University of Cologne

Tonal coarticulation in German learners of Standard Chinese

In East Asian tone languages, tonal realisation is considerably influenced by adjacent tones (e.g. Han & Kim, 1974 and Brunelle, 2009 on Vietnamese; Abramson, 1979 and Gandour et al. 1994 on Thai; Peng, 1997 on Taiwanese; Shih, 1988, Shen, 1990 and Xu, 1994 & 1997 on Standard Chinese), although the directionality, magnitude and temporal extent of tonal coarticulation vary across languages. Specifically, it has been found for Standard Chinese (Xu, 1997) that tonal coarticulation is predominantly progressive and assimilatory. This means that in native production the F0 of the initial syllable in a disyllabic word is largely unaffected. A number of studies have reported that advanced English learners of Standard Chinese who are able to produce lexical tones accurately in citation form have considerable difficulties producing these tones in sequence, e.g. in disyllables (Chen, 2000; Shih, 2010; Yang, 2011, among others). This opens up the question as to whether these difficulties are due to deviant tonal coarticulation patterns. Deviation might arise for a variety of reasons, e.g. because L1 has regressive coarticulation, or because L1 has a sparse tonal specification, allowing tones to spread onto adjacent syllables. In German, for example, tonal crowding triggered by right-edge boundary tones leads to the shifting of pitch events to the left. For German learners, then, the final portion of the f0 contour of a given tone may be affected by the presence of a following tone. This is not expected to be the case for native speakers.

Five native speakers and ten German learners read aloud three repetitions of 8 syllables, wēi, wéi, wěi, wèi, māo, máo, mǎo and mào, combining the segmental string weimao and the four tones in two conditions: (a) syllable in isolation and (b) first syllable of a disyllabic compound word (24 items × 3 repetitions × 15 speakers = 1080 data points). F0 values were extracted at 10 equidistant points for each syllable. The mean f0 value of the last 20% of each target syllable was predicted with a mixed-effects model featuring TONE {1,2,3,4}, GROUP {native, learner}, CONDITION {monosyllabic, disyllabic} and their two- and three-way interactions as fixed factors, and intercepts for SPEAKER {1,…15} and REPETITION{1,2,3} as random effects. Likelihood Ratio Tests did not reach significance when comparing the full model to a reduced model without the three-way interaction (χ2(1)=1.09, p=0.77), but reached significance when the crucial Group:Condition interaction was dropped (χ2(1)=25.18, p<0.0001). Taken together, the results can be interpreted as suggesting that the differences between the production of tones by learners and native speakers are stronger when an upcoming tone is present, a finding that is consistent with the hypothesis that learners are more likely than native speakers to produce anticipatory tonal coarticulation.

These results provide evidence for regressive coarticulation in L2 production, a pattern that is very different from the progressive coarticulation reported for native speakers. Whether this coarticulation is triggered by the presence of regressive assimilation patterns or sparse tonal specification in L1 (German) remains to be corroborated.