1 Overview of the volume

I first briefly describe the content of each chapter. Chapter 1 covers basic topics such as the nature and contrastive function of tone, mechanisms of pitch change and the tone-notation system. In Chinese, monosyllabic words can have different tonal melodies, which cannot be explained in terms of stress. Contrastive functions of tones are illustrated from Thai and Yoruba (with a contrast in meaning), Cantonese (with a contrast in grammatical category) and Somali (with a contrast in case). Wee presents cases where acoustic correlates of tones are not restricted to F0, but also involve intensity (as in whispered Chinese) and phonation (e.g. creaky and breathy). He asserts that tones are phonological in the sense of what they can do in languages: contrast, concatenate and alternate.

Chapter 2 explicates the autosegmentality of tones. Wee provides familiar examples where tone-bearing units (TBUs) are syllables or moras, and shows that a foot can also be a TBU, something which is not frequently discussed. In Kera, /LH/ is realised as LLH for words with a (ws)s foot structure, and as LHH for words with a (s)(ws) foot structure. The chapter covers basic topics such as floating tones, tone mobility, tone spreading and tone stability.

Chapter 3 deals with the development of the tone-notation system and the phonological representation of tone. The chapter presents the numerical scale tone-letter system (Chao Reference Chao1930), which uses five-level pitch values (lowest 1, highest 5). This letter system is used to show the tone inventories for the seven major groups of Han languages. From this inventory we can see that level, falling and rising tones are common, whereas dipping and peaking are rare. Downtrend and downdrift are covered, including the ‘berries’ intonation in English (Beckman & Pierrehumbert Reference Beckman and Pierrehumbert1986: 273). Wee presents the pitch tracks of monosyllabic words in Singapore English (Wee Reference Wee2008), showing that downdrift is weak if there is no preceding L. In addition, he suggests that downdrift is suppressed in Hong Kong Cantonese, a tonal language where tone is lexically contrastive.

Chapter 4 unpacks the discussion on whether tone sandhi is phonetic or phonological. It examines the question of whether the T3 sandhi (T3 → rising / _ T3) is phonological or phonetic in production and perception. In the production of Beijing speakers, pitch profiles of the derived rising tone are similar to the underlying rising tone T2 (Xu Reference Xu1997), which suggests phonological neutralisation. In perception, the two rising tones are not distinguishable. Another tone sandhi, T3 → [21] / _ {T1, T2, T4}, is also not purely phonetic, because [14] is better in terms of articulatory ease than [21] before T1 ([55]) and T4 ([51]). Sandhi rules in Tianjin, Hakha Lai and Taiwanese all offer examples which are too complex to be explained by phonetics alone. This reinforces the argument for sandhi being phonological, rather than just phonetic.

In Chapter 5, Wee introduces OT analyses of tonal patterns in different languages. For example, Margi tone spread is realised under the ranking *NoTone (‘do not have toneless TBUs’) ⪢ *MultiLink[TBU] (‘a TBU must not be associated with more than one tonal element’). The basics of OT are explained at the outset, including Eval, Gen and tableaux, so those who are not familiar with the theory can understand the analysis. Wee also introduces advanced theories in OT in the analysis of a seemingly arbitrary Taiwanese tone-sandhi circle: anti-faithfulness (Alderete Reference Alderete2001), comparative markedness (McCarthy Reference McCarthy2003) and contrast preservation (Łubowicz Reference Łubowicz2003). The chapter also discusses rule-ordering effects and opacity. For example, for Hakka Lai tone sandhi, extrinsic rule ordering is necessary. In Standard Chinese, lexicalisation of phrasal entities or compounds involves the deletion of the tone of the final syllable (T3 T3 → T2 Ø). Here tone sandhi must be ordered before H-deletion (counterbleeding). A full analysis in the OT framework (e.g. Sympathy Theory (McCarthy Reference McCarthy1999) or Stratal OT (Kiparsky Reference Kiparsky, Uyechi and Wee2010)) is not given, but readers who wish to further examine this topic are directed to related references.

Chapter 6 discusses segment–tone interactions. In addition to tonogenesis, discussed above, this chapter deals with tone and prosody. Unlike, for example, iambic pentameter in stress languages, rhythmic distribution of particular tones is rare in tone languages. One possible instance of metrical tone is verse from the Tang dynasty. Tang verse consists of four pentasyllabic or heptasyllabic lines in a stanza, and has sequences of Ping vs. non-Ping (or Ze) tones (from the Middle Chinese tone categories Ping ‘flat’, Shang ‘rise’, Qu ‘depart’ and Ru ‘enter’). Ping syllables can only rhyme with Ping syllables, and Ze only with Ze. However, the tone sequences of Tang verse may originally have been the syllable structure of Archaic Chinese before tonogenesis, so Wee concludes that evidence for purely metrical tone is weak.

Chapter 7 deals with musicality, the tone–tune interface, brain activity and tone acquisition. Wee presents evidence from Chaozhou and Beijing ‘operatic arts’ (Wee Reference Wee2007) for tone–tune mismatch, and compares musical notes and tonal notations. For example, a short level tone (tiap 5 ‘butterfly’) becomes a complex tune (with four musical notes), and a falling tone (an 51 ‘shore’) has a rising melody. Brain reactions to tone are only discussed briefly, but the section is full of interesting research results. Native speakers of tone languages show pitch consistency, so their pitch contours across a long time-lag (two different days) showed less variation in pitch profile than those of English speakers. Wee adds his own results on pitch profiles, and shows that Shicheng Hakka speakers’ pitch profiles remained almost identical after a 14-day lag, whereas British English speakers displayed marked differences.

In the remainder of this review, I will consider the book's major themes and proposals, and show how they are developed with respect to various languages.

2 Phonological representation of tone

Chapters 3, 4 and 6 develop the phonological representation of tone. §3.6 compares various tone models from the literature, which contain features ([high] and [low]) and register (high and low, represented by Wee as Hr and Lr respectively). Wee argues against the ‘register-dominance’ model, in which register dominates tone features (e.g. Yip Reference Yip1980), on the grounds that changes in register in languages such as Chaozho and Ewe do not affect tones. He also argues against the ‘parallel registers’ model, where tone features are each assigned a register (Duanmu Reference Duanmu1990). This model is adequate for languages with full rising and falling tones ([25], [214] in Standard Chinese), allowing a different register to be assigned for each tone feature. However, it predicts a non-existing pattern involving tonal contrasts in each register, for example between high rising/falling and low rising/falling contrasts. Such contrasts seem to exist in San Juan Copala (Trique), Hangzhou ([51]) and Lanzhou ([24], [42]). Wee claims this model is unnecessarily complicated, since these tones are in fact registerless rather than being a case of register-crossing. He adopts the ‘feature-subjacency’ model (Hyman Reference Hyman1987), where tone consists of register and contour, and the contour node dominates tone features separately from register. Yip (Reference Yip2002) also discusses the tone models of Duanmu (Reference Duanmu1990, Reference Duanmu2000) and Bao (Reference Bao1999), but Wee explores a larger set of possibilities in the feature geometry, and justifies the feature-subjacency model with novel examples from Hangzhou and Lanzhou.

The proposed model has an advantage in explaining OCP phenomena in tone languages. In Chapter 4, Wee applies the OCP to each level of the model, i.e. OCP[T], OCP[Rg], OCP[Ctr] and OCP[t] (respectively OCP for tone (H, L, R, F), register, contour and tone feature (h, l)), and illustrates the model with examples from Chinese and Taiwanese tone languages: Leling (OCP[Rg]), Dongshi (OCP[Ctr]) and Tianjin (OCP[t]). In Tianjin, for example, a sandhi rule R + H → LH is best explained in terms of the OCP applying to tone features, i.e. as lh.h → l.h, where lh = R, h = H, l = L.

Furthermore, the tone model is integrated with syllable structure in order to explain segment–tone interactions (§6.1). For this, Wee suggests an analogy between the onset–rhyme model of syllable structure and the register-contour model of tones. In many languages, register interacts with syllable onset. For example, Songjiang (Wu) has tones that differ in register, depending on the voicing of the onset consonant. For example, it has the falling tones [53] with a voiceless onset and [31] with a voiced onset. To account for this, Wee combines the tonal tier and the articulatory tier (Fig. 6.4). The [±voice] feature on the onset consonant determines the register of a tone, and the tone features are associated with the mora on the articulatory tier. Onset–tone interaction naturally follows this model, and language-specific details of association can vary. For example, in languages where there is no interaction, onset and tone are left unassociated. In languages where sonorants allow H-spreading (e.g. Ngzim), sonorants are not specified for register properties.

3 Tone–segment interaction and tonogenesis

Wee illustrates the interaction between segments and tones with examples of the historical development of tones from Ancient to Modern Chinese (§6.2). In his model, there are interactions between tone and other aspects of prosody (syllable structure, onset and coda segments, vowel length). He presents a facsimile of a woodcut page from a 19th-century Chinese dictionary (Shuowen Jiezi), which dates back to the Han dynasty (53 BCE–220), and was also reproduced in the Song Dynasty (960–1279) (Fig. 6.5). This allows us to trace phonological changes from the Han dynasty via the Song dynasty to Modern Chinese. Of particular interest is the facsimile of a page from Yunjing, a Chinese rhyme table (Fig. 6.6), which resembles the modern IPA chart. The table consists of columns of onset consonants arranged by place of articulation and VOT properties, and rows of tone categories.

Chinese tonogenesis provides a case of historical interaction of syllable types, vowel length, segment and tone. According to Wang (Reference Wang1980), before the 4th century tone categories were syllable types (open or closed) in Archaic Chinese. Later, both open and closed syllables split into long and short vowel categories, yielding the four Middle Chinese syllable types (Ping, Shang, Qu and Ru). Later the Ping category underwent register split into Yin and Yang (i.e. Hr and Lr), depending on onset voicing, giving the tone categories Yin Ping (T1), Yang Ping (T2), Shang (T3) and Qu (T4) in Standard Chinese (Ru split into various categories).

Wee also illustrates cases where tones arise from stress (Hong Kong English, Tibetan and Scandinavian languages). Hong Kong English has undergone tonogenesis through English–Cantonese language contact. It has developed its current tonal phonology through a tonal interpretation of stress. Wee puts forward several arguments to explain why Hong Kong English is tonal. For example, H tones occur freely, and are lexically stipulated. The placement of tones reflects English stress (Table 6.5), but H-tone placement is stable (maintáin, mainténance), unlike Standard English stress.

Wee's creative presentation of tonogenesis makes this part of the book essential reading for anyone interested in the topic. Seoul Korean can be added to his list, as an example of tonogenesis from laryngeal contrast, where its three-way laryngeal contrast among voiceless stops (plain, aspirated, tense) was formerly realised as differences in VOT, but has progressively shifted to F0 differences among younger generations (Silva Reference Silva2006, Kang & Han Reference Kang and Han2013).

4 Intonation in tone languages

Wee argues that lexical tones and intonational tones can be integrated in theory and description, and suggests that while lexical tones and intonational tones are similar, they differ in where they are associated. Tones can be docked to phrasal domains, not just to word-level domains. For example, H is consistently inserted between two phrases in Kimatuumbi (Odden Reference Odden1987). Conversely, tones can be used to indicate the word-boundaries, as in Singapore English, where word-final syllables always receive high tones (Wee Reference Wee2008), even within compound words (e.g. everybody and every body can be ambiguous). Based on this, Wee argues that tones can have the function of marking domains, across lexical and postlexical levels.

An interesting question arises as to how intonation melodies are realised in languages that also have lexical tones (§2.6). Wee shows that lexical tone languages can restrict the use of intonation to express a wide range of discourse-related meanings, so that F0 can be used only for lexical contrasts. In Cantonese, for example, while the use of intonation is severely restricted, more than thirty sentence-final particles are used to express discourse meanings. This is in accord with Gussenhoven's (Reference Gussenhoven2004: 228ff) observation that intonation contrasts are harder to maintain in languages with many lexical tone contrasts. Thai, however, which has five tones, employs various intonation patterns by modifying the shape of lexical tone contours (Table 2.7).

Intonation conveys discourse meanings in tone languages with smaller numbers of tones. In Hong Kong English, for example, word-level tones interact with intonation. Hong Kong English words contain at least one H tone at the word level (Fig. 2.11d). In declarative intonation, two-syllable words with an [H-Ø] pattern have a declining pitch contour, but the same words have a pitch rise when expressed in questions.

5 Representation of contour tones: level or shape?

The volume touches on the issue of whether a contour tone is merely an interpolation of level tones or whether the transition should be represented as a feature (§3.4). There has been controversy about whether the F0 movement of contour tones should be decomposed into [±high, ±low] level targets, or should be [±rising, ±falling] shape targets. To address this, Wee introduces the analyses of Woo (Reference Woo1969) and Wang (Reference Wang1967). Woo's seems to be the better account, because it requires a simpler inventory and accounts for autosegmental patterns (e.g. Mende [nìká] ‘cow’, [mbǎ ‘rice’). However, in Hangzhou (Hsieh Reference Hsieh, Hsiao, Hsu, Wee and Ho2008), there are two sandhi rules which support Woo's model (e.g. /24-X/ → [22–34]), and two others which support Wang's treatment of contour tone as a non-decomposable feature (e.g. /14-X/ → [13–41]). Here, the rising contour is squeezed into the first syllable as far as possible. Similarly, in Lingxi (Southern Min), the OCP applies to contour tones (e.g. /53–53/ → [33–53]). These examples show that we cannot simply reject contours as features. The Hangzhou and Lingxi examples are consistent with experimental findings in previous studies that show that rising slope is an important cue to the level vs. rising tone contrast (Massaro et al. Reference Massaro, Cohen and Tseng1985), and that contour tones have transition targets in the production of Mandarin (Xu & Wang Reference Xu and Emily Wang2001).