5. Grammaticalized function words in Shanghainese
5.2 Prosodic invisibility in Shanghainese
5.2.2 Blocking of redistribution
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
second tone feature of that contour from the sponsor syllable. In my previous work (T.-C. Huang 2015), I suggested that this can be understood as motivated by reduction of tonal complexity. Given the marked status of contour tones (Yip 2002), the redistribution of melodies happens for flattening non-final contours due to the notion proposed by J. Zhang (2002, 2007) that final position enjoys the length privilege in licensing such a marked type of tones. This can be expressed by what Zoll (1998) calls
“negative positional markedness” ― formulated as *CONTOUR-NONFINAL ― which punishes every instance of non-final contours. For example, T1 /HL/ being non-final in a certain domain, as in (HL.o), is assessed with one violation by *CONTOUR-NONFINAL; it needs spreading in the redistribution fashion, ending up as (H.L). If the fall is licensed at the right edge of some constituent, as in ((HL).o), there is no pressure for it to be redistributed. The latter case is how the blocking/invisibility phenomenon is analyzed in my previous study: it is considered the lack of any pressure to get flattened by redistribution, which I will review in details in the next subsection.
5.2.2 Blocking of redistribution
While melody redistribution generally applies after tone loss in the derivation of tonal neutralization. The application of redistribution can in fact be independent of tone loss.
There are occasions where melody redistribution is blocked from applying. That is, the initial contour tone stays in situ, surfacing with no flattening, as represented in (29), in which case the second syllable, though toneless, serves as a non-recipient for the redistributed melodies; it acquires some pitch through the association of L% at the later point in the derivation.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(29) Melody redistribution
a. Normal application: /HL.T/ → [H.L]
b. Application blocked: /HL.T/ → [HL.o)
In my previous survey, this blocking is found to occur when the second syllable belongs to certain types of function words, and the occurrence itself is variable, aside from inter-speaker variation. (30) shows the results of the survey, which are adapted from T.-C.
Huang (2015: 170). The occurrence of blocking is shown in the shaded area. As we can see from the table, the main condition for blocking concerns the type of Function words.
Sentence-final particles and modifier marker are non-recipients, the former being major non-recipients (annotated as FncMA) and the latter a minor non-recipient (annotated as FncMI) in terms of the proportion of blocking. By contrast, the remaining types of function words are always recipients (annotated as FncR ) of the redistributed melodies.
Tonal types also play an important role here. Of the three long tones in Shanghainese, only the fall /HL/ and the small rise /MH/ are vulnerable to blocking, more of the former being blocked than the latter, which is attested by two facts―(i) /HL/ can be blocked before the modifier marker while /MH/ cannot, and (ii) before sentence-final particles, /HL/ has the preference for blocking over spreading while /MH/ has a reversed preference. Note that the types of function words under the scope of the survey roughly matches those types discussed in the current dissertation, except numerals, classifiers, and phasal complements. Yet, based on my personal observation and communication, the behavior of these types is the same as object pronouns and directional complements;
namely, they are normal recipients and never block redistribution.
‧
(30) Variable blocking of redistribution in Shanghainese Fnc types categories of function words, namely, FncMA, FncMI, and FncR, my previous analysis assumes different prosodizations according to their distinct syntactic distribution, as schematized in (31). We can see all three categories of function words fail to construct a prosodic word of their own (putting numerals aside for the time being), and thus they are enclitics varying only in the prosodic host they are adjoined to. FncR are adjoined to ω, as in (31c), while the other two are adjoined to the subtypes of ϕ―the major non-recipient, FncMA, being Φ-adjoined, as in (31a), and the minor non-recipient, FncMI, φ-adjoined, as in (31b). This difference in prosodization lies in their different syntactic configuration. The crucial difference between FncMA and FncMI is that the former is right aligned with “CmmP” (i.e. Comma Phrase) while the latter is not. The notion CmmP is Potts’ (2003, 2005) proposal, referring to a syntactic constituent as “logically and compositionally independent of ... the at-issue entailments” (Potts 2003: 119) of the proposition expressed by the surrounding sentence. Instances of CmmP, as Potts proposes, include supplementary expressions, such as non-restrictive relatives and parentheticals, etc. Selkirk (2005) further suggests an inclusion of root clause and dislocated constituent such as topic expressions, both being the very projection of FncMA. By contrast, restrictive relatives, as what FncMI normal heads, do not have the
‧
status of being a CmmP. This projection, CmmP, is argued to be the syntactic constituent corresponding to intonational phrase according to Selkirk (2005); that is why FncMA is immediately dominated by intonational phrase, while FncMA is not. As for FncR, on the other hand, is usually the sister node of the preceding lexical host in the syntactic
( )Φ (Maximal) phonological phrase
( )φ (Minimal) phonological phrase
( )ω Prosodic word
c. ω-adjunction for FncR
[ ... Lex FncR ]lexP
( )Φ (Maximal) phonological phrase
( )ω Prosodic word
Given the different prosodizations, the previous analysis contends that all else being equal, only the prosodic domains that are higher than ω on the hierarchy can tolerate the final contour tone, the tolerance being proportional to the domain level on the hierarchy. For example, intonational phrase has the highest tolerance, maximal phonological phrase the next highest, and so on.
On the other hand, to what extent a contour tone can be tolerated also differs even
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
in the same position of the same prosodic domain―/LH/ being the least tolerable at non-final positions and /HL/ the most tolerable. The previous analysis holds that this difference is linked to the relative markedness of these contour tones in relation to one another. The more marked a contour tone is, the higher-level the prosodic domain it requires for licensing in the final position, given the duration-based account put forward by J. Zhang (2007). Therefore, given the markedness scale of the three long tones: LH
≻ MH ≻ HL (where ≻ represents “more marked than), the previous analysis proposes that the threshold of being tolerable in the final position is φ for /HL/, Φ for /MH/ and ι for /LH/.In other words, in φ-final only /HL/ can have no pressure to be redistribute, in Φ-final both /HL/ and /MH/ can, and in ι-final all stay unchanged.
This line of reasoning is crucially captured in the previous analysis by proposing the negative positional markedness as formulated in (32a), which enforces flattening of different non-final contour tones by redistribution of the tone segments within or across varied prosodic domains. This contour tone licensing constraints interacts with faithfulness to contour tones, as given in (32b).
(32) Contour-licensing constraints (T.-C Huang 2015: 178) a. *τ_)π
Assign a violation mark for every non-final contour tone of type τ within a prosodic domain of type π, where τ ⊃ {big rise, small rise, fall} and π ⊃ {ι, ϕ, ω}.
b. PRESERVECONTOUR (or PRESC for short)
Assign a violation mark for every contour tone in the input that does not have a correspondent in the output.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
The previous analysis shows that the presence/absence of blocking can be accounted for by manipulating the ranking of the faithfulness, PRESC, with respect to the set of positional markedness against non-final contours: PRESC outranks *τi_)πx if τi is redistributed across )πx, and conversely, PRESC is outranked by *τj_)πy if τj is blocked by the boundary )πy from applying redistribution. Accordingly, the proposed constraint ranking is as shown in (33). All the subtypes of *BIGRISE_)π are ranked higher than PRESC, given the attested fact that T3 as a big rise /LH/ is never tolerable non-finally in a prosodic constituent of any level (33a). As for T2, the small rise /MH/, since it can be licensed at the final position of a maximal projection (Φ), hence the crucial ranking:
*SMALLRISE_)Φ» PRESC » *SMALLRISE_)ι (33b). Last, T1, the only falling tone /HL/, can be licensed at the final position of a minimal phonological phrase (φ), an even lower constituent, therefore PRESC is ranked between *FALL_)φ and *FALL_)Φ(33c).
(33) Ranking of the blocking a. T3: no blocking
{*BIGRISE_)φ, *BIGRISE_)Φ,*BIGRISE_)ι} » PRESC b. T2:blocking by )Φ
{*SMALLRISE_)φ, *SMALLRISE_)Φ}» PRESC » *SMALLRISE_)ι c. T1: blocking by )φ
*FALL_)φ » PRESC » {*FALL_)Φ, *FALL_)ι}
While the previous analysis is workable and the proposed constraints are well-motivated, it is deficient in the knowledge of CP-periphery and therefore is flawed by the problematic formation of intonational phrase. This warrants a reanalysis of FncMA, the sentence-final particles.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
5.2.3 Prosodic licensing of the non-root C particle
My previous analysis of the blocking of redistribution in tonal neutralization offers an account that is characteristic of (i) syntactically driven difference in prosodization, and (ii) positional licensing of marked structure. Compared to the latter property, which is insightful and is thus held on to in the present analysis, the former can be relatively deficient because it presupposes a perfect match between some kind of syntactic configuration and a certain way of prosodization. In fact, there could be two cases where such perfect match does not exist.
The first case is where the function words in similar types of syntactic configurations have different behaviors with regard to the blocking. Cases like this occur in final particles. While the previous analysis assumes that sentence-final particles are situated in similar syntactic positions and accordingly should have identical prosodizations, which then grants them the same prosodic status to decide whether the contour tone can be licensed in situ. That is, all the sentence-final particles should be equally likely to block/apply the redistribution. However, this is not the case.
As illustrated in (34), there are subtypes of sentence-final particles. One particular subtype, including those that Paul (2014) assumes are situated in Attitude C, namely, the head of the highest complementizer layer (e.g. a0, ya0, le0, etc.), are highly likely to block redistribution of the preceding T2 /MH/, whereas the other types, the Force C and Low C for the most part (e.g. leh5, va3, etc.), are not as typical non-recipient of redistributed tones as the Attitude C. They barely or sporadically block the redistribution. Asymmetry between these subtypes is surprising because based on the previous analysis, all the subtypes of sentence-final particles are complementizer and are sandwiched between the right edge of a phonological phrase and the right edge of an intonational phrase, which should have made the small rise /MH/ preceding them
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
equally well-licensed.
(34) Subtypes of sentence-final particles in root C: no difference in prosodization a. Sentence-final particles in Attitude C: normal blocking
b. Sentence-final particles in Force C/Low C: sporadic or no blocking
This asymmetry might urge one to distinguish between the Attitude C and the other lower subtypes by means of, for example, assuming that only the highest root C corresponds to intonational phrase. This is exactly what Selkirk (2011) proposed to define syntax-phonology correspondence at clausal level. She suggests that, assuming multiple layers of complementizer in CP-periphery in the spirit of Rizzi (1997), the clause in correspondence with intonational phrase should be defined as the complement
of the highest layer, ForceP in the framework she adopts. This is illustrated in (35b).
(35) Clausal correspondence a. MATCH-CLAUSE:
A clause in syntactic constituent structure must be matched by a constituent of a corresponding prosodic type in phonological representation, call it ι.
b. [ForceP Force0 [ illocutionary clause ι] ] [ForceP Force0 ( illocutionary clause ι) ]
Given that the highest layer assumed in this dissertation is Attitude C, we may propose [[Ngu3 soa1=geh5 ce2TP] =a0 AttCP] ‘This dish was cooked by me.’
I cook=MOD dish =SFP
M Φ) H ι)
[[qieh4 coa2ve3 hoa2 TP] =va3ForceCP] ‘How about fried rice?’
I fried.rice dish =SFP
*/?MH Φ) H ι)
*/?MH Φ) o ι)
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
that only the complement of Attitude C forms an intonational phrase, following Selkirk’s proposal. This move makes the sentence-final particles in Attitude C outside the intonational phrase, because it is not a part of its own complement. The implementation is given in (36). As we can see, by doing so, sentence-final particles in Attitude C is successfully distinguished from the other subtypes of its like. The small rise /MH/ followed by an attitude final particle is now at the final position of an intonational phrase (ι), whereas the same contour tone followed by the other types of final particles remains at the final position of a maximal phonological phrase (Φ). The blocking asymmetry now can be explained by assuming that a small rise is well-licensed only in ι-final.
(36) Subtypes of sentence-final particles in root C: different prosodization a. Sentence-final particles in Attitude C: normal blocking
b. Sentence-final particles in Force C/Low C: sporadic or no blocking (=34b)
This revision, however, runs into the other problem that the previous analysis may pause, that is, the case where function words share the same tonal behavior and yet they are in distinct syntactic positions which would give rise to different prosodization for them. An example of such a case is the sentence-final particle geh5. This final particle is an equivalent to Mandarin de0 in the propositional assertion construction, which is analyzed in Paul (2014) as a non-root C, in contrast with the abovementioned subtypes [[Ngu3 soa1=geh5 ce2TP] ForceCP] =a0 AttCP] ‘This dish was cooked by me.’
I cook=MOD dish =SFP
M Φ) ι) H
[[qieh4 coa2ve3 hoa2 TP] =va3ForceCP] ‘How about fried rice?’
Eat fried.rice okey =SFP
*/?MH Φ) H ι)
*/?MH Φ) o ι)
‧
of root final particles. “Non-root” means that this particle is affiliated with an embedded clause, a position that cannot correspond to intonational phrase according to Selkirk (2011). In other words, geh5 as a non-root C final particle, its syntactic and prosodic statuses are similar to those of Low C/Force C particles, and yet it tonal behavior is not the like of them. Rather, its behavior with respect to blocking is more like Attitude C particles. This is illustrated in (37), the situation of Attitude C particles and Force C/Low C particles are repeated here for comparison.
(37)
a. Sentence-final particles in root Attitude C: normal blocking (=36a)
b. Sentence-final particles in root Force C/Low C: sporadic or no blocking (=36b)
c. Sentence-final particles in non-root C: optional blocking
This problem reveals that it is impracticable to define the prosodic invisibility by purely syntactic structure. Rather, we need a mechanism specific to the types of function words.
The prosodic licensing constraints serve as such a mechanism. The idea is that suppose the grammaticalization degree of the non-root C particle geh5 is higher than 6, by which it is a more grammaticalized element than Low C/Force C perhaps slightly lower than [[Ngu3 soa1=geh5 ce2TP] ForceCP] =a0 AttCP] ‘This dish was cooked by me.’
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Attitude C particles, whose degree of grammaticalization is given as 6.5. This Assumption that non-root C particle is highly grammaticalized is not a stipulation.
Pragmatically, it does not denote sentential aspect/tense/modality, so nothing like low C particles. It mainly conveys two kinds of meaning: one denotes the assertive declarative sentence type, and in this sense is close to Force C particles, while the other expresses subjective soft certainty and thus is more like Attitude C. In view of the correlation between subjectification and grammaticalization, the grammaticalization degree of the non-root C particle geh5 may be somewhere in between. Therefore, it is assumed to be categorized as Fnc>6 here.
With this higher grammaticalized status, we can follow the same rationale for the extrametricality in Sixian Hakka (see §3.2.2) and Taiwanese (§4.1.3.1), assuming that the invisible non-root C particle geh5 (Fnc≥6) must be immediately preceded by some right edge of a intonational phrase and it is not included in any instance of intonational phrase. In other words, the non-root C particle are clause-level extrametrical for not being parsed in a intonational phrase. This idea can be schematized in (38). This configuration clearly shows that the preceding small rise /MH/ is ensured to be properly licensed at the final position of a prosodically high-level domain, and thus there is no pressure for this contour tone to be redistributed.
(38) Extrametrical to the paradigmatic tone sandhi in Taiwanese ( … σ MH ι) Fnc≥6
This can also be captured by the alignment constraint we proposed for Sixian Hakka and Taiwanese, but this time, it is specific to grammatical words with degree of grammaticalization lower than or equal to 6, and in that way it dictates that the right
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
edge of a given intonational phrase be licensed by the right edge of a word in that degree, (i.e. Fnc≤6), which includes Low C particles/Force C particles, as well as the other types of function words that have even lower degree of grammaticalization.
(39) ALIGN-R(ι, Fnc≤6):
Assign one violation mark for every intonational phrase (ι) which is not right-aligned with some word with degree of grammaticalization lower than or equal to 6 (Fnc≤6).
With this alignment constraint ranked above ALIGN-R(ι, CLAUSE), the non-root C particle is manipulated to be located outside an instance of intonational phrase, by which the preceding MH can stay in situ. The following tableau illustrates this desirable results.
(40) ALIGN-R(ι, Fnc≤6) ≫ ALIGN-R(ι, CLAUSE) INPUT: (=37c)
/INPUT/ ALIGN-R(ι, Fnc≤6) ALIGN-R(ι, ForceP(Comp))
CAND 1 *!
CAND 2 *
5.3 Summary
This chapter addresses grammaticalized function words in Shanghainese. Shanghainese can be said to be phonologically highly attrited language in that the vast majority of [[Nong3 me1 we3teh4 ti2ce2 TP] =geh3CP(-root)] TP] ‘You’re a foodie.’
You quite capable to.order.dish =SFPFnc>6
CAND 1: */?MH ϕ) H Φ) ι)
CAND 2: */?MH ϕ) Φ) ι) o
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
function words are non-prominent in terms of tonal neutralization, and that the distinction between lexical and functional categories is obscured by the neutralization of non-initial tones of lexical words. Both can be analyzed as a result of stress assignment and prosodic licensing, captured by two sets of constraints ranking: ALIGN -R(ω, Fnc≤1).≫ SP-MAX-X0 and FTBIN ≫ σ-WITH-Σ (E). Besides, Shanghainese also exhibit prosodic invisibility in highly grammaticalized word, namely, sentence-final particles. This is attained through the constraint ranking ALIGN-R(ι, Fnc≤6) ≫ ALIGN -R(ι, CLAUSE).
‧ 6. Theoretical Remarks on Typology
Having settled the scenarios in each language, this chapter attempts to give a typological overview of those languages and draw some theoretical implications. The first two sections summarize the constraint rankings of neutral tone and prosodic invisibility. § 6.3 gives remarks on the prosodic well-formedness.
6.1 Neutral tone
Neutral tone has been assumed as derived from syllables unparsed into prosodic word.
This is captured by the conception of prosodic licensing, formulated as ALIGN-R(ω, Fnc≤n) ≫ SP-MAX-X0 ≫ ALIGN-R(ω, Fnc≤n-1). The number, which denotes degree of grammaticalization, is divergent across languages. This divergence is summarized below:
(1) Neutral tone typology
Languages High-ranked licensing constraint Neutral-toned types Sixian Hakka ALIGN-R(ω, Fnc≤6) ≫ SP-MAX-X0 ≫
‧
investigated in this dissertation, as shown in table (2), where the check mark “” means that this class of function words is neutral toned. The proposed constraint system thus embodies in a formalized way the general impression that Wu Chinese (as represented by Shanghainese), Min Chinese (as represented by Taiwanese) and Mandarin Chinese (as represented by Standard Mandarin) are particularly rich in stress-related phenomena such as neutral tones, while Hakka Chinese (as represented by Sixian Hakka) and Yue Chinese (as represented by Cantonese) are famous for scarce manifestation of stress and neutral tone phenomena.(2) Overview of neutral toned classes of function words across languages subclasses
Furthermore, with the edge alignment constraints formulated in stringency form, the currently proposed system captures both the universal and language-specific nature of
Furthermore, with the edge alignment constraints formulated in stringency form, the currently proposed system captures both the universal and language-specific nature of