PHONETICS AND PHONOTACTICS OF VOWEL LARYNGEALIZATION IN UPPER NECAXA TOTONAC Rebekka Puderbaugh University of Edinburgh [email protected] ABSTRACT and Tlachichilco Tepehua [33]. As a result, word final vowels frequently are devoiced, laryngealized, The present paper reports on an investigation of the or dropped entirely. phonetics and phonotactics of vowel laryngealiza- Little is known about the phonetics of laryngeal- tion in Upper Necaxa Totonac. Using analysis tech- ized vowels into Totonacan, but some preliminary niques inspired by corpus linguistics, an analysis acoustic analyses have been performed. In Pa- of transcribed dictionary forms revealed a highly pantla Totonac, non-modal vowels were found to significant relationship between vowel laryngealiza- have overall lower intensity than modal vowels, and tion and the consonant that followed immediately tokens produced with stiff voice appeared to be as- after the vowel. Specifically, laryngealized vowels sociated with the spread of voicing onto preceding were far more likely to occur before glottal stops stops [16]. In Misantla Totonac, Trechsel & Faber than non-laryngealized vowels were. Further to [31] measured the F1-F2 vowel space and the dif- these findings, an acoustic measure of voice quality ference in amplitude between F0 and F1 (a mea- (H1-H2) was extracted from multiple time points of sure they refer to as VQI, for Voice Quality Index). laryngealized and non-laryngealized vowels before The analysis revealed a high degree of interspeaker and after glottal stops and other consonants. A sta- variability in the F1-F2 space, but a more stable tistical comparison of H1-H2 values between vowels outcome with respect to VQI: laryngealized vowels in various contexts revealed that vowels followed by had lower VQI scores than non-laryngealized vow- glottal stops had stronger indications of non-modal els for both speakers. A recent study of duration and phonation regardless of their own laryngealization vowel space in UNT found that laryngealized vow- category, while vowel laryngealization itself did not els tended to have lower F0 than non-laryngealized affect H1-H2. In light of these findings, the the la- vowels [11]. ryngealized vowels may be interpreted as resulting Phonation types have highly variable acoustic from collocation with glottal stops instead of bear- profiles across langauges, with differences depend- ing contrastive larygnealization themselves. ing in part on whether the phonation is contrastive or allophonic, breathy, creaky, or otherwise modified. Keywords: Meso-America, documentation, laryn- Spectral and acoustic measures have been found to gealized vowels, phonotactics, phonation differentiate between modal and non-modal phona- tion types in several languages including English, 1. INTRODUCTION Korean, and Hmong [12], Mazatec, Mpi, and Chong [6], Zapotec [10], and Gujarati [9, 17], among oth- Upper Necaxa Totonac (ISO [tku], hereafter UNT) ers. The difference in amplitude between the first is a Totonacan language spoken by approximately and second harmonics, often reported as H1-H2, or 3400 people in four communities situated along the H1*-H2* when adjusted for first and second for- Upper Necaxa River in the Sierra Norte of Puebla, mant values, appears to be the most reliable mea- Mexico. Like other languages of the Totonacan fam- sure for differentiating phonation types across lan- ily, UNT is supposed to maintain a phonation con- guages [19, 29], though unadjusted H1-H2 may also trast between laryngealized and non-laryngealized be used [18]. In addition to spectral differences, the vowels [4]. This contrast may occur with any of timing of vowel phonation is affected by contrastive- UNT’s contrastive vowel qualities, as well as ei- ness: non-modal phonation lasts longer and is more ther of two contrastive lengths (short and long). highly differentiated from modal phonation when Vowel laryngealization may also indicate phono- the phonation differences are contrastive than when logical phrase boundaries, as is the case in other they are non-contrastive [6, 12]. In Hupa, allophonic related language varieties, such as Coatepec To- laryngealization spreads from following consonants tonac [21, 24, 25], Filomeno Mata Totonac [23], onto a portion of preceding vowels, but never laryn- gealizes the entire length of the vowel [13]. Acous- text. Chi-squared tests were performed on these tic analysis may therefore reveal evidence for inter- count data in order to determine whether laryngeal- preting laryngealization as either allophonic or con- ization of context segments was related to laryngeal- trastive in UNT. ization of target segments. In the present paper, the relationship between glottal stops and laryngealized vowels in UNT is 2.2. Results explored from two perspectives. First, a colloca- tional analysis of vowels and stops demonstrates that Count data for stops and vowels are reported in Ta- glottal stops are highly likely to be preceded by la- ble 1. Non-laryngealized vowels occurred in higher ryngealized vowels. Second, H1-H2 values from proportions across both oral and glottal stops, as vowels in a variety of segmental contexts are ana- would be expected given the higher proportion of lyzed, revealing that the presence of a glottal stop non-laryngealized vowels in the dictionary overall. has a greater effect on laryngealization than does the Figure 1 illustrates the distributions of stops. Both vowel laryngealization category itself. oral and glottal stops appeared in comparable pro- 2. SEGMENTAL COLLOCATES OF VOWEL portions before laryngealized and non-laryngealized LARYNGEALIZATION IN TRANSCRIPTION vowels. After vowels, over 88% of glottal stops occurred after laryngealized vowels, and more than 2.1. Methods 70% of oral stops occurred after non-laryngealized vowels. A complementary pattern was apparent in The materials for the present analysis were extracted vowels (Figure 2): 65% of laryngealized vowels from the digital version of the Upper Necaxa To- occurred before glottal stops, and more than 90% tonac Dictionary [5]. All forms were included in of non-laryngealized vowels occurred before oral the analysis, including head words, inflected forms, stops. and affixes that received their own entries. Table 1: Count data of stops and vowels in their The data required little pre-processing before in- preceding and following environments. clusion in the analysis because the orthography of UNT is transparent at the phonemic level. Dic- Before vowels After vowels tionary forms were copied into a plaintext file for- Oral 9676 5494 mat, then transliterated into IPA using grep and Glottal 2419 3400 regex (regular expressions) find-and-replace meth- Before stops After stops ods. Following a method similar to [3], transcrip- Laryng V 4603 3425 tions were accepted as given in the published dic- Non-laryng V 4295 8676 tionary. Because some segments were encoded by complex character sequences (i.e affricate digraphs, ejective fricatives, vowels with length, stress and Figure 1: Stops before and after vowels, divided larygnealization diacritics), phonemic symbols were by laryngeal type. Black bars represent non- laryngealized vowels; gray bars represent laryn- delimited by inserted spaces on either side. Word gealized vowels. boundary markers (#) were inserted at the end of each word to ensure that segments were not inter- preted as adjacent across word boundaries. The resulting list of prepared segment data was then converted into a single text string from which immediately preceding and following seg- ments were identified for all segments using the shift function from the data.table package [8] in R [27]. Segmental collocates of stops and vowels only are analyzed here. Vowels were divided into laryngeal- ized and non-laryngealized, while stops were di- vided into oral and glottal subsets. The results of the collocation analysis revealed a strong relationship between glottal stops and the la- ryngealization of vowel that precede them. These The tabulated data was analyzed using a Chi- data were further subdivided and cross-tabulated ac- squared test that revealed a highly significant rela- cording to the laryngealization of the relevant con- tionship between stop type and vowel laryngealiza- and all speakers were presented with words in the Figure 2: Vowels before and after stops, divided by stop type (oral/glottal). Black bars represent same list order. The procedure for recording the oral stops; gray bars represent glottal stops. word list was explained to speakers prior to begin- ning the recording. Speakers were asked to repeat each word three times within the frame sentence in ixla wanli’ ... chuwa [Sla wanli ... tSuwa] ’he said ... now’. During recording, speakers˜ had access to written forms and translations, as well as discussion with the author. Audio data were segmented and transcribed in Praat [7] according to the dictionary form of word items, rather than a close transcription of the pho- netic signal. Boundaries were placed between vow- els and adjacent glottal stops based on the regular- ity of oscillation period. H1-H2 measures were ex- tracted from vowel tokens using a Praat script de- signed to imitate the measures taken by the Voice- tion in both preceding and following environments. Sauce software developed at UCLA [32, 28]. Vowel Oral stops were more likely to precede laryngeal- quality was included in the linear models as a ran- ized vowels than would be expected given the over- dom intercept, allowing for the model to adjust all proportion of