Dissimilation in the second language acquisition of Mandarin Chinese tones

Abstract

This article extends Optimality Theoretic studies to the research on second language tone phonology. Specifically, this work analyses the acquisition of identical tone sequences in Mandarin Chinese by adult speakers of three non-tonal languages: English, Japanese and Korean. This study finds that the learners prefer not to use identical lexical tones on adjacent syllables, especially the contour tone sequences. It is argued that the Obligatory Contour Principle (OCP) was playing a role in shaping the second language Chinese tonal phonology even though it was not learned from these speakers’ native languages, nor found widely applied in the target language. The acquisition order of tone pairs suggests an interacting effect of the OCP and the Tonal Markedness Scale. This study presents a constraint-based analysis and proposes a four-stage path of OCP sub-constraint re-ranking to account for the error patterns found in the phonological experiment.

Keywords

Mandarin Chinese constraint re-ranking dissimilation Obligatory Contour Principle Optimality Theory second language tones Tonal Markedness Scale

I Introduction

In the past few decades, research on second language (L2) phonological acquisition has presented evidence that, not only does one’s native language or first language (L1) have an effect on L2 sound patterns, but also that interlanguage phonologies are restricted by Universal Grammar (UG). For example, the Ontogeny Phylogeny Model (OPM; Major, 2001) claims that the interlanguage phonology is made up of elements of the L1 (through transfer), the L2 (which is acquired), and universals (which include UG). Past research has uncovered patterns that appear to be derived, neither from learners’ native language nor from the target language, but instead from universal preferences (for reviews of relevant literature, see Epstein, et al., 1996; Eckman, 2004). Although evidence has been discovered that phonological universals play a role in L2 acquisition, it is unclear exactly how universals can be accessible to L2 learners and the extent to which phonological principles shape the interlanguage grammar.

Advancements in L2 phonology studies within the framework of Optimality Theory (OT; McCarthy and Prince, 1993a, 1995; Prince and Smolensky, 1993) have furthered our understanding of interlanguage properties and L2 learners’ access to phonological universals (see, amongst others, Broselow et al., 1998; Hancin-Bhatt, 2000; Lombardi, 2003). While most of the L2 phonological studies within OT have focused on the acquisition of syllable structure (see, amongst others, Broselow et al., 1998; Hancin-Bhatt and Bhatt, 1997; Hancin-Bhatt, 2000; Hayes, 2000; Shepherd, 2003; Cardoso, 2007, 2011), this study extends L2 phonological studies within the framework of OT to research on lexical tone acquisition and explores the effect of dissimilation in L2 production of modern standard Chinese (hereafter, Mandarin Chinese).

Dissimilation is a cross-linguistically common phenomenon in which similar or identical sounds in close proximity are avoided. Dissimilation can be found in static generalizations over a language’s lexicon, such as in Arabic roots where adjacent homorganic consonants are avoided (Alderete et al., 2013). Dissimilation can also be found in phonological processes, such as in Tashlhiyt Berber, where delabialization is triggered by two primary labial consonants in the same derived stem (Alderete and Frisch, 2007). Although it has been claimed that dissimilatory processes can occur with any phonological feature (Suzuki, 1998), the most common cases of dissimilation involve the dissimilation of tone, place of articulation, and laryngeal features (Alderete, 2003).

Studies on tone and prosodic phonology in the 1970s and 1980s have advanced a generative approach to dissimilation (see, amongst others, Leben, 1973; Goldsmith, 1976). In autosegmental phonology, dissimilation is understood as an effect of the Obligatory Contour Principle (OCP).

(1) Obligatory Contour Principle (Leben, 1973; Goldsmith, 1976; McCarthy, 1986): Adjacent identical autosegments are prohibited.

The OCP was originally developed to explain tonal dissimilation phenomena observed in Mende and other African tone languages (see Leben, 1973; Goldsmith, 1976). After its original conception, the OCP was argued to be a universal constraint, differing across languages only in ranking with respect to other constraints in languages in the spirit of OT (Yip, 2002). This study tests for evidence of the operation of the OCP in L2 tonal phonologies.

The acquisition of Mandarin Chinese lexical tones has consistently been regarded as a significant challenge for speakers of non-tonal languages (Lado, 1968). Previous research has found that English speakers acquiring tones tend to disfavor identical tone sequences (H Zhang, 2010). This difficulty with identical tone sequences is surprising, especially for identical contour tone sequences, since there is no equivalent of lexical contour tones in L1 English, and the OCP is not prominent in Mandarin Chinese (only seen in the T3 sandhi process). Therefore, if we suppose that a learner’s interlanguage grammar is made up solely of elements from the learner’s L1 and the L2, the difficulty found with identical contour tone sequences by L2 learners of Mandarin Chinese is left without an explanation. It is thus hypothesized that, even as adults, learners may still have some access to phonological universals, in particular the OCP. While OCP effects have been found in the L2 tones produced by English speakers by H Zhang (2010), no cross-linguistic studies have been carried out to test for the universality of these effects. The research in the present article seeks to fill this gap and determine whether and how the OCP functions in multiple interlanguage tones.

This study examines L2 tones produced by 60 speakers of American English, Tokyo Japanese, and Seoul Korean (henceforth English, Japanese, and Korean). Each of these languages represents one of three different types of non-tonal languages according to the characteristics of their word prosody (Jun, 2005): stress–accent languages (English), lexical–pitch–accent languages (Japanese), and non-stress and non-lexical–pitch–accent languages (Korean). Additionally, the speakers of English, Japanese and Korean make up the majority of the population of L2 learners of Mandarin Chinese worldwide (Hu, 2008). We look into how the three types of speakers learn the rich inventory of Chinese tones while simultaneously restricting some pitch patterns that exist in their L1 phonologies.

This article will first review basic concepts of Optimality Theory (Section II) and the Mandarin Chinese tone grammar (Section III). Sections IV and V will present the methods and results of an experiment designed to analyse the aspect of learners’ interlanguage grammars regarding tone pairs. Section VI will propose four stages in the development of a tonal grammar by a speaker of a non-tonal language.

II Optimality Theory and tone acquisition

1 Basic concepts of Optimality Theory

Analyses of Chinese tones within the theoretical framework of Optimality Theory (OT) have generated a good amount of insight (see, amongst others, Yip, 2002; J Zhang, 2009). Despite this, OT has had little influence within the subfield of L2 tone acquisition. This section introduces the basic concepts of OT through an example of an OT analysis of the Tonal Markedness Scale (TMS).

It is widely acknowledged that the simpler a tone structure, the more likely its occurrence in languages (Gordon, 2001; J Zhang, 2002). Contour tones are more complex (more marked) than level tones, since contour tones contain more pitch targets linked to a single Tone-Bearing Unit (TBU). Among the contour tones, falling tones are both easier to produce and perceive compared to rising tones (Cheng, 1973; Ohala, 1978). This phonetically-grounded tone asymmetry is reflected in the distributional property of tones in natural languages. For example, a survey of 187 genetically-diverse tone languages found that 37 languages do not allow rising tones, while only three do not allow falling tones (J Zhang, 2002). These differing tone inventories can be modeled within OT, using constraints such as those shown in items (2)–(5) below.

OT incorporates linguistic universals (represented with markedness constraints) in a formal theory of a speaker’s phonology, arguably explaining a wider range of phenomena in L2 phonological acquisition than rule-based theories (Broselow et al., 1998; Barlow and Gierut, 1999; Hancin-Bhatt, 2000; Lombardi, 2003; Eckman, 2004). OT grammars consist not of a set of rules as assumed in earlier generative frameworks, but rather of a set of ranked constraints, which define the optimal output corresponding to an input string. Two types of constraints are used in the process of evaluation to pick the optimal output: (1) markedness constraints, which require output forms to meet some criterion of structural well-formedness; and (2) faithfulness constraints, which require the output to be identical to the input in some particular way. Constraints are assumed to be universal: differences between languages are a result of different rankings of these universal constraints. Items (2)–(4) list Tonal Markedness Constraints (Ohala, 1978; Yip, 2002; Hyman and VanBik, 2004) relevant to the present discussion. *Contour is an abbreviation of *Rise and *Fall, which are separated in (2). High and Low pitch targets are abbreviated as H and L.

(2) *Contour: No contour tone (two varying pitch targets linked to a TBU) is allowed.

a. *Rise: No rising tone (two pitch targets LH linked to a TBU) is allowed.

b. *Fall: No falling tone (two pitch targets HL linked to a TBU) is allowed.

(3) *Level: No level tone (one pitch target linked to a TBU) is allowed.

(4) *Rise >>*Fall >>*Level: Rising tones are more marked than falling tones; falling tones are more marked than level tones.

Item (5) displays some typical faithfulness constraints used for tonal grammar. However, this study will be grouping relevant faithfulness constraints as Faith in discussions since the rankings of individual faithfulness constraints are not relevant to the argument.

(5) Example faithfulness constraints (Yip, 2002)

a. Max(T): Every tone in the input has a correspondent in the output (‘do not delete tones’).

b. Dep(T): Every tone in the output has a correspondent in the input (‘do not insert tones’).

c. Ident(T): Let α be a TBU in the input, and β be any TBU corresponding to α in the output. If α has tone T, then β also has tone T.

We follow Prince and Smolensky (1993) in assuming that the sound inventory of a language is derived by the ranking of faithfulness constraints relative to markedness constraints. Specifically, if faithfulness constraints related to property X are higher-ranked than markedness constraints related to X, then X will be allowed by the language. On the other hand, if markedness constraints related to property X are higher-ranked than faithfulness constraints related to X, then X will not be allowed by the language. Therefore, in a tonal language that does not allow rising tones, such as the African language Lama, the constraint *Rise is ranked higher than Faith, banning surface rising tones. Item (6) demonstrates a partial ranking of this grammar. Rising tones in the input never appear in the surface, while both falling and level tones are allowed. The ranking of *Fall relative to *Level cannot be determined, since this language allows both falling and level tones. This unspecified ranking is indicated with a dashed line. The dominance hierarchy thus can be represented as *Rise >> Faith >> *Fall, *Level.

(6) Tableau for a partial grammar which does not allow rising tones, with the sample input form /LH/

/LH/	*Rise	Faith	*Fall	*Level
LH	*!
☞ HL		*	*
☞ HH		*		*

Notes. The symbol * denotes a violation of the constraint and ☞ signifies the winner.

These constraints in different relative rankings can model at least three types of attested tone languages as shown in (7):

(7) a. Type 1 (Panoan languages) (Gordon, 2001): Contour tones are not permitted, but level tones are allowed.

*Rise, *Fall >> Faith >> *Level…

b. Type 2 (Lama) (J Zhang, 2004): Rising tones are not permitted, but falling and level tones are allowed.

*Rise >> Faith >> *Fall, *Level…

c. Type 3 (Mandarin Chinese): Both contour and level tones are allowed.

Faith >> *Rise, *Fall, *Level…

As we can see from the tonal grammars listed in (7), within the framework of OT, differences in languages arise not from different constraints, but from different rankings of the same set of constraints. Constraint ranking can not only be used to account for natural language inventories, but also explain sound patterns found in language acquisition.

2 OT in language acquisition studies: Grammar restructuring

The Tonal Markedness Scale (TMS), defined as *Rise >> *Fall >> *Level (Hyman and VanBik, 2004; Ohala, 1978), is reflected statically in the tone inventories of natural languages, and also reflected in the dynamic process of language acquisition. It is reported that Mandarin Chinese-speaking children produce level and falling tones earlier than rising tones (Li and Thompson, 1977; Zhu and Dodd, 2000). In OT, language acquisition can be described as the process of adjusting the ranking of these constraints to match that of the language being acquired. The acquisition order in L1 acquisition mentioned above can thus be understood to begin with a Type 2-like grammar as shown in (7), eventually demoting *Rise to converge on the adults’ Type 3 grammar.

As with children acquiring Mandarin Chinese as an L1, adults acquiring Mandarin Chinese as an L2 also show effects of the TMS when re-ranking their constraints. Studies that have focused on the order of acquisition of Mandarin Chinese tone types for L2 learners have reported various findings (see a review in Sun, 1998; H Zhang, 2013). The majority of these studies have found that L2 learners acquire the high level tone first (i.e. with the fewest errors), and the high falling tone earlier than the rising tone. Therefore L2 learners may experience a similar constraint re-ranking process as L1 learners.

At least initially though, the process of constraint re-ranking during language acquisition differs in L1 and L2 acquisition. In L1 acquisition, all markedness constraints are typically assumed to dominate all faithfulness constraints initially. Then, during the course of L1 acquisition, either only markedness constraints are demoted (Tesar and Smolensky, 1998, 2000), or markedness constraints are demoted while faithfulness constraints are promoted (Boersma and Levelt, 2000). On the other hand, since L2 learners start their acquisition with the L1 constraint rankings (see Young-Scholten, 1994; Archibald, 1998), L2 learners must eventually arrive at the target ranking through a series of re-rankings. Grammar restructuring during language acquisition is assumed to be error-driven (Tesar and Smolensky, 1998). That is, when the input forms are inconsistent with the grammar learners originally possessed, learners will adjust constraint rankings so as to produce the correct outputs. For example, if the input violates a high-ranked markedness constraint, this will lead to the demotion of that markedness constraint. A few OT studies on L2 acquisition of syllable structure suggest at least two types of constraint re-rankings: the gradual demotion of initially high-ranked markedness constraints for learners expanding their range of target structure (for example, Thai speakers acquiring English syllable structure; Hancin-Bhatt, 1998, 2000, 2008), and the gradual demotion of faithfulness constraints for learners restricting their range of target structure (for example, English speakers acquiring Japanese syllable structure; Hayes, 2000). The current study on tone acquisition presents a complex situation where learners must both restrict and expand their range of identical tone sequences. Details are provided in the following section.

III The acquisition of Chinese tone sequences

1 Mandarin Chinese tone grammar

While pitch is significant in all languages, it plays different roles for different languages. In tone languages, it is mainly used to specify lexical meanings of words. In intonation languages, it functions mainly at post-lexical levels to add functional meaning that is independent of word meanings (Gussenhoven, 2004). Mandarin Chinese is a typical tone language: a change of tone can cause a change in word meaning, making tone acquisition crucial for L2 learners. Mandarin Chinese morphemes are almost exclusively monosyllabic, and the tone taken by each syllabic morpheme is entirely arbitrary. This makes the acquisition of tones especially difficult.

As a Type 3 tonal language (see (7)), Mandarin Chinese has both lexical level tones and contour tones on full syllables: Tone 1 (T1, high level), Tone 2 (T2, mid rising), Tone 3 (T3, low level or low dipping), and Tone 4 (T4, high falling). In addition, there is a neutral tone with a short duration, which usually falls on unstressed syllables (such as function words). Following Duanmu (2000) and Yip (2002), this study assumes that the default form of Tone 3 is a low level tone.¹ For the sake of transparency, the four tones are abbreviated as H (T1), LH (T2), L (T3), and HL (T4) in the following discussions. Assuming a theory of autosegmental phonology, the TBUs in Mandarin Chinese will be taken to be syllables, and tones will be represented as single melodic units, following Yip (1989) and Bao (1999). A simplified version of the representations of each of the four Mandarin Chinese lexical tones borne by syllables is shown in (8):

(8) a. T1 b. T2 c. T3 d. T4

σ σ σ σ

| /\ | /\

H L H L H L

It is worth noting that contour tones (i.e. (8b) and (8d)) each behave as a single unit (as a whole tone) since the syllable node dominates the component pitch targets, H and L. This differs from the pitch contour in intonation languages. Following Pike (1948), Liberman (1975), Goldsmith (1976), and Pierrehumbert (1980), the phonetic pitch contours of English intonation should be treated as a series of level pitches, such as high and low. We assume the same for the other two non-tonal languages involved in this study, Japanese and Korean.² Therefore, it is assumed that non-tonal language speakers are familiar with linguistic structure such as (8a) and (8c). However, lexical contour tones such as (8b) and (8d) are new for these learners.

Disyllabic words are believed to be the most frequent type of word in Mandarin Chinese. All possible combinations of the four basic lexical tones occur in Mandarin Chinese words, with the exception of T3-T3 (L-L), due to the well-known Tone 3 sandhi process. When a T3 (L) is followed by another T3 (L), it becomes a rising tone, which is identical to T2 (LH) in phonetic shape. This rising tone resulting from the T3 sandhi process is referred to as ‘sandhiT3’ in this study. An example of the Tone 3 sandhi process is demonstrated in (9).

(9) Tone 3 sandhi

a. wŭ

‘five’

L citation form (T3)

b. wŭ.diăn

‘five o’clock’

L. L citation form (T3-T3)

LH. L surface form (sandhiT3-T3)

The sandhi process is acknowledged to be motivated by dissimilation (Yip, 2002).³ T3 sandhi is usually formulated as a rule (T3→T2/___T3) in traditional generative phonology. This process is arguably better described using the OCP in OT, since an OT account specifies the motivation for T3 sandhi and the range of sandhi processes, as shown below. Although OCP constraints operating over constituent tones are sometimes invoked in her analyses, Yip (2002) suggests that the dissimilatory cases in Mandarin Chinese ‘identify complete contours as units’. That is, the dominating OCP constraints operating in Mandarin Chinese are those that operate over whole tones rather than a tone’s component or constituent parts (for example, HL, rather than just H or just L). Item (10) presents the OCP constraints used in Yip (2002).

(10) OCP constraints used in Yip (2002):

OCP(WholeTone): Identical whole tones cannot occur on adjacent syllables.

OCP(ConstTone): Adjacent identical constituent tones are prohibited.

OCP(L): Identical low tones cannot occur on adjacent syllables (L-L is not allowed).

To account for the L2 tonal grammar discussed later, we propose to further break down the family of OCP (WholeTone) constraints into a set of OCP sub-constraints, shown in items (11) and (12):

(11) OCP(Level), or OCP(L) and OCP(H): Identical whole level tones (Low or High) cannot occur on adjacent syllables. Specifically, H-H and L-L are not allowed.

(12) OCP(Contour), including OCP(LH): Identical whole contour tones (such as Rising) cannot occur on adjacent syllables. Specifically, LH-LH and HL-HL are not allowed.

Mandarin Chinese prohibits low tone sequences (L-L) while allowing other tone pairs (LH-LH, HL-HL and H-H). In other words, the well-known Tone 3 sandhi found in Mandarin Chinese is the result of a high-ranked OCP(L) constraint, which restricts any occurrences of L-L. Ranking OCP(WholeTone) below Faith allows for all other identical tone sequences in the language. For the sake of simplicity, the analysis in items (13) and (14) restricts itself to candidates that do not change the second syllable of the word.⁴

(13) Tableau showing the prohibition of L-L surface tones (T3 sandhi)

/L-L/	OCP(L)	Faith	*Rise	OCP(WholeTone)
☞LH-L		*	*
L-L	*!			*

(14) Tableaux for H-H, LH-LH, and HL-HL

/H-H/	OCP(L)	Faith	*Rise	OCP(WholeTone)
☞H-H				*
L-H		*!

/LH-LH/	OCP(L)	Faith	*Rise	OCP(WholeTone)
☞LH-LH			**	*
L-LH		*!	*

/HL-HL/	OCP(L)	Faith	*Rise	OCP(WholeTone)
☞HL-HL				*
L-HL		*!

Tableau (13) shows how the input of a T3-T3 sequence, taken to be L-L underlyingly, results in a surface sequence of LH-L. The tableaux in (14) show how the input tone sequences H-H, LH-LH and HL-HL each produce output sequences that remain faithful to the input, since the high-ranked OCP (L) does not affect the surface form of these combinations. We therefore represent this part of Chinese tone grammar with the constraint ranking: OCP (L) >> Faith >> *Rise, OCP(WholeTone) (for a detailed account, see Yip, 2002: 210).

2 The L1 background

In this section we posit a simplified constraint ranking of L1 grammars in order to identify the tasks that the learners have to complete when acquiring Mandarin Chinese tones. As non-tonal languages, English, Japanese, and Korean share some basic characteristics, such as their representation of intonation contours. The tone sequences H-H and L-L are not uncommon in English, Japanese, or Korean intonation. For example, one of the typical pitch contours in the Korean Accentual Phrase, the most important prosodic unit in Korean, is H-H-L-H (for further information, see Jun, 1996, 2005; for English intonation, see Pierrehumbert, 1980; for Japanese prosody, see Venditti, 2005; Kubozono, 2011). Although all three of these L1s allow two adjacent TBUs bearing two consecutive H pitches or two consecutive L pitches, it is rare to find two identical contour tones within a single word in any of these L1s, such as the Mandarin Chinese LH-LH or HL-HL sequences.

Since learners’ L1s do not have lexical tones, this article will be basing the L1 constraint rankings on allowed phrase-level pitch patterns. Phrasal tone constraints may include Coincide constraints, which ensure that every phrase has phrasal tones (Zoll, 2004), as well as Alignment constraints, which require tones at specific positions (McCarthy and Prince, 1993b). The surface form of intonation is determined by markedness constraints on phrasal tones. Therefore, phrasal tone constraints must dominate Faith in intonational languages. In this study, we assume high-ranked phrasal tone constraints (such as Coincide and Alignment), but do not include them within most of the diagrams, as this study focuses only on the rankings of the constraints related to lexical tone acquisition. In English, Japanese, and Korean, markedness constraints such as *Contour and OCP(Contour) also dominate Faith, since two consecutive identical contour tones (such as HL-HL or LH-LH) within a word are not permitted. However, other markedness constraints, such as OCP(H) and OCP(L) are dominated by Faith. Since English, Japanese, and Korean have similar representations of intonation, we will be representing their L1 grammars with a single (partial) constraint ranking, seen in (15).

(15) Partial ranking representing English, Japanese, and Korean intonation grammars:

PhraseTone >> OCP(Contour), *Contour >> Faith >> OCP(H), OCP(L), *Level

Comparing L1 and L2, the learners must perform three tasks in order to correctly acquire identical tone sequences in Mandarin Chinese: (1) they must expand their set of allowed identical tone sequences to include LH-LH and HL-HL sequences; (2) they must restrict their range of possible identical tone sequences to exclude L-L, since L-L exists in all of their L1s but is not allowed in Mandarin Chinese; and (3) they must retain the H-H pitch sequence that already exists in their L1s. To reword this in terms of OT, these learners must (1) demote both the *Contour and OCP(Contour) below Faith, (2) promote the OCP(L) constraint above Faith, and (3) maintain the OCP(H) at its original low L1 ranking.

3 Hypotheses

As discussed above, the OCP effect is rare (low ranked) and only occasionally appears in the form of the OCP(L) in T3 sandhi situations in Mandarin Chinese. The general OCP(WholeTone) is masked by the faithfulness constraints in other identical tone combinations (H-H, HL-HL, LH-LH) and therefore is ‘inactive’ in the target language grammar. On the other hand, the first languages of the learners do not have lexical tones at all. The Mandarin Chinese individual lexical tones, sequences of the tones and the sandhi processes all pose significant difficulties to these learners. This study hypothesizes that, the L2 tone production is restricted by independent phonological constraints, such as the OCP. That is, during the course of L2 Chinese acquisition, OCP(WholeTone) effects may actively operate in interlanguage tones, although the sub-constraints of OCP(WholeTone) may not be re-ranked by L2 learners as a whole but rather moved separately. Therefore, even though the tone pairs H-H, LH-LH and HL-HL are equally allowed in Mandarin Chinese, learners may prefer not to use these identical tone sequences in their tonal production. The disfavor of identical tone sequences may be manifested both in the types of errors made as well as in the types of substitutions used. In other words, the learners may produce identical tone sequences such as HL-HL and LH-LH with more errors than non-identical tone sequences, and use non-identical tone sequences as substitutions when these errors do occur. This study seeks to determine whether (and, if so, how) the OCP is relevant in L2 tones by examining accuracy rates and errant production, thereby sketching out a path for the development of L2 tone phonology regarding the acquisition of identical tone sequences.

IV Methodology

To test the hypotheses outlined above, an experiment was constructed consisting of a pre-test followed by a main experiment. The pre-test, a reading task of monosyllabic morphemes, was designed to ensure that all participants were able to produce individual lexical tones correctly; 48 monosyllabic morphemes were used for both the pre-test and the main experiment.

1 Participants

Sixty-seven learners participated in the pre-test; 7 participants were excluded from participation due to low accuracy rates in the pre-test (below 85%), leaving a total of 60 participants to participate in the main experiment. Twenty participants were native English speakers (12 males and 8 females), 20 were native Japanese speakers (10 males and 10 females) from areas with Tokyo-type pitch accent, and 20 were native Korean speakers from Seoul (8 males and 12 females). All participants had been learning Mandarin Chinese for at least 6 months, but no more than 18 months at the time of data collection, placing them at approximately an intermediate level. The learners were recruited from the University of North Carolina at Chapel Hill, USA, and the International College at Zhejiang University, China. All participants claimed that Mandarin Chinese was the only tonal language they had studied. Learners participated in this study voluntarily, and each was paid for the recording session.

2 Procedure

Stimuli for the main experiment were disyllabic words bearing all 16 possible combinations of the four lexical tones. Each of the 16 combinations was equally represented using two words (consisting of different morphemes), resulting in 32 distinct words. Participants were asked to repeat the 32 words twice, resulting in 64 tokens collected per speaker per trial. The use of sonorants in test words was maximized in order to facilitate continuous pitch tracking. All syllables were CV(V) in structure, with no codas. Test words were at the lowest proficiency level according to the Handbook of the graded vocabulary for HSK (Hanyu Shuiping Kaoshi ‘Mandarin Chinese proficiency examination’) (1992). In order to ensure natural production, the 64 test words were embedded in the middle of sentences. To avoid anticipatory and carry-over effects by neighboring tones (Xu, 1997), the test tokens (labeled as XX in (16)) were both preceded and followed by de, a morpheme with a neutral tone, as shown in (16).

(16) Test sentence for the main experiment:

Wŏ juéde X X de dōngxi hĕn hăo

I think X X particle things very good

‘I think XX things are very good.’

Participants were given a list of sentences and were asked to produce each in Mandarin Chinese at a normal speed.

The sentences in the main experiment were randomly ordered in the reading list. Participants were provided with Pinyin transcriptions and Chinese characters, as well as the English, Japanese, or Korean translations of each sentence in the reading lists.

Participants were recorded with Version 5.2.17 of Praat (Boersma and Weenink, 2011) in a soundproof recording booth. The recording was paused between pages of the reading list.

This study refers to both Identical Tone Combinations (ITC) (i.e. T1-T1, T2-T2, and T4-T4) and Non-Identical Tone Combinations (NITC) (i.e. T1-T2, T2-T3, etc.). Correct surface production of underlying T3-T3 (L-L) sequences were labeled as ‘sandhiT3-T3’ (LH-L).

The goal of this experiment was:

to test for the evidence of the OCP by comparing the results of ITCs with those of NITCs;

to determine the movement of OCP(L) by observing surface production of sandhiT3-T3; and

to determine the ranking of OCP(H) and OCP(Contour) by comparing the results of various ITCs (specifically H-H, LH-LH, and HL-HL).

If the OCP is active in interlanguage grammars, we would expect to find significantly more errors and lower substitution rates of ITCs compared to NITCs.

3 Judgments and data analysis

Each learner produced 128 test syllables (64 test words with two syllables in each word) in the main experiment. The author, a native speaker of Mandarin Chinese, listened to participants’ productions and judge whether or not the tonal production was target-like, and marked each production as ‘correct’ or ‘incorrect’.

In order to ensure the reliability of correctness judgments and transcriptions of incorrect tones, both intra- and inter-rater agreements were calculated. All tonal production was judged and transcribed twice by the author, with a one month interval in between judgments. The agreement rate between these two judgments was 95.6%. For the inter-transcriber reliability, two other native speakers (referred to here as L and C) were hired to judge and transcribe 10% of the data independently. Both L and C were native speakers of Mandarin Chinese and had received training in linguistics. A pairwise comparison test indicates that the author and L had the highest level of agreement (93.8%), the author and C the second highest level of agreement (92.6%), while C and L had the lowest level of agreement (91.6%).

A tone was considered to be an error if either the contour (rising, falling, level) or register (pitch range, either high or low) of the tone was incorrect. For each incorrect production, the participant’s actual tonal production was also transcribed. Most of the errors fell within the Mandarin Chinese tone inventory. Due to the low rates of out-of-inventory production such as a mid-level tone (1.37%, 1.79% and 1.87% in English, Japanese and Korean speakers’ subsets respectively), the following discussion will only focus on the within-inventory mis-production.

Statistical analyses were used to calculate error and substitution rates. Hypotheses regarding universal constraints were tested using the SAS statistical package. The significance criterion adopted for declaring a significant difference is p < 0.05.

V Results

1 The OCP tests

Two OCP tests were conducted to detect possible effects of the OCP. As described above, the stimuli contain two types of tone combinations: Identical Tone Combinations (ITCs), and Non-Identical Tone Combinations (NITCs). Simply comparing the error rates of ITCs with that of NITCs may not be sufficient to test for effects of the OCP, since OCP testing requires an analysis of entire tone sequences. For example, when a learner intends to produce a target LH-LH sequence, he or she may change the first tone, the second tone, or both tones, and incorrectly produce an NITC. Alternatively, he or she might change both tones and incorrectly produce another ITC, such as HL-HL being produced for a LH-LH combination. In this case, even a high error rate does not necessarily indicate that the speaker has difficulties with identical tone sequences. A simpler explanation may be that the speaker prefers HL to LH. Similar discrepancies can occur in producing target NITCs, where the speaker may produce erroneous identical tone pairs as well as erroneous non-identical tone pairs. Because of this, this section analyses the types of substitutions made in L2 speakers’ production. Although individual differences were found in L2 tone production, we chose to look at these learners as a group, in order to posit generalizations regarding L2 tone acquisition.

a OCP Test 1: ITC and NITC change rates

The goal of Test 1 was to determine how often target ITCs were produced erroneously as NITCs, and how often target NITCs were produced erroneously as ITCs. This was referred to as their ‘change rate’. To compare change rates, we set X to be the percentage of wrongly-produced target ITC pairs which have been substituted with NITCs, and Y as the percentage of wrongly-produced target NITC pairs that have been substituted with ITCs, as illustrated in Figure 1. If X is significantly greater than Y, this would be consistent with the hypothesis that the OCP has an effect on L2 production.

Figure 1.

Obligatory Contour Principle (OCP) Test 1.

The SurveyFreq Procedure was used to calculate X and Y for each sub-dataset, and the Rao-Scott Chi-Square tests were used to calculate p-values. X values were significantly higher than Y values both in general as well as within each of the three groups of speakers. Tables 1 –3 below give details of Test 1. The fourth column shows the percentages of NITCs produced as ITCs, or ITCs produced as NITCs, out of the total number of production. Results of statistical tests are given in the last column.

Table 1.

Obligatory Contour Principle Test 1 results for English speakers.

Target stimulus type	Actual production	Raw percentages of actual production	Percentages	Hypo test
NITC	→NITC count: 964	92.69	75.31	Pr > Chisq < .0001Pr > F 0.0001
NITC	→ITC count: 76	Y = 7.31	5.94
ITC	→NITC count: 153	X = 63.75	11.95
ITC	→ITC count: 87	36.25	6.80

Table 2.

Obligatory Contour Principle Test 1 results for Japanese speakers.

Target stimulus type	Actual production	Raw percentages of actual production	Percentages	Hypo test
NITC	→NITC count: 972	93.46	75.93	Pr > Chisq < .0001Pr > F < .0001
NITC	→ITC count: 68	Y = 6.54	5.31
ITC	→NITC count: 138	X = 57.5	10.78
ITC	→ITC count: 102	42.5	7.96

Table 3.

Obligatory Contour Principle Test 1 results for Korean speakers.

Target stimulus type	Actual production	Raw percentages of actual production	Percentages	Hypo test
NITC	→NITC count: 932	89.62	72.81	Pr > Chisq < .0001Pr > F < .0001
NITC	→ITC count: 108	Y = 10.38	8.43
ITC	→NITC count: 137	X = 57.08	10.7
ITC	→ITC count :103	42.92	8.04

The change rate for ITCs was significantly greater than the change rate of NITCs, showing that all participants were better at producing a sequence of non-identical tones than a sequence of identical tones.

b OCP Test 2: The OCP and tonal markedness

The goal of Test 2 was to determine whether the proportions of Identical Tone Combinations (ITCs) in L2 production match those of the target stimuli in each sub-dataset. There were 16 possible tone combinations of four Chinese tones, all of which were evenly proportioned in the stimuli. The ITC sequences made up 3/16ths of the total number of stimuli (H-H, LH-LH, and HL-HL; note that L-L should be produced as sandhiT3-T3 which is LH-L). Therefore, if the OCP has no effect on L2 tones, we would expect the ITC production to also make up about 3/16ths (18.75%) of the total production.

The ITC rates were compared with their expected values using a binomial test. All ITC sequences together made up 12.72% of English speakers’ L2 tonal production, which was significantly lower than 18.75% (p < 0.0001); 13.28% of the Japanese speakers’ production, which was significantly lower than 18.75% (p < 0.0001); and 16.48% of the Korean speakers’ production (p = 0.0008). Korean speakers exhibited the highest rate of ITCs, but all three groups of speakers produced less than 18.75% ITCs, indicating that learners preferred not to use identical tone combinations in general.

However, when the ITC sequences were broken down and analysed by specific tone type, individual ITC sequence proportions varied widely. All three groups of speakers produced more H-H sequences than HL-HL, and more HL-HL than LH-LH sequences (see Figure 2).

Figure 2.

The occurrences of tone pairs in L2 production (OCP Test 2).

If the OCP had no effect on speakers’ production, we would expect each of these individual tone pair rates to be 1/16th, or 6.25%, of all total production. A binominal test indicated that H-H production across all three groups of speakers were significantly greater than the expected 6.25%, whereas the LH-LH and HL-HL rates were both significantly lower than the expected 6.25%. It was also found that participants produced many more HL-HL sequences than LH-LH sequences. Additionally, accuracy rates of tone pairs (as target tone sequences) showed the same tendency: H-H sequences were produced correctly significantly more often than HL-HL sequences, and HL-HL sequences were produced correctly significantly more often than LH-LH sequences. This finding held true across all three groups of speakers.

The higher occurrence of HL-HL over LH-LH was very likely motivated by the Tonal Markedness Scale. To confirm the effects of the TMS (*Rise >> *Fall >> *Level), the error rates of individual H, LH and HL production were calculated (see Figure 3). The SurveyFreq Procedure was used to calculate error percentages. The three groups demonstrated similar error patterns. A statistical test using the Rao-Scott Chi-Square test was conducted to account for multiple observations within participants. The final results are presented in Table 4. All three groups made more errors with HL than H. This difference was significant for the English and Japanese speakers and a trend for the Korean speakers. The error rates of LH were significantly higher than those of HL across all three groups of speakers.

Figure 3.

Error rates of individual tones in the main experiment.

Table 4.

Rao-Scott Chi-Square test results for Tonal Markedness Scale (TMS) hypothesis (Values of Pr > Chisq).

	Overall	English	Korean	Japanese
HL vs. H	< 0.0001	0.0037	0.2780	0.0004
LH vs. HL	< 0.0001	0.0044	< 0.0001	< 0.0001

The significantly better performance of H over HL and HL over LH was confirmed by analysing substitution patterns used by all three groups of learners. When learners made tone errors, they used H as the substitute tone for the target tone more often than HL, and they used HL as the substitute tone more frequently than LH.

2 The acquisition of Tone 3 sandhi

This section briefly reports participants’ acquisition of L-L sequences. In native Mandarin Chinese, the dissimilation of L-L is motivated by OCP(L). The underlying L-L sequence surfaces in Chinese as sandhiT3-T3 (LH-L). By comparing the accuracy rate of non-native tone production of sandhiT3-T3 sequences with that of other ITCs, we attempt to clarify how the re-ranking of OCP(L) progresses in comparison with other OCP constraints. Figure 4 shows the accuracy rates of the ITCs (namely, H-H, LH-LH, HL-HL) and that of the sandhiT3-T3 sequence ( LH-L).

Figure 4.

Accuracy rates of identical tone sequences.

The accuracy rate of the sandhiT3-T3 sequence was very high compared to that of other ITCs, suggesting that Tone 3 sandhi is acquired quickly.⁵ Most of the accuracy rates of sandhiT3-T3 sequences were lower than that of H-H sequences (with the exception of Japanese) but higher than that of HL-HL and LH-LH sequences.⁶

VI Discussion

1 OCP(WholeTone) or OCP(ConstTone)

The two OCP tests reported in the article present evidence supporting the operation of the OCP in L2 production of tones. To ensure that the effects we are seeing come from the OCP for whole contour tones (as in Yip, 2002) rather than the OCP for constituent or component tones motivated by OCP(ConstTone), we look closer at the error rates of relevant tone sequences. If the effects we are seeing come from OCP(ConstTone), the error rates of LH-LH and LH-LH would be very low. In addition, we would also expect LH-LH and HL-HL to be used as substitutions for other tone sequences often, since they consist of alternating H and L component tones. Meanwhile, the error rates of those sequences containing two identical component tones across syllable boundaries should be very high, specifically those of H-HL, LH-H, LH-HL, L-LH, HL-LH and HL-L. Out of all 16 possible tone combinations, only H-H and L-L can be counted as both identical whole tone sequences and as component tone sequences, and so were excluded from the analysis.

This study found support for an effect of OCP(WholeTone) over an effect of OCP(ConstTone). The error rates of LH-LH were the highest among the 16 tonal combinations across all three groups of participants. The error rates of HL-HL were lower than that of LH-LH, but still much higher than those of H-HL, LH-HL, and HL-L.⁷ This held true across all three groups of speakers. In addition, some sequences such as H-HL and LH-HL were often used as substitutions for other tone sequences. However, it was difficult to exclude from consideration the possibility of some effect of OCP(ConstTone), because the most often used substitutions also included LH-L and L-HL, both of which consisted of alternating H and L tones. To summarize, although L2 learners showed some effects of an OCP(ConstTone) constraint, they were largely overshadowed by effects of an OCP(WholeTone) constraint.

2 Stages in L2 tone phonology development

The results above show some evidence of an OCP (WholeTone) constraint in effect in learners’ interlanguage grammars, but this evidence seems dependent on the type of tone being analysed. The high accuracy rate of H-H and wide use of H-H as a substitute for other tones is in essence an ‘anti-OCP’ effect. This may be attributed to L1 transfer since H-H is a typical L1 tone sequence in Korean and in Japanese and is allowed in English. However, we believe that the preference for H-H is not solely due to L1 transfer because (1) compared to contour tones, high level tones are physically easier to produce and are more widely distributed in natural languages, and (2) phonetic studies show that strings of H-H sequences require the least articulatory effort (Shih and Lu, 2010). These reasons make it very likely that there is at least some type of universal preference for H-H sequences at play.

The asymmetry found between LH-LH and HL-HL likely cannot be attributed to L2 knowledge, since both LH-LH and HL-HL are equally grammatical in Mandarin Chinese, and cannot be attributed to L1 transfer, since both of them are unusual in English, Japanese and Korean. Because of this, it is argued that Tonal Markedness Scale is still accessible by adult L2 learners, and is interacting with the OCP. However, it should also be noted that the asymmetry found between LH-LH and HL-HL may have been activated by an asymmetry to exposure in the target language input, as there are more HL-HL words than LH-LH words in Mandarin Chinese and also in the teaching materials (Shang, 2000; Liu et al., 2009). The remainder of this section will account for these patterns and propose a four-stage model describing the development of tonal phonology in the interlanguage using the OCP sub-constraints.

As discussed in (15), the OCP(H) constraint is low-ranked for English, Japanese, and Korean. Because OCP(H) is also dominated by all other constraints in the Mandarin Chinese phonology, OCP(H) does not require re-ranking. It is not surprising that learners have little difficulty producing H-H sequences. Lexical contour tones (i.e. LH-LH and HL-HL tone sequences) are new for these learners, so the general OCP(WholeTone) constraint and OCP(Contour) are ranked high in the initial stage of acquisition. OCP(LH) is masked by the more general OCP(Contour) in L1s, since rising tone is a type of contour tone. OCP(LH) is also masked by OCP(Contour) but ranked low in the Mandarin Chinese phonology, since all contour pairs are allowed in the target language. Item (17) shows the partial rankings of the OCP constraints in the L1 and L2 phonologies.

(17) L1 and L2 phonologies:

L1 (initial state): OCP(Contour), OCP(LH) >> Faith >> OCP(H), OCP(L)

L2 (target state): OCP(L) >> Faith >> OCP(Contour), OCP(LH), OCP(H)

Item (17) suggests that OCP(Contour) must be demoted in order to approximate the target language ranking. HL-HL and LH-LH are both new to L2 learners and therefore difficult for them to acquire. However, results of the phonological experiment suggest that learners acquire HL-HL sequences earlier than LH-LH sequences because HL-HL sequences had a lower error rate than LH-LH sequences and HL-HL was used as a substitute for other tones more often than LH-LH. Following Boersma and Levelt (2000) and Boersma and Hayes (2001), who argue for a model of acquisition in which the rate of constraint re-ranking is dependent on the frequency with which a constraint is violated in the input data, we assume the markedness constraints that are more frequently violated are more quickly demoted than those that are less frequently violated. In this case of tone acquisition, every time OCP(LH) is violated, OCP(Contour) will also be violated, since LH is a type of contour tone, but not the vice versa. Therefore the general OCP(Contour) constraint is demoted before OCP(LH) is demoted. This would result in an interlanguage phonology in which, at some point during L2 acquisition, OCP(Contour) has been demoted, but OCP(LH) has not. This phenomenon of earlier demotion of a general constraint before the demotion of a specific constraint is argued to exist in both L1 and interlanguage grammars (Broselow, 2004). This intermediate stage is summarized in (18).

(18) Intermediate Stage: OCP(LH), OCP(L)>> FAITH>> OCP(Contour), OCP(H)

In this intermediate stage of tone acquisition, H-H and HL-HL sequences are allowed, but LH-LH is not, leading to more HL-HL sequences than LH-LH sequences. In the diagram (19) simulating this specific stage of L2 tone acquisition, the same group of input tone pairs surface with different outputs in the L1, the interlanguage, and the L2. Only H-H remains throughout the acquisition process. In the intermediate stage of the L2 acquisition of Mandarin Chinese tones, H-H and HL-HL are in learners’ outputs, but OCP(LH) remains high-ranked, which leads to the lack of LH-LH sequences in L2 tonal production for a period of time. Example outputs are cited from the most frequent production found in the current study.

(19) The re-ranking of OCP sub-constraints

L1 phonology OCP(Contour), OCP(LH) >> Faith >> OCP(H)
InputH-H/ HL-HL/ LH-LH		OutputH-H/ L-HL/ LH-L
Interlanguage phonologyOCP(LH) >> Faith >> OCP(Contour), OCP(H)
InputH-H/ HL-HL/ LH-LH		OutputH-H/ HL-HL/ LH-L
L2 (Chinese) phonologyFaith >> OCP(LH), OCP(Contour), OCP(H)
InputH-H/ HL-HL/ LH-LH		OutputH-H/ HL-HL/ LH-LH

In this case, the demotion of OCP constraints is affected by the Tonal Markedness Scale. The effects of OCP(LH) are originally masked by the more general OCP(Contour) constraint in both L1 and L2 phonologies, only becoming visible over the course of L2 acquisition. This situation represents what is known as ‘The Emergence of the Unmarked’ (McCarthy and Prince, 1994; Broselow et al., 1998). Very few participants in this study consistently produced all tone pairs, suggesting that most participants still have the rankings found in the developing interlanguage phonology.

As a summary, if we take the partial constraint ranking of the L1 intonation phonology as the initial stage and the native Mandarin Chinese constraint ranking as the final stage to set a frame, the proposed path of OCP constraint re-ranking includes at least two intermediate stages, among others, according to this study’s findings. As shown in Figure 4, the high accuracy rate of sandhiT3 indicates that restricting L-L pitch sequences is not difficult for language learners and the promotion of OCP(L) occurs no later than the demotion of OCP(Contour), if not earlier. This is labeled as Stage_n. Since learners were able to produce all individual tones, including lexical contour tones, with high accuracy, it is believed that *Contour is demoted very early in the re-ranking process, which is also shown in Stage_n. OCP(Contour) is demoted before OCP(LH), which is labeled as Stage_n+1.

(20) Summary of the demotion and promotion of the OCP constraints

Stage_initial: Partial L1 ranking

OCP(Contour), *Contour >> Faith >> OCP(Level), *Level

Stage_n: Early demotion of *Contour, promotion of OCP (L)

OCP(L), OCP(Contour) >> Faith >> OCP(H), *Level, *Contour

Stage_n+1: Demotion of OCP(Contour) before OCP (LH):

OCP(L), OCP(LH) >> Faith >> OCP(Contour), OCP(H), *Level

State_final: Partial L2 ranking

OCP(L) >> Faith >> OCP(Contour), OCP(H), *Contour, *Level

This is an idealized picture. The actual situation of L2 phonological acquisition is of course more complicated, as L2 learners usually exhibit wide variation. When L2 learners re-rank their universal constraints, this may result in an ‘unstable’ phonology with a seemingly inconsistent array of surface representations until the constraint rankings stabilize (see the ‘gradual learning algorithm’ of Boersma and Hayes (2001), which accounts for variation using probabilistic constraint rankings; see Broselow and Xu (2004) and Cardoso (2007, 2011), among others, for an application of this model to variation in second language learning).

3 An alternative account for the interacting effect of the OCP and the TMS

This article explains the interaction of the OCP and the TMS by linking individual tone types to OCP sub-constraints. Alternatively, we could account for this interaction using self-conjunction of tone markedness constraints, or weighted constraints (Pater, 2009). Due to limited space, the self-conjunction account will only be briefly discussed here.

As defined in (1), the OCP constraints, which can be stated as the prohibition of adjacent identical autosegments, do not actually reference specific markedness elements. On the other hand, the TMS (*Rising >> *Falling >> *Level) is concerned only with the complexity of individual tones and makes no claims about tone pairs at all. They are two independent constraints, but they jointly shape the learners’ interlanguage tonal phonology since, for example, we found two violations of LH to be more serious than two violations of HL in this study. In Alderete (1997), OCP effects are understood as the result of markedness constraints which are strengthened by the operation of ‘local conjunction’ of a constraint with itself, or ‘self-conjunction’ as discussed in Smolensky (1993). Within the context of L2 acquisition of Mandarin Chinese tone pairs, it is proposed here that co-occurrence restrictions can be explained by extending the original TMS ranking to the locally-conjoined tonal markedness constraints, as shown in (21).

(21) (*Rising) L ² >> (*Falling) L ² >> (*Level) L ²

By locally conjoining the TMS, the errors and substitutions participants made in this study can be explained with a single general principle: a sequence of two rising tones is the most difficult to produce (most marked); a sequence of two falling tones is of medium difficulty; and a sequence of two level tones is the least difficult. In this way, the promotion of the constraint OCP(L), which is proposed in Yip (2002), and the self-conjunction of the TMS ‘(*Rising) L ² >> (*Falling) L ² >> (*Level) L ²’ proposed in this study are sufficient to account for the dissimilation phenomena discussed in this article.

VII Conclusions

This study tested the hypotheses regarding the role of the OCP in shaping L2 tonal phonologies and presented a constraint-based analysis of the L2 acquisition of Mandarin Chinese tone pairs. An experiment conducted with English, Japanese, and Korean native speakers learning Mandarin Chinese as an L2 showed that L2 learners ‘disfavor’ identical tone sequences, with the exception of H-H tone sequence, which was produced with a high level of accuracy throughout the acquisition process. When learning contour pairs unfamiliar to their native language phonology but allowed in the target language, learners were better at producing HL-HL sequences than LH-LH sequences and prefer HL-HL to LH-LH in substitutions. This study suggests that OCP(WholeTone) interacts with the TMS (*Rise >> *Fall >> *Level) and that the OCP sub-constraints operate separately from one another. OCP(H) is kept at a low rank in the L1, the interlanguage, and the L2; OCP(L) is promoted to a ranking higher than Faith early on in the acquisition process, and OCP(Contour) is demoted to a ranking lower than Faith later on. The asymmetry of HL-HL and LH-LH in L2 tones motivates a proposal of demotion of the general OCP(Contour) constraint earlier than the specific OCP(LH) constraint. This OT account incorporates both L1 transfer (in the case of H-H) and phonological universals (the case of HL-HL and LH-LH) into a single grammar. The interacting effects of the OCP and the TMS found in L2 tonal phonologies can also be understood as a case of local conjunction of the TMS: (*Rising) L ² >> (*Falling) L ² >> (*Level) L ².

The model of constraint ranking assumed in this article provides us with the tools to explicitly model the unconscious knowledge underlying L2 tonal production. As a preliminary study of L2 tones within the framework of OT, this research presents an analysis of OCP effects in L2 tones, adding to the range of Optimality Theoretic studies on L2 prosody acquisition. More research within and beyond an OT framework is required to fully understand the properties of L2 tones. For example, a study on the relationship between tone perception and production, a more thorough exploration of the OCP(ConstTone) constraint effect in L2 tones, and a weighted constraint account of the interaction between the OCP and the TMS would help form a more complete picture of the properties of L2 tone pairs. It is hoped that this research will inspire more studies along this line.

Footnotes

Acknowledgements

I am grateful for the advice given by Dr. Jennifer Smith and the three anonymous reviewers of Second Language Research, whose insightful comments and suggestions have greatly benefited this article. A part of this article has been presented in the 2013 New Sounds International Symposium on the Acquisition of Second Language Speech. Many thanks to Chris Wiesen, Jie Cai, Jia Lin, Emily Moeng, and Alexandra Tran for their invaluable support, and to my colleagues and the participants at the University of North Carolina at Chapel Hill, Zhejiang University and George Washington University for all of their help.

Declaration of conflicting interest

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The University of North Carolina at Chapel Hill has provided university-internal doctoral grant to support this research.

Notes

References

Alderete

(1997) Dissimilation as local conjunction. Proceedings of the North East Linguistics Society 27: 17–32. Amherst, MA.

Alderete

(2003) Phonological processes: Dissimilation. In: Frawley

(ed.) International encyclopedia of linguistics. 2nd edition. Oxford: Oxford University Press, pp. 323–24.

Alderete

Frisch

(2007) Dissimilation in grammar and the lexicon. In: de Lacy

(ed.) The Cambridge handbook of phonology. Cambridge: Cambridge University Press, pp. 379–98.

Alderete

Tupper

Frisch

(2013) Phonological constraint induction in a connectionist network: learning OCP-Place constraints from data. Supplemental materials: Data supplement, software supplement. Language Sciences 37: 52–69.

Archibald

(1998) Second language phonology, phonetics, and typology. Studies in Second Language Acquisition 20: 189–211.

Bao

(1999) Tonal contour and register harmony in ChaoZhou. Linguistic Inquiry 30: 485–93.

Barlow

Gierut

(1999) Optimality Theory in phonological acquisition. Journal of Speech, Language, and Hearing Research 42: 1482–98.

Boersma

Hayes

(2001) Empirical tests of the gradual learning algorithm. Linguistic Inquiry 32: 45–86.

Boersma

Levelt

(2000) Gradual constraint-ranking learning algorithm predicts acquisition order. Proceedings of Child Language Research Forum 30: 229–37. Stanford, CA.

10.

Boersma

Weenink

(2011) Praat: doing phonetics by computer [computer program]. Version 5.2.17. Available at: http://www.praat.org (accessed March 2016).

11.

Broselow

(2004) Unmarked structures and emergent rankings in second language phonology. International Journal of Bilingualism 8: 51–65.

12.

Broselow

(2004) Differential difficulty in the acquisition of second language phonology. International Journal of English Studies 4: 135–63.

13.

Broselow

Chen

Wang

(1998) The emergence of the unmarked in second language phonology. Studies in Second Language Acquisition 20: 261–80.

14.

Cardoso

(2007) The variable development of English word-final stops by Brazilian Portuguese speakers: A stochastic Optimality Theory account. Language Variation and Change 19: 219–48.

15.

Cardoso

(2011) Onset-nucleus sharing and the acquisition of second language codas: A stochastic Optimality Theoretic account. Studia Linguistica 65: 198–231.

16.

Cheng

(1973) A quantitative study of tone in Chinese. Journal of Chinese Linguistics 1: 93–110.

17.

Duanmu

(2000) The phonology of standard Chinese. Oxford: Oxford University Press.

18.

Eckman

(2004) From phonemic differences to constraint rankings: Research on second language phonology. Studies in Second Language Acquisition 26: 513–49.

19.

Epstein

Flynn

Martohardjono

(1996) Second language acquisition: Theoretical and experimental issues in contemporary research. Behavioral and Brain Sciences 19: 677–758.

20.

Goldsmith

(1976) An overview of autosegmental phonology Linguistic Analysis 2: 23–68.

21.

Gordon

(2001) A typology of contour tone restrictions. Studies in Language 25: 405–44.

22.

Gussenhoven

(2004) The phonology of tone and intonation. Research surveys in linguistics. Cambridge: Cambridge University Press.

23.

Hancin-Bhatt

(1998) Optimality and stages of L2 syllables. Unpublished paper presented at the 1998 Second Language Research Forum, Honolulu, HI, USA.

24.

Hancin-Bhatt

(2000) Optimality in second language phonology: Codas in Thai ESL. Second Language Research 16: 201–32.

25.

Hancin-Bhatt

(2008) Second language phonology in Optimality Theory. In: Edwards

Zampini

(eds) Phonology and second language acquisition. Philadelphia: John Benjamins.

26.

Hancin-Bhatt

Bhatt

(1997) Optimal L2 syllables: Interaction of transfer and developmental effects. Studies in Second Language Acquisition 19: 331–78.

27.

HSK Department, Chinese government (1992) Handbook of the graded vocabulary for HSK. Beijing: Beijing Language Institute Press.

28.

Hayes

(2000) Reranking stages in OT analysis of the acquisition of Japanese as a second language. Carolina Working Papers in Linguistics 1: 1–1.

29.

(2008) Zuowei waiyu de hanyu jiaoxue [Teaching Chinese as a foreign langauge]. Foreign Language Education in China 1(2): 3–10.

30.

Hyman

VanBik

(2004) Directional rule application and output problems in Hakha Lai Tone. Language and Linguistics 5: 821–61.

31.

Jun

(1996) The phonetics and phonology of Korean prosody. New York / London: Garland.

32.

Jun

(2005) Prosodic typology. Oxford: Oxford University Press.

33.

Kubozono

(2011) Japanese pitch accent. In: van Oostendorp

Ewen

Hume

Rice

(eds) The Blackwell Companion to Phonology. Oxford: Blackwell, pp. 2879–2907.

34.

Lado

(1968) Linguistics across cultures: Applied linguistics for language teachers. Ann Arbor, MI: The University of Michigan Press.

35.

Leben

(1973) Suprasegmental phonology. Unpublished PhD dissertation, MIT, Cambridge, MA, USA.

36.

Thompson

(1977) The acquisition of tone in Mandarin-speaking children. Journal of Child Language 4: 185–99.

37.

Liberman

(1975) The intonational system of English. Unpublished PhD dissertation, MIT, Cambridge, MA, USA.

38.

Liu

. (2009) Integrated Chinese. 3rd edition. Boston, MA, USA: Cheng & Tsui Company.

39.

Lombardi

(2003) Second language data and constraints on manner: Explaining substitutions for the English interdentals. Second Language Research 19: 225–50.

40.

Major

(2001) Foreign accent: The ontogeny and phylogeny of second language phonology. Mahwah, NJ: Lawrence Erlbaum.

41.

McCarthy

(1986) OCP Effects: Gemination and antigemination. Linguistic Inquiry 17: 207–64.

42.

McCarthy

Prince

(1993a) Prosodic morphology I: Constraint interaction and satisfaction. Technical Report. Rutgers University Center for Cognitive Science, New Brunswick, NJ. [Available on Rutgers Optimality Archive, ROA-482].

43.

McCarthy

Prince

(1993b) Generalized alignment. In: Booij

van Marle

(eds) Yearbook of Morphology. Dordrecht: Kluwer, pp. 79–153. [Available on Rutgers Optimality Archive, ROA-7].

44.

McCarthy

Prince

(1994) The emergence of the unmarked: Optimality in prosodic morphology. Papers from the Annual Meeting of the North East Linguistic Society 24: 333–79.

45.

McCarthy

Prince

(1995) Faithfulness and reduplicative identity. In: Beckman

Dickey

Urbanczyk

(eds.) University of Massachusetts Occasional Papers in Linguistics 18. Amherst, MA: GLSA Publications, pp. 249–384. [Available on Rutgers Optimality Archive, ROA-103].

46.

Mei

(1977) Tones and tone sandhi in 16th century Mandarin. Journal of Chinese Linguistics 5: 237–60.

47.

Ohala

(1978) Production of tone. In: Fromkin

(ed.) Tone: A linguistic survey. New York: Academic Press, pp. 3–39.

48.

Pater

(2009) Weighted constraints in generative linguistics. Cognitive Science 33: 999–1035.

49.

Pierrehumbert

(1980) The phonology and phonetics of English intonation. Unpublished PhD dissertation, MIT, Cambridge, MA, USA.

50.

Pike

(1948) Tone languages. Ann Arbor, MI: University of Michigan Press.

51.

Prince

Smolensky

(1993) Optimality Theory: Constraint interaction in generative grammar. Malden, MA: Rutgers University Center for Cognitive Science. [Available on Rutgers Optimality Archive, ROA-537].

52.

Shang

(2000) Dui Hanyu Shuiping Cihui Dengji Dagang Jia Yi Ji Ci Shengdiao Guilv de Diaocha [A survey of the tone distribution]. Chinese Teaching in the World 2: 43–47.

53.

Shepherd

(2003) Constraint interactions in Spanish phonotactics: An optimality theory analysis of syllable-level phenomena in the Spanish language. Unpublished MA thesis, California State University, Northridge, CA, USA.

54.

Shih

(2010) Prosody transfer and suppression: Stages of tone acquisition. Speech Prosody 2010 Conference Proceedings. Chicago. Available at: http://speechprosody2010.illinois.edu/papers/100968.pdf (accessed March 2016).

55.

Smolensky

(1993) Harmony, markedness, and phonological activity. Unpublished handout to talk presented at the Rutgers Optimality Workshop 1, New Brunswick, NJ, USA.

56.

Sun

(1998) The development of a lexical tone phonology in American adult learners of standard Mandarin Chinese. Honolulu, HI: University of Hawai’i Press.

57.

Suzuki

(1998) A typological investigation of dissimilation. Unpublished doctoral dissertation, University of Arizona, Tucson, AZ, USA.

58.

Tesar

Smolensky

(1998) Learnability in Optimality Theory. Linguistic Inquiry 29: 229–68.

59.

Tesar

Smolensky

(2000) Learnability in Optimality Theory. Cambridge, MA: MIT Press.

60.

Venditti

(2005) The J_ToBI Model of Japanese intonation. In: Jun

(eds) Prosodic typology. Oxford: Oxford University Press, pp. 172–200.

61.

(1997) Contextual tonal variations in Mandarin. Journal of Phonetics 25: 61–83.

62.

Yip

(1988) The Obligatory Contour Principle and phonological rules: A loss of identity. Linguistic Inquiry 19: 65–100.

63.

Yip

(1989) Contour tones. Phonology 6, 149–74.

64.

Yip

(2002) Tone. Cambridge: Cambridge University Press.

65.

Young-Scholten

(1994) On positive evidence and ultimate attainment in L2 phonology. Second Language Research 10: 193–214.

66.

Zhang

(2010) Phonological universals and tone acquisition. Journal of Chinese Language Teachers Association 45: 39–65.

67.

Zhang

(2013) The second language acquisition of Mandarin Chinese tones by English, Japanese and Korean Speakers. Unpublished PhD dissertation, University of North Carolina at Chapel Hill, NC, USA.

68.

Zhang

(2014) The third tone: Allophones, sandhi rules and pedagogy. Journal of Chinese Language Teachers Association 49: 117–45.

69.

Zhang

(2015) Positional effects in second language Chinese tones. Journal of Chinese Language Teaching 12: 1–30.

70.

Zhang

(2002) The effects of duration and sonority on contour tone distribution: A typological survey and formal analysis. New York: Routledge.

71.

Zhang

(2004) The role of contrast-specific and language-specific phonetics in contour tone distribution. In: Hayes

Kirchner

Steriade

(eds) Phonetically based phonology. Cambridge: Cambridge University Press, pp. 157–90.

72.

Zhang

(2009) Contour tone distribution is not an artifact of tonal melody mapping. Studies in the Linguistic Sciences 33: 73–132.

73.

Zhu

Dodd

(2000) The phonological acquisition of Putonghua (Modern Standard Chinese). Journal of Child Language 27: 3–42.

74.

Zoll

(2004) Positional asymmetries and licensing. In: McCarthy

(ed.) Optimality Theory in Phonology: A Reader. Oxford: Blackwell, pp. 365–378.