DRAFT - Please cite the published version Sheísmo in Montevideo Spanish: Not (yet) identical to Buenos Aires Jim Michnowicz & Lucía Planchón North Carolina State University Abstract Studies have documented an ongoing change from /ʒ/ to /ʃ/ in Rioplatense Spanish, and research indicates that the change to /ʃ/ is complete for young speakers of Buenos Aires (BA) Spanish. However, sheísmo in the neighboring country of Uruguay has not been thoroughly studied. The present study finds that, unlike in BA, the change to /ʃ/ is not yet complete in Montevideo, as determined by persistent sex differences among young speakers (Chang, 2008; Cameron; 2011), and differences in voicing rates between /ʒ/~/ʃ/and phonologically voiceless /s/, indicating that observed voicing is not to due solely to gestural overlap (Rohena-Madrazo, 2015). Uruguay is at least one generation behind BA for this change, distinguishing the Spanish spoken in the two regions. Keywords: Rioplatense Spanish; Montevideo; Sheísmo; Sociolinguistics 1. Rioplatense Spanish1 The varieties of Spanish spoken in urban centers around the Río de la Plata form a macrodialect area, known as Rioplatense Spanish. These varieties, which include the urban areas of both Buenos Aires, Argentina (BA) and Montevideo, Uruguay, share many linguistic features that serve to distinguish the Spanish spoken in the region (Correa & Rebollo Couto, 2012). 1 We would like to thank the audience at HLS 2018, as well as the two anonymous reviewers, for their feedback on this study. All errors remain ours. Data collection was partially funded by a Graduate Research Grant from Sigma Delta Pi, the National Collegiate Hispanic Honor Society. 1 Figure 1. Map of the Rioplatense dialect region, including the urban centers of Buenos Aires and Montevideo. Source: https://www.openstreetmap.org/#map=9/-34.5689/-57.3899 Among the features that characterize Rioplatense Spanish are the predominance of vos in both oral and written language, coda /s/ weakening, as well as intonational and lexical differences that distinguish the region from the rest of Latin America (Lipski, 2004; Colantoni & Gurlekian, 2004; Weyers, 2009). Among the most frequently cited linguistic traits of Rioplatense Spanish is the presence of assibilated palatal fricatives, referred to as zheísmo or sheísmo, according to whether the palatal sibilant is voiced [ʒ] or voiceless [ʃ] (Wolf & Jiménez, 1979; Fontanella de Weinberg, 1983). The (de)voicing of /ʒ/ is the focus of the present study2. Importantly, Rioplatense Spanish is often assumed, implicitly or explicitly, to be synonymous with the Porteño Spanish spoken in BA. Historically, it is likely that BA was the nucleus of many of the linguistic changes that characterize Rioplatense Spanish (Colantoni, 2006; LangRigal, 2015), and there is no doubt that the varieties of Spanish spoken in BA and across the river in Montevideo are indeed very similar. Lipski (2004, p. 369) states “...[e]n un cierto sentido, el español de Uruguay es una mera extensión del habla porteña de Buenos Aires”, 2 In the rest of this study, we will follow recent previous work by Rohena-Madrazo (2015), Lang-Rigal (2015) and others in representing this phoneme as /ʒ/ for speakers of Montevideo and BAS where appropriate. 2 noting that even many inhabitants of the two cities cannot tell if someone is from BA or Montevideo based solely on linguistic factors. While we do not debate the veracity of this and similar statements grosso modo, the assumption that the Spanish of the two major Rioplatense cities is identical has rarely been examined in the literature. Therefore, one of the goals of the present investigation is to determine how Montevideo patterns vis-à-vis published accounts on BA Spanish (BAS) of one of the most characteristic linguistic traits of Rioplatense Spanish, the widely reported shift from zheísmo to sheísmo. Specifically, the shift from voiced /ʒ/ to voiceless /ʃ/ has been reported as being a completed change in BA, at least for the middle classes (Chang, 2008; Rohena-Madrazo, 2013). At the same time, there have been relatively few studies on palatal (de)voicing in Uruguayan Spanish, especially with the goal of comparing results from Montevideo with those of published studies on BA. The present study examines the social and linguistic factors that condition the use of /ʒ/~/ʃ/ in and around Montevideo, as a preliminary step in determining if the change can also be considered complete in Rioplatense Uruguayan Spanish, as has been reported for BAS. The rest of the article is presented as follows. In section 2, we review the literature on zheísmo/sheísmo in Rioplatense Spanish, with a particular focus on the role of speaker sex and age in determining patterns of use in BAS. We will also outline two methods employed in previous studies to determine the completion of a sound change, specifically through social (Cameron, 2011; Chang, 2008) or phonetic comparisons (Rohena-Madrazo, 2013). In section 3 we detail the methodology employed, and section 4 presents results. Sections 5 and 6 include discussion and final conclusions, respectively. 2. Zheísmo and sheísmo in Rioplatense Spanish 3 Both zheísmo and sheísmo refer to the pronunciation of /ʝ/ as a sibilant fricative, either voiced [ʒ] or voiceless [ʃ]. For example, calle ‘street’ may be pronounced as either [ˈka.ʒe] or [ˈka.ʃe]. Rioplatense assibilated palatals represent a variety of yeísmo, without distinguishing between orthographic “y” and “ll” (Canale & Coll, 2016). Assibilated palatals arose at various times in the development of spoken Latin to Spanish (Penny, 2002), and assibilation occurs sporadically in many modern dialects of Spanish (Lipski, 2004). It is in the Rioplatense region, however, that sibilant palatals represent the most common pronunciation of /ʝ/, and are widely considered to be the most indicative pronunciation of the region for Spanish speakers around the world (Lipski, 2004, p. 192). The assibilation of /ʝ/ is believed to have begun in BA sometime between the 18th and 19th centuries, and to have undergone a rapid expansion during the second half of the 20th century (Fontanella de Weinberg, 1973). Assibilated palatals are currently undergoing diffusion from BA to other regions of the country (Colantoni, 2006; Lang-Ridal; 2015). Studies on /ʒ/ in Argentinian Spanish have consistently reported a systematic devoicing to [ʃ], with younger speakers and women in particular leading the change (Wolf & Jiménez, 1979; Fontanella de Weinberg, 1983; Chang, 2008; Rohena-Madrazo, 2015; Lang-Ridal, 2015; among others). Similar trends have been reported in the few studies on Uruguayan /ʒ/ (Winkler, 1998; Barrios, 2002). While most studies have interpreted these patterns of more devoicing among women and younger speakers as evidence of language change in progress, Wolf (1984) argues that the pattern in BA was one of stable variation at the time of publication, with persistent social class differences that, “[m]uy probablemente, jamás lleve a un cambio lingüístico” (p. 185). More recent studies, however, have confirmed that a change is underway, and has reached completion for most social groups (Chang, 2008; Rohena-Madrazo, 2015). 4 Early studies on this phenomenon show a consistent effect of social factors on sibilant (de)voicing in BAS, with speaker age and sex offering the strongest correlations3. Examples of these early patterns, with women leading the change to /ʃ/ and men trailing by at least a generation, are seen in Figures 2 and 3, from Wolf & Jiménez (1979) and Fontanella de Weinberg (1983), respectively. Figure 2. Devoicing according to speaker sex and age in BAS. Adapted from Wolf & Jiménez (1979, pp. 129, 143). 3 Social class also plays an important, albeit secondary role, especially in early studies of (de)voicing. As noted in Wolf & Jiménez (1979, p. 129), “Es evidente que la correlación de las variables edad y sexo nos proporciona las más significativas conclusiones de esta investigación…” (emphasis in the original). For that reason, along with the role of sex and age in determining the completion of a sound change (Cameron, 2011), we focus on age and sex here. 5 Figure 3. (De)voicing of /ʒ/ by sex, age and education level (1=lowest). On the y-axis, a score of 100 = 100% voiced tokens; 200 = an intermediate, partially devoiced realization; 300=100% voiceless tokens (Fontanella de Weinberg, 1983, p. 100). Graph taken from Fontanella de Weinberg (1983, p. 112). Both Figures 2 and 3 show the same overall pattern, with men trailing behind women in the devoicing of /ʒ/. This trend is particularly striking in Figure 3, where Fontanella de Weinberg (1983) found that the behavior of men between the ages of 15-30 mirrors that of women between the ages of 31-70; at the time of data collection almost 40 years ago, men in BA were a full generation behind women. Wolf & Jiménez (1979) show the same basic pattern, with the youngest men producing a similar rate of devoicing (24%) to that of middle-aged women (30%). Importantly, some researchers have singled out sex differences in the frequency of an innovative form as an indicator of an on-going, incomplete change. Labov (2001) shows how sex 6 differences are small or non-existent at the beginning of a change, since everyone is using the same, more conservative form. Sex differences increase at the midpoint of a change, and then again decrease as the change nears completion, as the community converges on the new, innovative form (pp. 307-308). Cameron (2011) applies these concepts to palatal (de)voicing in BA, using data from Wolf & Jiménez (1979). Cameron’s (2011) analysis demonstrates how sex differences in (de)voicing are greatest among the middle age groups, and that these differences are reduced among the youngest speakers, as males steadily catch up to the devoicing rates of females (Figure 4). This trend leads Cameron (2011) to hypothesize that, at the time of Wolf & Jiménez (1979), the change from /ʒ/ to /ʃ/ was a mid-stage, and he predicts that, if the trend has continued to the present day, “...[t]he youngest speakers in present-day BA would show little or no gender differentiation...whereas those who are somewhere near or above 40 years of age would show continuing gender difference” (p. 224). In this way, the presence or absence of sex differences for an innovative form can indicate at what stage a change is, and if it can be considered to have been completed. 7 Figure 4. Sex differences for devoicing across age groups in BA. Data for the graph adapted from Table 10.4 in Cameron (2011, p. 224), which is originally based off of data from Wolf & Jiménez (1979). The lack of important sex differences among younger speakers, as predicted by Cameron (2011), is in fact exactly what is reported in more recent studies on sheísmo in BAS. Link (2009) found identical rates of devoicing for young men and women (ages 16-30), with both groups devoicing at 96% in a reading task. Lang-Rigal (2015) compared the speech of men and women across three cities in Argentina: BA, Tucumán and Córdoba. She finds that BA shows the most advanced stage of palatal devoicing among the three cities, and that there are no significant differences between men and women in that city. Tucumán, on the other hand, continued to show a strong gender divide, similar to that seen for BA in previous studies, suggesting that this city is at an earlier stage of the change to /ʃ/ (Lang-Rigal, 2015, p. 160). Córdoba in general lagged behind the other two cities, with no important gender differences, but much lower rates of devoicing, suggesting that Córdoba may not be fully participating in the change (p. 161). Additionally, Chang (2008), in his study of BAS likewise found no significant differences 8 among the younger men and women that make up his study, as speakers of both sexes produced almost categorically 100% devoiced tokens. Chang (2008, p. 62) observes “...although palatal devoicing was found to be correlated with age, it no longer seems to be correlated with gender...this result suggests that the change from [ʒ] to [ʃ], rather than being an ongoing change, is better characterized as a change that has already been completed”. Taken together, these studies indicate that the change to /ʃ/ in BA can be considered complete, at least for middle class speakers, as patterns of devoicing no longer show differentiation by sex for younger speakers. In the present study, we will likewise use sex differences as a diagnostic for comparing patterns of (de)voicing in Montevideo Spanish. In a recent study, Rohena-Madrazo (2015) addresses the question of sound change completion in BAS by examining phonetic variation at the individual speaker level. RohenaMadrazo (2015) notes the difficulty in determining the status of a sibilant fricative as either voiced or voiceless at the phonemic level, given that such phones regularly undergo phonetic voicing assimilation due to surrounding phonetic context. In other words, a phonologically unvoiced segment can be phonetically voiced to one degree or another when it occurs between two voiced segments, such as vowels. This process of phonetic voicing has been frequently found to affect uncontroversially voiceless sibilant phonemes, such as /s/ (Schmidt & Willis, 2011; and sources therein). Rohena-Madrazo (2015, p. 292) states “If /ʒ/ were to be consistently produced as 100% voiceless, then there would be no doubt that the devoicing change has been completed. However, the question remains, does the completion of the devoicing change necessarily imply or require 100% voicelessness in all utterances of /ʒ/?”. Rohena-Madrazo (2015) therefore uses the phonetic voicing of /s/ as a baseline for determining the phonological status of /ʒ/ and /ʃ/. If the palatal sibilant fricative shows voicing at 9 the same rate or less than /s/, then voicing is due to phonetic factors of coarticulation and gestural overlap, and an underlying /ʃ/ can be posited. The change would be considered complete. If, on the other hand, the palatal sibilant shows greater levels of voicing than voiceless /s/, it would be interpreted as evidence that the production of voiced [ʒ] is not due to phonetic factors alone, and instead reflects an underlying voiced phoneme /ʒ/. Rohena-Madrazo (2015) calls speakers whose voicing rates of /ʒ/~/ʃ/ are similar to their voicing of /s/ “devoicers”, while speakers who voice the palatal fricative at a rate greater than /s/ - e.g. at a rate greater than would be expected in the presence of phonetic voicing alone - are “voicers”. In an initial analysis, Rohena-Madrazo (2013) finds that the change to /ʃ/ is complete for young, middle-class speakers, since there were no significant differences in the voicing of /s/ and /ʒ/~/ʃ/ for this group. He does find, however, that upper-class young speakers still maintain higher rates of voicing for /ʒ/~/ʃ/ than for /s/, indicating that the change was not complete for speakers of that group, as their voicing of the palatal sibilant cannot be explained by phonetic processes alone. In his extended analysis of the data, however, Rohena-Madrazo (2015, p. 308) shows that the apparent preservation of voiced /ʒ/ among young upper-class speakers is due to the behavior of one speaker, a young woman “voicer” that lives “in a very luxurious high-rise building, in which each floor had only one apartment—this exceeds conventional standards of poshness even for La Recoleta [an upper class neighborhood in north BA]” (p. 310). RohenaMadrazo (2015) notes that more detailed ethnographic studies are warranted to further explore the motivations for some speakers to produce voiced forms in a way incongruent with their agematched peers, but overall characterizes the change to /ʃ/ as complete for most younger speakers, and in its final stages overall in BA, “perhaps with some correction from above (Labov, 1966:225)” (p. 312). Importantly, the change was found to be complete for lower- and middle- 10 class speakers, the same social groups represented in the present study on Uruguayan Spanish (see table 1). In this study, we also compare each speaker’s percent voicing of /s/ and /ʒ/~/ʃ/ as another criterion for determining the status of (de)voicing in Montevideo. Palatal (de)voicing in Uruguay has received much less attention from scholars, as noted previously, with only a handful of studies examining the possible shift to /ʃ/, and even fewer examining the phenomenon from within a variationist paradigm. Lipski (2004) observes that devoicing is not as common in Montevideo as in BA, but that Uruguayan Spanish is quickly catching up in the production of [ʃ]. Winkler (1998) reports that /ʒ/ remained the dominant form in Montevideo at the time of publication, but as in other studies, notes that devoiced variants are more common among women and younger speakers. Thun & Elizaincín (2000), using data from the Atlas diatópico y diastrático del Uruguay, report that younger speakers show increased devoicing, as has also been found for BA. Barrios (2002), in her analysis of sociolinguistic interviews, also finds increased sheísmo among younger speakers, and evidence of sex differences among older speakers indicate that, again like in BA, the change is led by women. She notes fewer differences based on speaker sex among young speakers. Fernández Trinidad (2010) presents an acoustic analysis of five female speakers of Rioplatense Spanish, from both BA and Montevideo. Interestingly, and underlying the assumed equivalence of BA and Montevideo in the literature, Ferández Trinidad (2010) makes no distinction between participants from the two cities, stating “[a] pesar del hecho de que pertenezcan a dos países diferentes, (Argentina y Uruguay, respectivamente) en el habla de las capitales se reconocen muy pocas diferencias diatópicas, al punto que dialectalmente constituyen una única unidad lingüística, el español platense o rioplatense” (p. 269). While her study focuses on describing the fine-grained acoustic correlates of voiced and voiceless variants, where differences between the two cities 11 might not be expected, the commonly held idea that the same process in both countries is identical has not been tested in the literature to the best of our knowledge. Based on patterns of (de)voicing reported in the previous literature, as well as both social and phonetic criteria for determining the completion of a sound change, the present study constitutes a preliminary analysis of palatal (de)voicing in and around Montevideo as compared to previous reports of BAS. Specifically, we seek to answer the following research questions: RQ1: What are the social and linguistic factors that influence (de)voicing of the palatal fricative /ʒ/~/ʃ/ in the Spanish of Montevideo, Uruguay and surrounding areas? RQ2: How does Montevideo compare to previous studies on BA? Has the change from /ʒ/ to /ʃ/ been completed in Montevideo, as determined by social and phonetic criteria, and as reported for BA? 3. Methodology 3.1. Participants: The present study is a semi-replica of Chang (2008), which investigated the speech of Argentinians in BA using a reading task that consisted of a series of Mafalda comic strips by Quino. There were 24 total participants (10 men, 14 women) who completed the task for the present study. Participants were recruited using familial connections, which included friends, coworkers, and clients of family. 18 participants were long term residents or natives of Montevideo, while three participants lived in Nueva Helvecia, Colonia during the time of the recordings, and three others were from Nueva Helvecia, but had been living in Montevideo for a 12 short time4. For purposes of the study, participants were divided into three age groups (older: age 54-86; middle: age 32-49; and younger: age 24-30). Demographic data (including sex, age, age group, residency, and level of education) for the participants can be found in Table 1. Participa nt Sex Age Age Group Residence Education AMWGR M 86 Older Montevideo Elementary AFWAR F 77 Older Montevideo High School AMWSP M 67 Older Montevideo University AMBOA M 66 Older Nueva Helvecia Some High School AFWCS F 65 Older Montevideo Masters AFBMI F 62 Older Nueva Helvecia University AFBJP F 57 Older Montevideo University AFBSC F 54 Older Montevideo Some High School MFBST F 49 Middle Montevideo High School MMBJC M 45 Middle Montevideo Elementary MMWGI M 43 Middle Montevideo University MFBPR F 36 Middle Montevideo Some High School MFBIN F 35 Middle Nueva Helvecia University MMWSP M 35 Middle Montevideo Masters MFBPI F 34 Middle Montevideo University MFWLY F 33 Middle Montevideo University MMWD V M 32 Middle Montevideo University 4 Nueva Helvecia, Colonia, is on the southern coast of Uruguay, across the Río de la Plata from Buenos Aires. See the map in Figure 1. As described below, there was no main effect of residence in the analysis, so all speakers were kept together in the rest of the study. 13 JFWAG F 30 Younger Montevideo University JFBJN F 28 Younger Montevideo High School JFWSB F 28 Younger Montevideo University JMBJI M 27 Younger Montevideo High School JMWSG M 27 Younger Montevideo University JMWMG M 26 Younger Montevideo University JFWMN F 24 Younger Montevideo High School Table 1: Information of Participants, in order of age 3.2. Analysis Before beginning the reading task, participants completed a demographic questionnaire. Participants were then given the reading task, consisting of thirty-two comic strips from Toda Mafalda (Quino) picked deliberately to encourage palatal production, following Chang (2008). Using the comic strips allowed for the speech to be conversational and more comfortable, as the participants saw familiar lexicon from a storyline that they had encountered prior to the investigation. The participants were recorded on an H2 Zoom digital recorder. Participants were asked to read as naturally as possible, at a normal reading pace. The recordings lasted between nine and twenty minutes, and due to the satirical nature of the comic strip, participants were encouraged to react as they normally would, which resulted in some tokens being disregarded due to laughter, etc. Overall, the comic reading task produced a range of between 65 and 75 tokens for most participants, for a total of 1628 instances of palatal fricatives5. All data were collected by the second author, a native speaker of Montevideo Spanish. 5 One participant only completed part of the task, producing a total of 27 tokens. 14 Tokens were manually identified and segmented in Praat (Boersma & Weenink, 2018), and each token was identified categorically, according to the waveform, spectrogram, presence or absence of a voicing bar and glottal pulses. This initial categorical coding was done to allow for a global view of the variants of /ʒ/ available to speakers in and around Montevideo, including non-sibilant variants, which appeared sporadically in the data (Table 2). The primary focus, however, consisted of an acoustic analysis of percent voicing of each token, carried out with a Praat script. Following File-Muriel & Brown (2011), the script automatically adjusted the window size so that the fricative token occupied precisely two-thirds of the analysis window, since the level of zoom can impact the measure of voicing. The script then utilized the Praat Voice Report to determine the percent voicing of each token6. Examples of token coding and delimitation showing both voiced and voiceless tokens is found in figure 5. Figure 5. Voiced (left) and voiceless (right) tokens of /ʒ/ in the word billetes. 6 Gradoville (2011) examines the validity of several acoustic measures of sibilant voicing, and determines that the Praat Voice Report (along with low-frequency-to-total intensities) provided “the best match for what can be observed in the spectrogram and auditorily” (p. 71), although he did note problems with “phantom pulses” (p. 70) that can cause the Voice Report to mistake periodic noise in the fricative as glottal pulses. Unfortunately, the solution (setting the pitch setting to 250Hz max) also resulted in the exclusion of data for one speaker in that study. We have opted to script the use of the Voice Report here for consistency, to address the problem of analysis window zoom, and ease of comparison with other studies, although any potential shortcomings must be kept in mind. 15 Percent voicing as a continuous variable formed the basis for the quantitative analysis detailed in the results7. A series of mixed-effects linear regression models were fitted to the data using lme4 (Bates, Maechler, Bolker & Walker, 2015) in R (R Core Team, 2017), with %Voiced as the continuous dependent variable, and independent variables of speaker sex, age group (older, middle, younger), place of residence (Montevideo vs. Nueva Helvecia), as well as the following linguistic factors to control for context: position in the word (initial vs. medial), grapheme (“y” vs. “ll”), and syllable stress (tonic vs. atonic). Given that all but two of the participants have high school or greater education, education level was not included in the statistical models. Speaker and word were included as random intercepts. P-values were computed with lmerTest (Kuznetsova, Brockhoff & Christensen, 2017). Finally, in order to compare participants voicing rates for /ʒ/ and /s/, approximately 50 tokens of intervocalic /s/ were extracted and analyzed for percent voicing from each recording, using the same script as for /ʒ/. A total of 1231 tokens of /s/ were analyzed, with the following distribution: 819 in atonic syllables, 412 in tonic syllables, and 322, 482 and 427 tokens in word initial, medial, and final position, respectively. 4. Results 4.1 Variants of /ʝ/ Table 2 shows the results of the categorical analysis, undertaken to document the variants available to speakers in Montevideo. Variant n % of data Affricate 5 0.31% Approximant 2 0.12% 7 An anonymous reviewer points out that duration can be a stronger cue for fricative voicing than glottal pulses; an additional mixed effects analysis with duration as the dependent variable found syllable stress to the be only significant predictor (p.0.43). Since no social factors significantly predicted differences in duration, the analysis will proceed with percent voicing. 16 Palatal lateral 1 0.06% Non-sibilant fricative 20 1.23% Voiceless sibilant fricative 1274 78.3% Voiced sibilant fricative 326 20% Total 1628 100% Table 2. Categorical distribution of variants The categorical results indicate that speakers produced very few variants other than [ʒ] or [ʃ], as the two sibilant fricatives account for 98% of the data. This is similar to the findings of Rohena-Madrazo (2015) for BA, where sporadic [ʎ], [dʒ] and [j] were also found, as well as the data from the Atlas diatópico y diastrático del Uruguay (ADDU) (Thun y Elizaincín, 2000). It is clear from these results, however, that the postalveolar sibilant fricatives [ʃ] and [ʒ] are by far the dominant forms in Montevideo Spanish. 4.2. Analysis of voicing Results of the best mixed-effects linear regression analysis are found in Table 3. We began with a maximal model (including all possible main effects and two-way interactions), and undertook model selection using anova() in R. Estimate Std. Error t-value p-value Intercept 0.20815 0.06474 3.215 0.003264 ** Sex = Male 0.20257 0.06163 3.287 0.003123 ** Age Group = Middle (ref=Younger) 0.15608 0.07474 2.008 0.047685 * Age Group = Older (ref=Younger) 0.33027 0.07697 4.291 0.000254 *** Stress = Tonic -0.07228 0.02839 -2.546 0.013754 * 17 Table 3. Results of best mixed-effects linear regression. Speaker and Word as random factors. n=1628. Std. Dev. of Speaker 0.14529; Std. Dev. of Word 0.09032. Speaker sex, age group and syllable stress were significant predictors of % voicing. Specifically, men showed significantly higher rates of voicing than women, and middle age and older speakers likewise showed higher voicing rates than younger speakers (significantly so for older speakers, with the difference between younger and middle age speakers marginally significant). An additional analysis with releveled age group shows that the difference between middle age and older speakers is also significant (p=0.024127). There were also significantly lower rates of voicing in tonic syllables. Residence, word position and grapheme did not significantly influence the percentage voicing in the present data. There were no significant interactions. Figure 6. %Voiced by Sex and Age Group. 18 The intersection of the two significant social factors, sex and age group, is clearly visible in Figure 6. Men consistently produced higher voicing rates than women across age groups, with younger women showing the most consistent devoicing (indicated by the median line and the size of the boxplot). A comparison of men’s values with those of women for the preceding generation (i.e. younger men compared with middle aged women, etc.) shows that men are a full generation behind women in the shift from /ʒ/ > /ʃ/. A similar result has been found for BAS (Wolf & Jiménez, 1979; Fontanella de Weinberg, 1983). Importantly, Figure 6 shows that important differences remain between sexes, even among the youngest speakers, which are confirmed in separate mixed-effects regression analyses of each age group, as seen in Table 4. Older (p=0.024579) Middle p=0.340620 Younger p=0.00333 Men 74.2% 46.4% 34.3% Women 43.7% 35.9% 17.1% Table 4. Mean voicing rates and significance for speaker sex across age groups. The differences observed in Table 4 again demonstrate that men are a full generation behind women in the move to /ʃ/, as men’s voicing rates correspond to those of women of the previous generation. The difference in voicing between men and women is significant for older and younger speakers. The middle age group shows more variation and overlap in voicing values, as seen in Figure 6, leading to a non-significant difference in the mixed-effects model, in spite of the difference in mean voicing. Overall, the differences based on speaker sex could have implications for determining if the change from /ʒ/ > /ʃ/ is complete in Montevideo (Chang, 2008; Cameron, 2011). We turn our attention to determining the status of completion of this change, as well as how Montevideo compares to BA, in the next section. 19 4.3. Completion of the change in Montevideo and comparisons with BA In this section, we examine the results according to the criteria for determining the completion of a sound change suggested by Chang (2008), Cameron (2011), and RohenaMadrazo (2015). To recap, Chang (2008) and Cameron (2011) base their determination on the social distribution of [ʃ], where the lack of significant sex differences indicates that the sound change is has been completed. Rohena-Madrazo (2015), on the other hand, bases his determination of change completion on a comparison of voicing rates of /ʒ/~/ʃ/ with the baseline phonetic voicing of /s/. If voicing rates for /ʒ/~/ʃ/ are equal to or lower than for phonologically unvoiced /s/, any residual voicing is due to phonetic assimilation and gestural overlap, and the change to /ʃ/ is posited to be complete. 4.3.1. Sex differences as a criterion for completed sound change Following Cameron (2011) and Chang (2008), we interpret significant differences in (de)voicing of /ʒ/ as evidence of an ongoing change in Montevideo. As seen in Table 3, sex was a significant factor in the analysis of %Voiced in Montevideo Spanish. Figure 6 shows that women are a full generation ahead of men in the devoicing of /ʒ/, and that sex differences remain even among the youngest speakers (Table 4). In order to examine this pattern in more detail, a mixed-effects conditional inference tree was run on the data using the lmertree function in the glmertree package in R (Fokkema, Smits, Zeileis, Hothorn & Kelderman, 2018). The package glmertree combines mixed-effects analysis from lme4 (Bates, Maechler, Bolker & Walker, 2015) and the recursive partitioning (trees) from partykit (Hothorn & Zeileis, 2015)8. For purposes of clarity, only social factors were included. 8 See Tagliamonte (2011) and Tagliamonte & Baayen (2012) for more on Conditional Inference Trees. 20 Figure 7. Mixed-effects Conditional Inference Tree of social factors. %Voiced. Starting at node 1, the primary breakpoint in the data is for age group, with younger speakers differing significantly from older and middle age participants. For middle age and older speakers, there is a split for sex (node 2), and a further split by age group for each sex. Interestingly, residence does play a significant, albeit very specific, role in the tree. Among older men, residents of Montevideo show lower percentages of devoiced tokens than their counterparts from Nueva Helvecia. Possible explanations for this pattern will be addressed in the discussion. Still, the most important node for the present discussion is node 11, which shows that significant sex differences are found among the youngest speakers, with women producing lower percentages of voicing than men. In the Uruguayan data, the persistent sex difference among young speakers suggests that, unlike in BA, the change to /ʃ/ is not yet complete in Montevideo. 4.3.2. Comparison with /s/ as a criterion for completed sound change. 21 The percent voicing for /ʒ/ tokens was compared to that of intervocalic /s/, following Rohena-Madrazo’s (2015) diagnostic for determining the completion of the devoicing to /ʃ/. A comparison of /ʒ/ and /s/ tokens by age and sex groups is found in Figure 89. Figure 8. Percent voicing of /ʒ/ and /s/ by age and sex. The only group that shows a higher voicing rate for /ʒ/ than /s/ are older men, suggesting that these speakers maintain a phonologically voiced phoneme. The other groups, however, show voicing rates at or below those of /s/. Following Rohena-Madrazo (2015), this result indicates that the voicing that does occur is due to phonetic factors such as gestural overlap, and therefore the change from /ʒ/ > /ʃ/ has been completed. In this sense, Montevideo is like BA, as both younger men and women have lower rates of voicing for /ʒ/ than for /s/. However, the results in Figure 8 indicate that younger men and women do not (de)voice at equal rates; the boxplot for 9 Overall, a higher rate of voicing for /s/ was found in the present data than in Rohena-Madrazo (2015), although the patterns for both /s/ and /ʒ/ are consistent across both studies. Intervocalic /s/ voicing in Montevideo should be examined further in future studies. 22 younger women bottoms out at 0% voicing, and maxes out at approximately 25% voiced. Younger men, on the other hand, have middle quartiles between 20% and 50% voicing. So, while both groups show the same phonological pattern of devoicing, on a phonetic level they still behave differently. In fact, it is interesting that many younger women do not even allow phonetic voicing of /ʃ/, suppressing voicing rates far below those of /s/. This trend can be clearly seen in an examination of individual voicing patterns, as in Figure 9. Figure 9. Percent voicing of /ʒ/ and /s/ by speaker. Three of the four younger women suppress the voicing of /ʒ/ far below that of phonologically unvoiced /s/, with the fourth producing relatively equal rates of voicing for both phonemes. This suggests that some younger women may be forcibly maintaining a voiceless [ʃ] 23 for social reasons. Two of the three younger men, however, show much more similar voicing rates between /ʒ/ and /s/, and one speaker (JMBJI) maintains voiced /ʒ/ at a greater rate than /s/. So while younger speakers of both sexes behave the same at a phonological level, in that as a group they voice /ʒ/ at a similar or lower rate than /s/, a comparison of the range of voicing in Figure 8 along with individual patterns in Figure 9 suggest that there are still important differences between younger men and women according to this criteria. 5. Discussion The results above allow us to answer our initial research questions, repeated here for convenience. RQ1: What are the social and linguistic factors that influence voicing of the palatal fricative /ʒ/~/ʃ/ in the Spanish of Montevideo, Uruguay and surrounding areas? Three factors were found to be significant predictors of the percentage of (de)voicing in Montevideo: age, sex and syllable stress, with younger speakers, women and tonic syllables all favoring higher rates of devoicing. Women also led the change to /ʃ/ in BA (Chang, 2008; Rohena-Madrazo, 2015), as is commonly found in studies on language change (Labov, 2001). Likewise, younger speakers show higher rates of devoicing in BA, as the change progresses across time (Wolf & Jimenez, 1979; Fontanella de Weinberg, 1983; Chang, 2008; RohenaMadrazo, 2015). The same patterns have also been found in the few studies on palatal (de)voicing in Montevideo (Thun & Elizaincín, 2000; Barrios, 2002; Winkler, 1998). Finally, tonic syllables have been shown to preserve contrasts to a higher degree than atonic syllables (see, for example, Penny, 2002 on the differing behavior of vowels and consonants according to syllable stress in the history of Spanish). Additionally, stressed syllables are overall of longer duration, which likewise makes them more likely to be voiceless; fricatives in tonic syllables 24 were significantly longer than in atonic syllables in the present data (see footnote 9). This trend is reflected here in the higher rates of devoicing in stressed syllables. RQ2: How does Montevideo compare to previous studies on BA? Has the change from /ʒ/ to /ʃ/ been completed in Montevideo, as determined by social and phonetic criteria, and as reported for BA? The results of this study indicate that the change from /ʒ/ > /ʃ/ is well underway in Montevideo Spanish, a result which links this variety with the dialects spoken in and around BA, as coastal regions of Uruguay, including Montevideo, are likely affected by the same inter-urban transmission of /ʃ/ taking place throughout Argentina (Lang-Rigal, 2015). Chang (2008) found 85% devoicing across younger speakers in BA, and Rohena-Madrazo (2015) reports a similar 81% devoicing for middle class younger speakers10. Younger speakers in the present study produced 76% devoicing on average, slightly lower than the results for BA in previous studies. One of the main driving questions behind the present study is whether we can consider Montevideo Spanish to be identical to BAS with respect to sheísmo. Previous studies of BAS have concluded that the change to /ʃ/ is complete, at least for middle class speakers in that dialect, using either social criteria (the lack of sex differences in Chang (2008) and Cameron (2011)), or a comparison with the phonetic voicing of /s/ (Rohena-Madrazo, 2015). At the same time, the few studies on Montevideo Spanish suggest that speakers in this city are following the same trends as in BA (Winkler, 1998; Barrios, 2002). Results indicate that sex differences in devoicing persist among the youngest speakers in Montevideo (Table 4, Figure 7), a finding that distinguishes Montevideo from BA, and supports Lipski’s (2004) observation that Montevideo 10 Upper class younger speakers in Rohena-Madrazo (2015) showed a lower mean devoicing rate, 71%, but individual speaker results show that this difference is due to the productions of one speaker (p. 308). With that speaker removed, younger speakers of both social classes behave virtually identically. 25 may be slightly behind BA in the change to /ʃ/. Likewise, Rohena-Madrazo (2015) finds that all but one of the younger speakers (mentioned in footnote 12) are ‘devoicers’; that is, they display rates of /ʃ/ voicing that are at or below the rates for /s/. While there is considerable variation among older speakers in those studies, younger speakers in BA were found to be remarkably consistent in their production of /ʃ/. A comparison of younger speaker mean voicing rates between the present Montevideo study and Chang’s (2008) BA data clearly demonstrates the differences between the two cities. Figure 10. Percent Voiced: Younger speaker comparison, Montevideo and BA. BA data from Chang (2008). Montevideo speakers on the left (JFWAG-JFWMN); BA speakers on the right (F1-M5). Figure 10 shows the mean percent voicing for younger speakers from the present study, compared to younger speakers from BA in Chang (2008). Montevideo speakers are on the left (speaker codes JFWAG through JFWMN), and Chang’s (2008) BA speakers are on the right (F1 26 through M5). The BA speakers are much more consistent in their (de)voicing, with an average rate of 15% voicing, and a range from 13% to 16%, regardless of sex. The Montevideo speakers, however, show much more variation. Only two speakers voiced at less than 20%, and the average voicing for this group was 24%, nine percent higher than the average rate for BA. Likewise, the range of voicing across Montevideo young speakers is from 11% to 39%, a range of 28%. The three highest voicing rates in Montevideo were all produced by men (speakers JMWSG, JMBJI and JMWMG), again reinforcing the persistent sex differences among younger speakers shown in the analysis in section 5. Likewise, a closer examination of individual patterns for /ʒ/ and /s/ in Figure 9 indicates that, of the 24 Montevideo participants, 15 can be considered ‘devoicers’ in the sense of RohenaMadrazo (2015). Of the nine ‘voicers’, five are men, including all three older men. In sum, results indicate that women are a full generation ahead of men in the adoption of /ʃ/ in Montevideo, a trend that continues among young speakers. Returning then to the question of whether Montevideo and BA are identical with respect to sheísmo, results indicate that the answer is no - at least not yet. Montevideo appears to be a generation behind BA in the adoption of /ʃ/, and in this way behaves similarly to other areas of Argentina, that are receiving the change through diffusion from BA (Lang-Rigal, 2015). A future study should examine interactional patterns between cities in the River Plate region, to determine how the change to /ʃ/ extends from BA to other regions. The present results suggest that geographic proximity may play a role. As seen in the conditional inference tree in figure 7, region (Montevideo vs. Nueva Helvecia) is only a significant factor for older men, with speakers from Nueva Helvecia showing lower voicing rates than their peers from Montevideo, although these results are tentative due to the low number of speakers from outside Montevideo. Nueva 27 Helvecia is a town in the Department of Colonia, the geographically closest part of Uruguay to BA. Future study should examine this possibility in more detail, including the role that transportation routes may play in diffusing linguistic changes from BA to Uruguay (see Trudgill, 1974). Likewise, comparative sociolinguistic analysis of palatal (de)voicing in cities across the Rioplatense region would shed light on the processes of transmission and diffusion that may be taking place with a variety of linguistic variables (see Tagliamonte & Denis, 2014). 7. Conclusions This study has shown that the change from /ʒ/ to /ʃ/ is not yet complete in Montevideo, as the percent (de)voicing in this dialect continues to show significant sex differences among all age groups, a result which distinguishes Montevideo from BA (Chang, 2008). While a comparison of /ʃ/ and /s/ voicing indicates that younger women most likely possess an underlyingly devoiced phoneme /ʃ/, the process is still variable for older speakers and men. Montevideo lags a generation behind BA in this change, suggesting that devoicing began later in Montevideo, as the result of diffusion from BA. These results also suggest that the future of sheísmo in Montevideo is likely identical to that in BA, where /ʃ/ will become essentially the only form within a generation or two, at least among most social groups. At the same time, the possible persistence of subtle phonetic or sociolinguistic differences should be investigated, as well as possible future divergences as the results of sociolinguistic processes. This research also underlines the need for more detailed studies of Rioplatense varieties outside of BA, without simply assuming a priori that all of these regional lects are identical to porteño Spanish. References 28 Barrios, G. (2002). El ensordecimiento del fonema palatal/ž. In G. Barrios y V. Orlando (Eds.). Marcadores sociales del lenguaje. Estudios sobre el español hablado en Montevideo (pp. 29-42). Montevideo: Gráficos del Sur. Bates, D., Maechler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software, 67(1), 1-48. doi:10.18637/jss.v067.i01. Boersma, P., & Weenink, D. (2015). Praat: doing phonetics by computer[Computer program] Version 5.4.22.http://www.praat.org/. Cameron, R. (2011). Aging, age, and sociolinguistics. In Manuel Díaz-Campos (Ed.). The Handbook of Hispanic Sociolinguistics (pp. 207-229). West Sussex, UK: Blackwell. Canale, G., & Coll, M. (2016). Historia y presente del yeísmo (rehilado) en el Uruguay. Lexis, 40(1), 5-40. Chang, C. B. (2008). Variation in palatal production in Buenos Aires Spanish. In M. Westmoreland & J. A. Thomas (Eds.). Selected proceedings of the 4th Workshop on Spanish Sociolinguistics (pp. 54-63). Somerville: Cascadilla Proceedings Project. Colantoni, L. (2006). Micro and macro sound variation and change in Argentine Spanish. In N. Sagarra & A.J. Toribio (Eds.). Selected proceedings of the 9th Hispanic Linguistics Symposium (pp. 91-102). Somerville, MA: Cascadilla Proceedings Project. Colantoni, L., & Gurlekian, J. (2004). Convergence and intonation: historical evidence from Buenos Aires Spanish. Bilingualism: Language and cognition, 7(2), 107-119. Correa, P., & Rebollo Couto, L. (2012). Sociolingüística rioplatense: principales fenómenos en variación. Español actual, 161, 213. Fernández Trinidad, M. (2010). Variaciones fonéticas del yeísmo: un estudio acústico en mujeres rioplatenses. Estudios de fonética experimental, 19, 263-292. 29 File-Muriel, R. J., & Brown, E. K. (2011). The gradient nature of s-lenition in Caleño Spanish. Language Variation and Change, 23(2), 223-243. Fokkema M, Smits N, Zeileis A, Hothorn T, Kelderman H (2018). “Detecting TreatmentSubgroup Interactions in Clustered Data with Generalized Linear Mixed-Effects Model Trees”. Behavior Research Methods, 50(5), 2016-2034. Fontanella de Weinberg, M. B. (1973). El rehilamiento bonaerense a fines del siglo XVIII. Thesaurus. Boletín del Instituto Caro y Cuervo, 18, 1-8. Fontanella de Weinberg, M. B. (1983). Variación y cambio lingüístico en el español bonaerense. LEA: Lingüística española actual, 5(1), 93-112. Gradoville, M. S. (2011). Validity in Measurements of Fricative Voicing: Evidence from Argentine Spanish. In S.M. Alvord (Ed.). Selected Proceedings of the 5th Conference on Laboratory Approaches to Romance Phonology (pp. 59-74). Somerville, MA: Cascadilla Proceedings Project. Hothorn, T., & Zeileis, A. (2015). partykit: A Modular Toolkit for Recursive Partytioning in R. Journal of Machine Learning Research, 16, 3905-3909. URL http://jmlr.org/papers/v16/hothorn15a.html Kuznetsova A, Brockhoff PB and Christensen RHB (2017). “lmerTest Package: Tests in Linear Mixed Effects Models.” Journal of Statistical Software, 82(13), pp. 1-26. doi: 10.18637/jss.v082.i13 (URL: http://doi.org/10.18637/jss.v082.i13). Labov, W. (2001). Principles of language change: Social factors. Malden, MA: Blackwell. Labov, W. (1966). The social stratification of English in New York City. 1st ed. Washington, DC: Center for Applied Linguistics. Lang-Rigal, J. (2015). Regional variation in the devoicing of the alveopalatal fricative in Argentina. Estudios de Fonética Experimental, 24(2015), 141-170. 30 Link, K. (2009). El yeísmo bonaerense–un análisis sociolingüístico. Retrieved from http://www.meta.narr.de/9783823368540/musterhausarbeit_235.pdf Lipski, J. M. (2004). El Español de América. Madrid: Cátedra. Penny, R. (2002). A history of the Spanish language. Cambridge University Press. Quino (2007). Toda Mafalda, 20th edition. Buenos Aires, Argentina: Ediciones de la Flor. Rohena-Madrazo, M. (2013). Variación y cambio de sonoridad de la fricativa postalveolar del español de Buenos Aires. In L. Colantoni & C. Rodríguez Louro (Eds.). Perspectivas teóricas y experimentales sobre el español de la Argentina, (pp. 37-57). Lingüística iberoamericana. Madrid : Frankfurt am Main: Iberoamericana/Vervuert. Rohena-Madrazo, M. (2015). Diagnosing the completion of a sound change: Phonetic and phonological evidence for /ʃ/ in Buenos Aires Spanish. Language Variation and Change, 27(3), 287-317. Schmidt, L. B., & Willis, E. W. (2011). Systematic investigation of voicing assimilation of Spanish/s/in Mexico City. In S.M. Alvord (Ed.). Selected proceedings of the 5th Conference on Laboratory Approaches to Romance Phonology (pp. 1-20). Somerville, MA: Cascadilla Proceedings Project. Tagliamonte, S. (2011). Variationist sociolinguistics: Change, observation, interpretation. Oxford: Wiley-Blackwell. Tagliamonte, S. A., & Baayen, R. H. (2012). Models, forests, and trees of York English: Was/were variation as a case study for statistical practice. Language variation and change, 24(2), 135-178. Tagliamonte, S. A., & Denis, D. (2014). Expanding the transmission/diffusion dichotomy: Evidence from Canada. Language, 90(1), 90-136. 31 Thun, H., & Elizaincín, A. (2000). Atlas lingüístico Diatópico y Diastrático del Uruguay (ADDU). Vol. I: Consonantismo y vocalismo del español. Fasc. A.1: Lateral palatal y fricativa medipalatal: lleísmo, yeísmo,ʒeísmo y ʃeísmo en el español uruguayo. Kiel: Westensee-Verlag. R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/. Trudgill, P. (1974). Linguistic change and diffusion: Description and explanation in sociolinguistic dialect geography. Language in society, 3(2), 215-246. Weyers, J. R. (2009). The impending demise of tú in Montevideo, Uruguay. Hispania, 829-839. Winkler, Juan. (1998). “El fonema /ž/ en el habla de Montevideo”. Anuario de Lingüística Hispánica. 14, 517-530. Wolf, C. (1984). Tiempo real y tiempo aparente en el estudio de una variación lingüística: ensordecimiento y sonorización del yeísmo porteño. In L. Schwartz & I. Lerner (Eds.), Homenaje a Ana María Barrenechea (pp. 175-196). Madrid: Castalia. Wolf, C., & Jiménez, E. (1979). El ensordecimiento del yeísmo porteño, un cambio fonológico en marcha. In A. María Barrenechea (Ed.). Estudios lingüísticos y dialectológicos: Temas Hispánicos (pp. 115-145). Buenos Aires: Hachette. 32