Contour-split Strategy

The split of the initial contour tone in the output form of TSS raises the question of whether the contour-split strategy is also available for Pattern B. It is important to address this issue, since there is the potential that using contour-split strategy to avoid the violation of COINCIDE(Contour, PrWd-Final) may undermine the analysis already established for Pattern B in the last section.

To begin with, we present a total of six possibilities that adopt the contour-split strategy for Pattern B to elude the violation of COINCIDE(Contour, PrWd-Final) (with three possible splitting patterns in trisyllabic words multiplied by two foot-parses in the discussion). All of them are contained in tableau (25) as the intended losers, since the intended winner adopts the strategy of moving the initial contour tone away from its source under the unbounded foot-parse. COINCIDE(Contour, PrWd-Final) becomes irrelevant to the choice between the word-final contour of the intended winner and the split level tones of all of the intended losers (25a-f), so it is omitted from this tableau.

(25) Possible splitting patterns for Pattern B in a comparative format Input Winner Losers ^ANCHOR-L

‧ 國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

According to Prince (2002a) and McCarthy (2008), in a comparative tableau like (25) every loser-favoring constraint ought to be dominated by some winner-favoring cons- traint. It follows that we need to check that every L has some W to its left across a solid line. This reveals that ANCHOR-L(tσ1, σ́) – which has been unranked with respect to the other three – must dominate LOCAL and FTBINMAX, or else each of the contour- split patterns in (25a), (25d) and (25e) would have an L without some W dominating it.

To put it another way, if the ANCHOR-L(t_σ1, σ́) were to be ranked below ^LOCAL and FTBINMAX,those unwelcome candidates would otherwise threat to beat the intended winner in the long run. Only with ANCHOR-L(t_σ1, σ́) ranked higher can we readily eliminate all of the three intended losers, together with those in (25b) and (25c). There is then no problem with the first five contour-split patterns. However, the remaining candidate still threatens to subvert our analysis, as revealed in row (25f), where there is an L that is not dominated by a W. This candidate can be a threat to the intended winner because it fares better on FTBINMAX, and is equally perfect in the eyes of ANCHOR-L(t_σ1, σ́) due to the fact that the first part of the split tone happens to coin- cide with the metrical head. The presence of this unwelcome candidate suggests that the current ranking remains inadequate, so I propose an anchoring constraint for the intended winner to beat candidate (25f). The definition of this constraint is as follows.

(26) ANCHOR-R(t_σ1, FT)

Let input tonal tier = t1t2t3 … tn and output tonal tier = T1T2T3 …Tm.

Assign one violation mark for every t_x if t_x is at the right edge of the tonal tier, which is associated with the word-initial syllable, and there is no Ty at the right edge of the metrical foot where t_x R Ty (McCarthy & Prince 1993a, 1995).

As tableau (27) illustrates, with ANCHOR-R(tσ1, FT) also ranked above FTBINMAX, we

‧

can readily eliminate the problematic candidate (27a) (copied from candidate (25f)), because the second half of its initial tone is not enclosed by the foot at all, and hence is in violation of the higher-ranked ANCHOR-R(t_σ1, FT). This forecloses the final way in which the initial contour tone of pattern B could take advantage of the splitting strategy to elude the violation of COINCIDE(Contour, PrWd-Final).

(27) ANCHOR-R(tσ1, FT) » FTBINMAX

Before we leave the discussion of Pattern B, something remains to be said about the newly proposed constraint, ANCHOR-R(tσ1, FT). First, given that none of the surface patterns of Shanghai tone sandhi infringes the stipulation made by ANCHOR-R(t_σ1, FT), as demonstrated in table (28), we can safely conclude that ANCHOR-R(tσ1, FT) must be undominated.

(28) No violation of ANCHOR-R(tσ1, FT) throughout

TSS TSYI Pattern B

Input σ_μμσσ σ_μσσ σ_μσσ

Output (σ́^μμσ)σ (σμσ́)σ (σμσσ́)

Second, ANCHOR-R(tσ1, FT) usurps the function of other two constraints. One of them is LOCAL, which was proposed in Chapter 3 to explain the locality of contour-split in TSS. ANCHOR-R(tσ1, FT) subsumes this effect because the split contour tones are pro-

L

‧

The other constraint that can be replaced is MAX(T)σ1. The tableau in (30) reveals that ANCHOR-R(t_σ1, FT) together with ANCHOR-L(t_σ1, σ́) amounts to the protection effect of MAX(T)σ1 on the initial tones if both of the anchoring constraints dominate *t.

(30) Substitution for MAX(T)_σ1

σμμ σ σ ANCHOR-L

In sum, ANCHOR-R(tσ1, FT) takes over the necessary work that has been attributed to LOCAL and MAX(T)σ1 respectively. The reasoning of Occam’s Razor demands that we dispense with those two constraints from the present analysis since ANCHOR-R(tσ1, FT) is proven to be sufficient.

H

‧

In this section I proceed to deal with Pattern A of TSYA. Section 4.3.1 gives an account of lapse licensing for the emergence of Pattern A in quadrisyllabic or longer words. Section 4.3.2 provides a co-phonological approach to the variable thresholds of Pattern A.

4.3.1 The Emergence of Pattern A

As introduced in section 4.1.1, the manipulation of the retained tone in Pattern A conforms to the processing of TSS, causing the initial contour tone to be re-distributed over the first two syllables. This contour-split pattern goes beyond our expectation, however, given that the initial contour tone, Yangru, is carried by a light syllable and therefore its best way of satisfying COINCIDE(Contour,PrWd-Final) must be to move as a whole to the end of the word (i.e. surfacing as Pattern B), given the top-ranked ANCHOR-L(t_σ1, σ́) and C^OINCIDE(σ́μ,FT-Final). This less harmonic output leads to the question of what motivates the contour split of Pattern A, as this operation seems im- possible under the current constraint ranking, as illustrated in (31). Even if we sup- pose that Pattern A obeys COINCIDE(Contour,PrWd-Final), COINCIDE(σ́^μ, FT-Final), ALLFTL and ANCHOR-R(t_σ1, FT), Pattern A will lose to Pattern B on ANCHOR-L(t_σ1, σ́) because the first part split from the initial contour tone (L) is inevitably misaligned by one syllable with the foot- head.

(31) Failure of Pattern A on ANCHOR-L(tσ1, σ́)