CHAPTER 3 OT ANALYSIS OF MEIXIAN HAKKA
3.5 Conclusion
國
立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
3.5 Conclusion
The chapter presents the OT analysis of Meixian Hakka tone sandhi. After investigating the tonal alternations from their internal structures, the tonal alternations of Meixian Hakka are grouped into two categories: assimilation and dissimilation. The general IDENT constraints are dominated by markedness constraints to have the sandhi.
Nevertheless, the IDENT-T is not dominated by basic OCP and NO-JUMP constraints. A more complex constraint conjunction is needed to solve the tonal alternations. Lin’s (2011) ideas by conjoining OCP and NO-JUMP with headedness constraints like *HD/Lr is followed. Another markedness constraint, OCP-T(HM) is also posited to deal with the assimilation of sequence of HM tone.
Aside from the tone sandhi grammar constraints, there are also several
undominated faitfulness constraints proposed in Meixian Hakka. First, IDENT-LL and IDENT-ML since LL and ML of Meixian Hakka tone are highly prominent and retain the tone value in the grammar. Second, IDENT-reg and IDENT-IT since the sandhi tones always retain the same register and initial target specifications. Third, IDENT-TD since tone duration in Meixian Hakka does not change. The natural characteristics of open and closed syllable determine the duration of the tone. The Hasse diagram (47) presents the grammar of Meixian Hakka tone sandhi by showing how the faithfulness and markedness constraint interact with each others.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
CHAPTER 4
OT ANALYSIS OF BANGKOK HAKKA
Chapter 4 presents an analysis of disyllabic tone sandhi in Bangkok Hakka.
Bangkok Hakka is a sub-dialect of Meixian Hakka spoken in Bangkok, Thailand. The data for this analysis are taken from Siripen (2008) which describes the phonetic value of its segments and tone and rule-based approach towards its tonal alternation phenomena.
This chapter is divided into several sections as follows: Section 4.1 gives the value of citation tones and sandhi tones in the combined model of Bao’s (1999) and Yip’s (2001).
Section 4.2 presents the tone sandhi pattern of Bangkok Hakka and its generalization.
Section 3 of this chapter provides the constraint based approach of OT and the application of constraint conjunction (CCT) to analyze the alternations of tones in
Bangkok Hakka. Several tonal structure preservation constraints to shape the structure of the tones are also provided in this chapter. Section 4.4 presents a Hasse Diagram and a summary table for the grammar of Bangkok Hakka tone sandhi. Conclusion of this chapter is provided in section 4.5.
4.1 Tone Inventory
In Siripen (2008), there are six citation tones in Bangkok Hakka: Yinping 33 (mid level/MM), Yangping 21 (low level/LL); Shangsheng 31 (low falling/ML); Qusheng 44
45
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(high level/HH); Yinru 32 (short low falling/ML) and Yangru 4 (short high level/H).
Furthermore, Bangkok Hakka has two sandhi tones: 35 (high rising/MH), and 53 (high falling/HM) from Yinping and Qusheng respectively. In this thesis, the analysis depends on the tonal preservation and change of register and tone melody values as proposed in Bao (1999). Therefore, the detail information for every tone and its features are provided below.
(1) Bangkok Hakka Tone Inventory.
Citation Tones Sandhi Tones MM, (Hr,l) MH (Hr,lh) LL (Lr, l)
ML (Lr, hl)
HH (Hr,h) HM (Hr, hl) ML (Lr, hl)
H (Hr,h)
4.2 Tone Sandhi Patterns
There are two citation tones that undergo tonal change in Bangkok Hakka. First, mid level tone (MM) alternates to high rising tone when it is preceded by low level tone (LL), low falling tone (ML) and short low falling tone (ML). Second, high level tone (HH) changes to high falling tone (HM) when it is preceded by mid level tone (MM) and short
‧
high level tone (H). The other tones (low level, mid falling, short mid falling and short high level) do not undergo tone alternations. To enable us to see better picture of tones that undergo alternations, and those that do not, a full chart (6x6) of the tonal
combination are presented in the table (2).
(2) Bangkok Hakka Tonal combinations S2
There are two tones which undergo tonal change. The mid level tone alternates to high rising tone when it is followed by low level tone, low falling tone and short low falling tone. The mid level tone sandhi pattern has a similar pattern to Meixian Hakka
47
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
mid level tone sandhi. The following table contains the mid level tonal alternations with their structural details.
(3) Mid level tone sandhi in Bangkok Hakka 1. MM-LL
[(Hr,l)-(l,Lr)]
ke phi ‘chicken skin’
→ MH-LL
[(Hr,lh)-(l,Lr)]
ke phi ‘chicken skin’
2. MM-ML [(Hr,l)-(hl,Lr)]
wu suj ‘black water’
→ MH-ML
[(Hr,lh)-(hl,Lr)]
wu suj ‘black water’
3. MM-ML [(Hr,l)-(hl,Lr)]
ke kut ‘chicken bone’
→ MH-ML
[(Hr,lh)-(hl,Lr)]
ke kut ‘chicken bone’
We see that there are two different forces that trigger mid level tone alternations:
tone melody dissimilation (MM-LL) and tone melody assimilation (MM-ML&ML). The trigger of the sandhi is the low register tone in head position of tonal combinations. The process can be observed from following trees.
(4) MM-LL tone sandhi (low tone melody dissimilation)
(5) MM-ML/ML tone sandhi (high tone melody assimilation)
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
In tree (4), the adjacent low tone melody dissimilates to high tone melody making them not identical. Meanwhile, tree (5) shows how the different tone melody value is assimilated from the head position/right syllable of the tonal pair.
Two tonal alternations are found in the high level tone of Bangkok Hakka: The high level tone alternates to high falling tone when it precedes mid level tone, and short high falling tones and the details are presented in the following table.
(6) High level tone sandhi in Bangkok Hakka 1. HH-MM
[(Hr,h)-(l, Hr)]
koŋ sin ‘made of steel’
→ HM-MM
[(Hr,hl)-(l,Hr)]
koŋ sin ‘made of steel’
2. HH-H
[(Hr,h)-(h,Hr)]
koŋ tshap ‘fence made of steel’
→ HM-H
[(Hr,hl)-(h, Hr)]
koŋ tshap ‘fence made of steel’
According to the type of the alternation, the HH-MM tone sandhi is categorized as assimilation of low tone melody that is triggered by the mid level tone in the head
position. On the other hand, HH-H is a dissimilation of high level contour. And the trigger for the assimilation and dissimilation are the tones in the right syllable as shown in the following trees.
(7) HH-MM tone sandhi (low tone melody assimilation)
49
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(8) HH-H tone sandhi (high level contour dissimilation)
Tree (7) shows the process of tone melody assimilation when the head/right syllable is mid level tone. The low tone melody of the head spreads to the non head syllable making it similar in term of their melody features. In tree (8) the dissimilation of the intersyllabic high level contour is triggered by short high level tone in head position, this makes its longer counterpart have another contour form in order to be different.
From the data above, we arrive to a generalization where there are two
mechanisms of the Bangkok Hakka disyllabic tone sandhi: intersyllabic dissimilation and intersyllabic assimilation at tone melody tier and contour tier. The dissimilation of the low tone melody is preferred over assimilation at the same tier with low register tone as the trigger. The alternations of tone in Bangkok Hakka is right headed where the
alternations only affects the left syllable. The tones of right syllable/head syllable are intact, and it is governed by undominated head position constraint, IDENT-HD (following Lin 2011). There are 3 low register tones (LL, ML, ML) and short high level tone (H) that do not undergo alternations despite they are provided the same environment. In addition, the citation tones and their sandhi counterparts always retain the same register value, intial tone value and tonal duration.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
4.3 OT Analysis
This section starts from the analysis of tone sandhi phenomena found in Bangkok Hakka. It is followed by several preservations of tonal features, and ends with a summary table of all possible candidates.
4.3.1 Obligatory Contour Principle Effect
The OCP effect that captures Meixian Hakka’s tonal alternation also play
important role in the tonal dissimilation of Bangkok Hakka. From the alternation analysis of MM-LL→MH-LL and HH-H → HM-H in section 4.2, we could see the dissimilation are triggered by different structures. The analysis begins with the case of MM-LL→MH-LL where the change of tone is enforced by a same set of constraints found in Meixian Hakka. As we have already tried in Meixian Hakka disyllabic tone sandhi, the single OCP-t(l) constraint is unable to capture the sequence of MM-LL in Bangkok Hakka.
Thus, a combination of the constraint and positional constraint, *HD/Lr is again posited to promote the tone alternation. The effect of OCP-t(l)&*HD/Lr can be seen from tableau (9).
51
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(9) keMM-phiLL → keMH-phiLL ‘chicken 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 * *
From tableau (9), we could see the alternation of MM to MH when it is followed by LL as the trigger. The similar adjacent tone melody is prohibited by the markedness OCP-t(l) & *HD/Lr. This constraint outranks IDENT-T to trigger the sandhi in order the grammar to take the dissimilated candidate as the well-formed one. Since *HD/Lr is ranked lower than IDENT-T, it is inactive in the grammar. The OCP-t(l)&*HD/Lr, however, should rank lower than IDENT-T-Lr, a faithfulness constraint that prevent changes for low register tones. The reason why the low register tone does not change is not clearly understood. They could by hypothetically more prominent tones in Bangkok Hakka tone system. However, OT provides the tool to preserve the unchanged tones in the grammar in form of faithfulness constraints.
The following are tableaus showing the selection of optimal candidate for low register tones under the same trigger.
(10) IDENT-T-Lr: Input-output of low register tone is identical
‧
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
Tableaus (11), (12) and (13) present ranking arguments between a tonal specific constraint and highly marked conjunction constraint. To preserve these tones from changes, OCP-t(l) & *HD/Lr needs lower ranking compared to IDENT-T-Lr. The IDENT -T-Lr constraint should be the winner in the ‘conflict’ against other markedness
constraints in the dialect.
The other dissimilatory tone sandhi, HH-H→HM-H, presents another
dimension/tier of tone alternation. It is because the tone sandhi has specific behavior for it is only found in HH-H tonal sequence but not in HH-HH pair which have similar feature except the duration of the head. Here, it is hypothesized that the trigger for this sandhi is the ru/short tones in the head position.
To let the tonal alternate but without affecting other pairs, the combination the OCP-c(h) and *ST which prevent a sequence of high level contour with the short tone in the head position is posited. It is necessary to mention the status of *ST constraint. This constraint is an inactive constraint in this dialect. This inactivity is due to the
characteristics of Hakka tonal inventory and most Chinese dialects. In the syllables of Bangkok Hakka, the ru/short or shu/long tones have been predetermined according to their syllable types.
(14) OCP-c(h) : Assign one violation mark for sequence of high level contour.
(15) *HD/ST: Assign one violation mark for every short tone in the head position
‧
(16) OCP-c(h)&*HD/ST: assign one violation that violates both OCP-c(h) and *HD/ST (17) tsuŋHH-thukH → tsuŋHM-thukH ‘poisoned’
HH- H
In tableau (17), it is recognized that *HD/ST constraint is violated by two possible candidates generated by GEN here however there is no effect here. The existence of the constraint in the tableau is intended to show us the interaction between the conjoined constraint and single markedness constraint. Candidate 2 does not surface since they violate the higher ranked OCP-c(h) &*HD/ST, then candidate 1 is the most optimal output.
Tableau (18) presents the ranking of IDENT-H which is higher than OCP-c(h)
&*HD/ST in order for a faithful candidate win. IDENT-H along with other specific tonal
55
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
identity constraints (IDENT-LL, ML, ML) are undominated in the ranking hierarchy of Bangkok Hakka tone sandhi.
4.3.2 No-Jumping Principle Effect
There are two sets of tonal change in Bangkok Hakka which are included in the intersyllabic tonal agreement/assimilation category. They are MM-ML/ML and HH-MM.
We could see the requirements of same tonal feature tone melody tier. The agreement of MM-ML/ML is firstly accounted for it has a similar environment found in Meixian Hakka. Thus this agreement of high tone melody can be captured by a similar CCT approach applied in Meixian Hakka. The combination of NOJUMP-t (assimilation) and
*HD/Lr which solve inadequacy of NOJUMP-t to deal with this tonal alternation is again
presented.
(19) wuMM-sujML → wuMH-sujML keMM-kutML → keMH-kutML MM-ML/ML
Hr-Lr l-hl
NOJUMP -t&*HD/Lr
IDENT-T NOJUMP-t
1. MH- ML/ML Hr-Lr
l[h-h]
*
2. MM-ML/ML Hr-Lr
[l-h]l
*!W L *
Tableau (19) shows that NOJUMP-t&*HD/Lr must outrank IDENT-T in order MM to alternate. Candidate 1 survives as it obeys the higher ranked constraint. On the other
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
hand, candidate 2 violates the constraint and it does not surface. In addition, since NOJUMP-t constraint is low ranked so despite being violated, it does not impact the candidate selection.
In the beginning of this section, it is mentioned that despite involving tonal melody agreement, NOJUMP-t or NOJUMP-t &*HD/Lr can not captured HH-MM sandhi simply because the head is high register MM. Positing a constraint that combining
NOJUMP-t and *HD/Hr itself also does not help as it will affect non tone sandhi candidate such as MM-HH and MM-H.
The tone sandhi is triggered by high register MM only. Due to the characteristic,
the tone sandhi case is becomes very specific. To solve the problem, this thesis follows de Lacy’s proposal (2002) that every tone in the language actually has markedness constraints depending on their prominence position. When these inactive constraints are conjoined with tonal alternations constraints such as NOJUMP-t and OCP-t then they play important role in ruling out the worst of the worst candidate.
(20) *HD/MM: No mid level tone in head position
(21) NOJUMP-t&*HD/MM: Assign one violation mark for every tonal pair which violates both NOJUMP-t and *HD/MM
57
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(22) tsieHH-suMM → tsieHM-suMM ‘borrowing book’
HH-MM Hr-Hr h-l
NOJUMP-t&
*HD/MM
IDENT-T NOJUMP-t *HD/MM
1. HM-MM Hr-Hr h[l-l]
* *
2. HH-MM Hr-Hr [h-l]
*!W L * *
Tableau (22) shows the assimilation of HH tone to HM as it is triggered by the MM tone in the head position. The MM tone itself is inactive as other tonal markedness in this dialect. Nevertheless the conjoining of *MM and assimilation constraint favors candidate 1.
4.3.3 OCP-t(l)& *HD/Lr vs. NOJUMP-t&*HD/Lr
In the generalization provided in the previous section, it is mentioned that low tone dissimilation is preferred to tone melody assimilation. This is proven in mid level tone alternation. A ranking argument where OCP-t(l) & *HD/Lr should rank higher than NOJUMP-t&*HD/Lr is provided in this sub-section.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(23) OCP-t(l) & *HD/Lr>> [NOJUMP-t&*HD/Lr
MM-LL Hr-Lr [l-l]
OCP-t(l) &
*HD/Lr
NOJUMP -t&*HD/Lr
1. MH-LL Hr-Lr l[h-l]
*
2. MM-LL Hr-Lr [l-l]
*!W L
Tableau (23) shows us how MH-LL is more optimal than the faithful MM-LL which violates OCP-t(l)&*HD/Lr constraint. This ranking shows us the preference of Bangkok Hakka grammar towards low tone melody dissimilation over tone melody assimilation.
59
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
4.3.4 Register Tier Preservation
In every disyllabic tonal alternation in Bangkok Hakka, it is evident that every tone that changes its value keep its original register value as in Meixian Hakka. For example, when it is a high register tone, it always has high register sandhi tone and the same situation also takes place for the low register tone. The stability of the register is expressed into a similar identity constraint as follow
(24) IDENT-reg- Assign one violation mark for every tone’s register value which is not identical from the input.
(25) Preservation of high register value MM-LL
Hr-Lr l-l
IDENT-reg OCP-t(l) & *HD/Lr
1. MH-LL Hr-Lr l[h-l]
2. MM-LL Lr-Lr [h-l]
*W
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(26) Preservation of low register value ML-LL
Lr-Lr hl-l
IDENT-reg OCP-t(l) & *HD/Lr
1. MM-LL Lr-Lr [h-l]
2. MH-LL Hr-Lr l[h-l]
*W
Tableau (25) and (26) present conditions where the tones may be different in Bangkok Hakka due to the markedness constraints but the changes are not allowed in their register tier so sandhi tones in Bangkok Hakka always have similar register from its input tone.
4.3.5 Initial Target Preservation
This thesis follows the assumption that every tone in Chinese dialects has initial target specifications (cf. Yip 2001). The assumption is kept in Bangkok Hakka as shown in the following sandhi tones.
1. Yinping MM → MH, *HM, *HH
2. Qusheng HH → HM , *LL, *MM, *MH
This means that it is obligatory that every tone in Bangkok Hakka preserve their initial target and any changes of it will violate a faithfulness constraint which is expressed as follows:
61
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(27) IDENT- IT: Assign one violation for every initial target of tone which is not identical to input.
(28) Preservation of the initial target of Bangkok Hakka tone.
MM-LL Hr-Lr l-l
IDENT-IT
1. MH-LL Hr-Lr l[h-l]
2. HH-LL Hr-Lr [h-l]
*!W
Tableau (28) shows the ranking argument for two candidates that undergo
changes. The tone’s initial target of candidate 2 in the tableau alternates from M to H so it violates the undominated IDENT-IT. Candidate 1 wins as the initial target remain the same input-output.
4.3.6 Tone Duration Preservation
Due to the syllable’s property which determine the length of tone to which it is
attached to the syllable, tone alternations in Bangkok Hakka are prevented from changes in their duration value. In this analysis, the identity which tones should attach to what
syllable is protected by two constraints (Zhang 1998
)
which both are highly ranked toprevent the mismatching tone length. However, the mismatching of the tonal duration in this dialect is prevented by a faithfulness constraint.
(29) IDENT-TD- assign one violation mark for every tone which changes its duration.
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
(30) Preservation of tone duration
HH-H koŋ tshap Hr-Hr
h-h
IDENT-TD OCP-T(H)
&*HD/ST
1. HM- H Hr-Hr [hl-l]
2. M- H Hr-Hr [l-h]
Tableau (30) presents two candidate sets which do not violate markedness constraint. Candidate two changes its tonal duration and thus it violates tone duration constraint so it is ruled out. Candidate 1 obeys both faithful duration and surface as an optimal output.
63
‧
國立 政 治 大 學
‧
N a
tio na
l C h engchi U ni ve rs it y
4.4 Constraint Ranking and Overall Summary of Tonal Pairs
In this chapter, the interaction faithfulness and markedness constraints for
analyzing the Bangkok Hakka tone sandhi is shown. The faithfulness constraints preserve the elements of the tone while the markedness constraint chooses certain sandhi form.
Following, we can find the Hasse diagram to express the relation among constraints as the reflection of Bangkok Hakka’s grammar of disyllabic tonal alternations.
(31) Hasse Diagram of Bangkok Hakka Tone sandhi
Table (31) provides an overall test for the constraint hierarchy. The test let us see how the constraints work with kinds of tone candidates available in tonal inventories of Bangkok Hakka.
IDENT-T-HD
IDENT-T-Lr, H IDENT-reg IDENT- IT IDENT-TD
OCP-t(l)&*HD/Lr NOJUMP-t &*HD/MM OCP-c(h) &*HD/ST
NOJUMP-t &*HD/Lr
NOJUMP-t &*HD/Lr