CHAPTER 2 LITERATURE REVIEW
2.4 Tone Sandhi and CCT (Lin 2011, Chen 2013)
國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
means a sandhi tone spreads to neighboring syllables as we may see from Wu Dialects (Chen 2000). Both dialects discussed in this thesis are of the contextual types. The tone sandhi is sensitive to adjacent elements and targets the melody tier of the adjacent elements.
2.4 Tone Sandhi and CCT (Lin 2011, Chen 2013)
Lin (2011) analyzes Hakka tone sandhi of Dongshi dialect by means of CCT.
Since Dongshi Hakka’s assimilation and dissimilation tone sandhi occurs when the head tone are low in register value, they cannot be captured by simple assimilation constraint (NOJUMP-t) and dissimilation constraint (OCP-C(l)). Thus, several CCTs on tone sandhi in adjacent domain are posited by Lin (2011) such as [NOJUMP-t&*HD/Lr]ADJ and [OCP-C(l)&*HD/Lr]ADJ to account for these kinds of tone sandhi phenomena. These combined constraints rank among the highest constraints while their single counterpart is inactive.
Chen (2013) also addresses the tone sandhi of Liujia Raoping Hakka and Ningdu Tiantou Hakka by positing several CCTs. Some of the tone sandhi in these two dialects are captured by a combination of agreement and sequential of markedness constraint with certain marked tones such as [OCP-h & *M] and Agree-t & *LL] which proves to be a good strategy to solve certain problematic tone sandhi in the dialects. Since similar cases are found in Meixian and Bangkok Hakka, where there are some characteristics of tonal feature in the head position of Meixian and Bangkok Hakka, this strategy in Lin (2011) and Chen (2013) are adopted in the analysis in Chapter 3 and 4.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
CHAPTER 3
OT ANALYSIS OF MEIXIAN HAKKA
This chapter presents a constraint based analysis of disyllabic tone sandhi of Meixian Hakka spoken in Meijiang District and Meixian County, China. The data of this thesis is mainly taken from Cheung (2011). The chapter is arranged into following sections: Section 3.1 provides the value of the citation tones and sandhi tones of Meixian Hakka. Section 3.2 shows patterns, process and generalization of the tonal alternations.
Section 3.3 presents the Optimality Theory analysis of the disyllabic tone sandhi patterns.
In this section, the alternations of tone structure are analyzed in detail. And the mechanism on how the conjunction constraints (CCT) are ranked against faithfulness constraints and other markedness constraints to produce the optimal sandhi tone is shown.
The preservation of certain tones, and tone features are also presented in this section.
Section 3.4 provides a Hasse diagram showing Meixian Hakka tone sandhi grammar, and a summary table of possible outputs. Section 3.5 is the conclusion for the analysis of tonal alternation grammar of Meixian Hakka.
3.1 Tone Inventory
According to the phonetic based research presented by Cheung (2011), there are six citation tones in Meixian Hakka: Yinping 33 (mid level/MM), Yangping 11 (low level/LL), Shangsheng 41 (high falling/HL), Qusheng 51 (high falling/HL), Yinru 41 (short high falling/HL), and Yangru 5 (short high level/H)1. Meixian Hakka tonal
1 In this thesis, Chao’s (1930) is converted to Yip’s transcription (2001) where tone value 4& 5 are marked as H, 3 as M, and 1&2 as L.
13
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
inventory is also composed of three sandhi tones: sandhi tones 35 (high rising/MH), 33 (mid level/MM), and 55 (high level/HH) of which derived from Yinping (MM),
Shangsheng (HL) and Qusheng (HL) respectively. There are several overlapping tone values in this thesis, but this thesis recognizes that the tones with similar tone value to have phonetic differences not a phonemic difference. For example, there are three high falling tones and two mid level tones. In order to solve this phonemic similarity, this thesis reconsider both tone value 41 of Shangsheng,and 41 of Yinru to be low falling/ML and short low falling/ML phonologically. This means, 41 and 41 are considered as phonetic variants of 31 and 31 respectively. This claim is also supported by the Cheung’s own research where the F0 value is slightly below the 4th point on the 5 point scale.
Therefore, it is highly possible the differences are only gradient (phonetic) differences rather than the categorical (phonemic) differences. Moreover, Qusheng is also considered as a high falling tone/HM rather than 51/HL. The reason for this consideration is that the fact that most of tone systems avoid contour tone pairs which have the same start or end points but different pitch differentials such as the overlap of HL and ML as found in this data (Bao 1999). So like Shangsheng and Yinru, 51 found in the data are considered as a phonetic variant rather than a phonemic category. This claim is also supported by Huang’s (1992) transcription where the tone value was given 53.
Another point to make is that there are two mid level tones here, Yinping citation tone and Shangsheng sandhi tone. In this thesis, it is assumed that the register value of Meixian and Bangkok Hakka sandhi tones are always identical and they are enforced by an undominated faithfulness constraint, IDENT-reg (see 3.3.4). Thus, these two tones are identified to have two different values: register value and tone melody value as we can
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
see later in (1). This argument can be proven by the alternation of tone from MM to MH and from ML to MM. The hypothesis is that register value of MH sandhi tone (Hr,lh) is preserved from its citation tone. The same condition applies to the sandhi tone MM
where the register value is derived from its citation tone ML which has low register value.
Furthermore, the tonal alternations in Meixian Hakka are approached from its internal structure. Since the tone alternations in this dialects affects its register, contour and the tone melody value of the tone, Bao’s (1999) tone structure system with Lin’s (2011) labeling is adopted. The following table presents the internal structure of tones in Meixian Hakka grammar.
(1) Meixian Hakka Tone Inventory Citation Tones Sandhi Tones
MM, (Hr,l) MH (Hr,lh)
LL (Lr, l)
ML (Lr, hl) MM (Lr,h)
HM (Hr,hl) HH (Hr, h)
ML (Lr, hl) H (Hr,h)
3.2 Tone Sandhi Patterns
There are three citation tones that undergo sandhi process in Meixian Hakka. First, the mid level tone (MM) alternates to high rising tone (MH) when it is preceded by low level tone (LL), low falling tone (ML) and short low falling tone (ML). Second, low
15
‧
falling tone (ML) alternates to mid level tone (MM) when it is preceded by low level tone (LL), low falling tone (ML) and short low falling tone (ML). Third, high falling tone (HM) alternates to high level tone (HH) when it is preceded by low level tone (LL), low falling tone (ML), short low falling tone (ML)and high falling tone (HM).
To allow us see a better picture of the tones that undergo alternations, and those that do not, a full chart (6x6) of the tonal combination is shown in the following table.
(2) Tonal combinations table S2
From the table above, it is evident that the mid level tone undergoes changes when it is followed by low level tone, low falling tone and short low falling tone. There are rules for each tonal alternation with its structural details in the following table.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(3) Mid level tone sandhi 1. MM-LL
[(Hr,l)-(l,Lr)]
tsu phi ‘pork skin’
→ MH-LL
[(Hr,lh)-(l,Lr)]
tsu phi ‘pork skin’
2. MM-ML [(Hr,l)-(hl,Lr)]
tsu tu ‘pork belly’
→ MH-ML
[(Hr,lh)-(hl,Lr)]
tsu tu ‘pork belly’
3. MM-ML [(Hr,l)-(hl,Lr)]
su tsok ‘desk’
→ MH-ML
[(Hr,lh)-(hl,Lr)]
su tsok ‘desk’
In the alternations of mid level tone, it is obvious that there are two different mechanisms of the tonal alternation: tone melody dissimilation (MM-LL) and tone melody assimilation (MM-ML & MM-ML). The tree diagrams showing the details on both phenomena are presented below.
(4) MM-LL tone sandhi (low tone melody dissimilation)
(5) MM-ML/ML tone sandhi (high tone melody assimilation)
Tree (4) shows the dissimilated intersyllabic low tone melody. The dissimilation process of (4) contradicts with tree (5) where the different intersyllabic features are assimilating from the head/right syllable of the tonal pair. In addition, ML and ML are treated in one analysis.
17
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
The low falling tone of the Meixian dialect alternates to mid level sandhi tone when it precedes the same sets as found in mid level tone alternations, they are low level tone, low falling tone, and short low falling tone. The alternations are shown in the following table:.
(6) Low falling tone sandhi 1. ML-LL
[(Lr,hl)-(l,Lr)]
pu theu ‘axe’
→ MM-LL
[(Lr,h)-(l,Lr)]
pu tteu ‘axe’
2. ML-ML
[(Lr,hl)-(hl,Lr)]
pu khau ‘make up examination’
→ MM-ML
[(Lr,h)-(hl,Lr)]
pu khau ‘make up examination’
3. ML-ML
[(Lr,hl)-(hl,Lr)]
tu pok ‘gambling’
→ MM-ML
[(Lr,h)-(hl,Lr)]
tu pok ‘gambling’
Regarding low falling tone alternations, the similar forces that trigger the sandhi in Meixian Hakka are also found. They are tone melody dissimilation with LL and tone melody assimilation with ML and ML. Following are trees which showing a detail rule analysis on both phenomena.
(7) ML-ML/ML tone sandhi (high tone melody assimilation)
(8) ML-LL tone sandhi (low tone melody dissimilation)
‧
Tree (7) provides the analysis of assimilation process where ML and ML high tone melody feature spreads to the left one. While Tree (8) shows how the adjacent low tones melody is prevented by the dissimilatory force as in (5).
There are four tonal alternations of high falling tone where it alternates to high level tone only when it precedes low level tone, low falling tone, high falling tone, and short low falling tone as shown in the following table.
(9) High falling tone sandhi 1. HM-LL
ki tsa ‘reporter’
→ HH-ML
[(Hr,h)-(hl,Lr)]
ki tsa ‘reporter
3 HM-HM
fu tsuk ‘rotten bamboo’
→ HH-ML/ML
[(Hr,h)-(hl,Lr)]
fu tsuk ‘rotten bamboo’
In term of the change of high falling tone, the tone melody dissimilation (HM-MM) will be first discussed, followed by tone melody assimilation (HM-MLML), and tonal dissimilation (HM-HM) of Meixian Hakka.
(10) HM-LL tone sandhi (low tone melody dissimilation)
19
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(11) HM-ML/ML tone sandhi (high tone melody assimilation)
(12) HM-HM tone sandhi (tonal tier dissimilation)
Tree (10) and (12) present dissimilatory type tone alternations, but they involve different structures in the process. In (10), the trigger is the low level tone at the head position while in (12), it is triggered by similar tone at the right position. In tree (11), the same assimilatory mechanism as in ML-ML/ML alternations is found.
In conclusion, there are two forces that trigger Meixian Hakka tone alternations:
dissimilation and assimilation forces at tone melody level and tone level. The dissimilation of the low tone melody is more preferred than assimilation of the tone melody when the register value of the head is low. Meixian Hakka tonal alternations bring the effect to the left syllable meaning only the left position tone changes its value, not the right ones. There are 3 tones (LL, ML, and H) that do not undergo changes despite being provided the same environment. Lastly, the citation tones and their sandhi counterpart always retain the same register value, initial target, and tonal type/duration.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
3.3 OT Analysis
The OT analysis of Meixian Hakka tone sandhi bases on the generalization made from the previous section. Several points are discussed here where it starts from the pattern of the tonal alternations, followed by certain preservations of tonal features and completed with a summary tables of all possible pairs. It is important to note that Meixian Hakka tone system is right dominant2 where the tone system in this dialect only allows the change of its left tone. The right (head) syllable tone does not undergo tonal change and this preservation of tone is expressed in the form of identity constraint IDENT-HD
(following Lin 2011). This constraint is undominated so that the right tone/head position tone should not be alternated.
3.3.1 Obligatory Contour Principle Effect
The Obligatory Contour Principle effect is found in Meixian Hakka disyllabic tone sandhi. The OCP governs the prohibition of certain similar elements to be adjacent which is normally associated with dissimilation. And there are four tonal pairs that undergo the process of dissimilation from the rule based analysis: MM-LL→MH-LL, ML-LL→MM-LL, HM-LL→HH-LL and HM-HM→HH-HM. The first three alternations where the triggers have low register value in the right position are taken into account initially because it is evident that they have similarity in term of sequence of low tone melody, and the tone in the non head position changes its value. This kind of intersyllabic low melody similarity can be captured by an OCP based constraint. The OCP based constraint should ranks higher than faithfulness constraint IDENT in the following tableau.
2 This right dominancy is due to several tone languages (Mandarin, Southern Min, Dongshi Hakka) whose second syllable always preserves its tone. There are also left-dominant languages such as Northern Wu dialects (Yue-Hashimoto 1987, Chen 2000, Zhang 2007).
21
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(13) OCP-t(l) - Assign one violation mark for every pair of tones which have a similar low tone melody value intersyllabically
(14) IDENT-T - Assign one violation mark for every tone in the output which have different value to its input
(15) MM-LL → MH-LL MM-LL
Hr-Lr l-l
OCP-t(l) IDENT-T
1. MH-LL Hr-Lr l[h-l]
*
2. MM-LL Hr-Lr [l-l]
*!W L
(16) ML-LL → MM-LL ML-LL
Lr-Lr hl-l
OCP-t(l) IDENT-T
1. MM-LL Lr-Lr h-l
*
2. ML-LL Hr-Lr hl-l
*W L
‧
Tableaus (15), (16), and (17) shows how OCP-t(l) provides a possible argument that by ranking it higher than IDENT-T, the expected candidate of the input of the pair which has sequence of low tone melody would not surface. However, an analytical problem occurs since there are other candidates which have similar low tone melody value such as MM-MM (Hr,l-l,Hr), ML-MM (Lr,hl-l,Hr) and HM-MM (Hr,hl-l,Hr) which do not undergo alternations. These remaining faithful candidates suggest that the higher ranking of OCP-t(l) over IDENT-T predicts the wrong candidate as shown by an example (18). It is a candidate pair that has adjacent low tone melody (MM-MM).
(18) Example of wrong candidate predicted by OCP-t(l)>> IDENT-T MM-MM→MM-MM
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Since it predicts wrong output, OCP-t constraint is inadequate to deal with the whole tonal pairings as shown in tableau (18). To solve this problem, the head (right syllable) structure is included into the analyses. It is found that the low register features of the right tone play role in the tonal alternation. This characteristic is also found in the literature of some Chinese languages where the high register tone such as high level tone does not trigger an alternation (Dongshi Hakka and Mandarin). To solve this problem, Lin’s (2011) idea of using a modified OCP-t constraint is applied in this thesis. This OCP based constraint is conjoined with the register value in the head/right syllable position.
(19) *HD/Lr - Assign one violation mark for tone in head position which has low register value
(20) OCP-t(l)&*HD/Lr : Assign one violation mark for every tonal pair which violate both OCP-t(l) and *HD/Lr
By using the conjoined constraint (CCT), we get a better identification of the whole dissimilated and non dissimilated tonal pairs. Following tableaus show how the CCT constraint plays its part in the ranking argument.
(21) tsuMM-phiLL → tsuMH- phiLL ‘pork skin’
MM-LL Hr-Lr l-l
OCP-t(l)&
*HD/Lr
IDENT-T OCP-t(l) *HD/Lr
1. MH-LL Hr-Lr lh-l
* *
2. MM-LL Hr-Lr l-l
*!W L * *
‧
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Tableaus (21), (22) and (23) show the tonal alternation which prevents sequence of low tone melody where the head is low register tone. The tonal alternation is triggered by the higher ranked of OCP-t(l)&*HD/Lr over IDENT-T. Moreover, the tonal pair in tableau (24) does not need to undergo changes since it has different register feature (high).
So, the faithful pair, MM-MM, is protected from dissimilation effect.
OCP-t(l)&*HD/Lr, however, affects LL and ML when they are paired with low level tone in the head position. However, they do not undergo changes, and their forms are preserved. The violation of the CCT is tolerated by following undominated tonal identity constraints.
(25) IDENT-LL: Input-output of low level tone is identical.
(26) IDENT-ML: Input-output of short low falling tone is identical.
(27)LL-LL→LL-LL LL-LL
Lr-Lr l-l
IDENT-LL OCP-t(l)&
*HD/Lr
IDENT-T
1. LL-LL Lr-Lr l-l
*
2. MM-LL Lr-Lr h-l
*!W L *
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(28)ML-LL→ML-LL ML-LL
Lr-Lr hl-l
IDENT-ML OCP-t(l)&
*HD/Lr
IDENT-T
1. ML-LL Lr-Lr hl-l
*
2. MM-LL Lr-Lr h-l
*!W L *
Tableau (27) and (28) show how the highest ranked IDENT-LL and IDENT-ML preserve the identity of LL and ML from changing enforced by OCP-t(l)& *HD/Lr. So despite being provided the same environment (low register head and similar low level melody), LL and ML cannot change or they will make serious violation. This king of ranking provides an argument that explains why LL and ML tones in Meixian Hakka do not undergo dissimilation. This kind of an argument that cannot be provided by the rule based model.
Next, the pair of HM-HM does not surface in Meixian. The high falling tone in the left syllable alternates to high level tone. It is found that the trigger of the alternation is HM. To solve this tone sandhi case, a constraint that can prevent two consecutive high falling tones to be adjacent another is needed. Here, another OCP based constraint that triggers the tone sandhi, namely OCP-T (HM) is posited.
(29) OCP-T(HM): assign one violation mark for every high falling tonal pair.
27
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(30) fuHM-kuiHM → fuHH-kuiHM ‘wealth’
HM-HM Hr-Hr hl-hl
OCP-T (HM) IDENT-T
1. HH-HM Hr-Hr h-hl
* 2. HM-HM
Hr-Hr hl-hl
*!W L
Tableau (30) shows how HM alternates to HH. It is due to the tonal tier
constraints, OCP-T (HM), ranks higher than IDENT-T. Other constraints are considered irrelevant and they are omitted from the tableau.
3.3.2 No-Jumping Principle Effect
As shown in the previous analysis, Meixian Hakka disyllabic tonal alternations also include the assimilation of the tone structures. The tone alternations of several pairs as: MM-ML/ML→MH-ML/ML, ML-ML/ML→MM-ML/ML and HM-ML/ML→HH-ML/ML are enforced by the assimilation process. The assimilation of high tone melody can be captured by NO-JUMPING Principle (Hyman & Van Bik 2004). This principle was firstly introduced in Hakka Lai to capture its tone alternation where the alternations occur intersyllabically. This concept is later adopted by Lin (2011) in a form of constraint to analyze the tonal assimilation of Dongshi Hakka. The same concept is then adopted in the thesis to deal with the agreement type of alternations.
(31) NOJUMP-t: Assign one violation mark for every tone melody value which does not agree intersyllabically.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
This assimilatory constraint, when ranked above faithful constraint IDENT-T like shown in the tableau (32), (33) and (34), seems to provide a ranking argument which enforce the outputs that have sequence of similar tone melody to surface, not the faithful ones.
(32)MM-ML/ML→MH-ML/ML MM-ML/ML
Hr-Lr l-hl
NOJUMP-t IDENT-T
1. MH-ML Hr-Lr l[h-h]
*L
2. MM-ML Hr-Lr [l-h]l
*!W
(33) ML-ML/ML → MM-ML/ML ML-ML/ML
Lr-Lr hl-hl
NOJUMP-t IDENT-T
1. MM- ML/ML Lr-Lr
[h-h]l
*
2. ML-ML/ML Lr-Lr [hl-h]l
*!W L
29
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(34) HM-ML/ML→HH-ML/ML HM-ML/ML
Hr-Lr hl-hl
NOJUMP-t IDENT-T
1. HH- ML/ML Hr-Lr [h-h]l
*
2. HM-ML/ML Hr-Lr h[l-h]l
*!W L
Tableaus (32), (33) and (34) show how markedness constraint NOJUMP-t that ranks higher than IDENT-T provides the expected outputs. But again a problem arises since since there are some pairs that do not undergo tone sandhi, such as MM-HM (Hr,l-hl,Hr), MM-H (Hr,l-h,Hr), and MM-HH (Hr,l-h,Hr). This means NOJUMP-t should not rank higher than IDENT-T. In agreement tone sandhi, it is also found that the trigger is the low register head. Thus, the same strategy is then applied where two markedness constraints are merged in to a conjoined constraint, NOJUMP-t &*HD/Lr. The least marked NOJUMP-t is ranked lower than IDENT-T.
(35) NOJUMP-t &*HD/Lr: Assign one violation mark for every tonal pairs which violate both NOJUMP-t and &*HD/Lr.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(36) tsuMM-tuML→ tsuMH-tuML ‘pork bowel’
suMM-tsokML → suMH-tsok ML/ML ‘desk’
MM-ML/ML Hr-Lr
l-hl
NOJUMP-t
&*HD/Lr
&*HD/Lr