Chapter 1 Introduction
1.4 Organization of the Thesis
The thesis is organized as below. In chapter 1, the introduction of this thesis is given. In
4 According to Puyuma mythology, there are two kinds of origins: bamboo-origin and stone-origin. Nanwang dialect, aside from other dialects in Puyuma, belongs to the bamboo-origin.
5
chapter 2, some literature reviews concerning traditional and modified Optimality Theory are provided. In chapter 3, the analyses of consonants adaptations are illustrated, while in chapter 4, the analyses of Puyuma native syllable and the adaptations of loanwords syllable structures are specified. Finally, in chapter 5, concluding remarks concerning the issues tackled in this thesis will be addressed.
6
Chapter 2 Literature Review
In this chapter, we first introduce the concept of classical Optimality Theory in §2.1.
After introducing the traditional OT, we will review the concepts of Rank-Ordering Model of EVAL (ROE) in §2.2 and Local Conjunction in §2.3. These concepts are used to analyze data in this thesis.
2.1 An Introduction to Optimality Theory
Optimality Theory is proposed by Prince and Smolensky (1993, 2004) and McCarthy and Prince (1993). It is composed of notions such as Input, Output, Candidates, Generator, Evaluator and set of Constraints. In this theory, when an input is given, the generator will generate an infinite set of possible candidates; these candidates will be examined by the Evaluator. Within the Evaluator, a number of constraints, which are violable, are set in hierarchical sequence. The higher-ranked constraints will be put on the left-hand side of the tableau (like Constraint A, B and C in comparison with Constraint D in Tableau (2)), while the lower ones will be placed on the right-hand side (like Constraint B, C and D in comparison with Constraint A in Tableau (2)). The idea is that among these possible candidates, a candidate will be ruled out if it violates the higher-ranked constraint; the
7
survival one, even though it might violate the lower-ranked constraint, is the most harmonic one with respect to the specific constraint ranking (grammar).
Therefore, as can be seen in Tableau (2), Candidate A violates the highest constraint (Constraint A) once (indicated by “*”); it is then ruled out (marked by “!”). Candidate B is also ruled out because it violates the relatively higher constraint (Constraint B) than Candidate C does. Even though Candidate C violates some of the constraints, it is still selected to be the optimal (harmonic) candidate (indicated by “ ”) because the violated constraints are lower-ranked.
(2) An application of Optimality Theory
Input: Candidate Constraint A Constraint B Constraint C Constraint D
Candidate A *! *
Candidate B *!
Candidate C * *
After the competition, one candidate will be selected as the optimal candidate which can be attested in the language. The above idea concerning Optimality Theory can be illustrated by the figure cited from Kager (1999:8) in (3):
8
(3) Optimality Theory schema:
2.2 An Introduction to Rank-Ordering Model of EVAL (ROE)
The Rank-Ordering Model of EVAL (ROE) is conceptualized by Coetzee (2006) in order to explain variations. It is different from the traditional Optimality Theory in that when dealing with variations, the constraints of the traditional OT needs to be re-ranked so as to explain the variations; on the contrary, with the help of ROE, the relative ranking does not have to alter. Furthermore, the frequency of the occurrence of the variants can be properly predicted in the ROE model. To achieve this, there is a ‘cut-off’ between the constraints. The cut-off separates the tableau into two layers. If all the candidates violate the constraints above the cut-off, the selection is just like that of the traditional OT; the one that violates the lower-ranked constraint is the most harmonic one. However, if some of the candidates can pass through the cut-off, the violations of the constraints above the cut-off are fatal, while the
Input
Output Candidate a
Candidate b Candidate c Candidate d Candidate…
C1 >> C2 >>… Cn
9
violations of the constraints below the cut-off incur not fatality. The candidates that pass through the cut-off are all possible candidates except that their frequencies of the appearances are different. The frequencies of the occurrences of the variants are predicted by the constraints below the cut-off. When the candidate infringes the higher ranked constraint below the cut-off, it would be the less frequent variant. The application of ROE is illustrated in (4).
(4) The illustration of the application of ROE
Cut-off are four constraints as well. Constraints C1 and C2 are above the cut-off, while constraints C3
and C4 are below the cut-off. Both candidates (c-d) are ruled out since they incur violation of the constraints C1 and C2 which are above the cut-off. Candidates (c-d) infringe constraints C3
and C4 located below the cut-off, but the infringements are not severe enough to rule out both candidates. Thus, candidates (c-d) are two possible variants. The candidate (b) is predicted to be the less frequent variant because it violates the dominant constraint under the cut-off while
10
candidate (a) is the more frequent variant since it violates the less dominant constraint under the cut-off. To indicate the frequencies of the appearances, the number is subscripted to the pointing hand. For instance, 1 means that the pointed candidate is the more frequent one;
on the contrary, 2 indicates that the pointed candidate is less frequent.
2.3 An Introduction to Local Conjunction
Local Conjunction is proposed by Smolensky (1993, 1995, 1997). Apart from traditional OT analysis, Local Conjunction allows that two separate constraints can conjoin with each other. The conjoined constraint is violated if and only if the two component constraints are violated within the same domain simultaneously. Furthermore, concerning the hierarchy of the constraints, the conjoined one should always outrank the component constraints, while the two composing constraints freely rank with each other. To be more concrete, the concept of Local Conjunction, stated by Smolensky (1995) is given below in (5).
(5) The Local Conjunction of C1 and C2 in domain D (Smolensky 1995; Padgett 2002) a. C1&C2 is violated when there is some D in which both C1 and C2 are violated.
b. Universally C1&C2 >> C1,C2
11
The application of Local Conjunction is illustrated in (6). As we can see in tableau (6), four candidates compete with each other. The constraint [C1 & C2]D is the conjoined constraint which outdoes two of its element constraints, namely, C1 and C2. D is the specific domain; it could be a segment, a syllable, etc. CandidateD is ruled out by the conjoined constraint [C1 & C2]D because it incurs violations of the constraints C1 and C2 at the same time. On the contrary, CandidateC and CandidateB are rejected by constraints C1 and C2
respectively. The violation of the conjoined constraint does not happen because both CandidateC and CandidateB do not violation two composing constraints simultaneously. At last, the CandidateA is selected as the optimal output.
(6) The illustration of the application of Local Conjunction
Input [C1 & C2]D C1 C2
a. CandidateA
b. CandidateB *!
c. CandidateC *!
d. CandidateD *! * *
Since Local Conjunction involves the conjunction of two constraints in a certain domain, questions concerning the restrictions of component constraints and possible domains would naturally arise. However, even though many scholars (such as Kirchner 1996, Fukazawa & Miglio 1998, Lubowicz 2002, Nathan 2001) have proposed their viewpoints, little consensus has been reached. (Padgett 2002) Besides, as pointed out in Padgett, the
12
framework of local conjunction should be relinquished because it will cause WOW (“worst of the worst”) effect, i.e., ‘the combined effects amount to a separately rankable, dominant constraint.’ Thus, the strict domination of constraint in classical Optimality Theory cannot hold. However, in this study, we still need to utilize this framework to account for the adaptations of several Japanese sounds. Take the adaptation of // for instance, both [] and [u] are selected as attested outputs. It will be shown that the conjoined constraint [IDENT-IO(round)& *SCHWA]Seg plays a decisive role. Constraint IDENT-IO(round) alone will ban the selection of round vowel [u] when the input is the unrounded vowel // while constraint *SCHWA forbid the appearance of []. The use of [IDENT-IO(round)& *SCHWA]Seg can enable us to select both [u] and [] at the same time when the input is unrounded vowel //.
13
Chapter 3
Segmental Adaptations
3.0 Overview
As we have mentioned in the introduction, loanword phonology has something to do with two different languages, namely, source language (Ls) and borrowing language (Lb).
When words are borrowed from Ls, they are very likely to undergo some changes so as to fit into the phonology system of Lb while at the same time be as faithful to the inputs as possible.
In other words, the illicit structures (from segmental to suprasegmental) of the loanwords from Lb’s perspectives would be adapted so that these words can fulfill the phonological requirements of Lb. But in the following discussion, we can find that the loanwords in Puyuma are peculiar that some illegal sounds are retained without being replaced by other native phonemes of Puyuma. We will discuss this phenomenon at the end of this chapter.
Furthermore, concerning the data, in this research, if the percentage of the attested data is less than 5 percent, they will be treated as exceptions.
In the following few sections, we will see the adaptation of segments. In §3.1, the segment inventories of both Puyuma and Japanese would be introduced. In §3.2, the adaptation of vowels will be discussed. In §3.3, the adaptation of onset consonants will be examined. In §3.4, we will focus on the adaptation of coda consonants. In §3.5, a discussion
14
of the emergence of the illicit sounds will be provided.
3.1 An Introduction to Segment Inventories
Before we move on discussing the adaptation of vowels and consonants, in this section, we first need to review the segment inventories of both Puyuma and Japanese.
3.1.1 Puyuma Segment Inventory
This section begins by reviewing all the vowels in Puyuma, then, we will have a brief summary of the consonants in the language.
The vowel inventory of Puyuma, as can be seen in (7), contains only four vowels in total. In addition to the four vowels, Puyuma has an allophone vowel, namely the mid back rounded vowel [o]. Concerning the allophone [o], both Huang (2000) and Teng (2008) indicate that it is attained when the phoneme /u/ is in specific condition; that is, preceding velar nasal. The rewrite rule is given in (8).
(7) Puyuma vowel inventory
Front Central Back
High
Mid [e]
Low
Note: The alphabet in the brackets is the orthographic writing.
15
(8) // [] / ______ [+velar, +nasal]: High round vowel transforms into mid round vowel when preceding a velar nasal.
As for consonants, according to Teng, there are eighteen consonants in total in Nanwang Puyuma dialect, which can be seen in (9). In Nanwang Puyuma, all the stops, except for the glottal stop, make contrast in whether they are voiced or not. There is only one fricative in Nanwang dialect which is the voiceless alveolar fricative /s/. The other fricative sound /h/, based on Huang, is used only in loanwords, so it is not counted as one of the native sounds in Nanwang Puyuma. The voiced counterpart of /s/, namely //, though absent in Nanwang, can be observed in other dialects, which replaces // [dr] in Nanwang dialect. As mentioned earlier, other dialects have more fricatives since many stops in Nanwang dialect have been changed into fricatives in these dialects. Unlike all other Formosan languages, three consonants, retroflex voiceless stop // [tr], retroflex voiced stop // [dr] and retroflex lateral // [lr] are widely used in Nanwang dialect.
16
(9) Puyuma consonant inventory
Bilabial Alveolar Retroflex Palatal Velar Glottal
Stop -vd [tr] [’]
+vd [dr]
Nasal [ng]
Fricative (h)
Laterals [lr]
Trill
Glide [y]
Note: The alphabets in the brackets are the orthographic writing.
3.1.2 Japanese Segment Inventory
Next, we continue reviewing the segments of Japanese. Likewise, an introduction of vowels is prior to that of the consonants.
The vowel inventory of Japanese is illustrated in Table (10). Japanese has five vowels.
The high back vowel // is very marked across languages because it is [+back] but
unrounded. Furthermore, vowel length in Japanese is also phonemic. Long and short monophthongs lead to different meanings, for example, [] ‘aunt’ and []
‘grandmother.’
17
They are all adhered during the transcription of data.
18
(12) // nasal [α place] / _____ C [α place]: The // would become variant which agrees with the place of the articulation of the following consonant.
(13) Alveolar alternation:
a. // [s] / _____ : Voiceless alveolar stop changes into voiceless alveolar affricate when preceding high back unrounded vowel.
b. // [] / _____ : Voiceless alveolar stop becomes voiceless alveo-palatal affricate before high front vowel.
c. // [] / _____ : Voiceless alveolar fricative turns into voiceless alveo-palatal fricative if followed by high front vowel.
d. // [] / _____ : Voiced alveolar fricative surfaces as voiced alveo-palatal affricate when preceding high front vowel.
(14) /h/ alternations:
a. // [] / _____ : Voiceless glottal fricative is realized as voiceless labial fricative when followed by high back unrounded vowel.
b. // [] / _____ : Voiceless glottal fricative surfaces as voiceless palatal fricative when followed by high front vowel.
On inspecting phonological rules (12-14), people might wonder whether we should still treat the corresponding allophones [s, , , , , ] as phonemes because their occurrences are highly predictable. From Tsujimura’s viewpoint, they should maintain their phonemic
19
status since there are data showing that they can appear in some other circumstances, for instance, [tes] ‘chess’ and [] ‘woman,’ which are not predicted by the mentioned rules. But, at the same time, they are also the allophones of the phonemes /t, s, z, h/. To conclude, it is prevalent for languages to have sounds ‘restricted to certain environments.’
(Tsujimura 2007)
3.2 Adaptations of Vowels
As mentioned, in the previous section, there are four vowels in Puyuma, namely, /a/, /u/, // and /i/. As for Japanese, there are five vowels in the inventory in total, that is, /a/, //, /e/, /o/ and /i/; vowel length is phonemic. From the field work data, we can notice that loanwords containing vowels [a] and [i] will be borrowed faithfully without any changes; thus, they will be examined only to testify whether the constraint ranking that we have proposed for other vowels is still workable. To begin with, we would analyze the loanwords involving long vowels.
3.2.1 The Adaptation of Japanese Long Vowels
As we have discussed previously, unlike Puyuma, Japanese makes contrast in vowel length. Even though the vowel length in Puyuma sometimes leads to differences in meaning, for example, [kaiju] ‘there’ and [kaiju] ‘over there,’ the usage of this kind is highly
20
restricted; it only occurs for emphatic purpose and words which can undergo this procedure are very little. Thus, vowel length in Puyuma is not considered as phonemic.
After examining the fieldwork data, all words with long vowels borrowed from Japanese transformed into the corresponding short vowels. The percentage and examples are given below in (15-16).
(15) Adaptation of long vowels
Japanese source Puyuma correspondent Number Total Percentage
V： V 82 82 100%
(16) Examples of the long vowels adaptation to short vowels (V V)
Japanese Puyuma Gloss
a. ‘soda’
b. ‘factory’
c. ‘English’
d. ~ ‘motorcycle’
e. ‘score’
To explain the adaptation of long vowels, the following constraints are needed. They are listed in (17-19).
(17) NLV: No long vowel is allowed.
(18) DEP-IO: Output segments must have input correspondents. (‘No epenthesis.’)
21
(19) MAX-μ-IO: Input moras should have output correspondents. (‘No mora deletion.’)
The tableau dealing with the adaptation of long vowels is in (20). Candidate (20a) retains the original long vowel. Candidates (20b-c) use different strategies to avoid having long vowel. Candidate (20b) removes one of the moras from the long vowel; thus, the long vowel /o/ becomes the short one. Candidate (20c) inserts an additional vowel so as to take away one of the moras from the long vowel. Candidate (20d) inserts an additional consonant in the coda position. But in this research, we cannot state that the inserted coda would take over one of the moras from the long vowel. The reason is that the stress in Puyuma is always placed at word-final syllable. Therefore, we cannot know if the coda consonant bears weight or not. As there is no other evidence showing that the coda consonant in Puyuma is moraic, we would like to leave the issue for future study. Notice that the uncertainty of whether the coda carries weight does not have any effect on this research since the insertion of a coda consonant already violates DEP-IO. The candidate that violates MAX-μ-IO, namely (20b), is survived. Even though Puyuma disfavors the long vowels, the remedy that adds extra segments to solve the condition of long vowels is also disfavored by the language. Therefore, to correctly rule out candidates (20a & 20c-d), we should pose the constraints NLV and DEP-IO above MAX-μ-IO.
22
(20) The adaptation of long vowels to short vowels (V V)
/ i/ ‘bank’ NLV DEP-IO MAX-μ-IO
Moreover, they can appear in similar circumstances, as can be seen in (22) below, that is to say, they are not phonologically conditioned. Therefore, they are treated as variations. In order to explain the collected data in variation, we will use the ROE model (Coetzee 2006) to analyze them. Furthermore, we should note that the ROE model is widely used throughout this thesis because in most of the cases, we cannot find specific phonological environment that determines the selections of more than one output. Thus, they are all treated as variations.
(21) Adaptation of vowel /e/ into [i] and [e]
Japanese source Puyuma correspondent Number Total Percentage e
5 As we are not sure whether the coda is moraic, the violation mark of MAX-μ-IO is in parenthesis.
6 The exception is //] ‘Abraham.’
23
However, as we can notice below, the use of this constraint does not necessarily mean that the mentioned sound cannot emerge after adaptation.
24
(24) IDENT-IO(back): Assign one violation mark for every feature change of [back] from input to output correspondent.
(25) IDENT-IO(high): Assign one violation mark for every feature change of [high] from input to output correspondent.
(26) *e: Assign one violation mark for every mid front unrounded vowel [e].
The analysis is illustrated in (27). As can be seen in (27), all the vowels are put into consideration. Since the optimal outputs are [e] and [i], we should find a constraint to rule out all the other vowels. By inspecting the feature matrices in (23), we can find that the difference between [e, i] and the other vowels [, a, u, o] is that except for [e] and [i], others are [+back] vowels. So, in order to exclude the [+back] vowels, the constraint IDENT-IO(back) should outrank all the other constraints. Furthermore, due to the fact that both [e] and [i] are observed in the data, to choose them both, there should be a cut-off under the constraint IDENT-IO(back). Finally, from the percentage provided in (21), we know that vowel [e] is optimal and vowel [i] is suboptimal. To demonstrate this phenomenon, the constraint which vowel [i] violates, namely IDENT-IO(high), should outrank the constraint that vowel [e]
violates, i.e., *e, so that we can ensure the selection is correct.
25 examples are shown in (28) and (29) respectively.
(28) Adaptation of vowel /o/ into [u] and [o]
Japanese source Puyuma correspondent Number Total Percentage
o
26
Except for the above constraints mentioned in (24-26), other constraints should be provided so as to explain why vowel /o/ changes into [u] and [o] after borrowing. The relevant constraints are listed in (30-34) below.
(30) *o: Assign one violation mark for every mid back rounded vowel [o].
(31) IDENT-IO(low): Assign one violation mark for every feature change of [low] from input to output correspondent.
(32) *SCHWA: No schwa. (Cote 2006)
(33) IDENT-IO(round): Assign one violation mark for every feature change of [round] from input to output correspondent.
(34) [IDENT-IO(round)& *SCHWA]Seg: A violation mark is given if and only if, within a scope of segment, IDENT-IO(round) that requires the feature [round] in the output has a correspondent in the input and *SCHWA that disfavors the appearance of the vowel [] are simultaneously violated.
The adaptation of vowel /o/ to vowel [o] and vowel [u] are shown in tableau (35). From the statistics in (28), we can know that both mid back rounded vowel [o] and high back rounded vowel [u] are possible outputs. The appearance of [o] is more frequent than that of [u]. On the contrary, other vowels should be discarded. So, to exclude both front vowels [e]
27
and [i] from being selected, the constraint IDENT-IO(back) plays a crucial role. The constraint IDENT-IO(low) functions to rule out low vowel [a]. As for the elimination of mid central vowel [], the local conjunction constraint [IDENT-IO(round)& *SCHWA]Seg is employed. The
reason why we do not use the constraint *SCHWA alone, putting it higher than the cut-off, to exclude [] is that the wrong prediction would occur in explaining the adaptation of the high back unrounded vowel // in the next section. All of the constraints mentioned above, that is, IDENT-IO(back), IDENT-IO(low) and [IDENT-IO(round)&*SCHWA]Seg, should be placed higher than the cut-off so as to rule out the unattested vowels. Eventually, concerning the choosing of [u] and [o], owing to the fact that the only difference between [u] and [o] is the feature [high], to choose [o] as the optimal, we should place the constraint IDENT-IO(high) higher than the other constraint *o; both of the constraints should be placed lower than the cut-off.
28
are analyzed in this section. As for the deletion of //, since it has something to do with the syllabification; it will be discussed in chapter 4.
(36) Adaptation of vowel // into [u] and []
Japanese source Puyuma correspondent Number Total Percentage
29
(37) Examples of the // adaptation
Japanese Puyuma Gloss
a. ‘beer’
b. ‘Gospel’
c. ‘ticket’
d. ‘gas’
e. ‘slipper’
f. ‘underwear’
In dealing with the adaptation of //, the constraints proposed up to now are not
enough. Therefore, more constraints are introduced in (38-39). The analysis of the adaptation of // is illustrated in tableau (40).
(38) [IDENT-IO(round)&*o]Seg: A violation mark is given if and only if, within a scope of segment, IDENT-IO(round) that requires the feature [round] in the output to have a correspondent in the input and *o that disfavors the appearance of mid back rounded vowel [o] are both violated.
(39) *: Assign one violation mark for every high back unrounded vowel [].
30
(40) The adaptation of // to [u] and []
// * IDENT-IO(back) IDENT-IO(low) [IDENT-IO(round) Seg &*o] [IDENT-IO(round) Seg & *Schwa] *Schwa IDENT-IO(round) *o
a. e *!
b. i *!
c. 2 *
d. a *!
e. 1 u *
f. o *! * *
g. *!
As can be observed in tableau (40), seven candidates are competing with each other.
As can be observed in tableau (40), seven candidates are competing with each other.