Loanword Phonology

Chapter 2 Literature Review

2.4 Loanword Phonology

Loanword phonology occurs when words from a language enter another one and, through the loan process, non-native sounds undergo certain degrees of transformations. Non-native sounds include both segments and suprasegments but since the present work focuses on segments, suprasegments are not mentioned anymore. The language where loanwords originally come from is called “donor language” or “source language” (henceforth LS). For the language where loanwords enter, it is called “recipient language” or “target language”

(henceforth LT). According to a great amount of literature concerned with loanword phonology, three distinct perspectives are going to be introduced.

First, adaptations happen primarily because of misperception (Dupoux, Kakehim, Hirose, Pallier & Mehler 1999, Peperkamp & Dupoux 2003). The LT users misperceive segments in the LS and thus adaptations can be observed. Generally, segments that do not exist in the LT

would be wrongly perceived. The wrongly perceived segments, in fact, are perceived as the most similar segments that exist in the LT.

Second, adaptations occur mainly due to the grammar that is not allowed in the LT. More specifically, the violation of the LT phonology results in the adaptations (Jacobs &

Gussenhoven 2000, LaCharite & Paradis 2005). In LaCharite & Paradis’ research, they argues that if there is a given LS phonological category (i.e. feature combination) that also can be found in the LT, the loaned sound sequence will be preserved without any transformations, even though there are phonetic differences. On the other hand, if the given LS phonological category cannot be found in the LT, which means the phonology category is ungrammatical in the LT,

the loaned sound sequence will be adapted based on the most similar phonological category in the LT.

Third, adaptations are conducted based on both perception and grammar evaluation.

(Kenstowicz 2005, Silverman 1992, Yip 2006,). This viewpoint is the intermediate between the aforementioned two perspectives. Through the perception, identical segments in the LS and LT

remain unchanged; non-identical or non-native segments would be replaced by the most similar ones that exist in the LT. After the segmental modification, loanwords are checked again in order to satisfy the phonotactics established by the LT. Silverman (1992) suggests there are two modification levels for words borrowed from one language to another. The first is “Perceptual Level” dealing with inventories, both sound and tonal inventories. Adaptations having nothing to do with inventories but being related to phonotactics would be handled in the second level,

“Operative Level.” The output of the Perceptual Level would be applied to the Operative Level so as to judge whether it violates phonotactic constraints or not. Silverman’s model of loanword phonology (1992:293) is given in (11).

(11) Silverman’s loanword adaptation model (1992:293)

Chapter 3

Sixian-Hakka Phonotactics

This chapter manages four phonotactic phenomena that can be observed in S-Hakka. The first one is how an empty nucleus position is filled, which will be presented in §3.1. The second one is about rhymes, the combination of a nucleus vowel and a coda consonant in a tautosyllable, which will be shown in §3.2. The third one is related to diphthongs, the combination of two vowels at the nucleus position, which will be dealt with in §3.3. Finally, palatalization will be discussed in §3.4.

3.1 Empty Nucleus

In S-Hakka, the underlying forms /, , / are realized as [, , ] in their surface representations (Chung 2004:72, 159-161). Regarding the derivation, this section aims to provide an optimality-theoretic account to explain how an empty nucleus position is filled and which segment is going to fill the vacancy. In §3.1.1 and §3.1.2, the phonetic descriptions for alveolar-dentals [, , ] and the derived vowel [] will be presented respectively, and several related researches will be reviewed as well. In §3.1.3, there will be an optimality-theoretic analysis.

3.1.1 The Phonetic Description for S-Hakka Alveolar-dentals

With respect to the natural class of S-Hakka alveolar-dentals (i.e. [, , ]) , Chung (2004:62) based on the place of articulation states that [t] and [t] are more like alveolars while [] is more like dentals. In contrast, Gu (2005:125) takes them as dentals and Liang (2004:6) considers them as alveolar-fronts (i.e. the place of articulation is slightly more advanced than the one of general alveolars). Chung (2004) assigns [+apical] to [, , ]

because they are pronounced with the tongue tip. Nevertheless, it is important to note that

“apical” is not taken as a feature in The Sound Pattern of English (SPE) (Chomsky & Halle 1968/1991).

Based on SPE (Chomsky & Halle 1968/1991), the feature [apical] which is adopted by Chung (2004) is better to be replaced by [distributed]. Roca & Johnson (1999) provide a clear definition for [distributed]. “The feature [distributed] refers to the “distribution” of the tongue over the passive articulator” (Roca & Johnson 1999:106). Roca & Johnson elaborately explain that only coronals are associated with [distributed] and its feature value, plus or minus, lies in the tongue body that precisely involves in. In other words, the feature value of [distributed]

relies on the size of air obstruction in the vocal tract. An obstruction that is shorter along the airflow (e.g. alveolars [, ]) is [-distributed], whereas an obstruction that is considerable longer along the airflow (e.g. palatal-alvelars [, ]) is [+distributed]. It is worth noting that Roca &

Johnson use the term “coronal apical” for sounds that are [-distributed] and “coronal laminal”

for sounds that are [+distributed].

The term “apical” adopted by Roca & Johnson (1999) and Chung (2004) is a little be different. Chung uses “apical” as a feature and assumes [+apical] is assigned when sounds are articulated with tongue tip. For example, Chung assigns [+apical] to both alveolars (e.g. [, ,

, ] in S-Hakka) that are articulated with the tongue against the alveolar ridge, and alveolar-dentals (e.g. [, , ] in S-Hakka) that are articulated with the tongue against the back of upper teeth and along the alveolar ridge. On the other hand, Roca & Johnson do not use “apical”

as a feature but use the term to describe sounds with the air obstruction which is caused by tongue tip and occurs at a particular point in the mouth like alveolar ridge. For instance, Roca

& Johnson assign [-distributed] to alveolar [t] while [+distributed] to alveolar-dental [].

According to the place of articulation, alveolars and retroflexes are [-distributed] whereas dentals, alveolar-dentals, palatal-alveolars and palatals are [+distributed]. From now on, the

thesis adopts [distributed] proposed by Roca & Johnson, rather than [apical] put forward by Chung, to do the feature specification because [distributed] is more widely adopted when researchers discuss phonetic features. In short, based on the definition of [distributed], S-Hakka alveolar-dentals [, , ] and alveolar-palatals [, , , ] are [+distributed].

3.1.2 The Phonetic Description for S-Hakka Derived Vowel

The vowel [] is named as “apical vowel” by researchers whose studies are for different languages. Chung (2004) who studies several Hakka dialects calls it an “apical vowel” because the vowel only appears after alveolar-dentals which are articulated with tongue tip. Gu (2005:130) who also focuses on Hakka dialects more specifically calls it an “apical front vowel”

and describes that it is pronounced with tongue tip approaching the back of upper teeth and with lip spreading⁴. Tseng & Chou (1999) conduct a research about the phonetic transcriptions for languages used in Taiwan, including Mandarin Chinese, Taiwanese and Hakka, and they take [] as apical vowel as well. Huang (2011) whose study is concerned with epenthetic vowels in Squliq Ayayal, a language in Austronesian family, also names [] as apical vowel;

furthermore, in her study, the apical vowel also has to co-occur with onset [], [] or [], which is similar to S-Hakka requiring [] to appear with onset [], [ or [].

With regard to S-Hakka vowel quadrangle, the one suggested by Chung (2004) is presented in (12). Chung assumes that the location of [] in the vowel quadrangle has to be more advanced than [], which means [] is [-back]. His argument is based on the fact that []

can only occur behind alveolar-dentals, so the place of articulation for [] approaches to the back of upper teeth and alveolar ridge rather than palatal which is the place of articulation for front vowels (e.g. //).

4 Gu (2005:130) does not transcribe the S-Hakka apical vowel as [] but as [ï].

(12) S-Hakka vowel quadrangle (Chung 2004:71) Front Central Back

High   

Mid   

Low   

However, according to the acoustic analyses of S-Hakka inventory (Liang 2004, Tan 2010, Zheng 2010), the location of [] is not like what Chung assumes to be in front of [], but it locates behind the vowel []. The experimental result in Liang’s study is presented in (13).

Besides, in Tan’s and Zheng’s studies, the vowel [] is found to gradually merge into the vowel [] by young language users. The mergence is carrying out because [] is very restrictive in distribution, [] is hard to be independently pronounced, and [] is not easy to be correctly perceived when being compared with [] and [].

(13) S-Hakka vowel quadrangle (F2: F1) (Liang 2004:97)

For the present work, vowel [] is considered as a central vowel and thus is [+back], which follows the acoustic analyses but differs from Chung’s assumption. In conformity with the acoustic experimental result makes analyses in the current study more logical and reasonable.

The vowel quadrangle adopted by the thesis is shown in (14).

(14) S-Hakka vowel quadrangle (The thesis)

Front Central Back

High   

Mid   

Low   

3.1.3 An Optimality-theoretic Analysis

Although phonologically it is said that nucleus is the only obligatory component in a syllable, S-Hakka consists of a syllable type in which only onset is filled in the underlying representation. In Chung’s study (2004:72, 159-161), S-Hakka allows three consonants that have no need to combine with vowels and can become as independent syllables: /, , /.

Chung also provides their surface representations: [, , ]. By comparing the underlying with their surface forms, it can be noticed that [] occurs. The occurrence of [] indicates that S-Hakka does not allow the “empty nucleus” which refers to the condition that the nucleus position is not filled by any segment; therefore, the language figures out its own way to solve the problem. The current study uses FILL^Nuc in (15) to capture the fact that the language requires a segment at the nucleus position.

(15) FILL^Nuc: Nucleus positions are filled with segmental materials.

Chung assumes that the empty nucleus position is occupied by [] for the reason that [+distributed]⁵ spreads from the preceding alveolar-dental consonant (i.e. [, , ]) to the following nucleus position. If based on Chung’s assumption a constraint is proposed in order to describe the derivation, AGREE[distributed] is given in (16) which asks a derived vowel to agree with its preceding consonant on [distributed].

(16) AGREE[distributed]: A derived vowel must agree with its preceding consonant on the feature value of [distributed].

However, AGREE[distributed] is not enough, since [, , ], [, , ] and [, , ]

5 In Chung’s (2004) study, he adopts [+apical] but it is replaced by [+distributed] in the thesis. Please refer to

§3.1.1 for more details.

all satisfy the constraint due to the fact that vowels [, , ] are all [+distributed], as can be seen in (17). Then, why it is [] rather than [] or [] that is derived? The present work takes vowel perspective of markedness and typology and the research results are significantly important.

The markedness relationships among vowels found by Lombardi are presented in (18).

(18) Markedness relationships among vowels (Lombardi 2002:5):

i. Languages may vary in whether low or nonlow vowels are less marked.

If low vowels are less marked, the epenthetic vowel will be //.

*low, *nonlow: re-rankable

ii. If nonlow vowels are less marked, other constraints choose among them:

Back vowels are less marked than front vowels: *Front >> *Back Mid vowels are marked: *Mid

Round vowels are marked: *[+round] >> *[-round]

Several things about Lombardi’s research are worth noting. First, Lombardi assumes that it is language-specific whether low or nonlow vowels are marked; as a result, in the markedness hierarchy, the top-ranking constraints usually are used to determine in a language whether it is low or nonlow vowel that is marked. Second, being concerned with tongue height, vowels are

6 The segment [] is [+distributed], which is based on Chung (2004). The segments [, ] are [+distributed], which is based on Clements & Hume (1995).

specified [nonlow] (e.g. [, , , , ] in Lombardi’s study), [mid] (e.g. [, ] in Lombardi’s study) or [low] (e.g. [] in Lombardi’s study). Third, Lombardi finds that mid vowels are universally marked but *[mid] is freely ranked among other markedness constraints (Lombardi 2002:11). Fourth, front vowels (e.g. [, e] in Lombardi’s study) are universally more marked than back vowels (e.g. [, , , , ] in Lombardi’s study), and round vowels are more marked than unround vowels. In the thesis, constraint rankings put forward by Lombardi are adjusted to fit the approach of binary values of phonetic features in order to have a consistent analytic approach with the succeeding sections. Thus, [nonlow] in Lombardi’s study is replaced by [-low], [low] by [+[-low], and [mid] by [-high, -low]. Furthermore, [front] in Lombardi’s study is substituted by [-back], and [back] by [+back]. Being concerned with S-Hakka markedness relationship of vowels, constraints with their rankings are provided in (19). It is inferred from the derived [] that S-Hakka makes vowels of [+low] more marked, rather than vowels of [-low].

(19) Markedness relationships among S-Hakka vowels (adjusted from Lombardi 2002:5):

i. Languages may vary in whether [+low] or [-low] vowels are less marked:

S-Hakka: *[+low] >> *[-low] (language-specific)

ii. If *[-low] vowels are less marked, other constraints choose among them:

*[-back] >> *[+back] (language universal) *[+round] >> *[-round] (language universal)

*[-high, -low] (language universal) (This constraint is freely ranked among other markedness constraints (Lombardi 2002:11))

The analysis through constraint competing is presented in (21) where how the empty vowel position is filled and which segment is going to fill in the vacancy are explained. Another constraint, IDENT-IO-V in (20), is given so as to make sure that the constraint hierarchy is applied for derived vowels.

20 violates FILL^Nuc. Candidates (21i.d) to (21i.f) are penalized by AGREE[distributed] since [, ,

] are not associated with [distributed]. By comparing candidates (21i.a) to (21i.c), it is found that the markedness constraints for vowels proposed by Lombardi (2002) can correctly predict the output which is (21i.a). Candidate (21ii.f) is chosen as the optimal output among other candidates, which results from the top-ranking of IDENT-IO-V.

Overall, the underlying forms /, , / are realized as [, , ]. The occurrence of a derived vowel is because of FILL^Nuc which is ranked highest. Furthermore, by virtue of AGREE[distributed] and other markedness constraints that are related to vowels, it is [] rather than other vowels that show up.

i. Input: /, , /

^F

^IL

LNuc

I

DENT

-I O -V A

GREE

[di st ri but ed] [ -hi gh, - low ] [ + low ] [ -l ow ] [- ba ck] * [ + ba ck]

 a.   * *

3.2 Rhymes

A rhyme refers to the combination of segments at the nucleus and coda positions. S-Hakka only allows voiceless stops and nasals to serve as codas, so consonants other than [, , , ,

, ] cannot be at the coda position. In (22), possible S-Hakka rhymes are presented, and those that is given a check “” indicates it is grammatical whereas a cross “” indicates unacceptable.

(22) Rhymes in S-Hakka

  

   

   

   

   

   

   

As shown in (22), alveolar codas [] are able to combine with every vowel to form rhymes. However, labial codas [] can only join to vowels that are not [+round], which is captured by OCP^R[labial] in (23). Moreover, velar codas [] that are [+high] can only join to vowels whose place of articulation is not linked to coronal, which is regulated by *Cor-Hi^R in (24). These two constraints are directly related to feature values of involved segments, so feature specifications of related sounds are provided in (25).

(23) OCP^R[labial]: Avoid a rhyme in which two segments are labial sounds.

(24) *Cor-Hi^R: Avoid a rhyme in which a coronal sound is followed by [+high].

(25) Feature matrices for segments involved in S-Hakka rhymes

         comprehended, they are separated based on different vowels. Additionally, in each tableau, there may be more than one optimal outputs. It is because the constraints function as phonotactic constraints, and candidates that do not violate phonotactics do appear in the language in question. The outcomes from the following tableaux, (26) to (31), correspond with table (22).

In (26), candidate (c) is penalized by *Cor-Hi^Rfor the reason that [] is a coronal sound and its following segment [] is [+high]. As for candidates (26a) and (26b), they incur no violation marks and hence are grammatical in S-Hakka. The constraints correctly predict that labial and alveolar codas, but not velar ones, can follow the vowel [].

(27) Input: /eC/

OCP^R[labial] *Cor-Hi^R

 a.

 b. 

c.  *!

In (27), candidate (c) is excluded. Candidate (27c) does not reach the requirement of *Cor-Hi^R because [] is a coronal segment and precedes [] which is [+high]. Candidates (27a) and (27b) satisfy the two constraints so without doubt they can be observed in the real language use.

(28) Input:C/

OCP^R[labial] *Cor-Hi^R

 a.

 b. 

 c. 

In (28), none of the candidates violate any of the constraints and thus the three candidates are all grammatical and acceptable in the language.

(29) Input: /uC/

OCP^R[labial] *Cor-Hi^R a. *!

 b. 

 c. 

In (29), candidate (a) is ruled out due to the fact that both [] and [] are labial sounds but OCP^R[labial] aims to avoid two adjacent labial segments within a rhyme. For candidates (29b) and (29c), they are the real outputs since no violation marks are assigned to them.

In (30), OCP^R[labial] plays a key role to cast away candidate (a) in which [] and its following sound [] are articulated with labials. No violations are brought out by candidates (30b) and (30c); as a result, both of them are real outputs.

(31) Input: /CCC/

i. Input: C

^F

^IL

LNuc

A

GREE

[di st ri but ed] [ -hi gh, - low ] [ + low ] [ -l ow ] [- ba ck] * [+ ba ck] O C P

[l abi al ] *C or -H i

 a.  * *

(31) Input: /CCC/ (Cont.)

In (31), how the vowel [] is derived through A^GREE[distributed] and the markedness constraints please refer to §3.1 for details. Candidate (31d) that has alveolar-dental as its onset contains no vowel to fill the empty nucleus position and hence it violates FILL^Nuc. Concerned with candidate (31c), it incurs the violation mark of *Cor-Hi^R for the reason that [] belongs to coronal sounds and is followed by [] which is [+high]. The constraints then correctly predict candidates (31a) and (31b) are real outputs since they satisfy both OCP^R[labial] and *Cor-Hi^R. To sum up, a permissible rhyme in S-Hakka has to obey both OCP^R[labial] and *Cor-Hi^R. A candidate could be chosen as the optimal output only when it incurs no violations of the two constraints. In this way, there may be more than one real outputs in a tableau and also there may be no ideal outputs in a tableau.

3.3 Diphthongs

A diphthong refers to the combination of two vowels at the nucleus position. Permissible diphthongs in S-Hakka are given below. In (32), the combination that is assigned with a check

“” is grammatical whereas with a cross “” is ungrammatical.

ii. Input:C

^F

^IL

LNuc

A

GREE

[di st ri but ed] [ -hi gh, - low ] [ + low ] [ -l ow ] [- ba ck] * [+ ba ck] O C P

[l abi al ] *C or -H i

 a.  * *

26 captured by *LONG/VOWEL in (33). Second, segments that are [-high] cannot appear together;

thus, OCP^V[-high] in (34) is proposed to prohibit [, , , , , ]. Third, [OCP[back]

& OCP[round]]^Vin (35), which requires two adjacent vowels not to have the same feature values on [back] and [round], is given to forbid [, , ,]. Notice that the constraints from (33) to (35) cannot be replaced by constraints that refer to the sonority hierarchy because S-Hakka includes both falling diphthongs (e.g. [, ]) and rising diphthongs (e.g. [, ]).

(33) *LONG/VOWEL: Assign one violation mark for every long or geminated vowel.

(34) OCP^V[-high]: Avoid a diphthong in which two vowels are both specified as [-high].

(35) [OCP[back] & OCP[round]]^V: A violation mark is given if and only if, within a diphthong, OCP[back] that avoids two segments to have identical feature values on [back] and OCP[round]

that avoids two segments to have identical feature values on [round] are simultaneously violated.

Following are several tableaux to show how acceptable diphthongs in S-Hakka are chosen through the constraint competition. In order to make analyses easier to be comprehended,

7 Whether the diphthong [] exists in S-Hakka is a controversial issue. No data in the present study show the combination so it is taken as ungrammatical. For the various researches discussing [] in S-Hakka, please refer to Chung (2004:161-164).

tableaux are separated according to the first vowel. Outcomes in tableaux (36) to (40) are able to correspond with table (32).

(36) Input: /V/