Productivity of Tone Sandhi Systems in Chinese Dialects
Chinese dialects are known for their complex patterns of tone sandhi. The challenge that these patterns pose for phonological theory lies in the following aspects: the ambiguous role of phonetics, the existence of phonological opacity, the prevalence of gradience, variation, and exceptions, and the potential relevance of syntactic structure. We have investigated these aspects of tone sandhi from the perspective of productivity in a number of Chinese dialects using the “wug” test paradigm.
Our study on the Beijing dialect focused on the issue of the relevance of phonetics in its tone sandhi system. We compared native Beijing speakers’ application of two exceptionless sandhi processes to novel words: the contour reduction 213→21/__T (T≠213), which has a clear phonetic motivation, and the perceptually neutralizing 213→35/__213 whose phonetic motivation is less clear. Results show that speakers apply the 213→21 sandhi with a greater accuracy than the 213→35 sandhi in novel words, indicating a synchronic bias against the phonetically less motivated pattern. We also showed that lexical frequency is relevant to the application of the sandhis to novel words, but it alone cannot account for the low sandhi accuracy of 213→35. The results of this study are published in Zhang and Lai (2010).
Taiwanese tone sandhi is noted for its opacity in the Kiparskian sense, and our study on Taiwanese focused on the effect of opacity on the productivity of the sandhi patterns. Through wug test experiments on sandhi applications in three different morphological processes — disyllables, single reduplication, and double reduplication, we found that sandhi productivity is negatively affected by opacity as well as unfavorable phonetic conditions and low lexical frequency. Opacity outweighs phonetics and frequency as a global effect; frequency effects are evident for everyday users of the language; phonetic effects only surface for occasional users for whom the frequency effects have been weakened due to the lack of use of the language. We have also modeled the simultaneous underlearning of exceptionless opaque patterns, overlearning of phonetic effects, and proper learning of lexical statistics by Taiwanese speakers using Maximum Entropy grammar that encodes learning biases against lexical listing constraints and phonetically unmotivated patterns. Results from the disyllables, single reduplication, and double reduplication can be found in Zhang, Lai, and Sailor (under review), Zhang, Lai, and Sailor (2011), and Zhang and Lai (2008), respectively.
As a dialect in close proximity to Beijing Chinese, Tianjin Chinese is undergoing a number of changes in its tonal system due to influences from Beijing, causing variations and exceptions to the sandhi patterns. In an acoustic study, we showed that the majority of the sandhis are not only variable, but also non-neutralizing, contra traditional descriptions. The acoustic study also showed that tonal coarticulation in Tianjin exhibits a number of well known crosslinguistic properties: progressive assimilation, regressive dissimilation, a greater progressive effect, and a number of High/Low asymmetries. Despite the dissimilatory nature present in both tone sandhi and tonal coarticulation, they seem to have different properties, indicating a complex relation between the two (Zhang and Liu, under review). We also investigated the effect of variability on the productivity of the sandhi patterns through a wug test, paying special attention to whether Tianjin speakers’ knowledge of tone sandhi matches the lexical patterns. Our results indicate that Tianjin speakers’ sandhi knowledge is again a combination of overlearning, underlearning, and proper learning. In particular, a sandhi pattern with a small percentage of lexical exceptions is overlearned and applies regularly in novel words; regular sandhis and a sandhi that only applies to a small percentage of lexical items are underlearned and apply less accurately in novel words; and a sandhi with a strong phonetic motivation is properly learned. We have constructed a biased Maximum Entropy learning model that aims to capture the different learning profiles. The results have been presented at GLOW-Asia-8 and the 2nd Pan-American/Iberian Meeting on Acoustics in 2010.
Shanghai tone sandhi is interesting in a number of aspects. First, its tone sandhi pattern is structure-sensitive. A disyllabic word with a modifier-noun structure may undergo a different type of tone sandhi from a disyllabic word with a verb-noun structure despite their identical underlying tones; the former undergoes left-dominant sandhi, whereby the tone on the initial syllable is preserved, while the latter often undergoes right-dominant sandhi, which preserves the tone on the last syllable. Whether a verb-noun word undergoes left- or right-dominant sandhi is dependent on its frequency of usage — the more frequent the word is, the more likely it will have a left-dominant sandhi pattern. Second, the left-dominant sandhi differs from the right-dominant sandhi in Shanghai in that it involves spreading the tone on the initial syllable over the entire disyllabic sandhi domain rather than a localized paradigmatic tone change. This spreading vs. paradigmatic change difference has been argued by Zhang (2007) to be a prototypical manifestation of the asymmetry between left- and right-dominant sandhis due to their different markedness. Therefore, a study of Shanghai affords us the opportunity to investigate the relevance of structural sensitivity to tone sandhi productivity and how this structural sensitivity may interact with the markedness and frequency effects that we have found for Beijing, Tianjin, and Taiwanese. We conducted a wug test on tone sandhi in different types of words in Shanghai and data analyses are currently underway.