In addition to the statistic results of our data, the relative hierarchy of a constraint in the ranking is also determined on the ground of relevant acoustic or contextual cues

Chapter 4

English Loanwords in Mandarin: OT Analysis

In this chapter, the interpretation patterns of English codas are projected to the

interaction of constraints in OT. The conformity and adaptation in loanword

phonology are attributed to constraint rankings. In addition to the statistic results of

our data, the relative hierarchy of a constraint in the ranking is also determined on the

ground of relevant acoustic or contextual cues.

4.1 Sound Inventories

Loanword phonology is studied to explain the sound changes of the word from

LS to LT. The native phonotactics leads to the markedness constraints which regulate

the well-formedness of an output in LT. The sound inventories of English and

Mandarin are listed in Tables (27) to (30).

Table (27) English Vowels (based on Heinz J 1992) i  u 

e      o  æ 

Table (28) Mandarin Vowels (based on Cheng 1973¹) i y  u

e    o a

English has three diphthongs, [a, a, ], and Mandarin has four, [ai, ei, au, ou].

As seen in the tables above, English and Mandarin differ in that English has the

length/tenseness contrast in high and mid vowels, while Mandarin has the rounding

contrast in non-low vowels. Hence vowel change is inevitable in the word-loaning

process from English to Mandarin.

Table (29) English Consonants (based on Heinz J 1992)

bilabial labio-dental inter-dental alveolar palatal velar glottal

+vcd b d 

stop

-vcd p t k

+vcd v  z  h

fricative

-vcd f  s 

-vcd t

affricate

+vcd d

nasal +vcd m n 

liquid/

retroflex

+vcd l/r

glide +vcd j w

1 The vowel chart and the upcoming consonant chart (29) are slightly modified from Cheng (1973).

Table (30) Mandarin Consonants (based on Cheng 1973)

bilabial labio-dental alveolar alveopalatal palatal retroflex velar +vcd

-asp p t k

stop

-vcd

+asp p t k

+vcd 

fricative

-vcd f s   x

-asp ts t t

affricate -vcd

+asp ts t t

nasal +vcd m n ²

liquid +vcd l

glide +vcd j w

Comparing the two consonant patterns, we single out a significant difference: the

contrast of voicing in English and that of aspiration in Mandarin obstruents. [] is the

only allowed voiced obstruent in Mandarin. For Mandarin speakers it is difficult to

distinguish an English voiced stop and a Mandarin unaspirated stop, which well

illustrates that in our data an aspirated stop in English is usually mapped to the same

aspirated stop in Mandarin, as in ‘[pa] (pie) → [pa] (派)’, and a voiced stop is

modified into an unaspirated stop in Mandarin, as in ‘[bs] (bus) → [pa.] (巴士)’.

Aspiration of voiceless obstruents are not viewed as the underlying contrast in the

consonant inventory of English, but as contextual allophones; an English voiceless

stop is aspirated when it is the onset of a stressed syllable, and is unaspirated else

where.

2 [] only occur in the coda position.

4.2 Basic Constraints

In (31-32), we introduce two markedness constraints on Mandarin syllable

structures, in particular, relevant to the restrictions on the coda position.

(31) *COMPLEXCODA: No consonant cluster is allowed in coda position.

(32) CODACONDITION: Syllables must have no coda, except that the coda is an

alveolar nasal, a velar nasal, or a retroflex in the [] syllable.

Constraint (31) is formulated in a negative statement, for the purpose of making

the contrast between English and Mandarin (i.e. something that English allows but

Mandarin does not). Mandarin allows no consonant cluster in the coda. Constraint (32)

reflects the fact that Mandarin codas are highly restricted to [n], [], and [r] in the []

syllable (details in 3.2). The two constraints are undominated since the Mandarin

output of an English loanword cannot violate the restrictions on Mandarin

phonotactics.

The output should not merely conform to the native phonotatctis, but also be as

close to its input as possible. Constraints that ask for such conformity are faithfulness

constraints, as listed below:

(33) MAX-IO(S): Every segment in the input must have a correspondent in the output.

(No deletion.)

(34) DEP-IO(S): Every segment in the output must have a correspondent in the input.

(No insertion.)

(35) IDENT-IO(F): Correspondent segment in the input and the output must be

identical in feature.

Constraint (33) prohibits the deletion of any input segment in the output, and

constraint (34) bans any insertion of it. Constraint (35) asks for phonetic identity

between correspondent segments. A violation of this constraint is incurred from a

difference in distinctive feature between the correspondent segments in input and

output.

4.3 Ranking and Tableaux

As observed in Tables (7) and (8) in Chapter 3, codas tend to be preserved in

most cases, despite a few exceptions. We plausibly assume the ranking below:

(36) *COMPLEXCODA, CODACONDITION >> MAX-IO(S) >> DEP-IO(S) >>

IDENT-IO(F)

*COMPLEXCODA and CODACONDITION are undominated in Mandarin loanword

phonology, and they are not ranked crucially within the hierarchy. Only relevant

constraint(s) will be listed in the following tableaux to save space. Since codas enjoy

the priority to be preserved in both monosyllabic and disyllabic loanwords, insertion

is inevitable, and therefore MAX outranks DEP. The ranking ‘MAX >> DEP’ is actually

the way Chinese loanword phonology is dealt with in literatures (Yip 1992, Guo 2001,

Wu 2001, Shih 2004, Miao 2005), but we will reconsider this argument later. Ranked

at the bottom is the IDENT-IO(F), since segmental change is bound to take place due

to different sound systems. Let us suppose the grammar is good and present an

example it can account for.

Tableau (37)³

Input: /wt/ watt Output: [wa.t] 瓦特

MAX-IO(S) DEP-IO(S) IDENT-IO(F)

a. wa *! *[+back]

)b. wa.t⁴ * *[+back]*[+aspirated]

c. wa.l * *[+back]*[+voiced]*[+lateral]!

In the candidate evaluation, we consider only outputs with legitimate segments.

In (37), candidate (a) incurs a violation of MAX-IO(S) for the lack of the coda [t], and

thus is eliminated. Candidates (b) and (c) both pass MAX-IO(S) for their preservation

of the codas but equally violate DEP-IO(S) once for the insertion of the vowel []. At

the bottom, candidate (b) violates IDENT-IO(F) twice for the [-back] of [a]⁵ and the

[+aspirated] of [t], while in addition to the same violation of [a], the [l] of candidate

(c) incurs one more violation of IDENT-IO(F) than candidate (b) for its different

feature values of [+voiced] and [+lateral] from [t]. The winner is thus candidate (b).

Let us turn to the case in Table (8). The coda of a disyllabic loanword can be

deleted, with a percentage of 21.82% in ‘_VC’ structure, which is much larger than

3 In a tableau, as repeated here, an asterisk ‘*’ stands for one violation of a certain constraint, and the exclamation mark ‘!’ after an asterisk represents the elimination of a candidate when a fatal violation occurs. The shading in the cells thus indicates that the violation content is no longer relevant. A solid vertical line separating two columns means the constraint on the left side outranks the one on the right side, while there is no ranking between two constraints separated by a dotted line.

4 Careful readers may have noted that an even better candidate is [wa.t], which is however not the actual output. This can be explained by the fact that there are a group of characters which are more likely to be chosen as the loanwords in Mandarin, such as 斯 ([s]), 特 ([t]), 克 ([k]), 尼 ([ni]), 里 ([li]), etc. The consideration of adopting preferable characters as loanwords sometimes overrides the choice of the phonetically closest one. We do not pursue the translation customs and focus on the processes governed by phonological factors.

5 The features of segments are based on Halle (1995), in which the features are presented by plus-minus contrast.

that of monosyllabic loanwords (6.29%). Consider the instance below:

Tableau (38)

Input: /kært/ carat Output: [k.la] 克拉

MAX-IO(S) DEP-IO(S) IDENT-IO(F)

?a. k.la *! *[-low]*[+back]*[+lateral]*[+low]

*)b. k.la.t * *[-low]*[+back]*[+lateral]*[+low]

Based on the proposed constraint ranking, in (38), MAX-IO(S) eliminates

candidate (a) since the coda of the second syllable is omitted in the output. Candidate

(b) should win despite its violations of DEP-IO(S) and IDENT-IO(F) ([-low] and

[+back] of [], [+lateral] of [l], and [+low] of [a]). This result, however, is not correct:

the actual output is candidate (b). Do we need two separate grammars to account for

monosyllabic and disyllabic loanwords? What about trisyllabic and quadrisyllabic

ones? Certainly, it would be ad hoc to say that monosyllabic and disyllabic loanwords

have two different loanword grammars. There should be something wrong with the

proposed grammar thus far and modification is necessary.

4.4 MINIMAL WORD

A critical constraint regarding prosodic level leads to this contrast:

(39) MINIMAL WORD: A word must consist of at least two syllables.

According to the Prosodic Hierarchy, a prosodic word must contain at least one

foot, and by Foot Binarity, a foot must be bimoraic or disyllabic. Mandarin, a

quantity-insensitive system, thus prefers two syllables as the size of a minimal word.

This has been proved through our monosyllabic data (as mentioned below Table 1,

74.52% of the outputs are disyllabic). As a result, in monosyllabic loanwords, when

the coda is the illicit obstruent, Mandarin speakers tend to insert a vowel after the

obstruent, instead of deleting it. In disyllabic loanwords, on the other hand, vowel

epenthesis is not obligatory. Deletion is an alternative adaptation strategy.

If this assumption is correct, then the preference for two syllables in a word is

determined by the effect of MINIMAL WORD. We thus modify the previous ranking as

follows:

(40) *COMPLEXCODA, CODACONDOTION >> MAX-IO(S), DEP-IO(S) >> MINIMAL

WORD >> IDENT-IO(F)

In (40), MAX-IO(S) and DEP-IO(S) are ranked in the same level. MINIMAL

WORD is inserted below MAX-IO(S) and DEP-IO(S) and above the lowest

IDENT-IO(F). Consider (41):

Tableau (41) (revised from (37)) Input: /wt/ watt

Output: [wa.t] 瓦特

MAX-IO(S) DEP-IO(S) MINWD IDENT-IO(F)

a. wa * *! *[+back]

)b. wa.t * *[+back]*[+aspirated]

c. wa.l * *[+back]*[+voiced]*[+lateral]!

In (41), the three candidates equally have one violation of MAX-IO(S) or

DEP-IO(S), which are not crucially ranked. Candidates (b) and (c) defeat (a) because

they conform to MINIMAL WORD while (a) does not. Candidate (b) wins for one less

violation of the lowest IDENT-IO(F) than candidate (c). Thus the selection of

candidate (b) has nothing to do with MAX-IO(S), as revealed by mistake in Tableau

(37).

Tableau (42) (revised from (38)) Input: /kært/ carat Output: [k.la] 克拉

MAX-IO(S) DEP-IO(S) MINWD IDENT-IO(F)

)a. k.la * *[-low]*[+back]*[+lateral]*[+low]

())b. k.la.t * *[-low]*[+back]*[+lateral]*[+low]

In (42), candidates (a) and (b) each violates a highest constraint, passes the

regulation of MINIMAL WORD, and equally incurs four violations of IDENT-IO(F).

Both are considered to be the winners, even though the actual output should be

candidate (a), which is a result of random selection. According to the corpus, the

adaptation of a stop coda is highly random, with 56.45% by preservation and 38.71%

by deletion. Similar examples to (37) are [tn.ni] (珍妮) from [dænt] (Janet),

[pei.k] (杯葛) from [bkt] (boycott), etc., where the coda [t] is deleted;

counterexamples are [ja.lwo.t] (夏洛特) from [rlt] (Charlotte), [tja.na.t]

(賈奈特) from [rnt] (Garnett), among many others, where the [t] is retained.

It should be noticed that in the two instances, [wa.t] and [k.la], by ranking

MINIMAL WORD over MAX-IO(S) and DEP-IO(S) we can have the same optimal

output. Why should MINIMAL WORD be ranked below MAX-IO(S) and DEP-IO(S)

then? The reason can be clarified in loanwords without a coda, exemplified by the

following case:

Tableau (43)

Input: /pa/ pie Output: [pai] 派

MINWD MAX-IO(S) DEP-IO(S)

?a. pai *!

*)b. pai.t **

In this tableau, MINIMAL WORD outranks MAX-IO(S) and DEP-IO(S), and

candidate (b) is wrongly chosen as the winner. It means any polysyllabic output is

better than [pai]. Apparently this constraint ranking is not applicable to

loanwords without codas.

Tableau (44)

Input: /pa/ pie Output: [pai] 派

MAX-IO(S) DEP-IO(S) MINWD

)a. pai *

b. pai.t *!*

In (44), MAX-IO(S) and DEP-IO(S) outrank MINIMAL WORD, and the output is

correctly predicted.

It has been addressed that in monosyllabic loanwords codas are mostly preserved

to form a disyllabic output, regardless of what the coda is. With a closer examination

of disyllabic loanwords, however, it is found that different classes (in terms of manner

of articulation) of consonants have different adaptation behaviors. In Tables (9) to

(26), we have statistic evidence showing the divergent adaptation strategies among

them. The contrasts are reflected in the grammar as well. In the next section, we

discuss those salient differences and make necessary revisions to the constraint

ranking established so far.

4.5 Fricative and Affricate Faithfulness

As seen in Table (19), fricatives and affricates as the simplex coda are preserved

in coda position in almost every disyllabic loanword, with the high percentage of

83.04%. When the fricative is a part of the consonant cluster in the coda position (the

‘_Vns’, ‘_Vnz’, and ‘_Vms’ in Table (23), the whole data in Table (24), the ‘_Vrs’

and ‘_Vrf’ in Table (25), and the ‘_Vls’, ‘_Vlf’, and ‘_Vlz’ in Table (26)), still, it

surfaces in comparatively more cases, with the percentage of 73.53%. By virtue of

friction noise made by the air escaping through a narrow obstacle, a fricative or an

affricate is highly salient and easily stands out from surrounding stops, liquids or

vowels. This acoustic saliency of fricatives and affricates makes them obligatorily

parsed in the loaning process, and thus we formulate the constraint below:

(45) MAX-IO([+del rel]): A segment with the feature [+delayed released] in the input

must have a correspondent in the output. (No deletion.)

This constraint should outrank other faithfulness constraints as below:

(46) *COMPLEXCODA, CODACONDITION >>MAX-IO([+del rel]) >> MAX-IO(S),

DEP-IO(S) >> MINIMAL WORD >> IDENT-IO(F)

An example is given below:

Tableau (47)

Input: /dsko/ disco Output: [ti.s.k] 迪斯可

MAX-IO

([+del rel]) MAX-IO(S) DEP-IO(S) IDENT-IO(F)

a. ti.k *! * *[-voiced]*[-round]

)b. ti.s.k * *[-voiced]*[-round]

In (47), candidate (a) is ruled out for the violation of MAX-IO([+del rel]).

Candidate (b) preserves the fricative coda by inserting a vowel to form an

independent syllable to satisfy MAX-IO([+del rel]), and wins as the optimal output.

The two violations of IDENT-IO(F) by both candidates are brought about by the

[-voiced] of [t] and the [-round] of [].

4.6 Nasal Faithfulness

Consider Tables (11) and (20). English nasals as the simplex coda are always

mapped to Mandarin in both monosyllabic and disyllabic loanwords. In Tables (14)

and (23), furthermore, where the nasal is part of the cluster coda, it still surfaces in

most cases, with the high percentage of 96.67% (excluding two ‘unidentified’ data).

In English [n m ] are allowed to be the coda, but in Mandarin only [n ] are

licit in the position. Despite the slight distinction, at least nasals are the only permitted

consonants in the coda position in Mandarin (if glides are taken as part of the

diphthong in nuclear position), and it is thus plausible that English nasal codas are

always perceived and parsed by Mandarin speakers. Even for the illicit [m], Mandarin

speakers preserve the [m] coda as well, mostly by means of feature change or vowel

insertion—in our corpus of both monosyllabic and disyllabic loanwords, [m] is 100%

parsed without exception. In terms of perceptual-similarity, on the other hand, nasals

as sonorants are perceptually salient and the deletion of which makes the alternation

large. We thus assume the constraint below:

(48) MAX-IO([+nasal]): A segment with the feature [+nasal] in the input must have a

correspondent in the output. (No deletion.)

MAX-IO([+nasal]) has no interaction with MAX-IO([+del rel]) and both are

ranked in the same level.

(49) *COMPLEXCODA, CODACONDITION >> MAX-IO([+del rel]), MAX-IO([+nasal])

>> MAX-IO(S), DEP-IO(S) >> MINIMAL WORD >> IDENT-IO(F)

As mentioned above, English [n] and [] are legitimate codas in Mandarin, and

therefore MAX-IO([+nasal]) is not violated in instances where the coda is [n] or [].

Compare the two tableaux below:

Tableau (50)

Input: /ndn/ engine Output: [in.ti] 引擎 MAX-IO

([+nasal]) MAX-IO(S) DEP-IO(S) IDENT-IO(F) a. i.ti

*! * *[+high]*[-voiced]

*[+aspirated]*[+dorsal]

)b. in.ti *[+high]*[-voiced]

*[+aspirated]*[+dorsal]

In (50), candidate (b) wins for the fatal violation of MAX-IO([+nasal]) by

candidate (a). The four violations of IDENT-IO(F) by both candidates are the [+high]

of [i], the [-voiced] and [+aspirated] of [t], and the [+dorsal] of []. However, the

same candidate is able to win in a grammar without MAX-IO([+nasal]) as well, as

revealed in (51):

Tableau (51)

Input: /ndn/ engine Output: [in.ti] 引擎

MAX-IO(S) DEP-IO(S) IDENT-IO(F) a. i.ti

*! *[+high]*[-voiced]

*[+aspirated]*[+dorsal]

)b. in.ti *[+high]*[-voiced]

*[+aspirated]*[+dorsal]

MAX-IO(S) alone is adequate in eliminating candidate (a). The assumption of

MAX-IO([+nasal]) seems to be untenable in loans with [n] or [] codas. Accordingly,

to testify the effect of MAX-IO([+nasal]), we should evaluate nasal codas with the

illicit [m].

Tableau (52)

Input: /hæmlt/ Hamlet

Output: [xa.mu.lei.t] 哈姆雷特 MAX-IO

([+nasal]) MAX-IO(S) DEP-IO(S) IDENT-IO(F) a. xa.lei.t

*! * * *[+dorsal]*[+high]

*[+aspirated]

)b. xa.mu.lei.t

** *[+dorsal]*[+high]

*[+aspirated]

In (52), MAX-IO([+nasal]) eliminates candidate (a) since the nasal coda of the

first syllable in the input does not surface in this output candidate. Candidate (b) thus

wins as the best output for its preservation of it. The three violations against

IDENT-IO(F) by the two candidates are incurred for the [+dorsal] of [x], the [+high] of

[ei], and the [+aspirated] of [t], though they are irrelevant in this case.

But for MAX-IO([+nasal]), candidate (a) would become the same competitive as

candidate (b), as shown in (53):

Tableau (53)

Input: [hæmlt] Hamlet

Output: [xa.mu.lei.t] 哈姆雷特

MAX-IO(S) DEP-IO(S) IDENT-IO(F)

*())a. xa.lei.t

* * *[+dorsal]*[+high]

*[+aspirated]

)b. xa.mu.lei.t

** *[+dorsal]*[+high]

*[+aspirated]

With equally two violations of MAX-IO(S) and DEP-IO(S), candidates (a) and (b)

should both win as the optimal outputs, which however is an incorrect prediction.

It may be noticed that a theoretically better candidate in this example is

[xan.lei.t] or [xa.lei.t], with the illicit nasal coda [m] changing into a licit [n]

or [], since there will be no violations of M^AX-IO(S) and DEP-IO(S) by them.

According to the statistics, segmental change (mainly ‘[m] → [n] or []’) is indeed

the major adaptation strategy of English [m] codas in disyllabic loanwords (vowel

insertion is mostly adopted for [m] in monosyllabic loanwords for bisyllabicity), with

93.75% in simplex codas and 100.00% in complex ones, as in ‘[æmpr] (ampere) →

[anpei] (安培)’ and ‘[ædms] (Adams) → [ja.ta.s] (亞當斯)’ respectively. Vowel

insertion is the secondary choice. In the cases with segmental change, like [n, ]

codas, the effect of MAX-IO([+nasal]) is not visible, since MAX-IO(S) alone is

enough in electing the correct optimal outputs. Though this case is one exception that

chooses vowel insertion as the repair strategy, it is however an appropriate example

that manifests the effect of MAX-IO([+nasal]).

4.7 Deletion of Postnuclear Retroflexes

Contrary to fricatives, affricates and nasals, it is found in our corpus that

retroflexes are mostly ignored in the postnuclear position. In disyllabic data, as Table

(22) reveals, the [r] in the English ‘_Vr’ type of codas is mostly deleted in their

Mandarin counterparts, representing a percentage up to 88.24%. In the data with the

‘_VrC’ form in Table (25), moreover, the postnuclear [r] is always unparsed in its

Mandarin adaptation, with no exception discovered. Even for monosyllabic loanwords,

in which codas are often preserved to reach bisyllabicity, the deletion of [r] also takes

precedence over the effect of MINIMAL WORD—in Table (13), the [r] in the ‘_Vr’

structure is mostly ignored by Mandarin speakers, accounting for 71.43%; in Table

(16), moreover, [r] is never parsed in data with the ‘_VrC’ structure.

When a structure in LS happens to be licit in LT, the structure should be

completely borrowed into LT. In our data, however, it is not always the case. English

has retroflex vowels [] and [], and Mandarin has []. However, English [] and []

are not faithfully preserved in Mandarin. Among the 90 English syllables with [] or

[] in monosyllabic and disyllabic syllables, only one is completely preserved (i.e.

[ma] (Mayer) → [mei.] (梅爾)). There are only three cases where the retroflex [r]

is vocalized into the retroflex vowel [] in Mandarin (e.g. [tk] (Turkey) →

[tu..ti] (土耳其)). In the remaining 86 cases the retroflex is surfaced as nothing

in the Mandarin output (e.g. [lez] (Laser) → [lei.] (雷射)), accounting for up to

95.56%. We are thus interested in the factor that leads to the seemingly unusual

adaptation tendency of postnuclear retorflexes, even overriding the faithful mapping

of retroflex vowels between English and Mandarin.

In the following subsection, we review a few previous works on the facts of

liquids.

4.7.1 The Facts of Liquids

During the production of liquids, ‘one part of the oral channel is blocked, while

another part remains unobstructed and allows the air to escape freely’ (Roca and

Johnson 1999). There are relevant remarks in the literature that correspond with such

a nature of liquids. Fay & Culter’s (1977) experiment shows that liquids possess

vowel-like formants so that they fail to stand out from the surrounding vowels. When

it comes to loanword adaptation, it makes sense that liquids tend to be ignored since

they are perceptually not salient.

Silverman (1992) exemplifies the distinction of the Perceptual Level and the

Operative Level with several Cantonese loanwords where [r] is part of the onset, as

reproduced below:

(54) a. break → [pik lik]

print → [p’i lin]

cream → [key lim]

b. printer → [p’n t’a]

broker → [puk k’a]

freezer → [fi sa]

(Silverman 1992, p. 290)

The [r]’s in (a) are preserved, while those in (b) are not present in the Cantonese

form. The answer to this asymmetry is that liquids are salvaged if the outputs reaches

bisyllabicity, and they are deleted when the output would still be disyllabic without

the interpretation of the liquid.

In our analysis, however, two problems related to his assertion rise. First, in

Mandarin loanwords, retroflexes are mostly preserved in a ‘C + retroflex’ onset, no

matter whether the English input is monosyllabic or disyllabic: within monosyllabic

loanwords, it is 100% preserved without exception; in disyllabic ones, it is preserved

in most cases, representing a predominant percentage of 89.09% (excluding 2

unidentified cases).

Secondly, though both retroflexes and laterals are categorized as liquids, the

adaptation tendencies of them in Mandarin are quite different. As mentioned above,

retroflexes in the coda position are mostly unparsed in the Mandarin forms, while

laterals are not like that. In monosyllabic data, as Table (11) shows, the [l] is mostly

parsed in the ‘_Vl’ structure by inserting a glide and a vowel after it, such as ‘[pl]

(Paul) → [pau.lwo] (保羅)’, with the percentage of 90.91%; in the data with the

‘_VlC’ cluster, as shown in Table (16), 75.00% is preserved. In disyllabic loans,

furthermore, Table (20) reveals that the lateral is preserved in data with the ‘_Vl’

structure in comparatively more cases, representing 65.38%; in Table (25), where the

lateral occurs as the postnuclear consonant in the cluster coda, however, only 22.22%

of the laterals are preserved. Based on the statistic results above, we come to the

conclusion that laterals in the coda position are inclined to be preserved in the

Mandarin forms, contrary to retroflexes. A similar example is observed by

Kenstowicz (2001) in the French loanwords in Fon.

Extending Silverman’s observation, Yip (1993) attributes the deletion of liquid in

the second half of an onset cluster to the fact ‘that the liquids are less salient than the

preceding obstruents and that this lack of salience renders them relatively vulnerable

to deletion.’ (Yip 1993, p. 268) However, this property of liquids is not reflected in the

onset position in our data, as also revealed by the statistic results above.

The Mandarin loanwords collected by Shih (2004) correspond with us in that

‘liquids assigned to the onset position are almost always preserved; only those

assigned to the coda position may be deleted.’ (Shih 2004, p. 81) He supposes that this

is because the data are mainly composed of transliterations of proper names, and

when it comes to proper names, people tend to preserve as much information from the

input in the output as possible, even if they are not salient. An assumption like this

would lead to at least two problems. Firstly, Mandarin adaptations of English proper

names accounts for a large percentage in his data. It is somewhat arbitrary to attribute

a particular adaptation behavior to the general fact of the linguistic data. An

explanation like this fails to account for the deletion of postnuclear retroflexes in

Mandarin adaptations, which are mostly transliterations of proper names too but tend

to be unparsed. Secondly, the LT of Silverman (1992) is Cantonese, but that of Shih

(2004) is Mandarin. It is plausible that the same input will be adapted differently by

speakers of two different dialects. Thus a retroflex can be preserved or deleted in the

onset in dialect A (Cantonese), when the prosodic constraint takes precedence;

whereas it may always be preserved in the onset in dialect B (Mandarin), when the

constraint of positional faithfulness of the onset ranks high. The above points are

missed in his statements.

The fact that postnuclear retroflexes do not surface in any forms in our data is

consistent with the findings above. First, retroflex is perceptually not salient and

hence they tend to be deleted in loanword adaptations. Second, retroflexes and laterals,

though both are classified as liquids, have different behaviors in loanword adaptations:

both are preserved in the onset position but the former tend to be deleted and the latter

be preserved in the coda position. Third, the asymmetry between the preservation of [r]

in the onset and the deletion of that in the coda is attributed to its occurring contexts:

word-initial positions are the phonologically prominent⁶ position but word-final

positions are not. It is thus plausible that whatever segment in a prominent position is

perceptually more salient than that in other positions. In the next section, the

‘unsaliency’ of postnuclear codas is projected into constraint interactions.

4.7.2 OT Analysis

As mentioned above, postnuclear retroflexes are apt to deletion and do not

surface in any form in both simplex and complex codas, which contradicts the

universal preference for retention in loanwords (Paradis 1995, Paradis and Darlene

1997). In the OT framework, Alderete’s (1999a, 2001) TAF constraints are well-suited

for the effect of the perceptual ‘unsaliency’ of postnuclear retroflexes in loanword

adaptation. To apply the TAF theory, which is initially defined in

morpho-phonological processes, to purely phonological alternations, we extend the

6 Refer to Beckman (1997b), in which relevant psycholinguistic evidence of the initiality effects is provided, including evidence from word recognition, word recalling, mispronunciation, etc.

theory from its application in OO- and BR-anti-faithfulness to IO-anti-faithfulness, as

described below:

(55) Faithfulness and Anti-Faithfulness for Postnuclear Retroflexes

a. MAX-IO(r/V_): A postnuclear retroflex in the input must have a

correspondent in the output. (No deletion.)

b. ¬MAX-IO(r/V_): It is not the case that a retroflex in the input must have a

correspondent in the output.

Between this pair, it is ¬MAX-IO(r/V_) that triggers the deletion of postnuclear

retroflexes in loanword adaptations. Like MAX-IO([+del rel]) and MAX-IO([+nasal]),

¬MAX-IO(r/V_) should outrank the general faithfulness constraints, as listed in (56),

which is the final version of Mandarin loanword grammar:

(56) *COMPLEXCODA, CODACONDITION >> MAX-IO([+del rel]), MAX-IO([+nasal]),

¬MAX-IO(r/V_) >> M^AX-IO(S), DEP-IO(S) >> MINIMAL WORD >>

IDENT-IO(F)

Within the constraint ranking, it is ¬MAX-IO(r/V_) that differentiates loanword

phonology from native phonology. The reason why we do not formulate a markedness

constraint to avoid the ‘[r] → []’ mapping, such as ‘*_Vr’, is twofold. First,

Mandarin does allow [r] in the coda position in the output, despite simply in the ‘ +

r’ combination. Second, the other liquid type, lateral, is in fact often mapped to

Mandarin [] in the output form. It is only the liquid coda [r] that is mapped to

nothing. Such an unfaithful mapping corresponds with Alderete’s (1999b, 2001) idea

that anti-faithfulness constraints ‘induce an alternation by requiring a violation of

related faithfulness constraint in word pairs’ (Alderete 2001, p. 9).

An example that illustrates the effect of ¬MAX-IO(r/V_) is given in (57):

Tableau (57)

Input: /jrk/ Yorkshire Output: [e.k.ja] 約克夏

¬MAX-IO(r/V_) M^AX-IO(S) DEP-IO(S) IDENT-IO(F) a. e..k.ja

*! **

*[+round]*[-round]

*[-back]*[-consonantal]

*[+dorsal]*[+aspirated]

*[+anterior]*[+low]

)b. e.k.ja

* **

*[+round]*[-round]

*[-back]*[+aspirated]

*[+anterior]*[+low]

In (57), candidate (a) is ruled out by incurring a violation of ¬MAX-IO(r/V_) for

its preservation of the postnuclear retroflex in the first syllable, interpreted as the mid

vowel []. Candidate (b) wins for its obedience to ¬MAX-IO(r/V_), despite its

violation against the lower MAX-IO(S). The 8 violations of IDENT-IO(F) by candidate

(a) are incurred by the [+round] of [y], the [-round] and [-back] of [e], the

[-consonantal] and [+dorsal] of [], the [+aspirated] of [k], the [+anterior] of [], the

[+low] of [a]. The six violations of I^DENT-IO(F) by candidate (b) are the same, minus

the two incurred by [].

4.8 Conclusion

In this chapter, we account for the statistic results of English coda adaptations

within OT. The adaptation tendencies observed in the previous chapter are thus

projected to constraint interaction, as listed below:

a. MINIMAL WORD indicates a tendency of changing monosyllabic loanwords into

disyllabic ones.

b. The relative ranking between MAX-IO(S) and DEP-IO(S) is not crucial so that

stop codas sometimes preserve and sometimes delete.

c. MAX-IO([+del rel]) indicates the obligatory preservation of fricatives and

affricates.

d. MAX-IO([+nasal]) indicates the obligatory preservation of nasals.

e. ¬MAX-IO(r/V_) indicates the obligatory deletion of postnuclear retroflexes.

The constraints interacting in the loaning process are ranked as

*COMPLEXCODA, CODACONDITION >> MAX-IO([+del rel]), MAX-IO([+nasal]),

¬MAX-IO(r/V_) >> M^AX-IO(S), DEP-IO(S) >> MINIMAL WORD >> IDENT-IO(F)

Syllable structure constraints rank high owing to the obligatory obedience to native

phonotactics, while IDENT-IO(F) ranks low since sound changes from LS to LT are

inevitable. There should be no ranking between MAX-IO(S) and DEP-IO(S), and the

tendency to preserve the coda in monosyllabic loans is attributed to the prosodic

effect of MINIMAL WORD. MAX-IO([+del rel]), MAX-IO([+nasal]) and

¬MAX-IO(r/V_) outrank M^AX-IO(S) and DEP-IO(S) owing to the perceptual saliency

of fricatives, affricates, and nasals and the weak perception cues to postnuclear

retroflexes. The existence of ¬MAX-IO(r/V_) constitutes evidence to prove that

loanword phonology is independent from native phonology.