Projects

Lexically-specific Sound Patterns: Borrowing or Lexical Diffusion

In this project I look at the diachrony of coda *-ŋ in Suzhou Chinese under the context of ‘Differing Literary and Colloquial Readings’ (“文白異讀”), a lexically-specific pronunciation alternation. In many varieties of Chinese languages there exist a Literary register containing technical/formal lexical items and recent borrowings from Standard Mandarin Chinese, and a Colloquial register containing the native vocabulary. On the surface, we observe certain Chinese characters having two pronunciations (Literary/Colloquial) while carrying very similar semantic meanings. This alternation has been considered as a prime example of Classical Lexical Diffusion (Wang 1969, Chen & Wang 1975), where the patterns essentially demonstrate a lexically gradual but phonetically abrupt sound change.

Based on the fieldwork data of Qian (1992), I argue that it is more appropriate to analyze the Literary forms as recent borrowings from Mandarin, and the Colloquial forms as direct descendants of language-internal historical forms. All Colloquial forms can be dated back to at least Middle Chinese (Pullyblank 1989, 1994) with phonetically-conditioned and phonetically-motivated sound change: (i). *-ŋ fronts to alveolar [n] following a non-back nucleus *ɘ or *ɪ; *ɘŋ/ɪŋ > [ɘn/ɪn]; (ii). *-ŋ deletes and transfers the nasality to a preceding low nucleus *a or *ɒ; *aŋ > [ã:]; *ɒŋ > [ɒ̃ŋ] >[ɒ̃:]. On the other hand, Literary forms showing the abrupt (i.e. irregular) change *aŋ > [ɘn] is a result of recent loans from Mandarin Chinese, as all characters showing the Literary/Colloquial pronunciation split have [ɘŋ] as their rhyme in contemporary Mandarin.

I show that Lexical Diffusion is not the only explanation for a lexically-specific pattern of sound change. The current case of Differing Readings can be analyzed as lexical borrowings with both empirical and theoretical adequacy: the presumably innovative change in Literary Readings does not spread to the Colloquial domain as a Lexical Diffusional analysis would predict — in fact, Literary pronunciations drastically recede and are replaced by more ‘conservative’ Colloquial forms in younger, more recent generations. In addition, the borrowing analysis does not invoke an ad-hoc analysis for the only irregular change *aŋ > [ɘn], but attributes it to language-external factors that do not undermine `the regularity hypothesis’ (Hockett 1965), the belief that all forms of sound change are phonetically-conditioned and fully regular.

Papers:

Talks/Posters:

 

Light-initial Tone Sandhi and Metrical Dominance in Suzhou Chinese

Previous phonological analyses of tone sandhi in Northern Wu dialects (Chan & Ren 1989, Duanmu 1999, Shi & Jiang 2013, among others) assume a left-aligned disyllabic trochaic foot as the relevant tone sandhi domain. In a disyllabic trochaic foot, the strong (initial) syllable retains all tonal material and the weak (second) syllable deletes its tone. Consequently, only the initial syllable determines the sandhi output of phonological words. A minimal example in Suzhou is patterns such as /LH/ + T  → [L.H], where an underlyingly /LH/ initial syllable plus syllables of any tone will surface as a disyllabic word with a Low-High sequence. This is referred to as Left Dominance.

My current research examines tone sandhi patterns of words with an initial ‘checked tone’ (‘entering tone’, “入聲”), which seems to show exceptions to the Left Dominance pattern. Historically, ‘checked tone’ was characterized by a stop coda *-p/t/k. In multiple attestations of contemporary Suzhou Chinese (Qian 1992, Ye 1993, Wang 2011), ‘checked tone’ syllables are transcribed as [(C)Vʔ], with a merged glottal stop coda [ʔ]. However, phonetic data from my recent fieldwork show that the coda stop closure is (1). neither present in isolation (e.g. [bɑ], ‘white’); (2). nor in conjunction with a following syllable (e.g. [bɑ.mɨ:], ‘white rice’; note the vowel length difference). The historical coda /ʔ/ has gone through complete neutralization, and the ‘checked tone’ categories in Suzhou are better represented as plain short (monomoraic) vowels in open syllables – /CVµ/, in contrast with ‘unchecked’ /CVµµ/ or /CVµNµ/ (subscripted µ indicates autosegmental association).

The emerging light-heavy syllable quantity contrast has significant bearings on the phonological system of contemporary Suzhou, lending metrical motivation for the exceptional sandhi patterns of checked-tone-initial prosodic words (I refer to this as ‘light-initial’ sandhi). The light-initial pattern differs from traditional Left Dominant sandhi patterns mainly in two ways: (1). Tones of the second syllable can influence the sandhi output in light-initial patterns, but not in Left Dominant patterns; (2). Only the first two moras may carry a H tone in a light-initial sandhi ([Lµ.HµLµ], [Hµ.Lµµ] but *[Lµ.HµHµ], *[Hµ.LµHµ]), whereas up to the first two syllables may carry a H tone in Left Dominant sandhi ([Hµµ.Hµµ], [Lµµ.Hµµ]). Both observations point to an alternation of metrical footing in the two sandhi patterns – a syllabic trochee for Left Dominant (i.e. heavy-initial) sandhi patterns, a moraic trochee (motivated in Kager 1993, Kager & Martínez-Paricio 2018) for light-initial sandhi patterns.

This project addresses a key debate in prosodic typology, viz. the interaction of tone, syllable quantity, and metrical structure (Kehrein et al. 2018 for overview). This study ties phonetic evidence (i.e. absence of coda closure; phonetically monomoraic open syllables) with a phonological argument (i.e. alternating syllabic and moraic trochees conditioned by initial syllable quantity), and gives an accurate phonological account to all possible tone sandhi patterns in contemporary Suzhou.

Papers:

Talks/Posters:

 

Computational evaluation of tonelessness in Suzhou

This project is a continuation of the above phonological analysis as it provides quantitative support for the toneless mora in Suzhou light-heavy (T.TØ) disyllables. So far, we know that light-heavy is the only weight profile in Suzhou that uniformly ends in low pitch in isolation (recall: *[Lµ.HµHµ], but [LµLµ.HµHµ]/[Lµ.Lµ.Hµ]/[Lµ.Hµ]). This points at a potentially toneless mora [T.TØ] in light-heavy sequences, yet the exact pitch interpretation for [Ø] is less clear, especially in word-medial positions.

Two major approaches exist for the interpretation of tonelessness: default tone (Chen 2000, H. Zhang 2016; Also see Yip 2002) and interpolation (Pierrehumbert & Beckman 1988; Gussenhoven 2004; Lee & Zee 2008). The former assumes that there is a ‘last-stage’ phonological process that supplies all yet-toneless TBUs with a default (often L) tone. The default tone is not phonologically active (i.e. does not participate in other processes) but is otherwise indistinguishable from ‘true’ tones. The interpretation account takes toneless TBUs to be without a tone throughout, and only interpretable in pitch based on interpolation to its surrounding. Isolation/phrase-final light-heavy disyllables cannot differentiate the two hypotheses (default L inserted to Ø vs. Ø interpolation to a phrase-boundary L%), therefore a study with more diverse tonal contexts is in order.

In the current study, I adopt the methodology of Kawahara & Shaw (2018) and M.Zhang et al. (2019). Toneless moras with different tonal contexts in Suzhou are sampled from trisyllabic or quadrisyllabic words (e.g. HØ.L…, HØ.H…). After pitch extraction and data compression via Discrete Cosine Transform (DCT, see Kawahara & Shaw), transformed F0 trajectories of the toneless moras are compared against (i). trajectories of fully-toned moras (e.g. HØ.L vs. HL.L), representing a ‘default L tone’ hypothesis, and (ii). simulated trajectories by applying naturally-observed variance to a straight ‘interpolation’ line, representing an ‘interpolation’ hypothesis. The goal of the simulation analysis is to assess whether the toneless trajectories more closely resemble ones with true phonological tones, or ones sampled from interpolation, or neither. The fully-toned and simulated tokens are also fitted to a Naive Bayes classification model. The classification model then assigns probabilities of ‘fully-toned/default tone’ vs. ‘simulated/interpolation’ to each toneless trajectory, yield a token-by-token estimate of toneless interpretation.

Data collection and annotation for the project is still ongoing. Preliminary data show that speakers differ in their ‘strategies’ to interpret tonelessness in specific tonal conditions. Overall, context [HØ.L…] heavily favors a ‘default L’ interpretation, while [HØ.H] contains more interpolation tokens. It is important to note that despite the general trend, there is no single unified way to interpret tonelessness in Suzhou, as both approaches are adopted by the speakers to different degrees.

 

Talks:

  • Variable Pitch Realization of Unparsed Moras in Suzhou Chinese: Evaluation Through F0 Trajectory Simulation and Classification. AMP 2022. UCLA.
  • Assessing tonal representation quantitatively: what does ‘toneless’ mean in Suzhou Chinese? Guest lecture at Chinese 7382: Chinese Phonology. The Ohio State University.

 

Strictly Local phonological processes and tones

Recent work on formal language theory has argued that most, if not all, phonological processes are Strictly Local – the phonological grammar only operates on (or “pays attention to”) contiguous strings of bounded length (Chandlee 2014, Chandlee et al. 2014, Chandlee and Heinz 2018, inter alia). Strictly Local phonological processes can be further divided into two subclasses: Input Strictly Local (ISL) maps and Output Strictly Local (OSL) maps. ISL processes, as the name suggests, are defined on contiguous and bounded input strings; OSL processes, on the other hand, are defined on contiguous and bounded output strings.

In the meantime, developments in phonological theory have also been focusing on the issue of locality/adjacency. Multiple seemingly non-local processes (e.g. nasal assimilation, tonal processes) have posit challenges for traditional SPE analyses, which have inspired approaches such as Autosegmental Phonology (Leben 1973, Goldsmith 1976). By treating phonological strings as bundles of relevant tiers (of distinctive features, prosodic constituents, tones, etc.), Autosegmental Representation characterizes non-local relations as local ones on specific relevant tiers (see Odden 1994). However, a closer mathematical investigation of the claim above reveals some peculiar results: while several linearly non-local tonal processes are indeed Strictly Local under autosegmental representation, there are also cases that are autosegmentally non-local but linearly local (See Koser et al. 2018, Chandlee 2018, Chandlee and Jardine 2019).

My present project assumes the formalism of Chandlee & Jardine (2019) and examines the locality status of three tonal processes in different Chinese languages: floating tone affixation in Cantonese (Chen 2000, Yip 2002), left dominant sandhi in Shanghai Wu (Duanmu 1999), and a combination of lexically-specified floating tones and left dominance in Suzhou Wu (my fieldwork). I show that simple floating tones (i.e. tones without segmental associations) and metrical dominance effects can both captured under ISL maps, while the combination case in Suzhou cannot. Crucially, the Suzhou case fails to conform to Autosegmental-Input Strictly Local (A-ISL) languages not because of the formalism under examination, but because of the property of representation itself – any other analytical framework in phonology (e.g. rewrite rules, Optimality Theory) would encounter the same issues unless the tonal representations are revised.

Strictly Local phonological maps therefore offer us a very unique machinery to study phonological locality: if one takes the strong hypothesis and claims that all phonological maps are Strictly Local, this sort of formal language theoretic analysis becomes a powerful “probe” in the sense that should exceptions occur, one should immediately look for revisions in the underlying representation, but not the grammar itself.

Paper:

Zhu, Yuhong. 2020. “Extending the Autosegmental Input Strictly Local Framework: Metrical Dominance and Floating Tones,” Proceedings of the Society for Computation in Linguistics: Vol. 3 , Article 38.

Poster:

Extending the Autosegmental Input Strictly Local Framework: Metrical Dominance and Floating Tones. Poster presented at the Society for Computation in Linguistics 2020 meeting.