They Talk Mutumutu: Variable Elision of Tense Suffixes in Contemporary Pitjantjatjara

Vowel elision is common in Pitjantjatjara and Yankunytjatjara connected speech. It also appears to be a locus of language change, with young people extending elision to new contexts; resulting in a distinctive style of speech which speakers refer to as mutumutu (‘short’ speech). This study examines the productions of utterance-final past tense suffixes /-nu, -ïu, -Nu/ by four older and four younger Pitjantjatjara speakers in spontaneous speech. This is a context where elision tends not to be sociolinguistically or perceptually salient. We find extensive variance within and between speakers in the realization of both the vowel and nasal segments. We also find evidence of a change in progress, with a mixed effects model showing that among the older speakers, elision is associated with both the place of articulation of the nasal segment and the metrical structure of the verbal stem, while among the younger speakers, elision is associated with place of articulation but metrical structure plays little role. This is in line with a reanalysis of the conditions for elision by younger speakers based on the variability present in the speech of older people. Such a reanalysis would also account for many of the sociolinguistically marked extended contexts of elision.


Introduction
This study investigates phonetic variability in utterance-final past tense suffixes /-nu, -ïu, -Nu/ in Pitjantjatjara, a Western Desert (Pama-Nyungan) language of Central Australia. Word-or phrase-final vowel elision has been documented in several Australian languages as a phonetic process in connected speech, but also as an outcome of historical sound change. In current day Pitjantjatjara, there are signs of both regular vowel elision in connected speech and an ongoing change in progress extending the contexts of this elision. Alongside these processes, there are signs of vowel elision in contexts which tend not to be perceptually salient to native speakers and listeners. The production of vowels in these contexts appears to be highly variable and not sociolinguistically marked. Our first aim in this paper is to provide a fine-grained acoustic phonetic analysis of vowel elision in one of these highly variable but not sociolinguistically or perceptually salient contexts. Our second aim is to describe any apparent change in progress between generations within this less salient context. Our research questions are: 1.
How are Pitjantjatjara speakers producing utterance-final past tense suffixes? 2.
Does a comparison of older and younger speakers suggest a change in progress? 3.
If there is evidence of a change in progress, what does this change look like?

Apical
Laminal Velar Alveolar Post-Alveolar Alveo-Palatal A phonetic/phonological overview of the sound system of Pitjantjatjara is provided by . Other phonetic studies of the language by Tabain and colleagues have investigated lexical stress , stop bursts (Tabain and Butcher 2015), nasal consonants (Tabain et al. 2016), intonation (Tabain and Fletcher 2012), and the alveolar-retroflex distinction (Tabain et al. 2020). Butcher (1996) also reports on some connected speech processes in Pitjantjatjara along with some other Australian languages. As well as these, Defina et al. (2020) investigated the prosody of clause chains in the language. This study builds on these by considering synchronic variation, and by examining processes which occur in spontaneous connected speech.
In this study, we are specifically focused on past-tense suffixes. There are three allomorphs /-nu, -ɳu, -ŋu/, depending on the conjugation class of the verb; we refer to these collectively as -Nu to represent the different nasal segments. The four conjugation classes are partially determined according to the metrical structure of the verbal stem (Wilmoth and Mansfield 2021). Verbal stems whose right edges are aligned with the right edge of a bimoraic metrical foot are assigned to the l-or Ø-class, and take the past tense allomorphs -ŋu and -ɳu, respectively; e.g., (carpa)φ-ŋu 'enter-PST' in the Ø-class, and (ampu)φ-ɳu 'hug- PST' in the l-class. Stems which are monomoraic or end in an unfooted syllable are assigned to the n-or ng-class, and take the -nu or -ŋu past tense allomorphs, respectively; e.g., (carpa)φ(cu-nu)φ 'insert-PST' in the n-class, (u-ŋu)φ 'give-PST' in the ng-class. This means that in the l-and Ø-classes, the past tense suffix is unfooted, whereas it is footed when affixed to verbs in the n-and ng-classes. This also means that the retroflex allomorph -ɳu A phonetic/phonological overview of the sound system of Pitjantjatjara is provided by . Other phonetic studies of the language by Tabain and colleagues have investigated lexical stress , stop bursts (Tabain and Butcher 2015), nasal consonants (Tabain et al. 2016), intonation (Tabain and Fletcher 2012), and the alveolar-retroflex distinction (Tabain et al. 2020). Butcher (1996) also reports on some connected speech processes in Pitjantjatjara along with some other Australian languages. As well as these, Defina et al. (2020) investigated the prosody of clause chains in the language. This study builds on these by considering synchronic variation, and by examining processes which occur in spontaneous connected speech.
In this study, we are specifically focused on past-tense suffixes. There are three allomorphs /-nu, -ïu, -Nu/, depending on the conjugation class of the verb; we refer to these collectively as -Nu to represent the different nasal segments. The four conjugation classes are partially determined according to the metrical structure of the verbal stem (Wilmoth and Mansfield 2021). Verbal stems whose right edges are aligned with the right edge of a bimoraic metrical foot are assigned to the l-or Ø-class, and take the past tense allomorphs -Nu and -ïu, respectively; e.g., (carpa) ϕ -Nu 'enter-PST' in the Ø-class, and (ampu) ϕ -ïu 'hug-PST' in the l-class. Stems which are monomoraic or end in an unfooted syllable are assigned to the n-or ng-class, and take the -nu or -Nu past tense allomorphs, respectively; e.g., (carpa) ϕ (cu-nu) ϕ 'insert-PST' in the n-class, (u-Nu) ϕ 'give-PST' in the ng-class. This means that in the l-and Ø-classes, the past tense suffix is unfooted, whereas it is footed when affixed to verbs in the n-and ng-classes. This also means that the retroflex allomorph -ïu is always unfooted, while the alveolar allomorph -nu is always footed. The Languages 2021, 6, 69 4 of 23 velar allomorph -Nu can be footed or unfooted depending on whether it is in the Ø-or ng-class. The above analysis is illustrated in Table 2; see Wilmoth and Mansfield (2021) for the full details. 1 As we will show below, metrical structure and place of articulation have relevance for variable elision of final vowels.

Vowel Elision in Pitjantjatjara and Yankunytjatjara
In this section we provide an overview of vowel elision in Pitjantjatjara and Yankunytjatjara. There is no discussion of elision in Pitjantjatjara in early sources such as Trudinger (1943), who was living in Pukatja, or Douglas (1957), who provides a sophisticated early phonetic and phonological description of Western Desert (encompassing a number of lects including that spoken at Pukatja).
In Yankunytjatjara, Goddard (1985, pp. 14-16) describes patterns of elision of 'weak vowels'. These are defined as unstressed vowels which follow an unstressed syllable and precede a stressed syllable (either within a word or across word boundaries). Following Wilmoth and Mansfield (2021), we understand 'weak' to refer to unfooted vowels. Goddard describes these elisions within prosodic words or phrases, and only in cases of vowel hiatus or where there are homorganic consonants on either side of the targeted vowel. Examples of this are shown in (1). Goddard makes no mention of utterance-final elision.
apu-ngka katu /(apuN) ϕ ka (katu) ϕ / → [apuNkatu] rock-LOC above 'on the hills' Langlois (2004), documenting teenage girls' Pitjantjatjara in Utju/Areyonga in the 1990s, describes similar patterns of elision as Goddard, as well as extensions to the contexts of elision, which speakers refer to as tjitjiku wangka 'children's talk' (p. 33). 2 Although the possible phonological contexts are not fully explicated, the examples suggest the following additional contexts which are relevant for this study: (a) Elision can occur utterance-finally; examples of utterance-final mulap(a) 'true' and watalp(i) 'almost' are given. The utterance final past tense elision we focus on in this paper falls within this category. (b) Vowels can be elided even when they would be expected to be within a foot, e.g., in the perlative suffix /-wanu/ being pronounced as [wan].
Those teenage speakers recorded by Langlois would now be in their forties, in between the two age cohorts examined in this paper. Anecdotally, it seems that Pukatja Pitjantjatjara is somewhat more conservative than Utju/Areyonga Pitjantjatjara, and many speakers have commented that various features now used by young people in Pukatja are in fact 'Utju way'. (Utju/Areyonga is approximately 400 km north of Pukatja and has a higher proportion of residents with other language backgrounds, such as Pintupi-Luritja, Warlpiri, or Arrernte; the latter is phonologically very different from Western Desert languages).
As this study is focussed on the particular context of past tense suffixes, we will not include a detailed analysis of the current contexts for vowel elision, and their frequencies among different speakers given different linguistic conditions. Final vowel elision is extremely frequent in the Pitjantjatjara currently spoken in Pukatja, particularly among young people, and broadly similar to what Langlois (2004) described in Areyonga Teenage Pitjantjatjara. It is hoped that this study can serve as the phonetic basis for further variationist research into this variable in other contexts.

From Subconscious Variation to Sociolinguistic Meaning
One of the main motivations for this study is the concerns that speakers have voiced regarding language change. Older Pitjantjatjara speakers often express worry for the future of their language. This is not only regarding potential language shift, but also what they perceive as changes initiated by young people which are not mulapa 'true' Pitjantjatjara. One of these changes is the shortening of words, i.e., final vowel elision, as can be seen in the following passage from speaker SL 3 . Here, we also see the source of the title of this paper: they talk mutumutu [short].
All of us education workers think so, all the old people and middle aged people, like M and K. We hear it and we say, "No! No, you're changing those words, it's like you're shortening them. Speak like this, say the words properly." Sometimes we tell the young ones in our families, to really, you know, speak Pitjantjatjara the long way, you know? "Say those Pitjantjatjara words the long way." But they talk mutumutu [short]. They shorten it when they're speaking. They cut up Pitjantjatjara. They shorten it. . . . They don't fully understand how to say it the long way, so they say it short. It's like that. (SW20190425-08-PitjKunpuKanyini) The same phenomenon of 'shortening' words is discussed in this excerpted passage, from participant NB: (3) Ka wangka nyanga paluru nganampa kuwari kampa kutjuparingu, ka tjitji malatj tjutang kuwari wangka kutjupa tjuta wangkanyi. Palu wangka nyanga paluru, wangka palulatuya ngaranyi paluru, tjukutjuku tjukutjuku panya wangka, mutumutu tjana palyan i, tjukutjuku tjukutjuku.
Our language has changed, and the children nowadays are speaking lots of different languages. The language is still here, but a little bit at a time, they're making it short, just a little bit at a time. (SW20190531-01-TjukurpaPt1) As these comments show, word-final vowel elision is a very salient sociolinguistic variable associated with young people, and a cause of concern for many older Anangu. 4 However, vowel elision is not limited to young people, and is not always noticed by speakers. The focus of this study was also selected in part due to the experiences of the first two authors. While transcribing with Pitjantjatjara speakers, we encountered many cases where we could not hear a final vowel or suffix, but they would reject transcriptions without it. Though they understand and support our goal of representing language variation and are generally happy to transcribe spontaneous Pitjantjatjara speech with non-standard spelling, they would often insist that an elided tense or case suffix was in fact fully present and should be written down. There was a mismatch in our perception of the speech signal due to our different linguistic experiences. As our proficiency as non-native Pitjantjatjara listeners has improved, at times we have found ourselves perceiving 'phantom' vowels or suffixes for which there is no spectrographic evidence. This perceptual illusion could be due to residual cues, or to listener expectations in the absence of audible phonetic cues. Listener expectation has been shown to motivate listeners to 'hear' vowels that are not phonetically present, as in several studies of perceptual vowel epenthesis in Japanese, for example (e.g., Dupoux et al. 1999;Kilpatrick et al. 2019).
If a vowel or suffix is absent or just barely perceptible, this might serve as the source of a sound change, according to a listener-oriented model of sound change (e.g., Ohala 1981). That is, while one generation might perceive a full suffix due to their expectations based on prior experience and/or perception of subtle acoustic cues, following generations may encounter more reduced realizations more often, and their phonological representation of the form of the suffix may eventually differ, leading to the absence of the vowel or suffix entirely.
Based on these and other experiences, and conversations with speakers such as those excerpted above, it appears that in some contexts word-final vowel elision is a stylistic resource associated with young people, while in other contexts (such as inflectional suffixes) it is below the level of consciousness. We focus on the latter context to shed some light on how the former, more sociolinguistically salient changes may have been initiated.

Materials and Methods
The data for this study comes from eight native Pitjantjatjara speakers, all female. While the gender variable would certainly be interesting to investigate with regard to these phenomena, as female researchers ourselves we abide by local norms in working most closely with women. The speakers were selected from a larger corpus to represent a 'younger' and 'older' cohort. The younger speakers were 17, 17, 24, and 26 years old at the time of recording. The older speakers were 52, 59, 62, and 69. Speakers were recruited through our various connections and relationships, as part of several ongoing research projects in Pukatja. Data collection for this project received ethics approval from the University of Melbourne Human Research Ethics Committee (project numbers 1953646.1 and 1647234.2). Permits to travel and conduct research on the APY Lands were granted by APY. All participants provided their informed consent and were paid for their time.
Speakers were recorded with a Sennheiser lapel microphone (model ew-100-eng-g3) or a Countryman ISOMAX headset microphone (model MHHP6HH05L), with either a Zoom H6 or H4n recorder depending on the session, sampled at 48 kHz. The data consists of spontaneous speech of different genres, including conversation, narrative, and semistructured elicitation such as storybook or cartoon retellings. For the younger speakers, and usually but not always with the older speakers, recordings were conducted within groups of peers, not alone with the researcher.
Utterance-final tokens of past tense verbs were extracted from the corpus, and any examples with overlapping speech, excessive background noise, or other factors such as coughing or laughing were excluded. The utterance-final context was chosen as it is a common context for elision and allows us to control for the linguistic environment. The verbs extracted were those which were initially transcribed with a full past tense suffix, or with the final nasal only, in a first pass (non-phonetic) transcription; due to the perceptual reasons mentioned above, verbs were not transcribed as missing suffixes entirely. Where perceptual cues may have been ambiguous, intended past tense suffixes were identifiable based on the context of the discourse, for example within a narrative told in the past tense. There were 280 tokens in total; the number of tokens per speaker is shown in Table 3. The verb class and lexical item of each token were coded by the first and second authors. The original orthographic transcription of the suffix was blinded so as not to bias the phonetic analysis. The third author then conducted a fine-grained acoustic phonetic analysis of each token. Each verb was analysed in terms of the realization of the nasal (present or absent and its quality), the realization of the vowel (present or absent and its quality), and any other notable aspects. This analysis was carried out using both auditory and acoustic analysis (observing cues in the spectrogram), using Praat version 6.1.34 (Boersma and Weenink 2020). Each analysis category is defined and illustrated below.

Denasalization
Denasalization is a process that is reported to be common in connected speech in Australian languages, especially in suffixes and especially in nasal-vowel-nasal-vowel sequences (Butcher 2006, p. 10). Butcher (1999, p. 480) describes denasalization as occurring when the 'lowering of the velum is often "left too late", with the result that orality perseverates into the nasal consonant'. This means it is not uncommon for nasal segments to appear on spectrograms as oral stop-like, having a less resonant nature than a prototypical nasal. Denasalised nasals can have less evident formant activity, or no evident formant activity, having far less resonance than a prototypical nasal. On spectrograms they may appear as completely stop-like, simply having a 'voice bar' with the same characteristics as a voiced stop, and in some cases in our data these were followed by a release burst as would be observed for a stop. Note that this is not the same as a prestopped nasal, which consists of short stop-like activity at the beginning of a nasal segment (i.e., Butcher 1999); instead denasalization applies to the entire segment. Figure 2 shows a devoiced nasal at the end a word (where the vowel is deleted). This example contains three /n/ segments in the word ninti-nu 'teach-PST'. The first in the sequence (n1) is a prototypical nasal, the second (n2) is less nasal-like (formants are weaker) and the last (n3) appears entirely denasalised. This segment has no formant activity and its very low energy is also evident on the waveform when compared with the other nasals in the sequence. There is a pre-stop at the beginning of the 'n3' segment which highlights the fact that denasalization is not necessarily exactly the same as having absolutely no nasal airflow. For a large proportion of this segment there is little spectral activity aside from a 'voice bar' which indicates vocal fold vibration, and then the voicing cuts out entirely towards the end of the segment. There is also an oral release burst evident at the very end of 'n3'. The auditory impression is similar to what English speakers describe as speaking while holding one's nose or having a blocked nose (i.e., there is far less nasality). and the last (n3) appears entirely denasalised. This segment has no formant activity and its very low energy is also evident on the waveform when compared with the other nasals in the sequence. There is a pre-stop at the beginning of the 'n3' segment which highlights the fact that denasalization is not necessarily exactly the same as having absolutely no nasal airflow. For a large proportion of this segment there is little spectral activity aside from a 'voice bar' which indicates vocal fold vibration, and then the voicing cuts out entirely towards the end of the segment. There is also an oral release burst evident at the very end of 'n3'. The auditory impression is similar to what English speakers describe as speaking while holding one's nose or having a blocked nose (i.e., there is far less nasality).

Approximation
This is also a common process for nasals in Australian Indigenous languages, as reported by Butcher (2006, p. 10). In the example below, the initial nasal in Nari-Nu 'lie-PST' is a prototypical nasal, the second (in the suffix) is approximated. Approximants are often described as having vowel-like acoustic structure (i.e., Ladefoged and Maddieson 1996, p. 372), and this is certainly evident in the example shown in Figure 3. The approximated nasal in the suffix appears more vowel-like on the spectrogram, which is caused by less supralaryngeal constriction. This is seen in measures highlighting greater amplitude and resonance, including more darkness in the spectrum and more energy in the waveform for the approximated /N/. There is also clearly evident formant structure in the approximated nasal. By contrast the word initial nasal has less energy due to the fact that there is vocal tract constriction for this articulation; it is lighter in the spectrum, the waveform dips, and while nasal formants are evident on the spectrum, they are not vowel-like. nasal in the suffix appears more vowel-like on the spectrogram, which is caused by less supralaryngeal constriction. This is seen in measures highlighting greater amplitude and resonance, including more darkness in the spectrum and more energy in the waveform for the approximated /ŋ/. There is also clearly evident formant structure in the approximated nasal. By contrast the word initial nasal has less energy due to the fact that there is vocal tract constriction for this articulation; it is lighter in the spectrum, the waveform dips, and while nasal formants are evident on the spectrum, they are not vowel-like.

Devoicing
Devoicing was a common occurrence in the data, both for nasal consonants and vowels. In Figure 4, showing the word a-nu 'go-PST', the final vowel is heavily devoiced and there is an evident difference between the initial voiced /a/ and the devoiced final /u/ vowel. The /a/ has clear formant structure, high energy activity on the waveform and is especially dark on the spectrum (highlighting resonance). The /u/ by contrast has formant structure typical of a close back vowel as would be expected, with low F1 and low F2 (Ladefoged and Maddieson 1996, p. 286), but has very little voicing-the energy on the waveform peters out, and there is only slight voicing in the initial portion of the spectrum indicated by the energy in the lower portion of the spectrogram which indicates vocal fold vibration.

Devoicing
Devoicing was a common occurrence in the data, both for nasal consonants and vowels. In Figure 4, showing the word a-nu 'go-PST', the final vowel is heavily devoiced and there is an evident difference between the initial voiced /a/ and the devoiced final /u/ vowel. The /a/ has clear formant structure, high energy activity on the waveform and is especially dark on the spectrum (highlighting resonance). The /u/ by contrast has formant structure typical of a close back vowel as would be expected, with low F1 and low F2 (Ladefoged and Maddieson 1996, p. 286), but has very little voicing-the energy on the waveform peters out, and there is only slight voicing in the initial portion of the spectrum indicated by the energy in the lower portion of the spectrogram which indicates vocal fold vibration.

Shortening
Vowels and consonants were categorized as short impressionistically, when they short relative to the majority of tokens produced by the speaker. Short vowels were often similar in length to English schwa. Other processes of vowel reduction, such as centralization, were not analyzed.

Shortening
Vowels and consonants were categorized as short impressionistically, when they short relative to the majority of tokens produced by the speaker. Short vowels were often similar in length to English schwa. Other processes of vowel reduction, such as centralization, were not analyzed.

Place of Articulation
The transcribed place of articulation of the suffix nasal was blinded for the phonetic analysis. The difference between alveolar, palatal and velar nasals tends to be very salient auditorily, and there are also acoustic cues that assist in determining place of articulation. Alveolar nasals lack acoustic cues found in VN and NV transitions for the other places of articulation; for example, when N is a palatal, F2 is especially high in preceding and following vowels. When the nasal is velar, it is common to see a velar pinch in F2 and F3 in VN transitions. The third author coded place of articulation based on these auditory and acoustic cues and we then compared this coding with the phonemic place of articulation of each lexeme.

Nasalization and Rounding
The example in Figure 5 shows a case in which there is both nasalization and rounding on the final vowel in the stem. Rounding on vowels is generally indicated by a lower F3. While rounding could be difficult to see visually on a spectrogram without comparing with an unrounded vowel counterpart, this example is useful because the rounding begins shortly after the vowel articulation has started, so it is easy to see the F3 movement which dips within the first part of the vowel; in this case the formant movement cannot be attributed to any other kind of coarticulation given that the vowel occurs word and phrase finally. It is also worth noting that the vowel here is especially long, possibly a compensatory strategy because of the loss of segments. Nasalization has a number of acoustic cues, though not all of these will always be evident on a spectrogram, especially in connected dynamic speech. Ladefoged and Maddieson (1996, pp. 288-89) describe a weaker second formant in nasalised vowels, and this is also somewhat evident in the example below when comparing the final vowel sound to the preceding vowel segments. Nasalization is also especially clear auditorily in these data, in particular because it contrasts with the typically phonetically non-nasalised vowel articulations produced by speakers of Aboriginal languages (i.e., Butcher 2006).
Labialization (rounding) of nasals is not especially clear spectrographically, and Nasalization has a number of acoustic cues, though not all of these will always be evident on a spectrogram, especially in connected dynamic speech. Ladefoged and Maddieson (1996, pp. 288-89) describe a weaker second formant in nasalised vowels, and this is also somewhat evident in the example below when comparing the final vowel sound to the preceding vowel segments. Nasalization is also especially clear auditorily in these data, in particular because it contrasts with the typically phonetically non-nasalised vowel articulations produced by speakers of Aboriginal languages (i.e., Butcher 2006).
Labialization (rounding) of nasals is not especially clear spectrographically, and hence we do not provide a visual example of this. Researchers who work on the acoustic cues between labialised and non-labialised consonants have reported, for example, that labialised versus non-labialised sounds are in fact quite similar acoustically, and this has actually led to delabialization of consonants being a historically common occurrence (Beeley 2015, p. 4). However, this rounding is relatively clear auditorily, and we feel that native listeners are picking up on these cues in interpreting the vowel as being present. We further refer to the work by Hansen and Hansen (1969, p. 157) discussed in Section 4, who also observed auditorily (for Pintupi) that the rounding feature is present on nasals even when the vowel is elided.

Results
The analysis showed wide-spread variation in the realization of the utterance final past tense suffix -Nu among both Older and Younger speakers. Realizations were highly gradient and ranged from fully articulated to completely elided suffixes. Figure 6 shows the distribution of tokens produced with both the nasal and vowel, with the nasal alone (vowel elision), or with neither (elision of the suffix). The proportions of these different realizations were similar across age groups: both groups produced the vowel in just more than half of cases; 59% and 56% of tokens for Older and Younger speakers, respectively. Productions with the nasal alone were also common among both age groups (32% and 36%, respectively) and both age groups were observed fully eliding the suffix (9% and 8%, respectively). Note the color-coding in the figure shows the proportion of tokens from each speaker, where it can be seen that all speakers produced each variant, except for MD who did not produce a fully elided suffix. This exception is likely due to the combination of rarity of full elision in the corpus and token numbers (26 in total for this speaker). In the following sections we describe the range of vowel and nasal realizations in more detail (Sections 4.1 and 4.2, respectively), as well as instances where both vowel and nasal are elided (Section 4.3). In Section 4.4 we compare the behavior of Older and Younger speakers.

Realization of Vowels
When the vowel was produced, it was produced with variable quality by both Younger and Older speakers. Realizations of the vowel were categorized as fully produced, devoiced, short, or short and devoiced. The proportions of each category are shown in Figure 7. When the vowel was present, both cohorts tended to produce it fully, with younger speakers producing a higher proportion of non-reduced vowel tokens (49% and 75% of tokens for Older and Younger speakers, respectively). An example of this is illustrated in Figure 8. Both cohorts also showed lenition of the vowel in some tokens, either through devoicing or shortening, or both. The Older speakers all showed similar tendencies. The vowel lenitions among the Younger speakers showed more individualized patterns. AR showed no vowel lenition; she either produced a full vowel or none at all. DM and TE both produced devoiced vowels, with and without shortening. In contrast, CB only lenited vowels by shortening them. These idiosyncrasies and the tendency for Younger speakers to favor full vowel production over reduced vowels may indicate an emerging trend towards categoricalization rather than the more gradient cline of productions shown by the older speakers. Further investigation would be needed to better understand these potentially age-graded patterns of behavior.

Realization of Nasals
The data also showed variation in the articulation of the nasal consonant. Nasals were coded as being fully produced, approximated, denasalized, devoiced, shortened, or produced with a different place of articulation. Combinations of these different processes were also observed. The proportions of each are shown in Figure 9.
When the vowel was produced, Older and Younger speakers tended to produce a full nasal consonant preceding it (86% of tokens among both cohorts). In some cases, the nasal was produced with degrees of lenition even when the subsequent vowel was fully produced. The lenition processes in these cases were: approximation, denasalization (observed with Younger speakers only), devoicing (observed with Older speakers only),

Realization of Nasals
The data also showed variation in the articulation of the nasal consonant. Nasals were coded as being fully produced, approximated, denasalized, devoiced, shortened, or produced with a different place of articulation. Combinations of these different processes were also observed. The proportions of each are shown in Figure 9. When the vowel was produced, Older and Younger speakers tended to produce a full nasal consonant preceding it (86% of tokens among both cohorts). In some cases, the nasal was produced with degrees of lenition even when the subsequent vowel was fully produced. The lenition processes in these cases were: approximation, denasalization (observed with Younger speakers only), devoicing (observed with Older speakers only), shortening (observed with Older speakers only), or a shift of the place of articulation to alveolar such that /ï/ or /N/ were realized as [n]. There was only one instance with multiple types of reduction of the nasal when the vowel was also produced; in this instance a younger speaker both devoiced and shifted the place of articulation to [n]. When the vowel was present, the most frequent process affecting the nasal was a change in the place of articulation.
Lenition of the nasal consonant was more frequent when the following vowel was not produced. In that case, Younger speakers lenited the nasal 44% of time, compared to 34% of the time among the Older speakers. Combinations of lenition processes were also more frequent when the vowel was absent. Older speakers tended to devoice and change the place of articulation, while Younger speakers, particularly DM, tended to denasalize and change the place of articulation. Simultaneous devoicing and denasalization was observed with both cohorts. Denasalization and change in place of articulation was observed with the younger cohort only (DM), devoicing and change in place of articulation was observed with the older cohort only, as was denasalization and shortening. We considered all of these processes to be types of reduction. Figure 10 shows a spectrogram of an example of a fully realized nasal with the vowel elided. Figure 11 shows the spectrogram of the word kunkun-ari-N(u) 'sleep-INCH-PST' produced by UT with a shortened nasal and no vowel. In the final vowel of the stem, formants 2 and 3 show a clear 'velar pinch', indicating the onset of a velar closure. There is then a short period of nasal consonant closure straight after the vowel where /N/ is labelled, but no evidence of a final vowel. It is clear that voicing ceases after the velar closure.   Nasal consonants were also regularly (phonetically) prestopped; this was observed in various contexts including the utterance-final past tense suffixes. Older speakers produced 15% of their past tense suffix nasal consonants with prestopping, while Younger speakers prestopped 8% of tokens. Prestopping was observed with fully produced nasals and those where the place of articulation was changed. Since this does not represent a lenition of the consonant, we will not examine it further here. Nasal consonants were also regularly (phonetically) prestopped; this was observed in various contexts including the utterance-final past tense suffixes. Older speakers produced 15% of their past tense suffix nasal consonants with prestopping, while Younger speakers prestopped 8% of tokens. Prestopping was observed with fully produced nasals and those where the place of articulation was changed. Since this does not represent a lenition of the consonant, we will not examine it further here.

Suffix Deletion: When Neither Nasal Nor Vowel Are Produced
Sometimes, speakers of both age groups fully elided the suffix, producing neither the nasal nor the vowel segment of the past tense suffix. This occurred in 9% of Older speaker tokens and 8% of Younger speaker tokens. Speakers elided the entire suffix with verbs of all lengths, including monosyllabic verb roots u-Nu 'give-PST' and ña-Nu 'see-PST'; this was unexpected given Pitjantjatjara has been analyzed as having a bimoraic word minimum (Goddard 1985, p. 13).
One of the questions raised in Section 1.3 was whether there are any remaining phonetic traces of the elided segments that may lead speakers to perceive suffixes as being present. When the syllable was elided, there were detectable traces on the final vowel of the verb stem 50% of the time for Older speakers, and 73% of the time for Younger speakers.
In some cases, only nasalization of the final stem vowel was detectable (8% for Older speakers and 9% for Younger). In most cases where there was some trace of the elided tense suffix, both nasalization and lip rounding were detectable (42% for Older speakers and 64% for Younger). See Figure 12 for the distribution of this across cohorts and speakers.

Comparison of Older and Younger Speakers
The phonetic analysis described above showed extensive variation in the production of utterance-final past tense suffixes by both Older and Younger speakers. In this section, we examine whether there were any significant differences between the productions of these two cohorts. Our focus is on the extent of elision or lenition, with the hypothesis that younger speakers may lenite suffixes more often or to a greater extent. We coded extent of lenition by adding one point for every process of lenition (e.g., devoicing, shortening, place reduction) of the nasal or vowel and three points if the nasal or vowel was fully elided. This created a quantifiable scale of reduction from 0 (fully produced nasal and vowel) through to 6 (fully elided suffix, with or without residual traces on the verb stem). For example, a value of 1 was given to any token with a devoiced vowel or nasal. A reduction value of 2 was given to a token with a devoiced nasal and vowel. A reduction value of 3 was given to a token with a full nasal and no vowel, or to a token with an alveolarized nasal and a devoiced and shortened vowel. Tokens with no vowel and a lenited nasal have a value of 4 or 5. Fully elided suffixes-with and without trace nasalization and lip rounding on the stem-were coded as 6. Tokens where the suffix nasal was phonemically alveolar /n/ were marked ineligible for place of articulation reduction (17% of tokens).
We examined whether there was a difference in extent of reduction between the two cohorts using a Linear Mixed Effects model within RStudio (R Core Team 2020; RStudio Team 2020) with the glmmTMB package (Magnusson et al. 2020). Extent of reduction was the dependent (ordinal) variable. Fixed effects were AgeGroup (Older or Younger), Footedness (is the suffix within a metrical foot or not), Place of Articulation (is the nasal consonant velar or not 5 , abbreviated PoA), and Length (is the verb stem monomoraic or not). These are all dichotomous variables and all contrasts were set to sum to zero. The starting model was the maximal model with all potential interactions. Interactions with Length were removed as they led to confounds (all monomoraic verb roots have suffixes within a metrical foot) or they were not significant and their removal improved the fit of the model. Random slopes for speaker with Footedness, Place of Articulation and Length were initially included, but excluded due to nonconvergence. The final model and all effects can be seen in Table 4. There was a significant effect of Place of Articulation; suffixes with velar nasals were more likely to be reduced. There was also a significant two-way interaction of AgeGroup with Footedness. Moreover, the three-way interaction AgeGroup, Footedness and Place of Articulation was significant. An examination of the effects plots ( Figure 13) shows that Older speakers are more likely to reduce if the nasal is velar AND the suffix is unfooted, but if one of those conditions does not hold-i.e., if the suffix is footed OR the consonant is coronal-they tend not to reduce the suffix.
Younger speakers were more influenced by place of articulation in their reduction processes. They were more likely to reduce the suffix if the nasal is velar. This difference in place of articulation was more marked for footed syllables, but beyond this there was no apparent influence of Footedness for the Younger speakers, unlike for Older speakers. Languages 2021, 6, x FOR PEER REVIEW 18 of 25 Figure 13. Effects plot of the three-way AgeGroup*Footedness*PlaceofArticulation effect. A higher reduction value indicates a greater tendency to reduce the past tense suffix. The plot shows the clear trend for Older speakers to reduce when the suffix is Not footed AND the nasal consonant is velar. The younger speakers show a tendency to reduce more with velar consonants in general and little difference according to footedness.

Discussion
This paper set out to determine how Pitjantjatjara speakers are producing utterancefinal past tense suffixes /-nu, -ɳu, -ŋu/ (RQ1), whether a comparison of older and younger speakers suggests a change in progress (RQ2) and if so, what form it takes (RQ3).

Variable Elision in Pitjantjatjara and Other Australian Languages
Regarding the first research question, it was found that the realization of past tense suffixes was highly gradient. Nasals were found to be fully produced, approximated, denasalized, devoiced, shortened, or completely absent; all forms of reduction that are known to occur in Australian languages (Butcher 1996(Butcher , 2008. In addition, some nasals shifted to an alveolar place of articulation, so that the nasal in the suffix was still fully present, but realized as /n/. Sometimes, nasal realizations involved a combination of these processes. Vowels were fully produced, devoiced, short, short and devoiced, or absent. In some cases, both the nasal and the vowel were absent and so the suffix was completely elided, in these cases residual cues such as nasalization or rounding were sometimes found on the final vowel of the verb stem. This wide variation in realization of both the nasal and vowel was observed with all speakers, regardless of age. While both age groups exhibited the full range of suffix productions from fully realized nasal and vowel all the way through to fully elided, there was nevertheless a difference between age groups in the conditions under which speakers in each group tended towards elision. Older speakers were more likely to elide the suffix if the nasal was velar and the final syllable was not footed, but not if it was footed or if the consonant was coronal. Younger speakers, on the other hand, were less influenced by metrical structure, while place of articulation was a clearer contributing factor. These observed patterns in the contexts of elision both fit with what is known about elision in Australian languages generally. For example, Butcher (1996, p. 91) notes that elision can be conditioned by particular segments or prosodic conditions; also observing that its occurrence is highly gradient and variable. Figure 13. Effects plot of the three-way AgeGroup*Footedness*PlaceofArticulation effect. A higher reduction value indicates a greater tendency to reduce the past tense suffix. The plot shows the clear trend for Older speakers to reduce when the suffix is Not footed AND the nasal consonant is velar. The younger speakers show a tendency to reduce more with velar consonants in general and little difference according to footedness.

Discussion
This paper set out to determine how Pitjantjatjara speakers are producing utterancefinal past tense suffixes /-nu, -ïu, -Nu/ (RQ1), whether a comparison of older and younger speakers suggests a change in progress (RQ2) and if so, what form it takes (RQ3).

Variable Elision in Pitjantjatjara and Other Australian Languages
Regarding the first research question, it was found that the realization of past tense suffixes was highly gradient. Nasals were found to be fully produced, approximated, denasalized, devoiced, shortened, or completely absent; all forms of reduction that are known to occur in Australian languages (Butcher 1996(Butcher , 2008. In addition, some nasals shifted to an alveolar place of articulation, so that the nasal in the suffix was still fully present, but realized as /n/. Sometimes, nasal realizations involved a combination of these processes. Vowels were fully produced, devoiced, short, short and devoiced, or absent. In some cases, both the nasal and the vowel were absent and so the suffix was completely elided, in these cases residual cues such as nasalization or rounding were sometimes found on the final vowel of the verb stem. This wide variation in realization of both the nasal and vowel was observed with all speakers, regardless of age. While both age groups exhibited the full range of suffix productions from fully realized nasal and vowel all the way through to fully elided, there was nevertheless a difference between age groups in the conditions under which speakers in each group tended towards elision. Older speakers were more likely to elide the suffix if the nasal was velar and the final syllable was not footed, but not if it was footed or if the consonant was coronal. Younger speakers, on the other hand, were less influenced by metrical structure, while place of articulation was a clearer contributing factor. These observed patterns in the contexts of elision both fit with what is known about elision in Australian languages generally. For example, Butcher (1996, p. 91) notes that elision can be conditioned by particular segments or prosodic conditions; also observing that its occurrence is highly gradient and variable.
Similar patterns of full and gradient word-final vowel elision are found in many Australian languages, beyond Western Desert. In Kayardild, Evans (1995) describes a regular process of 'prosodic truncation', whereby /a/ is deleted utterance-finally. Bowern (2013) also discusses utterance-final devoicing in Bardi, as well as variation in the elision of word-final vowels, which 'is both a phonetic and a phonological process' (p. 91). There are differences in the presence of final vowels between different dialects of Bardi, as well as a higher occurrence of final vowels in older materials as compared to modern materials. In Yidiny too, there is a semi-regular process of deletion of unfooted, word-final vowels (see Dixon 1977;Round 2017 inter alia).
In other languages, vowel elision is less phonologised. In an overview of connected speech processes in Australian languages, Butcher (2008) gives several examples of vowel elision word-and/or utterance-finally in Pitjantjatjara, as well as Warlpiri, DjambarrpuyNu, and Adnyamathanha. These languages differ in the precise contexts in which vowels are elided; a representative example from each is given in Table 5. In the first three examples, the final vowel in the first of two words is elided, while the final vowel of waraci is elided in the final example. In the Warlpiri example, a vowel is also elided utterance finally.  Hansen and Hansen (1969) do not describe regular patterns of vowel elision in their description of Pintupi 6 phonology, but they do note an occurrence of elision with remaining traces of rounding similar to what we observe here for Pitjantjatjara (Section 3.3): 'When word final, a nonstressed rounded nasal is a portmanteau phone signaling the sequence of the phonemes Nu or nu: NalkutiNu 'had just eaten" (p. 157); that is, the vowel itself is elided but its rounding feature is realized on the nasal.
The phenomenon of word-and/or utterance-final elision is common across Australian languages, and our phonetic analysis in this study shows how gradient these productions can be, and how such a change in progress might happen between generations of speakers.

Language Contact and Change
Regarding our second and third research questions, we found a difference between age groups suggesting a change in progress, whereby elision by younger Pitjantjatjara speakers is likely no longer driven by footedness. One question stemming from this observation is to what extent this is a contact induced change, or a language-internal shift. Utterance-final devoicing or elision is cross-linguistically a very common locus of sound change both in contact situations and language-internally, see for example Cser (2015). This is also the case in Australian languages, where such changes have occurred in the traditional languages mentioned above as well as in contact varieties or situations of language shift. For example, in young people's Dyirbal, Schmidt (1985, p. 64) found that the future suffix -ny was frequently elided completely (amongst other changes to verbal morphology). In Modern Tiwi, Lee (1987, p. 71) found that morpheme-final vowels were elided more often by younger speakers in vowel hiatus contexts.
A similar loss of final vowels has occurred in some case markers (e.g., dative -ki → -k) in the mixed language Light Warlpiri (O'Shannessy et al. 2019;O'Shannessy 2006); which likely had its source in phonetic variation in Warlpiri. In a study of Warlpiri connected speech, Ingram et al. (2008) found that there can be radical lenition at phrase edges, particularly in inflectional suffixes. In this context, there can be 'extensive anticipatory vowel assimilation, often the sole remaining cue to an otherwise completely lenited following consonant' (p. 94); this is similar to our findings in Section 3.3 and Hansen and Hansen (1969) observation mentioned above for Pintupi. Pentland and Ingram (2005) also describe utterance-final lenition processes in Warlpiri in some detail, including breathy voice, glottalization, and devoicing. In contrast to Pitjantjatjara, vowel or syllable deletion happened utterance finally, but not utterance-medially; note that this is also contrary to some Warlpiri examples in Butcher (2008), such as that presented in Table 5. This type of phenomenon likely instantiated an ongoing change in the realization of case markers in several Warlpiri-speaking communities, especially those where Light Warlpiri is now spoken (O'Shannessy et al. 2019;O'Shannessy 2006).
While regularizing and increasing vowel elisions are observed as linguistic changes in traditional languages such as Bardi (Bowern 2013) and contact languages such as Light Warlpiri (O'Shannessy et al. 2019;O'Shannessy 2006), the Pitjantjatjara situation is different. Our present findings for Pitjantjatjara do not suggest such an increase in the elision of utterance-final past tense vowels. Rather, what we see is a reanalysis of the associated contexts of elision. While we cannot be sure what has caused this shift, it appears to be language-internally driven based on a plausible reanalysis of the speech younger speakers hear. Older speakers produce highly gradient realizations of utterance-final past suffixes, with a tendency to elide with Ø-class verbs where the suffix is unfooted and contains a velar nasal. Younger speakers appear to have focused more on the velar context rather than the metricality of the suffix. This shift could be related to the relative frequency of each verb class. Ø-and l-class verbs with their unfooted suffixes are more frequent than ng-and n-class verbs. This means that suffixes with velar nasals are more often in the high elision unfooted context. However, a fuller investigation of the contexts of elision across the language and their frequencies would be needed to better understand the motivations behind the younger speakers' reanalysis.

Syncretism and Blocking Effects
One further point worth comparing between Pitjantjatjara and Light Warlpiri is in the forms of the ergative and locative case markers. The Warlpiri ergative -ngku varies between -ngku, -ngu and -ng in Light Warlpiri, whereas the locative -ngka varies between -ngka and -nga, but not -ng (O'Shannessy et al. 2019;O'Shannessy 2006, p. 20). The avoidance of locative -ng is interpreted as a blocking effect, so as to maintain a distinction between ergative and locative case. In contemporary Pitjantjatjara as spoken at Pukatja, we see similar variation in the realization of case markers. The ergative varies between -ngku, -ngu, and -ng (as well as -ngk before a vowel-initial word). However, there does not appear to be any blocking effect with the locative, which varies between -ngka, -nga, -ng, and -ngk, even though this results in syncretism with the ergative. Whether the -ng and -ngk forms are ergative or locative can only be determined by context. This is relevant for the current paper as a similar syncretism may arise in Pitjantjatjara tense suffixes if vowels are completely elided. One potential explanation for the observed contexts of elision could then be that speakers are avoiding syncretism or ambiguity with other tense suffixes, specifically the present tense forms. A comparison of the past vs. present endings is shown in Table 6. The l-class forms are indicated with *; these are the forms which would result in syncretism should the final vowel of the tense suffixes be wholly elided, leaving only /-ï/. A tendency not to elide the unfooted but retroflex l-class past tense -ïu might therefore also be interpreted as being due to a blocking effect. Table 6. Pitjantjatjara past and present tense endings.

Ø-Class l-Class ng-Class n-Class
Past /-Nu/ /-ïu/* /-Nu/ /-nu/ Present /-ñi/ /-ïi/* /-Nañi/ /-nañi/ This scenario is similar to a historical change in Kukata (Platt 1968). Kukata and Kukatja are two Western Desert varieties traditionally spoken further south of Pitjantjatjara/Yankunytjatjara with a verbal paradigm very similar to Pitjantjatjara. Kukata appears to have already undergone a change due to final vowel elision on present and past tense suffixes, with the exception of retroflex-initial suffixes -ïu/-ïi in the l-class. This is illustrated in Table 7. Forms where this final vowel has been lost in Kukata are indicated with *. The final /i/ or /u/ has been lost in Kukata in most present and past tense allomorphs, where the retained consonant (/ñ/ or /N/) distinguishes between the tenses. However, in the l-class forms, loss of the final /i/ or /u/ would mean that the present and past are indistinguishable. The observed patterns of elision by older Pitjantjatjara speakers are in line with this syncretism avoidance explanation. We found that while older speakers tend to elide unfooted syllables more than footed ones, this only held for the velar-initial Ø-class suffixes and not the retroflex-initial l-class suffixes. It should be noted, however, that the Pitjantjatjara elision is not as categorical as in Kukata and older speakers do not always elide the vowel in Ø-class suffixes, nor do they consistently produce the vowel in l-class suffixes, or even in the footed and not potentially syncretic n-and ng-class suffixes. A potential alternative explanation is a phonetic effect: Butcher has noted that peripheral segments (i.e., velar nasals) are 'particularly prone to lenition ' (1996, p. 91). It is not possible to decide between these two potential explanations based on the current study.
The younger speakers' tendency to elide whenever the suffix is velar-initial results in more elisions with the Ø-and ng-class verbs, neither of which would result in a syncretism with the present tense.

Sociolinguistic and Perceptual Salience
Word-final vowel elision is a very salient sociolinguistic variable associated with young people in the Pitjantjatjara speaking community. This shortening of words is often commented on as a cause for concern by older Anangu, as seen in the reflections excerpted in Section 1.3. In contrast, utterance-final elision of past tense suffixes and their vowels is not perceptually salient to Pitjantjatjara speakers and our results show older speakers and younger speakers are equally likely to elide these vowels. The contexts of elision do, however, differ between the two groups, with younger speakers relatively uninfluenced by metrical structure. This is in line with, and a potentially contributing factor to younger speakers eliding vowels in footed syllables more generally, a context which is not associated with elision by older speakers. This phenomenon may then become a salient point of difference, where younger people are eliding vowels which older people feel should not be elided, as far as the rhythmic nature of speech is concerned, such as in the non-traditional contexts observed by Langlois (2004).
One question is why the elision of these past-tense suffix vowels is so much less salient to native listeners than elision within lexical roots. A potential reason is a difference in predictability between lexical stems versus grammatical affixes. Speakers may be hearing illusory vowels and nasals because they expect the suffix to be there, as observed for Japanese listeners hearing illusory vowels between consonant clusters which are not allowable within Japanese phonology (Dupoux et al. 1999;Kilpatrick et al. 2019). Further research would be needed to determine whether the vowels in grammatical suffixes are more predictable than the final vowels of lexical stems in Pitjantjatjara. Another possibility is that the vowel, and nasal if elided, are still perceptible by native listeners because traces of them remain in the speech signal. Indeed, we detected nasalization and rounding in many cases where the suffix was elided, especially within the younger speakers' productions.
There are parallels here with the phenomenon of rhoticity attrition in Scottish English, whereby articulatory gestures may remain even in cases where they were not audible or observable acoustically. Using ultrasound, Lawson et al. (2014) discovered that words produced without an audible /r/ were often nevertheless produced with noticeable /r/ gestures. This suggested that speakers were continuing to produce /r/ speech gestures although they were no longer articulating them sufficiently in order to produce an audible /r/ signal or a visible spectrographic trace. The situation seems similar with Pitjantjatjara utterance-final vowel and nasal deletion in past tense suffixes (although we are reporting on cases with spectrographic traces). Speakers appear to often still be articulating the nasal and vowel gestures. These articulatory gestures are often producing detectable traces with nasalization and rounding noticeable in the audio signal. It is very possible that speakers are still producing these gestures even when they leave no audible or visible traces, though this would be a matter for another study using video recordings and/or articulatory analysis such as electromagnetic articulography (EMA) and/or analysis of nasal airflow with a nasal-oral airflow mask.

Conclusions
Word-final vowel deletion is a pervasive property of Pitjantjatjara connected speech. The extension of vowel deletion to a broader range of contexts is strongly associated with younger speakers and this is acknowledged as an ongoing change within the language. In this study, we examined the spontaneous productions of utterance-final past tense suffixes /-nu, -ïu, -Nu/ by four older and four younger Pitjantjatjara speakers. Elision in this context is not salient to native listeners who in fact were noted to report hearing 'phantom vowels' when vowels were elided in this context. Our results show that both older and younger speakers elide these vowels in approximately 40-50% of cases. Speakers in both groups produced highly gradient and variable realizations of the nasal (fully realized, approximated, denasalized, devoiced, shortened, shifted to alveolar, or absent) and of the vowel (fully realized, shortened, devoiced, absent). In many cases, when the suffix was elided there remained traces of nasalization and rounding on the final segment of the verb stem.
While both age groups showed similar rates of elision, the contexts of elision varied significantly between the two groups. Older speakers tended towards elision when the suffix was in an unfooted syllable, but only if the nasal was velar. In contrast, younger speakers were not influenced by metrical structure and rather tended towards elision whenever the nasal was velar. Thus, the patterns of elision by younger speakers appear in line with a reanalysis of the conditions for elision, but this reanalysis is nonetheless based on patterns found in the productions of older speakers. In the context of utterance-final past tense suffixes, both systems produce a similar rate of elision. However, the tendency for younger speakers to elide vowels in footed as well as unfooted final syllables is in line with a noticeable extension of vowel elision in lexical stems. This increase in vowel elision is likely one of the more perceptually salient differences commented upon by older speakers as part of ongoing language change. Along with the linguistic aims outlined above, we hope that this research can help to inform the community about how variation already present in Traditional Pitjantjatjara may have been one factor that has influenced this perceived change in the language.  Informed Consent Statement: Informed consent was obtained from all participants involved in the study.

Data Availability Statement:
Recordings cited here will be archived before the end of the current project with the Pacific and Regional Archive for Digital Sources in Endangered Cultures (PAR-ADISEC).