從優選理論分析東勢大埔客語之連讀變調

(1)

行政院國家科學委員會專題研究計畫成果報告

從優選理論分析東勢大埔客語之連讀變調

計畫類別：個別型計畫計畫編號： NSC94-2411-H-004-032- 執行期間： 94 年 08 月 01 日至 95 年 07 月 31 日執行單位：國立政治大學語言學研究所計畫主持人：蕭宇超報告類型：精簡報告處理方式：本計畫可公開查詢

中華民國 95 年 10 月 23 日

(2)

行政院國家科學委員會補助專題研究計畫

; 成果報告

□期中進度報告

（計畫名稱）

從優選理論分析東勢大埔客語之連讀變調

An Optimality Theory Approach to Tone Sandhi in Dapu Hakka

計畫類別：; 個別型計畫 □ 整合型計畫

計畫編號：NSC 94－2411－H－004－032－

執行期間：94 年 8 月 1 日至 95 年 7 月 31 日

計畫主持人：蕭宇超

共同主持人：

計畫參與人員：黃婷、邱昀儀、吳耿彰、彭國書

成果報告類型(依經費核定清單規定繳交)：;精簡報告 □完整報告

本成果報告包括以下應繳交之附件：

□赴國外出差或研習心得報告一份

□赴大陸地區出差或研習心得報告一份

□出席國際學術會議心得報告及發表之論文各一份

□國際合作研究計畫國外研究報告書一份

處理方式：除產學合作研究計畫、提升產業技術及人才培育研究計畫、

列管計畫及下列情形者外，得立即公開查詢

□涉及專利或其他智慧財產權，□一年□二年後可公開查詢

執行單位：國立政治大學語言學研究所

中華民國 95 年 8 月 28 日

(3)

An Optimality Theory Approach to Tone Sandhi in Dapu Hakka

Yuchau E. Hsiao

National Chengchi University

Abstract. This paper discusses tone sandhi in Dapu Hakka under the

framework of Optimality Theory. Assuming that the base tone and the sandhi tone is co-indexed, I posit a set of the tonal constraints, which incorporate the ideas of positional faithfulness and anti-faithfulness. OO correspondence and prosodic phrasing play a role in the tonal mapping as well. Alternative readings are then subject to a couple of floating constraints.

Keywords: tone sandhi, Dapu Hakka, OT, positional faithfulness, anti-faithfulness

Anchoring, foot, OO correspondence, floating constraints

1. Introduction

This paper investigates tone sandhi in Dapu Hakka, taking an Optimality Theory approach (Prince & Smolensky 1993). I establish a corpus of 2122 lines, which includes 1150 Yangping words, 1090 Yinping words and 1008 Yinqu words. The corpus is coded with information such as tone, foot, syntax and word category. This research looks into two issues. First, what is the difference between Ping sandhi and Qu sandhi? Second, how are the stylistic alternatives derived?

The rest of the paper is organized as follows. Section 2 sketches the tone system of Dapu. Section 3 reexamines the OCP effects and employs the idea of anchoring to account for the tone sandhi phenomena. Section 4 considers alternative readings in casual speech and formal speech from the perspective of OO correspondence. Section 6 discusses the function of prosodic structure in tone sandhi and concludes this paper by summarizing a partial constraint ranking.

2. Tone System of Dapu

There are seven base tones in Dapu Hakka. Six of them pertain to the traditional tone categories, including Yinping, Yanping, Yinshang, Yinqu, Yinru and Yangru. In addition, there is a synchronically developed new category, referred to as Chaoyinping. Consider the tone table in (1).

(1) Tone Inventory

Tone Categories Base Tones Sandhi Tones Examples

(4)

Yinping M MH fong ‘wind’

Yinshang ML keu ‘dog’

Yinqu HM H thien ‘electricity’

Yinru M- muk ‘eye(s)’

Yangru H- hok ‘learn’

Chaoyinping MH chia ‘umbrella’

The tone values in table (1) are recorded according to our fieldworks, based on three informants, Liu A-Xiang, Zhu A-Ding and Qio Jin-Sheng.1 In terms of the base tones, there are two rising tones (LM and MH), two falling tones (ML and HM), one mid level tone (M), and two checked tones (M- and H-). Like many Chinese dialects, Dapu Hakka distinguishes two types of syllables, checked syllables and smooth syllables. Checked syllables are those end in a voiceless stop, such as p, t, k and ?, while the rest are referred to as smooth syllables. The tones carried by checked syllables are known as checked tones. Although vowel length in Dapu Hakka is noncontrastive, the checked tone is phonetically short with a voiceless coda. The drop of the coda, however, will result in the compensatory lengthening of the vowel. In this case, the checked syllables become smooth syllables, and the checked tones, M- and H-, display falling pitches, ML and HL.2 What this means is that the checked tones are not level tones, but they are falling tones in nature. Accordingly, four of the base tones are [+falling], including HM, ML, M- and H-, a fact that is crucial to Yinqu tone sandhi in this dialect (as will be discussed in section 3).

The base tones do not always surface in the output. As shown in table (1), there are only three sandhi tones, M, MH and H, which occur in limited contexts. Consider the tone distribution in (2): (2) Tone Distribution T1 T2 LM M ML HM M- H- MH LM M LM LM LM LM LM LM M MH M MH M MH M M ML ML ML ML ML ML ML ML HM HM HM H H H H HM M- M- M- M- M- M- M- 1

The three informants are native speakers of Dongshi Dapu, respectively aged 59, 61 and 63. Two of them are housewives, while the other is a farmer. The fieldworks are conducted by myself and my research assistant Chiu Yun-yi (who is also a native speaker).

2

Similar patterns can be found in other Chinese dialects. For example, in Southern Min, the word to? ‘table’ carries a checked tone H- (sandhi form), which becomes HL in to-kha “table foot’, where the glottal stop is

(5)

H- H- H- H- H- H- H-

H-MH MH MH MH MH MH MH MH

Three patterns are observed in (2).3 First, LM occurs with its sandhi form, M, in the output before another LM. Second, M surfaces as MH when it precedes a LM, ML or M-. Third, HM changes to H when it is followed by a M-, H-, ML, or another HM. The following section will look into these patterns from a constraint-based perspective.

3. Anchoring and Nonanchoring

Tone sandhi in Chinese dialects is not a simple matter of output markedness, but it requires correspondence of tone category. In other words, the output tone must be co-indexed with the input tone; for example, the Yangping base tone is co-indexed with the Yangping sandhi tone. There is an abstract principle governing this pattern, as formalized in (3):

(3) IO-CoindexT:

If x=input and y=output, then y must be co-indexed with x.

This constraint sets forth two restrictions. First, the input-output mapping is only a choice between a base-sandhi pair within the same tone category, the LM-M pair, etc. Second, a base tone that does not have a sandhi form must remain invariant in the output; any change of the surface tone value would be a violation of the tone category. IO-CoindexT is always respected, and thus undominated. On the other hand, a constraint like (4) is ranked at the bottom, since tone sandhi by nature allows the output tone to have different values from the input tone.

(4) IO-Ident-T: the input tone and the output tone must be identical.

Yangping tone sandhi in Dapu bans adjacent LM tones in the output, and a handy resolution would be to appeal to OCP, for example, to posit a constraint like OCP-LM. However, a problem would arise from such formalization. Namely, while preventing a sequence of LM tones, OCP-LM can not predict which of the LM tones should change, and which should not. To provide a better account, we need to have a closer look at Chinese tones. In most Chinese dialects, each syllable carries a tone, and each syllable is a morpheme or a word. At this point, each tone is a morpheme, and tonal adjacency by nature follows from anchoring of morphological (and/or prosodic) edges. Anchoring, conceptually developed from alignment (McCathy and Prince 1993a,b, Prince. and Smolensky 1993, Selkirk 1995)4 and

3

For discussions of the Dapu tone sandhi rules under the derivational tradition, cf. Tung (1994) and Chiang (1998, 2002).

4

(6)

variously elaborated (Ito and Mester 1995, McCathy 2003, among others), may refer to the coinciding of same edges, as in (5a), or opposite edges, as in (5b).

(5)a. …x] b. … x][y … …y]

The schemas above are essential in capturing the generalization of tone sandhi. In particular, (5b) constitutes the basis for the constraints in (6-7):

(6) ¬ IO-Anchor-LM(x][y):

If x=LM and y=x, then the fact that x in the input is identical to x in the output is not true.

(7) IO-Anchor-LM(y][x):

If x=LM and y=any tone, then x in the input is identical to x in the output.

¬IO-Anchor-LM(x][y) reflects not only an OCP effect,5

which bans neighboring identical elements in the output, as in (8a), but also a process of anti-faithfulness, which allows a morpheme to idiosyncratically cause alternation of a neighboring morpheme (Alderete 2001), as in (8b). IO-Anchor-LM(y][x) requires positional faithfulness (Beckman 1998); that is, noninitial tones in the input are identical to noninitial tones in the output, as in (8c).

(8)a. *LM LM b. LM LM (input) c. … LM (input) d. LM LM (input) | | | |

*LM M (output) … LM (output) M LM (output)

¬ IO-Anchor-LM(x][y) dominates IO-Anchor-LM(y][x) to ensure that the first but not the second of a pair of LM tones surfaces with its sandhi form, as in (8d). The tableau in (9) illustrates the function of the partially ranked constraints:

(9) tshim mo ‘can’t find’ I: LM LM O: M LM

LM LM IO-CoindexT ¬IO-Anchor-LM(x][y) IO-Anchor-LM(y][x) IO-Ident-T

a. LM LM *! ☺b. M LM * c. M M *! ** d. LM M *! * * e. HM LM *! *

referring to the coinciding of prosodic and morphological edges.

5

For example, Zhang (1997) posits the constraint *33, and Lin (1999) proposes the constraint *L/L to account for Mandarin tone sandhi.

(7)

Candidate (a) violates ¬IO-Anchor-LM(x][y), as the first LM surfaces without change, whereas the second LM in candidate (c) occurs with its sandhi form in the output, in violation of IO-Anchor-LM(y][x). Both of the constraints are violated in candidate (d), where the surface pattern is LM M. As to candidate (e), it is eliminated by IO-CoindexT, since the Yangping changes to Yinqu, HM. As a result, candidate (b) is selected as the optimal output (indicated by the happy face,☺).

Yinping tone sandhi exhibits a similar pattern, except that it is triggered in three contexts, namely, before LM, ML or M-. These three tones belong to a natural class: they are contour tones, [+C],6 and low registered, [-U]. The base tone of Yinping is a low-registered mid level tone, M-U, which does not surface in the three contexts above, and the constraint in (10) serves to account for this fact.

(10) ¬IO-Anchor-M-U(x][y):

If x= M-U and y=[+C, -U], then the fact that x in the input is identical to x in the output is not true.

This constraint ranks between IO-Ident-TCat and IO-Anchor-LM(y][x), as in (11-13), where the low-registered mid level, M, correctly surfaces as a high-registered rising, MH, before a low-registered rising or falling.

(11) mai ng ‘buy fish’ I: M LM O: MH LM

M LM IO-CoindexT ¬IO-Anchor-M-U(x)[y] IO-Anchor-LM(y)[x] IO-Ident-T

a. M LM *!

☺b. MH LM *

c. M M *! * *

d. HM LM *! *

(12) san-ten ‘top of the hill’ I: M ML O: MH LM

M ML IO-CoindexT ¬IO-Anchor-M-U(x)[y] IO-Anchor-LM(y)[x] IO-Ident-T

a. M ML *! ☺b. MH ML * c. M M *! * d. HM ML *! * 6

(8)

(13) shiu-toh ‘desk’ I: M M- O: MH

M M- IO-CoindexT ¬IO-Anchor-M-U

(x)[y] IO-Anchor-LM(y)[x] IO-Ident-T

a. M M- *!

☺b. MH

M-*

c. M M- *! *

d. HM M- *! *

This partial constraint ranking also predicts two facts. First, a low-registered mid level tone remains invariant before another low-registered mid level tone, as in (14). Second, any base tone that does not have a sandhi form is retained in the output, as in (15).

(14) kin-shu ‘scripture’ I: M M O: M M

M M IO-CoindexT ¬IO-Anchor-M-U(x][y) IO-Anchor-LM(y][x) IO-Ident-T

a. MH M *!

☺b. M M

c. M MH *!

d. HM M *! *

(15) lia kai ‘this’ I: ML HM O: ML HM

ML HM IO-CoindexT ¬IO-Anchor-M-U(x][y) IO-Anchor-LM(y][x) IO-Ident-T ☺a. ML HM

b.M HM *! *

c. ML HM *! *

d. H M *!* **

Compare now the tableaux in (9) and (11). The former derives an M before LM, but the latter bans an M before LM. Hence there is an inconsistency in the output. This inconsistency can be resolved by assuming a register difference. Namely, the M in (9), the sandhi form of Yangping, is a high-registered tone, while the M in (10), the base form of Yinping, is a low-registered tone. This analysis has two advantages. First, it correctly predicts that only M-U is subject to the constraint in (10), but M+U is not. Second, it makes possible a consistent pattern that in both Yangping and Yinping, the I-O mapping involves a change of register, i.e., from [-U] to [+U].

Whereas the base tones of Yangping and Yinping are low-registered, that of Yinqu is high-registered. Yinqu tone sandhi requires a HM-to-H mapping before M-, H-, ML, or another HM. As mentioned in section 2, the checked tones are not level tones, but have the feature [+F]. The anchoring and non-anchoring of HM, therefore, reflect an OCP effect of [+F], as stated in

(9)

the constraints below:

(16) ¬ IO-Anchor-HM(x][y):

If x= HM and y=[+F], then the fact that x in the input is identical to x in the output is not true.

(17) IO-Anchor-HM(y][x):

If x=HM and y=any tone, then x in the input is identical to x in the output.

Again, a constraint like (16) indicates a requirement of positional anti-faithfulness. The schema in (18) is considered a violation of this constraint, where non-anchoring of HM is enforced regardless of the presence or absence of the following [+F] feature in the output. The constraint in (17) then requires faithfulness of non-initial HM tones.

(18) HM [+F] |

*HM [αF]

The tableaux in (19-23) illustrate the non-anchoring of HM in the relevant four contexts.

(19) thien-fa ‘telephone’ I: HM HM O: H HM /HM HM/ IO-Ident- TCat ¬ IO-Anchor- HM(x][y) IO-Anchor- HM(y][x) IO-Ident-T a. HM HM *! ☺b. H HM * c. H H *! ** d.HM H *! * * f.HM MH *! * * * (20) so-pa ‘broom’ I: HM ML O: H ML HM ML IO-Ident- TCat ¬ IO-Anchor- HM(x][y) IO-Anchor- HM(y][x) IO-Ident-T a. HM ML *! ☺b. H ML * c. H H *! ** d.HM H *! * * d.HM LM *! * * *

(10)

(21) mai nioh ‘sell meat’ I: HM M- O: H HM M- IO-Ident- TCat ¬ IO-Anchor- HM(x][y) IO-Anchor- HM(y][x) IO-Ident-T a. HM M- *! ☺b. H M-* c. H H *! ** d.HM H *! * * d.HM LM *! * * *

(22) tso sit ‘make food’ I: HM H- O: H HM H- IO-Ident- TCat ¬ IO-Anchor- HM(x][y) IO-Anchor- HM(y][x) IO-Ident-T a. HM H- *! ☺b. H H-* c. H H *! ** d.HM H *! * * d.HM LM *! * * *

3. OO Correspondence

Tone sandhi in trisyllabic strings or longer often involves alternative readings. Dapu Hakka distinguishes casual and formal readings, which can be characterized by way of OO correspondence (McCarthy and. Prince 1995). A relevant constraint is stated in (23).

(23) OO-Anchor-T(x, [, y, [):

If x= reference output and y=derived output,

then the initial tone in x is identical to the initial tone in y.

In Chinese dialects, a trisyllabic expression usually consists of a disyllabic word (or phrase) and a monosyllabic word. For example, the VP tshim nam-nin ‘find a man’ is made of the verb tshim ‘find’ and the noun nam-nin ‘man’. In terms of tone sandhi, correspondence between the mother and the daughters plays a role in deriving surface readings.

(24) [T1 [T2 T3]] IO T1 T2 T3 [T1], [T2 T3] OO

(11)

As schematized in (24), the monosyllabic [T1] and the disyllabic [T2 T3] serve as the reference outputs (RO), which may or may not affect the trisyllabic output. OO-Anchor-T(x, [,

y, [) indicates positional faithfulness and requires that the initial tones, T1 and T2, of the

reference outputs be identical to those in the trisyllabic derivative. This constraint ranks low in casual speech, but ranks high in formal speech. As in (25a,b), the verb tshim of the right-branching structure surfaces with its sandhi form, M, in the casual output (CO), but retains its base form, LM, in the formal output (FO).

(25) [tshim [nam-nin]] ‘find a man’

a. I: LM LM LM RO: LM, M LM CO: M M LM LM LM LM ¬IO- Anchor- LM(x][y) IO- Anchor- LM(y][x) OO- Anchor-T (x, [, y, [) IO- Ident-T ☺a. M

M LM * * ** b. LM M LM *! * * c. M LM LM *! ** * d. LM LM LM *!* * e.M M M **! * *** b. I: LM LM LM RO: LM, M LM FO: LM M LM LM LM LM OO- Anchor-T (x, [, y, [) ¬IO- Anchor- LM(x][y) IO- Anchor- LM(y][x) IO- Ident-T a. M

M LM *! * ** ☺b. LM M LM * * * c. M LM LM **! * * d. LM LM LM *! ** e.M M M *! ** ***

In a left-branching structure like (26), the casual and formal outputs are identical, regardless of the reranking of OO-Anchor-T(x, [, y, [).

(12)

(26) [[nam-nin] loi] ‘A man comes.’ a. I: LM LM LM RO: M LM, LM CO: M M LM LM LM LM ¬IO- Anchor- LM(x][y) IO- Anchor- LM(y][x) OO- Anchor-T (x, [, y, [) IO- Ident-T ☺a.M M LM * ** b.M LMLM *! * c.M M M **! * *** d. LM M LM *! * * * b. I: LM LM LM RO: M LM, LM FO: M M LM LM LM LM OO- Anchor-T (x, [, y, [) ¬IO- Anchor- LM(x][y) IO- Anchor- LM(y][x) IO- Ident-T ☺a.M M LM * ** b.M LMLM *! * c.M M M *! ** *** d. LM M LM *! * * *

The tableau evaluations in (26a,b) result in the same output by coincidence. Such coincidence is also observed in the left-branching Yinping structure, as shown in (27a,b).

(27) [[sam-chu] ti] ‘closet bottom’

a. I: M LM M RO: MH LM, M CO: MH LM M M LM M ¬IO- Anchor- M-U(x][y) OO- Anchor-T (x, [, y, [) IO- Anchor- LM(y][x) IO- Ident-T a. M LM M *! * ☺b. MH LM M * c. MH LM MH *! ** d. M M

M *! * * * e.M M MH *! ** * *

(13)

b. I: M LM M RO: MH LM, M FO: MH LM M M LM M OO- Anchor-T (x, [, y, [) ¬IO- Anchor- M-U(x][y) IO- Anchor- LM(y][x) IO- Ident-T a. M LM M *! * ☺b. MH LM M * c. MH LM MH *! ** d. M M

M *! * * * e.M M MH *!* * * *

Again, the casual and formal readings contrast in the right-branching structure, as in (28), where the noun sam ‘dress’ carries a sandhi tone, MH, in the casual output, but occurs as the base form in the formal output.

(28) [sam [ship-phet] ] ‘The dress is wet.’

a. I: M M-M- RO: M, M-M- CO: MH M-M- M M-M- ¬IO- Anchor- M-U(x][y) IO- Anchor- LM(y][x) OO- Anchor-T (x, [, y, [) IO- Ident-T a. M M- M-☺b.MH M-M- *! * * b. I: M M-M- RO: M, M-M- FO: M M-M- M M-M- OO- Anchor-T (x, [, y, [) ¬IO- Anchor- M-U(x][y) IO- Anchor- LM(y][x) IO- Ident-T ☺a. M M-M- * b.MH M-M- *! *

This left/right branching asymmetry is not found in Yinqu tone sandhi. As shown in (29b) and (30b), the high ranking of OO-Anchor-T(x, [, y, [) incorrectly selects the undesirable outputs (indicated by the serious face, ), while the real optimal outputs (indicated by the upset face, ) are ruled by this constraint. What happens here is that the casual and formal readings are the same, regardless of the branching direction.

(14)

(29) [khon [thien-shi]] ‘watch television’ a. I: HM HM HM RO: HM, H HM CO: H H HM HM HM HM ¬IO- Anchor- HM(x][y) IO- Anchor- HM(y][x) OO- Anchor-T (x, [, y, [) IO- Ident-T a. HM HM HM *!* ☺b. H H

HM * * ** c. H H

H **! ** *** d. HM H HM *! * * e.H HM HM *! * * b. I: HM HM HM RO: HM, H HM *FO: HM H HM HM HM HM OO- Anchor-T (x, [, y, [) ¬IO- Anchor- HM(x][y) IO- Anchor- HM(y][x) IO- Ident-T a. HM HM HM **! b. H H

HM *! * ** c. H H

H *! ** *** d. HM H HM * * * e.H HM HM *!* * *

(30) [[thien-fa] sien] ‘telephone line’

a. I: HM HM HM RO: H HM, HM CO: H H HM HM HM HM ¬IO- Anchor- HM(x][y) IO- Anchor- HM(y][x) OO- Anchor-T (x, [, y, [) IO- Ident-T a. HM HM HM *!* * ☺b. H H

HM * ** c. H H

H **! * *** d. HM H HM *! * ** * e.H HM HM *! *

(15)

b. I: HM HM HM RO: HM, H HM *FO: H H HM HM HM HM OO- Anchor-T (x, [, y, [) ¬IO- Anchor- HM(x][y) IO- Anchor- HM(y][x) IO- Ident-T a. HM HM HM *! ** b. H H

HM *! * ** c. H H

H *! ** *** d. HM H HM * * * e.H HM HM *!* * *

In other words, the OO correspondence constraint can not dominate the anchoring and non-anchoring constraints. The partial constraint ranking should look like (31), where OO-Anchor-T(x, [, y, [) is a floating constraint. It ranks as high as above ¬ IO-Anchor-LM(x][y) and ¬IO-Anchor-M-U(x][y) but below IO-Anchor-HM(y][x), and as low as below IO-Anchor-LM(y][x). This explains the fact that whereas Yangping and Yinping distinguish casual and formal readings, Yinqu does not.

(31) Dapu constraint ranking (tentative)

IO-CoindexT t | ¬ IO-Anchor-HM(x][y) | IO-Anchor-HM(y][x) / \ ¬ IO-Anchor-LM(x][y), ¬IO-Anchor-M-U (x][y) | OO-Anchor-T(x, [, y, [) IO-Anchor-LM(y][x) | IO-Ident-T

4. Prosodic Anchoring

When an expression is longer than three syllables, OO-Anchor-T(x, [, y, [) is insufficient in deriving the stylistic alternatives. Consider (32):

(16)

a. I: LM LM LM LM RO: M LM, M LM CO: M M M LM LM LM LM LM ¬IO- Anchor- LM(x][y) IO- Anchor- LM(y][x) OO- Anchor-T (x, [, y, [) IO- Ident-T a. M LM M LM *! * ** ☺b. M M M LM ** *** c. M M M M ** ****! d. LM M LM M *!* ** ** ** e. LM M M LM *! ** * ** f. M M LM LM *! * * ** b. I: LM LM LM LM RO: M LM, M LM *FO: M M M LM LM LM LM LM OO- Anchor-T (x, [, y, [) ¬IO- Anchor- LM(x][y) IO- Anchor- LM(y][x) IO- Ident-T a. M LM M LM *! * ** b. M M M LM ** *** c. M M M M ** ****! d. LM M LM M *!* ** * ** e. LM M M LM *!* * ** ** f. M M LM LM *! * * **

The reranking of OO-Anchor-T(x, [, y, [) is unable to select the correct formal output, M LM M LM, as candidate (a) in (32b) is removed by ¬IO-Anchor-LM(x][y). How then can the formal output be obtained? The answer can be related prosodic phrasing. In particular, the formal reading can be derived by way of footing. Two constraints are thus relevant:7

(33) FtMin: a foot is minimally disyllabic. (34) FtMax: a foot is maximally disyllabic.

FtMin dominates FtMax to ban monosyllabic feet. At this point, a trisyllabic string will always constitute a singly foot, as in (35).

7

(17)

(35) Trisyllabic footing

σσσσ FtMin FtMax

a. (σ)(σσ) *!

☺b. (σσσ) *

c. (σσ)(σ) *!

A quadrasyllabic string is usually parsed into a couple of disyllabic feet under this ranking, as in (36).8 (36) Quadrasyllabic footing σσσσ FtMin FtMax a. (σ)(σσσ) *! * ☺b. (σσ)(σσ) c. (σ)(σ)(σσ) *!* d. (σσ)(σ)(σ) *!* f. (σσσσ) *!

To avoid unnecessary digressions, the following discussions of quadrasyllabic tone sandhi will consider only the disyllabic foot. The difference between the casual reading and the formal reading in quadrasyllabic expressions is that the final tone in the foot retains its base form, a fact that can be captured by the constraint in (37).

(37) OI-Anchor-T(Σ, ]): the Σ-final tone in the output is identical to the input tone.

In this research, 98.6% of the Yangping data and 95% of the Yinping data are sensitive to the foot. The constraint in (37) ranks at the bottom in casual speech, but is promoted in formal speech, as in (38).

8

When a quadrasyllabic or trisyllabic compound is involved, lexical integrity will prevent the parsing of the disyllabic feet (cf. Hsiao 1991). Due to the limited pages of the short report, I will discuss this issue in a longer paper.

(18)

(38) [nam-nin][tshim tshien] ‘The man looks for the money.’ I: LM LM LM LM RO: M LM, M LM CO: M LM M LM LM LM LM LM ¬IO- Anchor- LM(x][y) OO- Anchor-T (x, [, y, [) IO- Anchor- LM(y][x) IO- Ident-T OI- Anchor- T(Σ, ]) a. (M LM)(M LM) *! * ** ☺b. (M M)(M LM) ** *** * c. (M M)(M M) ** ****! ** d. (LM M)(LM M) *!* ** * ** ** e. (LM M)(M LM) *! * ** ** * f. (M M)(LM LM) *! * * ** * I: LM LM LM LM RO: M LM, M LM FO: M LM M LM LM LM LM LM IO- Anchor- T(Σ, ]) OO- Anchor-T (x, [, y, [) ¬IO- Anchor- LM(x][y) IO- Anchor- LM(y][x) IO- Ident-T ☺a. (M LM)(M LM) * * ** b. (M M)(M LM) *! ** *** c. (M M)(M M) *!* ** **** d. (LM M)(LM M) *!* ** ** * ** e. (LM M)(M LM) *! * * ** ** f. (M M)(LM LM) *! * * * **

Yinping tone sandhi shows the same ranking of OI-Anchor-T(Σ, ]), as in (39). (39) [theu-ka][ten ki] ‘The boss waits for him’

a. I: LM M ML LM RO: LM M, ML LM CO: LM MH ML LM LM M ML LM ¬IO- Anchor- M-U(x][y) OO- Anchor-T (x, [, y, [) IO- Anchor- LM(y][x) IO- Ident-T OI- Anchor-T(Σ, ]) a. (LM M)(ML LM) *! * ☺b. (LM MH)(ML LM) * * b. I: LM LM LM LM RO: LM M, ML LM FO: LM M ML LM LM LM LM LM IO- Anchor- T(Σ, ]) OO- Anchor-T (x, [, y, [) ¬IO- Anchor- M-U(x][y) IO- Anchor- LM(y][x) OI- Ident-T ☺a.(LM M)(ML LM) * b.(LM MH)(ML LM) *! *

(19)

In Yinqu tone sandhi, however, OI-Anchor-T(Σ, ]) always ranks low. As in (40), the casual output and the formal output should be the same, and promotion of OI-Anchor-T(Σ, ]) selects the undesirable output in formal speech.

(40) mo mai thien-fa “not selling the telephone’

I: LM HM HM HM RO: LM, H H HM CO: LM H H HM LM HM HM HM ¬IO- Anchor- HM(x][y) IO- Anchor- HM(y][x) OO- Anchor-T (x, [, y, [) IO- Ident-T OI- Anchor-T(Σ, ]) a. (LM HM)(HM HM) *!* ** ☺b. (LM H)(H HM) ** ** I: LM HM HM HM RO: LM, H H HM *FO: LM HM H HM LM HM HM HM OI- Anchor- T(Σ, ]) ¬IO- Anchor- HM(x][y) IO- Anchor- HM(y][x) OO- Anchor-T (x, [, y, [) IO- Ident-T a. (LM HM)(HM HM) ** ** b. (LM H)(H HM) *! ** **

In brief, the partial constraint ranking can be enriched as follows:

(41) Dapu constraint ranking (final)

IO-CoindexT | ¬ IO-Anchor-HM(x][y) | IO-Anchor-HM(y][x) / \ ¬ IO-Anchor-LM(x][y), ¬IO-Anchor-M-U (x][y) | OO-Anchor-T(x, [, y, [) IO-Anchor-LM(y][x) OI-Anchor-T(Σ, ]) | IO-Ident-T

(20)

計畫成果自評

理論方面：首次以優選理論分析東勢客語的變調，驗證一套普遍語法制約，從非派生角度提出一個新的理論分析模式。語料庫方面： 1. 首次建立一個大型的東勢客語變調電子語料庫，共計二千一百餘行，成果如預期。 2. 此次建立之語料庫標注聲調、音步、句法、詞類，並使用十分普及的 Excel 軟體建立語料庫，可方便流通。待本計畫之相關論文出版後，即可將語料庫掛於網頁上提供學者進一步利用。論文發表方面：本報告已有其完整性，擬投稿國外期刊。本研究之階段性成果曾發表於一篇期刊及兩個研討會：蕭宇超、邱昀儀。2006。〈東勢客語變調：三至四字組分析〉。《語言暨語言學》。第七卷：第二期。中央研究院。。蕭宇超、邱昀儀。2004.〈從東勢客語看句法變調範疇定義之問題〉。美和第三屆客家學術研討會論文。美和技術學院客家研究中心。蕭宇超、邱昀儀。2004.〈東勢客語三至四字組變調〉。第二屆漢語方言小型研討會 -客家語研究。中央研究院。人才培育方面：參與碩士生包括黃婷、邱昀儀、吳耿彰、彭國書。在本計畫的薰陶下，邱昀儀的碩士論文即從語料庫角度分析東勢客語變調。

References

Alderete, J. 2001. Morphologically Governed Accent in Optimality Theory. New York: Routledge. Revised Ph.D. Dissertation. UMass, Amherst, 1998.

Beckman, J. 1998. Positional Faithfulness. University of Massachusetts, doctoral dissertation. ROA 234.

Chen, M. 1987. "The Syntax of Xiamen Tone Sandhi." Phonology Yearbook. 4: 109-150. Chen, M. 2000. Tone Sandhi: Patterns across Chinese Dialects. Cambridge University Press. Hsiao, Yuchau E. 1991. Syntax, Rhythm and Tone: A Triangular Relationship. Taipei: Crane

Publishing Co., Ltd.。

Hsiao, Yuchau E. 1995. Southern Min Tone Sandhi and Theories of Prosodic Phonology. Taipei: Student Book Co., Ltd.

Ito, J. and A. Mester. 1995. “Prosodic Faithfulness and Correspondence: Evidence from a Japanese Argot.” ROA-146.

(21)

Lin, Hui-Shan. 1999. An Optimality Theoretic Approach to Tone Sandhi in Mandarin, in Taiwanese, and in Mandarin-Taiwanese Code-mixing. M.A. Thesis, National Chengchi University

McCathy, J. 2003. “OT Constraints Are Categorical.” Phonology. 20: 75-138.

McCathy, J. and A. Prince. 1993a. Prosodic Morphology: Constraint Interaction and Satisfaction. ROA-482.

McCathy, J. and A. Prince. 1993b. “Generalized Alignment.” Yearbook of Morphology. 79-153. Dordrecht: Kluwer.

McCarthy, John. and Alan. Prince. 1995. “Faithfulness and reduplicative identity.” J. Beckman et al, eds. Papers in Optimality Theory. University of Massachusetts Occasional Papers 18. 249-384

Prince, A. and P. Smolensky. 1993. Optimality Theory: Constraint Interaction in Generative Grammar. Manuscript. Rutgers University.

Selkirk, E. 1986. "On Derived Domain in Sentence Phonology." Phonology Yearbook. 3: 371-405.

Selkirk, E. 1995. “The Prosodic Structure of Function Words.” University of Massachusetts Occasional Papers 18: Papers in Optimality Theory. 439-469. GLSA. University of Massachusetts, Amherst. 江敏華。1998。《臺中縣東勢客語音韻研究》。國立台灣大學碩士論文。江敏華。2002。〈東勢客家話的重疊結構與變調〉。《語言暨語言學》。第三卷：第三期： 543-567 頁。董忠司. 1994.〈東勢客家語音系統略述及其音標方案〉，台灣客家語研討會論文。新竹：國立清華大學。

從優選理論分析東勢大埔客語之連讀變調

行政院國家科學委員會專題研究計畫 成果報告