Tonal Processes Defined as Tone Sandhi Jie Zhang The University of Kansas Keywords: tone sandhi, typology, tone features, productivity, syntax-phonology interface 1. Introduction In this section, I provide the definition of tone sandhi (as a phonological process by which lexical tones exhibit contextually determined alternation) and a brief discussion of why tone sandhi is particularly worth investigating in Chinese dialects. The general thesis here is that, due to tone sandhi’s prevalence and complexity in Chinese dialects, it sheds light on fundamental issues in phonological theory such as the role of typology in synchronic theories, learnability, productivity, feature representation, the interface between phonology and other domains of linguistics such as phonetics and morphosyntax, and the processing of phonological alternation. It sets the stage for the rest of the discussion. 2. Typological patterns This section provides an overview of the typological patterns of tone sandhi in Chinese dialects. Drawing on works by Yip (1980/1990), Wright (1983), Yue-Hashimoto (1987), Chen (1991, 1992, 1996, 2000), Duanmu (1994), Bao (1999, 2004), Zhang (2007), and others, the section first categorizes the tone sandhi patterns as left-dominant and right-dominant, as determined by the edge of the sandhi domain that maintains the base tone, then discusses a typological asymmetry between left- and right-dominant sandhi patterns: left-dominant patterns, as attested in Northern Wu dialects, generally involves the spreading of the initial tone rightward, while right-dominant patterns, as attested in Min, Southern Wu, and Mandarin dialects, typically involve paradigmatic tone changes in nonfinal positions. Another prominent issue in the typology of tone sandhi is the prevalence of phonetically arbitrary patterns. Although there are typological tendencies for tone sandhi to be related to phonetic assimilation, dissimilation, and contour reduction, counterexamples abound. The commonly attested synchronic chain shift in tone sandhi systems is also a manifestation of phonetic arbitrariness. For instance, the tone sandhis in the “tone circle” pattern in Taiwanese Southern Min, as given in (1), have no phonotactic, and hence, phonetic motivation, as the base tone itself is not phonotactically illegal in the sandhi position. (1) Taiwanese Southern Min tone sandhi: 51 " 55 " 33 ! 24 in non-XP-final positions % ' 21 The typology of Chinese tone sandhi patterns raises a number of questions regarding what synchronic tone patterns mean for phonological theory. To what extent do they provide arguments for the formal representation of tone? Are all synchronic patterns equally productive and learnable for speakers? And if not, what are the factors that influence the productivity of sandhi patterns, and what does this entail for the nature of tone sandhi grammar? Sections 3 and 4 explore these issues in further detail. 1 3. Tone sandhi and tonal representation The patterns of tone sandhi have featured prominently in the debate on the nature of tonal representation. The representational issue typically revolves around the following questions: (a) What is the tone-bearing unit (TBU)? (b) What are the primitive features of tone? (c) Are contour tones represented by unitary features or sequences of level features? (d) If there are different layers of tonal features such as register (representing the overall pitch height) and contour (representing TBU-internal pitch change), what is the geometric relation among the features: independence, dominance, or sisterhood? This section summarizes some of the major debates regarding these questions, drawing on works such as Wang (1967), Woo (1969), Yip (1980/1990, 1989), Duanmu (1994), Chen (1996), Bao (1999), Tsay (1994), and others. But the section also notes that the discussion of the formal representation of tone has been stymied, for two reasons. First, the impressionistic tonal transcriptions that theorists use are easily open to reinterpretation according to any theoretical position. Second, due to tone’s diachronic malleability, the synchronic tone sandhi patterns are likely a combination of productive phonology and lexicalized patterns. This makes the practice of using specific synchronic sandhi patterns as arguments for tonal representation suspect. It is suggested that more fruitful approaches to tonal representation are likely found in phonetic and psycholinguistic studies. For instance, in a large body of work on the nature of timing and coarticulation of Mandarin tones in different syllable types and speaking rates, Xu and colleagues showed that contextual tonal variations are better accounted for by asymptotic approximations of tonal targets that include both static and dynamic targets rather than static targets alone. Therefore, the underlying units for contour tones consist of dynamic targets such as [rise] and [fall] with a linear movement specification, and the entire syllable, regardless of its segmental composition, is the TBU (Xu 1998, 2001, Xu and Wang 2005, Prom-on et al. 2009, etc.). Xu’s view on the unit nature of contour tones is supported by speech error data on Taiwan Mandarin tones in Wan and Jaeger (1998). 4. The productivity of tone sandhi patterns The productivity and learnability of various tone sandhi patterns have been investigated under the nonce-problem tests (“wug” tests) in a long series of work since the 1970’s. Hsieh (1970, 1975, 1976), Wang (1993), and Zhang et al. (2011), for example, have shown that the circular chain shift in Taiwanese Southern Min is not fully productive, indicating that the analysis for the sandhi pattern should not be in the form of underlying representation to surface representation derivation, but lexical or allomorph listing. Zhang and Meng (2016) and Yan and Zhang (2016) provide a comparison between Shanghai and Wuxi tone sandhi. Both dialects have a left-dominant spreading pattern, but Shanghai simply spreads the tone of the initial syllable, while Wuxi needs to substitute the initial tone with another tone before spreading, and the tone substitution is paradigmatic and involves a chain shift. Wug test results showed that the Shanghai pattern is generally productive; for Wuxi, the spreading aspect of the tone sandhi is likewise productive, but the substitution aspect of the sandhi is unproductive due to its chain shift nature. Another aspect tone sandhi productivity is whether the productivity of a sandhi pattern is positively correlated with the phonetic naturalness of the sandhi. Zhang and Lai (2010) tested the productivity difference between the less natural T3 sandhi and the more natural half-third sandhi in Mandarin and showed that, although both sandhis apply consistently in novel words, 2 the former involves incomplete application of the sandhi phonetically and is thus less productive. Zhang et al. (2011) showed that the duration properties of tones play a role in the productivity of the sandhi patterns in Taiwanese Southern Min. These results support the relevance of phonetics in synchronic phonological grammar. In general, the productivity studies of tone sandhi demonstrate that native speakers’ phonological knowledge of tone sandhi is not necessarily identical to the lexical patterns of tone sandhi. In other words, traditional tone sandhi descriptions need to be supplemented with experimental data to serve as the empirical basis for theoretical analysis. 5. The processing of tone sandhi patterns The nature of tone sandhi can also be investigated by observing the production and perception processing of tone sandhi words. Chien et al. (2016, 2017) used auditorily primed lexical decision tasks to investigate the representation of tone sandhi words that the listeners use in lexical access in Mandarin and Taiwanese Southern Min. The basic method is to ask the subjects to decide whether a disyllabic item is an existing word in the language when the word is preceded by different types of monosyllabic primes that share either the base tone or the sandhi tone of the first syllable of the test item. The reaction time to the lexical decision is used as the dependent measure for the priming effect. The results showed a striking contrast between Mandarin and Taiwanese tone sandhi: for Mandarin T3 sandhi, the base-tone primes (i.e., T3 primes) elicit significant priming compared to control primes, while for Taiwanese tone sandhi in the circular chain shift, the surface-tone primes elicit significant priming. Moreover, the priming effect is regulated by word frequency in Taiwanese, but not for Mandarin. These results indicate that T3 sandhi words in Mandarin are represented in their base tones, while the sandhi words in Taiwanese are represented in their sandhi tones. These are consistent with the wug test results that suggest that the chain shift patterns in Taiwanese are lexicalized, while the Mandarin T3 sandhi pattern can be productively derived. Another line of research uses the implicit priming paradigm to investigate the representation of tone sandhi words used during speech production (e.g., Chen et al. 2011, Chen and Chen, in press, Politzer-Ahles and Zhang, in press). In implicit priming, participants memorize small sets of words that are each paired with a cue. Participants are then asked to say the words as quickly and accurately as possible when they see the cues. Reaction times tend to be faster when the targets in a set are phonologically homogeneous in terms of some initial portion of the word. Politzer-Ahles and Zhang (in press) used