Chapter 4 Sixian-Hakka Loanword Phonology
4.1 Adaptations of Vowels
4.1.2 Adaptations of []
4.1.2.2 Back Rounded [ ] Adapted from Back Unrounded []
concern in the next section.
Conclusively, LS [] turns out to be LT [] when the preceding consonant is alveolar-dental. The acoustic properties of [] and [] are almost the same so it has not been confirmed that there is a language to distinguish between the two segments (Schwartz, Boe, Vallee & Abry 1997). Concerned with feature matrices, [] is identical with []. Significantly, based on the LT phonotactics, [] must co-occur with a preceding alveolar-dental, which results in two different conditions. If the input [] appears with a preceding alveolar-dental, doubtless the output will be []. On the contrary, if the input [] appears without a preceding alveolar-dental, what will happen? Next section, §4.1.2.2, provides explanations.
4.1.2.2 Back Rounded [] Adapted from Back Unrounded []
Apart from S-Hakka [] that can be the adapter of Japanese [], S-Hakka [] can be another adapted form as well. Then, why is not [] consistently replaced by only [] since, concerned with the feature values, [] and [] are identical to each other? The reason is contributed to the restrictive distribution of [] in S-Hakka. Because of the limited distribution of [], only after alveolar-dentals, there must be another vowel that can more widely occur and thus can replace []. The vowel that can widely occur to replace [] is []. Statistically, some data are provided in (58) and examples can be seen in (59).
42
(58) Statistics for the adaptation of [] after non-alveolar-dental sounds Japanese
(59) Examples for the adaptation of [] after non-alveolar-dental sounds Japanese loanwords S-Hakka correspondents Gloss
a platform
respectively; that is to say, [] and [] show the same degree of similarity to [] in terms of feature values. Then, why does [] rather than [] become the other adapter of []? By virtue of [] being preferred, it is supposed that IDENT-IO[back] outranks DEP-IO[labial]. An
11 The symbol “Ø” means “deletion” and two types are found. One type is the syllable deletion (e.g. []
→ [] ‘acrylic’; [] → [] ‘boat’). The other is that, within a syllable, the nucleus vowel is deleted and its onset becomes the coda of the preceding syllable (e.g. [] → [] ‘ice cream’;
[] → [] ‘bread’). The deletion is taken as an exception since no obvious phonetic contexts can be observed; furthermore, MAX-IO-Seg is attested to be dominant in the current work.
43
In (60), candidate (f) is disfavored owing to the illegal segment [] that violates the top-ranking . Candidate (60e) disobeys OK-σ because of the sequence []. Please remind that the distribution [] is very restrictive. In the LT, the segment [] is a derived vowel, not a phoneme, and its occurrence mainly lies in the feature [+distributed] spreading from an alveolar-dental onset to the following nucleus position, which results in the condition that []
only occurs after the alveolar-dentals (i.e. [s, ] in S-Hakka). The co-occurrence is captured by AGREE[distributed] and thus the appearance of [] in any candidate has to be judged by the constraint. Please refer to §3.1 for more details. Candidate (60d) incurs the violation of MAX-IO-Seg since the input [] has no corresponding output. With respect to candidates (60a) to (60c), it is apparent that only when IDENT-IO[back] and IDENT-IO[high] outranks DEP -IO[labial] can candidate (60a) be selected as the most harmonic output.
Overall, both S-Hakka [] and [] are the adapted forms of Japanese [] and the difference between the two adapters lies in the phonetic contexts where they can occur. When the LS []
enters into the LT, it transforms to [] with a preceding alveolar-dental onset but it transforms to [] elsewhere. Besides, when compared with [], [] is carefully chosen as the other adapter, even though both segments differ from [] for only one feature. The segment [] is chosen rather than [] mainly because IDENT-IO[back] ranks higher than DEP-IO[labial] does.
4.1.3 Summary for the Adaptions of Vowels
In §4.1, issues being dealt with are related to vowels. Several adaptations that transform from Japanese inputs to S-Hakka corresponding outputs are found and listed in (61). Regarding the constraints, all constraints adopted in §4.1 and their ranking are given in (62).
44
(61) Summary for the adaptions of vowels
Japanese sources S-Hakka correspondents
§4.1.1 V V
§4.1.2.1 (after alveolar-dental)
§4.1.2.2
(62) Constraint ranking for the adaptations of vowels:
OK-σ, *MAX-IO-SegIDENT-IO[back], IDENT-IO[high] >> DEP-IO[labial], IDENT-IO[long]
With regard to loanword phonology, the results obtained so far are congruent with the perspective from Kenstowicz (2005), Silverman (1992) and Yip (2006). These researchers all claim that loanword adaptations include both the transformation of non-identical sounds and the transformation in order to obey the phonotactics of LT. The best proof for the two types of transformations can be detected through the adaptation from [] to either [] or []. The sound [] is non-native in the LT so doubtless [] transforms to a segment that is native in the LT and that is []. The vowel [] transforms to [] because both of them are identical in terms of feature matrices. The adaptation from [] to [] evidences the change from non-native to native sounds. Then why there is another adapter []? LT phonotactics plays an important role. LT
phonotacticshighly restricts the distribution of [], only after alveolar-dentals, and thus []
which legally is able to occur in more different phonetic contexts appears. The adaptation from [] to [] proves not only the change from non-native to native sounds but also the compliance with the LT phonotactics.
4.2 Adaptations of Onset Consonants
In this section, attentions will be on onset consonants. S-Hakka allows every consonant in its inventory to appear at the onset position; however, it only permits six consonants, [, , ,
, , ], to occupy at the coda position. For this reason and to make analyses more clear and
45
effective, onset consonants will be discussed in §4.2 and coda consonants will be presented in
§4.3. In §4.2, onsets can be classified based on the manner of articulation (i.e. stop, fricative, affricate, nasal, flap and glide). Stops, fricatives, affricates and flaps will be discussed in §4.2.1,
§4.2.2, §4.2.3 and §4.2.4 respectively. On the other hand, by virtue of no transformations regarding onset nasals and glides, no analyses are presented.
4.2.1 Stops
4.2.1.1 Voiced Stops
It is certain that Japanese voiced stops [, , ] must undergo some degrees of transformations during the loan process because S-Hakka has no voiced stops. That means the LT prohibits the appearance of voiced stops, which is captured in (63). With respect to what voiced stops become throughout the process of loaning, statistics and a few examples are provided in (64) and (65) respectively.
(63) *VOICED/STOP: Assign one violation mark for every voiced stop.
(64) Statistics for the adaptations of voiced stops Japanese
46
As revealed in (64), every voiced stop has various correspondents in the outputs and statistically the correspondents account for different percentages. Coetzee (2006) proposes ROE mainly to account for variations in languages. By taking advantage of ROE, different correspondents can be logically explained. The present study discusses the three voiced stops and their correspondents one by one in the following three paragraphs.
(65) Examples for the adaptations of voiced stops
Japanese loanwords S-Hakka correspondents Gloss
a bonus constraints are put forward in (67) and (68) so as to cast away impossible candidates and the optimality-theoretic analysis is presented in (69).
(66) Feature matrices for [, , , , , ]
47
(67) [IDENT-IO[continuant] & IDENT-IO[voice]]Seg: A violation mark is given if and only if, within a scope of segment, IDENT-IO[continuant] that requires correspondent segments in input and output have identical values on [continuant] and IDENT-IO[voice] that requires correspondent segments in input and output have identical values on [voice] are simultaneously violated.
(68) [IDENT-IO[strident] & IDENT-IO[voice]]Seg: A violation mark is given if and only if, within a scope of segment, IDENT-IO[strident] that requires correspondent segments in input and output have identical values on [strident] and IDENT-IO[voice] that requires correspondent segments in input and output have identical values on [voice] are simultaneously violated.
(69) Variations for the input []
Input: [
*V
OICED
/S
TOP[I
DENT-I O [c ont inua nt ] & I
DENT-I O [voi ce ]]
Seg[I
DENT-I O [s tr ide nt ] & I
DENT-I O [voi ce ]]
SegI
DENT-I O [S G ] I
DENT-I O [s onor ant ] I
DENT-I O [c ont inua nt ] I
DENT-I O [na sa l] I
DENT-I O [s tr ide nt ] I
DENT-I O [v oi ce ]
a. 1 *
Concerned with candidate (69f), it is excluded because [] violates *VOICED/STOP which is above the cut-off and hence the violation is critical. Regarding candidate (69a), the present study assumes that IDENT-IO[voice] ranks not only below the cut-off but also at the bottom so the candidate wins out. The assumption is able to predict that [] can most frequently appear among other variants since it incurs the least significant violation. For candidates (69b) to (69d), they only violate IDENT-IO[F] constraints which are below the cut-off so they still can be found as the real outputs. Besides, based on the ranking of the various IDENT-IO[F] constraints, the frequency of occurrence for various correspondents can be predicted and the prediction agrees with the statistics shown in table (64). Importantly, the ranking of IDENT-IO[F] constraints function properly for other analyses in the following sections. As for candidate (69e), it is cast
48
away by both [IDENT-IO[continuant] & IDENT-IO[voice]]Seg and [IDENT-IO[strident] &
IDENT-IO[voice]]Seg because feature values of [] differs from those of [b] on [continuant], [strident] and [voice].
Second, with regard to the voiced alveolar stop [] and its correspondents, their feature values are listed in (70) and the analyses based on OT are given in (71).
(70) Feature matrices for [, , , ]
In (71), the most faithful candidate (e) is ruled out by *VOICED/STOP which is not violable.
The most harmonic candidate is (71a) because it violates IDENT-IO[voice] which is lowest ranked. For candidate (71a), its least insignificant violation among other candidates results in the highest percentage of occurrence for itself, which is the same as what has been shown in table (64). In addition to (71a), both candidates (71b) and (71c) are still possible, though less
49
favored. Candidates (71b) and (71c) are not discarded because their violation marks also land below the cut-off. For candidate (71d), it is excluded for the reason that input [] is specified [-continuant] and [+voice] but ] is specified [+continuant] and [-voice], which leads to the violation of [IDENT-IO[continuant] & IDENT-IO [voice]]Seg. The constraint ranking of IDENT -IO[F] constraints correctly predicts the outcomes.
Third, concerned with the voiced velar stop [] and its correspondents, the feature matrices are shown in (72) and an application of OT is given in (73). As can be seen in (73), the most serious violation of *VOICED/STOP makes candidate (c) ruled out. Candidate (73a) accounts for the majority since its violation is less fatal than candidate (73b) is.
(72) Feature matrices for [, , ]
To conclude, the various S-Hakka correspondents for each Japanese voiced stops can be explained by adopting ROE, a revised OT framework. Generally, voiced stops from the LS
enter the LT in the forms of their voiceless counterparts, and thus IDENT-IO[voice] is at the price. Furthermore, compared with other correspondents, voiceless stops are more well-formed so statistically they appear more frequently in the real language use.
50
4.2.1.2 Voiceless Stops
At the first glance over voiceless stops, it is confusing whether Japanese voiceless stops [, , ] would undergo transformations when they are borrowed into S-Hakka in which there are two sets of voiceless stops, unaspirated [, , ] and aspirated [, , ]. It is likely that the transformation would not happen because in loanword phonology it is said that sounds existing in both LS and LT undergo no adapted process. That means it can be predicted that Japanese [, , ] would not go through any transformations when they enter S-Hakka.
Nevertheless, the prediction turns out to be wrong. What the truth is the transformation takes place at certain phonetic contexts. Statistics and some examples are provided in (74) and (75).
(74) Statistics for the adaptations of voiceless stops Japanese
sources
S-Hakka
Correspondents Word-initial onset Word-middle onset
Number Total Percentage Number Total Percentage
Two things can be observed through (74). First, with respect to the LS voiceless stops, they will all transform to their aspirated counterparts, if they occupy the word-initial position. Second, if the LS voiceless stops are at the word-middle onset position, each of them will have two correspondents that account for different percentages.
13 The data are [] → [] ‘concrete’ and [] → [] ‘karat.’
14 The onset [] becomes coda [] of the preceding syllable and the vowel following the original onset [] is deleted. Data are [] → [] ‘boxing’; [] → [] ‘restaurant’; []
→ [] ‘post office’ and [] → [] ‘toast.’
51
(75) Examples for the adaptations of voiceless stops Japanese
j a kind of Japanese traditional clothes
k airplane
l Sake (a kind of Japanese traditional wine)
Speaking of Japanese word-initial voiceless stops, their acoustic property has to be taken into consideration. Voice onset time (VOT) of Japanese word-initial voiceless stops are controversial. Lisker & Acramson (1964) demonstrate a cross-linguistic research and categorize stop sounds into three types based on VOT. Their findings are shown in (76). VOT ranging from zero to twenty-five milliseconds is the quality of voiceless unaspirated stops, ranging from sixty to one hundred milliseconds is the quality of voiceless aspirated stops, and ranging from zero to minus twenty-five is the quality of voiced stops.
(76) VOT values of stop sounds (Lisker & Acramson 1964)
Stops Voiceless unaspirated Voiceless aspirated Voiced
VOT values 0~25 60~100 -25~0
Many studies (Han 1992, 1994, Homma 1981, Riney, Takagi, Ota & Uchida 2007) all report that Japanese word-initial voiceless stops usually are “more aspirated.” The VOT values of Japanese and S-Hakka word-initial voiceless stops are provided in (77). VOT values of Japanese word-initial voiceless stops are intermediate between the criteria of unaspiration and aspiration set up by Lisker & Acramson. Therefore, the thesis hypothesizes that Japanese
word-52
initial voiceless stops are [] owing to their more aspirated property. Based on the hypothesis, why Japanese word-initial voiceless stops all transform to S-Hakka [] but not [] can be accounted for. Note that Japanese word-initial stop in the form of []
are hypothesized by the present work, and they become [] when entering S-Hakka, so no OT tableau is provided because no segmental change occurs.
(77) VOT values of Japanese and S-Hakka voiceless stops at the word-initial onset position Word-initial onsets word-middle onset position has two correspondents in the LT, aspirated and unaspirated, and the correspondents account for different frequencies of occurrence. Nevertheless, by observing (74) more carefully, it can be found that the transformations do not show a consistency in percentage terms. Regarding the majority, [] and [] mostly are still [] and [], while []
largely becomes [].VOT values of Japanese voiceless stops at the word-middle onset position, as shown in (78), are unanimously grouped into the unaspirated category according to the criteria set up by Lisker & Acramson (1964). For the segments in question, optimality-theoretic analyses are provided in (79).
(78) VOT values of Japanese and S-Hakka voiceless stops at the word-middle onset position Word-middle onsets
[] [] []
Japanese (Homma 1981) 7 16 24
53
As shown in (79), IDENT-IO[SG] is below the cut-off so two possible outputs for each input are correctly predicted. However, which output is more well-formed is hard to tell. The difficulty lies in the fact that, in terms of percentage, voiceless stops as word-middle onsets are unable to show a similar pattern of transforming when they are borrowed into the LT. In comparison of results in percentage terms in table (74) with those in optimality-theoretic analyses in tableau (79), the hierarchy of output well-formedness is correctly predicted in (79i) and (79ii) where candidate (a) is more well-formed than candidate (b) is; however, the outcome in (79iii) is different from what has been presented in (74). Based on the statistical information provided in (74), [] is the most well-formed output and the second ideal output is []. On the contrary, based on (79iii), the most ideal output is [] and [] is chosen as the second one.
Because the frequency of occurrence differs from statistics in (74) to ROE prediction in (79iii), bombs “” are assigned. This question about the well-formed hierarchy of possible outputs may be dealt with by more data that can be collected. The present study leaves this question for further study and provides a very similar phenomenon found in the study of Japanese loanwords in Taiwanese (Nien 2009) in the next paragraph as a reference.
(79) Variations for the voiceless stops at the word-middle onset position
Taiwanese, like S-Hakka, has two sets of stop sounds in its phonetic inventory. One set of
54
it is [, , ] and the other one is [, , ]. When Japanese loanwords enter Taiwanese, concerned with voiceless stops and their distributions, statistics are given in (80).
(80) Statistics for the adaptations of voiceless stops from Japanese to Taiwanese (Nien 2009:45) Japanese
sources
Taiwanese
Correspondents Word-initial onset Word-middle onset
Number Total Percentage Number Total Percentage contrast, Japanese word-initial voiceless stops all transform to their aspirated counterparts when they are borrowed into S-Hakka. On the other hand, for word-middle voiceless stops as onsets, the conditions for alveolar and velar stops are like those found in S-Hakka where []
largely is still [] and [] mainly becomes []. As for word-middle onset [], when it enters Taiwanese, it largely transforms to [], which is different from what has be found in S-Hakka.
However, please note that the percentages for [] and [] at the word-middle onset are quite even in Taiwanese.
Summing up results of the discussion so far, the present study finds the adaptations of voiceless stops have to take the phonetic contexts into consideration. At the word-initial onset position, Japanese voiceless stops which are hypothesized as [, , ] based on their “more aspirated property” transform to S-Hakka [, , ]. On the other hand, at the word-middle onset position, each LS voiceless unaspirated stop has two possible outputs, aspirated and unaspirated, in the LT; however, the transformations do not show a consistent pattern. Both []
55
and [] largely remain unchanged whereas [] tends to change into []. The result about voiceless stops at word-middle onset position is not easy to explain and thus more data and researches are required.
4.2.2 Fricatives
Japanese has six fricatives: [,,, ,, ]. Both [] and [] are legal in S-Hakka so they are loaned without any transformational processes and thus they are no longer discussed.
However, since there are still four ungrammatical segments [,,, ] left, their adapters and how they are adapted will be the concerns of this section. The arrangement of this section is as follows. The adaptation of the voiced alveolar fricative [z] is discussed in §4.2.2.1, the voiceless alveolar-palatal fricative [] in §4.2.2.2, the voiceless palatal fricative [] in §4.2.2.3, and the voiceless bilabial fricative [] in §4.2.2.4.
4.2.2.1 Alveolar-dental [s] Adapted from Alveolar [z]
Japanese alveolar fricative [] is not a member in S-Hakka inventory so as the loan process occurs, input [] cannot be realized in the output, which is suggested in (81). Statistics and several examples are provided in (82) and (83) respectively. As can be seen in (82), [] can be adapted as either [] or [] but the former one accounts for the majority. The current research still adopts ROE (Coetzee 2006) to explain the variations.
(81) *: Assign one violation mark for every voiced alveolar fricative [].
(82) Statistics for the adaptation of []
Japanese
56
(83) Examples for the adaptation of []
Japanese loanwords S-Hakka correspondents Gloss
a size
b fuse wire
c small wine glass
Before the application of ROE in (85), the segments [, , ] are compared in terms of feature values in (84). In (85), it is obviously that because all IDENT-IO[F] constraints are below the cut-off, candidates (85a) and (85b) are possible outputs in the LT. Besides, IDENT-IO[voice]
is lowest ranked so candidate (85a) is correctly predicted to appear most frequently. As for candidate (85c), it incurs the violation of * that is fatal and thus it is ruled out.
(84) Feature matrices for [, , ] [], [] is also a possible output. Concerned with the frequency of appearance, [] appears more frequently than [] does because both IDENT-IO[distributed] and IDENT-IO[del rel]
outrank IDENT-IO[voice] and all the three constraints are below the cut-off.
57
4.2.2.2 Alveolar-palatal [] Adapted from Palatal-alveolar []
Japanese [] is an allophone derived from the phoneme // when the phoneme precedes []. The segment [] is absent in S-Hakka; therefore, the adaptation of [] is inevitable during the loan process. To capture that segment [] is not present in the LT, the constraint is suggested in (86). Statistics and a few examples are provided in (87) and (88) respectively.
(86): Assign one violation mark for every voiceless palatal-alveolar fricative [].
(87) Statistics for the adaptation of []
Japanese
(88) Examples for the adaptation of []
Japanese loanwords S-Hakka correspondents Gloss
a hello [telephone use]
b tomorrow
c a brain is short-circuited
d sashimi (a kind of Japanese food)
e yummy
As revealed in (87), the adapted form of [] is []. It is worth noting that both Japanese [] and S-Hakka [] are allophones, and both of them are derived when Japanese // and S-Hakka //
are followed by []. In other words, Japanese [] and S-Hakka [] are fixed phonetic sequences.
The fact that Japanese [] and S-Hakka [] have to occur with a following [] is a key factor when the present study analyzes the adaptation of [].
Prior to OT analysis be provided, what feature being able to distinguish [] from [] has to be decided because the contrast between palatal-alveolar (e.g. Japanese [, , ]) and alveolar-palatal (e.g. S-Hakka [, , ]) is very trivial. These two natural classes (i.e. palatal-alveolar and palatal-alveolar-palatal sounds) are adjacent to each other in terms of place of articulation.
58
The place of articulation for palatal-alveolar sounds is closer to alveolar ridge while for alveolar-palatal sounds is closer to hard palate. Ladefoged & Maddieson (1996:138) even describe “palatal” as “palatalized palatal-alveolar.” In general, to articulate alveolar-palatal, there is a larger portion of the tongue body to approach hard palate. As a result, in the current work, the feature [high] would be used to discriminate between palatal-alveolar and alveolar-palatal, which is in line with Chung (2004).
The distinctive feature between [] and [] has been confirmed (i.e. [high]) and the feature values of sounds that are similar to the input [] are compared in (89). An application of OT framework is given in (90).
In (90), candidate (e) is the most faithful one but it violates *which is serious and fatal. For candidate (90d), the sequence [] violates *[+anterior, +strident]/and thus is discarded. In
59
fact, because of the predictable vowel [] in the fixed sequence [], [] is impossible to be the adapted form of []. To obey *[+anterior, +strident]/, when alveolar-dentals (i.e. [s,, ]) are in front of vowel [], they transform to their phonetic variants, or allophones, that are alveolar-palatals (i.e. [, ,]). Candidate (90c) brings out the violation of MAX-IO-Seg because [] has no correspondent in the output. Candidate (90b) is disciplined by IDENT -IO[back] and DEP-IO[labial] owing to the transformation from [] to [], and by IDENT -IO[anterior] due to the adaptation from [] to []. Candidate (90a) violates DEP-IO[high] for []
turning out as [], and it wins out because DEP-IO[high] ranks lower than IDENT-IO[back].
In summary, the impossibility of [] as the real output is because the violation of OK-σ must be incurred. On the other hand, the candidate [] serves as the most harmonic output because it brings out the least fatal violation. Importantly, both the input (i.e. Japanese []) and the output (i.e. S-Hakka []) are allophones in the languages, and they are derived when the LS // and the LT // precede []. In other words, both [] and [] must co-occur with a following [i] so no vowel adaptation is necessary, which leads to fewer violation marks related to vowel changes.
4.2.2.3 Glottal [] Adapted from Palatal []
The voiceless palatal fricative [] is an allophone of /h/ in Japanese. When /h/ precedes [], [] can be seen. That is to say, [] is a sound sequence which is fixed and unalterable. The segment [] is n in S-Hakka so undoubtedly it would go through some transformational processes once it enters the LT. The ungrammaticality of [] in the LT is stated in (91).
(91): Assign one violation mark for every voiceless palatal fricative [].
60
Statistics are given in (92). As shown in (92) and all examples in (93), [] is loaned in the form of [] without any exceptions.
(92) Statistics for the adaptation of []
Japanese
(93) Examples for the adaptation of []
Japanese loanwords S-Hakka correspondents Gloss
a airplane
b cypress
c fuse wire
The adapted form of [] is []; however, regarding feature values, segments [ , ] are all similar to []. Feature specifications for [, , , ] are given in (94) and these similar sounds are taken into consideration when OT is applied in (95).
(94) Feature matrices for [, , , ]
61
(95) Input: [] → Output: [] ‘airplane’
O K -σ M
AX-I O -S eg D
EP-I O [hi gh ] M
AX-I O [a nt er ior ] M
AX-I O [di st ri but ed ] I
DENT-I O [a nt er io r] I
DENT-I O [ st ri de nt ]
a. * *
b. *! *
c. *!
d. *!
((46ix.)*[+anterior, +strident]/) * *
e. *!
In (95), candidate (e) is excluded owing to the ungrammatical segment []. For candidate (95d), it is not chosen as the optimal output because the vowel [] from the unchangeable input sequence [] makes the phonetic sequence [] cast away by *[+anterior, +strident]/. As for candidate (95c), [] has no correspondent in the output, which does not reach the requirement of MAX-IO-Seg. Then, the reason why candidate (95a) is selected rather than (95b) as the
In (95), candidate (e) is excluded owing to the ungrammatical segment []. For candidate (95d), it is not chosen as the optimal output because the vowel [] from the unchangeable input sequence [] makes the phonetic sequence [] cast away by *[+anterior, +strident]/. As for candidate (95c), [] has no correspondent in the output, which does not reach the requirement of MAX-IO-Seg. Then, the reason why candidate (95a) is selected rather than (95b) as the