1 Introduction
Anong is a Tibeto-Burman language on the edge of extinction, spoken in Fugong County in the Yunnan Province in China.Footnote 1 In 1995, there were 62 fluent speakers and about 300 semi-fluent speakers living in Mugujia village (Sun Reference Sun2005b). The highly endangered nature of Anong is evident in bilingual contact situations. Anong is only used on that rare occasion when all the speakers are fluent in Anong, otherwise Lisu is the language of communication (Sun Reference Sun2005b, G. Thurgood Reference Thurgood, Nowak, Yoquelet and Mortensen2003).Footnote 2
This acoustic study investigates the distinctions among the coronals, examining in particular the three-way distinction among the voiceless fricatives /s ʂ ɕ/, among the aspirated voiceless affricates /tsʰ tʂʰ tɕʰ/ and the unaspirated voiceless affricates /ts tʂ tɕ/. The fundamental question being asked is how the three-way distinction among the coronal fricatives and affricates is established acoustically. To answer this question, Anong coronals are examined in terms of their spectral properties, their gravity center frequencies and the formant frequencies of the following vowel.
The results of this study are compared with acoustic studies of coronal fricatives and affricates in other languages with a three-way distinction. Specifically, Anong is compared with Standard Chinese (Svantesson Reference Svantesson1986; Ladefoged & Wu Reference Ladefoged and Wu1984), Polish (Halle & Stevens Reference Halle, Stevens, Kiritani, Hirose and Fujisaki1997; Nowak Reference Nowak2006), and Komi-Permyak (Kochetov & Lobanova Reference Kochetov and Lobanova2007). Standard Chinese shares with Anong the same inventory of coronal fricatives and affricates; Polish and Komi-Permyak inventories are similar except there is only one series of voiceless affricates. Additionally, the Anong retroflex fricative is compared with retroflex fricatives in Toda (Gordon et al. Reference Gordon, Barthmaier and Sands2002) and in O'odham (Dart Reference Dart1993).
1.1 Anong vowels and consonants
Sun (Reference Sun2005a) distinguishes eight main vowels plus /y/, a marginal vowel restricted to a handful of Chinese borrowings. Sun (Reference Sun2005a: 34) observes that the vowel /ɯ/ is more central than the symbol ɯ would lead one to expect, and, as the acoustic analysis of Anong vowels has shown (Thurgood Reference Thurgood, Trouvain and Barry2007), a more appropriate IPA symbol for /ɯ/ would, in fact, be /ɨ/. Thurgood's study also shows that the vowel [a] can be analyzed as an allophonic variation of /ɑ/, reducing Sun's inventory by one. Table 1 presents a comparison of the two systems.
Table 1 Anong vowel inventory.
The consonant inventory of Anong is shown in table 2. This inventory has been undergoing some rapid changes, including the loss of the retroflex series /ʈʰ ʈ ɖ ɳ ɭ/, which became /tʰ t d n l/, respectively. Three observations about Anong coronals are particularly important for the present study. First, Sun (Reference Sun2005a: 28) describes the alveolo-palatal series /ɕ tɕʰ tɕ dʑ/ as laminal alveolo-palatals, articulated slightly back and sounding close to [ç cçʰ cç 
j]. Second, Sun (Reference Sun2005a: 28) reports that the retroflex series /ʂ tʂʰ tʂ/ is phonetically similar to [ʃ tʃʰ tʃ]. Third, no lip rounding has been noted, something one would have expected Sun to have noticed and comment on were it present.
Table 2 Anong consonant inventory (after Sun Reference Sun2005a: 27).
2 The present study
2.1 Data
The data used for the present study have been selected from recordings made by Sun Hongkai in 1999. Sun recorded one speaker, who at the time was over 60 years old. He was quite possibly one of the last fluent speakers of Anong. The recordings were made on a camcorder. The forms were elicited in isolation, each repeated twice in a row. The database used in this study consisted of 28 tokens of coronal fricatives (10 tokens of /s/, 10 tokens of /ʂ/, 8 tokens of /ɕ/). The fricatives were analyzed in front of /ɑ/, /u/, and either /ɨ/ or /ɛ/. Twenty-eight tokens of the aspirated affricates were measured (10 tokens of /tsʰ/, 8 tokens of /tʂʰ/, 10 tokens of /tɕʰ/). Whenever possible, word-initial contexts were chosen. However, when the coronals occurred intervocalically, the results were inspected for possible positional variation (cf. Kochetov & Lobanova Reference Kochetov and Lobanova2007). The aspirated affricates were analyzed in front of /ɑ/, /u/, and either /i/ or /ɨ/. For the unaspirated affricates /ts tʂ tɕ/, and in particular for /tʂ/, the data were sparse, something that Sun observed as early as in 1988 (see Sun Reference Sun1988). In the available word-list (table 3), /ts/ occurs only before /ɑ/ and /tʂ/ only before /u/.
Table 3 Word-list used to examine Anong coronal fricatives and affricates.
2.2 Measurements
Using SoundEdit 16 version 2, the recordings were downsampled to 22,050 Hz when measuring the consonants, and to 11,025 Hz when measuring the following vowel. The data were then analyzed using Scicon's Macquirer software system. A number of acoustic measurements were taken. First, FFT spectra were computed using a 1024-point frame (approximately 46 milliseconds) centered around the middle of each fricative and the fricative part of each affricate. Numerical spectra were then averaged together over the tokens of each consonant appearing in each environment. Second, following the method described in Gordon, Barthmaier & Sands (Reference Gordon, Barthmaier and Sands2002), the center of gravity was also calculated for each token and then averaged together for each coronal appearing in each context. Finally, the frequencies for the first three formants (F1, F2, F3) were measured at the onset and in the middle of the following vowel /ɑ/. The SPSS statistical package (version 11.0.3) was used to perform repeated measures of ANOVA and post-hoc comparisons.
3 Acoustic analysis of Anong coronal fricatives
3.1 Spectral properties
The averaged spectra of the three coronal fricatives are plotted in figures 1–3. For each fricative, the tokens are separated by the following vowels. In general, there is a considerable uniformity across the analyzed data in the spectral characteristics of the three coronal fricatives. Starting with /s/, spectra show a steady rise with the highest spectral peaks centered at approximately 7.2 kHz, at which point there is a steep decline in noise. Spectra for /ɕ/ show a very different display of noise distribution. The most pronounced peaks are between 2.4 kHz and 4 kHz. Before all three vowels, the noise displays a relatively flat spectrum between 4.1 kHz and 7.1 kHz and then a steep decline. Spectral peaks for /ʂ/ cluster in two locations. The most prominent peak cluster is between 2.2 kHz and 3.4 kHz, the less prominent peak cluster is between 5.1 kHz and 6.2 kHz.
Figure 1 Averaged acoustic spectra for /s/ before two vowels.
Figure 2 Averaged acoustic spectra for /ɕ/ before three vowels.
Figure 3 Averaged acoustic spectra for /ʂ/ before three vowels.
The overall spectral shapes clearly differentiate the three coronal fricatives from each other. As an illustration, averaged spectra for /s ʂ ɕ/ occurring before /ɑ/ are plotted in figure 4. In this context, the amplitude peak of /s/ is at approximately 7.2 kHz. Before the amplitude peak, the intensity of /s/ is below the intensities of /ɕ/ and /ʂ/; following the amplitude peak, the intensity of /s/ is above the intensity of /ɕ/ and /ʂ/. The fricatives /ʂ/ and /ɕ/ display high noise at lower frequencies: /ʂ/ shows the greatest amplitude approximately at 2.4 kHz, /ɕ/ later at 4.1 kHz. The spectra for the two fricatives also differ in that /ʂ/ shows a secondary peak at ca. 6.2 kHz, while /ɕ/ shows a relatively flat spectrum between 4.1 kHz and 7.1 kHz and then a steep decline. The intensity of /ʂ/ is somewhat lower than that of /ɕ/.
Figure 4 Averaged spectra of /s/, /ʂ/ and /ɕ/ before /ɑ/.
3.2 Centers of gravity
Gravity center frequencies for Anong coronal fricatives before different vowels and averaged across these vowels are shown in figure 5. For /ʂ/ and /ɕ/, the gravity center frequencies were measured before /ɑ/, /ɨ/ and /u/. For /s/, the gravity center frequencies were measured only before a following /ɑ/ and /ɨ/. The center of gravity for /s/ is the highest; it is always greater than 6 kHz. The centers of gravity for /ʂ/ and /ɕ/, are always smaller than 5.5 kHz. An analysis of variance indicates that fricative type affects the center of gravity values (F(2,23) = 43.513, p < .0001). Scheffe's post-hoc tests reveal the difference between /s/ and /ɕ/ and between /s/ and /ʂ/ to be significant (p < .05). Pairwise comparison between /ʂ/ and /ɕ/ does not reach statistical significance.
Figure 5 Gravity center frequencies for Anong coronal fricatives before different vowels and averaged across vowels.
3.3 Formant frequencies of /ɑ/ following the fricative coronals
Formant values for the first three frequencies (F1, F2, F3) of /ɑ/ following /s/, /ʂ/, and /ɕ/, calculated at the onset and in the middle, are given in table 4. For comparison, the formant frequencies of /ɑ/ following a non-coronal (taken from Thurgood Reference Thurgood, Trouvain and Barry2007) are also presented.
Table 4 Averaged frequencies of /ɑ/.
Following /ʂ/ and /ɕ/, the average F1 is lower at the vowel onset than following /s/. The lowering effect of the alveolo-palatal /ɕ/ on F1 persists into the middle of the vowel; the lowering effect of the retroflex /ʂ/ on F1 vanishes by the middle of the vowel. There is no difference in F1 values between the beginning and the middle of the vowel when the vowel /ɑ/ follows the alveolar /s/. The formant frequency values of /ɑ/ following /s/ pattern with the formant frequency values of /ɑ/ arrived at in the earlier study (Thurgood Reference Thurgood, Trouvain and Barry2007). Turning to F2, the formant values are consistently higher not only at the onset but also in the middle of the vowel when /ɑ/ follows /ɕ/. In contrast, when /ɑ/ follows /ʂ/, even though the F2 is raised at the onset, it is not raised in the middle of /ɑ/. Turning to F3, the retroflex /ʂ/ triggers lowering of this formant. The effect of /ʂ/ on /ɑ/ persists well beyond the portion of the vowel immediately adjacent to the fricative. Neither /s/ nor /ɕ/ causes F3 lowering.
3.4 Discussion
All three Anong coronal fricatives /s ʂ ɕ/ are well differentiated by their spectral shapes. Among the three, the alveolar fricative /s/ is the most distinct. The most prominent spectral peaks for /s/ occur at higher frequencies than for /ʂ/ and for /ɕ/, at approximately 7.2 kHz. In contrast, the fricatives /ʂ/ and /ɕ/ display their most prominent spectral peaks at lower frequencies, between 2.2 kHz and 4 kHz depending on the following vowel. The spectra for the two fricatives differ in that /ʂ/ shows a less prominent, secondary peak cluster between 5.1 kHz and 6.2 kHz, while /ɕ/ shows a relatively flat spectrum following the most pronounced noise peak and then a steep decline.
The spectral shapes of Anong coronal fricatives /s ʂ ɕ/ parallel the spectral shapes of Polish coronal fricatives /s ʂ ɕ/ obtained by Nowak (Reference Nowak2006). In Polish, as in Anong, the spectral shapes of the retroflex /ʂ/ are characterized by two clusters of spectral peaks. Depending on the following vowel, Polish spectral peaks cluster at frequencies between 2.8 kHz and 3.5 kHz and between 5.4 kHz and 5.7 kHz (Nowak Reference Nowak2006: 142). Two spectral peaks have also been reported for Mandarin Chinese /ʂ/ and Komi-Permyak /ʂ/. In Mandarin Chinese, the spectral peaks are located around 2.9 kHz and around 4.5 kHz (Ladefoged & Wu Reference Ladefoged and Wu1984: 273–274). In Komi-Permyak, they are located at lower frequencies, around 1.5 kHz and around 3.5 kHz (Kochetov & Lobanova Reference Kochetov and Lobanova2007: 70).
Anong shares with Polish the spectral shapes of the alveolo-palatal /ɕ/. In both languages, the spectral peaks are followed by a relatively flat spectrum and then a steep decline. In Polish, the spectral peaks are located close together at frequencies between 3.3 kHz and 3.9 kHz depending on the vowel context, the following flat part of the spectrum declines at frequencies higher than 6 kHz (Nowak Reference Nowak2006: 142). This spectral shape has not been observed for Komi-Permyak /ɕ/, where /ɕ/ spectra are not significantly different from /ʂ/ spectra (Kochetov & Lobanova Reference Kochetov and Lobanova2007: 72–73).
All three Anong coronal fricatives /s ʂ ɕ/ are well differentiated by the formant frequencies of the following vowel. The alveolo-palatal /ɕ/ triggers a significant lowering of F1 and a slight raising of F2 in the following /ɑ/. These changes in formant frequency values parallel changes reported in other languages. For example, for Polish lower F1 and higher F2 of a vowel adjacent to an alveolo-palatal have been reported (Halle & Stevens Reference Halle, Stevens, Kiritani, Hirose and Fujisaki1997, Nowak Reference Nowak2006). However, what makes Anong fricative coronals different is that their influence on the formant frequency values is found not only at the onset but also into the middle of the vowel.
In Anong, the retroflex /ʂ/ triggers a significant lowering of F3 in the following /ɑ/. F3 lowering of a vowel adjacent to a retroflex fricative has been reported for other languages, for example for Toda (Gordon et al. Reference Gordon, Barthmaier and Sands2002) and O'odham (Dart Reference Dart1993). However, in Polish (Nowak Reference Nowak2006) and Komi-Permyak (Kochetov & Lobanova Reference Kochetov and Lobanova2007), /ʂ/ does not trigger a lowering of F3. As suggested by Kochetov & Lobanova (Reference Kochetov and Lobanova2007: 73), the lack of F3 lowering can be explained by smaller curling of the tongue in the production of /ʂ/.
The gravity center frequencies have been found to differentiate /s/ from both /ʂ/ and /ɕ/ well, with /s/ showing significantly higher gravity center frequencies than those of /ʂ/ and /ɕ/. The high gravity center frequencies have been shown to correlate with the frontness of the constriction in articulation (Stevens Reference Stevens1998). This correlation is found cross-linguistically (Gordon et al. Reference Gordon, Barthmaier and Sands2002). Likewise in Svantesson's (Reference Svantesson1986) study of Mandarin Chinese, high gravity center frequencies of /s/ differentiate it from lower gravity center frequencies of /ʂ/ and /ɕ/. The difference between Anong and Mandarin Chinese, however, is that while in Mandarin Chinese the center of gravity is higher for /ɕ/ than for /ʂ/, in Anong the difference between the two coronals does not reach statistical significance. A lack of significant differences between the gravity center frequencies of /ʂ/ and /ɕ/ has been observed for Polish /ʂ/ and /ɕ/ (Nowak Reference Nowak2006) and for Komi-Pemyak /ʂ/ and /ɕ/ (Kochetov & Lobanova Reference Kochetov and Lobanova2007).
4 Acoustic analysis of Anong coronal affricates
4.1 Spectral properties of fricative components
Averaged spectra of the fricative components of /tsʰ/, /tʂʰ/ and /tɕʰ/ are plotted in figures 6–8. For each affricate, the spectra are separated according to the following vowel. The fricative component of /tsʰ/ has a relatively flat spectrum followed by the most prominent noise peak at about 5.6 kHz when followed by [u] and at about 6.8 kHz when followed by [ɨ]. Before [ɑ], the flat spectrum is first followed by a fall and only then by a rise with a peak at about 6.7 kHz. Regardless of the following vowel, the highest spectral peaks of /tsʰ/ are centered at higher frequencies.
Figure 6 Averaged acoustic spectra for /tsʰ/ before three vowels.
Figure 7 Averaged acoustic spectra for /tɕʰ/ before three vowels.
Figure 8 Averaged acoustic spectra for /tʂʰ/ before three vowels.
The spectral characteristics of /tɕʰ/ differ depending on the following vowel. When followed by a high vowel, /tɕʰ/ displays most pronounced peaks at lower frequencies followed by a slow decline in noise. When followed by the low vowel /ɑ/, the spectrum shows a cluster of secondary, less prominent peaks at higher frequencies. The spectra of /tʂʰ/ show a very similar display of noise distribution.
For a clearer view of the two types of affricates pattern together, averaged spectra for /tʂʰ/ and /tɕʰ/ occurring before the low vowel and before the non-back high vowel are plotted in figures 9 and 10. For both affricate types before the low vowel, a secondary cluster of peaks occurs between 5.7 kHz and 6.4 kHz. For both affricate types before the non-back high vowel, a relative flat spectrum occurs between 4.9 kHz and 7.9 kHz and then there is a decline.3
Figure 9 Averaged acoustic spectra for /tʂʰ/ and /tɕʰ/ before the low vowel.
Figure 10 Averaged acoustic spectra for /tʂʰ/ and /tɕʰ/ before a non-back high vowel.
4.2 Centers of gravity
Gravity center frequencies for Anong coronal affricates before different vowels and averaged across the vowels are shown in figure 11. An analysis of variance indicates that affricate type affects the center of gravity values (F(2,21) = 18.613, p < .0001). Before all vowels, the alveolar /tsʰ/ has the highest gravity center values, while the alveolo-palatal /tɕʰ/ has the lowest gravity center values. Pairwise post-hoc comparisons involving /tsʰ/, /tʂʰ/ and /tɕʰ/ show a significant difference at minimally p < .05.
Figure 11 Gravity center frequencies for aspirated coronal affricates before different vowels and averaged across the vowels.
For the unaspirated affricates, the comparison is more restricted. However, even though the data are limited in the number of tokens available, the gravity center frequencies of unaspirated /ts/, /tʂ/ and /tɕ/ pattern in the same way as the gravity center frequencies of aspirated /tsʰ/, /tʂʰ/ and /tɕʰ/. The alveolar affricate /ts/ has the highest gravity center values, while the alveolo-palatal affricate /tɕ/ has the lowest, with the gravity center values of the retroflex affricate /tʂ/ placed in between (see figure 12).
Figure 12 Gravity center frequencies for the fricative component of unaspirated affricates averaged across vowels.
4.3 Formant frequencies of /ɑ/ following a coronal affricate
Formant frequency values at the onset and in the middle of /ɑ/ following the three coronal affricates are given in table 5. There is no significant difference in formant frequencies between the /ɑ/ following the unaspirated /ts/ and /tɕ/, and the /ɑ/ following their aspirated counterparts, /tsʰ/ and /tɕʰ/. In table 5, formant frequencies of /ɑ/ from aspirated and unaspirated counterparts are collapsed. For comparison, the formant frequencies of /ɑ/ following a non-coronal (taken from Thurgood Reference Thurgood, Trouvain and Barry2007) are presented.
Table 5 Average frequencies of /ɑ/ following a coronal affricate.
As table 5 shows, the formant values of /ɑ/ following alveolar affricate /ts/ measured in the middle of the vowel do not differ from the formant values of /ɑ/ given in Thurgood (Reference Thurgood, Trouvain and Barry2007). F1 is lowered when /ɑ/ follows either the retroflex or the alveolo-palatal affricate. A lowering effect of both types of affricates occurs not only at the onset but also in the middle of /ɑ/. Additionally, the retroflex affricate has not been found to trigger lowering of F3 in the following vowel. As a result, the retroflex and the alveolo-palatal do not reliably differ in their effect on F1, F2 and F3 of the following vowel.
4.4 Discussion
The results show that of the three acoustic measurements used in this study, the best differentiation among the three coronal affricates is achieved by the gravity center frequencies. Gravity centers for alveolar /tsʰ ts/ are highest. Their relatively high frequency center of gravity is shared with high gravity centers for Mandarin Chinese /tsʰ ts/ (Svantesson Reference Svantesson1986). However, the gravity centers of the remaining two types of affricates do not pattern in the same way. In Anong, the alveolo-palatals /tɕʰ tɕ/ are characterized by the lowest gravity centers. In Mandarin Chinese, retroflex /tʂʰ tʂ/ have the lowest gravity centers.
Spectral shapes of the fricative component of /tsʰ/ are similar to spectral shapes of the corresponding fricative /s/, with both their spectral peaks located at higher frequencies. Spectra for /tɕʰ/ and for /tʂʰ/ show a very similar display of noise distribution with the most pronounced peaks between 2 kHz and 4 kHz. For both affricate types before the non-back high vowel, the spectral peaks are followed by a relatively flat spectrum and then a decline. However, before the low vowel, the spectra of /tɕʰ/ and for /tʂʰ/ are characterized by secondary peaks between 5.7 kHz and 6.4 kHz. Before the non-back high vowel, the spectral shapes of /tɕʰ/ and /tʂʰ/ are similar to the spectral shape of /ɕ/. Before the low vowel, the spectral shapes of /tɕʰ/ and /tʂʰ/ are similar to the spectral shape of /ʂ/.
The fricative component of /tʂʰ/ does not trigger lowering of F3 in the following vowel. Instead, it patterns with /tɕʰ/ in triggering a lower F1 in /ɑ/. A lower F1 reflects a longer constriction not characteristic of a retroflex (Halle & Stevens Reference Halle, Stevens, Kiritani, Hirose and Fujisaki1997, Stevens Reference Stevens1998). It again points to /tʂʰ/ being more like /tɕʰ/.
5 Conclusions
The study has examined the acoustic characteristics of Anong coronal fricatives and affricates, focusing on the three-way distinction among the voiceless fricatives /s ʂ ɕ/, the voiceless aspirated affricates /tsʰ tʂʰ tɕʰ/, and the voiceless unaspirated affricates /ts tʂ tɕ/. The study has shown a lack of parallelism between the coronal fricatives and affricates. With the fricatives, the spectral shapes and the formant frequencies of the following vowel display a three-way distinction among the fricatives /s ʂ ɕ/, but the gravity center frequencies only provide a two-way distinction, between /s/, on the one hand, and /ʂ/ and /ɕ/ on the other. However, with the affricates, the spectral shapes and the formant frequencies of the following vowel do not differentiate /tsʰ tʂʰ tɕʰ/ well; these acoustic measurements only show a two-way distinction, between /tsʰ/, on the one hand, and /tʂʰ/ and /tɕʰ/ on the other. However, the gravity center frequencies do preserve a three-way distinction.
As for retroflexes in Anong, the spectral shapes of /ʂ/ and the formant frequencies of the following vowel show that /ʂ/ is acoustically a retroflex. First, the spectral shape of /ʂ/ with its two spectral peaks not only differentiates it well from the remaining two coronals /s/ and /ɕ/, but it also patterns well with the spectral shapes of the retroflex fricatives in other languages (Komi-Permyak, Mandarin Chinese, Polish). Second, Anong /ʂ/ triggers a significant lowering of F3 in a following /ɑ/, a feature that has been associated with the presence of retroflexion. It is of interest that in Anong the effect of /ʂ/ on the following vowel is observed not only at the vowel onset but also into the middle of the vowel. In contrast, the /tʂʰ/, termed retroflex by Sun, is not a typical retroflex acoustically. First, the fricative part of /tʂʰ/ does not consistently pattern with the fricative /ʂ/. Second, the spectral peaks of /tʂʰ/ are not different from those of /tɕʰ/. Third, the vowel /ɑ/, when it follows /tʂʰ/, does not show a lowered F3, but a lowered F1 instead. Thus, in the speech of the one of the last fluent speakers of Anong, the retroflex affricate /tʂʰ/ has been lost along with other retroflex series. In contrast, the retroflex fricative /ʂ/ is still preserved.
Acknowledgements
I would like to thank Sun Hongkai for making this study possible. My thanks are also due to Graham Thurgood, David Bradley, Adrian Simpson, and an anonymous reviewer for their helpful comments. This work was supported by NSF Grant #606232 Endangered Languages in China.
