Chapter 4 Optimality Theory Analysis
4.1 Substitution of Vowels
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
61
Chapter 4
Optimality Theory Analysis
In this chapter, we develop an Optimality Theory (Prince and Smolensky 1993/2004) analysis of the correspondence between Japanese sources and Taiwanese loans. The analysis will focuses on the dominant or the systematic substitutes.
4.1. Substitution of Vowels
The loanword data in the present corpus show that segment substitution is preferred over deletion. MAX is therefore needed to prevent segment deletion.
(1) MAX: Assign one violation mark for every input that has no correspondent in the output.
First, consider the substitution of Japanese long vowels. While Japanese short vowels contrast with long ones phonemically, such difference in length between short and long vowels is phonetically in Taiwanese. Japanese long vowels do not surface in Taiwanese, and this is enforced by vowel shortening. The two constraints concerning the modification of the vowel length are given in (2) and (3).
(2) *V:: Assign one violation mark for every lengthened vowel.
(3) IDENT[long]: Assign one violation mark for every output that differs from its input correspondent in the feature [long].
‧
IDENT[long] is dominated by *V:, or loanwords with lengthened vowels will be the optimal output. With this ranking hierarchy, consider tableaux (4) and (5).
(4) Input: [oʤ
i:sãɴ] ‘old man (Japanese)’
Output like [ozisãŋ] (with a nasalized vowel) is also a possible candidate. Since our main concern here is to show the competition between long and short vowels, it will be ignored at this moment. However, we will discuss the choice between a nasalized and an oral vowel in substituting for Japanese nasalized vowels in section 4.2.2.
As shown in (4) and (5), two Japanese sources [oʤ
i:sãɴ] ‘old man’ and [o
ʤisãɴ]
‘uncle’ which have different meanings lose their contrast in Taiwanese loans.14 They are both borrowed as [ozisaŋ]. In tableau (4), the most faithful output candidate (a) is ruled out as its violation of *V: is fatal. Candidate (4b) defeats candidate (4c) in that (4b) conforms to the higher ranked constraint MAX. Candidates (4c) and (5b) are in fact not legitimate forms in Taiwanese. They are listed in the tableaux to show the ranking between MAX and IDENT[long]. That is, to obtain the optimal output candidate (b), MAX must dominate IDENT[long]. In tableau (5), candidate (a) satisfies all the
14 When evaluating candidates in a tableau, the focus is on the segments in boldface. All tableaux in this chapter follow this principle.
‧
constraints and is chosen as the optimal output.15
Next, consider the substitution of Japanese short vowels. There are five short vowels, [a], [i], [ɯ], [e], and [o], in Japanese. Only [ɯ] is illicit in the Taiwanese sound system, and it is clear that this vowel will be substituted. Concerning the markedness constraints of segments, it is generally agreed that languages have
constraints that prohibit the occurrence of certain vowels, such as *ɯ, *e, *o, … Since [ɯ] is prohibited in Taiwanese, the segmental markedness constraint *ɯ, as shown in (6), will be ranked at the top, while the markedness constraints like *e and *o will be ranked as low as needed.
(6) *ɯ: Assign one violation mark for every vowel [ɯ].
However, constraint *ɯ says nothing about which substitute will be picked up for the source vowel in question. To decide which vowel shall be the substitute for the Japanese [ɯ], distinctive features are employed, as illustrated in (7).
(7) Feature specifications of vowels
a # i u e o ə ɯ
High - - + + - - - +
Low + + - - - - -
-Back - + - + - + + +
Round - - - + - + -
-It is suggested by some researchers (e.g., Chung 1996) that the value of [back] of the low vowel /a/ is empty underlyingly in Taiwanese. However, the present study deals with the surface form of segments. The low vowel /a/ has in fact two allophones [a] and [#] at the surface. The vowel /a/ is realized as [a] when it combines with the
15 Constraints that are adopted in the present study are defined as prohibiting only literal feature change (McCarthy 2008a). Based on this, candidate (4c) and (5b) do not incur violation of IDENT[long], since the target vowel is deleted in those candidates.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
alveolar coda nasal [n] or the labial coda nasal [m]. In contrast, /a/ will be realized as [#] when it combines with the velar coda nasal [ŋ].
As shown in chapter 3, Japanese and Taiwanese differ in the high back vowel.
Japanese high back [ɯ] is unrounded, while the Taiwanese high back [u] is a round one. To account for the change of roundness, IDENT[round] is adopted, as in (8). As for the Japanese vowels [a], [i], [e], and [o], they remain intact in general. This suggests that Japanese vowels and their substitutes in Taiwanese are all identical in the place of articulation. To account for the identity of place features, IDENT[Place] is adopted, as given in (9).
(8) IDENT[round]: Assign one violation mark for every output that differs from its input correspondent in the feature [round].
(9) IDENT[Place]: Assign one violation mark for every output that differs from its input correspondent in the features [high], [back], and [low].
IDENT[round] shall be ranked lower than IDENT[Place], or the vowel [u] will never be chosen as the substitute for [ɯ].
With the new proposed constraints above, first consider the substitution of the Japanese vowel [ɯ].
(10) Input: [ɾɯ
:ba:] ‘louver’
*ɯ MAX IDENT[Place] IDENT[long]] IDENT[round]
e. liba [back]! *
☞ f. luba * *
g. loba [high]! * *
h. lɯba *! *
i. lba *!
‧
The constraints *V: is omitted in this tableau to save space. Since illicit vowels never surface in Taiwanese, the constraint *ɯ is undominated in the ranking hierarchy. The faithful output (d) is ruled out, as its violation of *ɯ is fatal. Candidate (e) is
eliminated by the constraint MAX. Candidates (a) and (c) both violate IDENT[Place], making the output less closer to the input, and thus are eliminated. Therefore, the optimal output is candidate (b).
Next, consider the vowel [e]. Although it is a licit vowel in Taiwanese,
adjustment is still observed. This sound is intact when it is in an open syllable or ends with a glottal stop [ʔ]. But it will change to [ja] when preceding an alveolar nasal or an alveolar stop, since *en/t is not common in early phases in Taiwanese. However, many Taiwanese speakers pronounce [jan] as [jen], some even drop the glide [j], and the sound surfaces as [en] (Hung 2005). Chung (1996) pointed out that the [jan/t] to [jen/t] change is “quite universal in [the] Chinese language.” However, Taiwanese prohibits the backness feature of two vowels that combine to a diphthong from being the same. On that account, the form [jen/t] will further go through glide deletion, yielding to [en/t] in the surface to prevent violation of Taiwanese phonotatics.
The [ian] to [en] change is in fact not a necessary phonological process. However, according to Yip (1993), the adjustments of loanwords are minimal so that the source and its loan can be as close as possible. Thus, loanwords like [enʤiɴ] ‘engine’ or [enka] ‘traditional Japanese song’ are pronounced as [enzin] and [enka] by Taiwanese speakers, respectively. Consider tableau (11) below.
(11) Input: [ẽnʤĩɴ] ‘engine’
MAX IDENT[long]
☞ a. enzin
b. janzin *!
c. ŋzin *!
‧
The adjustment of the nasality is ignored here, but will be discussed in section 4.2.2.
Constraints such as IDENT[round] and IDENT[long] are inactive and are omitted in this tableau. Constraints such as IDENT[Place] and IDENT[round] are inactive and are
omitted in this tableau. The monophthong [e] changes to a diphthong [ja] in candidate (b), which violates IDENT[long], as the length of [ja] is longer than that of [e].
Candidate (c) is eliminated by MAX for segment deletion.16 Candidate (a) is the optimal output as it satisfies all the constraints.
Finally, consider the Japanese vowel [o]. Most occurrences of this vowel will remain intact, but it will change to [ə] as the second part of a diphthong in Taiwanese, that is, [jə]. The /io/ sequence is typically pronounced as [jə] by Taiwanese speakers.
To account for this, constraint *jo is adopted, as in (12).
(12) *jo: Assign one violation mark for every sequence [jo].
(13) Input: [ʃ
ok
ɯba] ‘place of work’*jo IDENT[Place] IDENT[round]
c. sjok*ubaʔ *!
☞ d. sjək*ubaʔ *
e. sjek*ubaʔ [back]!
The glide [j] in the outputs is a result of reconfiguration (that is, [ʃ] is split to [s] and [j]). The phenomenon of reconfiguration will be discussed in section 4.2.3. In this tableau, candidate (a) is eliminated as its violation of *jo is fatal. Candidate (c) is excluded by IDENT[Place]. The optimal output is candidate (b), though it violates IDENT[round]. This suggests that *jo has to be ranked higher than IDENT[round].
To summarize, we have examined the substitution of the vowels in loanwords
16 In this candidate, the nasal [n] in the first syllable changes to [ŋ] (or maybe to [m], and become a nucleus) to satisfy Taiwanese phonotatics.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
67
with the deployment of markedness as well as faithfulness constraints and rankings.
The faithfulness constraints concerning the feature identities are adopted since the output is assumed to be as close to the input as possible. The other markedness constraints are employed to prevent the output from having illegitimate segments or structures. The overall rankings of all of the constraints related to the vowel
substitutions are provided in (14).
(14) *ɯ, *V:, *jo >> MAX, IDENT[Place] >> IDENT[round], IDENT[long]