國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
47
(39) The Hasse Diagram of rhymes without a nasal coda
PHONOTACTICS *COMP-NUC
AGREE[bk]-VG AGREE[rd]-VG IDENT[syl]
AGREE[bk]-GV AGREE[rd]-GV IDENT[F]-G MAX-CODA *[G2αF]σ
IDENT[F]-V MAX OCP-HIVOCOID DEP-LOWV
*GαF DEP-MIDV
To have an overview of all the rhyme patterns without a nasal coda, the constraint ranking in diagram (39) can correctly predict their presence and absence, except for an unallowed rhyme *[ɥa]. Because there are no other absent patterns that seem to be symmetric to this rhyme, it is reasonable to consider the absence of *[ɥa] an accidental gap.
4.2 Rhymes with a Nasal Coda
4.2.1 High Vowel Preceding a Nasal Coda
In section 2.2.1, it is mentioned that Lin (2007) has a hypothesis stating that there are two forms of the VX structure, diphthongs and “high/low vowel + nasal.” Based on her assumption, mid vowels never precede nasal codas in the underlying representation.
In other words, all the mid vowels preceding a nasal coda in the surface representations are inserted or derived from other underlying forms. Her theory asserts that an [ə] is
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
inserted between a high vowel and a nasal coda when they have different values for the [back] feature. This indicates that the structure of “high vowel + nasal” and that of
“glide + ə + nasal” belong to the same phonemic group (see 3.3). A brief chart is arranged in (40) below.
(40) The outputs of underlying “high vowel + nasal”
a. in *jən b. *un w n c. yn *ɥən
d. iŋ *jəŋ e. *uŋ *wəŋ f. *yŋ *ɥəŋ
There are two problems having to be solved in chart (40). First, there should be accurately one allowed rhyme in each item, but no rhymes can be surfaced out in (e) and (f). The other problem lies in (d). According to Lin’s theory, an [ə] should be inserted when a high vowel and its adjacent nasal coda have different values for the [back] feature, but it is [iŋ] that is surfaced rather than *[jəŋ], with [i] and [ŋ] differing in the [back] feature.
4.2.1.1 The Regular Patterns [in], [wən] and [yn]
Compare the patterns in (40a – c) to those in (40d – f), it is known that patterns with an alveolar nasal [n] are more regular than those with a velar nasal [ŋ]. Therefore, the patterns with [n] is going to be analyzed first. As Lin’s (2007) hypothesis, an [ə] is inserted between only when the high vowel and its following nasal coda have different values for the [back] feature. Based on this phenomenon, a preliminary descriptive generalization is introduced in (41), and the related constraints are illustrated in (42 –
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
49
[ə] are both [+back]. If we assume that all vowels should agree with its nasal coda in [back] feature, then inserting an [ə] between [u] and [n] does not help repair the ill form.
(41) A preliminary descriptive generalization for high vowel preceding a nasal coda A high vowel should have a different value for its [back] feature from its nasal coda.
This requirement is enforced by [ə]-insertion.
(42) AGREE[bk]-VHN
Assign one violation mark for every syllable whose high vowel and nasal coda do not agree in the [back] feature.
(43) DEP-ə
Assign one violation mark for every [ə] in the output which does not have a correspondent input.
(44) DEP-e
Assign one violation mark for every [e] in the output which does not have a correspondent input.
(45) DEP-o
Assign one violation mark for every [o] in the output which does not have a correspondent input.
(46) IDENT[bk]-N
Assign one violation mark for every nasal coda which has a different value for the [back] feature from its input.
‧
Assign one violation mark for every high vowel which has a different value for the [F] feature from its input, where [F] = [back] or [round].
Constraint (42) avoids a high vowel to have a different value for its [back] feature from the following nasal coda. Constraint (43) indicates the strategy of inserting an [ə], which is violable, so it is ranked low. Constraints (44) and (45) are developed to avoid inserting other mid vowels rather than [ə]. AGREE[F]-GV (which in fact includes two constraints) acts as a low-ranked constraint here, for [ə] does not agree in [back] and [round] features at the same time with any glide. Constraints (46 – 47) are here to prevent the vowel and the nasal from changing their features. Tableau (48) is the example of how these constraints are ranked.
(48) /un/ → [wən]
‧
In 4.1.1, it is proved that AGREE[F]-GV dominates IDENT[F]-V. For example, /jə/ is surfaced as [je] to satisfy AGREE[F]-GV by violating IDENT[F]-V. However, in tableau (48), candidate (d), which violates the lower-ranked IDENT[F]-V, is worse than candidate (a), which violates the higher-ranked AGREE[rd]-GV. Therefore, it is assumed that there should be a constraint that is ranked even higher to rule out (d) [yn]. For this reason, we rank the constraint IDENT[F]-HIV higher than AGREE[F]-GV, which specifically restricts a high vowel to change its feature. This is very reasonable because there are three high vowel phonemes in Mandarin but only one mid/low phoneme. Changing the features of a high vowel may lead to confusion.
There are still some facts worth mentioning. First, some may argue that there are several possible candidates not listed in tableau (48), such as *[uən], [wan], [u] and *[n].
In section 4.1.3, we have developed a constraint *COMP-NUC to avoid vowel clusters, so
*[uən] is absolutely worse than [wən], for *COMP-NUC >>IDENT[syl]. As for [wan], also mentioned in 4.1.3, inserting a low vowel is worse than inserting a mid vowel, for DEP -LOWV >> DEP-MIDV, so [wan] is also absolutely worse than [wən]. Candidates [u] and [n] violate MAX, which has been proved ranked higher than DEP-MIDV, so there are no reasons for them to be the optimal outputs.
Second, though [o] does not agree with [n] in [back], it does agree with the prenuclear glide [w]. Therefore, DEP-o still needs to be developed here and ranked higher than AGREE[F]-GV. Furthermore, in 4.1.3, a constraint DEP-MIDV is developed to assign violation marks for every mid vowel which does not have a correspondent input.
The constraint is there to ensure that a mid vowel is inserted between two high vocoids.
(e.g. /ju/ → [jow].) In this section, it is found that inserting an [ə] is better than inserting an [e] or [o], even though all of them are mid vowels. Therefore, there should be stricter constraints to define inserting which mid vowel is better or worse, just like what MAX and MAX-CODA in 4.1.2 do. The example of /ju/ becoming [jow] inserts an [o], which
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
therefore violates not only DEP-MIDV, but also the stricter DEP-o, as (49) shows.
Likewise, when it comes to the example of /wi/ → [wej], the constraint ranking works the same. Note that DEP-e is not active in example (49) because *[jew] is absolutely a worse choice than [jow], for it violates AGREE[F]-VG. Neither is DEP-o active in (50).
Just make sure that DEP-e and DEP-o are always ranked together.
(49) /ju/ → [jow]
/ju/ OCP-HIVOCOID DEP-LOWV MAX AGREE[F]-VG DEP-o DEP-ə
a. → jow *
b. jɑw *!
c. u *!
d. ju *!
e. jəw *! *
(50) /wi/ → [wej]
/wi/ OCP-HIVOCOID DEP-LOWV MAX AGREE[F]-VG DEP-e DEP-ə
a. → wej *
b. waj *!
c. i *!
d. wi *!
e. wəj *! *
By reviewing the mid vowel insertion between two high vocoids, it is found that
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
53
for the [back] features is repaired by inserting [ə]. Tableaux (49) and (50) have helped explain the difference. The choice of [wən] instead of *[wen] proves that DEP-e (as well as DEP-o) dominates DEP-ə, but [wej] is better than *[wəj], for AGREE[F]-VG is ranked even higher than DEP-e.
The constraint ranking is summarized as Hasse Diagrams in (51). Diagram (a) illustrated the constraint ranking for the [ə]-insertion between high vowel and nasal; (b) revises the constraint ranking for the [e]/[o]-insertion between two high vocoids in 4.1.3;
(c) concludes (a) and (b) by combining them together.
(51) The Hasse Diagrams of mid vowel insertion a. [ə]-insertion between high vowel and nasal
AGREE[bk]-VHN IDENT[bk]-N DEP-e DEP-o IDENT[F]-HIV
DEP-ə AGREE[bk]-GV AGREE[rd]-GV b. [e]/[o]-insertion between two high vocoids
OCP-HIVOCOID DEP-LOWV MAX AGREE[bk]-VG AGREE[rd]-VG
DEP-e DEP-o
DEP-ə c. Mid vowel insertion
OCP-HIVOCOID DEP-LOWV MAX AGREE[bk]-VG AGREE[rd]-VG
AGREE[bk]-VHN IDENT[bk]-N DEP-e DEP-o IDENT[F]-HIV
DEP-ə AGREE[bk]-GV AGREE[rd]-GV
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
In (51a), DEP-ə is ranked the lowest because inserting an [ə] is the best policy to repair the ill forms. At the same time, since AGREE[bk]-GV and AGREE[rd]-GV are both violated, for [ə] does not agree with the prenuclear glide, they are also dominated by DEP-e, DEP-o and IDENT[F]-HIV. On the other hand, AGREE[bk]-VHN and IDENT[bk]-N are only proved to dominate DEP-ə because they do not contradict with AGREE[bk]-GV and AGREE[rd]-GV.
Hasse Diagram (51b), revised from the Hasse Diagram (38) in 4.1.3, separates the constraint DEP-MIDV into DEP-e, DEP-o and DEP-ə, with DEP-e and DEP-o dominating DEP-ə. In these cases, the mid vowels inserted between the high vocoids are [e] and [o]
instead of [ə] in order to satisfy AGREE[bk]-VG and AGREE[rd]-VG, so they also dominate DEP-e and DEP-o. Combining (51a) and (51b) can obtain diagram (51c).
4.2.1.2 The Irregular Outputs of /iŋ/, /uŋ/ and /yŋ/
Based on the constraint ranking in (51), AGREE[bk]-VHN is a high-ranked constraint which is inviolable. However, the rhyme [iŋ] does violate this constraint. The output of /iŋ/ is predicted as *[jəŋ] through this constraint ranking, as (52) shows, but
*[jəŋ] does not appear in Taiwanese Mandarin. Here we do not consider other candidates like [joŋ] or *[jeŋ] because such forms are proved worse in the previous part.
(52) /iŋ/ → *[jəŋ]
/iŋ/ AGREE[bk]-VHN DEP-ə
a. 💣 jəŋ *
b. iŋ *!
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
55
Because the constraint ranking wrongly predicts the output of /iŋ/, it is assumed that there is a constraint ranked even higher to rule out the predicted output *[jəŋ]. To further observe the patterns, it is found that in addition to the absence of *[jəŋ], *[ɥəŋ], which is predicted as the output of /yŋ/, is also not allowed. Hence, we assume that the combination of [əŋ] is very marked in Taiwanese Mandarin, which blocks *[jəŋ] and
*[ɥəŋ] from surfacing out. The descriptive generalization is thus revised as (53), and the constraint is shown in (54) with its definition.
(53) The revised descriptive generalization for high vowel preceding a nasal coda When a high vowel has a different value for the [back] feature from its nasal coda, an [ə] is inserted, unless it leads to the sequence of [əŋ].
(54) *əŋ
Assign one violation mark for every sequence of [əŋ].
This constraint seems ad hoc. However, Taiwanese Mandarin is not the only dialect in which the sequence of [əŋ] has something to do with irregular patterns. In Beijing dialect, the word weng ‘old man’ is pronounced as [wəŋ] (Lin, 2007), where an [ə] is inserted even though the high vocoid and the nasal coda are both [+back]. This example supports that [əŋ] is exactly a highly-marked sequence. Because [iŋ] is a better candidate than *[jəŋ], the constraint *əŋ should dominate AGREE[bk]-VHN, as is illustrated in tableau (55).
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(55) /iŋ/ → [iŋ]
/iŋ/ *əŋ AGREE[bk]-VHN DEP-ə
a. → iŋ *
b. jəŋ *! *
The constraint *əŋ successfully rules out *[jəŋ] as well as *[ɥəŋ], a possible output of /yŋ/, but *[yŋ] is also not surfaced out even though it should be the optimal output according to the constraint ranking in (55). Now that neither *[yŋ] nor *[ɥəŋ] is allowed, then how does /yŋ/ surface out? According to the “phonemic group” introduced in 3.3, when X = [n] or [ŋ], there should be only one choice among ∅, [e], [ə] and [o], as the shadowed boxes in chart (56) illustrate. Nevertheless, the list in (57) below proves that there is something wrong under this assumption.
(56) Phonemic group indication chart (G = any)
V \ X ∅ j w n ŋ
(i, u, y) e ə o a ɑ
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
57
(57) The combinations of GVN, where V = ∅, e, ə, o. The arrows illustrate the derivation prediction.
a. *n *en n *on e. *un *wen w n *won
b. *ŋ *eŋ ŋ oŋ f. *uŋ *weŋ *wəŋ *woŋ
c. in jen *jən *jon g. yn ɥen *ɥən *ɥon d. iŋ *jeŋ *jəŋ joŋ h. *yŋ *ɥeŋ *ɥəŋ *ɥoŋ
The underlined gray patterns in (57c) and (57g) will be later discussed in 4.2.3.2, for the underlying form of the vowels are low ones. In addition to these patterns, it is found that there are two allowed patterns in (57b) and (57d), while none is allowed in (57f) and (57h). Because the rhyme [joŋ] in (57d) contains both [-back] and [+round]
features of [y], it is assumed that /yŋ/ is surfaced as [joŋ], and symmetrically, /uŋ/ is surfaced as [oŋ].
The fact that /yŋ/ is surfaced as [joŋ] indicates that the vowel /y/ in the input undergoes a process of splitting. Historically, [y] is the combination of [ju], with [j]
reflecting the [-back] feature and [u] the [+round] feature, and all of the three being [+high] (Duanmu, 2007). Splitting the [y] preceding [ŋ] into two segments does prevent a [-back] vowel from being adjacent to a [+back] nasal coda and avoid the sequence of [əŋ], so this is the best strategy to repair the ill-formed *[yŋ]. Splitting a segment violates a faithfulness constraint, though. Constraint (58) is therefore developed;
however, tableau (59) still wrongly predicts the output of /yŋ/ as *[juŋ], for [y] is split into [ju] instead of [jo].
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(58) INTEG
Assign one violation mark for every segment in the input which has more than one correspondent outputs.
(59) /yŋ/ → *[juŋ]
/yŋ/ *əŋ AGREE[bk]-VHN DEP-ə INTEG
a. 💣 juŋ *
b. yŋ *!
c. ɥəŋ *! *
There are two explanations for the ill-formed *[juŋ]. First, as discussed in 4.1.3, two high vocoids cannot be adjacent. Splitting [y] into [ju] does violate the constraint OCP-HIVOCOID. The strategy to repair the ill forms is to insert [e] or [o] between the high vocoids. By doing so, other ill-formed syllables are derived, such as *[jouŋ] and
*[jowŋ]. To rule out *[jouŋ], the constraint *COMP-NUC should be active here, but gliding the high vowel only leads to another ill form *[jowŋ], which should be ruled out by introducing a new constraint to prohibit complex codas. The only strategy to avoid all these ill forms is to change the value for the [high] feature of the vowel, and the optimal output [joŋ] can finally be derived. This analysis surely explains why [joŋ]
is chosen as an optimal output for /yŋ/, but there is another explanation, which is much more economic.
To make the other explanation of why /yŋ/ is surfaced as [joŋ], we have to focus on the output of /uŋ/ first. As the assumption of /yŋ/ surfacing as [joŋ], it is found that both [əŋ] and [oŋ] in (57b) are allowed while no patterns can be surfaced in (57f).
‧
derived from /uŋ/. Now that /uŋ/ is surfaced as [oŋ], it is not surprised for *[juŋ] to be repaired as [joŋ]. Therefore, by explaining the derivation of [oŋ], we can at the same time solve the problem of the surface form of /yŋ/.
Actually, the phenomenon that [u] is lowered as [o] when preceding a velar coda is also observed in other Sinitic languages, such as Southern Min and Cantonese. In Southern Min, there are three back vowels, namely [o], [ɔ] and [u]; however, only [ɔ]
can precede a velar coda [ŋ] or [k], which means that the combinations of a high back vowel and a velar coda *[uŋ] and *[uk] are prohibited in Southern Min (Chung, 1995).
On the other hand, in Cantonese, there is a distinction between the rhymes [ɔŋ] and [oŋ]
and between [ɔk] and [ok], as the examples in (60). However, [o] does not solelyexist in the rime and moreover, there is no *[uŋ] or *[uk]. Therefore, it is assumed that [oŋ]
and [ok] are also derived from /uŋ/ and /uk/ (Yip and Matthews, 2006).
(60) The minimal pairs of [ɔ] and [o] in Cantonese
mok [mɔk] ‘do not’ muk [mok] ‘wood’
dong [tɔŋ] ‘swing’ dung [toŋ] ‘move’
Although the stop coda [k] is not allowed in Mandarin, it is observed that in other languages where [u] is lowered when preceding the velar nasal [ŋ], like Southern Min and Cantonese, the velar stop [k] also triggers the [u]-lowering. Therefore, we conclude that two high back segments cannot be adjacent in the rhyme. The descriptive generalization is illustrated in (61), and constraints (62 – 63) are therefore introduced.
(61) The descriptive generalization for [u]-lowering
Two high back segments cannot be adjacent in the rhyme domain. This requirement is enforced by lowering the vowel.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(62) OCP-[+hi, +bk]
Assign one violation mark for every pair of adjacent high back segments.
(63) IDENT[hi]-V
Assign one violation mark for every vowel which has a different value for the [high]
feature from its input.
In addition to constraints (62) and (63), *əŋ and DEP-o are also active here as high-ranked constraints to avoid predicting ill-formed candidates like *[wəŋ] and *[woŋ] as an output, as illustrated in tableau (64). Note that candidate (64c) also violates DEP-ə, but this constraint is not considered here because DEP-o has been proved to dominate DEP-ə and the candidate should have been ruled out when evaluated by *əŋ. The Hasse Diagrams in (65) concludes the constraint rankings for the rhymes [iŋ], [oŋ] and [joŋ].
(64) /uŋ/ → [oŋ]
/uŋ/ OCP-[+hi, +bk] *əŋ DEP-o IDENT[hi]-V
a. → oŋ *
b. uŋ *!
c. wəŋ *!
d. woŋ *!
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
61
(65) The Hasse Diagrams of rhymes underlyingly “high vowel + [ŋ]”
a. The unallowed sequence [əŋ]
*əŋ
AGREE[bk]-VHN
INTEG DEP-ə b. [u]-lowering
OCP-[+hi, +bk] *əŋ DEP-o
IDENT[hi]-V c. Combining to predict [joŋ]
*əŋ DEP-o
AGREE[bk]-VHN OCP-[+hi, +bk]
INTEG DEP-ə IDENT[hi]-V
Diagram (65a) shows that *əŋ is ranked the highest because it is inviolable. In order to satisfy this constraint, AGREE[bk]-VHN can be violated to surface [iŋ]. However, when it comes to /yŋ/, INTEG is violated to satisfy AGREE[bk]-VHN, so AGREE[bk]-VHN dominates INTEG. That AGREE[bk]-VHN >> DEP-ə has been illustrated in (51).
Diagram (65b) shows that when two high back segments are adjacent, the best policy to repair this ill form is lowering the vowel instead of inserting other vowels.
Therefore, IDENT[hi]-V is ranked the lowest. Combining (65a) and (65b) can obtain (65c), where DEP-o >> DEP-ə is proved in previous analysis.
‧
4.2.2 [ən] and [əŋ] without a Prenuclear Glide
In 4.2.1.2, it is concluded that *[jəŋ] and *[ɥəŋ] do not surface out successfully as the outputs of /iŋ/ and /yŋ/ because the marked sequence of [əŋ] is not allowed in Taiwanese Mandarin. However, there are a lot of examples in Taiwanese Mandarin which do have this sequence in the rhyme, such as deng [təŋ] ‘light’, keng [khəŋ] ‘hole’, sheng [ʂəŋ] ‘voice, etc. Why are these syllables allowed while *[jəŋ] and *[ɥəŋ] are not?
To solve this problem, we have to mention the issue proposed in 2.2.1 again: Lin (2007) states that only high vowels and low vowels can precede a nasal coda in the underlying representation, and therefore the mid vowels preceding a nasal coda in the surface forms are inserted or derived from high/low vowels. Based on this hypothesis, we assume that the [ə]’s in the words like deng, keng, and sheng are also inserted. With this assumption, the underlying forms of the rhymes [ən] and [əŋ] are considered to be /n/ and /ŋ/ respectively.
This hypothesis does not violate Lin’s statement that there are no mid vowels preceding a nasal coda in the underlying representation, and it can also explain why the marked sequence [əŋ] can still surface out in syllables without a prenuclear glide (which have no vowels in the underlying forms), for DEP-ə is the lowest-ranked constraint among all of the constraints which prohibit a vowel to be inserted. To develop the constraints, a descriptive generalization is introduced first in (66), and constraint (67) is developed to ensure the existence of the nucleus. IDENT[syl] is high-ranked here to avoid syllabic nasals. *əŋ is ranked lower than NUC and IDENT[syl] here, for the sequence of [əŋ] is surfaced in order to fulfill the requirement of having a nucleus. Though inserting an [ə] causes [əŋ], it is still a better choice than inserting an [o]. Therefore,
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
63
(66) The descriptive generalization for inserting [ə] as a nucleus
A syllable should have a nucleus. This requirement is enforced by [ə]-insertion.
(67) NUC
Assign one violation mark for every syllable which has no nucleus.
(68) /ŋ/ → [əŋ]
/ŋ/ NUC IDENT[syl] DEP-o *əŋ DEP-ə
a. → əŋ * *
b. oŋ *!
c. ŋ *!
d. ŋ̩ *!
High vowels are not considered to be inserted in example (68) because we have known that both of *[uŋ] and *[yŋ] are ill-formed rhymes through the analysis in 4.2.1.2.
Even though [iŋ] can be surfaced out, it could be interpreted that [iŋ] will not be the optimal candidate because of its violation of AGREE[bk]-VHN. However, if we consider the other pattern of the same phenomenon, i.e. /n/ surfaced as [ən], inserting a high vowel could be a possible strategy to repair this ill form. To avoid this prediction, constraint (69) is developed, which is also high-ranked, as tableau (70) shows. The Hasse Diagram is illustrated in (71).
(69) DEP-HIV
Assign one violation mark for every high vowel in the output which does not have a correspondent input.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(70) /n/ → [ən]
/n/ NUC IDENT[syl] DEP-HIV DEP-ə
a. → ən *
b. n *!
c. n̩ *!
d. in *!
(71) The Hasse Diagram of the inserted [ə] as a nucleus
NUC IDENT[syl] DEP-HIV DEP-o
*əŋ
DEP-ə
As (71) shows, DEP-ə is dominated by NUC, IDENT[syl] and DEP-HIV because only DEP-ə is violable when it comes to a rhyme composed of only a nasal consonant in the input.
4.2.3 Low Vowel Preceding a Nasal Coda
4.2.3.1 Low Vowel Backing
In 4.2.1.1, it has been stated that high vowels should agree with the nasal coda in [back] feature while mid vowels do not have this restriction. This is proved by the fact
‧
[wən], where the vowel and the nasal coda still do not agree in [back] with each other.
Then what about low vowels? To answer this question, we just have to take a look at the patterns [an] and [ɑŋ]. Obviously, the low vowel is surfaced as [-back] when preceding [n] and [+back] when preceding [ŋ]. By this fact, it is known that low vowels, like high vowels, should also agree with the nasal coda in [back], but the strategy to repair the ill forms is different. As the strategy for low vowels to agree with the glide coda, they also change the value for the [back] feature to agree with the nasal coda.
In fact, it can be generalized that vowels except for mid ones should agree with the nasal coda in [back] feature, and the best strategy to repair the ill forms is to insert an [ə], unless the vowel cannot be glided (i.e. low vowels) and thus causes a complex nucleus. The analysis can be supported by the constraint ranking as below shows, where the lower index L stands for low:
AGREE[bk]-VH/LN, *COMP-NUC >> IDENT[F]-V >> DEP-ə
However, it is not natural for a constraint to restrict the nasal coda to agree with its preceding “high or low” vowels with the mid vowels excluded. Therefore, here we tend to define two different constraints to regulate a high vowel and a low vowel to agree with the nasal coda in [back] feature respectively. One is AGREE[bk]-VHN, which has been developed previously, and the other is illustrated in (73). The descriptive generalization is developed in (72), and an example is displayed in tableau (75).
Constraint (74) is to ensure that the low vowel is not raised in this environment. As what is mentioned above, *COMP-NUC is also active here for prohibiting an inserted [ə]
causing a complex nucleus. IDENT[bk]-N is also listed in the tableau to ensure that the nasal coda does not change.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(72) The descriptive generalization for low vowel backing
A low vowel should agree with its nasal coda in [back] feature. This requirement is enforced by vowel backing.
(73) AGREE[bk]-VLN
Assign one violation mark for every syllable whose low vowel and nasal coda do not agree in the [back] feature.
(74) IDENT[low]-V
Assign one violation mark for every vowel which has a different value for the [low]
feature from its input.
(75) /aŋ/ → [ɑŋ]
/aŋ/ AGREE
[bk]-VLN *COMP-NUC IDENT[bk]-N IDENT[low]-V IDENT[F]-V
a. → ɑŋ *
b. aŋ *!
c. an *!
d. aəŋ *!
e. eŋ *!
When analyzing the phenomenon of inserting an [ə] between a high vowel and a nasal coda in 4.2.1.1, we ranked IDENT[bk]-N and IDENT[F]-V the same high because we had not had any evidence to compare these two constraints yet. In this part, it is
‧
the feature of the nasal coda. Therefore, that IDENT[bk]-N >> IDENT[F]-V is proved here.
If we take the patterns with a prenuclear glide into consideration, it is found that the low vowel still agrees with the nasal coda in [back] feature even though the prenuclear glide has a different value for the [back] feature from it, as tableau (76) shows. This proves that AGREE[bk]-VLN >> AGREE[bk]-GV10 , while we have no evidence to prove which one of AGREE[bk]-VHN and AGREE[F]-GV is ranked higher in 4.2.1.1. The Hasse Diagram is illustrated in (77) below.
(76) /jaŋ/ → [jɑŋ]
/jaŋ/ AGREE[bk]-VLN IDENT[low]-V AGREE[bk]-GV IDENT[F]-V
a. → jɑŋ * *
b. jaŋ *!
c. jeŋ *!
(77) The Hasse Diagram of low vowel backing
AGREE[bk]-VLN IDENT[low]-V *COMP-NUC IDENT[bk]-N
AGREE[bk]-GV
IDENT[F]-V
Hasse Diagram (77) shows that when a low vowel precedes a nasal coda, it should agree with the coda in [back] feature even though it violates AGREE[bk]-GV and
Hasse Diagram (77) shows that when a low vowel precedes a nasal coda, it should agree with the coda in [back] feature even though it violates AGREE[bk]-GV and