On tone and syllable structure in Cantonese

1
On tone and syllable structure in Cantonese
Alan C. L. Yu University of Chicago
2
Goals

• To reexamine certain claims about the interaction
  between tone and syllable structure in the
  literature.
• The focus will be on the realization contour
  tones.

3
Phonetic approach to contour tone licensing
condition

• The longer the sonorous portion of the rime, the
  more complex a tone can be (Gordon 1999/2001,
  Zhang 2000).

The tolerance of contour tones on syllables
which are inherently less well suited to carrying
tonal information implies the tolerance of
contour tones on syllables which are better
suited to manifesting tone (Gordon 2001447) CVV
gt CVR gt CVO gt CV
4
Cantonese tonal inventory
Why do CV syllables carry rising tones but
CVO/CVVO syllables do not? (cf. CVO gt CV)
5
Gordon demonstrates that the rhyme duration of
smooth syllables is significantly longer than
that of checked syllables in Cantonese.
6
An Exception

• While contour tones are restricted to
  phonetically long syllables in Cantonese, they
  are suspended in certain derived forms
• CV(V)O syllables may carry a rising tone (25)
  when they undergo a process called Pinjam
  ??Changed Tone.

7
Pinjam ?? in Cantonese
8
An analysis

• Gordon (1999) argues that CVVO syllables may host
  contour tones in the Pinjam contexts because the
  constraint MAXmorpheme is ranked higher than
  the constraint that licenses contour tones in
  rimes contained solely of sonorants.

9
? in ?? ? ? in ?? p??k?33 to strike p??k?25
a racket
10
Problems

• No empirical confirmation is available.
• Languages vary in their responses to the
  realization of contour tone in syllables with
  insufficient tone-bearing ability.

11
Zhang (2001)

• A contour tone encounters a syllable with
  insufficient tone-bearing ability
• Where does Cantonese fall in this typology?

12
Contour tone-induced lengthening in Cantonese

• A production experiment

13
Subjects

• Six native speakers of Cantonese (three males and
  three females) currently residing in the US.

14
Methods

• Stimuli were monosyllabic Cantonese words of
  three different syllable types (i.e.
  plosive-final, vowel-final, and nasal-final),
  which were chosen to form (near-)minimal pairs or
  triplets (i.e. with identical rhyme).

15
Methods

• For CVV and CV(V)N syllables, the minimal
  triplets consisted of three tone types lexical
  low-mid-rising (25), lexical level or falling
  (22, 33, or 21) and Pinjam-derived low-mid-rising
  (25).

16
CVV stimuli
17
CV(V)N stimuli
18
Methods

• For the CVVO syllables, minimal pairs consisted
  of two tone versions level (22 or 33) and
  derived low-mid-rising (25) no lexical
  low-mid-rising is available.
• Both CVVO and CV(V)N tokens contain some short
  vowels.

19
CV(V)O stimuli
20
Methods

• The targeted syllables were presented as part of
  a disyllabic word phrase to ensure that the
  appropriate pronunciations were rendered.
• Certain pairs of targeted syllables are
  represented by the same Chinese characters the
  semantic and pronunciation differences are only
  apparent when the character is used in the
  appropriate context.
• ? sou33 to sweep vs. ?? t?s?k?55 sou25 a
  bamboo broom

21
Methods

• The subjects recited a list of Cantonese target
  disyllabic words/phrases in the carrier phrase
  ??? __ ??? /?? wui t?k? ___ p?i n?i t?æ?/ three
  times.
• (2 tones x 1 syllable type (CVVO) 3 tones x 2
  syllables (CVV CV(V)N)) x 10 tokens x 3
  repetitions 240 tokens

22
Example stimuli
23
Measurements
f?n25 powder
F0 Peak
Turning Point
24
Measurements

• excursion size pitch difference (in semitone)
  between adjacent f0 minimum and maximum in the
  target syllable
• excursion time time interval between adjacent
  f0 minimum and maximum in the target syllable
• excursion speed (excursion size/excursion time).

25
Hypotheses

• Null hypothesis no difference (i.e. the rhyme
  of a CV(V)O syllable remains the same regardless
  the type of tone it bears.
• Phonetic approach to contour tone licensing
• CV(V)O syllables with a derived rising tone are
  longer than their level-toned counterparts.
• No comparable lengthening should be observed in
  CVV or CV(V)N syllables since they are already
  phonetically long.

26
Duration analysis

• How the presence of a contour tone affects
  syllable duration

27
Level vs. Derived Rising

• CV(V)O with derived R is significantly longer
  than CV(V)O with a level tone.
• No comparable length difference is found in the
  other syllable types.

28
Discussion

• These results show that CVVO syllables under
  Pinjam are no exceptions to Gordons
  generalization.
• When a CVVO syllable acquires a contour tone, its
  intrinsic shortness is remedied via the
  lengthening of the rhyme in Cantonese.
• Similar patterns are found in Mitla Zapotec
  (Briggs 1961), Wuyi Chinese (Fu 1984), Gã (Paster
  1999).

29
Question

• While the presence of a contour tone affects the
  duration of the rhyme, what effect does the
  shortness of CVVO have on the realization of a
  contour tone?

30
Analysis of the f0 contour

• How syllable structure affects contour tone
  realization

31
The rising contours
32
Patterns of contour alignment

• The f0 peak is reached at approximately 90 point
  of the rhyme regardless of syllable structure.

33
Patterns of contour alignment

• The turning point in CV(V)O comes significantly
  earlier than that in CVV syllable.
• The difference in turning point alignment between
  obstruent- and nasal-final syllables did not
  reach the adjusted level of significance (p
  0.036).

34
The shape of the contour
35
Excursion size (the rise)

• The difference between the excursion sizes of the
  vowel- and nasal-final syllables reaches
  significance.
• The differences in excursion size between CV(V)O
  and CVV and between CV(V)O and CV(V)N did not
  reach significance.

p lt 0.01

36
Excursion time

• The excursion times for all three syllables are
  significantly different from each other.
• This suggests that a rising tone in a CV(V)O
  syllable must cover as large (if not larger) a
  pitch rise as CVV syllables but in a much shorter
  amount of time.

37
Excursion speed (the slope)

• The pitch excursion speed was significantly
  faster in the CV(V)O syllable than in the
  nasal-final syllable (p lt 0.001) and than in the
  CVV syllable.
• No difference in slope between CVV and CV(V)N is
  found.

38
Discussion

• The results suggest that rising tones are
  realized significantly differently across
  syllable structures.
• Tonal target undershoots are observed in
  obstruent-final and vowel-final syllables.
• Yet, the nature of the undershoot differs
  depending on syllable structure.

39
Onset truncation in CV(V)O

• When the syllable is obstruent-final, tonal
  target undershoot is in the form of
  onset-truncation.
• The beginning of the rise originates from an
  earlier point relative to the onset of the rhyme
  and a higher f0 in an obstruent-final syllable
  than in a nasal-final syllable.

40
End truncation in CVV

• When the syllable is vowel-final, tonal target
  undershoot results in end-truncation instead.
• While the rising tone in both vowel- and
  nasal-final syllables originate from the same
  point, the rise in a vowel-final syllable
  culminates at a lower f0 than observed in a
  nasal-final syllable.

41
Discussion

• Gordons assertion concerning the exception
  nature of rising tone in CV(V)O is not warranted.
• Rhyme duration is lengthened and the rising
  contour is reduced when an obstruent-final
  syllable acquires a rising tone from Pinjam.

42
Puzzle

• If contour reduction is possible, then why is
  lengthening necessary?
• Lexical rising tone syllable vs. its level tone
  counterpart?

43
Level vs. Lexical Rising

• CVV and CV(V)N syllables are longer when their
  bear a lexical rising tone than when they carry a
  level tone.
• The difference in duration between level and
  derived rising toned syllables and between
  lexical and derived rising toned syllables do not
  reach significance level.

44
Puzzle

• Syllables with long rime duration do not
  necessarily escape the lengthening effect, since
  CVV and CV(V)N syllables are longer when they
  bear a lexical rising tone than when they bear a
  level tone. Why should this be?

45
Some explanations

• Many researchers have characterized the observed
  differences in syllable duration with respect to
  the level, extent, and direction of fundamental
  frequency as physiologically conditioned.
• For example, studies on the maximum speed with
  which pitch can be changed found that subjects
  were able to perform pitch drops considerably
  faster than pitch elevation even though the pitch
  ranges are comparable (Ohala Ewan, 1973
  Sundberg, 1973, 1979 Xu Sun, 2002).

46
Perception of tone and duration

• Lehiste (1976), for example, found that
  listeners judge a dynamic (falling-rising or
  rising-falling), as supposed to a flat f0
  pattern, to be longer even when the stimuli are
  of equal acoustic durations.
• This finding was replicated in other studies on
  perceived duration of isolated vowels (Pisoni,
  1976 Wang, Lehiste, Chuang, Darnovsky, 1976).

47
Research question

• Does the perceptual effect found in earlier
  studies on dynamic tone and duration extend to
  tonal languages?

48
Methodology
49
Stimuli

• A 300 ms pa syllable was synthesized using
  SynthWorks.
• A 3-step duration continuum was created with 100
  ms. decreasing increments 300, 200, and 100 ms.
• The f0 of the syllable was manipulated to make
  five stimuli of varying f0 contours.

50
Subjects

• Seven native speakers of Cantonese (2 males 5
  females), all students at the University of
  Chicago, were paid a nominal fee to participate
  in the experiment.
• None of them report any speech or hearing
  problems.

51
Procedures

• Subjects were asked to rate each stimulus for
  duration on a 7-point scale, with 1 being the
  shortest duration and 7 being the longest.
• Subjects first completed a practice session of
  nine trials with the 33 tone stimuli at three
  durations presented three times in a random
  order.
• After practice, subjects completed an
  experimental session of 300 trials (5 tones x 3
  durations x 20 blocks) with the order of trials
  randomized within each block.

52
Procedures

• The presentation of the stimuli was controlled by
  the subject.
• Presentation of the next item starts after a
  response is recorded.
• The experiment was administered using E-Prime.

53
Results
To facilitate cross-subject comparison, the data
was first normalized by transforming the rating
value to a z-score scale.
54
Discussion

• The experimental results suggest that the height
  and slope of fundamental frequency can
  significantly affect the listeners perceived
  duration of speech sounds.

55
Discussion

• The results w.r.t dynamic f0 are consistent with
  previous findings.
• Dynamic f0 generally lengthens subjective
  syllable duration.

56
Discussion

• There is an intrinsic perceptual skewing
  introduced by dynamic f0 movement regardless
  whether the language is tonal or not.
• The contour-tone lengthening effect observed in
  synchronic typological patterns and diachronic
  changes may not be purely a matter of
  physiological constraints on tonal implementation.

57
Discussion

• The fact that lexical rising toned CVV and CV(V)N
  syllables have longer rhymes than their level
  toned counterparts may be because rising toned
  syllables are perceived as longer than level
  toned syllables.
• This perceptual skewing might have been
  lexicalized as part of the phonetic makeup of
  rising toned syllables.

58
Conclusion

• While duration and sonority are important factors
  for contour tone realization, the reverse
  relationship appears to be equally important.
• Duration and sonority might be just as likely to
  be influenced by tonal movement as contour tone
  realization may be dependent on duration and
  sonority.

59
The End