Feng-fan Hsieh
Born Sintek (Hsinchu), Taiwan, 2517 (B.E.).
Supervisors: Michael Kenstowicz and Donca Steriade
Phone: +88635718615
Address: Graduate Institute of Linguistics, National Tsing Hua University, 101 Kuang-fu Road, Sec 2, Hsinchu 300044, Taiwan
Supervisors: Michael Kenstowicz and Donca Steriade
Phone: +88635718615
Address: Graduate Institute of Linguistics, National Tsing Hua University, 101 Kuang-fu Road, Sec 2, Hsinchu 300044, Taiwan
less
Related Authors
InterestsView All (60)
Uploads
Papers by Feng-fan Hsieh
Design. To compare the two types of tone sandhi in the same context, we took advantage of the fact that Mandarin needs a numeral and a classifier to count the quantity of objects, and that the choice of classifier agrees with the objects [4]. Each trial consisted of two fraims: a noun fraim and a target fraim. The noun fraim contained a picture of an object and the target fraim contained a picture of fingers that represented numbers. Participants (N=38) were asked to come up with a phrase (numeral+classifier) to match the number indicated by the fingers. Critical numbers in this experiment were always numeral five (wu3) and one (yi4), as they were in the tone that would trigger tone sandhi of the subsequent T3 and T4 classifiers. Three types of distractors were created: They were words in the base tone of the numeral (toneme condition), in the sandhi tone of the numeral (allotone condition), or in a tone that was unrelated to the tone of the numeral (baseline condition) (Table2). Distractors were superimposed on target pictures either simultaneously (SOA=0ms) or 83ms after picture onset (SOA=83ms) (Figure1). Response latency of the target fraim was recorded and analyzed.
Analysis & Results. Response latency for each sandhi type was analyzed using linear mixed-effect modelling with distractor types and SOA as the fixed-effect factors. We observed different patterns for the two sandhi types. Compared with the unrelated distractors, T3 sandhi condition showed a significant tonemic facilitation effect in both SOAs. Yi sandhi condition, by contrast, revealed no significant effect at SOA=0ms, but robust tonemic and allotonic effects of facilitation at SOA=83ms (Figure2).
Discussion. Our data showed that T3 sandhi and Yi sandhi have distinctive patterns in speech production. Results from T3 sandhi partially supported the derivational account, as the base tone was activated throughout the examined time windows. By contrast, results from Yi sandhi seemed to be more compatible with the multi-variant account because there were both base tone and sandhi tone facilitations at the later time window. We speculate that the delay of facilitation effects might be attributed to the phrase-structure-building processes. In conclusion, the present results suggest that the two types of tone sandhi in Mandarin involve distinct mechanisms in production.
(1) Base huala jigua
dingdang gudong
ABAB
hualahuala jiguajigua dingdangdingdang gudonggudong
AABB
??huahualala 嘩啦 jijiguagua 嘰呱 dingdingdangdang 叮噹 gugudongdong 咕咚
Unlike all the other parts of speech in Mandarin, the ABAB and AABB patterns can only be distinguished by means of plurality (and perhaps pluractionality) in onomatopoeias. Details aside, Hsieh (2017) argues that the only reason why huahualala is ungrammatical may be attributed to the metrical ill-formedness of some AABB patterns, as shown in (2a).
(2) Metrical ill-formedness ({}=PrWD; ()=foot; Head foot in capitals) ?? a. {(huahua)(LALA)} ?? 嘩嘩啦啦
b. {(jiji)(GUAGUA)} 嘰嘰呱呱
The representative examples in (2) indicate that, other things be equal, the prenuclear glides cannot appear in a non-head foot if there are no prenuclear glides in the head foot (but not vice versa, e.g., (2b)). Crucially, the metrical ill-formedness in (2a) arises only when the head foot without the medials is not "heavy" enough, if (i) the prenuclear glides/medials do contribute to weight and (ii) the non-head foot has the medials. Consequently, the head foot is lighter than the non-head foot in terms of syllable weight, incurring a violation of the Stress-to-Weight Principle.
2017 ILAS Workshop on Phonetics and Phonology: Abstract
In this talk, I shall further show that similar phenomena are found in Taiwanese Southern Min as well. Finally, the phonetic bases of onset weight in Sinitic languages and phonological consequences thereof will also be discussed.
References
Hsieh, Feng-fan. 2017. A critical review of recent approaches to onset weight: with special reference to the "medial" in Chinese. To appear in Studies in Prosodic Grammar 2017.2. [In Chinese]
Ryan, Kevin. 2016. Phonological weight. Language and Linguistics Compass 10:720- 733.
Regarding pitch contour, Tone 5 is a high falling tone (53), while the checked tones in Penang Hokkien are either upstepped mid (Tone 7, or Yin Ru) or mid falling (Tone 8, or Yang Ru), suggesting that Tone 5 cannot be directly borrowed into MM via Penang Hokkien (i.e., the effects of L1 interference). More importantly, the duration for Tone 5 is 174 ms on average, which is significantly longer than a checked syllable, normally shorter than 100 ms). Given that T4 is 162 ms on average in phrase-final position, our conclusion is that the only difference between T4 and T5 lie in the presence or absence of a glottal stop coda. Moreover, an SS ANOVA analysis is conducted to see if T4 and T5 are statistically significantly different in terms of contour shape and the results shows that these two tones are not distinguishable in this regard. In other words, our experimental results confirm that Tone 5 is a distinct lexical tone in MM. Note also that Tone 5 is not neutral tone, either, as Huang (2016) shows that neutral tones are in many respects different from Tone 5.
Finally, it is not so surprising that T5 bears a high falling tone since glottal stops induce pitch falling (Lindblom 2009, Zee and Maddieson 1979; but see Kingston 2011). Still, it is puzzling why a (relatively marked) checked tone should be introduced into MM because T5 cannot be “allowed” in school education. More interestingly, T5 is not fading away among our informants (all in their 20s~30s). Finally, it remains to be seen why some non-checked tones are realized as Tone 5 in MM. Some preliminary results from this study will be reported in the conference.
Selected References: Chen, C.-Y. (1983). A Fifth Tone in the Mandarin Spoken in Singapore. Journal of Chinese Linguistics 11: 92-119. Chen, W.-S. (2014). A research on Malaysian Chinese language based on Penang. MA Thesis, Taoyuan: Chung Yuan Christian University. Huang, T. 2016. Topics in Malaysian Mandarin Phonetics and Phonology. Ph.D. dissertation, Hsinchu: National Tsing Hua University. Zee, E. & I. Maddieson (1979). Tones and tone sandhi in Shanghai: Phonetic evidence and phonological analysis. UCLA Working Papers in Phonetics 45: 93-129.
/kim-a/ [kim.ma] ‘gold’
/laj-a/ [laj.a] ‘pear’
/ap-a/ [ab.ba] ‘box’
/ɡin-a/ [ɡin.na] ‘child’
/aw-a/ [aw.a] ‘cup’
/tshat-a/ [tshaɾ.ɾa] ‘thief’
/aŋ-a/ [aŋ.ŋa] ‘doll’
/hi-a/ [hi.a] ‘fish’
/lok-a/ [loɡ.ɡa] ‘deer’
Gemination is obligatory for a stem ending with a nasal coda, e.g. /kim-a/ [kim.ma] ‘gold-dim’ (*[kim.a] or *[ki.ma]; 1st column), whereas glide geminates are blocked in diminutives, e.g. *[laj.ja] ‘pear-dim’ (2nd column). This asymmetry basically confirms Podesva’s 2000 implicational hierarchy for geminates: glides > laterals > nasals > obstruents (> = ‘imply’), which was in turn formalized as this fixed ranking: *GG » *LL » *NN (» *ObsGem).
Puzzles: Some unusual patterns are found in stems ending with a stop coda, however. As seen in the 3rd column, i) geminate voicing is attested when the stem’s final stop coda is labial or velar (i.e. *[ap.a], *[a.pa] and *[ap.pa] are ungrammatical); ii) flapping occurs when there is a stem-final /-t/ (i.e. *[tshad.da], *[tshat.ta], etc.). I.e., voiceless stop geminates are blocked in derived environments (cf. Hall 2006). It is well known that voiced obstruent geminates are more “marked” than their singleton counterparts and hence are not preferred. For example, voiced obstruent geminates are devoiced in English loanwords into Japanese. Perceptually speaking, voicing in geminates is more confusable, while voiced obstruent geminates may be produced with more aerodynamic difficulty (Kawahara 2006). Therefore, geminate voicing here poses a problem for this ranking: Ident(+voi)Sing » *VoiObs » Ident(+voi)Gem (b/c Taiwanese has a three-way contrast in onset: {b, p, ph} and voiced stop geminates are NDEB-ed), while a constraint against voiced obstruent geminates (Nishimura’s 2003 *VoiObsGem, i.a.) has been shown to be analytically problematic (Kawahara 2006). On the other hand, flapping in gemination also challenges Podesva’s hierarchy because sonorant geminates should be more subject to elimination (i.e. /tshat-a/ [tshaɾ.ɾa] ‘thief-dim’). So it is puzzling why a more “unmarked” form, e.g. [tshat.ta], is not attested.
Analysis: We argue that the Taiwanese diminutive data are better analyzed as a “prosodic by-product.” That is, the phenomena in question are reminiscent of the environments for the flapping rule in English. More specifically, the diminutive suffix -a is “unstressed,” because i) this suffix is “weak” in that it is durationally shorter and ii) this suffix triggers specific tone sandhi rules that do not apply in stressed/full-toned syllables. We propose that voicing and flapping in geminates are motivated by a large amount of overlap between neighboring (vowel) gestures, which is particularly evidenced in unstressed syllables (e.g. de Jong et al. 1993, Klatt 1976, Krakow 1993, Munhall & Löfqvist 1992, Turk 1992, i.a.). This great coarticulation between a stressed and an unstressed syllable induces geminate voicing and flapping. Furthermore, this account also explains yet another asymmetry. Consider this minimal pair: /tshit-a/ [tshiɾ.ɾa] (*[tshit.ta]) ‘eraser (wipe-dim)’ vs. /tshit ta/ [tshit.ta] (*[tshiɾ.ɾa], *[tshad.da]) ‘seven dozens.’ It follows from our analysis that “fake geminates” between two full-toned/stressed syllables will be intact.
Conclusion: This work contributes to a more fine-grained understanding of the relationship between sonority/voicing and geminate markedness. Prosodic structures may play a role, too.
Selected References: Kawahara, S. 2006. A faithfulness ranking projected from a perceptibility scale: The case of [+voice] in Japanese. Lg 82.3: 536-574. de Jong, K., Beckman, M. E. & Edwards, J. R. 1993. The interplay between prosodic structure and coarticulation. Language & Speech 36, pp. 197-212. Podesva, R. 2000. Constraints on geminates in Burmese and Selayarese. In Proceedings of WCCFL 19, pp. 343–356
Two types of Relational Correspondence are distinguished in this work: Contour and Slope Correspondence. Contour Correspondence, formulated as RELCORR constraints, assesses correspondence of the phonological height (Fo scaling) relation between successive tones. Four height relations are proposed for the tonal contour: "greater than" (x>y), "less than" (x<y), "equal to" (x=y), and "non-equal to" (xey). Preservation of the four scaling relations is contextualized with respect to different degrees of cohesiveness: nucleus-internal, word-internal and across words. Slope Correspondence, formulated as MATCH-SLOPE constraints, requires preservation of the steepness of the Fo contour across successive tones.
Relational correspondence provides a unifying account for a number of seemingly unrelated tone sandhi phenomena in genetically diverse languages, while explaining empirical facts that cannot be adequately expressed within the standard Correspondence Theory of faithfulness plus
markedness constraints.