普通话不同信息结构中轻声的语音特性

Phonetic Correlates of Neutral Tone in Different Information Structures

本文将普通话两音节轻声词和其对立的非轻声词置于五种不同的信息结构中,分析了轻声的语音产出编码方式。研究使用了一些新的语音特征,如描写声调调形的斜率特征、谐噪比等特征。统计分析发现,轻声的声学特征与信息结构、轻声前字调和轻声底层调等相关。在这些特征中,音高和时长的作用最大;在单念的时候,音高的作用大于时长;在语流中,轻声相关量不但与信息结构相关,还与轻声的底层调相关。另外一个发现是,与非轻声词相比,轻声词除了轻声音节的音高变化外,其前字音域明显拉大,从而使得前后音节产生更大的轻重对比。

Neutral tone is one of the characteristics of Mandarin. Most previous studies have examined neutral tone from various approaches, such as lexieology （ Chao 1968）, syntax （Chao 1929, 1932; Lin 1962）, and phonology （ Wang 1997, 2000; Lin 1999）, while others have focused on its acoustic realization （ Chao 1922, 1932, 1933; Luo and Wang 1967; Chen and Xu 2006） and its perceptual correlates （ Lin 1962, 1983; Cao 1986; Wang 2004; Li, et al. 2014; Li and Fan 2015）. However, most studies on the phonetic properties of neutral tone so far have mainly focused on neutral tone words （ or phrases） that are produced in isolation. Much less is known on how neutral tone is realized in different information structures. This study addressed this issue. Our research question is three-fold： 1） Do different information structures affect the acoustic realization of neutral tone？ 2） How neutral tone is realized in different information structures/contexts？ 3） What are the most important correlates of neutral tone in different information structures/ contexts ？ To answer these questions, this study examined the realization of disyllabic neutral tone words as well as their non-neutral tone counterparts in five different information structures： Isolation, Broad Focus, On Focus, Post Focus Near and Post Focus Far. All neutral tone or non-neutral tone target words （ or phrases） were embedded in corresponding carrier sentences （ except in the Isolation condition）. The phonetic encoding schemes of neutral-tone words （NW） and non-neutral-tone words （SW） were explored by comparing patterns of acoustic features of the target words （ or phrases）, such as Pitch, Duration, Intensity and Voice Quality. Novel features were also adopted, such as Harmonics-to-Noise Ratio （HNR）, and Slope of tone in describing the tonal contours. Linear Discriminant Analysis （LDA） discovered that the acoustic features of a neutral- tone syllable are heavily related to information structure, the tone of its previous syllable, and the citation tone of its own. Among all the parameters examined in this research, Pitch and Duration are the most reliable ones, confirming previous studies such as Lin and Yan （ 1990）. When produced in isolation, Pitch outweighs Duration in discrimination; while in sentences, the discriminant features are more diverse due to different information structures and citation tones of the neutral-tone syllables. More specific findings are as follows： The change of Pitch and Duration is closely related to information structures and present three groups： a. Isolation; b. Broad Focus and On Focus; c. Post Focus Near and Post Focus Far. Another interesting finding is that a greater contrast was observed between the first syllable and the neutral-tone syllable, through a significant expansion of the pitch range of the first syllable in the NW, compared to that in SW. The Formant, Root Mean Square Amplitude （RMS） and Intensity show significant differences between NW and SW only in Isolation. However, they are not quite efficient in distinguishing NW in sentences. Spectrum Tilt （ the ratio of band energy change） is a more robust parameter compared to band energy itself. In the discriminant analysis, the Spectrum Tilt of syllable final is more efficient than that of syllable initial; while the Spectrum Tilt of initial at a high frequency band is more efficient than that at a low frequency band. The Harmonics-to-Noise Ratio （HNR） is a very reliable and stable acoustic parameter in the discriminant analysis, Moreover, by comparing the HNR of NW and SW, the results showed that： a） In neutral- tone syllables, the initial tends to be more voiced while the final more devoiced; b） The initial in sentences gets voiced more frequently than in Isolation, while the final shows the other way around, which gets devoiced more frequeutly in sentences other than in Isolation. Results of this study will shed further light on the typological differences of the phonetic correlates of the weak and stressed elements in tone versus stress languages. The results will also benefit the teaching and learning of Mandarin as a second language.