CHAPTER 4 : Optimality Theory Analysis
4.1 Adaptations in Conformity to Mandarin Phonotactics
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
On the other hand, some of tokens violate Mandarin phonotactics (L1) rather follow such types of grammar: Korean phonology (L2) and an interlanguage phonology. The specific CV sequences that follow each of the Korean phonology and the interlanguage phonology are analyzed in this chapter.
The constraints in Korean, Mandarin and KM Kong-er are compared. First of all, constraints for adaptations conforming to Mandarin phonology are presented (4.1). Then, the marked adaptations that follow Korean phonology or an interlanguage phonology are analyzed by different constraint rankings in (4.2). 4.3 summarizes chapter 4.
4.1 Adaptations in Conformity to Mandarin Phonotactics
Mandarin speakers allow variations when they adopt Korean consonants in KM Kong-er. In this case, a consonant in the onset position can be preserved. The constraint MAX-SONSET is proposed to avoid onset deletion.
(1) MAX-SONSET:
Assign one violation mark for every input onset segment that does not have a correspondence in the output.
Since there is no variation form such as deleted onset, MAX-SONSET is located above the cut-off line, eliminating the redundant candidates.
‧
obsturents regardless of the form of the substitution (100%, 260/260). It is because the Korean tensed obstruent has a [-spread glottis] feature, which is perceived by Mandarin speakers as an alternative form during KM Kong-er adaptation. The markedness constraint *TENSE(C) is proposed to prevent the tensed Korean obstruents from occurring. To preserve the [-spread glottis] feature inherent in the Korean tensed obstruents, the local conjunction of constraint IDENT[tense]&*[+aspirated] is proposed.(2) *TENSE(C):
Assign one violation mark for every [tense] consonant in the output.
(3) [IDENT[tense]&*[+aspirated]]ONSET:
Assign one violation mark for every onset that violates both IDENT[tense] and
*[+aspirated].
The tensed consonant inputs are only mapped to the unaspirated outputs. This marked result leads to the proposal of a local conjunction constraint [IDENT[tense]&*[+aspirated]]ONSET. In the domain of the onset consonant, IDENT[tense]
and *[+aspirated] are locally conjoined. The markedness constraint *[+aspirated] is inactive individually. It is activated when the faithful constraint IDENT[tense] is violated (Łubowicz, 2002; Hsiao, 2015). The conjunct constraint is higher-ranked than each of IDENT[tense] and *[+aspirated] than either IDENT[tense] or *[+aspirated]. It functions
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
contrast in other type of adaptations, a faithfulness constraint IDENT[aspirated] allows variations of aspiration contrasts below the cut-off line.
Korean tensed obstruents are divided into two groups: one for tensed stops /p’, t’, k’/ and the other for tensed affricates and fricatives /tɕ’, s’/. These two groups are in different tableau due to the variations: /p’, t’, k’/ are only mapped to [p, t, k] and /tɕ’, s’/
are possibly mapped to either unaspirated counterparts [tɕ, s] or retroflexed counterparts [tʂ, ʂ].
Three constraints appear above the cut-off line in tableau (4). The constraints below the cut-off line are omitted. There is no ranking between the constraints since a domination relation does not decide the optimal candidate.
(4) /nun.p’i.tɕʰi/ → [nəŋ.pi.tɕhi] (能逼七) ‘the color of the eyes-SUBJ’
Input: /p’i/ MAX-SONSET *TENSE(C) [IDENT[tense]&*[+aspirated]]ONSET
☞ a. pi
b. phi *!
c. p’i *!
d. i *!
In the tableau (4), (4b) is ruled out by [IDENT[tense]&*[+aspirated]]ONSET. (4c) is also eliminated because tensed obstruents are banned by the markedness constraint *TENSE(C) in KM Kong-er. Lastly, candidate (4d) is ruled out by the constraint MAX-SONSET due to the deletion of the onset consonant.
‧
corpus, variants such as the retroflex post alveolars [tʂ] and [ʂ] also appear. This thesis employs the idea of cue constraint (Boersma, 1997). Boersma (1997) uses natural class to find out the roles of articulation and perception in phonology, and the constraints such as [+REPLACE] and [+GESTURE] are suggested. Hsiao (2011) also uses [+RETROFLEX] to discuss accent formation. We borrow the concept of the neutral feature [RETROFLEX] to propose the constraint *[retroflex]. It is used below the cut-off line to induce retroflex variants.(5) *[retroflex]:
Assign one violation mark for every retroflex consonant.
In tableau (6), a syllable /s’a/ is given as an example for illustration.
(6) /tɕʰʌt.s’a.raŋ/ → [tɕʰjou.sa.laŋ] (求撒朗) ~ [tɕʰjou.ʂa.laŋ] (求莎朗) ‘ first love’
In the tableau (6), previously proposed constraints are placed above the cut-off line to rule out the ill-formed candidates. They eliminate candidates (6c) and (6d). In this case,
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
constraint [IDENT[tense]&*[+aspirated]]ONSET is irrelevant since there is no aspirated /s/
in Mandarin. The constraint would rule out the aspirated affricate [tɕh] if an input is the tensed affricate /tɕ’/. At the same time, the retroflexed candidate (6b) is preserved as a variation by the cue constraint *[retroflex] below the cut-off line. As a result, the tensed fricative /s’/ in the output is mapped to two outputs [sa] (6a) and [ʂa] (6b).
4.1.1.2 Coda Deletion
Except for the codas /n/ and /ŋ/, other Korean consonants in the coda position are all illicit in Mandarin phonotactics. Accordingly, five consonants /p, t, k, m, l/ in the Korean coda position are disallowed by CODA CONDITION.
(7) CODA CONDITION:
Assign one violation mark for every coda segment that is not an alveolar nasal /n/, nor a velar nasal /ŋ/.
In KM Kong-er adaptation, there are two dominant strategies for illicit codas, deletion (48.73%) or replacement (20.90%), and one dominant strategy for licit codas:
retainment (30.37%). Let us first examine the adaptation of the illicit codas /p, t, k, m, l/.
The majority of the tokens are deleted, and are also replaced by variants such as /n, ŋ/. In order to generate these variations, the faithfulness constraints MAX-SCODA and IDENT[nasal]CODA are proposed below the cut-off line.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(8) MAX-SCODA:
Assign one violation mark for every input coda segment that does not have a correspondence in the output.
(9) IDENT[nasal]CODA:
Assign one violation mark for every input-output disparity of the [nasal] feature in the coda position.
The constraint MAX-SCODA is positioned below the cut-off line to let a candidate without a coda be mapped as a variant. Adjacent to MAX-SCODA, IDENT[nasal]CODA is positioned below the cut-off line to preserve the candidates replaced by the licit codas (n, ŋ).
IDENT[nasal]CODA dominates MAX-SCODA since the frequency of the variant with coda-deletion is higher than one with replacement. Tableau (10) illustrates an adaptation the result of an adaptation with the stop coda /p/.
(10) /pʰap.kʰon/ → [pʰa.kʰoŋ] (怕空) ‘pop-corn’
Input: pʰap CODA CONDITION IDENT[nasal]CODA MAX-SCODA
☞1 a. pʰa *
☞2 b. pʰan *
☞2 c. pʰaŋ *
d. pʰap *!
CODA CONDITION functions to rule out illicit codas (10d). The low ranking of M
AX-‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
c), which occur as the result of replacement, to be selected as variations. A solid line demonstrates the ranking of the constraints, indicating IDENT[nasal]CODA >> MAX-SCODA
based on the frequency of the variants.
Next, tableau (11) illustrates the adaptation of the lateral coda /l/. In the same way as the illustration with obstruent codas in (10), the lateral coda /l/ in (11d) is eliminated by the CODA CONDITION, and the deletion strategy (11a) and replacement strategy (11b-c) are chosen to be variations by the constraints below the cut-off line. The hierarchical relationship between IDENT[nasal]CODA and MAX-SCODA successfully induces the ranking of the outputs.
(11) /sal.tɕ’ak/ → [ʂa.tʂa] (沙炸) ‘lightly’
Input: sal CODA CONDITION IDENT[nasal]CODA MAX-SCODA
☞1 a. sa *
☞2 b. san *
☞2 c. saŋ *
d. sal *!
e. sap *!
In the case of the coda /m/, identical constraint ranking leads to a different candidate ranking because of the nasality. In the KM Kong-er corpus, the replacement of the coda /m/ by [n] or [ŋ] (80.43%) appears more than deletion (19.57%). With the previously proposed two constraints, a coda deletion strategy becomes the second frequent result
‧
during the EVAL. A syllable /pam/ is given as an example for illustration in tableau (12) below.
(12) /pam/ → [phaŋ (旁) ~n ~∅] ‘night’
Input: pam CODA CONDITION IDENT[nasal]CODA MAX-SCODA
☞1 a. pan IDENT[nasal]CODA and MAX-SCODA below the cut-off line lead the candidate (12ab) to be selected as an optimal result (replacement) and the candidate (12c) to be a second-most frequent variation (deletion).
Lastly, let us examine the strategy for the licit codas /n, ŋ/: retainment (30.37%).
The result shows that two codas are mostly retained, also showing replacement and deletion in order. To make achieve the result, the faithfulness constraint IDENT[anterior]CODA is placed below MAX-SCODA with a solid line (MAX-SCODA >>
IDENT[anterior]CODA). IDENT[anterior]CODA ensures the frequency hierarchy among the candidates since the codas /n/ and /ŋ/ can be differentiated by the [anterior] feature.
Tableau (14) illustrates the adaptation of the syllable /saŋ/ below.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(13) IDENT[anterior]CODA:
Assign one violation mark for every input-output disparity of the [anterior] feature in a coda position.
(14) /haŋ.saŋ/ → [xaŋ.saŋ] (夯桑) ‘always’
Input: saŋ CODA CON ID[nasal] CODA MAX-SCODA ID[anterior]CODA
☞1 a. saŋ
☞2 b. san *
☞3 c. sa *
d. sam *! *
e. sal *! * *
f. sap *! * *
As in (14), when the input is /ŋ/, a candidate with the coda /ŋ/ (14a) is not fatal according to any constraint, and thus becomes the most optimal and frequent output (retainment). The coda /n/ is fatal by IDENT[anterior]CODA (14b), and so is the second-most frequent variation (replacement). Lastly, the deletion of the input /ŋ/ (14c) is fatal by MAX-SCODA, and so it the least frequent variant, but it is still possible for it to be an output (deletion).
4.1.1.3 Glide Insertion
When there is a syllable with the Korean post-alveolar affricates /tɕ, tɕh, tɕ’/ in the source, glide insertion appears in the Kong-er adaptation (63.45%). /tɕ, tɕh, tɕ’/ is also
‧
proposed to adjust the Korean affricates into Mandarin affricates with the on-glide /j/.Also, IDENT[anterior]ONSET is proposed to prevent any alveolar affricate mapping with the dental sound [ts, tsh, s] in this section. Lastly, IDENT[aspirated] and DEP-S are proposed to allow for aspiration and segment insertion.
(15) [+high, -back]CX:
Assign one violation mark for every CX sequence where C and X do not agree in [+high, -back] (X=G or V)
(16) IDENT[anterior]ONSET:
Assign one violation mark for every input-output disparity of the [anterior] feature in the onset position.
(17) IDENT[aspirated]:
Assign one violation mark for every input-output disparity of the [aspirated] feature. (18) DEP-S:
Assign one violation for each segment in the output that does not appear in the input.
[+high, -back]CX is proposed above the cut-off line to eliminate ill-formedness. The constraint eliminates the disparity between CX in the [+high, -back] feature regardless of the left-right or right-left directions. This constraint is borrowed from the idea of the locus of violation (McCarthy, 2003; Hsiao, 2015), which is related to the OCP (obligatory contour principle) effect. Unlike the idea originally purposed to ban adjacent identical elements, [+high, -back]CX is proposed to allow an element to follow the other adjacent
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
element. Regarding the constraint, thus, a [+high, -back] consonant such as an alveolar-palatal affricate is obligatory adjacent to a [+high, -back] vocoid such as /i(j), y(ɥ)/.
IDENT[anterior]ONSET is also placed above the cut-off line to eliminate ill-formedness, and it is neither mapped to glide insertion nor to a retroflex counterpart. Also, IDENT[aspirated] and DEP-S are positioned below the cut-off line since the aspiration contrast and segment insertion are not fatal in the adaptation but induce variations. A syllable /tɕa/ is given as an example in (19).
(19) /tɕa.k’u/ → [tɕhja.ku] (恰古) ‘incessantly’
Input: tɕa [+high, -back]CX ID[anterior]ONSET *[retroflex] DEP-S
☞1 a. tɕja *
☞1 b. tɕhja *
☞2 c. tʂa *
d. tsa *!
e. tɕa *!
f. tʂja *! * *
In tableau (19), the input /tɕa/ is mapped to three variations: candidates (19a-b), in which glide insertion arises but differ in whether the output is aspirated or not, and candidate (19c), which is a retroflexed post-alveolar counterpart. IDENT[anterior]ONSET functions to eliminate the dental [tsa] in (19d). [+high, -back]CX rules out both of (19e) and (19f) due to the disparity of the [+high, -back] feature in CX (X=glide or vowel).
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
IDENT[anterior]CODA dominates DEP-S below the cut-off line with a solid line since glide insertion is the most frequent output (63.45%), and retroflex is the second-most frequent variant (28.43 %).
In the case of glide insertion before the either vowel /o/ or /e/, markedness constraints are required due to the fact that Mandarin phonotactics does not allow the single nucleus /o/ or /e/ in a syllable. Thus, the markedness constraints *C[o], *C[e] are proposed here.
(20) *C[o]: Assign one violation mark for every syllable with the single nucleus /o/.
(21) *C[e]: Assign one violation mark for every syllable with the single nucleus /e/.
Since the violation of *C[o] and *C[e] is fatal in the KM Kong-er adaptation, the constraints are placed above the cut-off line. A syllable /tɕe/ is given as an example for the illustration in (22). Since constraints are not ranked with a dotted line, the exclamation mark in parenthesis indicates that either violation is fatal.
(22) /tɕe.pal/ → [tɕje.pa] (皆巴) ‘please’
Input: tɕe *C[e] [+high, -back]CX ID [anterior]ONSET *[retroflex] DEP-S
☞1 a. tɕje *
☞1 b. tɕhje *
c. tʂe *! *
d. tɕe *! *(!)
e. tse *! *(!)
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
In tableau (22), candidates (22a-b) are selected as optimal variations and the retroflexed candidate (22c) is eliminated by the constraint *C[e] in this case. A possible candidate [tʂei] is omitted in this tableau since [tʂei] does not appear in Mandarin, and it can be simply eliminated by a phonotactic constraint like *tʂei.
Lastly, glide insertion also appears before the vowel /ʌ/. Also, the retroflexed counterpart is the second-most frequent candidate. Related to the vowel /ʌ/, diphthongization mapping [ʌ]-to-[ou] is analyzed in the next section.1 Since [ʌ] is not allowed in Mandarin phonotactics, a related constraint is proposed in (23).
(23) *[ʌ]: Assign one violation mark for every output [ʌ].
*[ʌ] is placed above the cut-off line to rule out any candidates with the illicit vowel [ʌ].
A syllable /tɕʰʌ/ is given as an example in (24).
1 The Korean vowels /ʌ/ and /ɯ/ do not have corresponding vowels in Mandarin, which leads to the employment of other strategies, such as vowel lowering and vowel diphthongization, the use of which are described in the following sections. In this subsection, we will focus on glide insertion.
‧
*C[o] also successfully defeats the candidates (24a-b) since they have single nucleus [o]
in the output. As a result, the outputs with glide insertion [tɕʰjou]/[tɕjou] (24c-d) are selected as the most optimal variants, and the outputs with retroflexation [tʂʰou]/[tʂou]
(24e-f) become secondary optimal outputs. To disallow [ʌ]-to-[e] mapping from appearing, IDENT-V[+back] is used and prevents the candidate [tɕje] (24j).
(25) IDENT-V[back]:
Assign one violation mark for every input-output disparity of the [back] feature.
‧
IDENT-V[+back] is also used in the later section related to the vowel adaptations.
4.1.2 Vowel Adaptation 4.1.2.1 Vowel Lowering
The Korean high back vowel /ɯ/ does not exist in the Mandarin phonetic system.
The strategy that Mandarin speakers mainly use for adapting the vowel is lowering (80.30%). They also use rounding (17.84%) for variation. IDENT-V[back], previously proposed, cannot rule out any candidate above the cut-off line since both lowering and rounding arise with the [back] vowels /ɤ/ and /u/. Therefore, OT analysis requires new constraints: *[ɯ], IDENT-V[round], IDENT-V[-low] and IDENT-V[high] as shown below.
(26) *[ɯ]: Assign one violation mark for every output including the vowel [ɯ].
(27) IDENT-V[round]:
Assign one violation mark for every input-output disparity of the [round] feature in a vowel.
(28) IDENT-V[-low]:
Assign one violation mark for every input-output disparity of the [-low] feature in a vowel.
(29) IDENT-V[high]:
Assign one violation mark for every input-output disparity of the [high] feature in a vowel.
The constraint *[ɯ] is proposed to eliminate the faithful output [ɯ], which is absent in Mandarin phonotactics. IDENT-V[-low] is proposed to rule out the mapping [ɯ]-to-[a].
Lastly, IDENT-V[round] and IDENT-V[high] preserve the variation below the cut-off line.
A syllable /kɯ/ is analyzed in tableau (30).
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(30) /kɯ.kʌ/ → [khɤ.kou] (可勾) ‘that one’
Input: kɯ *[ɯ] *C[o] IDENT-V[-low] IDENT-V[round] IDENT-V[high]
☞1 a. kɤ *
☞1 b. khɤ *
☞2 c. ku *
☞2 d. khu *
e. ko *! *
f. ka *! *
g. kɯ *!
In tableau (30), the faithful output (30g) is eliminated by the Mandarin phonotactics constraint *[ɯ]. IDENT-V[back] above the cut-off line (omitted in the tableau) ensures adjustment for [ɤ] (31a-b) and [u] (30c-d). IDENT-V[-low] eliminates [kɯ]-to-[ka]
mapping in (30f). Since lowering the output [ɤ] is more frequent than rounding the output [u] in the corpus, IDENT-V[round] dominates IDENT-V[high], inducing the candidate ranking. (30a-b) are optimally selected and (30c-d) are selected as a variation in the adaptation.
4.1.2.2 Diphthongization
Diphthongization is correlated with the Korean vowels /o, e, ʌ/. Since /o/ and /e/
cannot occur as a single nucleus in Mandarin, diphthongization along with either an on-glide or an off-on-glide is the main strategy for these two source vowels in adaptation. Also,
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
neither /o/ nor /e/ are ever deleted in the process of adaptation, and thus, the constraints related to the result are introduced as in (31)-(33).
(31) IDENT-V[-high]:
Assign one violation mark for every input-output disparity of the [-high] feature in vowels.
(32) MAX[o]:
Assign one violation mark for every vowel /o/ in the input that does not have a correspondence in the output.
(33) MAX[e]:
Assign one violation mark for every vowel /e/ in the input that does not have a correspondence in the output.
IDENT-V[-high] functions to preserve mid vowels as well as IDENT-V[-low] works vowels. Especially for the three vowels which follow the pattern of diphthongization, these two constraints rule out both [o/e/ʌ]-to-[u] and [o/e/ʌ]-to-[a]. On the other hand, MAX[o] and MAX[e] can be proposed since both vowels are retained in the result (100%).
A syllable /mo/ is taken as an example in (34). Only the diphthongized output [mou] and [mwo] are chosen.
‧
Previously proposed constraints successfully eliminate ill-formed candidates (34c-g). Since the ranking of the constraints of IDENT[round] and DEP-Sdoes not conflict or induce frequent ranking, no domination relation is constructed in tableau (34).
Next, let us examine the Korean vowel /e/. In KM Kong-er adaptation, it is possibly mapped to both /je/ and /ei/ (100%). Thus, MAX[e] is posited above the cut-off line to prevent the deletion of vowel /e/ and to eliminate any mapping without [e]. An example is illustrated in (35).
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(35) /o.re/ → [ou.lje] (偶勒) ‘for a long time’
Input: re *C[e] IDENT-
V[-low] IDENT-
V[-high] MAX[o] I
DENT-V[round] DEP-S
☞1 a. lei *
☞1 b. lje *
c. le *! *(!)
d. la *! *(!)
e. lɤ *!
f. lu *! *(!) *
In the tableau (35), the input /re/ is possibly mapped to both [lei] (35a) and [lje] (35b).
The faithful candidate [le] (35c) is ruled out by *C[e], and the other candidates [la], [lɤ]
and [u] (35d-f) are also ruled out by previously proposed constraints.
Lastly, the vowel /ʌ/ is diphthongized to [ou/wo] in most part of the whole (77.45%). Also, some tokens also adapt [ɤ] (22.55%). To produce the result, IDENT-V [-high] is activated. The input /kʌl/ is chosen as an example for the illustration in (36).
‧
(36) /mo.tɯn.kʌl/ → [mou.tɤŋ.kou] (某等勾) ‘everything-OBJ’ Input: kʌl *[ʌ] *C[o] IDENT-
In tableau (36), the faithful candidate [kʌ] (36e) is ruled out by *[ʌ]. IDENT-V[-low] and IDENT-V[-high] ensure that the source vowel /ʌ/ to undergoes a process of diphthongization to become [kou] (36a) and [kwo] (36b), but it is not mapped to either [ka] (36f) or [ku] (36g) at the same time. Also, the constraint ranking indicates that [kʌ]-to-[kɤ] (36c) is possible mapping.
Even though the tableau successfully demonstrates that neither diphthongized output (36a-b) nor lowered output (36c) are fatal, the frequency based on the ranking below the cut-off line is not identical with the actual result. According to the tableau, (36c) must be the most optimal and frequent output, and (36a-b) should be the second most frequent variation. McCarthy (2008) once noted that OT grammar emphasizes the possible outputs instead of the actual outputs. Thus, we conclude that output such as [kɤ]
in (36c) is predicted to be the most optimal output if more data is collected.
‧
The constructed constraint ranking of the adaptations conforming to the Mandarin phonotactics is presented in (37).
(37) Ranking of Adaptation Conforming to Mandarin Phonology
MAX-SONSET,
*TENSE(C), [IDENT[tense]&*[+aspirated]], CODA CONDITION,
IDENT[anterior]CODA, *[retroflex]
|
IDENT-V[round], DEP-S
|
IDENT-V[high], IDENT[aspirated]
Above the cut-off line, the constraints are not ranked since they do not conflict with each other. They only rule out ill-formed candidates. *C[o], *C[e], CODA CONDITION, [+high, -back]CX are proposed to eliminate a candidate in violation of Mandarin phonotactics. *[ʌ], *[ɯ], *TENSE(C) function to rule out faithful candidates which Mandarin phonotactics does not have. MAX-SONSET, MAX[o], MAX[e] are proposed since Kong-er adaptation does not allow onset deletion and or vowel [o]/[e] deletion. Lastly,
‧
[IDENT[tense]&*[+aspirated]], IDENT[anterior]ONSET, IDENT-V[back], IDENT-V[-low]
and IDENT-V[-high] functions to rule out wrongly mapped outputs.
Below the cut-off line, the constraints are hierarchized as IDENT[nasal]CODA >>
MAX-SCODA >> IDENT[anterior]CODA for coda adaptation, which is successful in inducing adaptation result according to the coda types. Also, *[retroflex] ensures retroflexed counterpart in the tense de-tensing and the glide insertion. Under the constraints, I DENT-V[round], DEP-S are placed for the vowel adaptation, to allow [ɯ]-to-[u] adaptation and diphthongization. IDENT-V[high] is dominated at the bottom to ensure the result that vowel lowering [ɯ]-to-[ɤ] is more frequent than vowel rounding [ɯ]-to-[u]. Lastly, IDENT[aspirated] is also dominated at the bottom since KM Kong-er does not allow for aspiration contrast.