languages

Article Men Finally Got It! Rhotic Assibilation in Mexican Spanish in Chihuahua

Natalia Mazzaro * and Raquel González de Anda

Department of Languages and Linguistics, University of Texas at El Paso, El Paso, TX 79968, USA; [email protected] * Correspondence: [email protected]



Received: 17 August 2020; Accepted: 3 October 2020; Published: 14 October 2020


Abstract: Rhotic assibilation is a common sociolinguistic variable observed in different Spanish speaking countries such as Argentina, Ecuador, and México. Previous studies reported that rhotic assibilation alternates with the flap and/or with the trill. In this study,we explore three aspects of rhotic assibilation in the Spanish of the state of Chihuahua, Mexico/El Paso, TX, United States: (1) Its diachronic development; (2) the linguistic and social factors that affect this variation and; (3) the possible effect of contact with English in this variable. Fifty-eight participants, including Spanish monolingual and Spanish-English bilingual subjects, performed one formal and two semi-informal speech production tasks. Acoustic and perceptual analysis of the tokens showed that the variation is not binary (standard vs. non-standard variant), but that it includes other rhotic variants with varying degrees of frication. Variation is restricted to phrase-final position and heavily favored by preceding front vowels (/e/ and /i/). These effects have a clear aerodynamic and articulatory motivation. Rhotic assibilation is not receding, as previously reported. It continues to be a prestigious variable prevalent amongst females, but also present in male speakers. The comparison between bilingual and monolingual speakers shows that contact with English does not significantly affect the occurrence of assibilation.

Keywords: rhotic assibilation; sociolinguistics; sociophonetics; Mexican Spanish; variation

1. Introduction Rhotic assibilation is a pervasive feature of most varieties of Spanish. It has been reported in Argentina (Colantoni 2006); Bolivia (Morgan and Sessarego 2016); Costa Rica (Vásquez Carranza 2006); Ecuador (Bradley 2004); Spain (Henriksen and Willis 2010); Dominican Republic (Willis 2007); and Mexico (Amastae et al. 1998; Bradley and Willis 2012; Eller 2013; Lope Blanch 1967; Perissinotto 1972; Rissel 1989). Although assibilation is a common feature in all these dialects, social and linguistic factors seem to influence the phonological feature in unique ways. In this study, we focus on rhotic assibilation in the Spanish of the state of Chihuahua, Mexico/El Paso, TX, United States. We explore three aspects of rhotic assibilation: (i) The development of this feature more than two decades after the first (and only) study was conducted in this same dialect (Amastae et al. 1998); (ii) the linguistic and social factors that affect this variation; and (iii) whether contact with English affects the variable in bilinguals. The rest of the paper is organized as follows: Section1 reviews the phonetic and phonological characteristics of rhotics, the sociolinguistic literature on rhotic assibilation in Mexico City and Ciudad Juárez, Chihuahua. Section2 details the methodology for the study. Section3 introduces the results, and Section4 answers the research questions and relates the findings of the present study with the literature reviewed. Finally, we conclude and discuss the future directions of this research.

Languages 2020, 5, 38; doi:10.3390/languages5040038 www.mdpi.com/journal/languages Languages 2020Languages, 5, x FOR2020, 5PEER, 38 REVIEW 2 of 21 2 of 21

1.1. Phonological, Articulatory, and Acoustic Characteristics of Spanish Rhotics 1.1. Phonological, Articulatory, and Acoustic Characteristics of Spanish Rhotics With regardWith to regard rhotic to rhoticdistribution, distribution, there there is isa acl clearear distinction distinction between between the the syllabic syllabic positions positions that that can be occupiedcan be occupied by taps by and taps andtrills. trills. Taps Taps / /ɾ /,/, as as in in caro caro/ká o/ká/ ‘expensive’,ɾo/ ‘expensive’, and trills and/r¯/, as trills in carro /r/,̄ / kasár¯ oin/ carro /káro/̄ ‘car’,‘car’, contrast contrast in in word-internal word-internal intervocalic intervocalic position position (Hualde (Hualde 2005). In 2005). word-initial In word-initial position and position after and after a consonanta consonant in ina different a different syllable, only only the the trill trill occurs: occurs: Rosa Rosa/r¯ósa/ /r‘rose’ósa/̄ and ‘rose’ Israel and/is r¯Israelaél/. The /isr tapaél/.̄ The occurs in onset clusters: Prosa /p ósa/ ‘prose’ and in word-final position before a vowel: Ser amigos tap occurs in onset clusters: Prosa /pɾósa/ ‘prose’ and in word-final position before a vowel: Ser /sé amígos/ ‘to be friends’. A variable rhotic appears in coda position (within a word or across word amigos /séboundaries)ɾ amíɡos/ followed‘to be friends’. by a consonant A variable or a pause, rhotic i.e., appears when resyllabification in coda position is not possible.(within Whilea word a tap or across word boundaries)is more frequently followed found by in a these consonant contexts, or an a emphatic pause, trilli.e., can when also occurresyllabification1 e.g., arte a[r¯]te is ornotá[ ]tepossible. While a tap‘art’, is amor more amo[ frequently] or amo[r¯] ‘love’.found Following in these Hualde contexts,(2005 ),an we emphatic will use the trill symbol can/ ralso/ to represent occur1 thee.g., arte a[r]tē or á[non-distinctɾ]te ‘art’, rhoticamor found amo[ inɾ] final or amo[r position]̄ ‘love’. where there Following is no possible Hualde resyllabification. (2005), we Wewill will use also the use symbol the neutral term ‘rhotic’ rather than ‘tap’ and ‘trill’ to refer to it. /r/ to represent the non-distinct rhotic found in final position where there is no possible Articulatorily, standard taps and trills2 share the same place and manner of articulation; they are resyllabification.both usually We realized will also asvoiced use the and neutral alveolar term3. The ‘r mainhotic’ difference rather betweenthan ‘tap’ taps and and ‘trill’ trills is to that refer taps to it. Articulatorily,are produced standard with a single taps contact and between trills2 theshare tip ofthe the same tongue place towards and the manner alveolarridge, of articulation; while trills they are both usuallyare produced realized with severalas voiced (usually and two alveolar or three)3. The such main rapid contactsdifference (Hualde between 2005). taps Navarro and Tom trillsás is that taps are produced(Tomás 1970 ,with pp. 1, a 116) single gave contact a more detailed between description the tip of of the the Spanish tongue trill thattowards specifies the the alveolar position ridge, while trillsof theare tongue produced tip, dorsum, with andseveral root. He(usually stated thattwo the or tip three) of the tonguesuch bendsrapid upwardscontacts to (Hualde touch the 2005). upper-most part of the alveolar ridge, the tongue root retracts towards the back of the oral cavity, and Navarro theTomás tongue ([1918] body adopts 1970, app. hollow 1, 116) or concave gave shape.a more detailed description of the Spanish trill that specifies the positionThe realization of the of trillstongue requires tip, a dorsum, complex combination and root. ofHe articulatory stated that movements the tip ofof the the tongue-tip tongue bends upwards andto touch aerodynamic the upper-most forces (Solé 2002 part). of Due the to theiralveolar articulatory ridge, complexity, the tongue trills root are masteredretracts latetowards in theL1 the back of the oralacquisition cavity, and and arethe not tongue present body in the adopts babbling a stage. hollow Not onlyor concave do native shape. speakers find trills challenging, The realizationbut L2 learners of maytrills also requires find them a complex difficult tocomb acquireination and someof articulatory may never succeedmovements in rolling of the their tongue- [r]s (Solé 2002). Because the production mechanism of trills is quite complex and requires precise tip and aerodynamic forces (Solé 2002). Due to their articulatory complexity, trills are mastered late articulatory and aerodynamic conditions, small changes in their production can lead to perceptible in the L1 acousticacquisition differences. and are These not small present changes in the in the babbling production stage. of trills, Not which only coulddo native be due speakers to contextual find trills challenging,or prosodic but L2 features, learners create may a favorablealso find environment them difficult for the to instability acquire of and this soundsome andmay the never creation succeed of in rolling theirnew [r]s variants. (Solé In2002). fact, theBecause rich literature the production on rhotic variationmechanism in different of trills dialects is quite of Spanishcomplex provides and requires precise articulatoryevidence for theand instability aerodynamic of trills and, conditions, most importantly, small thechanges methodological in their challenges production they present can lead to in their analysis (see Method section). perceptible acoustic differences. These small changes in the production of trills, which could be due Acoustically, the tap and the trill share the same characteristics: A lowered third formant (Colantoni to contextual2001) or and prosodic brief periods features, of occlusion, create one a favora occlusionble for environment the tap and more for than the oneinst forability the trill. of However,this sound and the creationtaps of and new trills variants. differ in the In duration fact, the of therich segment, literature with on taps rhotic being shortervariation than in trills. different Quilis(1993 dialects) of Spanish providesreported an evidence average of for 20 ms.the for instability taps and 60 of ms. trills for trills.and, Themost length importantly, of the trill is the affected methodological by the challengesfollowing they present vowel: Itin is their shorter analysis before [i] (see and Method longer before section). [a] (Sol é 2002). Assibilated and standard rhotics4 tend to alternate synchronically and diachronically (Solé 1992). Acoustically, the tap and the trill share the same characteristics: A lowered third formant Synchronically, apical trills exhibit non-trilled variants, taps, approximants, and fricatives. Solé (2002) (Colantoniperformed 2001) and a study brief to periods replicate of the occlusion, phonological one variation occlusion between for the voiced tap trills and and more fricatives. than one Her for the trill. However,results taps showed and that trills trills differ may becomein the duration fricatives (orof the assibilated) segment, if the with finely taps controlled being shorter articulatory than trills. Quilis (1993)or aerodynamic reported requirementsan average forof trills20 ms. are notfor met,taps suggesting and 60 ms. that fricativesfor trills. involve The length a less complex of the trill is affected byarticulation the following and allow vowel: a wider It rangeis shorter of oropharyngeal before [i] and pressure longer variation before than [a] trills. (Solé Thus, 2002). assibilated Assibilated and standard rhotics 4 tend to alternate synchronically and diachronically (Solé 1992). Synchronically, apical trills exhibit non-trilled variants, taps, approximants, and fricatives. Solé 1 The distribution of trills and taps is further discussed in Hualde(2005). (2002) performed2 Since there a isstudy variation to affecting replicate the system the ofphonological rhotics (Hualde 2005 variation), when we refer between to ‘standard’ voiced variants trills we mean and prestige fricatives. forms used in careful speech, prescribed by schools and used by newscasters. Her results3 Lipski showed(1994) and that Hualde trills(2005) discussedmay otherbecome (non-standard) fricatives realizations (or including assibilated) dorsalization if and the pre-aspiration finely ofcontrolled the articulatory trill,or neutralization,aerodynamic retroflexion, requirements strengthening offor rhotics trills in codasare andnot onset met, clusters. suggesting that fricatives involve a 4 Solé (1992) actually refers to word-final rhotics as “trills” but, as stated earlier, we prefer to use Hualde’s (2005) neutral term less complex“rhotic”. articulation Solé also refers and to assibilatedallow a variants wider as “fricatives”.range of oropharyngeal pressure variation than trills. Thus, assibilated rhotics occur when the vibrating tongue-tip fails to make contact with the palate, or

1 The distribution of trills and taps is further discussed in Hualde (2005). 2 Since there is variation affecting the system of rhotics (Hualde 2005), when we refer to ‘standard’ variants we mean prestige forms used in careful speech, prescribed by schools and used by newscasters. 3 Lipski (1994) and Hualde (2005) discussed other (non-standard) realizations including dorsalization and pre- aspiration of the trill, neutralization, retroflexion, strengthening of rhotics in codas and onset clusters. 4 Solé (1992) actually refers to word-final rhotics as “trills” but, as stated earlier, we prefer to use Hualde’s (2005) neutral term “rhotic”. Solé also refers to assibilated variants as “fricatives”. Languages 2020, 5, 38 3 of 21 rhotics occur when the vibrating tongue-tip fails to make contact with the palate, or when apical vibration fails, allowing the high velocity air to flow continually through the aperture generating frication (Solé 2002). She further stated that assibilated rhotics in phrase-final position result from the difficulty of sustaining trilling with the lowered decreased subglottal pressure that occurs at the end of a statement. Since this is a natural phenomenon, Solé (2002) argued that the co-occurrence of trilling and frication (or standard and assibilated rhotics) should be a common cross-linguistic pattern. In fact, rhotic assibilation (mainly in phrase-final position) has been reported in many dialects of Spanish (Blecua and Cicres 2019; Canfield 1981; Quilis 1981), in other Romance and non-Romance languages such as Brazilian Portuguese (Silva 1996), Czech (Howson et al. 2014), and in Farsi (Ladefoged and Maddieson 1996). Solé (2002) replicated in the laboratory the variation observed in different languages.

1.2. Sociolinguistic Literature on Rhotic Assibilation in Mexico City Assibilated taps and trills were first studied in Mexico City by Lope Blanch(1967). The author (Lope Blanch 1967) argued that assibilation was a recent phenomenon that appeared after the 1950s in the speech of women. He suggested that assibilation might have been imported from Spain, because it had been observed there and in several other Latin-American countries. Lope Blanch(1967) proposed that rhotic assibilation is a natural process in the evolution of the phonological system of languages. Unfortunately, this last point was not developed further, but it is explored in the laboratory by Solé (2002), summarized above. The first synchronic sociolinguistic analysis of rhotic assibilation in Mexico City was published by Perissinotto(1972). The study was based on 110 h of recorded conversation (the number of participants was not stated) collected between 1963 and 1969. His results showed a high overall percentage of assibilation (68.2%). Female speakers had a higher rate of the assibilated rhotic [ˇr] compared to their standard variant [r] ([ˇr] 81.8% vs. [r] 18.2%) while male speakers had a reverted pattern of variation: A higher percentage of the standard variant compared to the assibilated one ([r] 61.1% vs. [ˇr] 38.9%). By comparing the distribution of the assibilated variant across age and socio-economic status, Perissinotto (1972) found that assibilation was more common in the younger age group and in the high and middle socio-economic levels. In other words, assibilation was a prestigious innovative variant adopted by women of the higher classes. More than three decades after the first study, Martín Butragueño(2006) performed an analysis of rhotic assibilation in 54 native speakers of Mexico City. The overall percentage of assibilation in absolute final position was 27%, compared to the 68.1% reported by Perissinotto(1972). A multivariate analysis of rhotic assibilation showed a number of linguistic and social factors that favored assibilation: (1) Absolute final position; (2) formal style; (3) mid and high education; (4) older generation; and (5) females. Martín Butragueño(2006) claimed that given the lower percentage of assibilation found in his study, as compared to Perissinotto’s (1972), rhotic assibilation seemed to be a receding case of language change. This conclusion was reinforced by the fact that assibilation was more frequent in older women, followed by the adult and younger groups (36%, 32%, and 17%, respectively). This is a tendency in the other direction of what Perissinotto(1972) had found with assibilation being more frequent in younger speakers, followed by adults and older speakers (73.5%, 64.5%, and 31.3%, respectively). The comparison of assibilation found by Perissinotto(1972) and Mart ín Butragueño(2006) seems to indicate that rhotic assibilation was a short-lived fashion in Mexico City. Martín Butragueño’s (2014) second study analyzed the speech of 54 interviews and reading tasks from the Sociolinguistic Corpus of Mexico City. The subjects were distributed along the different social categories of age, sex, and education level. The study took into consideration all rhotic positions including coda, onset and consonant cluster, word final, and absolute final. A multivariate analysis confirmed what had been reported previously for rhotic assibilation. Presented in order of importance, assibilation was favored by: (1) Phrase-final position; (2) interview (as opposed to reading); (3) consonant clusters (especially /tr/); (4) people with medium level of instruction (10–12 years); (5) female Languages 2020, 5, 38 4 of 21 speakers; and (6) older speakers (55+ years old). Other coda positions were not selected as significant by the multivariate analysis and the author reported very few cases of assibilation in positions other than absolute final and in consonant cluster /tr/. Languages 2020, 5, x FOR PEER REVIEW 4 of 21 Martín Butragueño(2014) presented some spectrograms of trill and taps to illustrate the di fferent types of rhotics foundselected in Mexico as significant City Spanish. by the Besidesmultivariate the standardanalysis variants,and the author he also reported found approximant very few cases of variants, which theassibilation author definedin positions as weakenedother than absolute versions final of standardand in consonant rhotics, cluster and assibilated /tr/. rhotics. He proposed that rhoticMartín assibilation Butragueño be (2014) described presented as change some spectrogra from an approximantms of trill and taps to a to fricative illustrate sound. the different According to thetypes author, of thisrhotics becomes found evidentin Mexico in theCity cases Spanish. where Besides the rhotic the standard starts as variants, an approximant he also found approximant variants, which the author defined as weakened versions of standard rhotics, and (with a clear sonority bar and no noise) and continues as a fricative with high frequency noise and an assibilated rhotics. He proposed that rhotic assibilation be described as change from an approximant almost inexistentto F0. a fricative However, sound. as demonstratedAccording to the by author, Solé (this2002 becomes) in the evident laboratory, in the therecases where is compelling the rhotic starts cross-linguistic evidenceas an approximant suggesting (with that a trills clear alternate sonority bar with and fricatives no noise) (or and assibilated continues variants),as a fricative so itwith is high plausible that assibilatedfrequency rhotics noise resultand an from almost “a failureinexistent ofsustaining F0. However, trilling as demonstrated with the lowered by Solé decreased (2002) in the subglottal pressurelaboratory, that occurs there at is thecompelling end of across-linguistic statement” ( Solevidenceé 2002 suggesting). that trills alternate with fricatives (or assibilated variants), so it is plausible that assibilated rhotics result from “a failure of sustaining 1.3. Sociolinguistictrilling Literature with onthe Rhotic lowered Assibilation decreased insubglottal the State pressure of Chihuahua, that occurs Mexico at the end of a statement” (Solé 2002). Rhotic assibilation reached the north of Mexico after the 1960s, with the first study conducted in Ciudad Juárez1.3. in Sociolinguistic 1998 (Amastae Literature et al. on 1998 Rhotic). Assi Thisbilation study in showedthe State of that Chihuahua, assibilation Mexico was present in Ciudad Juárez, but with a lower frequency of overall occurrence than reported in Mexico City. Rhotic assibilation reached the north of Mexico after the 1960s, with the first study conducted in The percentage ofCiudad assibilation Juárez in 1998 all final (Amastae positions et al. (including 1998). This wordstudy finalshowed and that phrase-final) assibilation was present 6%, in while the percentageCiudad of assibilation Juaréz, but with in absolute a lower final frequency position of overa wasll 22%. occurrence Compared than toreported the 68.1% in Mexico reported City. The by Perissinotto(1972percentage), assibilation of assibilation was muchin all final less positions frequent (including in Ciudad word Ju finalárez. and Amastae phrase-final) et al. was(1998 6%,) while conducted a multivariatethe percentage analysis of assibilation on 72 speakers in absolute and final found position that was assibilation 22%. Compared was to a fftheected 68.1% by: reported (1) by Gender: HigherPerissinotto rate of assibilated (1972), assibilation rhotics in was women much (women less frequent 28% in and Ciudad men 16%);Juárez. (2)Amastae social et class: al. (1998) Greater percentageconducted of assibilation a multivariate in the higheranalysis socio-economic on 72 speakers classand found as compared that assibilation to themiddle was affected class 5by: (1) Gender: Higher rate of assibilated rhotics in women (women 28% and men 16%); (2) social class: (21% and 18%, respectively); (3) age: Adults (36–55 years) and older participants (56+) showed higher Greater percentage of assibilation in the higher socio-economic class as compared to the middle class5 rates of assibilation(21% (31% and and18%, 19%,respectively); respectively) (3) age: than Adults younger (36–55 years) participants and older (15%); participants and (4) (56+) education: showed higher Assibilation was morerates of frequent assibilation in speakers (31% and with 19%, higherrespectively) levels th ofan formal younger education participants (university (15%); and 23% (4) andeducation: high school 14%).Assibilation All these e wasffects more suggested frequent that in speakers assibilation with washigher also levels a prestigious of formal education variant in (university Ciudad 23% Juárez. Amastaeand et al. high(1998 school) argued 14%). thatAll these assibilation effects suggested in Ciudad that Ju assibilationárez was was a recent also a “changeprestigious from variant in above” (Labov 1972Ciudad) imported Juárez. Amastae from Mexico et al. (1998) City argued by the that higher assibilation classes in and Ciudad transmitted Juárez was by a women.recent “change According to Labovfrom(1972 above”), a change(Labov 1972) from imported above is afrom linguistic Mexico change City by that the entershigher theclasses language and transmitted from by above the level ofwomen. consciousness; According that to is,Labov speakers (1972), are a change generally from aware above of is the a linguistic linguistic change form andthat theyenters the language from above the level of consciousness; that is, speakers are generally aware of the linguistic manipulate its use depending on the context and/or their interlocutors. The upper classes use these form and they manipulate its use depending on the context and/or their interlocutors. The upper new linguistic formsclasses in use order these to dinewfferentiate linguistic themselvesforms in order from to thedifferentiate lower classes, themselves while from lower the classeslower classes, use these forms inwhile order lower to sound classes moreuse these formal forms and in order similar to soun to thed more upper formal classes. and similar to the upper classes. Twenty years afterTwenty Amastae’s years after study, Amastae’s Mazzaro study, and Mazzaro Gonz andález González de Anda (2019)(2019 investigated) investigated the relationship the relationship betweenbetween the the perception perception and and productionproduction of of rhotic rhotic assibilation assibilation [ř] and [ˇr]and deaffrication deaffrication of the voiceless of the voiceless post-alveolarpost-alveolar a ffaffricatericate [ ʃ ]] in in the the Spanish Spanish of Chihuahua, of Chihuahua, Mexico. Mexico. Thirty-three Thirty-three native Spanish native speakers Spanish speakersfrom from the the state state of ofChihuahua Chihuahua completed completed production production and perception and perception tasks to tasks establish to establish whether those whether those thatthat produced produced the the variants variants were were able able to perceive to perceive them. them.The production The production data were dataelicited were by asking participants to narrate the fairy tale Caperucita Roja ‘Little Red Riding Hood’ and by asking them to elicited by asking participants to narrate the fairy tale Caperucita Roja ‘Little Red Riding Hood’ and talk about their favorite food. Results showed that the overall percentage of [ř] production was by asking them togreater talk aboutthan that their of [ favoriteʃ] (17.15% food. and 11.83%, Results respectively). showed that The the authors overall explained percentage that the of increased [ˇr] production was greateroverall rate than of that [ř] might of [ ]be (17.15% due to assibilation and 11.83%, only respectively). being investigated The in authors absolute explained final position, that which the increased overallis the rate context of [ˇr]might where most be of due theto variation assibilation occurs. only The beingsocial factors investigated that turned in absolute out to be final significant position, which iswere the gender context and where generation, most ofwith the female variation and younge occurs.r speakers The social favoring factors the that use of turned assibilation. out The to be significantrelationship were gender between and generation, production with and femaleperception and of younger assibilation speakers was very favoring subtle; the overall, use the of assibilation. Theperception relationship of the betweenvariant was production very low (9.1%). and perception Those who ofproduced assibilation assibila wastion very were subtle; not the ones

5 Amastae et al.(1998 ) did not analyze the speech of participants of lower socio-economic class. 5 Amastae et al. (1998) did not analyze the speech of participants of lower socio-economic class. Languages 2020, 5, 38 5 of 21 overall, the perception of the variant was very low (9.1%). Those who produced assibilation were not the ones that perceived it the most, which seemed to suggest that rhotic assibilation was below the level of consciousness and social awareness (Labov 2001). To the best of our knowledge, no other sociolinguistic studies have been done on rhotic assibilation in this geographic area. The present research will further investigate the social and linguistic factors that affect the production of this variable. We also add another layer of analysis to increase the accuracy of coding of tokens, by supplying the auditory classification of tokens with a spectrographic analysis. Acoustic information will allow us to identify if there are other variants that could also be alternating with assibilated rhotics. This study is designed to explore different aspects of the production of rhotic assibilation in the state of Chihuahua/El Paso, TX speech community6. Our specific questions are the following: i. What is the current state of rhotic assibilation more than two decades after the first (and only) study was done? In this study, we expect to find a lower frequency of rhotic assibilation, since previous studies on this same dialect (Amastae et al. 1998) stated that this variable was receding. ii. Which rhotic variants are found in Chihuahua Spanish? Given that Amastae et al.(1998) suggested that assibilation is receding, we predict that we will find a much smaller percentage of rhotic assibilation than he did in 1998, and a large percentage of standard variants (taps and trills). It is possible that we could find other variants reported in other dialects of Mexican Spanish such as approximants, retroflex, and fricatives (Martín Butragueño 2006). iii. What are the most important linguistic and social correlates of rhotic assibilation? Gender is expected to be one of the most significant factors that influence the variable under study. The fact that female speakers favor assibilation has been consistently reported by Amastae et al. (1998); Lope Blanch(1967); Mart ín Butragueño(2006); and Perissinotto(1972). Besides gender, we anticipate older speakers to produce more assibilation than the other age groups, which is due to the receding status of the variant (Amastae et al. 1998). The literature on rhotic assibilation agree that the phrase-final position is the most favoring phonetic context for assibilated rhotics. Thus, in this study, we focus on rhotic production in absolute final position. Finally, the preceding vocalic context was found to be significant in our previous analysis of rhotic assibilation (Mazzaro and González de Anda 2016), so we predict that with more data, this factor will show a clearer and more robust effect. iv. Does being a Spanish-English bilingual affect the use of this variable? Previous work (Dalola and Bullock 2017) found that being bilingual affects the social perception and production rates of variables in the L2. However, our participants are all L1 speakers of Spanish, so we do not expect the L2 English to affect the rate of assibilation in bilingual compared to monolingual speakers. v. What is the effect of the formality of the task (style): Reading vs. narrative vs. conversation on the occurrence of assibilation? Given the prestige attached to assibilated rhotics (Amastae et al. 1998; Perissinotto 1972), we predict that tasks that are more formal would elicit higher instances of assibilation. Therefore, the reading task will elicit a higher percentage of assibilation than the narrative and the informal conversation. As the formality of the task decreases, so will the frequency of assibilated rhotics. vi. Does assibilation remain a prestigious feature of speech? We expect that rhotic assibilation will continue to be a prestigious feature of speech. This is based on the vast amount of literature that report the assibilated variant to be used in higher social classes and subjects with higher levels of formal education (Amastae et al. 1998; Martín Butragueño 2006, 2014; Perissinotto 1972).

6 We consider El Paso, TX part of the state of Chihuahua’s Spanish-language speech community because of their geographic proximity (separated only by the Rio Grande River). We are aware that there is lack of research that compares the Spanish of the state of Chihuahua and that of El Paso, but we posit that they are one dialect, at least at the phonetic/phonological level. Languages 2020, 5, 38 6 of 21

2. Materials and Methods

2.1. Speakers Participants of this study are native Spanish speakers recruited in the El Paso, Texas—Ciudad Juárez, Mexico border area. A total of 58 subjects participated in this study: 36 women and 22 men, with an age range between 18 and 69 years. To compare our data with Amastae et al.’s (1998) generational groups, the participants were divided into four groups: Generation 1: <20; Generation 2: 21–35; Generation 3: 36–55; and Generation 4: >56. Participants were asked to complete an adult language background questionnaire that elicits information about their place of birth, language(s) of schooling, and language use. The questionnaire contained a section that asked for participants’ self-proficiency ratings in both Spanish and English, and only those who reported to use mostly/only Spanish in their daily everyday interactions (at home, at work, and in social situations) were selected to participate in the study. We considered bilingual those participants who reported to know another language and self-assessed their English knowledge as higher than basic. The demographic information of the participants considered in the statistical analysis is presented in Table1.

Table 1. Participants’ demographic information.

Participants’ Information Participants n 58 Age range in years 18–69 Generation 1 (<20 years) n 23 (13 females, 10 males) Generation 2 (21–35 years) n 18 (11 females, 7 males) Generation 3 (36–55 years) n 12 (7 females, 5 males) Generation 4 (>56 years) n 5 (5 females) Male: Female n 22:36 Median/Mean age in years 22/30 Ciudad Juárez (25), n From location El Paso, TX (20), Chihuahua—Capital (5), Chihuahua—Interior (Delicias, Parral, Balleza, Cd. Guerrero, Cuauhtémoc, Jiménez) (8) Bilingual: Monolingual n 29:29:00

All subjects gave their informed consent for inclusion before they participated in the study. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Ethics Committee of the University of Texas at El Paso IRB (Project identification code [274707-10]). Thirty-four participants were recruited at the University of Texas at El Paso (UTEP). Some were students enrolled in beginner-level ESOL (English for Speakers of Other Languages). Seven participants were students at the Universidad Tecnológica de Ciudad Juárez. The remaining 17 participants were contacted using the ‘friend-of-a-friend’ technique (Milroy 1987), whereby potential informants are contacted through common friends, an approach that is particularly appropriate for the community under study. Although rhotic assibilation in Mexican Spanish is a feature mainly observed in higher social classes, specifically mid-high to high class (Amastae et al. 1998; Perissinotto 1972), our study also includes participants of the mid-low or middle social classes. Most of the participants from El Paso who attend UTEP fall within the mid-low or middle social classes. The participants from the state of Chihuahua who attend UTEP are probably from mid to high classes, since it is very expensive for Mexican citizens to afford college north of the border. However, because we do not have an accurate tool to determine participants’ social class, we will not investigate the effect of this factor henceforth. Languages 2020, 5, 38 7 of 21

2.2. Independent Variables Based on the literature reviewed, we investigated a number of linguistic and social factors considered significant in previous studies, in addition to other factors that proved important in our previous analysis of this feature (Mazzaro and González de Anda 2016). Among the linguistic factors, we analyzed the influence of the preceding context on the occurrence of assibilation. Mazzaro and González de Anda(2016) found that the distribution of rhotic assibilation across different vocalic contexts was different; [ˇr] was more frequent when preceded by [e] and [i]. Our 2016 study also confirmed previous findings that rhotic assibilation was localized to phrase-final position (Amastae et al. 1998; Lope Blanch 1967; Martín Butragueño 2006; Perissinotto 1972). Thus, we decided to focus our present study on this context only. Though our original plan was to study the influence of ‘stress’, we found very few tokens of syllable coda -r in unstressed position (N = 8) and, therefore, we could not investigate this effect. Among the social factors, we analyzed the influence of sex, age, and education on the occurrence of rhotic assibilation. We also explored the effect of task: Reading (formal) vs. narrative of a fairy tale (semi-informal) vs. talking about the favorite food (informal) on the frequency of rhotic assibilation.

2.3. Data Collection and Recording Equipment During the production stage of the study7, participants’ speech was elicited by completing three tasks. First, participants completed a word-list reading task (AppendixA). The stimuli were presented on a computer screen. Words from the list were presented one at a time using MS Power Point slides. For this task, participants were instructed to read words in isolation with a pause after each word. The stimuli consisted of 36 content words with a rhotic in final position and preceded by /a/, /e/, /i/, or /o/ e.g., comer (to eat); posar (to pose). The list contained 23 distracters, some of which had /r/ in different positions in the word or in consonant clusters e.g., bicho (bug); pared (wall). Word final rhotics were always in stressed syllables, so the ‘stress’ factor was not investigated in the word list. In the second task, speakers narrated the fairy tale Caperucita Roja ‘Little Red Riding Hood’ fairy tale. To elicit the story, participants were shown wordless pictures of children’s book based on Perrault’s version of the tale. First, participants were shown the pictures as a refresher. When they felt ready, participants recorded the narrative using the images as guidance. The third production task consisted of an informal interview in which the researcher asked participants about their favorite food, where they eat it, and/or who prepares it. This last section was considered the most casual style of speech, since speakers were free to talk without prompts. All the sessions were conducted in a sound-treated room. Informants were recorded directly onto a laptop computer using Audacity 2.1.2 and a Blue Snowball USB microphone. The speech was sampled at 44.1 K, and phrase-final rhotic tokens were analyzed with PRAAT (Boersma and Weenink 2018).

2.4. PRAAT and Spectrographic Inspection of Tokens Recordings were transcribed with PRAAT (Boersma and Weenink 2018) using regular orthography on tier five (Figure1). The transcriptions were force-aligned with Praatalign (Lubbers and Torreira 2016). After alignment, the text grids were manually inspected for accuracy. PraatAlign generated two tiers based on the audio and transcription: (1) The fourth tier contained the segmental analysis, and the third tier contained the word boundaries. The top tier contains our auditory classification of the rhotics in absolute final position, and the second tier contains the classification of the syllable into stress and unstressed. Our perception of rhotics was based on both auditory analysis and inspection of the spectrograms and oscillogram where necessary.

7 Participants also performed a perception task. This study focuses on the production data only. The perception study was published elsewhere Mazzaro and González de Anda(2019). Languages 2020, 5, x FOR PEER REVIEW 8 of 21 Languages 2020, 5, 38 8 of 21 Languages 2020, 5, x FOR PEER REVIEW 8 of 21

FigureFigure 1. 1. TranscriptionTranscription of of the the narrative narrative in in PraatAlign. PraatAlign. Female Female speaker speaker (UT007). (UT007). Figure 1. Transcription of the narrative in PraatAlign. Female speaker (UT007). 2.5.2.5. Coding Coding 2.5. Coding RhoticsRhotics in in absolute final final position position were were coded coded by twoby trainedtwo trained phoneticians phoneticians who are who Spanish are Spanish speakers. speakers.TokensRhotics were Tokens classifiedin absolutewere classified as eitherfinal position non-assibilatedas either were non‐ assibilatedcoded or assibilated. by twoor assibilated. trained Non-assibilated phoneticians Non‐assibilated tokens who were tokens are coded Spanish were as a codedtrillspeakers. (t) as with a Tokenstrill multiple (t) werewith closures multipleclassified (Figure closures as 2eithera) or (Figure asnon tap‐assibilated (r) 2a) with or a as single or tap assibilated. (r) closure with (Figure a Non single‐assibilated2b). closure Assibilated (Figure tokens rhotics were2b). Assibilated(rs)coded were as thosea rhoticstrill where(t) (rs)with thewere multiple presence those closures where of friction the(Figure presence or noise 2a) or was of as friction perceivedtap (r) orwith noise strongly a single was throughout perceivedclosure (Figure thestrongly rhotic 2b). throughout(FigureAssibilated1). Rhotics the rhotics rhotic that (rs) (Figure were were partially those1). Rhotics where assibilated that the werepresence were codedpartially of asfriction (m)assibilated ‘mix’ or noise (Figure were was2 c).coded perceived Mixed as rhotics(m) strongly ‘mix’ are (Figurethosethroughout that 2c). begin Mixed the withrhotic rhotics one (Figure are or several those 1). Rhotics that closures begin that (like with were a standardone partially or several rhotic) assibilated closures and end were (like with codeda friction.standard as If(m) rhotic) a token‘mix’ andlacked(Figure end closure with2c). Mixed friction. and was rhotics If accompanieda token are thoselacked bythat closure weak begin noise and with was it wasone accompanied codedor several as (a) closures ‘approximant’by weak (like noise a (Figure itstandard was 2codedd). rhotic) If theas (a)rhoticand ‘approximant’ end was with heard friction. as(Figure a trill If a2d). (with token If several the lacked rhotic closures) closure was heard and with was frication,as a accompanied trill (with it was several coded by weak asclosures) (ts) noise (Figure withit was2e). frication, coded Rhotics as itthat(a) was ‘approximant’ were coded heard as (ts) as (Figure a(Figure voiced 2d). 2e). glottal If Rhotics the fricative rhotic that was or were aspiration, heard heard as aas weretrill a voiced (with coded severalglottal as (h) (Figureclosures)fricative2f). orwith Ambiguous aspiration, frication, wereorit was unintelligible coded coded as (h)as (Figure(ts) tokens (Figure were2f). Ambiguous2e). classified Rhotics as or that (n) unintelligible andwere were heard later tokens as discusseda voiced were classified glottal with a fricative thirdas (n) coder. and or wereaspiration, Only later the discussedtokenswere coded for with which as a(h) third there (Figure coder. was 2f). complete Only Ambiguous the agreementtokens or forunintelligible which were included there tokens was in complete thewere statistical classified agreement analysis, as (n) were and so included (n)were tokens later inarediscussed the excluded statistical with from analysis,a third the statisticalcoder. so (n) Only tokens analysis. the are tokens excluded for which from there the statistical was complete analysis. agreement were included in the statistical analysis, so (n) tokens are excluded from the statistical analysis.

(a) (b) (a) (b) Figure 2. Cont. Languages 2020, 5, 38 9 of 21 Languages 2020, 5, x FOR PEER REVIEW 9 of 21

(c) (d)

(e) (f)

FigureFigure 2.2. ((aa)) TrillTrill (t)(t) inin thethe wordworddador dador(donor). (donor). MaleMale speakerspeaker UT086;UT086; ((b)) taptap (r)(r) inin thethe wordword soporsopor (sleepiness).(sleepiness). Female speaker UT079; ( c)) partially assibilated (m) (m) rhotic in the word besar (to kiss). FemaleFemale speakerspeaker UT079;UT079; ((dd)) approximantapproximant ((aa)) inin thethe wordword toto weigh weigh(pesar). (pesar). MaleMale speakerspeaker UT080;UT080; ((ee)) assibilatedassibilated trill trill (ts) (ts) in in the the word wordbeber beber(to (to drink). drink). Male Male speaker speaker UT085; UT085; (f) aspirated (f) aspirated rhotic rhotic (h) in (h) the in word the tutorword(tutor). tutor (tutor). Female Female speaker speaker UT105. UT105.

2.6.2.6. StatisticalStatistical AnalysisAnalysis TheThe effect effect of of the the linguistic linguistic and extra-linguisticand extra‐linguistic variables variables was tested was using tested binary using logistic binary regression logistic inregression R (R Core in Team R (R 2020 Core). WeTeam used 2020). a mixed We effectused modela mixed that effect included model ‘speaker’ that included as a random ‘speaker’ variable, as a usingrandom the variable, lmer function using (theBates lmer et al.function 2011). (Bates The independent et al. 2011). variablesThe independent were added variables to the were regression added modelto the regression one at a time model until one the at BIC a time (Bayesian until the Information BIC (Bayesian Criterion) Information for the Criterion) model did for not the improve model further.did not Theimprove order further. in which The the order variables in which were the added variables to the were model added was determined to the model by was random determined forests, usingby random the cforest forests, and using varimp the functions cforest and in the varimp party functions package also in the in Rparty (Strobl package et al. 2008 also). in Random R (Strobl forests et al. evaluate2008). Random the relative forests importance evaluate of eachthe relative independent importance variable, of using each a procedureindependent that variable, tests whether using any a independentprocedure that variable tests is whether a sensitive any predictor independent of the responsevariable variable.is a sensitive To do so,predictor random of forests the response follow a processvariable. of To trial do and so, errorrandom working forests with follow random a process samples of trial of the and data error (Tagliamonte working with and random Baayen samples2012). of the data (Tagliamonte and Baayen 2012). 3. Results

3. ResultsThe dataset consists of 2223 tokens of (r) in absolute final position. The word-list reading task yielded a total of 1762 tokens, the narrative task yielded a total of 235 tokens, and the semi-spontaneous The dataset consists of 2223 tokens of (r) in absolute final position. The word‐list reading task speech yielded 226 tokens. The overall percentage of fully assibilated rhotics including all three tasks yielded a total of 1762 tokens, the narrative task yielded a total of 235 tokens, and the semi‐ was 49.6% (N = 1103). However, as explained in the coding section, a thorough acoustic and perceptual spontaneous speech yielded 226 tokens. The overall percentage of fully assibilated rhotics including analysis of the tokens showed that there were other more subtle rhotic variants with varying degrees of all three tasks was 49.6% (N = 1103). However, as explained in the coding section, a thorough acoustic frication. In order to capture the difference in their realization and to provide a more accurate picture and perceptual analysis of the tokens showed that there were other more subtle rhotic variants with of assibilation in this speech community, we classified them according to their perceptual and acoustic varying degrees of frication. In order to capture the difference in their realization and to provide a more characteristics. Table2 presents the distribution of all /r/ variants in the three tasks combined. accurate picture of assibilation in this speech community, we classified them according to their perceptual and acoustic characteristics. Table 2 presents the distribution of all /r/ variants in the three tasks combined. Languages 2020, 5, x FOR PEER REVIEW 10 of 21

Languages 2020, 5, 38 10 of 21 Table 2. Frequency of occurrence of all rhotic variants (three tasks combined). Languages 2020, 5, x FOR PEER REVIEW 10 of 21 Languages 2020, 5, x FOR PEER REVIEW IPA Table 2. Frequency of occurrenceOccurrence of all rhotic variants 2Description of (three 21 tasks combined). Languages 2020, 5, x FOR PEER REVIEW 10 of 21 IPATable 2. Frequency of Occurrence occurrence of all Description rhotic variants (three tasks combined). 1.1. Phonological, Articulatory, and Acoustic Characteristics of Spanish Rhotics N % Table 2. Frequency of occurrence of all rhotic variants (three tasks combined). [rIPA]̄ Trill OccurrenceN% 108Description 4.9% Multiple closures without friction With regard to rhotic distribution, thereIPA[Languagesr]¯ Trill is a cl2020ear, 5, xdistinction FOR PEER REVIEW Occurrencebetween 108 theDescription 4.9% syllabic Multiple positions closures that without friction 10 of 21 [ ɾ] Tap N % 852 38.3% A single tap can be occupied by taps and trills. Taps /[r [ɾ]̄ ]Trill/, Tap as in caro /káɾo/ ‘expensive’,N 108 852 % 4.9% and 38.3% Multipletrills A /r single /,̄ cl osuresas tap in withoutcarro friction [[[rřɾ]̄] TrillTap Assibilated Table rhotic 2. Frequency 108 852 of 4.9%occurrence 38.3% Multiple 1103 of A all single rhoticclosures 49.6% tap variants without (three Afriction fully tasks combined). fricativized rhotic /káro/̄ ‘car’, contrast in word-internal intervocalic[[ˇr]Assibilatedɾ] Tap position rhotic (Hualde 852 2005). 1103 38.3% In word-initial A 49.6% single tap A fully position fricativized and rhotic [ř]IPA Assibilated rhotic Occurrence 1103 49.6% DescriptionA fully fricativized rhotic after a consonant in a different syllable, only[[[ˇrř]]̸ / ]Assibilated * *the PartiallyPartially trill rhoticoccurs: assibilated assibilated Rosa rhotic 1103/rósa/̄ rhotic 88 49.6% ‘rose’ A 4%and 88fully fricativizedIsraelA 4%rhotic /isr rhotic thataél/.̄ starts AThe rhotic with a closure(s) that starts and finishes with with a noiseclosure(s) and finishes with noise [ř]̸ * Partially assibilated rhotic N 88 % 4% A rhotic that starts with a closure(s) and finishes with noise [ř]̸ * Partially assibilated rhotic 88 4% A rhotic that starts with a closure(s) and finishes with noise tap occurs in onset clusters: Prosa /pɾósa/[[[ ɹ[r ]]‘prose’] ]̄ AlveolarAlveolar TrillAlveolar approximantandapproximant approximantin word-final 108 50 50 position4.9% 2.2% 2.2% 50Multiple beforeA rhotic A closures 2.2% rhotic witha vowel: with no without closure noA Ser closure frictionrhotic with no closure [ ɹ ] Alveolar approximant 50 2.2% A rhotic with no closure ɡ ɦ[ ɾ] Tap 852 38.3% A single tap amigos /séɾ amí os/ ‘to be friends’. A variable[[[ ɦ ]] ] Glottal]GlottalGlottalrhotic Glottal fricative fricative appears fricative in coda 15 position 15 15 0.7% 0.7% (within Frication 0.7% 15 Frication thata Frication word 0.7% follows that or follows that a partiallyacross follows Frication a partially devoiced a partially devoicedthat vowel devoiced follows vowel vowel a partially devoiced vowel [̌ ř] Assibilated rhotic 1103 49.6% A fully fricativized rhotic word boundaries) followed by a consonant[[ r̄ ̌̄]]̌ AssibilatedAssibilatedor a pause, trill trill trill i.e., when 7 7resyllabification 7 0.3 0.3 Multiple 0.3 Multiple closures Multipleis notclosures with possible. closures friction with with friction friction [ r[̄ ř]̸ *Assibilated Partially assibilated trill rhotic 88 4% 7A rhotic that 0.3 starts with Multiple a closure(s) and closures finishes with with noise friction ** WeWe tentativelytentatively use use this this symbol symbol to represent to represent the partially the partially assibilated1 assibilated rhotic, whichrhotic, has which not been has not been While a tap is more frequently found in thes[ɹ] *Alveolare We contexts, tentatively approximant usean thisemphatic symbol50 to trill represent2.2% can theAalso rhotic partially occur with assibilated no closuree.g., rhotic, arte which has not been described in describedthedescribed* literature.We in tentativelyin the the literature. literature. use this symbol to represent the partially assibilated rhotic, which has not been a[r]tē or á[ɾ]te ‘art’, amor amo[ɾ] or amo[r]̄ [‘love’.ɦ] Glottal Following fricative Hualde15 (2005),0.7% weFrication will use that followsthe symbol a partially devoiced vowel ̌ /r/ to represent the non-distinct rhotic [rfound]̄ TableTableAssibilateddescribed 2 2 showsin shows trillfinal an inan overall thepositionoverall literature.low 7low percentage where percentage0.3 ofthereMultiple trills of trillsinis closuresabsolute noin absolute withpossible final friction position final position (4.9%) while (4.9%) taps while are taps are muchTable *more We2 tentativelyshowsfrequent an inuse overall this this positionsymbol low percentage to (38.3%). represent However, ofthe trills partially inassibilated absolute assibilated rhotics final rhotic, position arewhich by hasfar (4.9%) notthe been whilemost taps are resyllabification. We will also use the neutralmuch term more ‘r hotic’frequent rather in this than position ‘tap’ and(38.3%). ‘trill’ However, to refer toassibilated it. rhotics are by far the most muchcommon moreTabledescribed inin frequent absoluteabsolute 2in theshows literature. infinal final this position an position position overall (49.6%), (38.3%). (49.6%), lowwhich However, which percentage is expected is assibilated expected given of rhotics thatgiventrills this arethat inis by theabsolutethis far favorable theis the most favorable final common position (4.9%) while taps are Articulatorily, standard taps and trillsenvironment2 share the for this same variant. place The and other manner non-standard of articulation;variants were less they frequent in our data: Partially inenvironment absoluteTable final2 forshows positionthis an variant. overall (49.6%), Thelow otherpercentage which non-standard is expected of trills in givenvariants absolute that werefinal this positionless is the frequent favorable(4.9%) in while our environment data:taps are Partially for muchassibilated 3more rhotic 4%; frequent approximant in 2.2%; this glotta positionl fricative 0.7%; (38.3%). and assibilated However, trill 0.3%. assibilated rhotics are by far the most are both usually realized as voiced and alveolarthisassibilatedmuch variant. more. The rhotic The frequent main other 4%; difference approximant non-standardin this position between 2.2%; variants (38.3%). glotta taps were However,l fricative and less frequenttrills assibilated0.7%; is andthat in our assibilatedrhotics data: are Partially bytrill far 0.3%. assibilated the most rhotic A closer look at the distribution of the data across factors can help us determine if there is any taps are produced with a single contactcommon 4%;betweencommon approximant thin einabsolute tip 2.2%;absolute of thefinal glottal tongueposition fricativefinal towards(49.6%), position 0.7%; which andthe assibilated alveolar is(49.6%), expected ridge, trill given 0.3%.which that thisis isexpected the favorable given that this is the favorable relationshipA closer between look at phonetic the distribution categories of and the thedata sociolinguistic across factors dimensions can help usinvestigated determine in if this there is any environmentA closer look for this at variant. the distribution The other non of the‐standard data across variants factors were less can frequent help us in determine our data: Partially if there is any while trills are produced with several (usuallyenvironmentrelationshipstudy. The two following between or forthree) conditional phoneticthis such variant. inferencecategoriesrapid contactsThe tree and (Figureother the (Hualde sociolinguistic3) non-standard illustrates 2005). a link dimensions between variants genderinvestigated were and lessin this frequent in our data: Partially assibilated rhotic 4%; approximant 2.2%; glottal fricative 0.7%; and assibilated trill 0.3%. Navarro Tomás ([1918] 1970, pp. 1, 116)relationship study.generationgave aThe more and following between their detailed effect phonetic conditional on description the categories production inference of and of the rhtree theotic Spanish sociolinguistic (Figurevariants. trill 3)A conditionalillustrates that dimensions inferencea link investigated between tree allows gender in this study. and assibilatedA closer lookrhotic at the 4%; distribution approximant of the data across 2.2%; factors glotta can helpl fricative us determine 0.7%; if there andis any assibilated trill 0.3%. Thegenerationthe visualization following and conditional oftheir complex effect inference interactions on the production tree between (Figure predictors.of3) illustratesrhotic variants. For a each link predictor,A between conditional genderits influence inference and generationin the tree allows and specifies the position of the tongue tip, dorsum,relationship and root.between He phonetic stated thatcategories the tipand of the the sociolinguistic tongue bends dimensions investigated in this theirtheoutcome visualization eAffect iscloser ontested. the ofIf productionlook thecomplex influence at ofinteractionsthe rhoticis statisticallydistribution variants. between significant A conditional predictors. of (lowthe p -value), inferencedata For each acthe treepredictor,ross data allows are factors split its the influencein visualization two can helpin the of us determine if there is any upwards to touch the upper-most part of thestudy. alveolar The following ridge, the conditional tongue rootinference retracts tree (Figuretowards 3) illustratesthe back a link between gender and complexgroups. The interactions process repeats between with predictors. each group Foruntil each there predictor, are no more its influencesignificant inpredictors. the outcome The is tested. outcomegeneration is tested.and their If theeffect influence on the production is statistically of rhotic significant variants. (lowA conditional p-value), inference the data tree are allows split in two of the oral cavity, and the tongue body adoptsrelationshipIfresulting the influencea hollowbinary issplits orbetween statistically concave along with significantshape.phonetic their respective (low categoriesp -value),p-values theare dataandplotted are thein split a tree-likesociolinguistic in two groups.figure. The The processdimensions investigated in this groups.the visualization The process of complex repeats interactions with each between group untpredictors.il there For are each no predictor,more significant its influence predictors. in the The The realization of trills requires a complexrepeatsnumbers comb with in squares eachination group are of the untilarticulatory node there numbers are movements no and more only significant serve of asthe a predictors. tongue-reference for The identification. resulting binary The ‘n’ splits along study.resultingoutcome The binaryis tested. following splits If the along influence withconditional is theirstatistically respective significantinference p-values (low areptree‐value), plotted (Figure the indata a aretree-like 3) split illustrates in figure. two The a link between gender and number on each of the terminal nodes represent the number of data points in each group (see tip and aerodynamic forces (Solé 2002). Duewithnumbersgroups. to their their inThe respective squares articulatoryprocess are prepeats-values the comp nodewith are lexity, plottedeachnumbers group trills in and a until tree-like are only there mastered serve figure. are asno a Thelatemore reference numbers significant for in identification. predictors. squares are The the The node ‘n’ generationTagliamonte and and Baayen their (2012) effectfor a more on detailed the explanationproduction of conditional of rh oticinference variants. trees). A conditional inference tree allows in the L1 acquisition and are not present innumbers the babbling and only stage. serve asNot areference only do fornative identification. speakers The find ‘n’ trills number on each of the terminal nodes thenumberresulting visualization on binary each splitsof the ofalong terminal complex with nodestheir respectiveinteractions represent p ‐thevalues number between are plotted of data inpredictors. apoints tree‐like in figure.each For group The each (see predictor, its influence in the challenging, but L2 learners may also findrepresentTagliamonte numbersthem difficult thein andsquares number Baayen to areacquire of the data(2012) node pointsand for numbers asome more in each and maydetailed group only never serveexplanation (see succeed asTagliamonte a reference of in conditional for andA—approximant identification. Baayen inference(2012 The )trees). for ‘n’ a more rolling their [r]s (Solé 2002). Because the productionoutcomedetailednumber explanation on mechanism iseach tested. of ofthe conditional terminal ofIf trillsthe nodes isinfluence inference quite represent complex trees). isthe statistically andnumber requires of data significantpointsh—aspiration in each group (low (see p -value), the data are split in two Tagliamonte and Baayen (2012) for a more detailed explanation of conditional inferenceA—approximant trees). precise articulatory and aerodynamic conditions,groups. Thesmall process changes repeatsin their productionwith each can group lead to unt ilm—partially there assibilated are no more significant predictors. The h—aspiration perceptible acoustic differences. These small changes in the production of trills, which could be due r—tapA—approximant resulting binary splits along with their respective p-valuesm—partially are assibilated plotted in a tree-like figure. The to contextual or prosodic features, create a favorable environment for the instability of this sound and rs—assibilatedh—aspiration numbers in squares are the node numbers and only server—tap as a reference for identification. The ‘n’ the creation of new variants. In fact, the rich literature on rhotic variation in different dialects of t—trillm—partially assibilated Spanish provides evidence for the instabilitynumber of trillson eachand, most of theimportantly, terminal the nodes methodological represent ts—assibilatedther—tap number rs—assibilated trill of data points in each group (see challenges they present in their analysis Tagliamonte(see Method section). and Baayen (2012) for a more detailed explanationrs—assibilatedt—trill of conditional inference trees). Acoustically, the tap and the trill share the same characteristics: A lowered third formant t—trillts—assibilated trill (Colantoni 2001) and brief periods of occlusion, one occlusion for the tap and more than one for the ts—assibilated trill A—approximant trill. However, taps and trills differ in the duration of the segment, with taps being shorter than trills. h—aspiration Quilis (1993) reported an average of 20 ms. for taps and 60 ms. for trills. The length of the trill is affected by the following vowel: It is shorter before [i] and longer before [a] (Solé 2002). m—partially assibilated Assibilated and standard rhotics 4 tend to alternate synchronically and diachronically (Solé r—tap 1992). Synchronically, apical trills exhibit non-trilled variants, taps, approximants, and fricatives. Solé

(2002) performed a study to replicate the phonological variation between voiced trills and fricatives. rs—assibilated Her results showed that trills may becomeFigure fricatives 3. Conditional (or inference assibilated) tree in R showing if the the finelydistribution controlled of phrase-final rhotic given gender and generation. t—trill articulatory or aerodynamic requirements for trills are not met, suggesting that fricatives involve a less complex articulation and allow a wider range of oropharyngeal pressure variation than trills. ts—assibilated trill Figure 3. Conditional inference tree in R showing the distribution of phrase-final rhotic given gender Thus, assibilated rhotics occur when the vibratingFigureFigure tongue-tip 3. 3. Conditional Conditional fails inference to make tree tree in in contactR Rshowing showing with the the distribution thedistribution palate, of phrase of or phrase-final ‐final rhotic rhotic given gendergiven gender and generation. andand generation. generation.

1 The distribution of trills and taps is further discussed in Hualde (2005). 2 Since there is variation affecting the system of rhotics (Hualde 2005), when we refer to ‘standard’ variants we mean prestige forms used in careful speech, prescribed by schools and used by newscasters. 3 Lipski (1994) and Hualde (2005) discussed other (non-standard) realizations including dorsalization and pre- aspiration of the trill, neutralization, retroflexion, strengthening of rhotics in codas and onset clusters. 4 Solé (1992) actually refers to word-final rhotics as “trills” but, as stated earlier, we prefer to use Hualde’s (2005) neutral term “rhotic”. Solé also refers to assibilated variants as “fricatives”.

Figure 3. Conditional inference tree in R showing the distribution of phrase-final rhotic given gender and generation. Languages 2020, 5, 38 11 of 21

Languages 2020Figure, 5, x 3FOR shows PEER thatREVIEW assibilation (rs) is most frequent in female participants of generation11 of 21 4 (older group). They record the highest levels of assibilation of all groups and the lowest rates of the standard Figure 3 shows that assibilation (rs) is most frequent in female participants of generation 4 (older variants: Trills (t) and taps (r). Because they have such high levels of assibilation, they also score a group). They record the highest levels of assibilation of all groups and the lowest rates of the standard lower overall use of other non-standard variants such as partial assibilation (m), assibilated trills (ts), variants: Trills (t) and taps (r). Because they have such high levels of assibilation, they also score a aspiration (h), and approximants (a). Middle-aged women (generations 2 and 3) present a low rate of lower overall use of other non‐standard variants such as partial assibilation (m), assibilated trills (ts), aspirationassibilation (h), and and approximants other non-standard (a). Middle variants‐aged women in their (generations speech, while 2 and they 3) have present a higher a low userate ofof taps and assibilationtrills. The and di otherfference non between‐standard the variants middle-aged in their femalespeech, groupswhile they and have the younger a higher use one of is taps that and generation 1 trills.has The less difference assibilation between overall the middle and a more‐aged frequent female groups occurrence and the of younger other variants one is that such generation as approximants 1 hasand less partially assibilation assibilated overall and rhotics. a more Their frequent behavior occurrence resembles of other that variants of the such younger as approximants male speakers who and presentpartially a assibilated varied use rhotics. of non-standard Their behavior variants resembles including that assibilated of the younger and partiallymale speakers assibilated who rhotics, presentapproximants a varied use and of non aspiration.‐standard Males variants from including generation assibilated 2 differ and from partially generation assibilated 3 in their rhotics, frequency of approximantsuse of assibilation and aspiration. with generation Males from 3 generation showing the 2 differ lowest from percentages generation of 3 assibilationin their frequency of all of groups. In use otherof assibilation words, femaleswith generation decrease 3 theirshowing use ofthe assibilation lowest percentages down the of ageassibilation scale while of all males groups. increase In their otheruse words, of assibilation. females decrease In addition, their use while of assibilation older females down present the age scale an almost while exclusivemales increase use oftheir assibilated use variants,of assibilation. younger In femalesaddition, and while males older also females use other present non-standard an almost and exclusive possibly use less of phoneticallyassibilated salient variants,variants younger including females partial and assibilation, males also use approximants, other non‐standard and aspiration. and possibly less phonetically salient variantsIn the including following partial conditional assibilation, inference approximants, tree (Figure and4 aspiration.), we explore the relationship between precedingIn the following vowels conditional (/a/, /e/, /i/ ,inference/o/)8 and tree diff erent(Figure rhotic 4), we variants explore to the determine relationship if certain between patterns of 8 precedingdistribution vowels can (/a/, be /e/, elucidated. /i/, /o/) and different rhotic variants to determine if certain patterns of distribution can be elucidated.

a—approximant

h—aspiration

m—partially assibilated

rs—assibilated rhotic

ts—assibilated trill

FigureFigure 4. Conditional 4. Conditional inference inference tree tree in inR Rshowing showing the the distribution distribution of of phrase-finalphrase‐final rhoticrhotic givengiven preceding precedingvocalic vocalic context context/a/, /e /a/,/, /i //e/,, /o //i/,. /o/.

FigureFigure 4 shows4 shows that assibilated that assibilated rhotics rhotics (rs) occur (rs) more occur frequently more frequently with /i/, /e/, with and/ i/a/,/, / ewhile/, and /o//a /, while presents/o/ presents a lower arealization lower realization of assibilated of assibilated rhotics and rhotics a higher and percentage a higher percentageof partial assibilation of partial and assibilation approximants.and approximants. GivenGiven the low the overall low overall percentage percentage of partially of partially assibilated assibilated rhotic, assibilated rhotic, assibilated trills, approximants, trills, approximants, and andaspiration, aspiration, we collapsed we collapsed all these all thesenon‐standard non-standard variants variants with assibilated with assibilated rhotics for rhotics subsequent for subsequent analysisanalysis and andstatistics. statistics. Taps Tapsand trills and trillswere werecollapsed collapsed into one into group one group called calledstandard standard variants. variants. Table Table3 3 showsshows the the distribution distribution standard standard vs. non vs.‐standard non-standard variants variants across acrossthe linguistic the linguistic and extra and‐linguistic extra-linguistic factorsfactors analyzed analyzed in this in study. this study. Factors Factors are presented are presented in order in of order importance of importance as determined as determined by the by the randomrandom forest forest analysis, analysis, which which places places the most the important most important predictors predictors at the top at and the topthe least and theimportant least important onesones at the at bottom the bottom (see page (see page14). 14).

8 Because only 3 tokens with a preceding /u/ were found in narrative and the informal speech (sur ‘south’ appeared twice and 8 Becausetour once),only 3 these tokens tokens with were a preceding excluded from/u/ were analysis. found in narrative and the informal speech (sur ‘south’ appeared twice and tour once), these tokens were excluded from analysis. Languages 2020, 5, 38 12 of 21

Table 3. Distribution of standard and non-standard (r) across linguistic and social factor groups.

Standard Non-Standard FG1: Preceding Context N % N % Total count /a/ 591 42% 804 57% 1395 /e/ 88 36% 153 64% 241 /i/ 38 21% 145 79% 183 /o/ 243 60% 161 40% 404 FG2: Task N % N % Total count Word-List 678 38% 1084 62% 1762 Narrative 153 65% 82 35% 235 Conversation 129 57% 97 43% 226 FG3: Gender N % N % Total count Female 498 35% 915 65% 1413 Male 462 57% 348 43% 810 FG4: Generation N % N % Total count Gen 1 (<20) 298 40% 448 60% 746 Gen 2 (21–35) 378 49% 395 51% 773 Gen 3 (36–55) 247 51% 241 49% 488 Gen 4 (>56) 37 17% 179 83% 216 FG5: Education N % N % Total count Level 3 (high school) 180 40% 264 60% 444 Level 4 (univ.) 780 43% 999 56% 1779 FG6: Bilingualism N % N % Total count Bilingual 474 37% 807 63% 1281 Monolingual 486 52% 456 48% 942 FG7: Location N % N % Total count El Paso 355 40% 523 60% 878 Cd. Juárez 513 49% 534 51% 1047 Chihuahua 92 31% 206 69% 298

The analysis of preceding context shows a clear relationship between mid and high front vowels and assibilation (/i/ 79% and /e/ 64%). A preceding /a/ also has a high rate of assibilation (57%), but not as high as /e/ and /i/. On the contrary, the back vowel /o/ did not yield as much assibilation as the other vowels. The overall percentage of assibilation including all three tasks was very high (56.86%). The distribution of assibilation by individual task shows that the top contributor of the high frequency of assibilation is the word-list reading task. Because the word list is considered a formal task, this result would suggest that assibilation is a prestigious feature of speech. However, the rates of assibilation in the other two tasks (narrative and conversation) appear to be contradictory, since the conversation has higher percentage of rhotic assibilation (43%) than the narrative (35%). A binary logistic regression analysis (Table4) showed that while the di fference between the reading task and the other two tasks was significant, there was no significant difference between the narrative and the conversation. The next important predictor of assibilation is gender. Our study confirms what previous literature (Amastae et al. 1998; Lope Blanch 1967; Martín Butragueño 2006, 2014; Perissinotto 1972; Mazzaro and González de Anda 2016) has reported: Female speakers present a higher percentage of assibilated Languages 2020, 5, 38 13 of 21 rhotics than males (65% vs. 43%). The significance of this and other factors was analyzed using a mixed effects binary logistic regression model in Table4.

Table 4. Binary logistic regression model coefficients with previous segment, task, and gender as predictors.

Estimate Std. Error z-Value p-Value (Intercept) 0.44 0.63 0.71 0.47 − − Preceding segment (reference level is /a/) Preceding segment /e/ 0.85 0.19 4.4 <0.001 Preceding segment /i/ 1.91 0.25 7.74 <0.001 Preceding segment /o/ 0.59 0.16 3.68 <0.001 − − Task (reference level is the Narrative) Conversation Word-List 0.43 0.26 1.69 0.09 − − 0.79 0.21 3.8 <0.001 Gender (reference level is female) Male Generation 0.82 1.04 0.79 0.43

Gender: Generation interaction (reference level is female) 0.28 0.25 1.09 0.27 Male: Generation 1.11 0.5 2.22 0.027 − − Log Likelihood AIC 2192.3 BIC 2249.4 1086.2 −

The next social factor in order of importance is ‘age’. The distribution of rhotic assibilation across generation groups presents an interesting effect with higher rates of the assibilated variant in the extreme groups (Gen1 60%; Gen2 51%; Gen3 49%; Gen4 83%). This pattern shows that assibilation is not receding, as reported by previous studies in Mexico City (Martín Butragueño 2006, 2014) and in Ciudad Juárez (Amastae et al. 1998). Figure5 compares our percentages of assibilation per generation group Languages 2020, 5, x FOR PEER REVIEW 13 of 21 with those of Amastae et al.(1998). Since Amastae et al.(1998) collected his data from sociolinguistic interviews, we compare their data with the conversation about favorite foods, which is the closest task sociolinguistic interviews, we compare their data with the conversation about favorite foods, which in nature. is the closest task in nature.

Rates of rhotic assibilation two decades apart

100 90 80 70 64 60 50 45 40 42 33 30 31

20 19 16 14 10 0 Gen 1 Gen 2 Gen 3 Gen 4

1998 2020

Figure 5. Comparison of the distribution of rhotic assibilation across generation groups in Amastae et al. Figure 5. Comparison of the distribution of rhotic assibilation across generation groups in Amastae (1998) and our 2020 data. et al. (1998) and our 2020 data.

The percentages of assibilation in the present have increased in all groups, especially in the youngest and the oldest. One reason that may account for the higher overall rates of assibilation in our data could be that our sample of participants also included Spanish speakers from the rest of the state of Chihuahua and El Paso, which have higher percentages of assibilation than speakers from Ciudad Juárez do. Besides overall percentages, it is important to examine what each generation is doing. Amastae et al.’s generation 3, which in 1998 had the highest proportion of assibilation, is now our generation 4, also with the highest rate of assibilation. It is important to point out that our generation 4 contains only female participants, which may also account for the apparent present day increase of assibilation in this particular generation, since female speakers are the most frequent users of assibilation. Amastae et al.’s generation 2, which in 1998 had the lowest percentage of assibilation, is now our generation 3, also with the lowest rate of assibilation. So, the higher peak and the lower dip have moved with the speakers, i.e., they have not remained static with the generation. The fact that assibilation is appearing in younger generations including males (as shown in Figure 3) may suggest that the variable is going through a revival process. However, the variability found is more complex than we had expected, because we discovered a vast array of variants co‐occurring with the assibilated rhotic. Although an acoustic analysis would help elucidate this trend more accurately (Mazzaro forthcoming), our spectrographic examination of the tokens showed that the variant used by older women in generation 4 is more phonetically salient (higher concentration of noise and longer duration), whereas the variants used by younger people are more subtle and context dependent. The effect of education is not very strong, as shown by small differences in percentages of assibilation in participants with university education (56%) and high school (60%). In fact, this was the last factor in order of importance listed by the random forest (Figure 6). We believe that the reason for the lack of effect in education could be that the participants in this study belonged to the two highest levels of education. Future studies should examine the influence of education on rhotic assibilation in participants with lower levels of formal education. The effect of bilingualism was analyzed despite the fact that our bilingual subjects are native Spanish speakers and received most of their formal education in Spanish. We wanted to make sure that the variation observed in our data was not the result of a cross‐linguistic influence between English and Spanish. Thus, we compared bilingual speakers’ data with data collected from Languages 2020, 5, 38 14 of 21

The percentages of assibilation in the present have increased in all groups, especially in the youngest and the oldest. One reason that may account for the higher overall rates of assibilation in our data could be that our sample of participants also included Spanish speakers from the rest of the state of Chihuahua and El Paso, which have higher percentages of assibilation than speakers from Ciudad Juárez do. Besides overall percentages, it is important to examine what each generation is doing. Amastae et al.’s generation 3, which in 1998 had the highest proportion of assibilation, is now our generation 4, also with the highest rate of assibilation. It is important to point out that our generation 4 contains only female participants, which may also account for the apparent present day increase of assibilation in this particular generation, since female speakers are the most frequent users of assibilation. Amastae et al.’s Languagesgeneration 2020, 2,5, whichx FOR PEER in 1998 REVIEW had the lowest percentage of assibilation, is now our generation 3, also14 of with 21 the lowest rate of assibilation. So, the higher peak and the lower dip have moved with the speakers, i.e., monolingualthey have not subjects. remained There static was with a higher the generation. rate of assibilation The fact in that bilinguals assibilation (63%) is than appearing in monolinguals in younger (48%),generations but the includingrandom forest males analysis (as shown gave in this Figure factor3) less may weight suggest after that others the variable were considered is going (Figure through 6).a In revival other process.words, although However, bilingual the variability speakers found had a ishigher more percentage complex than of assibilation we had expected, in their speech, because bilingualismwe discovered does a vastnot seem array to of significantly variants co-occurring influence withthe occurrence the assibilated of the rhotic. non‐standard Although variant. an acoustic analysisThe distribution would help elucidateof assibilation this trend across more geographical accurately (Mazzaroareas is intriguing,), our spectrographic especially examination because we of findthe higher tokens rates showed of assibilation that the variant in Chihuahua used by older (capital women and interior in generation combined) 4 is more than phonetically in Ciudad Juárez salient (69% and 51%, respectively), with El Paso somewhere in between (60%). We suspect that other factors (higher concentration of noise and longer duration), whereas the variants used by younger people are could be interacting with location—age, for example. While the age distribution of the subjects from more subtle and context dependent. Chihuahua was fairly even, the subjects from El Paso and Ciudad Juárez were on average younger. The effect of education is not very strong, as shown by small differences in percentages of As stated earlier, older speakers have higher rates of assibilation than all other groups, so having assibilation in participants with university education (56%) and high school (60%). In fact, this was the more older speakers may have inflated the percentages found in Chihuahua. The random forest last factor in order of importance listed by the random forest (Figure6). We believe that the reason for analysis (Figure 6) shows that location is not a very strong predictor of assibilation once other factors the lack of effect in education could be that the participants in this study belonged to the two highest are considered. Nevertheless, we believe that the influence of geographical space on the influence of levels of education. Future studies should examine the influence of education on rhotic assibilation in rhotic assibilation should be re‐examined with a more balanced distribution of subjects across participants with lower levels of formal education. geographical locations.

Figure 6. Conditional permutation variable importance for the random forest with all predictors of Figure 6. Conditional permutation variable importance for the random forest with all predictors of rhotic assibilation. rhotic assibilation. The effect of bilingualism was analyzed despite the fact that our bilingual subjects are native The random forest indicates that the most important predictor of rhotic assibilation is the Spanish speakers and received most of their formal education in Spanish. We wanted to make sure that individual speaker. This is due to the high degree of variability in the production of assibilated rhotics. For instance, in the narrative and the conversation, the rate of assibilation per individual ranged from 0% to 100%. Appendix B includes a table with number of tokens and percentage of assibilation per speaker in the narrative and the conversation combined. Besides individual speakers, the next most important factors are preceding context, task, gender, generation, location, bilingualism, and education. The influence of each of these factors on the occurrence of the variable under study has been discussed above. Finally, we conducted several mixed‐effects binomial logistic regression models fitted to the data to determine the most important predictors. Factors were added one at a time, until the model did not improve anymore. Table 4 shows that beyond the individual speaker, the most decisive predictors of rhotic assibilation are: Previous segment, task, and gender with an interaction with generation. Languages 2020, 5, 38 15 of 21 the variation observed in our data was not the result of a cross-linguistic influence between English and Spanish. Thus, we compared bilingual speakers’ data with data collected from monolingual subjects. There was a higher rate of assibilation in bilinguals (63%) than in monolinguals (48%), but the random forest analysis gave this factor less weight after others were considered (Figure6). In other words, although bilingual speakers had a higher percentage of assibilation in their speech, bilingualism does not seem to significantly influence the occurrence of the non-standard variant. The distribution of assibilation across geographical areas is intriguing, especially because we find higher rates of assibilation in Chihuahua (capital and interior combined) than in Ciudad Juárez (69% and 51%, respectively), with El Paso somewhere in between (60%). We suspect that other factors could be interacting with location—age, for example. While the age distribution of the subjects from Chihuahua was fairly even, the subjects from El Paso and Ciudad Juárez were on average younger. As stated earlier, older speakers have higher rates of assibilation than all other groups, so having more older speakers may have inflated the percentages found in Chihuahua. The random forest analysis (Figure6) shows that location is not a very strong predictor of assibilation once other factors are considered. Nevertheless, we believe that the influence of geographical space on the influence of rhotic assibilation should be re-examined with a more balanced distribution of subjects across geographical locations. The random forest indicates that the most important predictor of rhotic assibilation is the individual speaker. This is due to the high degree of variability in the production of assibilated rhotics. For instance, in the narrative and the conversation, the rate of assibilation per individual ranged from 0% to 100%. AppendixB includes a table with number of tokens and percentage of assibilation per speaker in the narrative and the conversation combined. Besides individual speakers, the next most important factors are preceding context, task, gender, generation, location, bilingualism, and education. The influence of each of these factors on the occurrence of the variable under study has been discussed above. Finally, we conducted several mixed-effects binomial logistic regression models fitted to the data to determine the most important predictors. Factors were added one at a time, until the model did not improve anymore. Table4 shows that beyond the individual speaker, the most decisive predictors of rhotic assibilation are: Previous segment, task, and gender with an interaction with generation. The binary logistic regression model shows that preceding context, especially /i/, favors rhotic assibilation. A following /e/ is also a favoring environment for assibilation to occur. However, a preceding /o/ disfavors the occurrence of assibilation. These preceding contexts were compared to the reference vowel /a/. The next factor that predicted assibilation was task, particularly the word list, which promotes assibilation compared to the reference narrative. The conversation is not significantly different from the narrative. Gender and generation are not shown as significant predictors by themselves; however, their interaction proves to be significant. The coefficient estimate shows that the older male speakers disfavor assibilation. In addition, we know from the conditional inference tree (Figure2) that older females have the highest rates of assibilation from all groups.

4. Discussion and Conclusions This study was designed to explore different aspects of the variation affecting rhotics in phrase-final position in the Chihuahua/El Paso speech community. In this section, we address the questions posed at the beginning of the study. The first question concerned the current state of rhotic assibilation in Chihuahua Spanish. We compared our study with Amastae et al.’s (1998) and contrary to our expectations, we found that rhotic assibilation is not receding, but it is quite frequent in the population and, particularly, in the younger groups of both genders. In general, assibilation continues to be more prevalent amongst females, but male speakers assibilate, too. In fact, as the title of this study suggests, we consider the frequency of assibilation in male speakers a sign that the variable is losing its feminine connotation and becoming a variant available across genders. So, if assibilation is not receding, are we seeing a revival of the assibilated rhotic imported from Mexico City in the 1960s (Amastae et al. 1998)? We argue that the Languages 2020, 5, 38 16 of 21 variation adopted by the younger groups seems to have lost its social connection with femininity and sophistication; i.e., it may not be the same as the original variant from the 1960s. As shown by the present results, young speakers of both genders use the fully assibilated rhotic, but younger males are also using a more diverse set of variants, which are more ‘subtle’ (less salient) than the fully assibilated rhotic. Although we are currently working on the results of the acoustic analysis to confirm this statement (Mazzaro), we observed that the variants that alternate with the fully assibilated rhotic (partially assibilated rhotic, approximants, glottal fricatives, assibilated trills) have shorter durations and lower intensity than the fully assibilated one. We speculate that there is a closer connection between the fully assibilated rhotics found in older females and the variable imported from Mexico City, while the variation found in younger generations is more phonetically motivated. There might be a socio-economic explanation for the difference between the variation found in younger and older groups. Ciudad Juárez has been considered a place of opportunity due to the many jobs created by international companies and factories (known as maquiladoras) that boomed in the 1990s, when NAFTA was approved. Since the year 2000, immigration rates to the city increased considerably as a result of the arrival of immigrants from other regions (Cruz 2012) looking for work. The improved economic situation of the people from Ciudad Juárez vis-à-vis the rest of the country, may have increased the local pride in being from Ciudad Juárez, which may favor the adoption of local features of speech, rather than linguistic innovations from abroad. Our second question concerned the type of rhotic variants found in Chihuahua Spanish. We based our analysis on a combination of auditory transcription and spectrographic inspection of tokens. As hypothesized, the standard variants (taps and trills) are the most frequently used variants across participants. Besides the standard rhotics, we identified five non-standard variants: Assibilated rhotics, approximants, partially assibilated rhotics, aspiration, and assibilated trills; the last three variants had not been initially anticipated. We explored whether these variants were also influenced by social factors, as the fully assibilated rhotic was. As already mentioned, we found that while older females preferred the fully assibilated variant, younger generations, and particularly men, had a wider variety of variants in their speech. The results presented here confirm what Tagliamonte(2011) suggested, “everyone marks their place in the social hierarchy in phonetically distinguishable ways.” Our third question explored the most important linguistic and social correlates of rhotic assibilation. Our analysis based on random forest and binary logistic regression models indicated that beyond the individual speaker, which is the highest source of variability in the data, the preceding context is the most important factor influencing the variable under study. A preceding /i/ and /e/ significantly favored the occurrence of rhotic assibilation. Conversely, a preceding /o/ disfavored assibilation and had a wider range of variants co-occurring with it. We believe that the highest rate of assibilation next to /i/ and /e/ could be explained in articulatory terms. To produce /i/ and /e/ the dorsum of the tongue moves up and toward the front. However, to realize a trill or a tap the position of the tongue root retracts towards the back of the oral cavity, and the tongue body adopts a hollow or concave shape that allows the tongue tip to cup towards the alveolar ridge to produce a brief contact (tap). Because the position of a high-front vowel requires the tongue dorsum and root to go up, while the trill or tap requires the tongue root to stay down and retract, the co-articulation between these two sounds may cause modifications in the production of the rhotic. According to Solé (2002) small changes in aerodynamic conditions that govern the production of the trill may result in a fricative sound. The fact that assibilation is more common in phrase-final position can also be phonetically motivated, as it results from the difficulty of sustaining trilling with the lowered decreased subglottal pressure that occurs at the end of a statement (Solé 2002). With regard to the most important social correlates of rhotic assibilation, we confirmed our hypothesis that gender and age would significantly affect the variant under study. Concerning gender, our results agreed with those of previous studies that females favor assibilation more than males. In reference to age, we found that older female speakers tend to produce more assibilated rhotics that the other groups. However, the finding that youngest female and male speakers also had high percentages of assibilation was not anticipated by our hypothesis. Languages 2020, 5, 38 17 of 21

The fourth question examines whether language contact with English affected the occurrence of this variable in bilingual speech. The random forest analysis placed the ‘bilingualism’ factor in penultimate position, showing that it was one of the least important factors affecting rhotic assibilation. In addition, the binary logistic regression model stopped improving before adding this predictor. Our hypothesis was confirmed, as we did not find a significant difference in the rate of production of rhotic assibilation in bilinguals compared to monolinguals. A future study exploring how the other factors impinge on the realization of assibilated rhotics across bilingual vs. monolingual groups could help us understand finer grained differences depending on sociophonetic competence. Concerning the effect of style (reading vs. narrative vs. conversation) on the occurrence of assibilation, we found that the more formal reading task (word list) favored assibilation compared to the narrative and the conversation9. However, there was no significant difference in rate of assibilation between the narrative and the conversation. This was an interesting finding, as we expected the assibilation rate to be higher in the narrative, which was considered more formal than the conversation. In other words, style seems to play a role when comparing reading vs. other tasks, but there does not seem to be a special sensitivity of participants to the difference between the narrative and the conversation. One possibility could be that speakers did not consider the narrative with visual prompts more formal than chatting about their favorite food. Another possibility could be related to the low perception and sociolinguistic awareness of rhotic assibilation in this population (Mazzaro and González de Anda 2019). Since rhotic assibilation seems to be below the level of consciousness and social awareness, we would not expect a high degree stylistic shift within the more casual tasks. The final question concerns the status of assibilation as a prestigious feature of speech. The significantly higher rate of assibilation in the formal reading task suggests that rhotic assibilation continues to be a prestigious feature, as stated in our hypothesis. In addition, we found women to be ahead of men in the use of rhotic assibilation, which supports the sociolinguistic principle proposed by Labov(2001) that in changes from below the level of consciousness awareness, women tend to lead men in their use of prestigious variants. However, in order to obtain a more comprehensive picture of the status of rhotic assibilation in Chihuahua Spanish, we need to supply this analysis with information about social class. When talking about gender differentiation, Labov(2001) argued that the behavior of women is far from uniform across the speech community and that there is an intimate and complex interaction between style, gender, and social class. The results of this study clearly show that women are the leaders of this linguistic change, but it also demonstrates that younger men are adopting the change. We find men’s use of a wider set of variants particularly intriguing and, as stated earlier, we need to examine whether they have achieved the status of sociolinguistic variants and the social meanings they contain. To conclude, the purpose of this study was to examine the current state of rhotic assibilation in Chihuahua Spanish two decades after Amastae et al.(1998) conducted the first study. The results of our analysis reveal that the feature is frequent among the youngest groups and it does not show signs of disappearing as previously stated (Amastae et al. 1998; Martín Butragueño 2006; Perissinotto 1972). On the contrary, our study found much higher rates of rhotic assibilation than Amastae et al. did in 1998, both overall and across age groups. Despite the high degree of variability across individual speakers, we found that specific linguistic (preceding context) and extra-linguistic factors (style, gender, and generation) are strong predictors of the variable in question. Rhotic assibilation is phonetically motivated by phrase-final position and preceding non-low front vowels /i/ and /e/, which cause the tongue body to move up and front making it more difficult to sustain trilling. As stated earlier, the production of a trill requires fine articulatory and aerodynamic conditions that are highly sensitive to contextual and prosodic conditions. Imperfect

9 One of the reviewers suggested that reading differs in many ways from the other tasks, in that it involves other cognitive processes and makes subjects focus more on pronunciation because of the lack of content. In other words, the difference between the reading task and the other two tasks could be due to the formality plus the additional cognitive processes involved in reading. Languages 2020, 5, 38 18 of 21 articulatory positioning and changing aerodynamic conditions can lead to acoustic variations that provide favorable grounds for assibilation to occur. An interesting finding of this study is that the variation does not entail a simple binary opposition between assibilated and non-assibilated rhotic, but that the variable involves a more complex array of variants (partially assibilated rhotics, approximants, glottal fricatives, assibilated trills). Interestingly, men are the main users of these variants, which we speculate are less salient (or weaker) forms of rhotic assibilation. The next step of this investigation is to conduct an acoustic analysis of these fine-grained, sub-phonemic variations in pronunciation and to determine whether they have social meaning.

Author Contributions: Conceptualization, N.M. and R.G.; Data curation, N.M. and R.G.; Investigation, N.M. and R.G.; Methodology, N.M. and R.G.; Project administration, N.M. and R.G.; Resources, N.M.; Supervision, N.M.; Writing—original draft, N.M.; Writing—review & editing, R.G. Both authors have read and agreed to the published version of the manuscript. Funding: This research received no external funding. Acknowledgments: The authors wish to thank Martin Lazzari for his help with the statistical analysis, Laura Colantoni, Jon Amastae and Richard Teschner for their comments on earlier drafts of this paper. We wish to express our appreciation to two anonymous reviewers for their thorough review of this paper and their clever recommendations. Last, but not least, a big thank you to all the participants in this study. Errors and omissions are entirely our responsibility. Conflicts of Interest: The authors declare no conflict of interest.

Appendix A Stimuli used were content words with a rhotic in coda final position and preceded by /a/, /e/, /i/, or /o/: Bazar, billar, dejar, besar, guerrear, guisar, gozar, beber, duchar, poder, datar, gustar, pedir, podar, dador, pagar, seguir, donar, botar, casar, pasar, pesar, pitar, posar, telar, tirar, tocar, quemar, copar, sopor, sentir, sacar, curar, tutor, pudor, ayer. Distracters (inserted randomly between the words listed above): Bicho, cerro, arresta, es hembra, es cebra, pared, horrendo, cero, paré, reza, resta, red, renta, cerca, drogui, círculo, terso, dicho, tribu, tropel, virtud, champú, cura.

Appendix B Number of tokens with rhotics in phrase-final position, realized as standard rhotic (tap or tril), assibilated and percentage of rhotic assibilation per individual subject.

Table A1. Number and percentage of realizations of the standard and assibilated by individual speaker.

Speaker Number of Tokens Standard Assibilated Percent Assibilation UT007 21 9 12 57.1% UT030 3 1 2 66.7% UT032 24 24 0 0.0% UT034 5 2 3 60.0% UT035 13 11 2 15.4% UT036 10 5 5 50.0% UT037 6 4 2 33.3% UT038 5 3 2 40.0% UT039 8 7 1 12.5% UT040 4 2 2 50.0% UT041 5 4 1 20.0% UT060 2 1 1 50.0% UT061 9 4 5 55.6% UT062 10 7 3 30.0% UT067 10 6 4 40.0% UT070 10 10 0 0.0% Languages 2020, 5, 38 19 of 21

Table A1. Cont.

Speaker Number of Tokens Standard Assibilated Percent Assibilation UT072 10 8 2 20.0% UT073 4 3 1 25.0% UT074 8 5 3 37.5% UT076 1 1 0 0.0% UT077 9 6 3 33.3% UT079 16 5 11 68.8% UT080 2 1 1 50.0% UT081 3 0 3 100.0% UT082 8 0 8 100.0% UT083 9 4 5 55.6% UT084 8 5 3 37.5% UT085 10 10 0 0.0% UT086 9 6 3 33.3% UT087 6 6 0 0.0% UT088 10 8 2 20.0% UT089 4 2 2 50.0% UT090 12 8 4 33.3% UT091 6 1 5 83.3% UT092 6 2 4 66.7% UT093 9 4 5 55.6% UT094 7 6 1 14.3% UT096 5 4 1 20.0% UT097 7 2 5 71.4% UT098 3 1 2 66.7% UT099 5 4 1 20.0% UT100 6 0 6 100.0% UT101 12 8 4 33.3% UT102 9 4 5 55.6% UT103 2 1 1 50.0% UT104 5 5 0 0.0% UT105 17 10 7 41.2% UT106 6 2 4 66.7% UT107 15 12 3 20.0% UT108 5 0 5 100.0% UTCJ1 3 1 2 66.7% UTCJ3 11 6 5 45.5% UTCJ4 7 6 1 14.3% UTCJ5 11 3 8 72.7% UTCJ6 8 6 2 25.0% UTCJ7 5 4 1 20.0% UTCJ8 17 12 5 29.4% Total 461 282 179 38.8%

References

Amastae, Jon, Alberto Escalera, Araceli Arceo, Zenia Gutiérrez, Homero Silva, and Bertha Talamantes. 1998. Cambio desde abajo, cambio desde arriba. Memorias del IV Encuentrio Internacional de Linguistica en el Noroeste 2: 269–80. Bates, Douglas, Martin Maechler, and Ben Bolker. 2011. Lme4: Linear Mixed-Effect Models Using S4 Classes. R Package Version 0.999375-42. Available online: http//CRAN.R-project.org/package=lme4 (accessed on 10 October 2020). Blecua, Beatriz, and Jordi Cicres. 2019. Rhotic variation in Spanish codas: Acoustic analysis and effects of context in spontaneous speech. In Romance Phonetics and Phonology. Oxford: Oxford University Press. Boersma, Paul, and David Weenink. 2018. Praat: Doing Phonetics by Computer [Computer Program]. Version 6.0.43. Available online: http://www.praat.org/ (accessed on 10 October 2020). Languages 2020, 5, 38 20 of 21

Bradley, Travis G. 2004. Gestural timing and rhotic variation in Spanish. In Laboratory approaches to Spanish phonology 7: 195. Bradley, Travis G., and Erik W. Willis. 2012. Rhotic variation and contrast in Veracruz Mexican Spanish. Estudios De Fonética Experimental 21: 43–74. Canfield, D. Lincoln. 1981. Spanish Pronunciation in the Americas. Chicago: University of Chicago Press. Carranza, Luz Marina Vásquez. 2006. On the Phonetic Realization and Distribution of Costa Rican Rhotics. Revista De Filología y Lingüística De La Universidad De Costa Rica Vol.32 Núm.2 2007. Available online: https://www.researchgate.net/publication/315716488_On_the_phonetic_realization_and_distribution_of_ Costa_Rican_rhotics (accessed on 10 October 2020). Colantoni, Laura Marcela. 2001. Mergers, Chain Shifts and Dissimilatory Processes: Palatals and Rhotics in Argentine Spanish. Master’s dissertation, University of Minnesota, Minneapolis, MN, USA. Languages 2020, 5, x FOR PEER REVIEW 4 of 21 Colantoni, Laura. 2006. Increasing periodicity to reduce similarity: An acoustic account of deassibilation in rhotics. Paper presented at the 2nd Conference on Laboratory Approaches toselected Spanish as Phoneticssignificant and by Phonology,the multivariate analysis and the author reported very few cases of Somerville, MA, USA, September 17–19; pp. 22–34. assibilation in positions other than absolute final and in consonant cluster /tr/. Cruz, Rodolfo. 2012. Cambios fronterizos y movimientos migratorios en la fronteraMartín norte Butragueño de México. (2014) In presentedMéxico some spectrograms of trill and taps to illustrate the different ante los Recientes Desafíos de la Migración Internacional Mexico City. Editedtypes of by rhotics T. Ram foundírez Garcin Mexicoía and M.City Spanish. Besides the standard variants, he also found Ángel Castillo. Mexico City: Consejo Nacional de Población, pp. 157–84.approximant variants, which the author defined as weakened versions of standard rhotics, and assibilated rhotics. He proposed that rhotic assibilation be described as change from an approximant Dalola, Amanda, and Barbara E. Bullock. 2017. On Sociophonetic Competence: Phrase-final Vowel Devoicing in to a fricative sound. According to the author, this becomes evident in the cases where the rhotic starts Studies in Second Language Acquisition Native and Advanced L2 Speakers of French. as an approximant39: 769–99.(with a [clearCrossRef sonority] bar and no noise) and continues as a fricative with high Eller, Wendianne Alice. 2013. Sociolingüística del Español gay Mexicano: Variacifrequencyón Fó nica,noiseEstereotipos, and an almost Creencias inexistent F0. However, as demonstrated by Solé (2002) in the y Actitudes en una Red Social de Hombres Homosexuales. Mexico City: Universidadlaboratory, there Nacional is compelling Autónoma cross-linguistic de evidence suggesting that trills alternate with fricatives México. (or assibilated variants), so it is plausible that assibilated rhotics result from “a failure of sustaining Henriksen, Nicholas C., and Erik W. Willis. 2010. Acoustic characterizationtrilling of with phonemic the lowered trill production decreased subglottal in pressure that occurs at the end of a statement” (Solé Jerezano Andalusian Spanish. Paper presented at the 4th Conference on2002). Laboratory Approaches to Spanish Phonology, Somerville, MA, USA, January 1; pp. 115–27. 1.3. Sociolinguistic Literature on Rhotic Assibilation in the State of Chihuahua, Mexico Howson, Phil, Ekaterina Komova, and Bryan Gick. 2014. Czech trills revisited: An ultrasound EGG and acoustic Rhotic assibilation reached the north of Mexico after the 1960s, with the first study conducted in study. Journal of the International Phonetic Association 44: 115–32. [CrossRef] Ciudad Juárez in 1998 (Amastae et al. 1998). This study showed that assibilation was present in Hualde, José Ignacio. 2005. The Sounds of Spanish with Audio CD. Cambridge: Cambridge University Press. Ciudad Juaréz, but with a lower frequency of overall occurrence than reported in Mexico City. The Labov, William. 1972. Sociolinguistic Patterns. Philadelphia: University of Pennsylvaniapercentage of Press.assibilation in all final positions (including word final and phrase-final) was 6%, while Labov, William. 2001. Principles of Linguistic Change Volume 2: Social Factorsthe. percentage Language of in assibilation Society. Hoboken: in absolute final position was 22%. Compared to the 68.1% reported by Wiley-Blackwell. Perissinotto (1972), assibilation was much less frequent in Ciudad Juárez. Amastae et al. (1998) Ladefoged, Peter, and Ian Maddieson. 1996. The Sounds of the World’s Languagesconducted. Oxford: a multivariate Blackwell. analysis on 72 speakers and found that assibilation was affected by: (1) Lipski, John M. 1994. Latin American Spanish. Boston: Addison-Wesley. Gender: Higher rate of assibilated rhotics in women (women 28% and men 16%); (2) social class: Greater percentage of assibilation in the higher socio-economic class as compared to the middle class5 Lope Blanch, Juan M. 1967. La influencia del sustrato en la fonética del español de México. Revista De Filología (21% and 18%, respectively); (3) age: Adults (36–55 years) and older participants (56+) showed higher Española 50: 145. [CrossRef] rates of assibilation (31% and 19%, respectively) than younger participants (15%); and (4) education: Lubbers, Mart, and Francisco Torreira. 2016. Praatalign: An Interactive PraatAssibilation Plug-in was for Performingmore frequent Phonetic in speakers with higher levels of formal education (university 23% Forced Alignment. A Detailed Manual for Version 1.8. Available online:and https: high// www.scribd.comschool 14%). All these/document effects/ suggested that assibilation was also a prestigious variant in 399059498/Praat-Plugin-Book-2-0 (accessed on 7 August 2020). Ciudad Juárez. Amastae et al. (1998) argued that assibilation in Ciudad Juárez was a recent “change Martín Butragueño, Pedro. 2006. Líderes lingüísticos en la Ciudad de México.from In above”Líderes (Labov Lingüísticos: 1972) Estudiosimported De from Mexico City by the higher classes and transmitted by Variación y Cambio. Mexico City: El Colegio De México, pp. 158–208. [CrossRefwomen. According] to Labov (1972), a change from above is a linguistic change that enters the language from above the level of consciousness; that is, speakers are generally aware of the linguistic Martín Butragueño, Pedro. 2014. Fonología Variable del Español de México.México City: El Colegio de México. form and they manipulate its use depending on the context and/or their interlocutors. The upper Mazzaro, Natalia. Forthcoming. Men are doing it differently: Acoustic analysis of rhotic assibilation in Chihuahua classes use these new linguistic forms in order to differentiate themselves from the lower classes, Spanish. while lower classes use these forms in order to sound more formal and similar to the upper classes. Mazzaro, Natalia, and Raquel González de Anda. 2016. A receding variation: RhoticTwenty assibilation years after Amastae’s in Chihuahua study, Mazzaro and González (2019) investigated the relationship Spanish. In Hispanic Linguistic Symposium 2016. Washington, DC: Georgetownbetween the University. perception and production of rhotic assibilation [ř] and deaffrication of the voiceless Mazzaro, Natalia, and Raquel González de Anda. 2019. The perception-productionpost-alveolar connection: affricate/ [t ʃ /] deain theffrication Spanish of Chihuahua, Mexico. Thirty-three native Spanish speakers and rhotic assibilation in Chihuahua Spanish. In Issues in Hispanic andfrom Lusophone the state Linguistics of Chihuahua series: completed Recent production and perception tasks to establish whether those that produced the variants were able to perceive them. The production data were elicited by asking Advances in the Study of Spanish Sociophonetic Perception. Edited by Whitney Chappell. Amsterdam: John participants to narrate the fairy tale Caperucita Roja ‘Little Red Riding Hood’ and by asking them to Benjamins Publishing Company. talk about their favorite food. Results showed that the overall percentage of [ř] production was Milroy, Lesley. 1987. Language and Social Networks. Oxford: Blackwell. greater than that of [ʃ] (17.15% and 11.83%, respectively). The authors explained that the increased Morgan, Terrell A., and Sandro Sessarego. 2016. A phonetic analysis of intervocalicoverall/r rate/ in highlandof [ř] might Bolivian be due to Spanish. assibilation only being investigated in absolute final position, which Spanish in Context 13: 195–211. [CrossRef] is the context where most of the variation occurs. The social factors that turned out to be significant were gender and generation, with female and younger speakers favoring the use of assibilation. The relationship between production and perception of assibilation was very subtle; overall, the perception of the variant was very low (9.1%). Those who produced assibilation were not the ones

5 Amastae et al. (1998) did not analyze the speech of participants of lower socio-economic class. Languages 2020, 5, 38 21 of 21

Perissinotto, Giorgio S. A. 1972. Distribución demográfica de la asibilación de vibrantes en el habla de la ciudad de México. Nueva Revista De Filología Hispánica 21: 71–79. [CrossRef] Quilis, Antonio. 1981. Fonética Acústica de la Lengua Española. Madrid: Gredos. Quilis, Antonio. 1993. Tratado de Fonología y Fonética Españolas. Madrid: Gredos. R Core Team. 2020. R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing, Available online: https://www.R-project.org/ (accessed on 10 October 2020). Rissel, Dorothy A. 1989. Sex, attitudes, and the assibilation of /r/ among young people in San Luis Potosí, Mexico. Language Variation and Change 1: 269–83. [CrossRef] Silva, Adelaide Hercilia Pescatori. 1996. Para a Descriçao Fonético-Acústica das Líquidas no Português Brasileiro: Dados de um Informante Paulistano. Master’s dissertation, University of Campinas, Campinas, Brazil. Solé, Maria-Josep. 1992. Experimental phonology: The case of rhotacism. Phonologica 1988: 259–71. Solé, Maria-Josep. 2002. Aerodynamic characteristics of trills and phonological patterning. Journal of Phonetics 30: 655–88. [CrossRef] Strobl, Carolin, Boulesteix Anne-Laure, T. Kneib, Augustin Thomas, and Zeileis Achim. 2008. Conditional Variable Importance for Random Forests. BMC Bioinformatics 9: 307. Available online: http://www.biomedcentral. com/1471--2105/9/307 (accessed on 10 October 2020). Tagliamonte, Sali. 2011. Variationist Sociolinguistics: Change, Observation, Interpretation. Hoboken: John Wiley & Sons, vol. 39. Tagliamonte, Sali A., and R. Harald Baayen. 2012. Models, forests and trees of York English: Was/were variation as a case study for statistical practice. Language Variation and Change 24: 135–78. [CrossRef] Tomás, Tomás Navarro. 1970. Manual de Pronunciación Española. Madrid: CSIC. First published 1918. Willis, Erik W. 2007. An acoustic study of the ‘pre-aspirated trill’ in narrative Cibaeño Dominican Spanish. Journal of the International Phonetic Association 37: 33–49. [CrossRef]

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).