Chapter 2 Literature Review
2.4 Optimality Theory
2.4.4 Local Conjunction
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
24
at the lexical level. However, if they fail to associate with an actual term, they then search at the postlexical level where they may or may not successfully associate with an actual term. The separate operations at lexical and postlexical levels show the spirit of Stratal OT.
2.4.4 Local Conjunction
Local conjunction, which is proposed by Green (1994) and Smolensky (1995), combines two constraints into one. Smolensky (1995) proposes the formalized description, as in (32).
(32) Local conjunction
The local conjunction of C1 and C2 in domain, C1 and C2 is violated when there is some domain of type D in which both C1 and C2 are violated.
(Smolensky 1995:10)
Local conjunction rules out the worst of the worst output, which is called the WOW effect. The constraint conjunction is violated only when both of its members are violated. For example, the conjoined markedness constraints prohibit the worst of the worst marked output. Morris (2002) and Łubowicz (2002, 2005) indicate the need to conjoin markedness and faithfulness constraints. The concept is that the conjoined markedness member is activated only when the faithfulness member is violated. Wee (2002) proposes the faithfulness-faithfulness conjunction for tone. The constraint ID-R and ID-C is violated when both the tonal register and the tonal contour change. Hsiao (2015) proposes that both faithfulness-faithfulness and markedness-faithfulness conjuncts are required for the complex tonal chain shifts in Taiwanese.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
25
This study proposes a conjoined constraint, which is comprised of two markedness constraints, namely NoShare-σ and *CCCODA. NoShare-σ forbids syllables from being shared by two beats, whereas *CCCODA eliminates coda clusters. The conjoined constraint, NoShare-σ & *CCCODA is violated only when a syllable with a coda cluster is shared by two beats.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
26
Chapter 3
Language-to-Music Mapping: Onset Cluster
3.1 Introduction
This chapter compares the linguistic mapping and the language-to-music mapping in onset clusters. Segmental changes are observed from Mandarin-accented English in reading and singing. I posit the mapping schema in (33).
(33) Language-to-music mapping: segment Linguistic input
Production grammar Linguistic output
Perception grammar
Musical inputProduction grammar
Musical outputAs proposed in (33), the linguistic output is the English word pronounced by Mandarin speakers. This linguistic output is then perceived as the musical segmental inputs. The perceived musical input is then produced as the musical output, which is sung by the speakers. Segmental changes are observed from the linguistic mapping and language-to-music mapping.
There are two linguistic output variants of monosyllabic syllables with onset clusters. Linguistic output1 preserves the onset cluster and the syllable number
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
27
remains one. Linguistic output2 involves vowel insertion so that the syllable number becomes two. These linguistic outputs are respectively assigned to one and two musical beats. This chapter investigates how musical beat assignment affects segmental changes and how the prosodic word conditions musical beat assignment in the musical input.
3.2 Data Design
The database is designed to compare segmental changes of onset clusters in the linguistic mapping and the language-to-music mapping. The Mandarin informants include two males and three females, aged between 59 and 72, and are of senior secondary to higher education in Taiwan. All of them have learnt English for at least six years. In order to examine segmental changes between the linguistic output and the musical output, the informants are asked to read and sing the assigned target words. The procedures of reading in step1 and singing in step 2 are introduced in 3.2.1 and 3.2.2 respectively.
3.2.1 Step 1: Reading
The informants read the target words on a piece of paper in step 1. They can see and read the words to ensure that their linguistic output comes from their own input instead of from other peoples’ linguistic output.
The target words include onset clusters with different combination of consonants on the sonority scale.1 The sonority sequencing principle (SSP) is introduced by Sievers (1881) and Jespersen (1904). They propose that in the onset cluster a more
1 The collected onset clusters do not reveal clear correlation to the SSP, which, however, effective on coda clusters, as will be discussed in Chapter 4.
‧
sonorous consonant stands closer to the syllable peak than one that is less sonorous.
Carr (1993), and Broselow and Finer (1991) list stop and fricative as separate classes.
As in (34), stop is the least sonorous, while glide is the most sonorous.
(34) Sonority scale
The data in (35) are examples of target words with onset clusters. The onsets in (35) conform to SSP. In other words, the sonority of the first consonant in the onset cluster should be lower than that of the second consonant, which is closer to the syllable peak. For instance, the onset cluster [bl] is the combination of a less sonorous stop followed by a more sonorous liquid.
(35) Target words with onset clusters
SSP Example target words
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
29
In step 1, the informants read the target words, such as in (36-39).
(36) Target word: blue Output: [blu], [bulu]
(37) Target word: play Output: [phleɪ], [phuleɪ]
(38) Target word: green Output: [grin], [gurin]
(39) Target word: class Output: [klæs], [kəlæs]
There are two kinds of linguistic outputs. One is without vowel insertion and the other is with vowel insertion. The variation in pronouncing the onset clusters is schematized as (40).
(40) Linguistic input Linguistic output
(blu) Linguistic output1 (σ) /blu/ ‘blue’
(bulu) Linguistic output2 (σσ)
As in (40), the prosodic words are formed in the output. For example, the linguistic input /blu/ yields either (blu) or (bulu) in the output. Linguistic output1 preserves the onset cluster and remains monosyllabic, whereas linguistic output2 inserts a vowel into the cluster and becomes disyllabic.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
30
(41) Statistics of the linguistic outputs
Table (41) shows the linguistic input to linguistic output mapping. There are totally 59 syllables with onset clusters. Fifty-one, or 86%, of them surface as one syllable; eight, or 14%, of them insert a vowel, changing the syllable number to two. The percentage shows that the informants tend to correctly pronounce syllables with onset clusters.
3.2.2 Step 2: Singing
In step 2, the informants sing the linguistic outputs they produce in step 1. As mentioned in 3.2.1, there are two kinds of linguistic outputs, namely, linguistic output1 and linguistic output2. Each of the linguistic outputs is respectively assigned with one and two beats in the language-to-music mapping. Take (blu) for example.
(42) Linguistic output1 (σ) to singing mapping Linguistic output1:(blu)
As shown in (42), whether linguistic output1, (blu), is assigned with one musical beat, as in (42a), or with two musical beats, as in (42b), the singing outputs surface as [blu], where no vowel insertion occurs. The prosodic word structure is removed in the singing output.
2 Some of the target words are with coda, for example, green. However, since coda is not the focus in this chapter, only onsets and nuclei are listed.
Linguistic input Linguistic output1 Linguistic output2 Total CCV2 (CCV) (51/86%) (CVCV) (8/14%) 59 (100%)
a. Singing output b. Singing output blu
∣
q
blu ╱╲
q q
‧
(43) Statistics of the linguistic output1 (σ) to singing mapping
Linguistic output1 Singing output (q) Singing output (q q)
(CCV)
The table in (43) shows that there are totally 51 onset clusters that are produced as linguistic output1. These onset clusters are respectively assigned with one and two musical beats. When they are associated with one beat, 50, or 98%, of them are sung as monosyllabic CCV. When they are associated with two beats, still 100% of them are sung as monosyllabic syllables.
Consider the output2 mapping, as in (44).
(44) Linguistic output2 (σσ) to singing mapping Linguistic output2: (bulu)
When (bulu) is assigned with one musical beat, as in (44a), the singing output is truncated as [blu]. When the same linguistic outputis assigned with two musical beats, as in (44b), the singing output is still (blu). The structure of the prosodic word is
‧
(45) Statistics of the linguistic output2 (σσ) to singing mapping
Linguistic output2 Singing output (q) Singing output (q q)
(CVCV)
There are totally eight onset clusters that are produced as linguistic output2 in the database. Each of them is assigned with one and two musical beats respectively. As in table (45), when they are associated with one beat, seven, or 87.5%, of them are sung as CCV. Only one, or 12.5%, of them is sung as CV. None of them are sung as disyllabic syllables. When they are associated with two beats, still seven, or 87.5% of them are sung are sung as monosyllabic syllables.
In brief, two patterns are in order. First, whether the monosyllabic output1 is associated with one or two musical beats, there is no vowel inserted to resolve the onset cluster. Second, whether the disyllabic linguistic output2 is assigned with one or two musical beats, the inserted vowel in reading is deleted in singing.
3.3 Language-to-Music Mapping
I propose a model for the language-to-music mapping, as illustrated in (46).
Production and perception grammars are shown in the linguistic mapping and language-to-music mapping.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
33
(46) Segmental Model for the Language-to-Music Mapping:
Linguistic input I
Linguistic output O1 O2
Musical input I1 I2 I 3 I4
♩ ♩♩ ♩ ♩♩
Musical output O1 O2 O3 O4
The input on the top is a monosyllable with an onset cluster. The production grammar is shown in the linguistic input to output mapping. There are two linguistics outputs, namely, linguistic output1 and linguistic output2. Linguistic output1 is with monosyllable where no vowel is inserted. Linguistic output2 is pronounced with two syllables where a vowel is inserted to prevent complex consonants. Both of the linguistic outputs are parsed into prosodic words.
The mapping from the linguistic output to the musical input demonstrates a need for the perception grammar. The linguistic outputs are perceived as the musical inputs, and each of them are assigned with one and two musical beats respectively. The prosodic word structure, which may affect beat assignment, is formed in the musical input. Whether linguistic output1 is assigned with one or two beats, the musical segmental input remains monosyllabic. As for linguistic output2,itsinserted vowels are deleted whether they are assigned with one two beats.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
34
Finally, the musical inputs are mapped to the musical outputs, where the prosodic word structure is removed.
Take the linguistic input, /blu/, for example. /blu/ is produced as (blu) or (bulu)
in the linguistic outputs. The monosyllabic (blu) is referred to as linguistic output1, whereas (bulu) is referred to as linguistic output2. The linguistic outputs are perceived as the musical inputs, where both the monosyllabic (blu) and the disyllabic (bulu) are assigned with one and two beats respectively. Finally, the musical beat association yields the musical output, where the prosodic word structure is removed.
3.4 L
I-to-L
OMapping: Production Grammar
This section analyzes the linguistic input (LI) to linguistic output (LO) mapping under the framework of Optimality Theory (Prince and Smolensky 1993/2004). There are two linguistic outputs. One is with vowel insertion and the other is without vowel insertion. Take /blu/ ‘blue’ for example. The output of /blu/ can be either (blu) or (bulu). The relevant constraints are given in (47-50).
(47) MAX-C
Assign one violation mark for every input consonant that does not show in the output.
(48) DEP-V
Assign one violation mark for every output vowel that does not show in the input.
(49) *CC
Assign one violation mark for every syllable that has a consonant cluster.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
35
(50) ALIGN-E (LEX, )
Assign one violation mark for every lexical word whose edges are not aligned with the edges of a prosodic word.
The database shows that 86% of the onset clusters are pronounced faithfully to their input. Therefore, we need faithfulness constraints, MAX-C and DEP-V that demand faithful relation between the input and the output. These constraints forbid consonant deletion and vowel insertion so that the output segments remain the same as the input. *CC is a markedness constraint that forbids complex consonants in a syllable. *CC competes with the faithfulness constraints and select the outputs that do not have consonant clusters. Therefore, in order to surface [blu], MAX-C and DEP-V should be ranked higher than *CC so that the consonant cluster can be preserved.
Syntactic lexical words are aligned with the prosodic words in the output, which is governed by ALIGN-E(word, ).
Tableau (51) shows the competition of these constraints.
(51) LI-to-LO mapping: output1, (blu)
/blu/ ALIGN-E (word, )
MAX-C DEP-V *CC
a. (blu) *
b. (bulu) *!
c. (bu) *!
d. blu *! *
In tableau (51), MAX-C and DEP-V are ranked higher than *CC. (51b) inserts a vowel and incurs a fatal violation of DEP-V. (51c) deletes a consonant, and thus is ruled out by MAX-C. The output in (51d) is not a prosodic word, so it is ruled out by ALIGN-E(word, ). (51a) that is faithful to the input is selected, in spite of a violation of *CC.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
36
The database shows that 14% of the coda cluster are produced with an inserted vowel like (bulu). To obtain (bulu), *CCshould be ranked higher than DEP-V. In this case, consonant clusters are avoided by inserting a vowel instead of deleting any consonant. Therefore, both *CC and MAX-C ranks higher than DEP-V. Tableau (52) examines the selection of the optimal output.
(52) LI-to-LO mapping: output2, (bulu)
/blu/ ALIGN-E (word, )
*CC MAX-C DEP-V
a. (bulu) *
b. (blu) *!
c. (bu) *!
d. blu *! *
Consider the output2 in (52). (52b) preserves the onset cluster [bl], so it is ruled out by *CC.(52c) conforms to *CCby deleting [l], and thus is ruled out by MAX-C.
(52d) is not a prosodic word, so it is ruled out by Align-E(word, ).
The constraint rankings for the linguistic input-to-output production grammars are summarized in (53).
(53) Linguistic input to linguistic output production grammars
a. Output1: [blu]→(blu) ALIGN-E(LEX, ), MAX-C, DEP-V >> *CC
b. Output2: [blu]→(bulu) ALIGN-E(LEX, ), *CC, MAX-C >>DEP-V
The target word surfaces with two syllables when *CCis ranked higher than MAX-C and DEP-V. When *CCisranked at the bottom, the target word is faithful to the input and remains monosyllabic. ALIGN-E(LEX, ) is the dominant constraint governing both outputs.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
37
3.5 L
O-to-M
IMapping: Perception Grammar
3.5.1 Lo
1-to-M
IMapping
As mentioned in section 3.4, there are two kinds of linguistic outputs. Output1
(LO1) is monosyllabic, whereas output2 (LO1) is disyllabic. This section examines the linguistic output1 (LO1) to the musical input (MI) mapping.
When LO1 is mapped to MI, it is respectively assigned with one and two musical beats. I first discuss the case where LO1 is assigned with one musical beat. Three relevant constraints are proposed in (54-56).
(54) NOSTRAY
Assign one violation mark for every stray element, e, in the output.
(55) NOSHARE-B
Assign one violation mark for every musical beat that is shared by two or more syllables.
(56) NOSHARE-σ
Assign one violation mark for every syllable that is shared by two or more musical beats.
The constraint NOSTRAY is an undominated constraint, which ensures that every syllable or musical beat is associated. MAX-C is ranked above DEP-V, such that consonant deletion is avoided while vowel insertion is possible. NOSHARE-σ and NOSHARE-B are ranked at the bottom, as a syllable is often linked to multiple musical beats and vice versa. A partial constraint ranking is proposed in (57).
(57) Lo1-to-MI partial constraint ranking:
NOSTRAY, MAX-C >> DEP-V >> NOSHARE-B, NOSHARE-σ
‧
The tableau in (58) shows how the constraint ranking works.
(58) LO1-to-MI mapping: onset cluster (♩) syllable and a stray musical beat. (58a) incurs no constraint violation and is selected as the optimal output.
When LO1 is assigned with two beats, the mapped MI is still (blu), as shown in tableau (59).
‧
The syllable in (59a) is shared by two musical beats, violating the bottom-ranked NOSHARE-σ, but it is still selected as the optimal output. In (59b), there is one unlinked musical beat, and thus is eliminated by NOSTRAY. DEP-V and MAX-C then rule out (59c) and (59d) respectively.
3.5.2 Lo
2-to-M
IMapping
This section discusses the mapping from linguistic output2 (LO2) to the musical input (MI). There are two syllables in LO2, which are linked to either one or two musical beats in MI.
When the LO2, (bulu), is assigned with one musical beat, it is mapped to monosyllabic (blu),as shown in tableau (60).
‧
to the insertion of [u]. However, the same linguistic output in (60d) involves no vowel insertion, and thus DEP-V is inactive. NOSTRAY and MAX-C rule out (60b) and (60c) respectively. NOSHARE-B favors (60a) over (60d), where the musical beat is shared by two syllables. As a result, (60a) emerges.When the LO2, (bulu), is assigned with two musical beats, it is still mapped to monosyllabic (blu), which is linked to two musical beats. However, the tableau evaluation in (61) makes an incorrect prediction.
‧
NOSHARE-σ undesirably rules out (61a), the real optimal output, as indicated by the parenthesized white right-headed hand symbol, (), and (61d) is wrongly selected, as indicated by the parenthesized black right-headed hand symbol, . To exclude the unwanted candidate in (61d), I propose the constraint in (62).
(62) ALIGN-R(♩, ):
Assign one violation mark for every musical beat, ♩, that is not linked to the rightmost syllable in a prosodic word, .
This constraint must be ranked higher than NOSHARE-σ so that all the musical beats are associated with the final syllable. The enriched constraint ranking is provided in (63).
‧
(63) Lo2-to-MI constraint ranking (enriched)
NOSTRAY, MAX-C >> DEP-V >> ALIGN-R(♩,) >> NOSHARE-B, NOSHARE-σ violation of ALIGN-R(♩,). Eventually, (64a) is selected as the optimal output, where the leftmost [u] is deleted while the rightmost [u] is shared by two musical beats.
3.6 M
I-to-M
OMapping: Production Grammar
The mapping between the musical input and the musical output is subject to the production grammar that preserves syllable-beat association and removes prosodic structure in the musical output. Three constraints are posited in (65-67), and the constraint ranking is summarized in (68).
‧
Assign one violation mark for every output association line that is not identical to that in the musical input.
(66) *PROSST:
Assign one violation mark for every prosodic structure in the output.
(67) MAX-PROSST:
Assign one violation mark for every prosodic structure in the musical input that does not have a correspondent in the musical output.
(68) MI-to-MO constraint ranking:
ID-ASSOC,*PROSST >> MAX-PROSST
ID-ASSOC must be undominated so that the syllable-beat association in the musical input is retained in the musical output. *PROSST must outrank MAX-PROSST to ensure that there is no prosodic structure in the musical output. The tableaux in (69-70) show how the constraint ranking works.
‧
Both (69b) and (70b) are ruled out by *PROSST, as they preserve the prosodic words in the output. On the other hand, (69c) and (70c) are eliminated by ID-ASSOC, since the association lines are changed.
3.7 Summary
The mapping of an onset cluster from the linguistic input to the linguistic output involves interactions between segmental faithfulness and markedness, as well as prosodic alignment. Two output variants are observed: LO1 is monosyllabic and faithful, while LO2 is disyllabic, with a vowel inserted to resolve the onset cluster. I have proposed a set of relevant constraints, which are subject to flexible rankings, as illustrated by the Hasse diagrams in (71-72).
(71) LI-to-LO1 mapping: ()
ALIGN-E (LEX, ) DEP-V MAX-C
*CC
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
45
(72) LI-to-LO2 mapping: ()
ALIGN-E(LEX, ) *CC MAX-C
DEP-V
In the mapping to the musical input, the two linguistic outputs (LO1 and LO2) are assigned with one or two musical beats, as in (73).
(73) LO1-to-MI Mapping: (♩, ♩♩) LO2-to-MI Mapping: (♩, ♩♩)
In spite of the fact that there are two constraint rankings in the LI-to-LO mapping, the perception grammar in the LO-to-MImapping lies in a single constraint ranking, as in (74).
(74) LO-to-MI mapping
NOSTRAY MAX-C
DEP-V ALIGN-R(♩, )
NOSHARE-B NOSHARE-σ
The mapping between the musical input (MI) and the musical output (MO) is governed by three constraints, ID-ASSOC,*PROSST,and MAX-PROSST, as illustrated in tableaux (69-70), where the optimal outputs preserve the musical beat association but remove the constructions of the prosodic word.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
46
Chapter 4
Language-to-Music Mapping: Coda Cluster
4.1 Introduction
This chapter continues to discuss the Mandarin-accented English in reading and singing, with a focus on coda clusters. Again, the segmental changes are examined in both the linguistic mapping and the language-to-music mapping. Unlike the onset clusters, the English coda clusters collected encounter the sonority sequencing principle (SSP). The segmental changes in relation to the coda clusters further support the argument for the separation of the perception grammar and the production grammar. Again, the linguistic output, the English words read by the Mandarin speakers, is perceived as the musical segmental input for singing. There are two linguistic output variations of the monosyllabic syllables with coda clusters.
Linguistic output1 preserves the coda cluster and remains monosyllabic. Linguistic output2 involves vowel insertion so that the syllable surfaces with two syllables. The linguistic outputs are formed into prosodic words. Both of the linguistic outputsare respectively assigned with one and two musical beats. This chapter looks closely into how musical beat assignment affects the segmental changes and how the SSP and prosodic wordhood restrict the musical beat assignment in the musical input.
4.2 Data Design
The target monosyllabic words are designed to compare segmental changes of coda clusters in linguistic mappings and language-to-music mappings. The informants are the same as those mentioned in Chapter 3. They are asked to read and sing the