Phonetic patterns in Oto-Manguean tonal systems

Christian T. DiCanio [email protected]

Haskins Laboratories http://linguistics.berkeley.edu/∼dicanio

5/14/14

Christian DiCanio (((Haskins))) Oto-Manguean 5/14/14 1 Introduction Typology and tonal systems

How useful is a typological perspective for the study of tonal phonetics?

1 Structural diversity is abundant. Structural differences among languages contribute to phonetic variation in tone production/perception, even across well-known languages.

2 The phonetic timing of tones differs dramatically. There is substantial cross-linguistic variation in how tones are coordinated in larger utterances/units.

3 Models of speech production should be inclusive with respect to such cross-linguistic variability.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 2 Introduction Typology and tonal systems

How useful is a typological perspective for the study of tonal phonetics?

1 Structural diversity is abundant. Structural differences among languages contribute to phonetic variation in tone production/perception, even across well-known languages.

2 The phonetic timing of tones differs dramatically. There is substantial cross-linguistic variation in how tones are coordinated in larger utterances/units.

3 Models of speech production should be inclusive with respect to such cross-linguistic variability.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 2 Introduction Typology and tonal systems

How useful is a typological perspective for the study of tonal phonetics?

1 Structural diversity is abundant. Structural differences among languages contribute to phonetic variation in tone production/perception, even across well-known languages.

2 The phonetic timing of tones differs dramatically. There is substantial cross-linguistic variation in how tones are coordinated in larger utterances/units.

3 Models of speech production should be inclusive with respect to such cross-linguistic variability.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 2 Introduction Typology and tonal systems

How useful is a typological perspective for the study of tonal phonetics?

1 Structural diversity is abundant. Structural differences among languages contribute to phonetic variation in tone production/perception, even across well-known languages.

2 The phonetic timing of tones differs dramatically. There is substantial cross-linguistic variation in how tones are coordinated in larger utterances/units.

3 Models of speech production should be inclusive with respect to such cross-linguistic variability.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 2 Introduction Typology and tonal systems

How useful is a typological perspective for the study of tonal phonetics?

1 Structural diversity is abundant. Structural differences among languages contribute to phonetic variation in tone production/perception, even across well-known languages.

2 The phonetic timing of tones differs dramatically. There is substantial cross-linguistic variation in how tones are coordinated in larger utterances/units.

3 Models of speech production should be inclusive with respect to such cross-linguistic variability.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 2 Introduction E. Zsiga, R. Nitisaroj 347

Structural differences? Mandarinfalling slightly and vs. remaining Thai as low or lower than the mid tone for the first half of the syllable, then rising steeply in the second half. The low tone falls steadily, reaching the bottom of the pitch range at the end of the syllable. The falling tone is realized as a rise-fall contour, and the rising tone as a fall-rise contour. Contextual tonal ￿ ariations 67 &## 160 Figure 1 Tone 1 Contour shapes of Thai tones Tone 2 %$" ! in citation form. Representative examples from one speaker 140 % Tone 3 %"#

Tone 4 120 (Hz) 0

f %%" " &'()*+%, 100 %## #

!$" 80 $

0 25 50 75 100 !"# # #'! #'% #'& #'( #'" Normalized time (%) )*+,-.)

Figure 2 . Mean f 0 contours (averaged over speakers and tokens ; n ￿ 48) of four Mandarin tones in the monosyllable / m a / produced in isolation . The time is normalized , with all tones plotted with their average duration Creating these complex contours from simple H and L associated to the syllable proportional to the average duration of Tone 3 . is certainly possible. Such phonetic mapping rules would be complex, however. A What does a falling tone look like? Whatsingle accounts H associated for to the asyllable delayed would havefall to be mapped in into a level-rising scooped contour, while an H linked as part of a falling tone would correspond to a quick Thai? 3 . Results rise to the top of the pitch range. The complexity of attempting to map H and L 3 . 1 . Tones produced in isolation auto segments onto the actual contours of the tones of Thai leads Abramson (1979) to reject a compositional analysis of the contours. He argues (p.7) that the data “lend To establish the canonical forms of the four Mandarin tones produced byno the phonetic plausibility” to arguments for the specification of rising and falling tones (Figures speakers in this from study , tones Xu produced (1997); in isolation Zsiga were analyzed and first Nitisaroj . Fig . 2 showsas sequences (2007)) of H and L autosegments. the average f 0 contour of the syllable / m a / in the four tones , obtained by averaging over all tokens produced by all eight speakers (48 utterances for eachFor phonetic support of the argument one would expect to be able to devise a tone) and plotted as functions of average relative time of each segment , with theformula by which the dynamic tones were obviously to be derived from the shape average duration of Tone 3 as the reference duration (100%) . The cross symbol ‘‘ ￿ ’’of the static tones. Even the citation forms, let alone the F0 curves of running indicates boundaries between m and a . Tone 1 starts with a high f valuespeech, provide no acoustic basis for such a claim. It seems psychologically far Christian DiCanio (((Haskins))) / / / / Oto-Manguean 0 tone 5/14/14 3 (near 130 Hz) and stays around that level throughout the syllable . Tone 2 starts withmore reasonable to suppose that the speaker of Thai stores a suitable tonal shape as part of his internal representation of each monosyllabic lexical item. (p.7) a low f 0 (near 110 Hz) , then falls slightly before rising (starting at 20% into the vowel) throughout the remainder of the syllable . Tone 3 starts with an f 0 valueMorén and Zsiga argue, however, that a mora-based representation as in (2) slightly lower than the onset of Tone 2 , falls to the lowest f 0 of all the four tonesdoes allow a straightforward mapping from autosegments to the acoustics of Thai (about 90 Hz) right at the vowel midpoint , then rises sharply to the end oftones. the During moras with no phonological tone, pitch falls gradually to or within syllable . Tone 4 starts with the highest f 0 value of the four tones (140 Hz) , continuesthe mid range. Phonologically-specified moras reach a high or low pitch inflection to rise before reaching the maximum about one fifth of the way into the vowel , at then their right edge. Thus mid tones, with no phonological specification, remain fairly falls sharply to the end of the syllable . level throughout their duration. High tones are specified only on the second mora, In terms of duration , Tone 4 is the shortest tone (214 ms on average) , Toneand 3 isthus remain in the mid range during the first half of the syllable, begin to turn the longest (349 ms) , while Tones 1 and 2 have intermediate durations , with Toneupward 2 at syllable midpoint, and reach their high point at the rightmost edge of the (273 ms) longer than Tone 1 (247 ms) . These duration patterns match those found in previous studies (e . g ., Lin , 1988) . Language and Speech Downloaded from http://las.sagepub.com at Bibliotheques de l'Universite Lumiere Lyon 2 on April 20, 2009 The f 0 patterns of tones produced in isolation reflect relatively directly the canonical forms of the tones . In multisyllabic utterances , the canonical forms will be distorted by various factors , including the adjacent onset and of fset values of the neighboring tones . As suggested earlier and demonstrated by Xu (1994 a ) , it is High-level tone (55) Mid-level tone (33) Low-rising tone (24) Low-falling tone (21) High-falling tone (51) 330 F2 F2 F2 F2 F2 300 Introduction 270 Coordination differences? Mandarin vs. Taiwanese

(Hz) 240 0 f 210 Mandarin falling tones (left, (Xu, 1994)) undergo greater coarticulation 180 than Taiwanese falling tones (right, (Peng, 1997)). (a) (b) (c) (d) (e) 150 Riang Fa • 330 180- 180 F2 F2 F2 F2 F2 x _ / x / _ / x 300/ x Compatible _ _ Context •, / / X_ _ / X X 270! 160

240 (Hz) 0 /_ / / / _ x f x 210x x Conflicting _ _ Context _/ ! /! X _X X 180X } (f) (g) (h) (i) (j) 150 120 120 –20 30 80 130 180 230 280 330 –20 30 80 130 180 230 280 330 –20 30 80 130 180 230 280 330 –20 30 80 130 180 230 280 330 –20 30 80 130 180 230 280 330 Duration (ms) Duration (ms) Duration (ms) Duration Duration (ms) Duration (ms)

Panels (a) – (e) Panels (f ) – ( j) FIG. 2. Tone patternsof trisy!labi½words and phrasesused for F0 mea- — ￿ — I phrase-initial — ￿ — 55 surements. Compelhie, Faster — ￿ — M phrase-medial Christian DiCanio — ￿ — (((Haskins))) 33 Oto-Manguean tone 5/14/14 4 – – ￿ – – F phrase-final -- Compatible, – – ￿ – –Slowe• 21 - - - ￿ - - - U utterance-final •-• Conlhcfng.Faster C. Recording Figure 5 . Subject F2 : Mean duration and f 0 of each target tone followed by dif ferent tones (panels f – j) in dif ferent prosodic positions (panels a – e) ...... Conf•ct'ng, Slower For eachof the 16 patternslisted in Fig. 2, two words or phraseswere found. They were repeatedfive timesand FIG. 3. Mean F0 slopes,mean F0 heightand mean durationfor the printedin Chinesein randomorder. The speakersrecorded risingtone and fallingtone in compatibleand conflictingcontexts. For the wholelist oncewith eachword spokenin isolation,and eachline, the midpointrepresents mean F0; the slopeof the line repre- a secondtime with eachword spokenin a carrier sentence, sentsthe meanslope; and the end of the line representsmean duration. w6 gabsunT•hl ze•rno hufshi?"Let me tell you what--is all about." It was expectedthat the presenceof where a contour changesfrom rising to falling or vice the carrier sentencewould increasethe speakingrate, thus versa,as importantmeasurements (G•rding and Zhang, increasingthe magnitudeof coarticulatorytonal variation. 1986; G•trding, 1987; Shen, 1990; Shen and Lin, 1991}. Each speakerproduced 320 utterances. However, becausethe durationsof the secondsyllables in During the recording,the overallspeaking rate of the the trisyllabicwords examinedin this study were rather speakerswas controlledby regularlyoccurring pure tone short (around 100 ms), there were sometimesonly a few pulsesplayed through a loudspeakerby a separatecassette vocal pulsesin the vowel,and the contourwas often very player. The speakerswere requiredto produceeach sen- closeto a straightline. So it was often difficultto decide tencebetween two pulsesoccurring at an interval of 3 s. wherethe turningpoint is. For this reason,turning points The recordingwas made with a condensermicrophone and werenot measuredin thisstudy, and the useof slopeas the an Otari tape recorderat a tape speedof 7.5 in./s. major measurementfor tonal contoursignored any possi- ble curvaturewithin a contour.Also becausethis study is concernedmainly with comparingtonal variationsunder D. F0 analysis differentconditions rather than examiningthe exact con- All the utteranceswere digitizedby the HaskinsPCM tour shapes,slope measurement should be sufficient. system(Whalen et al., 1990) at a 20-kHz samplingrate. The extraction of the F0 values was done manually to E. Results and discussion achievehigh accuracy.A label was placedon each vocal pulsein the waveformof eachtarget syllabledisplayed on For each word or phrasespoken by each speakerat the computerscreen. Then, a computerprogram was used one of the speakingrates (with or without eartier), the to measurethe time intervalsbetween neighboring ]abels meanslope, mean F0, and meanduration were obtained by and convertthe intervalsinto frequencyvalues. For sylla- averagingover the individual values of those measure- bles with initial •nasals, the lateral, or unaspi- ments for all five tokens of the word. Figure 3 gives a rated stops, F0 measurementstarted from the point of schematicrepresentation of the mean values of all three release,i.e., wherea nasalor lateralsegment ends and the measurementsby targettone and tonalcontext for all con- vocalicsegment starts. For syllableswith initial aspirated ditionsand speakers.For eachline in the figure,the mid- stopsor affricatcs,F0 measurementstarted from the onset point representsmean F0; the slopeof the line represents of voicein the syllable.For all the target syllables,the F0 mean slope;and the end of the line indicatesmean dura- measurementended right beforethe closureor constriction tion. Severalthings are immediatelyapparent from this for the initial consonantof the following syllable. figure.( 1) For the risingtone, the slopeis positivein com- For each token, three raw measurements were ob- patiblecontext but negativein conflictingcontext. [In fact, tained for the secondsyllable, namely, duration, mean F0, 11 out of the 16 words in this condition (rising tone in and slopeof the F0 contour.Duration measuresthe time in conflictingcontext} had negativemean slopes].For the millisecondsbetween the first marked vocal pulse and the falling tones, the slope is steeperin compatiblecontext last. Mean F0 is the averageof the F0 valuesfor all the than in conflictingcontext. And, althoughthe overallmean vocal pulses.Slope is the coefficientof a simplelinear re- slope remainednegative even in the conflictingcontext, gressionline for eachtoken, with time of the vocalpulses three out of the 16 words in the conflictingcontext had as independentvariable. positivemean slopes. (2) In compatiblecontext, the slopes Previous tone studies often used the location and F0 are steeperat a fasterspeaking rate (with carrier) than at valueof the turningpoint in a tone contour,i.e., the point a slower speakingrate (without carrier) for both tones;

2242 J. Acoust.Sec. Am., VoL 95, No. 4, April 1994 Yi Xu: Productionand perceptionof coarticulatedtones 2242 Introduction

A typological perspective will reveal the extent to which both structural and language-specific differences contribute to phonetic patterns related to tone.

Oto- possess a unique collection of structural properties and phonetic patterns which challenge some of the established ideas within the tonal phonetics literature.

1 Strong evidence for the mora as the TBU and the unit of planning, as opposed to the syllable (Prom-on et al., 2009; Xu and Prom-on, 2014; Zhang, 2004).

2 Active processes of dissimilation and range expansion in coarticulation, as opposed to assimilation/reduction (Xu, 1994; Gandour et al., 1994, 1999; Peng, 1997).

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 5 Introduction

A typological perspective will reveal the extent to which both structural and language-specific differences contribute to phonetic patterns related to tone.

Oto-Manguean languages possess a unique collection of structural properties and phonetic patterns which challenge some of the established ideas within the tonal phonetics literature.

1 Strong evidence for the mora as the TBU and the unit of planning, as opposed to the syllable (Prom-on et al., 2009; Xu and Prom-on, 2014; Zhang, 2004).

2 Active processes of dissimilation and range expansion in coarticulation, as opposed to assimilation/reduction (Xu, 1994; Gandour et al., 1994, 1999; Peng, 1997).

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 5 Introduction Roadmap

1 Properties of the Oto-Manguean stock

2 Tonal domains and alignment

3 Tonal coarticulation

4 Discussion

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 6 Oto-Manguean stock Language families in

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 7 Oto-Manguean stock Oto-Manguean languages

With 177 languages, Oto-Manguean is the largest language family in the Americas (and 9th largest in the world).

A majority of these languages are spoken in the state of . In fact, 157 of the 285 languages spoken in Mexico are found in Oaxaca.

Extensive diversity within language family largely correlates with biological diversity in the areas where it is spoken. Oaxaca is the most biologically diverse state in Mexico with the greatest number of endemic vascular plants (de Ávila, 2010).

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 8 Oto-Manguean stock

!"#$%& !"#$%& '()*+,-. /01*+2(- /01*+2 3,04#+ 5#02(*+2 6&7&8&2(- /($(*+2 912(*+2 5:&2:&8*+2 6&7&8&2( ;(7&*+2(- ;(7&*+2 5:(*0-& <*&=/(-%#+(-. <*&=6("+(- 5:02:0"+2 6("+ /(*8(*$0-2( <*&"> /($(:#( ?+)*+,-. 5:0-(-*+2(- 5:0-(-*+2 @#A*0(A(=38(7B @#A*0(A( 38(7(-+2 5:&,&*+%(- 5:0(7(-+2 5:&,&*+%(

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 9 Oto-Manguean stock Tone in Oto-Manguean languages

All are tonal and many have very large tonal inventories. At least three tones are reconstructed at the earliest levels (Kaufman, 1990; Rensch, 1976).

Laryngeal/glottal features which are often orthogonal to tone (Silverman, 1997).

Complex onsets are possible, but most languages lack codas. Most languages have polysyllabic words.

Complex morphology on verbs and with personal clitics which frequently involves tone (Campbell et al., 1986; Palancar, 2009; Suárez, 1983) and classic processes of tone sandhi (Goldsmith, 1990; Pike, 1948).

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 10 Oto-Manguean stock Complex tonal systems

How many tones occur in Oto-Manguean languages?

Language Tones Northern Pame 2 (Berthiaume, 2004) Mazahua 4 (Knapp Ring, 2008) Tlacoatzintepec Chinantec 7 (Thalin, 1980) Itunyoso 9 (DiCanio, 2008) Yoloxóchitl 10 (DiCanio et al., 2012a) San Juan Quiahije Chatino 11 (Cruz, 2011) Chiquihuitlan Mazatec 17 (Jamieson, 1977) Quiotepec Chinantec 19+ (Castillo Martínez, 2011)

But how do you count? Is the TBU the stem? the syllable? the mora?

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 11 Oto-Manguean stock Quetzalapa Chinantec

Five tone levels with contours (rising tones excluded). Words courtesy of Isabel Alhondra.

Tone Word Gloss 55 tsoU ‘his/her fault’ 44 tsoU ‘illness’ 33 tsoU ‘he/she goes’ 22 tsoU ‘straight’ 21 tsoU ‘sin’ 32 tsoU ‘male’ 42 tsoU ‘people’

What is the TBU here though? Are there only 5 levels (1/mora)?

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 12 Moraic alignment in Mixtec Variation in tonal alignment

How we count tones is tied to the phonological domains for tone. What evidence is there for such domains in speech production in Oto-Manguean languages? (Phonology and the phonetics of alignment)

Intonational pitch accents are anchored to segmental targets/onsets (Atterer and Ladd, 2004; Ladd et al., 1999; Ladd, 2004).

Lexical tones are aligned to syllables (Gao, 2008, 2009; Prom-on et al., 2009; Xu, 1998; Xu and Prom-on, 2014).

Lexical tones are aligned to moras (Myers, 2003; Morén and Zsiga, 2006).

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 13 Moraic alignment in Mixtec Syllables or moras?

Similar alignment across CVN and CV syllables at different speech rates in Mandarin. Tonal contrasts are aligned to syllables (Xu, 1998).

Contour tone licensing is insensitive to moraic structure, but sensitive to rime sonority (Zhang, 2004). Contour tones surface on syllables with longer duration of voicing and even are sensitive to polysyllabic shortening (Lehiste, 1970; Turk and Shattuck-Hufnagel, 2000).

Earlier F0 maxima observed for H and HL tones in Kinyarwanda than for the LH tone, suggesting moraic alignment (Myers, 2003).

The inflection points of Thai tonal contours align at the right edge of moras. Trajectories only begin in the second mora (Morén and Zsiga, 2006; Zsiga and Nitisaroj, 2007).

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 14 Moraic alignment in Mixtec Case study: Yoloxóchitl Mixtec

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 15 Moraic alignment in Mixtec Tonal phonology

Like other Mixtecan languages, all roots are minimally composed of bimoraic couplets, consisting of either monosyllabic stems with long vowels (CVV) or disyllabic stems with shorter vowels (CVCV) (Castillo García, 2007).

Five possible tones on the initial mora: 1, 3, 4, 13, 14

Nine possible tones on the final mora: 1, 2, 3, 4, 13, 14, 24, 32, 42

Pattern consistent across word types: /Bi3ta42/‘soft’ vs. /˜n˜u3˜u42/‘night’ /˜n˜u3˜u2/‘town’ vs. /˜n˜uP3˜u2/‘fire’ /nu14u3 /‘face’ vs. /Sa14tu3/‘soft corn tortilla’

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 16 Moraic alignment in Mixtec

Table: Tone in YM (4 = high, 1 = low) Level nda1a1 Sa3a3 nda4a4 ‘flat’‘fast’‘black’ Falling nda3a2 nda4a2 ‘sloping’‘where’ Rising ta1a3 ndo1o4 nde3e4 nda13a3 tu13u4 ‘man’‘sugarcane’‘strong’‘went up’‘stripped’

Rise+Fall kwe13e2 Sa14a3 ndi1i42 ñu3u42 ‘linger’‘new’‘pink’‘night’ High+Rise nde4e13 kwi4i14 ka4a24 ‘they enter’‘is peeling’‘slips’ Rise+Rise ndo14o13 kwi14i14 ka14a24 ‘to not stay’‘is not peeling’‘does not slip’

If the syllable is the unit of tone planning, how many distinct types?

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 17 Moraic alignment in Mixtec Alignment study

“Complex” contours with initial rises should show earlier alignment than simple rises.

If tone is aligned to moras in Mixtec, alignment of contour tones should be similar between monosyllabic and disyllabic words, as both are bimoraic.

If tone is aligned to syllables, then alignment of contour tones in monosyllables need not correspond to the alignment in disyllables.

Examined F0 alignment in large elicited corpus of 261 words x 6 repetitions x 10 speakers.

LMER with word size, normalized time, and tone as DVs, speaker as a random effect.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 18 Moraic alignment in Mixtec Expectations for alignment – parity across word types

0.484939271 0.834889968 220 200

150 ) z H (

h c t i 100 P 80 v1

v1 v2

0.4849 0.8349 Time (s)

Test: to what extent do F0 contours differ across word types?

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 19 Moraic alignment in Mixtec Results

There is no general effect of word size. However, there was a significant tone x word size interaction (tone /4/)

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 20 Moraic alignment in Mixtec Falling tones are similar

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 21 Moraic alignment in Mixtec Rising tones are similar.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 22 Moraic alignment in Mixtec Complex vs. simple rises

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 23 Moraic alignment in Mixtec

Late target attainment of tone /1/ in /1.4/, but early rise of tone /13/ in /13.4/.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 24 Moraic alignment in Mixtec Double rises

Complete rise attained in first mora of vowel in monosyllables.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 25 Moraic alignment in Mixtec Discussion: Alignment

No general effect of word size on alignment – not predicted if the syllable is the unit of tone planning.

Interactions between word size and tone with respect to F0 height (not time), for melodies /1.3, 1.4, 4.14, 4.2, 4.4/.

Strong evidence for alignment to the mora, even within a monosyllabic long vowel.

Strong similarity across word sizes also suggests phonetic alignment to the mora.

Counter Zhang’s (2004) argument that tonal licensing is not constrained by moraic structure. Alignment was not considered in his proposal.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 26 Moraic alignment in Mixtec Conclusions - Alignment

Moraic structure not simply assumed to account for the distributional differences in Mixtec, but it is supported by phonetic data examining alignment.

Typological considerations into the size of tonal inventories need to look carefully at the nature of the tone-bearing unit in particular languages, lest we mischaracterize apparent (or hidden) complexity.

We just didn’t know that languages could do this!

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 27 Moraic alignment in Mixtec Conclusions - Alignment

Moraic structure not simply assumed to account for the distributional differences in Mixtec, but it is supported by phonetic data examining alignment.

Typological considerations into the size of tonal inventories need to look carefully at the nature of the tone-bearing unit in particular languages, lest we mischaracterize apparent (or hidden) complexity.

We just didn’t know that languages could do this!

YM has a large inventory of tones, but it’s not as many as you might assume.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 27 Triqui tonal coarticulation Tone production and variability

The relative timing of tone varies by context and language.

How are tones coordinated with one another in Oto-Manguean languages?

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 28 Triqui tonal coarticulation What constrains the production of a sequence of lexical tones?

1 Tonal context: the production of a tone is sensitive to the F0 height and slope of adjacent tonal targets.

2 Speech rate: rate impinges on the temporal demands for producing certain F0 shapes.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 29 Triqui tonal coarticulation What constrains the production of a sequence of lexical tones?

1 Tonal context: the production of a tone is sensitive to the F0 height and slope of adjacent tonal targets.

2 Speech rate: rate impinges on the temporal demands for producing certain F0 shapes.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 29 Triqui tonal coarticulation Tonal context and compatibility

Phonetic contexts which cause more abrupt F0 transitions between adjacent syllables are more likely to perturb F0 than contexts causing gradual transitions (Xu, 1994).

Conflicting: the tone offset does not match the following onset, e.g. Rise + Low.

Compatible: the tone offset matches the following tone onset, e.g. Fall + Rise.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 30 Triqui tonal coarticulation

Greater coarticulatory effects occur in conflicting contexts than in compatible contexts in Mandarin Chinese (Xu, 1994).

Riang Fa • 180- 180 x _ / x / _ / x / x Compatible _ _ Context •, / / X_ _ / X X ! 160

/_ / / / _ x x x x Conflicting _ _ Context _/ ! /! X _X X X } 120 120

Duration

FIG. 2. Tone patternsof trisy!labi½words and phrasesused for F0 mea- surements. Compelhie, Faster -- Compatible,Slowe•

•-• Conlhcfng.Faster C. Recording ...... Conf•ct'ng, Slower For eachof the 16 patternslisted in Fig. 2, two words or phraseswere found. They were repeatedfive timesand Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 31 FIG. 3. Mean F0 slopes,mean F0 heightand mean durationfor the printedin Chinesein randomorder. The speakersrecorded risingtone and fallingtone in compatibleand conflictingcontexts. For the wholelist oncewith eachword spokenin isolation,and eachline, the midpointrepresents mean F0; the slopeof the line repre- a secondtime with eachword spokenin a carrier sentence, sentsthe meanslope; and the end of the line representsmean duration. w6 gabsunT•hl ze•rno hufshi?"Let me tell you what--is all about." It was expectedthat the presenceof where a contour changesfrom rising to falling or vice the carrier sentencewould increasethe speakingrate, thus versa,as importantmeasurements (G•rding and Zhang, increasingthe magnitudeof coarticulatorytonal variation. 1986; G•trding, 1987; Shen, 1990; Shen and Lin, 1991}. Each speakerproduced 320 utterances. However, becausethe durationsof the secondsyllables in During the recording,the overallspeaking rate of the the trisyllabicwords examinedin this study were rather speakerswas controlledby regularlyoccurring pure tone short (around 100 ms), there were sometimesonly a few pulsesplayed through a loudspeakerby a separatecassette vocal pulsesin the vowel,and the contourwas often very player. The speakerswere requiredto produceeach sen- closeto a straightline. So it was often difficultto decide tencebetween two pulsesoccurring at an interval of 3 s. wherethe turningpoint is. For this reason,turning points The recordingwas made with a condensermicrophone and werenot measuredin thisstudy, and the useof slopeas the an Otari tape recorderat a tape speedof 7.5 in./s. major measurementfor tonal contoursignored any possi- ble curvaturewithin a contour.Also becausethis study is concernedmainly with comparingtonal variationsunder D. F0 analysis differentconditions rather than examiningthe exact con- All the utteranceswere digitizedby the HaskinsPCM tour shapes,slope measurement should be sufficient. system(Whalen et al., 1990) at a 20-kHz samplingrate. The extraction of the F0 values was done manually to E. Results and discussion achievehigh accuracy.A label was placedon each vocal pulsein the waveformof eachtarget syllabledisplayed on For each word or phrasespoken by each speakerat the computerscreen. Then, a computerprogram was used one of the speakingrates (with or without eartier), the to measurethe time intervalsbetween neighboring ]abels meanslope, mean F0, and meanduration were obtained by and convertthe intervalsinto frequencyvalues. For sylla- averagingover the individual values of those measure- bles with initial fricatives•nasals, the lateral, or unaspi- ments for all five tokens of the word. Figure 3 gives a rated stops, F0 measurementstarted from the point of schematicrepresentation of the mean values of all three release,i.e., wherea nasalor lateralsegment ends and the measurementsby targettone and tonalcontext for all con- vocalicsegment starts. For syllableswith initial aspirated ditionsand speakers.For eachline in the figure,the mid- stopsor affricatcs,F0 measurementstarted from the onset point representsmean F0; the slopeof the line represents of voicein the syllable.For all the target syllables,the F0 mean slope;and the end of the line indicatesmean dura- measurementended right beforethe closureor constriction tion. Severalthings are immediatelyapparent from this for the initial consonantof the following syllable. figure.( 1) For the risingtone, the slopeis positivein com- For each token, three raw measurements were ob- patiblecontext but negativein conflictingcontext. [In fact, tained for the secondsyllable, namely, duration, mean F0, 11 out of the 16 words in this condition (rising tone in and slopeof the F0 contour.Duration measuresthe time in conflictingcontext} had negativemean slopes].For the millisecondsbetween the first marked vocal pulse and the falling tones, the slope is steeperin compatiblecontext last. Mean F0 is the averageof the F0 valuesfor all the than in conflictingcontext. And, althoughthe overallmean vocal pulses.Slope is the coefficientof a simplelinear re- slope remainednegative even in the conflictingcontext, gressionline for eachtoken, with time of the vocalpulses three out of the 16 words in the conflictingcontext had as independentvariable. positivemean slopes. (2) In compatiblecontext, the slopes Previous tone studies often used the location and F0 are steeperat a fasterspeaking rate (with carrier) than at valueof the turningpoint in a tone contour,i.e., the point a slower speakingrate (without carrier) for both tones;

2242 J. Acoust.Sec. Am., VoL 95, No. 4, April 1994 Yi Xu: Productionand perceptionof coarticulatedtones 2242 Triqui tonal coarticulation Tonal coarticulation and speech rate

While all languages have some anticipatory and carryover effects, the latter is typically stronger (in Vietnamese, Thai, Mandarin) (Brunelle, 2009; Gandour et al., 1994; Xu, 1997).

Generally, tonal coarticulation is assimilatory in nature, but anticipatory coarticulation is sometimes dissimilatory. High tones raise before low tones in Hokkien (Chang and Hsieh, 2012), Taiwanese (Peng, 1997), and Mandarin (Xu, 1997; Tilsen, 2013).

As rate increases, fewer F0 targets are successfully reached and contours flatten, as in French (Fougeron and Jun, 1998), Croatian, and English (Bradlow et al., 2003; Smiljanić and Bradlow, 2005).

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 32 Triqui tonal coarticulation Case study: Itunyoso Triqui

Spoken in the town of San Martín Itunyoso, in Oaxaca, Mexico, it is one of three Triqui variants.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 33 Triqui tonal coarticulation Tone in Itunyoso Triqui

Nine lexical tones contrast in word-final syllables, but only level tones occur in non-final syllables. Syllable structure is open with the exception of two possible glottal codas (DiCanio, 2008, 2012a, 2014).

There are morphological tone changes, but no tone sandhi across words (DiCanio, submitted).

Tone IPA Gloss Tone IPA Gloss 4 B:e4 ‘hair’ 43 li43 ‘small’ 3 n:e3 ‘plough’ 32 n:e32 ‘water’ 2 n:e2 ‘to lie (tr.)’ 31 n:e31 ‘meat’ 1 n:e1 ‘naked’ 45 joH45 ‘my forehead’ 13 jo13 ‘light, quick’

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 34 Triqui tonal coarticulation Tones in open syllables, from (DiCanio, 2012b)

Tone 220 T1 T2 T3 T4 T43 T32 T31 200 180 PItch (Hz) 160 140

0.00 0.05 0.10 0.15 0.20

Time (seconds)

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 35 Triqui tonal coarticulation Method: stimuli

How do tonal context and rate influence tone production in Triqui?

Four tones were chosen and embedded in four different tonal contexts in natural sentences of 3 words.

Example: n˜ı3P˜ı3 s˜ıh45 ja3kweh3 ‘The man knows Oaxaca.’

The medial word was always monosyllabic while the adjacent words were always disyllabic.

The resulting sentences were natural carrier sentences in the language of the form: Verb + Subject + Modifier (adj, object), c.f. Scholz (2012).

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 36 Triqui tonal coarticulation Method: tonal contexts and rate

Adjacent tone (L/R) Medial tone /2.2/ /45/ /3.3/ /4/ /3.32/ /43/ /3.45/ /2/

Each of the 16 sentences were repeated five times by 8 speakers (4 male, 4 female) in two rate conditions (normal, fast).

Normal rate: 3.45 45 3.45, [atS˜ıh45 s˜ıH45 tSa3kah45] ‘The man is asking for a pig.’ Fast rate: 3.45 45 3.45, [atS˜ıh45 s˜ıH45 tSa3kah45]

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 37 Triqui tonal coarticulation Measures

Time-normalized F0 data for the three vowels extracted using Voicesauce (Shue et al., 2009), at 20 points.

F0 data examined with several linear mixed effects model fit for each tone in each position (Left, Medial, Right) with 4 factors (Offset.difference X Time X Duration X Rate). Speaker was treated as a random effect with random intercepts and slope set for the effect of rate.

The 20 time points were treated as continuous and recentered.

Conservative measure of distance of z-score log F0 from speaker’s average for a given tone (a measure of variation).

All results discussed here are significant at p < .01 via model comparisons using a X2 test with analysis of variance.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 38 Triqui tonal coarticulation Context effect Medial target: level tones

Dissimilatory lowering of T2 and raising of T4 t2 t4

2

1

Adjacent tones t2

0 t3 t32 t45 z-score normalized log F0 log normalized z-score -1

-2

1 2 3 4 5 6 7 8 9 1011121314151617181920 1 2 3 4 5 6 7 8 9 1011121314151617181920 Time (normalized)

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 39 Triqui tonal coarticulation Context effect Medial target: contour tones

Dissimilatory raising of T45 t43 t45

2

1

Adjacent tones t2

0 t3 t32 t45 z-score normalized log F0 log normalized z-score -1

-2

1 2 3 4 5 6 7 8 9 1011121314151617181920 1 2 3 4 5 6 7 8 9 1011121314151617181920 Time (normalized)

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 40 Triqui tonal coarticulation Context effect Summary - medial target

Significant effects of the adjacent tones on the medial tone’s F0 trajectory, most notably for an adjacent /45/ and /32/.

Not assimilatory, but dissimilatory.

Tone /2/ lowers between contour tones /(3).45/.

Tone /45/ (and tone /43/ to a lesser extent) raises between tones /2.2/.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 41 Triqui tonal coarticulation Rate effect Results II: Effect of rate on tone production

Average Normal:Fast speech rate = 1:1.26

Speech rate by speaker 7

6

Rate condition 5 fast normal 4

3 Speech rate (syllables/sec) Speech

BCM CCM ERM JLG MCM RDM RMS WCM Speaker

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 42 Triqui tonal coarticulation Rate effect

F0 range expansion

Medial tones at different speech rates

2.5 T2 T4 T43 T45 Normal Fast 2.0 1.5 1.0 0.5 z-score normalized F0 normalized z-score 0.0 -0.5 -1.0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Time (normalized)

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 43 Triqui tonal coarticulation Discussion Discussion

Tones remain distinct across contexts. This is different from Mandarin, where tones change drastically (Xu, 1994), but similar to findings in Thai (Gandour et al., 1999). Triqui and Thai both lack tone sandhi.

More coarticulation occured when the following tone was a contour than when it was level.

More dissimilatory effects were found in anticipatory contexts (medial), which is in-line with work on Taiwanese (Peng, 1997), Mandarin (Xu, 1998), Malaysian Hokkien (Chang and Hsieh, 2012), and Tianjin Chinese (Zhang and Liu, 2011).

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 44 Triqui tonal coarticulation Discussion Contrast preserving

Expanding the pitch range and processes of dissimilation during in fast speech may aid the listener in the perception of tonal contrasts.

Manuel’s output constraint: “languages generally tend to tolerate less contextually induced changes in acoustic phonetic output if they are likely to lead to confusion of contrastive phones.” (Manuel, 1990)

Languages for which increased variability in F0 does not result in decreased lexical identification do not undergo range contraction at a faster speech rate, as it is detrimental for perception.

But if so, why doesn’t Mandarin do this?

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 45 Triqui tonal coarticulation Discussion Contrast preserving

Expanding the pitch range and processes of dissimilation during in fast speech may aid the listener in the perception of tonal contrasts.

Manuel’s output constraint: “languages generally tend to tolerate less contextually induced changes in acoustic phonetic output if they are likely to lead to confusion of contrastive phones.” (Manuel, 1990)

Languages for which increased variability in F0 does not result in decreased lexical identification do not undergo range contraction at a faster speech rate, as it is detrimental for perception.

But if so, why doesn’t Mandarin do this?

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 45 Discussion/Conclusions Discoveries from Oto-Manguean languages

Structural differences between languages influence tonal alignment.

The target of a tone need not be what we consider the typical unit of speech planning (the syllable) (Krakow, 1999; Goldstein et al., 2007).

Coarticulatory dissimilation may be a more common feature of languages with a larger number of tone levels and/or languages without tone sandhi processes.

Range expansion during fast speech preserves these tonal contrasts too.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 46 Discussion/Conclusions Investigating complexity

There is not only a unique complexity to the phonology of Oto-Manguean tonal systems, but also unique phonetic processes.

1 Our attempts to understand and model tonal processes should come to grips with this.

2 Suggests the need for a fusion between fieldwork and experimental research on tone (or at least a fusion of researchers).

3 Models of tone production should attempt typological coverage.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 47 Discussion/Conclusions Investigating complexity

There is not only a unique complexity to the phonology of Oto-Manguean tonal systems, but also unique phonetic processes.

1 Our attempts to understand and model tonal processes should come to grips with this.

2 Suggests the need for a fusion between fieldwork and experimental research on tone (or at least a fusion of researchers).

3 Models of tone production should attempt typological coverage.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 47 Discussion/Conclusions Investigating complexity

There is not only a unique complexity to the phonology of Oto-Manguean tonal systems, but also unique phonetic processes.

1 Our attempts to understand and model tonal processes should come to grips with this.

2 Suggests the need for a fusion between fieldwork and experimental research on tone (or at least a fusion of researchers).

3 Models of tone production should attempt typological coverage.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 47 Discussion/Conclusions Investigating complexity

There is not only a unique complexity to the phonology of Oto-Manguean tonal systems, but also unique phonetic processes.

1 Our attempts to understand and model tonal processes should come to grips with this.

2 Suggests the need for a fusion between fieldwork and experimental research on tone (or at least a fusion of researchers).

3 Models of tone production should attempt typological coverage.

Not every language show these patterns, but the patterns show us what constraints speakers control in tone production.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 47 Discussion/Conclusions Future directions

1 Modelling of coarticulation and rate effects in Triqui in a general production model (TADA) under NSF grant (Whalen & Xu).

2 Investigating tonal coarticulation data in Yoloxóchitl Mixtec.

3 Investigating tonal variability in Mixtec corpus data using forced alignment (DiCanio et al., 2012b, 2013).

4 Investigating the use of dynamic cues by native listeners in tone perception.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 48 Discussion/Conclusions Acknowledgements

The research was partly funded through a grant to Haskins Laboratories (Douglas Whalen, PI) on phonetic documentation in endangered languages.

Hosung Nam, Doug Whalen, Jonathan Amith, Rey Castillo García

The Yoloxóchitl Mixtec and Itunyoso Triqui communities, kùruaa nihírèh! ([ku2Ru4a43 ni3Pi4ReP1])! Thank you!

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 49 Appendices Duration of tone in YM

Little difference in duration among tones in monosyllables.

300

200

Duration (ms) Duration 100

0 complex rise concave double rise fall level rise Tonal melody type

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 50 Appendices Triqui coarticulation Duration differences for the Triqui vowels

No difference in duration between stressed vowels, but unstressed vowels were shorter.

300

250

200

Speech rate fast 150 normal Duration (ms) Duration

100

50

0

Left (stressed) Right (unstressed) Mid (monosyllabic) Vowels

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 51 Appendices Triqui coarticulation

Tones did not differ substantially in duration in medial monosyllabic words.

300

250

200

Speech rate fast 150 normal Duration (ms) Duration 100

50

0

/2/ /4/ /43/ /45/ Tone

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 52 Appendices Triqui coarticulation Left target: level tones

Anticipatory raising with T2, Left syllable t2 t3

2

1

Following tone t2

0 t4 t43 t45

z-score normalized log F0 log normalized z-score -1

-2

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Time (normalized)

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 53 Appendices Triqui coarticulation Left target: contour tones

Assimilatory raising of T45, Left syllable t32 t45

2

1

Following tone t2

0 t4 t43 t45

z-score normalized log F0 log normalized z-score -1

-2

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Time (normalized)

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 54 Appendices Triqui coarticulation Summary - left target

Significant effects of the following tone on the preceding tone’s F0 trajectory, most notably for a following /45/ and /43/.

Tone /2/ raises before contour tones /43, 45/, but is unaffected by following level tones.

Tone /45/ is realized with an earlier F0 peak before contour tones /43, 45/ than before following level tones.

It is not simply the presence/absence of a higher tone which causes tonal coarticulation here, but the presence of a contour.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 55 Appendices Triqui coarticulation Right target: level tones

Assimilatory changes in tone /3/ in melodies /3.32/, /3.45/, Right syllable t32 t45

2

1

Preceding tone t2

0 t4 t43 t45

z-score normalized log F0 log normalized z-score -1

-2

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Time (normalized)

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 56 Appendices Triqui coarticulation Larger summary

Strong anticipatory effects on word preceding medial monosyllable and strong carryover effects on following word too.

Dissimilatory effects on word which varies in tone in the elicitation frame, but only with highest/lowest adjacent targets.

Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 57 References Atterer, M. and Ladd, D. R. (2004). On the phonetics and phonology of “segmental anchoring” of F0: evidence from German. Journal of Phonetics, 32(2):177–197. Berthiaume, S. (2004). A Phonological Grammar of Northern Pame. PhD thesis, University of Texas at Arlington. Bradlow, A. R., Kraus, N., and Hayes, E. (2003). Speaking clearly for children with learning disabilities: sentence perception in noise. Journal of Speech, Language, and Hearing Research, 46:80–97. Brunelle, M. (2009). Northern and Southern Vietnamese tone coarticulation: A comparative case study. Journal of Southease Asian Linguistics Society, 1:49–62. Campbell, L., Kaufman, T., and Smith-Stark, T. C. (1986). Meso-america as a linguistic area. Language, 62(3):530–570. Castillo García, R. (2007). Descripción fonológica, segmental, y tonal del Mixteco de Yoloxóchitl, Guerrero. Master’s thesis, Centro de Investigaciones y Estudios Superiores en Antropología Social (CIESAS), México, D.F. Castillo Martínez, R. (2011). El sistema tonal del chinanteco de San Juan Quiotepec, Oaxaca. Master’s thesis, Centro de Investigaciones y Estudios Superiores en Antropología Social (CIESAS). Chang, Y.-C. and Hsieh, F.-F. (2012). Tonal coarticulation in Malaysian Hokkien: a typological anomaly? The Linguistic Review, 29:37–73. Cruz, E. (2011). Phonology, tone, and the functions of tone in San Juan Quiahije Chatino. PhD thesis, University of Texas at Austin. de Ávila, A. (2010). Mixtec plant nomenclature and classification. PhD thesis, University of California, Berkeley. Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 57 References DiCanio, C., Amith, J., and Castillo García, R. (2012a). Phonetic alignment in Yoloxóchitl Mixtec tone. Talk Presented at the Society for the Study of the Indigenous Languages of the Americas Annual Meeting. DiCanio, C., Nam, H., Whalen, D. H., Bunnell, H. T., Amith, J. D., and Castillo García, R. (2013). Using automatic alignment to analyze endangered language data: Testing the viability of untrained alignment. Journal of the Acoustical Society of America, 134(3):2235–2246. DiCanio, C. T. (2008). The Phonetics and Phonology of San Martín Itunyoso Trique. PhD thesis, University of California, Berkeley. DiCanio, C. T. (2012a). Coarticulation between Tone and Glottal Consonants in Itunyoso Trique. Journal of Phonetics, 40:162–176. DiCanio, C. T. (2012b). Cross-linguistic perception of Itunyoso Trique tone. Journal of Phonetics, 40:672–688. DiCanio, C. T. (2014). Cue weight in the perception of Trique glottal consonants. Journal of the Acoustical Society of America, 135(2):884–895. DiCanio, C. T. (submitted). Tonal classes in Itunyoso Trique person morphology. In Tone and Inflection, Empirical Approaches to Language Typology. Mouton de Gruyter. DiCanio, C. T., Nam, H., Whalen, D. H., Bunnell, H. T., Amith, J. D., and Castillo García, R. (2012b). Assessing agreement level between forced alignment models with data from endangered language documentation corpora. In INTERSPEECH-2012. Fougeron, C. and Jun, S.-A. (1998). Rate effects on French intonation: prosodic organization and phonetic realization. Journal of Phonetics, 26:45–69. Gandour, J., Tumtavitikul, A., and Satthamnuwong, N. (1999). Effects of speaking rate on Thai tones. Phonetica, 56:123–134. Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 57 References Gandour, J. T., Potisuk, S., and Dechongkit, S. (1994). Tonal coarticulation in Thai. Journal of Phonetics, 22(4):477–492. Gao, M. (2008). Mandarin Tones: An Articulatory Phonology Account. PhD thesis, Yale University. Gao, M. (2009). Gestural coordination among vowel, consonant and tone gestures in Mandarin Chinese. Chinese Journal of Phonetics, 2:43–50. Goldsmith, J. (1990). Autosegmental and metrical phonology. Oxford: Blackwell. Goldstein, L., Chitoran, I., and Selkirk, E. (2007). Syllable structure as coupled oscillator modes: Evidence from Georgian vs. Tashlhiyt Berber. In Proceedings of the 16th International Congress of Phonetic Sciences, pages 241–244. Saarbrücken, Germany. Jamieson, A. R. (1977). Chiquihuitlan Mazatec Tone. In Merrifield, W. R., editor, Studies in Otomanguean Phonology, pages 107–136. Summer Institute of Linguistics, University of Texas at Arlington. Kaufman, T. (1990). Early otomanguean homelands and cultures: some premature hypotheses. University of Pittsburgh Working Papers in Linguistics, 1:91–136. Knapp Ring, M. H. (2008). Fonología segmental y léxica del Mazahua. Instituto Nacional de Antropología e Historia (INAH). Krakow, R. A. (1999). Physiological organization of syllables: A review. Journal of Phonetics, 27:23–54. Ladd, D. R. (2004). Segmental anchoring of pitch movements: autosegmental phonology or speech production? In Quené, H. and van Heuven, V., editors, On Speech and Language: Studies for Sieb G. Nooteboom, pages 123–131. Netherlands Graduate School of Linguistics. Ladd, D. R., Faulkner, D., Faulkner, H., and Schepman, A. (1999). Constant “segmental anchoring” of F0 movements under changes in speech rate. Journal of the Acoustical Society of America, 106(3):1543–1554. Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 57 References Lehiste, I. (1970). Suprasegmentals. MIT Press, Cambridge, MA. Manuel, S. Y. (1990). The role of contrast in limiting vowel-to-vowel coarticulation in different languages. Journal of the Acoustical Society of America, 88(3):1286–1298. Morén, B. and Zsiga, E. C. (2006). The Lexical and Post-Lexical Phonology of Thai Tones. Natural Language and Linguistic Theory, 24:113–178.

Myers, S. (2003). F0 Timing in Kinyarwanda. Phonetica, 60:71–97. Palancar, E. (2009). Gramática y textos del hñöñhö, Otomí de San Ildefonso Tultepec, Querétaro., volume 1. Universidad Autónoma de Querétaro: Plaza y Valdés. Peng, S.-h. (1997). Production and perception of Taiwanese tones in different tonal and prosodic contexts. Journal of Phonetics, 25:371–400. Pike, K. L. (1948). Tone Languages. University of Michigan, Ann Arbor. Prom-on, S., Xu, Y., and Thipakorn, B. (2009). Modeling tone and intonation in Mandarin and English as a process of target approximation. Journal of the Acoustical Society of America, 125(1):405–424. Rensch, C. R. (1976). Comparative Otomanguean Phonology. Number 14 in Language Science Monograph. Bloomington: Indiana University. Scholz, F. (2012). Tone sandhi, prosodic phrasing, and focus marking in Wenzhou Chinese. PhD thesis, Leiden University. Shue, Y.-L., Keating, P., and Vicenik, C. (2009). VOICESAUCE: A program for voice analysis [computer program]. Journal of the Acoustical Society of America, 126(2221(A)). Silverman, D. (1997). Laryngeal complexity in Otomanguean vowels. Phonology, 14:235–261. Smiljanić, R. and Bradlow, A. R. (2005). Production and perception of clear speech in Croatian and English. Journal of the Acoustical Society of America, 118(3):1677–1688. Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 57 Appendices Triqui coarticulation Suárez, J. A. (1983). The Mesoamerican Indian Languages. Cambridge University Press. Thalin, A. (1980). Tlacoatzintepec Chinantec Syllable Structure. SIL-Mexico Workpapers, 4:1–8. Tilsen, S. (2013). Inhibitory mechanisms in speech planning maintain and maximize contrast. In Yu, A., editor, Origins of Sound Change: Approaches to Phonologization. Oxford University Press. Turk, A. and Shattuck-Hufnagel, S. (2000). Word-boundary-related duration patterns in English. Journal of Phonetics, 28:397–440. Xu, Y. (1994). Production and perception of coarticulated tones. Journal of the Acoustical Society of America, 96(4):2240–2253. Xu, Y. (1997). Contextual tonal variations in Mandarin. Journal of Phonetics, 25:61–83. Xu, Y. (1998). Consistency of tone-syllable alignment across different syllable structures and speaking rates. Phonetica, 55:179–203. Xu, Y., Gandour, J. T., and Francis, A. L. (2006). Effects of language experience and stimulus complexity on the categorical perception of pitch direction. Journal of the Acoustical Society of America, 120(2):1063–1074. Xu, Y. and Prom-on, S. (2014). Toward invariant functional representations of variable surface fundamental frequency contours: Synthesizing speech melody via model-based stochastic learning. Speech Communication, 57:181–208. Zhang, J. (2004). Contour Tone Licensing and Contour Tone Representation. Language and Linguistics, 5(4):925–968. Zhang, J. and Liu, J. (2011). Tone sandhi and tonal coarticulation in Tianjin Chinese. Phonetica, 68:161–191. Zsiga, E. and Nitisaroj, R. (2007). Tone features, tone perception, and peak alignment in Thai. Language and Speech, 50:343–383. Christian DiCanio (((Haskins))) Oto-Manguean tone 5/14/14 57