• 沒有找到結果。

The Phonological Representation of Mandarin Vowels

N/A
N/A
Protected

Academic year: 2021

Share "The Phonological Representation of Mandarin Vowels"

Copied!
53
0
0

加載中.... (立即查看全文)

全文

(1)

One of the fundamental goals of every phonological theory is to account for the nature of the basic units of speech sounds, and the relationships between these units and their contextual variants. This relationship is equally crucial to phonological theory whether it is called ‘phonemes and allophones’, ‘underlying and surface forms’, or ‘input and output’. However, purely structural analyses of phonological systems can often produce several hypotheses regarding the phonemic inventory and its surface reflexes in any particular language, all of which are supportable by the contrast and alternation patterns of the language. In this paper we look at four such hypotheses regarding the underlying vowel system of Mandarin, all based on Beijing Mandarin: the six-vowel system of C. Cheng (1973), the five-vowel systems of R. Cheng (1966) and of Lin (1989), and the four-vowel system of Wu (1994). We then present distri-butional, phonetic, and psycholinguistic evidence (the latter based on a corpus of 238 syntagmatic speech errors or ‘slips of the tongue’ involving vowels) that the vowel system of the dialect of Mandarin currently spoken in Taiwan cannot be accounted for by any of these hypotheses. We then propose a new 5-vowel system for Taiwan Mandarin, based on the distributional, phonetic, and especially the psycholinguistic facts. We conclude that phonological theories which are compatible with psycholin-guistic evidence such as the data presented here are those most likely to be modeling the actual cognitive representations and processes of real speakers.

1. INTRODUCTION

One of the fundamental goals of any phonological theory is to be able to account for the nature of the basic units of speech sounds, and the rela-tionship between these units and their contextual variants, both universally and in particular languages. Whether we talk about this relationship in terms of ‘phonemes and allophones’, or ‘underlying and surface forms’, or ‘input and output’, we are making the same basic assumption: In any language there is an inventory of possible abstract sound categories in which words are represented, which will be manifested differently in different phonetic contexts due to both universal and language-specific factors (Goldsmith (1995: 2)). In psycholinguistics we would talk about these same phenomena in terms of the units in which phonological form is represented in lexical entries, and the relationships between these units to either the incoming acoustic signal, or to their manifestation in contextual speech. In any particular language, the actual analysis of the abstract phonological units and their contextual manifestations is frequently the source of much theo-retical controversy.

This is certainly true of the system of vowels in Mandarin Chinese. There I - P I N G W A N A N D J E R I J A E G E R

Journal of East Asian Linguistics 12, 205–257, 2003.

 2003 Kluwer Academic Publishers. Printed in the Netherlands.

T H E P H O N O L O G I C A L R E P R E S E N TAT I O N O F TA I WA N M A N D A R I N V O W E L S : A P S Y C H O L I N G U I S T I C S T U D Y

(2)

has been a longstanding controversy in the literature over the number of underlying vowel categories in Mandarin, and the relationship of the myriad of surface vowel forms to these phonemic categories (e.g., Chao (1934, 1968), R. Cheng (1966), C. Cheng (1973), Pullyblank (1983), Lin (1989), Wang (1993), Wu (1994)). The reason for the continuance of this contro-versy is that most phonetic manifestations of vowels in Mandarin occur in a fairly narrow range of contexts, which suggests that they probably can be reduced to a smaller set of basic vowel categories. However, a standard structural analysis of the contexts in which the various vowel alter-nates occur allows for several different resolutions to the problem, each one of which can be motivated on the basis of theoretical and language-internal consistency. Thus if one uses strictly ‘language-internal’ evidence (Ohala (1986)), it may be the case that this controversy cannot be definitively resolved. Furthermore, while many of these theories regarding the under-lying vowel system intend to capture the basic vowel system for all dialects of Mandarin, they are for the most part based on the Beijing dialect. Thus even if the analysis adequately captures that dialect, it may be the case that other Mandarin dialects are different enough from Beijing Mandarin that they can be accounted for only by a different underlying vowel system. The purpose of the present paper is to bring some distributional, phonetic, and particularly psycholinguistic evidence to bear on this issue, looking specifically at the vowel system of the dialect of Mandarin spoken in Taipei, Taiwan (hereafter TM). First we will review four theories regarding the underlying vowel system of Mandarin, which are intended to be applic-able to all dialects of Mandarin: the 6 vowel system of C. Cheng (1973), the five-vowel systems of R. Cheng (1966) and Lin (1989, 1992), and the four-vowel system of Wu (1994). We then look at some distributional and phonetic facts regarding the vowel system of TM, and show that certain sound changes have caused the Taiwan and Beijing dialects to have different vowel systems, particularly in terms of the high central unrounded (or ‘apical’) vowels, and in terms of the phonemic affiliation of the lower-mid front unrounded vowel [ε]. We then argue for this new hypothesis regarding the underlying system of TM by presenting evidence from a corpus of 238 syntagmatic phonological speech errors or ‘slips of the tongue’, where either one vowel is substituted for another, or where the environment adjacent to a vowel is changed, causing the vowel to shift in quality. The questions we ask are the following:

(1) Which vowels can be substituted for each other in syntagmatic phono-logical errors? The rationale for this question is that if two vowel phones can be substituted for each other without any change in the contextual 206 I - P I N G W A N A N D J E R I J A E G E R

(3)

environment, then they must derive from different underlying categories, since slips of the tongue never produce illegal surface phonetic forms. (2) How do vowel phones change if the contiguous environment is changed by an error? The rationale for this question is that if the intended vowel phone is changed to a different vowel phone consistently when the context changes in some particular way, then these two phones are likely to be contextually determined manifestations of the same abstract vowel category.

This paper is organized as follows. In the following section we will discuss the four relevant theories mentioned above regarding the basic vowel categories of Mandarin, and then lay out the facts regarding the vowel phones in TM, including the restrictions on their contextual occurrences and ways in which they differ from Beijing Mandarin. In the third section we will present our methodology for the collection and analysis of slips of the tongue in detail. Section four presents our psycholinguistic results, and the analysis of those results both in terms of the four competing hypotheses and in terms of our proposed model for TM. In section five we summarize our study, and discuss the phonological analysis supported by this study in detail, relating it to a psycholinguistic model of speech production planning.

Because we want to make it clear that the issues being addressed in this paper are relevant to a phonological analysis of Mandarin regardless of the phonological theory in which such an analysis is performed, we will be eclectic in our terminology throughout this paper. That is, we will use the terms ‘phoneme’, ‘underlying representation’, ‘input’, ‘abstract phonemic category’, and ‘cognitive category’ interchangeably to indicate the abstract form in which vowels are represented in lexical entries, and we will use the terminology ‘allophone’, ‘surface form’, and ‘phonetic man-ifestation’ interchangeably to indicate the form of the vowel which is actually spoken or is the ‘output’ of the phonological ‘rules’ or ‘constraints’ component during processing. We take the stance that, from a psycholin-guistic point of view, all of these ways of talking about vowels should be reducible to the same cognitive entities.

2. MANDARIN VOWELS

2.1. Previous Theoretical Analyses

Out of the many analyses of the Mandarin vowel system to appear previ-ously, we have selected the following four proposals to review here: C.

(4)

Cheng (1973), who proposed a six-vowel system based on phonemic evidence; R. Cheng (1966) who proposed a five vowel system also based on phonemic analysis; Lin (1989), whose 5-vowel system is developed from autosegmental phonology and underspecification theory; and Wu (1994), whose 4-vowel system is based on a feature geometry analysis. We selected these four theories for three reasons. First, they cover the range of hypotheses regarding the number of underlying vowels in Mandarin, from 4–6.1Second, they represent a range of types of phono-logical theories, from phonemic analyses through autosegmental and feature geometry approaches. Third, they all propose to represent the basic structure of the Mandarin vowel system regardless of dialect, and thus it is appropriate for us to compare the vowel system of TM to these proposals.

C. Cheng’s (1973) analysis begins with the assumption that there are 12 surface vowels in Mandarin: [i, ι, ι, y, u, e, ε, ə, o, , a, ɑ]. (All of these symbols are used according to their IPA value, except the two ‘apical’ vowels: [ι], which is a high central unrounded vowel, sometimes called a ‘dental’ vowel, and [ι], a high central retroflexed vowel.) His 6-vowel phonemic system groups allophones as follows:

(1) The 6-vowel system of C. Cheng (1973) /i/ → [i] // → [ι], [ι] /y/ → [y] /u/ → [u] // → [e], [ə], [o], [] /ɑ/ → [a], [ɑ], [ε]

C. Cheng argues that the ability of a vowel to occur singly or in CV syllables is evidence that the vowel should be analyzed as an underlying phoneme. Since the vowels [i, ι/ι, y, u, , a] can occur alone or finally, he treats them as the underlying phonemic centers, as opposed to the other vowel phones which either cannot occur alone or must be conditioned by a preceding or following segment (the details of this conditioning will be discussed in Section 2.2 below). One phonetic difference between his analysis and the other analyses to be discussed here is that he considers the manifestation of the low vowel which occurs in open syllables to be the back vowel [ɑ], whereas all the other theories argue that the low vowel which occurs in open syllables is the central [a]. A phonemic difference is that he considers the two apical vowels as being derived from an under-lying high central unrounded vowel phoneme, //; it will be shown below that all four theories differ as to the status of the apical vowels. Finally, 208 I - P I N G W A N A N D J E R I J A E G E R

(5)

he considers Mandarin to have vowel sequences rather than vowel-glide or glide-vowel sequences, whereas many other analyses claim that when a high vowel is found adjacent to a non-high vowel, the high vowels are realized as glides.

The 5-vowel system of R. Cheng (1966) assumes the same 12 surface vowels as in the 6-vowel system of C. Cheng (1973); R. Cheng also does not include glides in his system. However, he has one less phoneme due to deriving both [i] and the two apical vowels from the phoneme /i/ as shown in (2).

(2) The 5-vowel system of R. Cheng (1966) /i/ → [i], [ι], [ι]

/y/ → [y] /u/ → [u]

/ə/ → [e], [ə], [o], [] /a/ → [a], [ɑ], [ε]

Lin (1989) presents a 5-vowel system which is similar to R. Cheng’s in that it posits the same 5 phonemes /i, y, ə, a, u/. She makes an argument for selecting these phones as the basic allophones because she claims that they are the phonetic centers of each of the phonemes. However, her analysis differs from R. Cheng’s in several ways. First, she states that there are 13 surface vowels, which include all of those in R. Cheng plus the lower-mid back rounded vowel [ɔ]: [i, ι, ι, y, u, e, ε, ə, ɔ, o, , a, ɑ]. Second, she derives the two apical vowels by rule from an epenthesis and assimi-lation process; she considers them to be a vocalic extension of the onset consonant, following Chao’s (1934) proposal. Third, she considers [ε] to be a manifestation of two different phonemes; in the phonetic sequences [jεn] and [ɥεn] she argues that the [ε] is underlyingly /a/, whereas in the sequences [jε] and [ɥε], it is underlyingly /ə/. Finally, she does derive glides from high vowels adjacent to non-high vowels, so that, for example, the word /kuai55/ ‘obedient’ would occur as [kwaj55] in its surface form; similarly the glide [ɥ] is derived from the vowel /y/. Thus her phonemic analysis is as follows.

(3) The 5-vowel system of Lin (1989) /i/ → [i], [j]

/y/ → [y], [ɥ] /u/ → [u], [w]

/ə/ → [e/ε], [ə], [o/ɔ], [] /a/ → [a], [ɑ], [ε]

(6)

The surface vowels [ε] and [ɔ] associated with the /ə/ phoneme are argued to be lowered variants of [e] and [o] respectively. Lin cites as evidence that the [ε] in sequences of [palatal glide + ε + n] is derived from the /a/ phoneme, the fact that these same morphemes, when they undergo [r] suf-fixation, surface with the vowel [a] ((1989: 68); see also C. Cheng (1973)). So for example, /ian/ has as its surface form [jεn]; but for the sequence /ian/ + /r/, the nasal is deleted and the phonetic form is [jar]. The other mani-festations of the /ə/ phoneme are derived by assimilation, as will be discussed in detail in Section 2.2 below; Lin’s feature system will be dis-cussed in Section 5.3 below.

Finally, Wu (1994) presents a somewhat different analysis of the vowel system of Mandarin, positing four underlying vowels. As in Lin’s analysis, Wu derives glides from high vowels, and posits that there are 13 surface vowels; however the set of vowels Wu is working with is somewhat dif-ferent from the others: [i, , ɯ, y, u, ə, e, , o, ɒ, æ, a, ɑ]. Her phonemic groupings are as follows:

(4) The 4-vowel system of Wu (1994)

Simple Vowels Complex Vowels

/i/ → [i], [], [ɯ], [j] /ui/ → [y], [ɥ] /u/ → [u], [w]

/ə/ → [e], [ə], [o], [ɒ], [] /a/ → [a], [ɑ], [æ]

Wu distinguishes the same two apical vowels as in the other theories, but uses the symbols [] for the dental vowel and [ɯ] for the retroflex vowel. She derives these two vowels from the phoneme /i/ as does R. Cheng, as opposed to setting them up as a separate phoneme (C. Cheng) or deriving them by rule (Lin). For the high front rounded vowel [y], she derives this vowel from a sequence of the two underlying phonemes /ui/. She argues that this vowel should not have the same status as the other high vowels since it has a limited distribution, failing to occur in post-vocalic position (i.e., as a postnuclear glide), where the other high vowels do occur. She con-siders the lower back rounded vowel which is part of the schwa phoneme to be [ɒ] instead of [ɔ] (e.g., [kwɒ35] ‘country’), and analyzes the front raised version of the phoneme /a/ as phonetically [æ] rather than [ε] as in the other three theories (e.g., [jæn35] ‘language’).

This brief review of four diverse hypotheses regarding the basic vowel system of Mandarin shows several areas of difference. Looking first at the high vowels, C. Cheng posits four distinct phonemes, /i, y, , u/, whereas R. Cheng and Lin posit three phonemes /i, y, u/; R Cheng derives the two apical vowels from /i/, but Lin derives them by rules of epenthesis and

(7)

assimilation. Wu posits only two high vowels /i, u/; she derives the apical vowels from /i/, but derives [y] from a sequence of /ui/. Thus these analyses raise questions regarding the relationship between the apical vowels and /i/, as well as the status of /y/ as an independent phoneme.

As far as the low vowel is concerned, all four theories consider [a, ɑ] to be conditioned variants of each other; all but C. Cheng consider [a] to be the basic variant, which occurs in open syllables with no conditioning environment. All four theories indicate that there is a front allophone in this set; C. Cheng, R. Cheng, and Lin analyze this segment as [ε], whereas Wu calls it [æ]. Lin further argues that there is also an [ε] which is derived from the /ə/ phoneme and occurs in open syllables after palatal glides. So for the low vowels, there is some question about which is the central allo-phone, and about the status of [ε].

Finally, with the mid vowels, all four theories consider [e, ə, o, ] to be conditioned variants of this phoneme. The low back rounded vowel [ɔ] does not occur in the theories of C. Cheng or R. Cheng, whereas Wu includes this vowel, saying that it is phonetically [ɒ]. Lin argues that there are lowered versions of [e] and [o] which occur as part of this phoneme as [ε] and [ɔ]. Thus the number of surface variants of the mid vowel, as well as the actual phonetic quality of some of these variants, is at issue. The status of [ε] is also relevant to this phoneme.

2.2. Description of Vowel Occurrence and Phonetics in

Taiwanese Mandarin

2.2.1. Overview

In the dialect of Mandarin under study here, there are the following 12 surface vowels.

TABLE 1

Vowel phones in Taiwan Mandarin

Front Central Back

Unround Round Unround Unround Round

High i y  u

Mid e ə  o

Lower-Mid ε ɔ

(8)

The most obvious difference between this set of vowels and the set discussed in Section 2.1 above is that this dialect does not distinguish the two apical vowels (i.e., dental apical [ι] vs. retroflex apical [ι]). We will return to a detailed discussion of this point below.

The contextual occurrences of these vowel phones are given in Table 2.2

As can be seen from Table 2, in TM the five vowels [i, y, u, , a] can occur alone as complete syllables, and the greatest number of contrasts which can occur with the same onset is also these five, e.g., [li, ly, lu, l, la]. The vowels [i, , y, u, ε, ɔ, , a] can occur in open syllables; the low vowel in open syllables is [a] rather than [ɑ] in TM. The vowels [i, , y, ε, ə, a] can occur in syllables closed with the nasal [n]; the vowels [i, o, ɑ] occur in syllables closed with the nasal [ŋ]. The vowels [o, ɑ] can occur in syllables closed with the glide [w]; the vowels [a, e] can occur in syllables closed with the glide [j]. A syllable can begin with one of the following vowels: [i, y, ə, o, a, ɑ].

As discussed above, in many traditional and recent analyses of glides, it is claimed that the occurrence of glides can be predicted in every envi-ronment by rule (e.g., Lin (1989), Duanmu (1990, 2000), Li (1999)). This is equally true of TM as of other dialects of Mandarin. Glides are analyzed as coming from underlying high vowel phonemes: a high vocalic segment alternates with a glide when adjacent to a nonhigh nucleus vowel, causing an alternation of [i] with [j], [u] with [w], and [y] with [ɥ]; the high central phone [] does not have a glide alternant.

We will now turn to a detailed discussion of the distribution and phonetics of sets of vowels, organized by vowel height.

2.2.2. High Vowels

In all of the analyses of Mandarin presented in Section 2.1 above, the researchers have made a distinction between the dental apical vowel [ι] or [] and the retroflex apical vowel [ι] or [ɯ]. The major claim has been that the high central vowel after dental obstruent onsets differs in quality from the high central vowel following retroflex obstruent onsets. However, in TM many speakers do not produce strongly retroflexed consonants (see note 6), and more importantly, they do not produce the strongly retroflexed vowel after retroflex consonants which is more typical of the Beijing dialect. Thus these two vowels have fallen together in TM, and are phonetically indistinguishable. The following spectrogram compares the two syllables [ts] and [t ], in order to illustrate this claim. The utterances in this spec-trogram were spoken by a 21-year-old male who is a native speaker of

(9)

TABLE 2

Possible co-occurrence of Mandarin vowels with all onsets and codas

i  y u e ε ə ɔ o  a ɑ single i y u  a _(C/G) in yn ən oŋ an ɑŋ iŋ ow aj ɑw p, ph pi pu pa

pin pən poŋ pan pɑŋ

piŋ pej pow paj pɑw

t, th ti tu t ta

tin tən toŋ tan tɑŋ

tiŋ tej tow taj tɑw

k, kh ku k kh a kən koŋ khan kɑŋ kej kow kh aj kɑw f fu fa fən foŋ fan fɑŋ fej fow fɑw s, ts, tsh s su s sa sən soŋ san sɑŋ

sej sow saj sɑw

, t , t h,  u  a ən oŋ an ɑŋ ej ow aj ɑw , t , t h i y in yn iŋ x xu x xa xən xoŋ xan xɑŋ

xej xow xaj xɑw

m mi mu ma

min mən moŋ man mɑŋ

miŋ mej mow maj mɑw

n ni ny nu n na

nin nən noŋ nan nɑŋ

niŋ nej now nɑŋ nɑw

l li ly lu l la

lin lən loŋ lan lɑŋ

liŋ lej low laj lɑw

(C)j_ jε joŋ ja jɑŋ jεn jow jɑw _(C)ɥ_ ɥε ɥεn ɥoŋ (C)w_ wej wɔ wa wən woŋ wan wɑŋ waj

(10)

TM. (Naturally these spectrograms, as well as those given in (8) below, are meant to serve only as illustrations of the neutralization of this contrast in the dialect of some speakers of TM; full documentation that this neu-tralization is widespread in TM would of course require data from multiple speakers, with detailed acoustic and statistical analyses.)

(5) Spectrogram comparing [ts] and [t ]

An analysis of the formant values in the spectrogram showed that that vowel following the dental onset has formant values of approximately F1: 300, F2: 2245, and F3: 3295. For the vowel following the retroflex onset the values are approximately F1: 300, F2: 2285, and F3: 3400. The first formants of the two vowels are nearly identical, and the second formants differ by 80 Hz. The third formant is about 105 Hz higher for the vowel following the retroflex consonant, showing that there is clearly no retroflexion on the vowel, since retroflexion typically causes lowering of all three formants, especially F3 (Jakobson, Fant and Halle (1951), Ladefoged (2001: 212–214)). Since a variation of 100 Hz is within the normal frequency variance for a formant of a single vowel, we would argue that there is no phonetic distinction between the high central vowel after dental vs. retroflex onsets in TM, and thus we notate this vowel as [] in every case.

(11)

Turning to the distributional facts, it can be seen that the phones [i, y, u] contrast in open syllables following /l/, as further illustrated in (6a). However, [] contrasts only with [u], since [] can only occur after dental and retroflex affricates and fricatives, whereas [i, y] cannot occur in these environments, as shown in (6b,c). Thus the phone [] does not contrast with either [i] or [y].

(6) a. [lu51] ‘deer’ [li51] ‘power’ [ly51] ‘green’ b. [su51] ‘pale’ [s51] ‘four’ *[si] *[sy] c. [ u51] ‘tree’ [ 51] ‘yes’ *[ i] *[ y]

In sum, the phonetic and distributional facts show that /i, y, u/ contrast and are therefore members of three different phonemic categories. The high central vowel [], which does not have dental vs. retroflex variants in TM, contrasts with [u], but is in complementary distribution with both [i] and [y]. Thus it cannot be determined by the distributional facts alone which of these two phones it should be grouped with in a single phonemic category, if either. The psycholinguistic data to be discussed below will be crucial for supporting one hypothesis over another.

2.2.3. Low Vowels and the Lower-Mid Vowels

In TM, the low central vowel [a] occurs in open syllables, or in closed syllables before the coda consonants [n, j] as illustrated in (7a). The low back vowel [ɑ] only occurs before the velar consonants [ŋ, w], as illustrated in (7b). There is no phonetic [æ] in TM; the low front phone which occurs in the context of [j__n] is the lower-mid front vowel [ε]. However, there are some problems with analyzing the [ε] as part of the /a/ phoneme as opposed to including it in the mid-vowel phoneme.

First, the phones [a] and [ε] contrast in open syllables following palatal glides, as shown in (7c); this is parallel to the fact that [a] and [ɔ] contrast in open syllables following the labial-velar glide, as shown in (7d).

(7) a. [ta55] ‘to take (a bus)’ [tan55] ‘to serve’ [taj55] ‘stupid’ b. [tɑŋ55] ‘to be’ [tɑw55] ‘knife’

c. [ja51] ‘surprise’ [jε51] ‘leaves’

d. [wa21] ‘roof ’ [wɔ21] ‘I’

e. [jεn55] ‘smoke’ [ɥεn55] ‘hatred’ *[jan] *[ɥan] Because [a] and [ε] contrast in otherwise identical environments, there is reason to believe that [ε] should not be grouped with the /a/ phoneme, at least according to standard theories of phonemic contrast.

(12)

syllables, such as those shown in (7c), Lin analyzes the [ε] as a lowered version of [e] and thus an alternant of the mid-vowel phoneme. But the [ε] which occurs in the environment [j__n] or [ɥ__n], as illustrated in (7e), is argued to be derived from /a/, since the phone [a] cannot occur in this environment. Evidence for the latter came from the [r] suffixation process, which brings out a [a]–[ε] alternation in the same root, as discussed in Section 2.1 above. However, in TM this [r] suffixation process does not occur, and thus speakers would have no reason to believe that the [ε] in the environment [palatal glide + ε + n] is any different from the [ε] that occurs in open syllables following the palatal glides. Furthermore, there is no phonetic difference between the two manifestations of [ε], as shown in the example spectrograms in (8). These two utterances were spoken by the same male native speaker of TM who produced the spectrograms in (5) above.

(8) Spectrograms of [jε] vs. [jεn]

An analysis of these two spectrograms shows that the steady-state vowel formants in [jε] are approximately F1: 605, F2: 2050, and F3: 3280; for the vowel in [jεn] the formant values are F1: 610, F2: 2005, and F3: 3200.3 Clearly these two vowels show no obvious phonetic differences which could be attributed to their being affiliated with different underlying vowel categories. Thus the second controversial point which will be looked at in

(13)

detail in the psycholinguistic data below is whether or not the [ε] behaves as if it belongs to both the mid and low phonemes, or whether it shows a strong affiliation with only one of these two phonemes.

2.2.4. Mid and Lower-Mid Vowels

The occurrence of the mid vowel phones in TM depends on the status of adjacent segments, as shown in Table 2. The words in (9) below were chosen to illustrate some of the co-occurrence patterns of mid vowel phones with both onset and coda consonants.

(9) a. [51] ‘hungry’ [kh

51] ‘lesson’

b. [ən55] ‘mercy’ [pən51] ‘stupid’

c. [oŋ55] ‘old man’ [ow55] ‘Europe’ [koŋ51] ‘together’ [kow21] ‘dog’ d. [kwɔ35] ‘nation’

e. [kej21] ‘to give’ [tjε55] ‘father’

[lɥε51] ‘to plunder’

f. [kwej51] ‘expensive’ [t jow51] ‘uncle’

g. [joŋ51] ‘to use’ [ɥoŋ51] ‘to use’ [woŋ51] ‘jar’ h. [jεn35] ‘along’ [ɥεn35] ‘round’ [wən51] ‘to ask’ It can be seen from Table 2 that the mid vowel phones never occur after the palatal obstruents [ , t , t h]. The examples in (9a) illustrate the fact that the vowel phone [] occurs alone or in CV syllables where the C is neither a glide nor a labial consonant.4The examples in (9b) illustrate the fact that [ə] always occurs before the alveolar nasal [n], and can occur either with no onset or with obstruent onsets; glide onsets will be discussed below. While in some dialects of Mandarin [ə] can occur before the velar nasal [ŋ], in either onsetless syllables or syllables with onsets, in the dialect of TM spoken by the subjects in this study, all instances of [(CG)əŋ] have been neutralized to [(CG)ən]. In (9c), the examples show that [o] always occurs before [ŋ] or [w], in either onsetless syllables or in syllables with obstruent onsets; again, glide onsets will be discussed below. As seen in (9d), when [w] is in the prenuclear glide slot in codaless (C)GV syllables, the only mid vowel phone allowed in this position is [ɔ]. Example (9e) illustrates the fact that when there is a palatal glide [j] in the coda position, the mid vowel [e] occurs, but when there is a prenuclear palatal glide the vowel is realized as [ε]. (Recall that the rounded palatal glide [ɥ] occurs only in prenuclear, not postnuclear position.) When glides both precede and follow a mid vowel, the coda glide determines the quality; in (9f) the [e] in [kwej51] is conditioned by the coda [j], but the [o] in [t jow51] is

(14)

condi-tioned by the coda [w]. When there is a glide onset and a nasal conso-nant coda, there are three different patterns: When the coda nasal is [ŋ], only the [o] mid vowel is allowed, as illustrated in (9g). When the coda nasal is [n], the front vowel [ε] occurs after the two front glides [j] and [ɥ], and the central vowel [ə] occurs after the labial-velar glide [w], as shown in (9h).

We summarize the allophonic variation of mid and lower-mid vowels as below. (Since no mid vowels occur following palatal obstruent onsets, this has not been indicated separately in the table.)

(10) [] occurs in (C)V syllables where C is neither a labial obstruent

[] nor a glide.

[ə] occurs in (C)V[n] or [w]V[n] syllables. [e] occurs before [j] in CV[j] or [w]V[j] syllables. [ε] occurs after [j, ɥ] in (C)GV([n]) syllables.

[o] occurs in (C)(G)VF syllables where the final is [ŋ] or [w]. [ɔ] occurs after [w] in (C)GV syllables.

In the psycholinguistic data we will look for evidence that all of these phones are associated with each other; in particular we will see whether [ε] is more closely related to the mid vowels vs. the low vowels.

2.2.5. Summary

In Table 3 we summarize all possible contrasts of surface vowels. It can be seen from this table that there are many phones in Mandarin which contrast, and thus are candidates for being grouped into separate phonemic categories. On the other hand, there are many non-contrasting phones which are phonetically similar and could be candidates for being grouped in phonemes. For example, [] does not contrast with either [i] or [y]; [e, ε, ə, ɔ, o, ] do not contrast with each other. The phone [u] does not contrast with [o, ɔ, ə, ɑ]; similarly [i] does not contrast with [e, ε]. Thus it is clear that the distributional facts alone do not resolve these phones into a unique phonological system. We now turn to a pre-sentation of our psycholinguistic study, which will offer support to a specific resolution of this phonemic system.

3. METHODOLOGY AND DATA COLLECTION

The use of speech errors, or ‘slips of the tongue’ to look at issues in linguistic theory as well as speech production planning models has a long

(15)

history (Fromkin (1973, 1980), Dell (1980), Cutler (1982), Stemberger (1983), Garrett (1988), Levelt (1989), Baars (1992), to name a few). A slip of the tongue (hereafter ‘SOT’) is defined as a one-time error made during speech production planning, such that the speaker produces an utterance which is at odds with his or her intended utterance, differing from the intended utterance in terms of lexical choice, syntactic structure, or phonological structure. While most of this research has focused on errors from English, Dutch, or German, a few recent studies have looked into cross-linguistic data (Berg (1987), Wells-Jensen (1999)), and we have recently begun publishing on speech errors in Mandarin (Wan (1997, 1999, 2000, 2001, 2002), Wan and Jaeger (1998); see also Shen (1993), Chen (1993, 1999), Yang (1997)). SOTs have been shown to be invaluable evidence for the cognitive status of specific linguistic units and processes in specific languages, and this is our rationale for using the SOT method-ology for looking at the vowel system of TM.

The current study is based on 238 SOTs involving vowels, selected TABLE 3

All possible contrasts between surface vowels

i  y u e ε ə ɔ o  a ɑ i – + + – – + – + + + +  – + – – – – – + + – y + – – + – – + + – u – – – – – + + – e – – – – – + – ε – – – – + – ə – – – + – ɔ – – + – o – – +  + – a – ɑ ‘+’ = contrast, ‘–’ = no contrast.

(16)

from a corpus of approximately 3000 SOTs collected by the first author from native speakers of TM between 1995 and the present. Some of the errors were gathered from tape-recorded free conversations, during which the subjects did not know they were being recorded. After each conversa-tion, the subjects were informed that the conversation had been recorded, and permission was obtained to use their data. Other errors were gathered following the ‘pen and paper’ method. That is, when slips occurred in situations where speakers were not being recorded (e.g., in normal con-versations, during lectures, etc.), the errors were immediately written in a notebook. For each error the researcher recorded the complete utterance including self-corrections, and relevant contextual information; portions were written in IPA phonetic transcription when appropriate. Thus SOTs will be reported below in terms of the actual pronunciations subjects produced during the error utterance.

Errors were collected from approximately 80 different speakers. Subjects ranged from monolingual to trilingual, with TM as their first language and English and Taiwanese as their other language(s) if any. However, all the errors were collected when the speakers were conversing in TM; any errors which showed a bilingual influence were not included in the data set to be examined in this paper. Therefore we believe that these errors accu-rately reflect the processing involving in speaking TM.

The usual way to collect SOT data is to rely on the native-speaker linguist’s intuitions as to what categories in the native language were heard by the native listener (Fromkin (1973)). This methodology is subject to some listener bias (Cutler (1982)). Furthermore, Mowrey and MacKay (1990) found that in SOTs induced in the laboratory by having speakers repeat ‘tongue twisters’ several times in succession, some phonetic differences between erroneously produced and intentionally produced consonants could be detected using electromyography. In the present study, one might argue that the perception by a native speaker that a syllable was spoken with a particular vowel is a more valid psycholinguistic measure than the actual phonetic properties of the utterance. However, the first author subjected several instances of erroneously produced vowels and the same vowels produced intentionally in the same environment to formant value analysis, and found no significant differences between the two (Wan (1999)). We thus assume that the data to be discussed below are sufficiently reliable to support our analyses.

In the earlier study by Wan (1999), it was found that SOTs in TM can occur at any stage of the speech production planning model (see Section 5.2 below for the details of our planning model). As indicated in the intro-duction, in this paper we are interested in errors which occur during the

(17)

phonological planning stage, where underlying phonological representations are being assigned a surface phonetic form. There are two types of errors which can occur during this stage on which we are focusing. First, one vowel can be erroneously substituted for another. Second, consonants can be sub-stituted, added or omitted next to a vowel, causing the vowel quality to shift; this can also involve a metathesis of a glide and vowel, causing the vowel quality to shift. These errors can occur contextually, that is, there can be a source for the error in the utterance itself; they can also be non-contex-tual, with no source in the error utterance. When an error is contexnon-contex-tual, and the error occurs in a word spoken before the source of the error, this is called an ‘anticipation’; if the error occurs in a word spoken after the source, it is a ‘perseveration’. Consonants and vowels can also be exchanged.

In all of the examples which we will present below, the following nota-tional format will be used: In the row headed by ‘I’, the intended TM utterance is given in IPA with the English glosses below; tones are given in tone numbers after the segments as follows: Tone 1 = 55, Tone 2 = 35, Tone 3 = 21,5Tone 4 = 51. Then in the row headed by ‘E’, the error utter-ance is given. Under the error utterutter-ance is a translation of the intended utterance. In every case, the phonological error resulted in an utterance which was meaningless, in the sense that it was either ungrammatical or uninterpretable, so we have not given translations of the error utterance. In the intended and error utterances, the ‘source’ unit(s) of the error (that is, the units which caused the error) are in boldface; the ‘target’ unit(s) (that is, the units planned for the intended utterance which were produced erro-neously in the error) are underlined; and the actual ‘error’ (the elements spoken erroneously) are both boldfaced and underlined. In cases where a consonant has been erroneously added, substituted, or omitted next to a vowel, causing the vowel to be realized in a different surface form, both the vowel and consonant are underlined as the ‘target’, and both boldface/underlined as the ‘error’. This is not meant to imply that the vowel was erroneously produced, but that the change in the vowel was made to accommodate the consonantal error (see Example (12) below). The two basic types of errors are as follows:

1. Phonological errors where one vowel is erroneously substituted for another, either due to the influence of another vowel in the same utter-ance, or non-contextually. If one vowel phone can be substituted for another, with the contiguous environment remaining the same, this proves that the vowel phones come from different underlying vowels. SOTs always produce sequences of segments which are legal in the language being spoken; if

(18)

the two phones were members of the same phoneme, substituting one for the other would produce an allophone disallowed in this environment, i.e., a disallowed phonotactic sequence, which never occurs. An example is given in (11). (11) I: thɑŋ55-mu21 k51-lu21-s → Tom Cruise E: th ɑŋ55-mu21 k51-lu21-su ‘Tom Cruise’

In this error the vowel [u] from either the syllables [mu21] or [lu21] is perseverated and substituted for the vowel [] in the syllable [s55], showing that [u] and [] are reflexes of differing vowel phonemes.

2. The second type of error which is of interest in this study is phonolog-ical errors in which the environment before or after the vowel is changed; that is, a contiguous segment is erroneously substituted, added, or omitted, or a vowel and contiguous segment metathesize. If changes occur in the vowel quality when these errors occur, then the two vowel phones can be taken to be conditioned variants of each other, as illustrated in (12).

(12) I: maj51-t i 55-tha21 Macintosh

E: maj51-t i 55-th 21 ‘Macintosh’

In (12), the [ŋ] from the syllable [t iŋ55] is perseverated and added to the end of the syllable [tha21]; in this context the [a] is realized as [

ɑ], suggesting that these two phones derive from the same phoneme.

Another possible way of interpreting such errors might be that when a consonant is erroneously substituted, added, or deleted, causing an illegal sequence of CV or VC, the vowel will be shifted to some other legal vowel phone in this environment, but that this is not necessarily an argument that the vowel produced in the error is an allophone of the intended vowel. And, as mentioned above, it certainly is a well-documented fact in the SOT literature that slips rarely produce an illegal sequence in the language (Fromkin (1973), Nooteboom (1973), Stemberger (1983)). The main problem with this argument, and particularly in terms of the example given in (12) above, is that in most cases there are a number of possible vowel phones which could occur legally in the environment which has been newly created by the error. So in the case of (12), the vowel phones [i, o, ɑ] could all legally occur in the vowel slot in the sequence [th_

ŋ]. If the shift in vowel were simply caused by the speaker ‘cleaning up’ the illegal

ŋ

(19)

string, we would expect that any of these vowel phones could be substituted for the intended phone [a] with equal frequency. But in fact in nearly every error where a vowel shifts quality when the consonant environment changes, the vowel alternates only with a single vowel phone (or in the case of the mid vowels, the same set of vowel phones) and not with other vowel phones which would be legal in this environment. In Section 4.3 below we document 85 cases in which the vowel phones [a] and [ɑ] alternated with each other in TM speech errors where the consonantal environment was shifted, and only one possible counterexample where [a] alternated with [ε]. This is extremely strong evidence that the analysis in terms of allo-phonic variation is the correct one, and certainly is in line with previous research on SOTs which documents that such allophonic variation is the norm (Stemberger (1983)).

An important methodological consideration has to do with making deci-sions on the classification of error types when more than one analysis is possible. Consider for example the error in (13).

(13) I: tan55- in55 wɔ21 fɑŋ35-ts →

worry I house

E: tan55- in55 wɔ21 fan35-ts ‘worry about my house’

One could make at least three arguments regarding this error. First, one could argue that the alveolar place feature from the two [n]s in the sylla-bles [tan55] and [ in55] had perseverated and affected the velar place of [ŋ] in [fɑŋ35], causing [ŋ] to be manifested as [n] and thus causing the vowel phone [ɑ] to front to [a]. Second, one could argue that the whole segment [n] had been substituted for [ŋ] also causing this vowel change. Finally one could argue that the rhyme [an] was perseverated from [tan55] and substituted for the rhyme [ɑŋ] in [fɑŋ35].

In making decisions on this type of error we use the ‘Minimal Movement Principle’ of Laubstein (1987), who argues that by far the most common unambiguous phonological error type is the substitution, addition, or omission of a single segment. Unambiguous feature errors are extremely rare in all languages, and Shattuck-Hufnagel and Klatt (1979) showed con-clusively that most errors where one segment changes to a different segment are in fact errors of whole segments rather than features. Similarly, Fromkin (1973) argues that features are not ‘independently controlled’ in speech pro-duction, and therefore are not as susceptible as segments to error. Furthermore, all speech error studies have found that unambiguous errors involving syllables, rhymes, or other sub-syllabic units are extremely rare;

(20)

however, one segment is more likely to substitute for another if both target and source have identical adjacent segments (the ‘repeated phoneme effect’, Dell (1984)). Thus all of these principles would lead us to analyze errors such as that in (13) as a case where the segment [n] has substituted for the segment [ŋ], causing the vowel change; the fact that both nasal codas were adjacent to the low vowel phoneme simply makes this nasal segmental substitution error more likely to occur because the codas share a common adjacent environment.

4. FINDINGS

4.1. Overall Findings

In looking at our data, we will ask the following questions.

1 Which vowel phones can substitute for each other in SOTs, without necessitating any other changes in the syntagmatic string?

2. When the environment contiguous to a vowel phone changes, which vowel alternations occur?

3. If some vowel phones are judged to be contextual variants of the same phoneme, is one of the allophones the more basic variant? In making this decision, we are making a decision about the structure of the under-lying vowel system of TM.

4. Finally, do our results concur with any of the four analyses of Mandarin vowels presented in Section 2.1 above, or do they require us to develop a new analysis appropriate for the vowel system of TM?

We will present general results here, and then discuss the various sets of vowel phones by vowel height in the following sections.

Our data revealed 41 errors in which one vowel substituted for another, with no other change in the error word, as illustrated in (11) above. Furthermore there were another 15 errors in which a glide, which we assume is derived from a high vowel, substituted for a vowel but then was realized as its vocalic alternant, or vice versa. An example is given in (14).

(14) I: εn21-li35 →

far-away E: εn21-ly35

‘far away’

In this error, the front rounded vowel [y] is realized as its glide alternant [ɥ] in the first syllable [ɥεn21] since it is adjacent to a non-high vowel. However, when it substitutes for the vowel [i] in the second syllable [li35],

ɥ ɥ ɥ ɥ ɥ ɥ

(21)

it then takes on its vocalic alternant, so that the error is produced as [ly35]. If we combine all the vowel-vowel substitutions with the vowel-glide substitutions, we find the pattern of vowel substitution shown in Table 4 (N = 56). This table will be discussed in detail below.

Second, we tabulated every error in which one vowel phone shifted to another due to a change in the environment in which the vowel occurred. The results are shown in Table 5 (N = 155). Again we will discuss this table in detail in the following sections.

Finally, there were 27 other errors in which vowels and glides inter-acted. In some cases a vowel was added or omitted, causing a vowel to become its glide alternant or vice versa. In other cases a vowel or glide was added in a new location, but was required to take on the opposite syllabicity due to its new syntagmatic status. These errors included 15 involving the [i-j] alternation, 8 involving the [w-u] alternation, and 1 involving the [y-ɥ] alternation, which is given in (15). There were also 3 errors where a vowel substituted for a coda nasal, causing the glide alter-nant to occur (two involving [w] and one involving [j]; [ɥ] does not occur in coda position so is not eligible for this type of error).

TABLE 4

The number of errors in which one vowel phone substituted for another

i  y u e ε ə ɔ o  a ɑ i 19 5 1 1 1  7 y 7 u 2 1 e 2 ε ə 4 ɔ o 6  a ɑ

(22)

(15) I: thjεn55 ɥεn35 ti51 t wan21 →

sky turn ground turn

E: thjεn55 yn35 ti51 t wan21 ‘The whole world is turning over.’

In (15), the vowel [ε] was deleted in dissimilation from the vowel [ε] in the preceding word; when the /y/ found itself in syllable nucleus position, the vowel [y] rather than the intended glide [ɥ] was produced. We take these errors to be extremely strong evidence that glides derive from high vowel phonemes. One other possible interpretation would be that when the vowel was erroneously deleted, the illegal sequence *CGV was produced, with no vowel nucleus; in this case the glide was replaced by some vowel by a sort of ad hoc on-line processing constraint. However, if this sort of process was common, we would expect to find that there would be no consistent relationship between the glide and the vowel which substituted for it (or vice versa), when in fact in every error of this sort the

relation-TABLE 5

The number of errors in which a vowel phone shifts to a different vowel quality when the adjacent environment changes

i  y u e ε ə ɔ o  a ɑ i 9  y u e 3 06 7 1 3 ε 11 1 4 1 01 ə 3 3 2 ɔ 5 8 o 2  a 85 ɑ

(23)

ship between [i]-[j], [y]-[ɥ], and [u]-[w] was consistent (N = 42). Furthermore, there were no errors in our corpus which involved the creation of an illegal phonotactic sequence; for example, there were no errors where a vowel was deleted, producing a syllable such as *CF with no nucleus. This suggests that the processing mechanism is treating the glide as if it is a nucleus in underlying structure, so if there are two potential nuclei in sequence in a syllable, the non-high vowel may be deleted. The gliding which appears on the surface is a later process which occurs after the SOTs have occurred, if a sequence of high + non-high vowel (or vice versa) remains in the surface form. For all these reasons we will take as a given in the remainder of this paper the hypothesis that glides are derived from high vowel phonemes, which is consistent with the theories of Lin (1989) and Wu (1994) discussed above. We will now turn to a detailed discussion of groups of vowel phones, starting with the high vowels.

4.2. High Vowels

Following the above guidelines, the high vowel errors will be interpreted in terms of the following questions:

1. In substitution errors involving high vowels [i, y, , u], are these phones all equally likely to be substituted for each other?

2. When the environment before or after these four high vowel phones changes, do any of these phones alternate with each other?

3. Are there any errors in which the vowel [y] shows evidence of being underlyingly a sequence of [u] and [i], as in Wu’s (1994) proposal? That is, are there errors in which it appears that only part of this sequence participates?

4. What are the underlying high vowel phonemes of TM?

Looking first at Question 1, we found 25 cases involving substitution among the high vowels [i, u, y, ], and another 13 substitution errors involving interactions of glides with high vowels. Of the 25 simple vowel substitutions, 10 involved the interaction between [i] and [y], as exempli-fied in (16); 9 of the vowel/glide interaction errors also involved [i/j] and [y/ɥ].

(16) I: phiŋ35-t yn55 phiŋ35-t yn55 → average average

E: phiŋ35-t yn55 phiŋ35-t in55 ‘pretty much average’

(24)

In Example (16), the vowel [i] is perseverated and substituted for the vowel [y], but since [t in] is an allowable sequence in TM, no further changes in the word are required. The next most common substitution (N = 7) involved an interaction between [u] and [], as illustrated in (17).

(17) I: tha55 51 tswɔ51 ts51-t u51-tshan55 he is do cafe

E: tha55 51 tswɔ51 tsu51-t u51-tshan55 ‘he has a café’

In Example (17), the vowel [u] is anticipated and substituted for the vowel [], producing the legal sequence [tsu]. An equally common substitution involved [u] and [y]; 5 vowel substitution errors and 2 vowel/glide errors involved this pair of phones, as illustrated in (18).

(18) I: lu51 t y51 wɔ21-mən xɑw21 t in51 →

deer distance us very close

E: ly51 t y51 wɔ21-mən xɑw21 t in51 ‘The deer is very close to us.’

In this error, the vowel [y] is anticipated and substituted for the vowel [u]. The remaining 3 vowel substitution errors and 2 vowel/glide errors involved the pair [i] and [u], as exemplified in (19).

(19) I: pu35 jɑw51 thjɑw35 phi35 not want naughty

E: pi35 jɑw51 thjɑw35 phi35 ‘don’t be naughty’

In Example (19), the vowel [i] is anticipated and substituted for the vowel [u]. We found no cases where [i-] or [y- ] substituted for each other. This is to be expected, since [] only occurs after dental and retroflex obstru-ents, and [i, y] never occur in these environments.

These data suggest that /i, y, u/ are separate phonemes, and that [] and [u] belong to distinct phonemes. Table 4 also shows that there are vowel substitution errors involving [i]-[ə, , a], and [u]-[, a], verifying that these phones belong to separate phonemes. However, the vowel substitu-tion data do not tell us whether [] should be grouped with either [i], or [y], or neither.

In order to look into this issue, we need to look at errors where the environment changes, causing a concomitant change in the vowel. However, it turns out that unambiguous errors of this sort are difficult to find for high vowels, since the consonantal environment which is relevant for

(25)

con-ditioning the occurrence of these three high front or central vowel phones, palatalization, could be considered to be a function of the vowel quality. Thus we are in danger of falling into a circular reasoning pattern. Consider the following examples:

(20) a. I: tjεn51-t i55 pwɔ35-s51 pan55 → electrical-engineering Ph.D. class

E: tjεn51-t i55 pwɔ35- i51 pan55

‘Ph.D. program in the Department of Electrical Engineering’ b. I: koŋ55-tswɔ51 t i55-xwej51 →

job opportunity

E: koŋ55-tswɔ51 ts55-xwej51 ‘job opportunity’

Wan (1999) argued that these errors show either a substitution of the feature of palatalization (as in (20a)) or a deletion of the feature of palatal-ization (as in (20b)) in the consonant, due to the influence of other palatal or non-palatal consonants in the utterance. In the previous analysis it was not claimed that (20a) showed a substitution of [ ] for [s], since these two segments are considered to be derived from the same underlying phoneme; similarly in (20b), [t ] and [ts] are considered allophones of the same phoneme. This grouping of the dentals and palatals into single phonemes was argued for due to the fact that whenever the palatal feature is deleted (as in (20b)), the consonant reverts to the dental place of articulation; sim-ilarly, if a palatal glide is inserted after a dental obstruent, it is realized as a palatal obstruent. However, one could argue that these are vowel sub-stitution errors. In (20a), one could claim that the [i] from [t i55] was perseverated and substituted for the [] in [s51]. But in (20b) it would be difficult to argue that there is a substitution of the vowel [] for [i], since there is no [] elsewhere in the utterance which could serve as the source of the error, and such substitutions are overwhelmingly contextually moti-vated. Therefore, if we accept the Wan (1999) analysis of these errors for the sake of argument, then our corpus yields 9 errors in which the conso-nantal environment shifts from dental or retroflex to palatal (as in (20a)) or vice versa (as in (20b)), and in every case, the high front unrounded vowel [i] alternates with the high central unrounded vowel []. There are no cases in our errors where the environment shifts and [] shows an alternation with the other palatal vowel, [y]. Specifically, there are no cases where the palatal environment is added, such as in (20a) above, and the intended vowel [] is realized as [y], even though the sequences [ y], [t y] and [t hy] are legal. Furthermore, there are no errors in our data where the palatal

(26)

environment is removed before [y]; we suspect that this is because this would result in an illegal sequence of a non-palatal consonant + /y/, and /y/ does not have a non-palatal allophone to revert to. Two other examples involving consonant addition may serve to strengthen this argument.

(21) a. I: ən35 jɑw51 ts55-tsu35 → people need content

E: ən35 jɑw51 55-tsu35

‘People need to be content with what they have’ b. I: i35-tiŋ51 jɑw51 khan51

san55- 35 fən55 u55 →

must need study 30 minute book

E: s35-tiŋ51 jɑw51 khan51

san55- 35 fən55 u55 ‘(you) must study for 30 minutes’

In (21a), the nasal [n] is perseverated from [ ən] and added after the vowel [] in [ts55-tsu35].6 As shown in Table 2, the vowel [] can only occur in open syllables in TM, and thus sequence *[tsn] is not legal; therefore the vowel shifts to its legal allophone [i], which can occur before [n]. This causes the preceding affricate to palatalize from [ts] to [t ]; (note that the sequence [t yn] is also a legal sequence, but does not occur in this error). In (21b) we find the opposite case: the [s] from [san55] is anticipated and added into the onset slot of [i35]; since the sequence of dental fricative or affricate + [i] is not allowed in TM, the vowel [i] shifts to the appropriate allophone [] after [s]. Thus there is clear evidence that the high central vowel [] should be grouped together with the high front unrounded vowel [i] into a single phoneme, since it is in complementary distribution with this phone, and these two phones alternate when the consonantal environment shifts. There is no reason to consider grouping it with the rounded vowel [y]; although [y] and [] do not contrast, they also never alternate in any of these speech errors.7

Moving on to Question 3, we find that there are no cases where the vowel [y] splits, and only one part is involved in an error. In every case where [y] is involved in an error, it behaves identically to every other vowel phone in the data, i.e., like a whole unit rather than a sequence of units. This is illustrated in Examples (15), (16), and (18) above. More evidence for the fact that [y] is not derived from a sequence of /ui/ is that these three vowels [i, y, u] freely substitute for each other as whole segments; we have no cases where, for example, a sequence of vowels substitutes for a single vowel, or vice versa. We do find it a bit odd that we found only one error where [y] and [ɥ] alternate with each other when the

(27)

ment shifts, as we found several for [i-j] and [u-w]; but this may be due to the fact that the front rounded vowel/glide phoneme is far less frequent than the other two high vowel phonemes. Finally, we are in principle opposed to an analysis which derives the high front rounded vowel/glide from a sequence of underlying high vowels, since in all other theories the only possible sequences of vowels in phonological representations must involve a high with a non-high vowel. This proposal makes a very dif-ferent claim about the possible underlying structures of morphemes in Mandarin, one which would require more supporting evidence. We there-fore see no reason to consider the vowel /y/ as anything other than a phoneme in its own right.

In summary, our findings show that the vowels [i, y, u] and the vowels [, u] belong to separate phonemes, since they can be substituted for each other. The vowels [i, ] alternate with each other when the preceding envi-ronment alternates between palatal and non-palatal, showing that they are allophones of a single phoneme. There is no evidence to support the claim that the vowel [y] is derived from a sequence of /ui/, since [y] never splits into components in errors, and can substitute for both [u] and [i] in errors.

The final remaining question is: What is the basic phonemic system of the vowels of TM? Because the phone [i] occurs in far more environments than [], we will assume that [i] is the basic variant, which gives us the universally more common vowel system /i, y, u/ (as compared to */y, , u/).

Comparing our findings to the four theories discussed in Section 2.1 above, we find that our analysis is different from all four in that we claim that there is a single high central unrounded vowel [] in TM rather than two distinct ‘apical’ vowels. Other than this difference, however, our analysis is most similar to the 5-vowel system of R. Cheng (1966), who considers /i/, /y/, and /u/ to be distinct phonemes, and derives the apical vowels from /i/. We differ from C. Cheng who considers /i/ and // to be distinct phonemes. We also differ from Lin who derives the apical vowels from an epenthesis and assimilation process. Finally, we differ from Wu in that, even though she considers /i, u/ to be phonemes, and derives the apical vowels from /i/, she considers [y] to be derived from a sequence of /ui/; we find no psycholinguistic evidence for this latter analysis.

4.3. Low Vowels

Errors involving the low vowels will be interpreted in terms of the following questions:

(28)

1. Do the vowel phones [a] and [ɑ] ever substitute for each other in an error, with no other changes?

2. If the environment contiguous to one of these vowel phones changes, does the central phone [a] shift to the back phone [ɑ] and vice versa?

3. Are there any cases where either of the two low vowel phones [a, ɑ] alternate with the lower-mid front vowel [ε] when the environment changes, as would be predicted by the four analyses discussed in Section 2.1 above?

4. If [a] and [ɑ] alternate, which is the basic variant?

Looking first at Question 1, we find no cases in our data where [a] substituted for [ɑ] or vice versa with no other changes in the word, as shown in Table 4. It can be seen from this table that there are substitution errors involving [a]-[i, u, e, ə], and [ɑ]-[o], verifying that these high and mid vowel phones contrast with the low vowel phones, and thus should not be grouped into phonemes with [a] or [ɑ].

Table 5 shows a clear answer to our second question: We have 85 errors in which the environment contiguous to one of these two low vowels changes, causing the vowel to shift to the other variant. Seventy-three of these errors involve the substitution of one segment for another in the coda slot following one of these phones; in 30 cases, the non-velar codas [n, j] are substituted for the velar codas [ŋ, w], causing [ɑ] → [a], as illus-trated in (22), and in 39 cases the reverse occurs, as illusillus-trated in (23); in the remaining 4 cases the nasals [n] and [ŋ] exchange with each other, causing the vowels to shift.

(22) I: jεn21-kwɑŋ55 → eye-light

E: jεn21-kwan55 ‘judgment’

In Example (22), when the nasal [n] is perseverated and substituted for the nasal [ŋ], the vowel [ɑ] changes to [a].

(23) I: nan35-thi 55 s

21 l → hard-listen dead

E: n 35-thi 55 s 21 l ‘very bad music’

In Example (23), when the nasal [ŋ] is anticipated and substituted for the nasal [n], the vowel [a] changes to [ɑ].

ŋ

(29)

Addition and omission errors give further evidence regarding Question 2. Our data contain 9 errors in which the vowel [a], which is in an open syllable in the target word, is realized as the back [ɑ] when the velar codas [w, ŋ] are erroneously added after the vowel, as exemplified in (24).

(24) I: lwɔ35-ma21 → Rome

E: lwɔ35-m 21 ‘Rome’

In Example (24), the glide [w] is perseverated and added after [a], and [a] changes to [ɑ].

Similarly, we found 3 errors in which [ɑ] is realized as [a] when the coda glide [w] or nasal [ŋ] is erroneously omitted, as illustrated in (25).

(25) I: tsu51-t jɑw51 pi21 sej35 tow55 ta51 → assistant than anyone much powerful E: tsu51-t ja51 pi21 sej35 tow55 ta51

‘The assistant is more powerful than anyone else’

In Example (25), the glide [w] in [t jɑw51] is omitted in dissimilation from the [w] in [tow], causing the [ɑ] to occur in an open syllable so that it must be realized as the central [a] variant. From these substitution, addition, and omission errors we find very strong evidence that [a] and [ɑ] are contextual variants of the same phonemic category, as would be predicted by all four theories discussed above.

However, Question 3 is more problematic. First, we found no errors in our corpus where [ε] substituted for [a] or [ɑ]. One could argue that this supports the claim that [ε] could be grouped with the low vowels into one phonemic category. However, when we look at errors where the environ-ment changes and vowels shift quality, we find a very different pattern. In every case except one, when the environment contiguous to [ε] changes so that this is no longer a legal phone in the new environment, the [ε] shifts to one of the other mid vowels; similarly, if the environment con-tiguous to one of the mid vowels is changed so that [ε] is the appropriate allophone, all the other mid vowels shift to [ε] (N = 20); these figures are shown in Table 5, and will be discussed in detail in Section 4.4 below. This includes 15 errors in which the erroneous environment is exactly the environment where Lin (1989) argues the [a-ε] alternation should occur. Consider the following example.

(30)

(26) I: tsan51-tsaj51 i55-pjεn55 → stand in one-side E: tsan51-tsaj51 i55-pən55

‘(People are) standing together to one side’

In this example, the prenuclear glide [j] is omitted in the syllable [pjεn55], in dissimilation from the [j] in [tsaj51]. Because there is no longer a palatal prenuclear glide in this syllable, the vowel must shift to another phone which is allowed in the sequence [p__n]. The crucial factor is that both [pən] and [pan] sequences are allowed in TM, and yet the error the speaker actually produced was [pən], which argues strongly that the underlying phonological representation of the target word is /piən/ and not /pian/. It is clear from the unambiguous data discussed above regarding the [a-ɑ] alternation, that the default production of an error word will maintain the same phoneme as in the target word if possible, when the environment next to the phoneme changes, even if some other phoneme can occur there. Otherwise we might expect, for example, that the [a] in [nan35] in Example (23) above might change to an [o] when the [ŋ] substitutes for the coda [n], since the sequence [noŋ35] is legal in TM. However, this sort of random vowel change after a consonant substitution doesn’t occur in our data. Thus, since we have 20 errors in which [ε] alternates with other mid-vowel phones when the environment is erroneously changed, and only one error in which this does not occur, this is very strong evidence for the affiliation of [ε] with the mid vowel phoneme in TM and not the low vowel phoneme.

The one counterexample is given in (27). (27) I: t o 55-t jεn55 t jow51- i 35 l →

middle just okay

E: t o 55-t j 55 t jow51- i 35 l ‘(putting it) in the middle is okay’

In this error the consonant [ŋ] was either anticipated or perseverated and substituted for the consonant [n] in the syllable [t jεn], giving the illegal sequence *[t jεŋ], since [ε] doesn’t occur before the velar nasal. In this case, the vowel [ε] was realized as the low back [ɑ] in the error. If we wanted to argue that the [ε] derives from the mid vowel phoneme, we would have expected that the error phone would be [o], since a sequence of [t joŋ] is legal for at least some TM speakers, as in the word [t joŋ21], an archaic word meaning ‘embarassed’; however, this is the only word with this sequence in TM. If the speaker who made the error in (27) knew this word, then we would have to conclude that there may be some affiliation

ŋ ŋ

(31)

of [ε] with /a/ for this speaker, since in this one case the two alternated and the expected [o] did not occur. However, if this speaker did not know this word, then this would be a simple case of the speaker substituting the [n] with [ŋ], and then producing the closest legal sequence possible, which would be [t jɑŋ]; in this case this error would not be as strong a counterexample to the claim that [ε] and [ɑ] are not affiliated with each other. Even if this were a counterexample, we would argue that because there were 20 errors which showed an alternation between [ε] and the other mid vowels, and only this one case showing the [ε]-[ɑ] alternation, we would argue that there is much more evidence for the inclusion of [ε] in the mid vowel than the low vowel phoneme in general. We will return to this point below in our discussion of mid vowels.

Finally, turning to Question 4, there is clear evidence that [a] is the basic variant of the phoneme which includes the [a-ɑ] alternants. This is because whenever the coda consonant is omitted following one of these two phones, only the phone [a] is realized; since this is the phone which occurs when there is no conditioning environment (i.e., no environment which would cause a backing or fronting assimilation), it must be the basic variant. This phenomenon was illustrated in Example (25) above.

In summary, there are 85 errors in our corpus in which the [a] and [ɑ] phones alternate; these alternations were caused by the substitution, omission, or addition of consonants after the vowel, such that only [a] occurred before [n, j] or syllable finally, and only [ɑ] occurred before the velar consonants [w, ŋ], clearly a place of articulation assimilation. Because only [a] occurs in open syllables with no conditioning environment, the low vowel phoneme is considered to be /a/. Finally, we found no evidence that [ε] should be considered a member of the low vowel phonemic category; we will return to this point in the next section.

4.4. Mid Vowels

Data regarding the grouping of mid vowels will be interpreted in terms of the following questions.

1. Can any of the mid vowel phones [e, ε, ə, ɔ, o, ] be substituted for each other without any other change in the contiguous environments? 2. Which of these phones alternate with each other when the contiguous

environment is changed?

3. What are the details of the evidence discussed in Section 4.3 that [ε] should be grouped with the mid vowels rather than the low vowels? 4. How many vowel phonemes are represented by these mid-vowel phones,

(32)

Regarding Question 1, we found no cases in our data where any of these mid vowel phones substituted for each other with no other changes in the syllable. Table 4 shows the following substitutions involving mid vowels: [ə]-[i, a], [e]-[a], [o]-[ɑ], []-[u, i], indicating that mid vowels belong to distinct phonemes from both the high vowels [i] and [u], as well as the low vowels [a] and [ɑ].

In our data we have 60 errors which involve alternations among [e, ε, ə, ɔ, o, ], as shown in Table 5, which provide evidence for Question 2. In 19 cases, one consonant substitutes for another either in the onset or coda of the target word, causing the target mid vowel to shift to one of the other phones. This includes 6 errors involving the [ə]-[e] alternation as exemplified in (28), 5 errors involving [ə]-[ε] as exemplified in (29), 4 errors involving [ε]-[o] ((30)), 3 errors involving [ə]-[o] ((31)), and one error involving [ε]-[ɔ], given in (32).

(28) I: nej51-ljεn51 → reserved E: n 51-ljεn51

‘reserved’

In Example (28), the dental nasal [n] was either perseverated or antici-pated and substituted for the glide [j] in the coda of the target word [nej51]. Since the [e] allophone can only occur before the glide [j], the vowel is realized as [ə] which is the correct allophone before [n].

(29) I: pu35 ɥεn51 tu35-su55 → not want study

E: pu35 n51 tu35-su55 ‘(she) doesn’t want to study’

In this error, the vowel [u] was perseverated or anticipated and substituted for the [ɥ] in the onset slot of the target word [ɥεn51]; in this context the /u/ is realized as the glide [w]. Since [ε] cannot occur after [w], the correct allophone for this environment, [ə], was spoken in the error. Notice that this is another case where [a] would have been legal in the error utterance, since the utterance [wan] is legal in TM; however, in this error what was actually spoken was the vowel [ə].

(30) I: min35- ɥoŋ35 → Ming-Xiong E: min35- ɥ 35

‘Ming-Xiong (county name)’ ən

數據

Figure 1.  A speech production planning model.

參考文獻

相關文件

(a) Find the unit vectors that are parallel to the tangent line to the curve at the point.. (b) Find the unit vectors that are perpendicular to the

(c) Draw the graph of as a function of and draw the secant lines whose slopes are the average velocities in part (a) and the tangent line whose slope is the instantaneous velocity

remember from Equation 1 that the partial derivative with respect to x is just the ordinary derivative of the function g of a single variable that we get by keeping y fixed.. Thus

When the transfer is complete, a STOP bit (P) is sent by releasing the data line to allow it to be pulled up while SCL is constantly

Proof : It is clear that three numbers are vertices of triangle inscribed in the unit circle with center at the origin... So, it is clear that axiom

The format of the URI in the first line of the header is not specified. For example, it could be empty, a single slash, if the server is only handling XML-RPC calls. However, if the

a single instruction.. Thus, the operand can be modified before it can be modified before it is used. Useful for fast multipliation and dealing p g with lists, table and other

Courtesy: Ned Wright’s Cosmology Page Burles, Nolette & Turner, 1999?. Total Mass Density