CHAPTER 2 THEORETICAL AND TONAL BACKGROUND
2.4 Previous Analyses of Shanghai Tone Sandhi
2.4.1 Tonal Basics
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
– initial retention, realization on initial stress (Shanghai, Suzhou) – final retention, realization on final stress (Xiamen, Yantai)
It is noteworthy that in (19), Shanghai tone sandhi is grouped under the category in which tone retention and realization is in the same position, and therefore no tone movement occurs.
2.4 Previous Analyses of Shanghai Tone Sandhi
The major task of this section is to review the previous works on Shanghai tonal phonology. Beginning with an introduction to the tonal inventory of Shanghai, I will in turn discuss two approaches that used in the literature – an autosegmental approach (section 2.4.2), and an OT approach (section 2.4.3).
2.4.1 Tonal Basics
According to the literature (Xu et al. 1981, Xu et al. 1988, Qian 1992), Shanghai contains five lexical tones in citation forms, with three of smooth (Yinping, Yinqu and Yangqu) and two of checked (Yinru and Yangru). Smooth tones are solely carried by smooth syllables (i.e. open syllables and syllables closed by a nasal), whereas checked tones are, in complementation, carried exclusively by checked syllables (i.e. syllables ending in a glottal stop.) Checked tones are for this reason significantly shorter than smooth tones. The following table shows the distribution of the five tones with respect to syllable types1.
1 The transcriptions of the tones in Chao’s tone letters differ slightly in different sources. For example, Yinping is transcribed as 51 or 52 by different authors. Nonetheless, the same contour shape can be inferred; the tones are for this reason transcribed here in an impressionistic way. For a discussion of some limitations regarding the use of Chao’s tone letters, please refer to Duanmu (1990, 103ff).
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(20) Citation tones in Shanghai
Tonal types Smooth tones Checked tones
Syllable types CV, CVN CVʔ
Tonal categories Yinping Yinqu Yangqu Yinru Yangru
Tonal values HL MH LH Hʔ LMʔ
As a typical Wu Chinese Language, the five tones are divided into the high register and the low register. It has been instrumentally confirmed that such a register contrast is tightly related to phonation differences in Shanghai (Cao & Maddieson 1992, Ren 1992, Zhu 1999). More specifically, low-register tones occurs only on syllables with breathy voice (or murmur, I shall use them interchangeably), while high-register tones are compatible with syllables with modal (clear) phonation. The tone-phonation cor- relation is rooted in a well-known fact: breathy voice lowers F0 (see Hombert 1978, Homber et al. 1979, Gordon & Ladefoged 2001, Silverman 2002).
In addition, Ren (1992), through a series of acoustic measurements, perception tests, and physiological investigations, arrived at the conclusion in (21). In brief, he argues that such contrasts are manifested neither on the vowel nor on the tone because murmur is most salient in the vocalic onset and fades away before the middle point of a vowel. In particular, if murmur were represented with a feature under the tonal node (Yip 1993), it would be expected that breathy voice lasts throughout the entire tone- bearing portion. Therefore, phonation is the inherent property of the onset consonant.
(marked with ‘ ̤’ underneath a consonant). Zhu (1999) further notes that the modal vs.
murmur distinction is attested in sonorants and onsetless vowels as well.
(21) Phonation contrasts among the obstruent categories (Ren 1992:150) Initial position Non-initial positions
Phonation type Tonal register Phonation type Tonal register taa More adducted H-register More adducted Neutralized
‧
As (21) indicates, the phonation distinction and the register contrast are restricted to the initial position. In non-initial positions, murmur is lost, resulting in obstruent voi- cing and, more importantly, the neutralization of the tonal register.
2.4.2 An Autosegmental Approach
2.4.2.1 Tone Sandhi of Smooth Tones as Deletion and Reassociation
Despite the fact that Shanghai is probably the best-documented Wu Chinese Lan- guage in the phonological literature, most of the previous analyses have only focused on the tone sandhi of smooth tones (TSS, see Zee & Maddieson 1979, Yip 1980, 1989, Wright 1983, Selkirk & Shen 1990, Duanmu 1993, 1994b, among others). Virtually all of the works agree that TSS is derived from a two-step process from the perspe- ctive of autosegmental phonology (Goldsmith 1976, Pulleyblank 1986): (a) deletion of non- initial tones, and (b) re-association of the remaining initial contour tone to the syllable string from left to right in a one-to-one fashion, as illustrated in (22).
(22) Derivation: Yinping + T kʰe.sɨ ‘boiled water’
This analysis appears to entail the assumption that syllables are the tone-bearing unit (TBU), as posited implicitly by most analysts (for example, Selkirk & Shen 1990).
However, Duanmu (1993, 1994b) argued for a different view of the TBU of Shanghai, σ σ
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
in an attempt to explain why the re-distribution of contour tones occurs in Shanghai, and not in Mandarin. Two assumptions were made in his work for this purpose: (a) the TBU is mora instead of syllables in Chinese dialects, and (b) syllables in Shanghai are monomoraic, different from the case of Mandarin where most syllables (specifically, non-neutral-toned syllables) are bimoraic. Given these assumptions, Duanmu claimed that the contour re-distribution in (22) can be treated simply as a result of a one-to-one mapping between tone segments and TBUs.
The evidence drawn by Duanmu in support of those assumptions is the fact that Shanghai contains fewer bi-segmental rhymes than Mandarin. However, in Chapter 3 I will argue that such evidence is insufficient to assert that all syllables in Shanghai are light syllables, and as the present analysis will show, contour re-distribution may be unnecessary to be explained by stipulating mora as the TBU.
2.4.2.2 Yinru and Yinqu
Whether the TBU is posited to be syllable or mora, the traditional deletion-cum- reassociation approach to TSS fails to derive the tone sandhi of checked tones (TSC), especially in the case of Yinru, where the surface contour, M-H, has little to do with the initial level tone, H. Consequently, to account for the sandhi-patterns of Yinru, the traditional analysis has posited that both the presence of the M on the first syllable and the H on the second syllable ought to have their source underlyingly, therefore the need to posit a putative underlying representation that never occurs as a citation tone, i.e. /MH/. This assumption leads to a merger of different tonal categories: Yinru and Yinqu are underlyingly identical. This merger is consistent with the received view that a checked tone is an allotone of some smooth counterpart. Needless to say, having posited the tonal value of Yinqu, /MH/, as the underlying tone for Yinru, an extra rule is required to derive from it to the actual citation tone: /MH/ → [H], which is specific
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
to the context of checked syllables. This is basically the analysis – with minor varia- tions – of all current proposals (see Zee & Maddieson 1979, Yip 1980, Wright 1983, Selkirk & Shen 1990, Duanmu 1993, Jin 1997, Li 2003, among others).
Although this modus operandi is quite standard in linguistic practice, I submit that this putative base form is not only “nothing but an artifact of the conventional analysis” in the words of Chen (2000:224), but also brings in an asymmetry between the checked tones in two regards. First, given that Yinru and Yangru are carried by checked syllables in citation forms, it is quite puzzling that the contour-reduction rule is applied to the underlying contour tone of Yinru, but is blocked for that of Yangru.
Second, if both the checked tones are rising contours, a question arises as to why they behave differently in the sandhi process. Such problems reveal that to posit an abstract underlying form identical to what is manifested in Yinqu is of no help in accounting for TSC with the deletion-cum-reassociation analysis.
On the other hand, there are two arguments against the hypothesis that Yinru is only an allotone of Yinqu in Shanghai. First, according to Ping’s (2001b) experimental results, Yinru is phonetically implemented as a level tone, different from Yinqu and the other checked tone, Yangru, which are undoubtedly rising tones. The F0 tracings for the checked tones in isolation are shown below, where the normalized F0 curves are plotted against normalized time and so are logarithmic Z-scored (LZ) normalized F0 trajectories. Note that in the case of Shanghai, a contour tone requires at least a pitch excursion of 1 in LZ, according to the criterion set by Hsieh (2007: 81).
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(23) Normalized F0 (in Hz) and LZ Normalized F0 for citation checked tones
The (LZ-)normalized F0 tracks show that Yinru is a high-level tone (44). Note that
“the final pitch jump is presumably due to the glottalization” (Hsieh 2007: 79).
The other argument is drawn from the sandhi process in Old Shanghai, which can be heared only from people over 70 years old nowadays. The table in (24) shows that in Old Shanghai, the tone sandhi of Yinru is not parallel to that of Yinqu. As far as disyllabic domain is concerned, when the second syllable carries Yinshang, Yinqu, or Yangqu, the tone sandhi of Yinru exclusively surfaces as H-H, but the tone sandhi of Yinqu surfaces as another two sandhi patterns more commonly – M-HM and H-HL.
The asymmetry is more apparent when the second syllable carries checked tones: the surface tone pattern for Yinru tone sandhi is M-H, whereas the pattern for Yinqu tone sandhi is M-HL. These asymmetries appear to be in disfavor of the view that Yinru is grouped as the same toneme with Yinqu in Shanghai.
(24) Yinqu tone sandhi vs. Yinru tone sandhi in Old Shanghai σ2
σ1 Yinping Yinshang Yinqu Yangqu Yinru Yangru
Yinqu M-HL M-HL
H-H
H-HL H-H
M-HL
Yinru M-HL H-H H-H M-H
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
2.4.3 An OT Approach
In the literature, the OT approach to Shanghai tonology is twofold. One is to re- cast the deletion-cum-reassociation analysis in a non-derivational framework, works of which attribute the initial-tone preservation and the contour-tone re-distribution to positional faithfulness and positional markedness. For instance, Yip (2002) assumes that the initial syllable in Shanghai receives the metrical head, so that the preservation of the initial tone in the sandhi process can be explained with a higher-ranked faithful- ness constraint sensitive to the head position, MAX-HEAD-TONE. To account for the split of the initial contour, Yip further proposes LICENSE-CONTOUR, which is on the empirical basis that contour tones prefer to be licensed at the final position (cf. Zhang 2001; for other contour-licensing constraints, see Li 2003, Zoll 2003, inter alia).
This analysis itself runs into problems, however. On one hand, faithfulness cons- traints like MAX-HEAD-TONE entails that the metrical structure has been built in the input, which contradicts the basic premise in OT that phonological processes such as syllabification, metrification, etc. arise from the interaction of output-oriented cons- traints. On the other hand, what the contour-licensing constraints can do is nothing but to split the initial contour tone into level tones, the deployment of which is irrelevant to this constraint. Then, in face of a domain containing more than two syllables, it is impossible to guarantee that the general sandhi pattern in Shanghai, where the initial contour tone is re-distributed over the first two syllables, can be derived as expected.
Nonetheless, in the present analysis, positional faithfulness and positional markedness will be deemed the crucial factors in Shanghai tonology, and the problems they may cause will be one of those that we shall settle.
In addition to recasting the traditional deletion-cum-reassociation analysis, a few of the previous studies also employ the perspective of OT to cope with the tone sandhi of checked tones (TSC), which fails to be interpreted in a deletion-cum-reassociation
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
way. The relevant analyses can be grouped into two, with one treating TSC tones as a sub-grammar or minor system, and the other as a result of a single constraint ranking.
Both of them are discussed in what follows.
2.4.3.1 A Different Ranking for TSC
Chen (2000) regards TSC as the tonal manifestation of the weak~strong accent- tual pattern. Under this view, the high-pitch tone on the second syllable derived from TSC is the phonetic realization of the pitch accent. This analysis implies that TSC is a different system from TSS, since the pitch accent requires a set of constraint ranking regarding the stress-driven tone, say, *Hd/L » *Hd/M » *Hd/H (de Lacy 2002), which must be ranked higher than other tone-placement constraints in TSC, and must be ranked lower in TSS. That is, an account of different systems calls for constraint re- ranking, which is not favorable in the OT literature, though.
The pitch-accent analysis incurs another problem. Given that Shanghai is widely acknowledged to present a left-prominent system, then something must be said about the pitch accent on the second syllable. Accordingly, by assuming checked syllables and smooth syllables are heavy and light, respectively, Chen argues that the accent, or stress, does not fall on the leftmost syllable in TSC because that syllable is a checked syllable. To capture this phenomenon of quantity-sensitivity, he posits that the weight- to-stress constraint WSP (Prince 1990, defined in (25b)) crucially dominates the left- prominence constraint LEFTMOST, (defined in (25a)), as shown below.
(25) a. LEFTMOST: Assign stress on the leftmost syllable.
b. WSP: Stress cannot fall on a light syllable.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(26) Right-prominence in TSC (cf. Chen 2000:222)
σμσμμ SWP LEFTMOST
a. σμσ́μμ *
b. σ́μσμμ *!
This constraint ranking, however, leads to a couple of unwanted predictions, as listed in (27), all of which are unattested in Shanghai. Therefore, the pitch-accent analysis appears to be insufficient empirically.
(27) Wrong predictions made by the constraint ranking in (23)
a. The position of the stress holds consistent, whether the initial syllable carries Yinru or Yangru.
b. In forms lacking smooth syllables, the tress falls on the initial syllable.
c. In Yinru tone sandhi, the stress falls on the third syllable if that syllable is a smooth syllable.
d. In trisyllabic Yangru tone sandhi, the stress falls on the second syllable if that syllables is a smooth syllable.
In addition to Chen, Li (2003) also treats TSC as a distinct sub-system, so his analysis suffers from the same defect: the discrepancy between TSC and TSS cannot be attain- ed without re-ranking some constraints. To make matters worse, Li’s analysis can only interpret Yangru tone sandhi, leaving Yinru tone sandhi unsolved.
2.4.3.2 A Single Ranking for TSC and TSS
To account for the contour-tone displacement in Yangru tone sandhi, Hsieh (2007) proposes a smooth-syllable-specific IDENT(T), written as IDENT(T)INITIALLONGSYLL. The line of reasoning is that smooth syllables are longer than checked syllables, so it is
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
reasonable that the preservation of tonal identity is more important as far as smooth syllables are concerned. With this constraint higher than other tonal faithfulness cons- traints (e.g., the relational correspondence, RELCORR(x<y)), the displacement of the initial contour tone in Yangru tone sandhi is attained, without the resort to constraint re-ranking. In sum, although Yinru tone sandhi remains unsolved under the proposed constraint ranking, this account founded on a single ranking appears to be preferable than the previous ones.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
CHAPTER 3
YINRU TONE SANDHI
3.1 Descriptive Basics of Shanghai Tone Sandhi
Overall, tone sandhi in Shanghai is typical of the left-dominant tone systems in Yue-Hashimoto’s taxonomy (1987): only the initial tone matters in determining the surface tone pattern of the entire domain, regardless of the syllable counts therein; the underlying tones of the non-initial syllables are irrelevant and are lost in the process of tone sandhi. It is then possible for two otherwise contrastive tones to be completely neutralized if they occur after the same initial tone. As illustrated in (1), for example, an underlying contrastive pair, sɨHL ‘book’ in (1a) and sɨMH ‘water’ in (1b), associated with Yinping and Yinqu respectively, become non-contrastive with respect to their tone values when preceded by another syllable which is domain-initial and happens to be lexically associated with a tone in the same tonal category, say, Yangqu in this case.
(1) Neutralization of non-initial tones
a. du ‘picture’ + sɨ ‘book’ = du.sɨ ‘storybook’
LH HL L-H
b. du ‘big’ + sɨ ‘water’ = du.sɨ ‘flood’
LH MH L-H
Hence, Shanghai tone sandhi is remarkably different from other Chinese dialects in that the number of possible distinct tonal strings does not increase with the number of syllables. Rather, the tonal string remains invariant under the condition that the initial tone is identical. It follows that given the five lexical tones in Shanghai, the possible distinct tonal strings of a polysyllabic domain (of no matter how many syllables), can be reduced to only five tone-sandhi patterns, with three of smooth tones and two of
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
checked tones, depending on which lexical tone the initial syllable carries. The tone- sandhi patterns of smooth tones (TSS) are, under the traditional account, quite regular and transparent relative to the tone-sandhi patterns of checked tones (TSC), which are further comprised of two diverse sandhi patterns different in the derivation – the tone sandhi of Yinru (TSYI) and the tone sandhi of Yangru (TSYA). For ease of discussion, the description of these patterns will be broken down into three parts: the regular TSS is briefly sketched in section 3.1.1, and section 3.1.2 is devoted to one of the irregular TSC, namely, TSYI. The details of the other irregular pattern, TSYA, will be deferred until Chapter 4.
3.1.1 TSS as Tone Spreading
When the initial syllable in a given domain carries one of the three smooth tones, namely Yinping, Yinqu and Yangqu, the syllabic string of such domain surfaces as one of three TSS patterns. As shown in (2), the initial contour tone splits into two halves over the first two syllables in all of the TSS patterns, with the first half in situ, and the second half right on the following syllable. In other words, TSS appears to be an ex- tension of the initial contour, of which the array of data given in (3) is illustrative.
(2) TSS patterns: Qian (1992: 619)
Initial σ σσ σσσ σσσσ
Yinping HL H-L H-L-L H-L-L-L
Yinqu MH M-H M-H-L M-H-L-L
Yangqu LH L-H L-H-L L-H-L-L
‧
‘boiled water’ ‘opening ceremony’ ‘avaricious person’
base HL-MH HL-MH-Hʔ HL-LH-HL-HL
‘fifty’ ‘five o’clock’ ‘heavily swollen’
base LH-LMʔ LH-MH-HL LH-HL-LMʔ-MH
sandhi L-Hʔ L-H-L L-H-Lˀ-L
In derivational terms, the patterns of TSS are derivable via tone deletion and tone spreading in a straightforward way. Tone deletion eliminates all but the initial tone;
tone spreading then re-aligns the remaining tone segments with the syllables in a left- to-right, one- to-one fashion. Syllables left toneless after the tone spread, if any, are linked to L by default (Selkirk & Shen 1990, Yip 1980, 1995, Duanmu 1993, among others). This derivation is illustrated by examples like the following.
(4) Derivation of TSS
ɕjɔ.niɲ ɕjɔ.ka.ba.ʨʰi
‘child’ ‘shabby’
base tones MH.LH MH.HL.LH.MH
tone deletion MH. o MH. o. o. o tone spread M. H M. H. o. o default L -- M. H. L. L
o = unspecified for tone
‧
Although this derivation is applicable throughout the TSS patterns as we have seen, it runs into problems when a tonal string begins with a checked tone, which will be de- monstrated first through one of the checked tones, Yinru, in the next section.
3.1.2 TSYI as Tone Movement
When the domain-initial syllable is associated with Yinru (i.e. the high-register one of the two checked tones), the polysyllabic string of the entire domain invariably has a tonal envelope of TSYI. Consider the pattern in (5), where TSYI always has a M on the first syllable and a H on the second syllable, no matter how many syllables are included in the domain, of which the data is laid out below.
(5) TSYI pattern: Qian (ditto)
Initial σ σσ σσσ σσσσ
Yinru Hʔ Mˀ-H Mˀ-H-L Mˀ-H-L-L
(6) Examples of TSYI a. disyllabic
ɪʔ.dɔ haʔ.kaŋ
‘together’ ‘to utter nonsense’
base Hʔ-LH Hʔ-MH
‘all over the place’ ‘to talk through one’s hat’
base Hʔ-HL-MH-MH Hʔ-HL-LH-MH sandhi Mˀ-H-L-L Mˀ-H-L-L