CHAPTER 4 OT ANALYSIS
4.5 Tonal Constraints for Mandarin Regular Tone Sandhi
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
In short, the occurrence of focus may result in prosodic restructuring, and this section shows that the phrasing of the focal domain is determined by the interaction between alignment and foot form constraints.
4.5 Tonal Constraints for Mandarin Regular Tone Sandhi
Mandarin, like Min and Southern Wu dialects, belongs to a right-prominent tonal system. That is because these dialects tend to maintain the identity of the rightmost tone of words or even prosodic domains, while allowing tones to change on the non-final tones in sandhi contexts (Chen 1996). In addition, adjacent L tones are prohibited in Mandarin. Based on the above properties, Lin (2000) proposes a set of constraints to account for Mandarin tone sandhi.
(34) OCP-L: adjacent L tones are prohibited.
(35) IO-IDENT: the output tones must be identical to the input tones.
(36) PARSER: parse the rightmost tone within a domain.
Tableau (37) demonstrates the ranking of PARSER, OCP-L, and IO-IDENT and the evaluation of possible candidates.
‧
In (37), candidate (37b) is ruled out because ban, which is the rightmost tone within the Mandarin tone sandhi domain, is changed to a sandhi tone. On the other hand, candidate (37c) violates the OCP-L constraint because there are two adjacent L tones in the output. Consequently, candidate (37a) is selected as the optimal output. The ranking for Mandarin tone sandhi is PARSER >> OCP-L >> IO-IDENT.
However, there is a problem in the PARSER >> OCP-L ranking. That is, PARSER and OCP-L dominate IO-IDENT, but their ranking with respect to one another is unknown. Since candidate (37a) does not violate PARSER and OCP-L, the switching of these two constraints by no means changes the optimal output to candidate (37b) or (37c). In other words, if OCP-L dominates PARSER, the optimal output is still (37a).
Since they are unranked with respect to one another, the constraint ranking is PARSER, OCP-L >> IO-IDENT.
In addition, Parse is the anti-deletion faithfulness constraint. However, it has the
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
form of a markedness constraint because it evaluates only the output and not the input-output relation in the early OT literature. The Parse constraint can be replaced by the IDENT constraint (McCarthy 2008). Hence, I adopt IDENT-TONE-RIGHT to substitute PARSER.
(38) IDENT-T-R: an output tone at the right edge must be identical to that in the input.
(Hsiao 2000)
As discussed in examples (7) and (8), trisyllabic strings, either right branching or left branching, will always constitute a single foot. Tableaus (39) and (40) demonstrate the tonal evaluation of the present constraint ranking in trisyllabic strings.
Examples are taken from (7) and (8) in Chapter 4. The tonal output of (39) is (L.(LH.L)) while the tonal output of (40) is ((LH.LH.)L).
(39) zhao xiao li ‘Looking for Small Li.’
zhao xiao li (L(LL))
IDENT-T-R OCP-L(ft) IO-IDENT
a. (LH (LH L)) **!
☞b. (L (LH L)) *
c. (L (L L)) *!
d. (L (LH LH)) *! **
‧
The present constraints can account for the tonal output of right branching structure as in (39), but cannot account for the tonal output of left branching structure as in (40).
For (40), candidate (40a) violates IO-IDENT twice because it contains two LH tones.
Candidate (40b), which has only one violation in the IO-IDENT, is wrongly selected as the optimal output.
Obviously, the present constraints cannot make a correct prediction in the odd-numbered syllables in Type B, indicating that a new constraint is needed to rule out candidate (40b) in tableau (40). Pulleyblank (2004), in order to explain the unstability of M tone in Yoruba, argues that a language specific tonal scale is needed, where M is the least marked tone, H is the most marked tone, and L is intermediate5
5 For example, rí īgbɑ́ ‘see a calabash’ is realized as rígbɑ́ in connected speech, the initial H tone being retained at the expense of the M. M tones are also lost in favor of L tones, as can be seen in a case such as sē ɑ̀jé ‘engage in witchcraft’, realized as sɑ̀jé to avoid vowel hiatus. (Pulleyblank, 1986)
.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
This proposal is inconsistent with the recent suggestion by de Lacy (2002a, b) that the tonal prominence scale is based strictly on descending pitch height, with H more prominent than M, which is in turn more prominent than L. Pulleyblank further claims that there is no single, cross-linguistic valid scale of tonal prominence and languages may vary in the precise tonal scales that they adopt.
Yin (1991) postulates that the default tone in Mandarin is the high level tone (H) based on the evidence from kinship reduplicates, onomatopoeic words, the reading of foreign alphabets and letters, national phonetic alphabet, and a disguised language.
Some evidence is given below in (41).
(41) a. sawL sawH ‘elder brothers’ wife’
b. phiH liH phaH laH ‘sound of fire-crackers etc.’
c. eyH tçyaH biH təŋL üLH çiH ‘A + B = C’
d. nanLH khanHL → nəH khəH ‘ugly’
Because the high level tone (H) has a high frequency of appearance in Mandarin, we assume that a L tone is marked in Mandarin. A markedness constraint, *L, which prohibits L tone in surface representation, may be a help in the evaluation of tri-tonal strings because (40b) contains more L tones than (40a). However, a LH tone is more
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
marked than a L tone due to the fact that a LH tone is a contour tone. Since a markedness constraint aims to forbid the marked form in a language, it is the LH tone which should be prohibited instead of the L tone. Thus, the *L constraint needs to be reconsidered.
In light of this, a properly defined tone sandhi domain is necessary. To deal with the tri-tonal strings such as (σ(σσ)) and ((σσ)σ), this study adopts the IDENT-BOT
constraint proposed by Lin (2005), as defined below.
(42) IDENT-BOT: corresponding tones in the prosodically related bases and outputs must be identical. (Lin 2005)
The correspondence of the related base and output is given in the following figure:
(43a) (σ(σσ)) (43b) ((σσ)σ)
Input Tone Input Tone Input Tone Input Tone
L.L L.L.L L.L L.L.L(LH.L)1 (L.(LH.L)2)3 (LH.L)1 ((LH.LH)2.L)3
Base Tone Output Tone Base Tone Output Tone
BOT-I
DENTITYBO
T-I
DENTITY‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
In (43a) and (43b), the two adjacent L tones which trigger tone sandhi serve as the base, (LH.L)1. However, the main difference between (43a) and (43b) lies in the position of the additional L tone. In (43a), a L tone precedes the base (LH.L)1, so the output is (L.(LH.L)2)3, whereas in (43b), a L tone follows the base (LH.L)1, so the second tone undergoes tone sandhi and thus changes to a LH tone, ((LH.LH)2.L)3. The IDENT-BOT constraint requires that the base tones, (LH.L)1, and the output tones, (LH.L)2 or (LH.LH)2, are identical. Tableau (44) shows how IDENT-BOT operates in a trisyllabic foot like ((σσ)σ).
(44) ju qi shou ‘Raise hands.’ Reference Output: (juLH qiL) ju qi shou
((LL)L)
IDENT-T-R OCP-L IDENT-BOT IO-IDENT
☞a. ((LH LH) L) * **
b. ((L LH) L) **! *
c. ((L L) L) *!
d. ((L LH) LH) *! ** **
In (44), the tone sandhi domain of ju qi shou is ((σσ)σ). The base (ju qi) and its correspondence in the ouput ((ju qi) shou) is evaluated by IDENT-BOT. For (44a) and (44b), the former incurs fewer violations than the latter with respect to the IDENT-BOT
‧
constraint. In (44a), only the second output tone in ((LH.LH.)L) is not identical to the second base tone in (LH.L), whereas in (44b), the two output tones (L.LH) within ((L.LH.)L) are not identical to the base tones in (LH.L). This is why candidate (44b) incurs more violations than candidate (44a). As a result, (44b) is ruled out and (44a) is correctly selected as the optimal output.
(45) zhao xiao li ‘Looking for Small Li.’ Reference Output: (xiaoLH liL) zhao xiao li evaluation. As exhibited in the tableau, both (45a) and (45b) satisfy the IDENT-BOT
constraint, but (45a) is ruled out by the IO-IDENT constraint. Thus, the optimal output is (45b).
In sum, the previous proposed *L constraint is invalid and the IDENT-BOT
constraint is adopted to replace *L as it plays a crucial role in the evaluation of trisyllabic tonal domains such as (σ(σσ)) and ((σσ)σ). The present tonal constraint
‧
ranking for Mandarin tone sandhi is listed below.
(46) Tonal Constraint Ranking for Mandarin Tone Sandhi (in regular speech):
IDENT-T-R, OCP-L >> IDENT-BOT >> IO-IDENT