CHAPTER 2 : Literature Review
2.2 Optimality Theory
2.2.1 Local Conjunction of Constraints
2.2.1 Local Conjunction of Constraint
A mechanism of complex constraint, called the ‘local conjunction of constraints’
(Smolensky, 1993/1995), is used in this thesis. A locally conjoined constraint C1 and C2
within a local domain of D ([C1&C2]D) is violated when both of its conjuncts are violated.
Universally, the conjunct constraints C1 & C2 is higher-ranked than each of C1 and C2. The definition of local conjunction and the ranking of constraints is given in (4).
(4) Local Conjunction (Smolensky 1995: 4)
a. Local Conjunction of C1 and C2 in domain D, C1 &l C2, is violated when there is some domain of type D in which both C1 and C2 are violated.
b. Universally, C1 &l C2 >> C1, C2
In the case of the conjunction of a markedness constraint and a faithfulness constraint, the markedness constraint, which is low-ranked and inactive, is high-ranked and activated when the faithfulness constraint is violated (Łubowicz, 2002; Hsiao, 2015).
In this thesis, the markedness constraint *[+aspirated] is activated when the faithfulness constraint IDENT[tense] is violated.
The local conjunction of constraints allows for a wide range of complex phenomena from the Coda Condition (Smolensky 1995), to dissimilation and the Obligatory Contour Principle (Alderete 1997, Itô and Mester 1998), chain shift (Kirchner 1996, Moreton and Smolensky 2002), vowel harmony (Baković 1999) and opacity (Itô and Mester 1998, Moreton and Smolensky 2002). This thesis applies the mechanism to permit complex phenomena related to tensed obstruents and their unaspirated counterparts in chapter 4.
‧
2.2.2 Rank-ordering Model of EVAL (ROE)
The Rank-ordering Model of EVAL (ROE) is a framework under the OT proposed by Coetzee (2006). In this framework, language phenomena are more than categorical, which producing various variations of one linguistic form. Otherwise, classical OT cannot explain the variation appearing in real linguistic phenomena since the constraint ranking only selects the most optimal candidate as an output without a consideration of other possibilities. Instead of disregarding the non-optimal candidates, the ROE discusses the candidates which appear in reality, and treat them as part of a well-formed phonology.
As classical OT, all of the candidates in the ROE are evaluated by EVAL, and a ranking hierarchy within EVAL decides the well-formedness of the candidate set. The more well-formed candidate is the more high-ranked as decided by the ranking hierarchy, and it is more likely to be selected as the optimal output. It also occurs at a higher frequency in the language than the less well-formed candidates.
Since the well-formed candidates (variation) cannot be infinitely chosen, the criterion called ‘critical cut-off’ is used. In the ROE, the critical cut-off line separates the constraint set into two strata: constraints ranked above the line and below the line.
Constraints ranked above the line function just as a constraint set does in the classical OT, which links to the most optimal and grammatical output among the candidates without any variations. However, constraints ranked below the line are different. Unlike as in classical OT, candidates cannot be ruled out by the constraints ranked below the cut-off line, and can even be considered to be a grammatical and a possible output. In this way, variations arise in the ROE.
‧
constraints MAX(segment) and *σ̃ /{i,u} are placed below the cut-off line to ensure the variation (5ab). Since there is solid line between MAX(segment) and *σ̃ /{i,u}, a domination relation exists. The candidate (5a) which violates the constraint on the right side (*σ̃/{i,u}) is the first optimal candidate. Also, it occurs more frequently than the second optimal candidate (5b) does. The two constraints DEP[-high] and [+back]eliminate the ungrammatical candidates (5c-d) above the cut-off line.
The cut-off line can induce three possible situations as shown in (6). As in the first scenario in (6a), candidate 1 violates C3, and candidate 2 violates C4. Since C3 and C4 are below the cut-off line, both candidate 1 and candidate 2 are selected as variations of the input. The dominant constraint below the cut-off line decides more optimal/frequent candidate, relatively.
Another possible situation is shown in (6b), where only the best optimal candidate is chosen without any appearance of variation. The cand2 and can3 are both disfavored by the constraints above the cut-off line, and so are deemed ungrammatical.
The last situation in which a cut-off line is induced is shown in (6c), where no variants appear as output either. Since both cand1 and candidate 2 in (6c) are violating
‧
constraints above the cut-off line (C2, C1), the ROE functions exactly as in the classical OT.
(6) Three possible scenarios regarding the critical cut-off line (Coetzee, 2006) (6a) Variation
In KM Kong-er adaptation, some input-output mappings induce variations in the forms and some do not. Thus, the ROE is applied in the KM Kong-er analysis to reflect actual result of the adaptation.
‧
This research builds a corpus of Kong-er lyrics containing 955 phonological words (2690 syllables) from 11 songs. The phonological words in the corpus are constructed out of 57 tokens of monosyllable words, 328 of disyllable words, 370 of tri-syllable words, and 200 of polysyllable words (maximum seven syllables). The Korean writing system stipulates that every word is separated by a space except for function words which are attached to the preceding lexical morpheme as in the example (1) shown below.
(1) 수지-가 저녁-에 사과-를 먹었어요.
Suji-SUBJ evening-at apple-OBJ ate.
‘Suji ate (an) apple in the evening.’
Thus, this research chose the phonological words as a domain of analysis for two reasons. First, the separation based on the Korean writing rules visually influences Mandarin speakers who create KM Kong-er to perceive a phonological word as a unit.
Second, the K-pop songs that this corpus collected are auditorily punctuated to the p-word level to deliver clarity in the meaning. Thus, phonological words influence both the visual and auditory perception of Mandarin speakers.
All of the Kong-er lyrics were collected from four K-pop Kong-er websites1 managed by Mandarin speakers in Taiwan. The managers of the sites are all female, with an age of between 17-27, and with a mid-to-advanced level of Korean (2-3 years of
‧
studying). In the interviews, which were independently conducted, the Mandarin speakers stated that they relied on both visual inputs from the written source and auditory perception of the songs during the adaptation.
There are two kinds of adaptations in the corpus: one where the data conform to Mandarin phonotactics (L1) and the other where the data violate Mandarin phonotactics by using other transcription such as Zhuyin (the Chinese transliteration system for Taiwan Mandarin) and English. For the adaptations conforming to Mandarin phonotactics, five phonological patterns are observed and discussed namely obstruent de-tensing, coda deletion, glide insertion, lowering and diphthongization. For the adaptation violating Mandarin phonotactics, two types of phonotactic violation are observed: one that follows Korean grammar and the other that follows neither Mandarin grammar nor Korean grammar, which is called interlanguage grammar.
A simplified schema of KM Kong-er adaptation is illustrated in figure 1. Mandarin grammar plays the most important role in KM Kong-er, and two types of violations occupy a minor part of whole, but produce a pronunciation closer to the Korean source.
Figure 1. Three types of grammar adopted in KM Kong-er Korean Grammar
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
3.1 Adaptations in Conformity to Mandarin 3.1.1 Consonant Adaptation
Three phonological patterns for the consonants are observed in the Kong-er corpus, which are correlated with obstruent de-tensing in the onset position, coda deletion, and glide insertion2.
3.1.1.1 Obstruent De-tensing
The Korean consonant has an abundant inventory of obstruents. There is a three-way contrast for the obstruents: lax (plain), tense and aspiratedsounds. Examples are /p, pʰ, p’/, /t, tʰ, t’/, /k, kʰ, k’/ in stops, /tɕ, tɕʰ, tɕ’/ in affricates and /s, s’/3 in fricatives.
Mandarin does not include [tense] contrast, but does includes [spread glottis] contrast.
Examples are /p, ph/, /t, th/, /k, kh/ in stops, /tɕ, tɕʰ/ and /ts, tsh/ in affricates.
In KM Kong-er adaptation, Korean tensed obstruents are de-tensed to [-spread glottis] counterparts regardless of the way of transcription. For example, the obstruent tense stop /p’/ in /ɯn.p’it/ ‘silveriness’ is adjusted to a de-tensed unaspirated counterpart [p] as in 恩比 [ən.pi]. Other examples related to obstruent de-tensing are shown in table (2) below.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Overall, 87.97% (139/158) of the Korean tensed stops /p’, t’, k’/ in the corpus are replaced by the Mandarin unaspirated [p, t, k] as in (2a-c), and the rest of the Korean tensed stops (12.03%) are replaced by their unaspirated counterparts in the Zhuyin and English forms. Likewise, the Korean tensed affricate /tɕ’/ and the tensed fricative /s’/ are dominantly replaced by their unaspirated counterparts in Mandarin (2d-e), in 100%
(45/45) for the tokens of the affricate [tɕ] and 96.49% (55/57) of the tokens of the fricative [s]. Only two tokens of Korean tensed fricatives are adapted into Zhuyin/English transcription. To sum up, every token with a tensed consonant (260/260) adopts an unaspirated counterpart both in the Mandarin form or in the Zhuyin/English form. The statistics of obstruent de-tensing are shown in (3). The most dominant results from each category are shadowed in the table.
4 The transcription method for Mandarin-IPA is referred in Lin, Y. H. (2007).
(2) Obsturent De-tensing
Korean source Kong-er in Mandarin Gloss
a. nun.p’it ɹun.pi4 (潤筆) ‘the expression of the eyes.’
b. t’o tou (都) ‘again’
c. k’ot kou (勾) ‘flower’
d. ʌ.tɕ’ʌ.na ou.tɕjou.na (偶九那) ‘Oh dear! (exclamation)’
e. jip.s’u.ri ji.su.li (以速李) ‘lip-PAR’
‧
(3) Statistics of Obsturent De-tensing Korean tensed
consonant Kong-er in Mandarin substitution Number Total Percentage
stop [p’, t’, k’]
fricative [s’] unaspirated
[s] 55
unaspirated counterpart 260 100%
As in table (3), 260 syllables are tensed obstruents in the corpus. 260 tokens are de-tensed and replaced by an unaspirated counterpart. It is noticeable that only an unaspirated counterpart is chosen to be adapted when the Mandarin phonotactics also has an aspirated counterpart.
There is a different view on the traditional three-phoneme contrast in Korean, which is related to the fact that a tensed-obstruent only maps to an unaspirated phoneme (Kingston and Diehl, 1994; Kim and Duanmu, 2004). They argue that the feature [+tense]
is not needed for the Korean phonology system because the feature [voiced] and the [aspirated]/[spread glottis] inherent in the Korean phonology system can decide their contrasts based on tonogenesis theory. Based on their assertion, it is not necessary to
‧
import the markedness feature [+tense]/[+constricted glottis] into the Korean phonological inventory system, as shown in the table (4).
(4) Comparison of the three-phoneme contrast in Korean Example Traditional view Kim & Duanmu (2004)
p’ tensed voiceless unaspirated p lax/unaspirated voiced unaspirated ph aspirated voiceless aspirated
According to Kim & Duanmu (2004), Korean tensed obstruents can be described with [-voice] and [-aspirated] features. Following such classification, the tensed obstruents in the corpus are 100% mapped to the unaspirated counterpart.
3.1.1.2 Coda Deletion
The Korean phonological inventory allows seven consonants to occur in a coda position, including the three stops /p, t, k/, the three nasals /m, n, ŋ/, and one liquid /l/.
However, the Mandarin phonological inventory only allows two nasals in a coda position:
/n/ and /ŋ/. In KM Kong-er adaptation, Mandarin speakers can use three strategies on Korean seven codas: deletion, replacement and retainment.
First, the dominant strategy is coda deletion (422/866, 48.73%). Overall, the deletion occurs regardless of the types of Korean coda (stop, nasal, or liquid). It occurs in different percentages depending on the coda types. The type of coda of which the greatest number of the deletions occur is that of the stops /p, t, k/ (99.22%). The type of which the second largest amount of the deletions occur is that of the liquid /l/ (97.13%). The type
‧
of which the least number of deletions occur is that of the nasal coda /m, n, ŋ/ (17.39%).
Examples of coda deletion are shown in (5).
(5) Coda deletion
Korean source Kong-er in Mandarin Gloss
a. pʰap.kʰon pʰa.kʰoŋ (怕空) ‘popcorn’
(6) Statistics of coda deletion
Korean coda group Number Total Percentage
stop Vp, Vt, Vk 127 128 99.22%
liquid Vl 203 209 97.13%
nasal Vm, Vn, Vŋ 92 529 17.39%
Total 422 866 48.73%
As in (6), 99.22% of the stop codas are deleted, and 97.13% of liquid coda is also deleted.
‧
coda adaptation. The coda type which is subject to the most replacements is that of the nasal type /m, n, ŋ/ (174/529, 32.89%). In contrast, only six tokens in liquid /l/, and one token in the stops /p, t, k/ are replaced by the Mandarin nasal codas /n, ŋ/. Examples of coda replacement are shown in (7).(7) Coda replacement
Korean source Kong-er in Mandarin Gloss
a. mo.tɯn mwo.təŋ (摸登) ‘every’
b. tɕuŋ.tok zun.tou (尊斗) ‘addiction’
c. pa.ram.man pa.laŋ.man (巴朗曼) ‘only wind’
d. ma.mɯl ma.mən (馬悶) ‘heart-OBJ’
In table (7), the nasal coda /n/ is replaced by the other nasal coda /ŋ/ (7a), and the nasal coda /ŋ/ is replaced by nasal coda /n/ (7b). The Korean nasal coda /m/ and the liquid coda /l/ are also replaced by the Mandarin nasal codas /n, ŋ/ (7cd). The statistics of the coda replacement are shown in (8).
(8) Statistics of coda replacement
Korean coda group Number Total Percentage
obsturent Vp, Vt, Vk 1 128 0.78%
liquid Vl 6 209 2.87%
nasal Vm, Vn, Vŋ 174 529 32.89%
Total 181 866 20.90%
The last strategy used in coda adaptation is retainment (263/866, 30.37%). The retainment only occurs in the case of the codas /n, ŋ/, which are legal codas in both of the
‧
regularities. The overall statistics of coda adaptation are presented in table (9).(9) Statistics of coda adaptations Korean
coda structure
Kong-er in
Mandarin substitution Number Total Percentage
stop Vp Vt Vk deletion 127
replacement 100 22.88%
Total
deletion 422
866
48.73%
replacement 181 20.90%
retainment 263 30.37%
These results demonstrate that the phonetic features of a coda have largely influenced the process of Kong-er adaptation, especially in the case of coda deletion and replacement. Unlike the stop codas /p, t, k/ that are mostly deleted (127/128, 99.22%), a large amount of /m/ codas are turned into other nasal codas, either [n] or [ŋ] (74/92, 80.43%). Also, 2.87% of the tokens of the Korean liquid coda /l/ are replaced by Mandarin nasals.
In this regard, ‘sonority’ can explain the result. Sonority is one way for sound to be
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
an articulation. The more sonorant the sound is, the higher the tendency there is for it to be perceived as voiced. Based on the investigation of Yavas and Gogate (1999), the phonemic awareness of coda consonants is deeply related to the sonority relations in the language acquisition of children. They also found that the perception and segmentation of less sonorant codas are easier than that of more sonorant ones. According to the general patterns of the sonority scale shown in (10), Mandarin speakers follow strategies in the adaptation of Korean codas simply to retain nasality, and/or sonority.
(10) Sonority Scale
Vowels > Liquids > Nasals > Fricatives > Stops
Overall, Mandarin speakers either delete illicit codas in Mandarin, or replace illicit /m/ codas with Mandarin codas /n, ŋ/, following Mandarin phonotactics (L1).
3.1.1.3 Glide Insertion
In the Kong-er corpus, there are 328 tokens of Korean alveolo-palatal affricates (/tɕ, tɕh, tɕ’/) in the onset position5. The glide /j/ was inserted in 125 out of 197 tokens (63.45%) during Kong-er adaptation. For instance, ‘tɕa’ in /nun.toŋ.tɕa/ ‘the pupil of the eye’ is substituted with ‘tɕja’ as in [nu.toŋ.tɕja] (努東加). Other examples of glide insertion are shown in (11) below.
5 131 syllables from 328 syllables, where the vowel /i/ is followed after alveolo-palatal affricates, are excluded in the sum total of the number of glide insertion.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
(11) Glide Insertion for the vowels /a, ʌ, e/
Korean source Kong-er in Mandarin Gloss
a. tɕa.k’u tɕʰja.ku (恰古) ‘repeatedly’
b. tɕʰam tɕʰjaŋ (嗆) ‘truly’
c. ku.rɯm.tɕʰʌ.rʌm khu.ləŋ.tɕhjou.loŋ (苦冷秋龍) ‘as clouds’
d. tɕe.pal tɕje.pa (皆巴) ‘please’
e. tɕʰe tɕʰje (切) ‘just as it is’
In (11), the glide /j/ is inserted between the alveolar affricates /tɕ, tɕh/ and the vowels /a, ʌ, e/. /tɕa/ and /tɕʰa/ in (11ab) is replaced by [tɕʰja], implying that contrast of the [spread glottis] is not crucial in the adaptation. Glide insertion before the vowel /a/
accounts for the major part of the results (78.57%, 55/70). In (11c), the glide /j/ is inserted between alveolar affricates and the Korean vowel /ʌ/ (68.33%, 41/60) and mapped to [tɕhjou]. Lastly, the glide /j/ in (11d-e) is inserted before the Korean vowel /e/ (84.21%, 16/19), inducing [tɕje] and [tɕʰje].
However, unlike case of the glide insertion, 56 tokens (28.43%) of the Korean alveolo-palatal affricates are adjusted into Mandarin retroflex/post-alveolar affricates [tʂ, tʂh]. For example, ‘tɕo’ from /tɕo.a/ ‘to like’ is replaced by ‘tʂou’ as in [tʂou.a] (周阿).
The result is categorized by the onset and rhyme sequence as shown in table (12).
‧
(12) Statistics of glide insertion Korean affricate
onset + vowel Kong-er in Mandarin substitution Number Total Percentage
_a(C)6 _ja(C) 55 categorized as ‘others’. Some are replaced by dental counterparts such as [ts] and [tsh], and others are different in vowel from the source. Such tokens are irregular and scarce in the corpus; therefore, they are not discussed in this thesis.
6 (C) here denotes the possibility of a coda.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
The sequences, including the vowels /a/, /e/ and /ʌ/, take glide insertion as a dominant strategy. On the other hand, the sequences, including the vowels /o/, /u/ and /ɯ/, adopt other strategies, such as substitution into a retroflex (/tɕo/ → [tʂou]).
In the Korean phonology/phonetic system, the alveolo-palatal affricates [tɕ, tɕh, tɕ’]
cannot appear with a following glide /j/ (Sin et al., 2012). In contrast, in the Mandarin system, an alveolo-palatal affricate occurs only before the glides /j/, /ɥ/ (Lin, 2007). Thus, glide insertion in Kong-er indicates that Mandarin speakers follow the L1 phonology rather than follow the L2 phonology.
3.1.2 Vowel Adaptation
Two phonological patterns of the vowels are observed in the Kong-er corpus, which are related to lowering and diphthongization.
3.1.2.1 Lowering
A dominant strategy for the Korean high back vowel /ɯ/ is lowering. Since the high back vowel /ɯ/ does not exist in Mandarin, Mandarin speakers adjust every token of /ɯ/
to other vowels existing in the phonological system. The majority of the tokens of the /ɯ/
vowel are lowered (216/269, 80.30%), either into the mid back vowel [ɤ] or into the mid central vowel [ə]7. For example, ‘kɯ’ from /kɯ.kʌn.man/ ‘only for that’ is replaced by
‘kɤ’ as in [kɤ.kou.man] (哥勾曼). Other examples are shown in (13).
‧
Korean source Kong-er in Mandarin Gloss
a. kɯ.njaŋ kɤ.njɑŋ (可娘) ‘just’
b. kʌ.rɯm khou.lɤŋ (叩冷) ‘step’
c. na.mɯn na.mən (哪悶) ‘left over’
The rest of the tokens of the /ɯ/ are rounded (48/269, 17.84%) to [u]. For example,
‘tɯ’ in /pu.tɯ.rʌ.un/ ‘soft’ is replaced by ‘tu’ as in [pu.tu.lou.wən] (不督嘍溫). Table (14) shows the overall statistics related to the high back vowel /ɯ/.
(14) Statistics of lowering Korean
combination
Kong-er in
Mandarin substitution Number Total Percentage
(C)ɯ(C)
(C)ɤ/ə(C) 216
269
80.30%
(C)u(C) 48 17.84%
Others (ɨ syllabic consonant) 5 1.86%
As shown in (14), lowering to the vowel [ɤ/ə] is the most selected strategy for the Korean /ɯ/ vowel (80.30%). Rounding to [u] is the second most used strategy (17.84%).
Both [ɤ/ə] and [u] share the [+back] feature with the source /ɯ/ vowel. This implies that Mandarin speakers decide to preserve the [+back] feature of the source vowel, choosing vowels adjacent to /ɯ/ in figure 2. Again, Mandarin speakers follow the L1 phonology.
‧
Figure 2. Back vowel chart of Korean (underlined) and Mandarin (bold)3.1.2.2 Diphthongization
Three monophthongs /o, ʌ, e/ are discussed in this session. The vowels /o, e/ cannot occur as a single nucleus in Mandarin and /ʌ/ does not exist in the Mandarin phonetic inventory. For these three vowels, diphthongization is chosen to be the dominant strategy in the case of these three vowels during Korean adaptation. In the KM Kong-er corpus, there are 898 syllables that contain monophthongs /o, ʌ, e/ in total, of which 88.86% were diphthongized to either [ou] or [ei]. For example, ‘kʌ’ and ‘to’ in /tɕu.kʌ.to/ ‘to die’ is replaced by ‘kou’ and ‘tou’ respectively as in [tɕjou.kou.tou] (揪勾兜). Other examples of each vowel are presented in (15).
(15) Diphthongization
Korean source Kong-er in Mandarin Gloss
a. so.ri sou.li (搜李) ‘sound’
b. ʌ.tɕe ou.tɕje (歐皆) ‘yesterday’
c. tu.so.ne tu.song.nei (督宋內) ‘in two hands’
As in (15a-b), /o/ and /ʌ/ are prone to be diphthongized to [ou], and /e/ is replaced by [ei] as in (15c). For the source vowel /ʌ/, other variants such as the lowered vowels [ɤ,
ɯ u ɤ o ʌ
ɑ
‧
in Mandarin substitution Number Total Percentage
(C)o(C) _ou/wo(C) 277 second most dominant strategy (10.24%). Some variants such as glide insertion account for 0.89% of the whole. Also, the vowel /e/ in input is never deleted in the output, inducing variation in either the on-glide [je] or off-glide [ei] insertion. Vowel diphthongization implies the importance of the L1 (Mandarin phonology) again in KM Kong-er adaptation.
3.2 Adaptations in Violation of Mandarin Phonotactics
The data shows that some syllables (168/2690) are written in other types of transcription such as Zhuyin, which is the alphabet system for Mandarin in Taiwan, and
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
English. Among such tokens, two types of grammar are observed: one following Korean grammar (L2), and the other an emerging interlanguage grammar.
3.2.1 Adaptations in Conformity to Korean Grammar
60.7% (102/168) of the Zhuyin/English tokens belong to the adaptations following Korean phonology. The specific sequences from the tokens are grouped into velars, labials, dental consonants, and some vowels, such as /i, ʌ, o/, in this section.
3.2.1.1 Velars /k, kh, k’, h/ + Vowel /i/
In Mandarin phonotactics, velar consonants such as /k/, /kh/ and /h(x)8/ are not
In Mandarin phonotactics, velar consonants such as /k/, /kh/ and /h(x)8/ are not