逆向，而是因不同字調而定；（6）為正確詮釋字調協同發音的現象，不單語音的因素、音韻的動機也該一併考量在內。最後，我們探討構成北四縣客語變調的語音動機以及變調後的陰平T1 [31]與變調的目標調陽平T5[31]是否完全混同？本研究主張北四縣客語共時的陰平變調模式是綜合不同協同發音的歷時發展的音韻化結果。其一，前字上升陰平的高調尾透過順向協同發音的過程、提高後字調整體調高、但同時也因自身的基頻波峰延遲作用的影響（F0 peak delay effects）弱化了本身的上揚調形；其二，在另一上升陰平調T1 [35]之前、已弱化的前字上升陰平調受到後字陰平調在調形上的逆向異化作用影響、而逐漸變為下降的中降調；其三，已弱化的前字陰平調、在舒聲高平調 T3 [55]、以及入聲高平調 T6 [55]之前，受到後字兩個高平調在音高上的逆向異化作用而使得前字陰平調的整體音高大幅降至最低點。最後，變調產生的位置─詞首，是韻律上較弱的位置，出現在詞首的字調較易受到後字調的影響而變異較大，而且在詞首位的音節音長較短、無法提供足夠的時間讓陰平調完整產出調尾的高調。另外要探討的變調議題是，在變調陰平調T1 [31]與陽平調T5 [31]的調形是否完全混同的議題上，發現不論是兩種字調的基頻調形或是速率調形都顯示：兩種字調在統計上均有顯著的不同；因此北四縣客語的變調是屬於漸變式的類型，而非調類取代式的類型。. 關鍵字: 北四縣客語、字調表徵、單字調、字調協同發音、變調、字調混同、多項式模型

關鍵字

北四縣客語；字調表徵；單字調；字調協同發音；變調；字調混同；多項式模型

並列摘要

This dissertation investigates how native speakers of Northern Sixian (NS) Hakka produce the six lexical tones in isolation, in tonal coarticulation, and in tone sandhi, motivated by the disagreement in the descriptions of the tone system of NS Hakka among previous studies due to differences in research approaches adopted or procedures for processing acoustic data, and to the sparsity in studies of interaction between adjacent tones and the nature of tone sandhi in NS Hakka. To achieve our research goal, methodologically, tones on mono-syllabic and disyllabic words produced by 6 male and 6 female middle-aged native speakers were collected for acoustic and statistical analysis. For the speech corpus, we addressed the issue of efficacy of F0 normalization strategies for speaker variability. The study found a log z-score normalization method optimal in reducing F0 variation across speakers from the raw data, and this normalization strategy was applied to converting all the raw F0 data for acoustic and statistical analysis into log z-score values so as to attain an accurate understanding of the acoustics of tones in NS Hakka. Meanwhile, we also proposed a mixed method of polynomial and statistical modeling, which affords acoustic indices for the mean F0, linear slope, and curvature of tonal contours, and estimate differences in these aspects between tones. Regarding the issue of tonal representations for the tone system of NS Hakka, our data on the T-value normalized (Shi, 1990) F0 values for the isolated tones showed that the tone values calculated in this study for the six individual tones are highly consistent with those suggested by Chang (1995) to the exception of the checked Yang Ru tone. The study clarified the confusing tonal contrast between Yin Ru and Yang Ru found in Chang (1995), who annotated the contrast as [52] vs. [53], by proposing the contrast as [52] vs [55], a contrast of more differential distinctiveness. With respect to the issues associated with tonal coarticulation, the study found that: (1) both carryover and anticipatory tonal coarticulation were attested for NS Hakka with the former predominantly assimilatory and the latter dissimilatory, as reported in most previous studies for other tone languages; (2) however, unique to NS Hakka, word-final falling tones, such as T2 [51], T4 [52], T5 [31], were susceptible to assimilatory carryover effects primarily triggered by a word-initial rising T1, whereas the word-final rising T1[35] was prone to dissimilatory carryover effects induced by the two high level T3 [55] and T6 [55] in the word-initial position; based on (2) therefore, (3) a Contour/Level asymmetry was suggested regarding better effect undergors for carryover coarticulation, a pattern different from the High/Low asymmetry reported in previous studies on tonal coarticulation for other tone languages; (4) in terms of anticipatory coarticulation, patterns were quite consistent with findings from the past studies for other tone languages in that a High/Low asymmetry was found with respect to the better effect triggers, i.e. the low onset of a word-final mid-falling T5 [31] triggering a pitch raising effect on the word-initial tones, and with respect to better effect undergoers, i.e. the high pitch target of a word-initial tone that was particularly raised due to the dissimilatory anticipatory effects from the word-final T5 [31]; (5) the effect size associated with different coarticulation directions depended on the specific tones. Carryover effect size was not necessarily bigger than the anticipatory effect size; and finally (6) to make accurate interpretation of tonal coarticulation, not only phonetic groundings but also phonological motivations should be taken into account.. Finally, the acoustic study on tone sandhi in NS Hakka addressed the phonetic motivation underlying the pattern of the Yin Pin T1 sandhi in NS Hakka, and the completeness of tonal neutralization of the sandhi T1 [31] with the lexical T5 [31], to which T1 is alleged to turn in the sandhi contexts. We proposed that the synchronic phonological tone sandhi pattern in NS Hakka was a phonologized result of phonetic interactions between the word-initial Tone 1 [35] and the following sandhi context tones. First, a reduced rising form of the word-initial T1 was induced due to a high offset loss when the high offset raised the pitch level of the following sandhi context tones, and in the meantime, it was absorbed or realized in the following sandhi context tones. Second, in the T1T1 sequence, a contour dissimilatory effect was on the word-initial T1 in a reduced form by the following T1, whereby an already-reduced rising T1 in the word-initial position gradually turned a mild mid-falling tonal contour. Third, when T1 preceded the two high level T3 [55] and T6 [55], a pitch level dissimilation occurred on the realization of T1 in reduced form by the two following high level tones; in other words, a dissimilatory effect maximally lowered the pitch of the word-initial T1 to the lowest pitch range. Finally, the word-initial position, where tone sandhi occurs, is “prosodically weak,” implying that tones in this position tend to be affected by the word-final tones, and also leaving less time to fully realize the high offset target of the word-initial T1. Regarding the issue of completeness in neutralization, the study found the nature of tone sandhi in NS Hakka was gradient instead of categorical given the statistically significant differences in the realizations of tonal F0 contours and velocity contours between the sandhi T1 [31] and the lexical T5 in the same sandhi contexts. Key words: Northern Sixian Hakka, tonal representations, isolated tones, tonal coarticulation, tone sandhi, tonal neutralization, and polynomial modeling