Tone Alignment in San Juan Quiahije Chatino1 J. Ryan Sullivant
University of Texas at Austin Presentation to LASSO XL 1 October, 2011
0. Aims
[email protected]
This presentation seeks to provide an alternative for the representation of tone of San Juan
Quiahije Chatino outlined in Cruz (2011). I propose that that all tones in the language are formed
by tonal complexes in which a tone target can be aligned to either the beginning or the end of the complex. This analysis is motivated by phonemic evidence including neutralizations, perceptual difficulties, and instrumental phonetic data investigating the f0 realizations of the unassociated portions of tonal complexes. 1. Introduction
1.1 Introduction to San Juan Quiahije Chatino
San Juan Quiahije Chatino (QC) is a topolect (ISO 693-3 ctp) of the emergent Eastern
Chatino language (distributed across ISO 693-3 ctp, ctz, cly, ctn) (Campbell 2011; Kaufman
2006) spoken in the Juquila district of southwestern Oaxaca, Mexico. QC is spoken by some 3,500 people (Cruz 2011) in its eponymous municipality.
Like other Zapotecan languages, QC features lexical tone; however the inventory of
Eastern Chatino tone categories and sandhi rules is much larger than those of other Zapotecan languages.
QC, like most Chatino languages has undergone a process of non-final vowel syncopation
(Campbell 2011) in which the historic disyllabic roots have been reduced to monosyllables.
Thus, nearly all simplex words are monosyllabic, meaning that the stem (the domain of tone in most of Eastern Chatino)(v. Villard 2009; McIntosh 2010) is coterminous with the syllable in QC2.
1
I would like to thank my colleagues Hilaria Cruz and Emiliana Cruz, native speakers of Quiahije Chatino for their expert assistance in stimuli preparation and patience with experimental methodology; I would like to thank Anthony Woodbury for helpful comments on the Alignment Proposal and for introducing me to tonal feet. I gratefully acknowledge support for our work through Endangered Language Documentation Programme grant MDP0153 to the University of Texas at Austin, offered by the Hans Rausing Endangered Language Project at the School of Oriental and African Languages, University of London. 2 Evidence from 'fake compounds' suggests that the tonal domain has been reanalyzed in QC as the syllable (Cruz 2011:253ff). While a compelling topic in itself, this matter is outside the scope of this paper and my speaking of syllables associating with tone categories should be seen as an expositional shortcut and not a theoretical stance.
1.2 San Juan Quiahije Chatino Tonology
A recent phonemic analysis of QC tone has been written, though its treatment of some
tones does not appear to capture how certain tones fall together phonemically and makes incorrect assumptions about the phonetic structure of these tones. 1.2.1 Cruz (2011) for simplex words
QC has fourteen phonemic tone categories. Each is composed of up to two of the four
tonal primitives in the language (L M H and the superhigh 0) and may be associated with one of two unlinked floating tones (+H or +0). Table (1) shows the inventory of tones in QC.
Table 1: The Tonemes of QC (from Cruz (2011:124)) type
toneme
comments
isolation
Level
/H/
slight decline
[H]
/M/
slight decline
[M]
/H+0/ /L/ /_/
Rising
breathy, not sandhi-affected breathy, sandhi-affected
[H] [L] [L]
/M0/
steep rise
[M0]
/MH/
slight rise
[M^]3
/+H/
/LM/ /L+0/ Falling
slight decline
steep rise
long, with "elbow" long, with "elbow"
[MH] [LM]
[L+0]
/0L+0/
sharp fall
[0L]
/ML/
sharp fall
[ML]
/HL+0/ /ML+H/
sharp fall sharp fall
[HL]
[ML]
Not included in Table 1 is the super-high tone [0] which is an allotone of the weak L tone (/_ /) after a syllable associated with a floating super-high (/+0/) tone. 1.2.2 Cruz (2011) for complex words
Tone categories are not distributed evenly across word classes. For example, the Low
tones of nouns may be specified (/L /) or unspecified (/_ /), but only unspecified Low tones (/_ /) 3
The carat is used by Cruz to indicate the rising nature of the /MH/ tone, as opposed to the falling nature of the /M/.
are found in verbs. More interesting distributional lacunae can be found in complex words. 1.2.2.1 Inflection for 1IN
Verbs may be inflected for person to indicate their subject, and inalienably possessed
nouns for their possessor (1). While the inflection of the 2S and 1S forms are indicated by a replacement of the stem's tone with another (often predictable) tone4, the 1IN inflection is indicated by the addition of an enclitic (1)
a. sti /_ /
/+H/
to the base (i.e. third-person) tone (Table 2).
b. sti/LM/
sti[L]
c.
sti[LM]
d. st
/_/ [L]
/M/ [M]
-
-
/+H/ [MH]
father.of.3
father.of.2S
father.of.1S
father.of-1IN
'his/her/its father'
'your father'
'my father'
'our (inclusive) father'
Table 2: Person inflection for nouns and verbs (from Cruz (2011) a.
Stem tone
2s tone
1s tone
1in tone
/H/
/LM/
/M0/ or / _ /
/H-+H/
/H/
/L+0/ or /M/
/M-+H/
/ _ / or /M0/
/0-+H/
b.
/+H/
/+H/ or /LM/
d.
/MH/
/H/
/L+0/ or /MH/
f.
/_/
/LM/
/ _ / or /M0/ or /H/
/ML-+H/ /HL+0/
c. e.
g.
/M/
/M0/
/LM/
/L/
/+H/
/M0/
/M0/
h.
/ML+H/
/LM/
/ML+H/ or /M0/
j.
/LM/
/+H/
/ML/ or / _ /
i.
/HL+0/
/M0/
/L+0/
/H-+H/
/M-+H/
/ML-+H/ /HL+0/
/LM-+H/
Note the neutralization of the distinction between the tone pairs (a-b) and (c-d) in the 1in forms, as well as the phonemic shift from /M0/ to [0] in line (e.). 1.2.2.2 Compound words
QC-speaking researchers have reported difficulty diagnosing the tone category of the first
element of a compound word, especially during running or quick speech (Emiliana Cruz and 4
The first-person inflections and enclitics both bring about the nasalization of the final vowel of the stem, if oral, and will phonemically lower a high vowel.
Hilaria Cruz, p.c.). For example, they report that it is usually difficult to determine whether the
cran morpheme tu/MH/ ('hole' or 'mouth') of compounds such as tu/MH/tʔwa/L/ 'mouth', tu/MH/jne/LM/ 'throat' and tu/MH/
/M/
'sky' is tone /M/ or /MH/.
1.2.3 Criticisms of Cruz (2011)
The neutralizations found in 1in person inflection and in the perceptual difficulties above
suggest that the tone pairs /M/-/MH/, /H/-/+H/ and /M0/-[0] may fall together in some
phonological class. Furthermore, a production study of one speaker's f0 peaks of various QC
tones, failed to find significant differences in f0 peaks of /M/ and /MH/ words (Sullivant 2011). Furthermore, the analysis in Cruz (2011) proposes that the difference between the two
natural classes of phonetic 'level' and 'rising' tones is in the number of tonal primitives linked to each syllable (Table 3).
Table 3: 'Level' and 'Rising' tones according to Cruz (2011) type
phonetic tone
phonemic sources
Level
[0]
/_/ after /H+0/, /HL+0/, /L+0/, /0L+0/ slight decline
[H]
/H/, /H+0/ in certain contexts
slight decline
[M]
/M/
slight decline
[M0]
/M0/
steep rise
[M^]
/MH/
slight rise
Rising
[MH]
/+H/, /L/ in certain contexts
comments
steep rise
Such a proposal fails to capture the differing shapes of tones /ML/ (which is a fall from a high level down to the bottom of the speaker's range) and /LM/, which is a low rising tone which
features a prominent elbow midway through the syllable. It also suggests that the rising tones
have two f0 targets whereas the level tones have only one f0 target, which at first blush appears to be incorrect as the some of these purported f0 targets fail to surface. 1.2.4 Non-surfacing Cruz (2011) targets.
Figures (1-2) show instances of the non-surfacing of a tonal target indicated by the Cruz
(2011) analysis. In Fig. 1, none of the purported [M] tones of the three [M0] tones surface as such. In Fig. 2, the purportedly rising [MH] tone is actually falling throughout the syllable.
Figure 1: The initial [M] of a tone [M0] fails to surface /[M0]_ 400 350
Pitch (Hz)
300 250 200 150 sja[M0]
na[M0]
lwi[M0]
ʔĩ[ML]
ʃwa[M^]
ⁿt̻ha[M]
ha[L] la[MH]
ʔĩ[H]
0
4.189 Time (s)
Figure 2: The initial [M] of a tone [MH] fails to surface /[M0]_
Pitch (Hz)
420
300
200 150 sja[M0]
nːa[MH]
kni[M]
kwa[H]
ha[L] ʔne[ML] ka[ML]
su[H]
ʔĩ[0]
0.3032
3.476 Time (s)
Table 4: 'Level' and 'Rising' tones according to the Alignment Proposal type
Level
Rising
tonal complex phonemic sources (Cruz (2011) analysis)
comments
{0_}
/_/ after /H+0/, /HL+0/, /L+0/, /0L+0/
slight decline
{M_}
/M/
slight decline
{H_}
/H/, /H+0/ in certain contexts
slight decline
{_0}
/M0/
steep rise
{_H}
/+H/, /L/ in certain contexts
steep rise
{_M}
/MH/
slight rise
1.3 The Alignment Proposal
The neutralization of the pairs of tones in complex environments and in rapid speech can
be captured by a representation in which each of these tones is formed by only one tonal primitive, but this tonal primitive will be differentially aligned with respect to either the
beginning or the end of the associated vowel, which can account for the falling f0 of the level tones and the rising f0 of the rising tones (Table 4). 1.3.1 The Syllable-Internal Tonal Complex
In Table 4, we can see that each tone category is comprised of a single tonal complex (TC)
(following Akinlabi and Liberman (2000)) in which only one tonal target is specified5. Under the Alignment Proposal, each tone category in QC is composed of a TC which has two slots
which may be associated with one tonal primitive each. The first slot is anchored to the vowel onset and the second to the vowel offset. The phonetic [0] of Cruz (2011) is under the
Alignment Proposal the natural result of a floating 0 tone landing on the first unoccupied slot of the following TC (2), and thus I analyze Cruz's [0] as the TC { 0 _ }. (2)
a.
b.
ʔwa/HL+0/
ʔĩ/_/
{ H L }+0 { _ _ } ʔwa
banana
ʔĩ
RN.3
ʔwa[HL]
ʔĩ[0]
{H L} {0_} ʔwa
ʔĩ
'his/her/its banana'
The Alignment Proposal does not refer to moras since there is no evidence that heavy and light syllables exist in QC, as there is no vowel length contrast and no coda consonants except /ʔ/.
Since monosyllabic words cannot be divided into monomoraic and dimoraic groups, positing two mora for each vowel of QC would achieve little.
Additionally, in one lexical tone language where f0 peaks have been demonstrated to align
to moras (and where the mora can be motivated on segmental grounds), the f0 peaks align to the right edge of each mora, meaning that on dimoraic vowels (which is what all QC vowels will
have to become, much like some analyses of Mandarin (v. Duanmu 1993)), peaks align with the vowel midpoint and offset, not the vowel onset and offset as seen in QC (Zsiga and Nittisaroj 2007).
1.3.2 How is f0 Determined for Unassociated Portions of the TC?
While this analysis jibes with the neutralization and confusability of these tones in
complex and rapid environments, and indicates where the salient pitch peak of each tone category should be, an important question is left unanswered: how does one arrive at the 5
In their treatment of Yoruba contour tones, the end of each tonal complex is necessarily specified, an extraneous requirement for QC. Since no tonal complex in QC stretches across multiple syllables, I do not refer to TC as tonal feet.
observed f0 gestures from the tonal targets in Table 4? Does a speaker plot out only the
specified tonal targets and interpolate between the points? Can the lower f0 values on the
unspecified side of the TC be accounted for by a sagging transition between the tonal targets of adjacent TC?
2. Experiment 1: Straight-line Interpolation between Peaks of Unlike Tone Levels 2.1 Introduction
To prove that the unassociated side of a TC is unspecified, a series of sentences containing
left-aligned and right-aligned tones placed in tone environments of differing levels was recorded with an adult female native speaker of QC. The recording materials can be found in the appendix.
2.2 Methods The sound files were annotated in a Praat textgrid and the f0 values and time
measurements were taken for the peak of the target vowel, the following (for left-aligned tones) or preceding (for right-aligned tones) f0 excursion, and the vowel offset (for left-aligned tones) or onset (for right-aligned tones). 86 tokens were measured. 2.3 Results
Figure 3 shows the f0 values predicted for vowel onsets/offsets by a straight-line
interpolation between f0 excursions across an unassociated portion of the target vowel.
The x-
axis gives the predicted f0 value for the given point and the y-axis indicates the actual f0 of that point. The dashed line indicates where data points would fall should the interpolation accurately predict the f0 values with complete accuracy, points below the line indicate f0 values that were
below predictions. A linear regression gives the formula of the best-fit line as (y= 0.9773x + 9.8913) with a correlation of R2 = 0.69.
Figure 3: Predicted and actual f0 values of unassociated V onsets or offsets 340
Actual f0 Value (Hz)
320 300 280 260 240 220 200 180 180
200
220
240
260
280
Predicted f0 Value (Hz)
300
320
340
2.4 Discussion
The linear regression shows that f0 values at the unassociated portion of a TC (the onset
for right-aligned tones, and the offset for left-aligned tones) can be predicted with a reasonable degree of certainty by an interpolation of the associated target of the TC and the f0 excursion
associated with the nearest target of the preceding or following TC, thus lending credence to the Alignment Proposal's prediction that there is no f0 target on these portions of the TC. 3. Experiment 2: Sagging Transitions between Peak of Like Tone Levels 3.1 Introduction
Even though the f0 values between unlike f0 peaks has been shown to be predicted by a
straight-line interpolation, an inspection of pitch traces (such as Fig (1)) show that sequences of like tones are not domains of static f0. I propose that in such cases, the transition between f0 peaks is not a straight-line interpolation (i.e. a pitch spread) but rather a sagging transition
between peaks (cf. Pierrehumbert (1980)). If such a transition exists, we would expect that the
f0 value of the intervening trough to be a function of the surrounding f0 peaks and the interpeak distance.
3.2 Methods
The data set recorded for experiment 1 was annotated in a Praat textgrid and f0 and time
measurements were made of the peaks of the two like-toned syllables in each sentence and the
intervening trough. The sentences were created to maximize the number of sonorant segments in the two target words and to vary the interpeak distances by having some sentences with late peak target words (i.e. [_H][_H]) or early tones ([H_][H_]), distant peaks ([H_][_H]) and near peaks ([_H][H_]). Due to recording errors (the speaker produced a verb in a different aspect than
intended) and the elimination of tokens whose peaks were judged to have merged6 (thus having no intervening trough), a total of 72 tokens were measured. 3.3 Results
The f0 values of the troughs were found to correlate to the f0 values of the surrounding
peaks. Figures 4 and 5 show the trough values plotted against the magnitude of the preceding peak (Fig. 4) or the following peak (Fig. 5). Linear regressions found that the trough value
could be predicted as a function of the f0 value of the preceding peak (y=0.8272x + 31.538, R2=0.8961) or the following peak (y=0.8741x + 16.366, R2=0.8587). On each graph, the dashed line indicates peak f0 = trough f0.
6
Notably, peak merger has not yet been described in QC, nor any Chatino language.
Figure 4: Peak 1 f0 and Trough f0 (Hz)
Figure 5: Peak 2 (f0) and Trough f0 (Hz)
400
400
350 Trough f0 (Hz)
Trough f0 (Hz)
350
300
250
300
250
200
200 200
250
300
350
400
Peak 1 f0 (Hz)
200
250
300
350
400
Peak 2 f0 (Hz)
3.4 Discussion
As the f0 value of the trough between like-toned peaks can be predicted by the height of
the surrounding f0 peaks, irrespective of the interpeak distance and the phonemic tone level of the surrounding peaks (that is, the same ratio is present between the troughs and peaks of sequences of [M] [H] and [0] tones) suggest that the trough need not be indicated in the
phonemic representation of the tone, and hence these trough would be associated with empty portions of QC TCs.
4. General Discussion and Conclusions
The instrumental data lead us to conclude that the Alignment Proposal indeed captures
essential phonetic facts of QC. In agreement with the experimental data, associated portions of
the TC are indicated by an f0 target associated with either the beginning or the end of the vowel,
unassociated portions of TC between unlike target levels are spelled out by interpolation between associated targets and by a sagging transition between like tone targets. Furthermore, the
Alignment Proposal captures phonemic facts of the language, especially in the way it reifies the natural classes that are observed in tone neutralization and confusion in complex environments and running speech.
Future research is necessary to test the perceptual salience of the Alignment Proposal, as
well as to investigate the power of the representation theory against all the tone sandhi observed in QC. Additionally, the Alignment Proposal may be able to inform reconstructions of protoEastern Chatino or proto-Chatino tonal systems.
References AKINLABI, AKINIYI and MARK LIBERMAN. 2000. Tonal Complexes and Tonal Alignment. NELS 31:1-20.
CAMPBELL, ERIC. 2011. Internal Diversification and Subgrouping of Chatino. University of Texas, Austin. UT Austin. Unpublished ms.
CRUZ, EMILIANA. 2011. Phonology, Tone, and the Functions of Tone in San Juan Quiahije Chatino. Ph.D.
Dissertation. University of Texas at Austin. DUANMU, SAN. 1993. Rime length, stress, and association domains. Journal of East Asian Linguistics 2, 1-44. KAUFMAN, TERRENCE. 2006. Oto-Manguean Languages. In Keith Brown ed. Encyclopedia of Language and
Linguistics, Second Edition. 9:118-124. Oxford: Elsevier.
MCINTOSH, JUSTIN. 2011. Grammatical sketch of Teotepec Chatino. Master's thesis, University of Texas at Austin, Austin, Texas.
PIERREHUMBERT, JANET. 1980. The phonology and phonetics of English intonation. Doctoral dissertation. Distribution: Bloomington, IN:Indiana University Linguistics Club.
SULLIVANT, J. RYAN. 2011. An Investigation into Quiahije Chatino Tone. Ms. University of Texas at Austin. VILLARD, S
PHANIE.
2009. Grammatical sketch of Zacatepec Chatino. Master's thesis, University of Texas at
Austin, Austin, Texas. ZSIGA, ELIZABETH and RATTIMA NITTISAROJ. 2007. Tone features, tone perception, and peak alignment in Thai. Language and Speech. 50(3), 343-383
Appendix
The recording materials are presented here in a four-line interlinearization. The first line
indicates Cruz (2011)'s phonemic categories and the second her phonetic labels for the resulting
words after all tone sandhi rules have been applied and all floating tones have either been docked to the appropriate syllable or else removed. Abbreviations: COMP = Completive Aspect
(essentially a perfective), DEM = demonstrative, HAB = Habitual Aspect, NEG = negative, NOM=
nominalizer,
PROG
= Progressive Aspect, RN= relational noun (an inalienably possessed
noun indicating among other things animate direct objects). 1.
no/_/
na/M0/
lwi/M0/
ʔĩ/_/
no/_/
ʃwe/+H/
DEM
HAB.look.for
L.
RN
NOM
little
no[L] na[M0]
lwi[M0]
ʔĩ[ML]
no[L]
ʃwe[MH]
'At that time, Luis looks for the children.' 2.
no/_/
kʔju/H/
kla/_/
ⁿdja/M/
kna/M/
NOM
man
old
PROG.become
thief
no[L]
kʔju[H]
kla[ML]
ⁿdja[M]
'Over there, the old man is training to become a thief'
ʔ/M0/
kna[M]
ʔ[M0] DEM
ʔ/LM/
ʔ[LM] DEM
3.
sja/M0/
nːa/+H/
sja[M0]
although
ʃka/+H/
nːa[MH]
no/_/
ʃka[MH]
kwa/MH/
no[ML]
kwa[M^]
sja/M0/
nːa/+H/
kni/M/ kwa/H/
ha/L/-ʔne/LM/
ka/_/su/H+0/
ʔĩ/_/
although
PROG.cry
bird
NEG-HAB.do.2S
attention
RN
NOM
DEM
sja[M0]
nːa[MH]
kni[M] kwa[H]
ha[L]-ʔne[LM]
DEM
jna/H/
lo/L/
ke/L/
NEG-COMP.look.for
resin.incense
surface.of
rock
ha[L]-na[H]
jna[H]
lo[L]
no/_/
nːa/MH/
no[L] nːa[M^]
NOM PROG.cry
hi/MH/
ʔĩ/_/
skunk
RN
hi[M^]
ʔĩ[ML]
HAB.stand
ka[ML]su[H]
ha/L/-na/H/
'S/he didn't look for resin on the rock.' 6.
quiet
[LM]
other
'Although that bird is crying, you're not paying it attention.' 5.
ⁿd
[M0]
/LM/
PROG.listen
'Although the other guy is listening, he doesn't do anything.' 4.
ⁿd
/M0/
ʔĩ[0]
ke[MH]
ʔ/M0/
ʔ[M0] DEM
'The skunk is shouting at him/her.' 7.
sja/M0/
nːa/+H/
although
PROG.listen
sja[M0]
nːa[MH]
a/H+0/
kla/_/ ha/L/-ʔne/LM/
ka/_/su/H+0/
ʔĩ/_/
mule
old
attention
RN
a[H]
kla[0] ha[L]-ʔne[LM]
NEG-PROG.do
'Although the mule is listening, it's not paying attention.' 8.
ktʃĩ/L/
ⁿdja/M/
hi/MH/
ʃĩ/LM/
town
PROG.become
skunk
wide
ktʃĩ[L]
ⁿdja[M]
hi[M^]
ʃĩ[LM]
'The skunks in town are getting fat.' 9.
no/_/
na/H/
no[L] na[H]
NOM COMP.look.for
ka[ML]su[H]
a/H+0/
kla/_/
kwa/MH/
ʔĩ/_/
ʃʔna/H/
mule
old
DEM
RN
owner.of
a[H]
'The mule looked for its owner.'
kla[0]
kwa[M^]
ʔĩ[ML]
ʃʔna[H]
ʔĩ[0]
10.
ha/L/-na/H/
ʃka/+H/
a/L/
NEG-COMP.look.for
other
possum
ha[L]-na[H]
ʃka[MH]
kwa/H/
a[L]
kwa[H] DEM
'The other didn't look for that possum.' 11.
sja/M0/
na/M0/
lwi/M0/
ʔĩ/_/ ʃwa/MH/ ha/L/la/L/-ⁿ a/M/
although
HAB.look.for
L.
RN
sja[M0]
na[M0]
lwi[M0]
ʔĩ[ML] ʃwa[M^] ha[L]la[MH]-ⁿ a[M] J.
'Although Luis looks for Juan, he doesn't find him.' 12.
kla/ML/ʔn
NEG-HAB.find
no/_/
ʃwe/+H/
thirty-five
tortilla
COMP.look.for
NOM
little
tʃaʔ/MH/no/_/
nːa/MH/
kni/M/
ʃĩ/LM/
ʔĩ/_/
no/_/
ʃwe/+H/
because
PROG.cry
bird
wide
RN
NOM
little
no/_/
ʃwe/+H/
ʔĩ/_/
little
RN
kla[ML]ʔn
[ML]
jha[0]
na[M0]
no[ML]
ʃwe[MH]
tʃaʔ[M^]no[L]
nːa[M^]
kni[M]
ʃĩ[LM]
ʔĩ[ML] no[L] ʃwe[MH]
tʃaʔ/MH/no/_/
lo/L/
nːa/L/
nːa/+H/
because
surface.of
griddle
PROG.listen NOM
tʃaʔ[M^]no[L]
lo[L]
nːa[MH]
nːa[MH]
no[L] ʃwe[MH]
'Because the children are listening to him/her on the griddle.' 15.
sja/M0/
nːa/MH/
hi/M/
ha/L/la/L/-ʔne/LM/
ka/_/su/H+0/
ʔĩ/_/
although
PROG.cry
skunk
NEG-PROG.do.2S
attention
RN
sja[M0]
nːa[M^]
hi[M]
ha[L]la[MH]-ʔne[LM] ka[ML]su[H]
ʔĩ[0]
'Although the skunk is crying, you're not paying it attention.' 16.
RN
na/M0/
'Because the fat bird is crying to the children.' 14.
ʔĩ[H]
jha/_/
/ML+0/
'The children looked for thirty-five tortillas.' 13.
ʔĩ/_/
no/_/
nːa/+H/
jna/H/
lo/L/
NOM
PROG.listen
resin.incense
surface.of
no[L]
nːa[MH]
jna[H]
'At that time, the resin incense is listening.'
lo[L]
ʔ/LM/
ʔ[LM] DEM
ʔĩ[L]
