1. What can a Tonal Template do for Phonetics? Te-hsin Liu Department of Linguistics Paris 8 University liu.tehsin@gmail.com

2. The issue Contour tones are said to be phonetically more difficult and time-consuming to produce than level tones (Yip 2002, Zhang 2002). However, if African tonal languages do confirm this assumption, Chinese languages show two surprising characteristics: 1. First, while there is no system with only contour tones in African tonal languages, there is no system with only level tones in Chinese, and some have no level tones at all. This runs counter to the claims of markedness theory: the presence of any single marked element without its unmarked counterpart is ruled out in phonological systems

3. The issue Second, the high level tone is not shorter than the falling contour tone in Mandarin. (Kratochvil 1968, Xu 2004). Figure (1): Mandarin four tones pronounced in isolation (Xu 2004)

4. The goal of our work Answering the above paradoxes: Why is there no system with only level tones in Chinese, and why can some dialects lack level tones? Why are level tones not shorter than contour tones in Chinese?

5. On the linearity of tone (1) Why level tones are long in Chinese? Mapping mora-tone hypothesis (Duanmu 1990): a level tone is associated to two moras. However, there is no vowel length contrast in Chinese So duration is intrinsic to tone In autosegmental phonology, duration is translated by temporal positions Tones are temporal positions

6. On the linearity of tone (2) What allows us to say that, in a falling contour tone HL, H precedes L? If there was no linearity in the tonal domain, H+L would give rise to a mid tone, just as I+A=/e/ and not /ja/ or /aj/! If H+L does not give rise to a mid tone but a falling contour tone, it shows that H precedes L, and that there is an intrinsic linearity.

7. On the linearity of tone (2) In Hyman (1993), the difference between a mid tone and a contour tone is encoded by tonal nodes However, what is the nature of tonal root node?

8. In autosegmental phonology, the only thing that encodes duration and linearity is skeleton

9. Hypothesis of a HLHL tonal skeleton There is a universal tonal periodic skeleton HLHL (Carvalho 2002), analogous to the syllabic skeleton CVCV (Lowenstamm 1996) Chinese tones are constrained by a portion of this periodic skeleton: a tonal template HLHL. Chinese contour tones can be analyzed as a succession of level tones defined by an intratonal government relationship H/L encoding the notion of register.

10. Hypothesis of a HLHL tonal skeleton The vertical line indicates the tonal head. The register is low if and only if the head is low; it is high if and only if the head is high. The four citation tones in Mandarin:

11. Implications of the present hypothesis (1) Level tone is long, since it constitutes a contour just as a long vowel or a geminate consonant, associated to two positions. However the ‘melody’ is the syllable in the present hypothesis; the tone is the skeleton.

12. Implications of the present hypothesis (1) Phonetic argument (Kratochvil 1968, Xu 2004): Tone 1 is longer than tone 4 when pronounced in isolation Figure (1): Mandarin four tones pronounced in isolation (Xu 2004) Phonological argument: Template Satisfaction Condition (McCarthy&Prince 1986) Languages tend to fill in empty positions; The long duration of level tones suggests not only the existence of a tonal skeleton, but also that of a tonal template.

13. Implications of the present hypothesis (2) -Tonal markedness Analogy between tones and syllables within a CVCV approach: Just as CV is unmarked compared with.VC., the falling tone HL is unmarked by comparison with the rising tone LH, because the latter supposes two empty positions on its right and left sides. Typological argument: In a statistics on 187 tonal languages, Zhang (2002) noticed that 37 languages have a falling tone without the rising one. Only three languages have a rising tone without the falling one: Margi, Lealao Chinantec and Zengcheng. Argument from language acquisition: The low falling tone is acquired by 2;9 by Cantonese speaking children, and the low rising tone is learned at 3 years old (A. Tse 1992).

14. Implications of the present hypothesis (2) -Tonal markedness Level tones are more marked than contour tones: Their lexical representation involves not only two empty tones just as the rising tone, but also a median empty tone; i.e. two successive empty L-slots (= VCC). Their existence should imply the presence of contour tones, but the opposite is false. Typological argument: A language can have only contour tones without level tones, as in Chengtu, Shanghai, Zhenhai, guiyang, longquan, Pingyao and Wuxi. No Chinese dialect has only level tones.

15. Arguments in favor of the existence of a tonal skeleton (1) Fanqie (reverse cut) languages (data from Bao 1990) Reduplication usually acts on skeletal units: thus, phonological length is transferred from the base to the reduplicative affix, i.e. in Mokilese, and length is encoded, in autosegmental theories, by skeletal units. Tones are stable and reduplicated under the substitution of onset, rime and coda. So tones are skeletal units in Chinese.

16. Arguments in favor of the existence of a tonal skeleton (2) Realization of neutral tone According to a phonetic study of Shih (1987), the realization of toneless syllables preceded by four lexical tones is as follows: Preceding tone Toneless syllable 55Hstarts high, then falls 35LHstarts high, then falls, but not as low as after 55 21(4)HL(H)starts fairly low, then rises 53HLstarts fairly low, and falls even lower

17. Arguments in favor of the existence of a tonal skeleton (2) Analysis: L2 is the landing site for neutral tone The toneless syllable associates to the plain tone of S1, and to the following position if any.

18. Further implications (1) The role of purely phonetic factors in shaping phonological systems has to be reassessed since they do not explain: (i) why there are systems with only contour tones in Chinese; (ii) why level tones are long in Chinese. Actually, contrary to African tonal languages, tones might be positional objects in Chinese.

19. Further implications (2) The hypothesis of a tonal template raises an important issue on cognitive grounds: Just as English and French-speaking children who use fixed templates, an abstract phonological object that is emergent and available very early on, to build their phonology (Vihman 2001, Wauquier 2005), the tonal template might have the same function in Chinese.

20. Further implications (3) This conjecture is confirmed by Li & Thompson (1977) and Clumeck (1980), whose studies show that Mandarin-speaking children acquire tones before the acquisition of segments. J. Tse (1978), working on the language acquisition of a Cantonese-speaking child, confirms as well that tones are acquired before segments. In other words, Chinese-speaking children might use the tonal template to build their phonology, consonants and vowels being inserted to the template progressively.

21. Thank you for your attention!