Buenas no[tʃ]es y mu[ts]isimas gracias: A Sociophonetic Study of the Alveolar Affricate in Peninsular Spanish Political Speech

Abstract

While variation in the southern Peninsular Spanish affricate /tʃ/ has been considered in the context of deaffrication to [ʃ], this study examines an emergent variant [ts] in the context of sociolinguistic identity and style in political speech. Based on a corpus of public speech from Madrid and Andalusia, Spain, this study examines the phonetic and sociolinguistic characteristics of the affricate, finding variation in the quality of the frication portion of the segment through an analysis of segment duration (ms), the center of gravity (Hz), and a categorical identification of realization type. The results suggest that both linguistic variables, like phonetic environment, stress, lexical frequency, and following vowel formant height, as well as extralinguistic variables, like speaker city, gender, political affiliation, and speech context, condition use. Based on these findings, it appears that production of the alveolar affricate [ts] is an incipient sociolinguistic marker in the process of acquiring social meaning. It is particularly associated with female speech and prestige norms that transcend regional identification. This alveolar variant serves as an additional sociolinguistic resource accessible for identity development among politicians and offers insight into ongoing change in the affricate inventory of southern and northern-central Peninsular Spanish.

1. Introduction

Political speech offers insight into the agency and intentions of politicians, while at the same time highlighting resources that speakers can employ as part of their process of identity construction. The indexicality vested in certain terms is reflected in the persona that these individuals develop on a national stage. Given the public nature of their language use, the ways in which politicians express solidarity, converge with certain groups and diverge from others is often subject to close scrutiny, offering insight into the meaning of these differing forms. The current study selects a particular sociolinguistic phenomenon that undergoes a process of phonetic differentiation in the speech of certain politicians, namely the Spanish affricate /tʃ/, in the productions of politicians from both Madrid and Andalusia, Spain. This sociophonetic analysis does not only focus on the linguistic and extralinguistic factors conditioning variable use: it also performs an acoustic analysis on an emergent variant, the alveolar [ts], that has received little mention in descriptions of Peninsular Spanish (e.g., Samper-Padilla 2011) and only recent attention in perceptual research.1

Sociolinguistic style develops not just from the words speakers use, or the phonetic forms that they select, but also from the cloud of culturally charged items with which they surround themselves. All of these tools for self-representation, termed “bricolage” by Eckert (2008), serve to distinguish individuals in an indexical web of meaning, as described by Silverstein (2003). Specifically, in the realm of political speech, this can mean aligning oneself with an ideological stance or a certain subset of voters by using regional variants associated with the working class or rural speakers to show solidarity (e.g., Hernández-Campoy and Cutillas-Espinosa 2013; Pollock and Wheeler 2022; Holliday 2017), or by contrasting with one’s opponents through divergence in form usage (Cruz-Ortiz 2022; Pollock 2023a; Pollock, forthcoming). At the same time, politicians navigate personal motivations and party-based norms to develop a coherent identity in the face of political opposition (e.g., Hall-Lew et al. 2012; Hall-Lew et al. 2017; Pollock and Wheeler forthcoming).

Sibilant fricatives and affricates have been studied broadly within Romance (Recasens and Espinosa 2007) and other typologically distinct languages (Gordon et al. 2002). Within Peninsular Spanish, several variable phenomena have been identified that are relevant to the Spanish post-alveolar affricate, although focus is usually placed upon deaffrication in the speech of elderly, rural speakers (e.g., Henríquez-Barahona and Fuentes-Grandón 2018; Herrero de Haro 2017b; Samper-Padilla 2011). Alongside production of /tʃ/ (e.g., in hecho [hetʃo] ‘made/did’) as the deaffricated Andalusian [ʃ] (e.g., [heʃo]; Villena-Ponsoda 2013), researchers have also identified a voiced post-palatal [dʒ] in the Canary Islands (e.g., [hedʒo]; Almeida 2019) and a voiced pre-palatal [ʤ] in Murcia (e.g., [heʤo]; Torrano Moreno 2017). For the most part, however, the prototypical production of middle-aged urban speakers in central and southern Spain has been described as unvaryingly post-alveolar [tʃ] (Hualde 2005). The current study, which examines the use of a voiceless alveolar affricate [ts], also considers the distinction between a sociolinguistic “indicator”, which Labov (1972, p. 237) describes as having no social value, and a “marker”, which has acquired social meaning and occurs in a stratified way.

In this study, politicians with similar positions were selected in the two largest national political parties in Spain: the left-leaning Partido Socialista Obrero Español (PSOE) ‘Spanish Socialist Workers’ Party’ and the right-leaning Partido Popular (PP) ‘People’s Party’. Of the 32 individuals under consideration—eight each from Madrid, Malaga, Cordoba, and Seville—all productions of the affricate /tʃ/ were collected from a 35.8 h corpus of political speech. Sibilant affricates were examined based on their center of gravity (COG, in Hz), following vowel formant locus, auditory identification, and duration (as a percentage of the frication period in comparison with overall segment length), as well as extralinguistic factors including political affiliation, speech context, gender, and region.

This article is organized in the following way. Section 2 discusses existing literature on political speech, particularly in Peninsular Spanish, as well as the state of research into Spanish affricates and fricative-like segments cross-linguistically. This section concludes by posing three phonetic, linguistic, and socially motivated questions. Following that, Section 3 describes the speakers under consideration, the method of acoustic analysis, relevant dependent and independent variables, and the phonetic and statistical analyses employed to examine variation. Next, Section 4 describes the results of the study, including mixed-effects models to determine the factors influencing variable use, as well as the acoustic norms of the Spanish affricate. Section 5 addresses the phonetic and sociolinguistic ramifications of the results, referencing previous research and returning to the questions posed in Section 2. Finally, Section 6 concludes, offering some thoughts for future research and means of expanding the study of both Spanish affricates and sociolinguistic style in political speech.

2. Identity, Political Speech, and the Spanish Affricate

2.1. Sociophonetics and Political Speech

Early descriptions of sociolinguistic style by researchers like Labov (1972) cast the concept as attention paid to speech, with naturalistic interviews yielding more vernacular productions, while reading tasks such as word lists showed a higher rate of normative, non-vernacular variants. More recent studies, such as the classifications of identity described by Bucholtz and Hall (2005) and Coupland (2001), have problematized the representation of style as a binary distinction between vernacular and non-vernacular speech. Instead, they cast social meaning as an association of certain forms and speech decisions, part of a web of indexicality that allows speakers to take on the identity of others who are known to use these forms, following the description of Silverstein (2003). Through this process, linguistic style becomes just one aspect of the broader bricolage of a speaker’s social persona, as Eckert (2008) describes, contributing to a coherent vision of a speaker as having a consistent social identity and belonging to certain social subgroups.

This understanding of speaker agency emerges in part from earlier descriptions of speaker behavior relating back to audience expectations (e.g., Bell’s (1984) audience design theory) and accommodative behavior toward favorable and non-favorable social meanings (e.g., Giles et al.’s (1991) accommodation theory), and connects to a third-wave sociolinguistic approach associated with speaker design theory (e.g., Schilling-Estes 2013). This approach views speakers as having a multitude of linguistic and social goals that are carried out through variable uses of stylistic resources, according to the context in which they find themselves and the social associations pertinent to a specific speech encounter. Politicians, whose audiences and linguistic environments can vary greatly, from close supporters to international listeners and from one-on-one interviews to broadcast audiences of millions, use language to present themselves in a way that builds solidarity and crafts a consistent ideological persona. It is worth mentioning that, when accommodating to listeners, politicians base linguistic decisions not necessarily on the productions of specific individuals, but rather on the idealized norms associated with certain groups of speakers (e.g., Eckert 2008; Coupland 2001).

Numerous English-language researchers have focused on public and political speech as a means of accessing linguistic variation, from Bell’s (1984) work with New Zealand news anchors and Coupland’s (2001) discussion of a Cardiff disc jockey to more recent work looking at British politicians (e.g., Kirkham and Moore 2016), Scottish politicians (Hall-Lew et al. 2017), American politicians (Hall-Lew et al. 2012), Queen Elizabeth (Harrington 2007), Barack and Michelle Obama (Holliday 2017), Indian-American journalist Fareed Zakaria (Sharma 2018), and even Oprah Winfrey (Hay et al. 2010). However, there has also been work in recent years to probe some of these same sociolinguistic and even sociophonetic questions in Peninsular Spanish. Hernández-Campoy and Cutillas-Espinosa (2010, 2013) analyzed the speech of a former female president of Murcia, Spain, finding that she produced certain regional variants at a rate higher than even working-class rural male speakers, who were most closely associated with vernacular, regional identity in her community. The authors described this as a means of reinforcing her rural, working-class roots despite holding the highest office in her province, showing solidarity as a socialist politician toward her constituents. Hernández-Campoy and Cutillas-Espinosa (2010) argued that left-leaning Spanish politicians tended to use vernacular variants more than conservatives, a finding that Pollock and Wheeler (2022) also identified for a former female president in Andalusia, with respect to the elision of intervocalic /d/ in past participles.

However, other Peninsular Spanish research has offered less definitive evidence for these tendencies than Hernández-Campoy and Cutillas-Espinosa’s work. For example, in a study of Andalusian television presenters on social media, Fernández de Molina Ortés (2020) tracked the way that users’ identity work was interpreted by commenters. Of the four presenters studied, those with consistent identities and regional variant use received the most positive perceptual classifications, while dialect-switching was critiqued as a means of showing disloyalty to one’s region. In an examination of political speech production and perceptual norms in Andalusia and Madrid, Pollock (2023b) described an overall set of tendencies centered more around individual speaker motivations and goals than stable, party-based behaviors. However, Pollock (2023a) did find a tendency for regional variants to be produced by and perceived as associated with conservative political actors, which may be part of an ongoing shift in Peninsular politics toward populist and alt-right voices using novel linguistic patterns to appeal to rural and working-class voters in the south of Spain. At the level of forms of address and politeness, Pollock (forthcoming) further supports these findings by showing the extent to which impoliteness behavior increased on Spanish social media leading up to and during the COVID-19 pandemic. These studies suggest the importance of regional variants in political speech, due at least in part to the type of identity associated with these forms, as well as a complicated political reality in which linguistic and behavioral norms are in flux.

There is strong evidence to suggest that political actors are governed by the same linguistic expectations and norms as members of the broader speech community (e.g., Hernández-Campoy and Jiménez-Cano 2003). However, as Pollock (2023a) found in a perceptual analysis of regional Andalusian forms, potential voters are more likely to rate voices negatively once they are aware that they come from politicians. These speakers are at times subjected to unique norms of interpretation, which can also be reflected in the way they opt to speak. As Cruz-Ortiz (2022) described, there are certain norms in the Andalusian community associated with political speech, most particularly the prevalence of intervocalic /d/ deletion. She determined that politicians are much more likely to elide /d/ than even the most traditionally vernacular speakers in a given community. This tendency is also supported by the findings of Pollock and Wheeler (2022) for the former Andalusian president, Pollock (2023a) for a larger cohort of Madrid and Andalusian politicians, and Pollock and Wheeler (forthcoming) for a cohort of politicians in Galicia, Spain. These studies identify norms associated with an identity as a political actor in Spain that, at times, transcend both regional and normative speech expectations.

2.2. The Spanish Affricate and Acoustic Measures of Frication

Since the fifteenth century, there has been a marked reduction in sibilants within the Spanish inventory. While early Spanish had a robust eight-part sibilant system with six fricative and as many as two affricate components (i.e., /ts/, /dz/, /s/, /z/, /ʃ/, /ʒ/, /dʒ/, and /tʃ/), this had reduced to six components by the 16th century (i.e., $/\underset{⎴}{\mathrm{s}}/$, $/\underset{⎴}{\mathrm{z}}/$, /ʃ/, /ʒ/, /dʒ/, and /tʃ/) with the collapse of the alveolar/dento-alveolar fricative distinction. By the end of the seventeenth century, this had further reduced to two components (i.e., $/\underset{⎴}{\mathrm{s}}/$ and /tʃ/) as voicing distinctions were lost and the palatal fricative merged with /h/ (Penny 2002, p. 130; Bradley and Lozano 2022). As a result, modern Spanish has no phonemic voicing contrasts among sibilant fricatives and affricates (and, with respect to the occlusive portion of the affricate segment, word-initial voiced stops are produced with negative voice onset time (VOT) and voiceless obstruents have minimal VOT without aspiration, as described by Abramson and Lisker (1972), among others).

This historical reduction in systematicity in the sibilant range may offer the opportunity for increased variability in productions of the Spanish affricate. Speakers who alter their production of the affricate may have a reduced risk of being misinterpreted for producing an unexpected affricate or sibilant. This, in turn, offers an opportunity for the development of social meaning. As Labov (1972) described, linguistic features begin as indicators used in the same way across speakers. However, as they experience stratification (and thus an acquisition of indexical meaning), they become markers that can point to a speaker’s social class, gender, or even style, as speakers make use of the connection between the marker and its social associations.

Generally, the Spanish affricate is described in phonetic and phonological texts as /tʃ/, a segment with a prototypically voiceless, post-alveolar production (although note that Hualde and Colina (2014, p. 30) reference the possibility of production as a fricative [ʃ] or alveolar affricate [ts]). Almeida (2019) also described the length of the occlusive and fricative sub-components of the segment as normally being comparable. There is some discussion of variable production across Spanish-speaking communities. Deaffrication, well-documented in southern Spain, seems to be on the decline, particularly given its association with an older generation of speakers (e.g., Villena-Ponsoda 2013; Regan 2020). While focus in the current article centers on Peninsular Spanish, other work has also shown that reduction and deaffrication exists across the Spanish-speaking world (e.g., Díaz-Campos et al. 2023; Mazzaro 2022). In the Canary Islands, studies have shown that voiced post-palatal variants with increased tongue–palate contact can occur (Almeida 2019), whereas, in the southwest of Spain, a voiced pre-palatal [ʤ] is produced by mainly older Murcian speakers (Torrano Moreno 2017). Additionally, Pollock (2023a, 2023b) identified the presence of [ts] as a realization of the Spanish affricate in Andalusian and Madrid varieties that deserves further examination, and work by Del Saz, Vida-Castro and others identified overlap between fronted variants of the alveopalatal affricate and the production of a [t^s] affricate in contexts of /st/ cluster reduction (see Footnote 1).

Outside of the peninsular context, there are also references to the production of an alveolar affricate. Flores (2018) is one of several authors to describe this variant in Chile. In her examination of public radio speech, she identifies the variant as being associated with higher-prestige contexts and occurring in more frequent lexical items than the normative post-alveolar variant. Notably, these two allophones show not just audible differences, but also clear acoustic ones. When Flores (2014) compares the two, the segment of frication in alveolar variants has an average starting frequency concentration around 7000 Hz, while frication in normative post-alveolar productions starts closer to 5000 Hz. The presence of an alveolar affricate, attested in other dialectological accounts of Chilean Spanish, is often described as the result of contact with Indigenous languages like Huilliche in the region, which have a contrastive distinction between /tʃ/ and /ʈʂ/ (Henríquez-Barahona and Fuentes-Grandón 2018).

Much discussion of the Peninsular Spanish affricate has focused on deaffrication, particularly that found in the speech of elderly, rural speakers in parts of Seville, Granada, and Malaga (Henríquez-Barahona and Fuentes-Grandón 2018; Herrero de Haro 2017b; Samper-Padilla 2011). Additional linguistic and extralinguistic factors influence the occurrence of this variant. For example, Melguizo-Moreno (2006) found that, among older, working-class men with little education in Granada, the deaffricated variant [ʃ] mostly occurred word-initially. Villena-Ponsoda (2008, p. 148) described a regional divide, with speakers from southwestern parts of Andalusia like Jerez de la Frontera still producing high rates of deaffrication (83% of cases), while stigma and association with elderly rural speech had caused the variant to decline precipitously in northern and eastern parts of the region. According to his findings, fricative production makes up less than a quarter of tokens from speakers in Granada and Malaga.

From a methodological perspective, it is also important to point out how research has offered new means of approaching affricates using a continuous classification system. Phonetic work by researchers like Jongman (1989) and Jongman et al. (2000) showed how certain cues, including spectral, amplitudinal, and temporal ones, permit differentiation between fricative productions. More recently in Spanish, Díaz-Campos et al. (2023) considered affricates from a diachronic corpus of Caracas, Venezuela, dividing the segments into two sub-components, comparing the period of occlusion and that of frication through an analysis of temporal duration. They described the changes they identified over the two-decade span of the corpus as involving a process of retiming. While they did not find complete deaffrication in Venezuela, they did suggest that there is ongoing change moving in that direction. This study offers a way of conceptualizing affricates not through categorical means, but rather through a continuous analysis of their component parts, seeing the extent to which the balanced production of occlusion and frication described by Almeida (2019) in affricates is actually borne out in naturalistic speech.

In order to analyze a potentially alveolar production of the Spanish affricate, the current study turned its focus toward acoustic phonetic examinations of frication to identify the measures most often used to distinguish production norms. For example, in two varieties of Catalan, Recasens and Espinosa (2007) considered fricative and affricate production in an electropalatographic and acoustic analysis. They found that production of the alveolar affricate [ts] differed with respect to segment duration from the alveopalatal [tʃ], being generally the longer of the two segments. Meanwhile, in their study of seven typologically distinct languages, Gordon et al. (2002) used several measures to distinguish alveolar and post-alveolar frication, including segment duration, a spectral slice analysis, and the center of gravity (

Table 1. Sibilant fricative duration and COG norms (Gordon et al. 2002).

	/s/		/ʃ/
Language	Duration (ms)	COG (Hz)	Duration (ms)	COG (Hz)
1. Aleut	361.8	5219	-	-
2. Apache	172.3	5461	175.9	4859
3. Chickasaw	123.6	5163	112.8	4679
4. Gaelic	130.4	4884	110.7	4396
5. Hupa	276.4	4797	217.7	4440
6. Montana Salish	171.8	4601	178.9	4134
7. Toda	198.3	4529	239.6	4704
Average	204.9	4950.6	172.6	4535.3
standard deviation	79.4	317.5	48.3	239.2

).

Gordon et al. (2002) also tracked the trajectory of the formant locus in the following vowel to determine what coarticulatory effect might be present as a result of the preceding fricative. The hypothesis was that coarticulatory effects of fricative tongue placement would also appear in the left portion of the following vowel, with a corresponding difference in formant height for the same vowels produced after a more alveolar or palato-alveolar affricate. Other acoustic correlates were also tracked by these authors. For example, across all seven languages, the average duration was found to differ by only 32 milliseconds (ms), with /s/ being generally longer than /ʃ/ (similar to the findings for Catalan mentioned previously). Neither the duration measure nor the spectral slice were able to offer a clear and consistent distinction between production types. Center of gravity (COG), on the other hand, which averages the frequency of a spectral slice and tends to correlate with the place of articulation, was reliable when distinguishing between the fricative segments, with an average 400 Hz distinction between the two.

Other authors have identified it to be useful in the analysis and differentiation of coronal affricate productions, including Vida-Castro and Villena-Ponsoda (2016) for Spanish and Li and Li (2019) for Chinese. For the affricate [ts], Vida-Castro and Villena-Ponsoda found that a lower COG (i.e., 4619 Hz) was more likely to be perceived as the alveolar affricate /tʃ/, while productions with a higher COG (i.e., 6768 Hz and 4995 Hz) were more often perceived as associated with reduction in a word-internal /st/ cluster. This could have resulted from the process of post-aspiration in Spanish, where /st/ is produced as [t^h] and is reinterpreted as [ts], following from the pattern of aspiration where coda /s/ spirantizes to [^h]. Meanwhile, Li and Li (2019) found consistent differences between gendered productions of /ts^h/ (F = 9076 hz, M = 7989 hz) and /tɕ/ (F = 6596 hz, M = 5703 hz) for Mandarin Chinese. Across all of these examples, an /s/-like segment consistently tends to have a higher COG than a further backed post-alveolar or palatal segment. It is for this reason that, while the current study also examines duration and formant locus, COG is the main measure used to distinguish between affricate production norms in Spanish.

2.3. Research Questions

Three key questions guided the current analysis of Peninsular Spanish /tʃ/ and its allophonic production as either alveolar [ts] or normative post-alveolar [tʃ]: the first focused on phonetic variation, the second on statistical variation in linguistic and extralinguistic factors, and the third on implications for the sociolinguistic value of the affricate.

First, what variation exists at the phonetic level for the peninsular affricate? Based on acoustic measures, how do realization types differ, and to what extent is an alveolar affricate being produced by political actors?

Second, what variation exists among politicians, and to what extent is it influenced by linguistic and extralinguistic factors? While these are speakers who make performative use of language as a public means of identity construction, it is unclear the extent to which this variation occurs consistently across this population.

Finally, does the alveolar affricate function more as a sociolinguistic indicator or as a full sociolinguistic marker? If there are signs that the alveolar variant has become socially stratified, this suggests that there is a degree of community awareness of this phenomenon, at least insofar as it is associated with the broader speech patterns of a specific social group. If this is the case, it also welcomes future examination of stylistic variation with respect to this phenomenon beyond political speech to determine how widespread this sociophonetic variable is in Andalusian and northern-central Peninsular Spanish.

3. Methodology

3.1. Speaker Selection

For this study, 32 politicians were identified who formed a community of practice across southern and central Spain, sharing high-ranking positions in both regional and national governing bodies. Speakers included presidents, senators, and mayors who took part in interviews and speeches broadcast at the regional and national levels. Speakers were balanced by gender, city of origin, and political party. They were born in or near one of four cities—Malaga, Seville, Cordoba, and Madrid—and were affiliated with one of two major national parties—either the left-leaning PSOE or the right-leaning PP. See Supplementary Table S1 for a full list of speakers, including information on their sociolinguistic background and speech context.

Three types of speech situations were recorded in order to examine differences in language use based on the linguistic environments in which politicians found themselves. This included scripted speeches, often delivered before parliamentary bodies, as well as unscripted interviews with male interlocutors and with female interlocutors, both of which often came from broadcasts of major television news shows. Scripted speech was expected to reflect more carefully chosen decisions than those found in extemporaneous interviews, while gendered differences were expected to arise in the interviews based on the interlocutor, as Flores (2014) found for radio speech in Chile. Following previous studies that suggested that the earliest moments in interviews are the least vernacular and the most governed by normative expectations, coding began four minutes into each audio file (Díaz-Campos et al. 2018). Audio clips were 22 min in length on average, yielding 35.8 total hours.

Audio dated between the years 2011 and 2019 was collected, which roughly correlates to the span of time when the PSOE led the regional Andalusian government, and the PP was in charge of national governance and the regional government of Madrid. To be specific, the audio data span two periods of transition: the national PP government came into power in December of 2011 and then lost its control of the national government in June of 2018 following a vote of no confidence. By the end of the timespan considered in these audio data, a major shift had occurred at both the Andalusian and national levels that caused the parties in charge of national and regional governance to swap. As this was the first time that the PSOE lost control of the Andalusian government in over three decades, and as the alt-right Vox party experienced a marked growth in support, these speech data reflect a moment of substantial political change within Spain (Rama et al. 2021).

3.2. Acoustic Analysis of the Affricate

In total, 3175 tokens of the affricate were analyzed based on their production as either the alveolar affricate [ts] or the post-alveolar [tʃ]. All instances of the affricate were coded across each file following the four-minute mark, with the only (rare) exceptions being in cases where an interviewer’s speech or an audience’s applause overlapped with the speaker’s production. Outside of these contexts, affricates were produced clearly, without any cases of deaffrication or ambiguity. For normalization purposes, a comparable number of tokens of /s/ were also coded in various contexts throughout each file. In addition to the center of gravity, spectral examination was included as a means of better differentiating between tokens. On the whole, because broadcast-quality recording equipment was used, background noise was negligible and audio quality was high.

The first step in identifying applicable affricate contexts within the corpus involved the use of the speech-to-text transcriber available through Watson (IBM 2019). This software produced a simple transcript for each file, allowing the identification of all tokens of <ch> and a timestamp. Based on these data, Praat TextGrid files were created for audios that included intervals for each of the 3175 words identified with affricate tokens (Boersma and Weenink 2023). From there, intervals were manually created and affricate segments were identified.

For each token, six acoustic measures were collected. This included (1) the duration of the segment of occlusion and (2) the duration of the segment of frication in each affricate, based on visible wave form and spectrographic patterns. Additionally, both (3) F1 and (4) F2 were collected in vowels directly following the affricate, on the left edge of the vowel segment adjacent to the frication portion of the affricate to determine if there was a coarticulatory influence on this vowel (as examined by Gordon et al. (2002)). This measure was treated as an independent variable, to determine whether the differences Gordon and colleagues identified were also present in southern and northern-central varieties of Peninsular Spanish. This variable was collected in addition to the preceding and following environment, a categorical factor. Next, (5) the COG of the frication portion of each affricate token was measured. Finally, (6) a classification of the affricate was created based on the audible presentation of each segment by the researcher, a native-like speaker of the Peninsular variety of Spanish.

A script was developed in Praat to collect one COG value for the central 50% of aperiodicity in each fricative, as well as the F1 and F2 from the following vowel (collected at a single point, 20% from the leftward boundary to avoid acoustic noise at the edge of the segment), and the duration of tokens of /tʃ/. An example of a segment coded as post-alveolar is shown in Figure 1, while that of a segment coded as alveolar is given in Figure 2—in both cases, the COG and minimum frequency of aperiodic noise are provided (in Hz), showing how a more alveolar production (as found in typologically distinct languages) tends to have a higher COG than more backed ones.

Duration was determined, as shown in the red boxes in Figure 1 and Figure 2, by identifying the space between the conclusion of voicing in the preceding segment and the beginning of voicing in the following segment. Both the waveform and spectrogram were consulted to identify the end of the occlusion period (i.e., onset of high-frequency energy in the spectrogram and aperiodicity in the waveform) to begin the segment, and the onset of periodicity of the following vowel to end the segment. COG, on the other hand, was measured by creating an interval around only the densest aperiodicity of the fricative portion of the segment, then capturing the spectral slice of the middle 50% of this segment and identifying a single COG value for each token’s frication (i.e., the red box in Figure 1 and the blue box in Figure 2).

3.3. Dependent and Independent Variables

With respect to the dependent variables, two continuous and one categorical measure were used to describe variation in mixed-effects regression models. The continuous measures included the “percent of frication duration”, which divided the duration of the frication period of each affricate segment by the overall duration of the segment in an aim to account for speech rate. Additionally, COG (Hz) was examined as a second dependent variable, and was normalized based on the average COG of a speaker’s /s/ production (clarified in Section 3.4). Finally, the categorical measure represented the coding of affricate productions as alveolar [ts] or post-alveolar [tʃ], depending on acoustic examination and auditory identification. As no cases of deaffrication were identified, a categorical binary analysis was used.

Five linguistic and four extralinguistic independent variables were included in the analyses. The linguistic environment was examined through the inclusion of both preceding and following context, categorical measures that differed from the continuous F1 and F2 measures of the following vowel. The hypothesis in including these independent variables was that surrounding front vowels may have a coarticulatory effect that would correlate with the production of a more alveolar affricate token. Next, lexical accent was coded based on the tonicity of the syllable with the affricate. Pollock (2023a) identified a role of stylistic variation in the production of the Andalusian affricate—it is hypothesized that productions in tonic position have increased social salience, leading to higher rates of the fronted variant in the categorical analysis of [ts] and [tʃ] as a means of providing emphasis and indexing femininity, youth, and political affiliation. If this is an incipient change, more fronted productions of [tʃ] in tonic position would also be expected.

Next, two continuous linguistic variables were included in the analysis: a measure tracking segment duration and lexical frequency. First, in the models where percent of frication duration was not treated as a dependent variable, it was included as an independent variable to identify influences of frication length. This measure was expected to reflect the findings of Gordon et al. (2002), in that the alveolar segment is typically longer than the post-alveolar one. Subsequently, lexical frequency, coded as the number of times a word occurred in the overall corpus, was expected to identify often-used items that might be subject to higher rates of stylistic variation.

Extralinguistic variables included items coded in supplementary Table S1: a speaker’s city of origin, gender, and political affiliation, as well as the speech context of the audio file. Two additional variables were included as random effects to account for assumptions of the mixed-effect model related to the independence of tokens: the speaker and the unique lexeme. The breakdown of variables and possible individual factors are provided in

Table 2. List of dependent and independent variables, as well as random effects in the analysis.

Variable Type	Variable Name	Factors
Dependent variables	Frication period COG (Hz)	Continuous
	Percent frication duration	Continuous
	Place of production	Alveolar [ts]	Post-alveolar [tʃ]
Independent variables	Preceding context	Front vowel	Central vowel	Back vowel	Alveolar consonant
	Following vowel type	Front	Central	Back
	Following vowel F1 (Hz)	Continuous
	Tonicity	Tonic	Atonic
	Lexical frequency	Continuous
	City	Seville	Cordoba	Malaga	Madrid
	Gender	Male	Female
	Party	Left (PSOE)	Right (PP)
	Context	Scripted	Unscripted male	Unscripted female
Random effects	Speaker	n = 32
	Lexeme	n = 228

.

3.4. Data Analysis

In an effort to account for differences between speakers’ articulatory tracts, a means of normalization was necessary for affricate productions. Toda (2007) applied one such technique to reduce interspeaker variation among French and Japanese speakers for fricatives. This author found, through a quantitative analysis, that normalization preserves acoustic differences while reducing individual variation, leading to localized improvements (particularly in French). However, Toda also identified divergent tendencies between vowel categories and sibilant productions that may have resulted from coarticulatory effects. In order to avoid this possible influence, the current study similarly carries out normalization, albeit through the Lobanov (1971) method by comparing the COG of affricate frication with that of speakers’ productions of /s/.

This normalization method compares a given COG value (COG_n[V]) to the average COG of analyzed sibilants for a given audio file (Mean_N), then divides by the standard deviation (S_N) to acquire a normalized value centered around zero. The formula is provided in (1), as follows:

COG_n[V]^N = (COG_n[V] − Mean_N)/S_N

(1)

By including both fricative and affricate productions in the normalization model, the resulting difference in the apico-alveolar frication and post-alveolar frication period of the affricate are identified, while individual speaker differences are reduced. This, in turn, offers a basis for comparison within the articulatory tract of a single speaker. The resulting figures were then rescaled, following an altered version of the NORM standard found in the R vowels package formula (Kendall and Thomas 2010). This employed the formula in (2), where COG_N is the normalized value derived from (1) for a token’s COG, and COG_NMAX/MIN are the maximum and minimum normalized values for the entire group of speakers, as follows:

F′1 = 3000 + 1770 (COG_N − COG_NMIN)/(COG_NMAX − COG_NMIN)

(2)

To account for the sibilant COG, 3000 Hz was set as the baseline value, and variance was established as the standard deviation of /tʃ/ and /s/ in the current data (i.e., 1770 Hz). This effectively establishes a maximum scaled COG at 4770 Hz and a minimum at 3000 Hz.

This analysis used both the acoustic measurement of COG in the central 50% of the frication portion of affricates, as well as the binary categorical classification of tokens based on auditory coding with confirmation through visible acoustic measures in the waveform and spectrogram. The resulting data were analyzed using two mixed-effects regression models in the Rbrul software for R (version 4.0.2, Johnson 2009). Stemming from the variable rule approach to sociolinguistics, this type of model includes random effects, which help reduce imbalances across naturalistic data by accounting for dependencies (Johnson 2014).

Regression models offer several important statistics to describe how linguistic and extralinguistic factors influence variation (e.g., Tagliamonte 2012). The p-value of a variable indicates the strength of evidence against the null hypothesis, which would suggest that there is no difference between groupings of data, with p-values under 0.05 being treated as significant. The Rbrul statistical software offers a p-value for each variable. In logistic regressions, directionality can be attributed to an effect based on the factor weight. In linear regressions, this is shown in the coefficient. Factor weights range between 0 and 1, with values above 0.5 favoring use of the application value and those below 0.5 disfavoring it, while coefficients are centered around 0 but follow the same pattern. Finally, the range in factor weights, which reflects a difference in log-odds between factors, serves to indicate the magnitude of effect in logistic regressions. Greater differences between the highest and lowest weight for a single factor, also known as higher factor ranges, suggest that there are greater differences across the factors of a given significant variable. The models produced for each of the two dependent variables identify linguistic and extralinguistic factors that play a role in conditioning variable use. Model selection was carried out through comparisons of the Akaike Information Criterion (AIC) and log-likelihoods, favoring models with fewer variables when possible.

4. Results

In total, 3175 allophones of the Spanish affricate were collected for analysis, in addition to 3466 tokens of /s/ to permit normalization by speaker and audio file. On average, this meant that each of the 32 speakers produced around 33 affricates in each of the three speech contexts under consideration (i.e., scripted, unscripted with male interlocutor, and unscripted with female interlocutor). As shown in

Table 3. Average of acoustic measures for the affricate allophones and fricative tokens.

Production	Count	COG	Scaled, Normalized COG	Duration	Percent Frication	Scaled, Normalized F1 (next vowel)	Scaled, Normalized F2 (next vowel)
[tʃ]	2307	3979 Hz	4012 Hz	108.9 ms	50.3%	418.9 Hz	1422.5 Hz
[ts]	868	4783 Hz	4081 Hz	119.9 ms	51.6%	467.8 Hz	1491.2 Hz
All /tʃ/ tokens	3175 (33/file)	4199 Hz	4030 Hz	111.9 ms	50.7%	432.3 Hz	1441.3 Hz
All /s/ tokens	3466 (35/file)	4610 Hz	-	86.9 ms	-	-	-

, differences were identified between productions of /s/ and the affricate by duration (ms) and COG (Hz). Note that COG values of frication segments of each affricate were subjected to the normalization process described in Section 3.4. Additionally, formant loci for F1 and F2 were collected for following vowels as a continuous measure. A Lobanov normalization was performed on these F1 and F2 values using the NORM standard found in the R vowels package formula (Kendall and Thomas 2010) by first applying the formula in (1) used for COG, and then rescaling following the formula in (2) with the constants adjusted to the values expected for vowels (i.e., F1: 250, 500; F2: 850, 1400; Kendall and Thomas (2010)).

The average COG across the 3466 tokens of the sibilant fricative was 4609.6 Hz, and the duration was 86.9 ms. Of the 3175 productions of the affricate, 868 tokens were coded as alveolar [ts], which had an average COG of 4783 Hz, a duration of 119.9 ms, and a frication period taking up 51.6% of the segment. The remaining 2307 tokens were classified as post-alveolar with an average COG of 3979 Hz, a duration of 108.9 ms, and a frication period taking up 50.3% of the segment. While the normalization process for the affricate reduced some difference, there was still an over 65 Hz difference in tokens identified as alveolar and post-alveolar. Finally, formant values were only calculated in vowels following the affricate: the average normalized, scaled F1 was 418.9 Hz for vowels following [tʃ] and 467.8 Hz for those following [ts], while the average normalized, scaled F2 was 1422.5 Hz for vowels following [tʃ] and 1491.2 Hz for those following [ts].

With relation to the coarticulatory effects of the frication period of the affricate in the transition period to the following vowel, described by Gordon et al. (2002), the formant values in Table 3 suggest a relatively minimal average difference in vowel quality following the two allophones. Studies of differences in vowel height in English have suggested a perceptual threshold of at least 60 Hz necessary to distinguish between a low and mid vowel (Labov et al. 2013). While there is not an established threshold for vocalic differences in Spanish, Herrero de Haro’s (2017a) perceptual study of allophones of /e/ before contexts of elided /s/, /r/, and /θ/ suggests that an average difference of at least 60 Hz (and often closer to 100 Hz) is noticeable to listeners. Thus, while the differences in F1 and F2 suggest that preceding alveolar affricates are followed by lower and (very slightly) backed vowels, this difference should be taken with caution until further study can identify the extent to which listeners perceive these differences.

Meanwhile, an examination of the raw COG of /s/ and affricates suggests a trend between types of production of Spanish sibilants, represented in Figure 3. Politicians who produced tokens of /s/ with higher COG frequencies tended to produce both [tʃ] and [ts] with higher COGs as well. For both genders, the normative post-alveolar allophone tends to have a lower COG across most speakers. However, gendered differences also play a role in distinguishing productions, with most female tokens coded as [tʃ] having an average COG frequency similar to that of the male-produced /s/ or [ts]. This underscores the necessity of normalizing COG values to account for these anatomical differences across speakers.

A second measure referenced by Gordon et al. (2002) as a means of classifying productions of sibilant fricatives was segment duration. In their comparison, the authors found that alveolar /s/ tended to be longer than post-alveolar /ʃ/. As the current study focuses on affricates rather than fricatives, there may be a reduction in the magnitude of durational differences, as the two manners of production are not identical. However, as shown in Figure 4, the correlation identified by Gordon et al. is borne out in the current data. While the alveolar affricate had a greater overall variance in duration, average productions of this allophone tended to be longer than post-alveolar tokens, with men producing frication in tokens of [ts] on average 18 ms longer than in [tʃ], and women producing frication 9.5 ms longer.

Before developing a logistic regression model, descriptive comparisons of social factors in the data were made to identify general tendencies. The first comparison involved the differences in affricate realization type by speaker city and gender (Figure 5). Overall, the figure shows a reduced rate of alveolar affricate production among men, particularly in Malaga, both overall and in comparison to female speakers. While female politicians in Madrid and Cordoba produce around half of their affricates as alveolar [ts], those in Malaga and Seville produced around a quarter in that way, while men had near or below half as frequent a production of the alveolar variant.

The final comparison involved speakers’ political party and city of origin, allowing a consideration of differences based on political affiliation, as referenced in the discussion of Hernández-Campoy and Cutillas-Espinosa (2013), among others (Figure 6). Unlike many previous features discussed in contexts of political speech, both Figure 5 and Figure 6 suggest a minimal effect of regional variation, with speakers from Madrid and the northern Andalusian city of Cordoba grouping together. To be specific, for political affiliation and gender, conservatives and women from these two cities have a particularly high rate of alveolar production. The variant is infrequent in Malaga across genders and political parties, and relatively infrequent in Seville, as well as among most socialist and male politicians.

Based on the dependent and independent variables identified in Section 3.3, three models were developed to track variation in the data through three different means: continuously in a mixed-effects linear regression using the percent frication duration in each segment as a method of normalizing duration, followed by the normalized, rescaled COG (Hz) of the frication period of affricates, and then using a binary independent variable in a mixed-effects logistic regression, with segments classified as alveolar frication being treated as the application value. These models considered all linguistic and extralinguistic variables to determine which influenced COG values and affricate selection among this cohort of political actors. The two random effects were also included to account for speaker and lexical variation, as well as interactions between social factors that were theoretically motivated (i.e., the interaction between speaker and interlocutor gender, or those between political affiliation, speaker gender, and city).

In the first linear regression model, presented in Table 4, seven variables were selected as significantly conditioning frication duration in the segment, five of which were linguistic and two extralinguistic. First, the impressionistic realization type followed the trend identified by Gordon et al. (2002), in that [ts] had a longer frication period than the post-alveolar affricate. Keeping this in mind, there was an inverse relationship between duration percentage and preceding and following context—both non-front preceding vowels and consonants favored [ts], while, in following position, front vowels favored it. This suggests a process of movement, with longer frication segments reflecting a shift in the tongue from the back of the mouth or consonant production into a front vowel. While fricative sounds are traditionally seen as resistant to coarticulatory effects (Recasens 2018), duration may reflect the degree of emphasis placed upon production and speak to the indexical social meaning being attached to production norms. More research into the effects of coarticulation in this context is merited. The fourth significant variable, lexical frequency, was continuous, showing that more frequent items tended to favor longer productions. Lastly, the scaled F1 of the following vowel also showed a relationship with duration, in that lower vowel height overall favored longer frication durations.

Among the social factors selected in this model were political party and speech context. In the case of the former, socialists tended to produce affricates with longer frication periods, while, for the latter, scripted speeches were the only context to disfavor longer frication periods. No interactions were selected as significant predictors of variation in this model.

Table 4. Mixed-effects linear regression with percent frication duration in the affricate as the continuous dependent variable and speaker and lexeme as random effects.

Variable	Factor	Coefficient	Tokens	Frication as a % of Segment Duration
Political party (p = 0.003)	Socialist (PSOE)	0.024	1428	53.5%
	Conservative (PP)	−0.024	1747	48.4%
Range (%)				5.1%
Preceding context (p < 0.001)	Consonant	0.014	235	52.8%
	Back vowel	0.002	1783	51.5%
	Central vowel	0.001	79	50.8%
	Front vowel	−0.016	1078	48.9%
Range (%)				3.9%
Following vowel (p < 0.001)	Front vowel	0.014	502	52.3%
	Central vowel	−0.006	1036	50.9%
	Back vowel	−0.008	1637	50.1%
Range (%)				2.2%
Realization type (p < 0.001)	[ts]	0.008	868	51.6%
	[tʃ]	−0.008	2307	50.3%
Range (%)				1.3%
Speech context (p = 0.001)	Unscripted male	0.006	1047	51.1%
	Unscripted female	0.001	988	51.1%
	Scripted speech	−0.006	1140	50.0%
Range (%)				1.1%
Lexical frequency (p < 0.001)	Continuous	Coefficient
	+1	0.001
Scaled following vowel F1 (p = 0.010)	Continuous	Coefficient
	+1	0.001
n = 3175; df = 15; Log-likelihood = 3784; AIC = −7538; R2 fixed = 0.221; R2 total = 0.323

Table 4. Mixed-effects linear regression with percent frication duration in the affricate as the continuous dependent variable and speaker and lexeme as random effects.

Variable	Factor	Coefficient	Tokens	Frication as a % of Segment Duration
Political party (p = 0.003)	Socialist (PSOE)	0.024	1428	53.5%
	Conservative (PP)	−0.024	1747	48.4%
Range (%)				5.1%
Preceding context (p < 0.001)	Consonant	0.014	235	52.8%
	Back vowel	0.002	1783	51.5%
	Central vowel	0.001	79	50.8%
	Front vowel	−0.016	1078	48.9%
Range (%)				3.9%
Following vowel (p < 0.001)	Front vowel	0.014	502	52.3%
	Central vowel	−0.006	1036	50.9%
	Back vowel	−0.008	1637	50.1%
Range (%)				2.2%
Realization type (p < 0.001)	[ts]	0.008	868	51.6%
	[tʃ]	−0.008	2307	50.3%
Range (%)				1.3%
Speech context (p = 0.001)	Unscripted male	0.006	1047	51.1%
	Unscripted female	0.001	988	51.1%
	Scripted speech	−0.006	1140	50.0%
Range (%)				1.1%
Lexical frequency (p < 0.001)	Continuous	Coefficient
	+1	0.001
Scaled following vowel F1 (p = 0.010)	Continuous	Coefficient
	+1	0.001
n = 3175; df = 15; Log-likelihood = 3784; AIC = −7538; R2 fixed = 0.221; R2 total = 0.323

In the second linear regression model, displayed in Table 5, eight variables were selected as significantly conditioning increased COG (Hz), including five that were linguistic, three that were extralinguistic, and one interaction. Based on the findings of Gordon et al. (2002), a higher frequency of the COG of a segment was expected to correlate with a more alveolar production. The eight variables in this model are organized in order of the range in COG between their factors, with city coming first. Speakers from Madrid produced the highest rescaled COG in affricates overall, while those in Seville produced the lowest—however, the model described all speakers from Andalusia as producing an affricate that was less likely to have a higher frequency. Next, the model selected the preceding sound, with central and front vowels and alveolar consonants favoring a higher COG, and back vowels being more likely to pair with reduced COG. Following that, the realization coding was found to significantly predict frication segment COG frequency, with coding as an alveolar [ts] favoring a higher COG. After this, the following vowel was selected, with both front and central vowels again favoring a higher COG, while back vowels disfavored it.

The fifth variable selected in the model was tonicity, with tonic contexts favoring a higher COG, showing a potential correlation between the alveolar production and tonicity. Next, the speech context was selected, with scripted and female unscripted speech favoring a higher COG, and unscripted speech with a male interlocutor disfavoring it. Following that, the final extralinguistic factor selected in the model was speaker gender, with men having a higher normalized COG overall than women. The last linguistic factor was the rescaled, normalized F2 of the following vowel. More front vowels (i.e., those with increased F2 values) favored production after the affricate with a higher COG, associating the alveolar variant with coarticulatory effects. Finally, there was an interaction between speaker gender and speech context, with female politicians disfavoring COG raising in speeches with male interviewers, and male politicians disfavoring raising in scripted speeches and when talking with female interviewers, as demonstrated in Figure 7.

Table 5. Mixed-effects linear regression with the normalized, rescaled COG value as the continuous dependent variable and speaker and lexeme as random effects.

Variable	Factor	Coefficient	Tokens	Rescaled COG (Hz)
City (p = 0.001)	Madrid	63.838	793	4101.1
	Malaga	−5.442	695	4032.8
	Cordoba	−14.479	717	4024.1
	Seville	−43.917	970	3976.8
Range (Hz)				124.3
Preceding context (p < 0.001)	Central vowel	35.486	79	4110.5
	Alveolar	8.322	235	4106.0
	Front vowel	1.517	1078	4043.9
	Back vowel	−45.326	1783	4009.4
Range (Hz)				101.1
Realization coding (p < 0.001)	[ts]	26.8	868	4081.4
	[tʃ]	−26.8	2307	4011.7
Range (Hz)				69.7
Following vowel (p = 0.010)	Front	10.21	502	4073.0
	Central	8.4	1036	4050.1
	Back	−18.61	1637	4005.7
Range (Hz)				67.3
Tonicity (p = 0.001)	Tonic	19.406	543	4077.4
	Atonic	−19.406	2632	4021.2
Range (Hz)				56.2
Speech context (p < 0.001)	Scripted	16.916	1140	4047.9
	Female Unscripted	−0.303	988	4037.7
	Male Unscripted	−16.612	1047	4005.6
Range (Hz)				42.3
Gender (p = 0.021)	Men	20.921	1521	4041.7
	Women	-20.921	1654	4020.8
Range (Hz)				21
Interaction: Speaker Gender * Speech context (p = 0.001)	Male:Scripted	18.081	593	4050.0
	Female:Scripted	13.735	547	4045.6
	Female:Female Interloc	4.345	542	4042.0
	Male:Male Interloc	−4.345	482	4039.9
	Male:Female Interloc	−13.735	446	4032.5
	Female:Male Interloc	−18.081	565	3976.4
Range (Hz)				73.6
Scaled normalized following vowel F2 (p = 0.015)	Continuous	Coefficient
	1	0.026
n = 3175; df = 20; Log-likelihood = −21,322; AIC = 42,683; R2 Fixed = 0.105; R2 Total = 0.148

Table 5. Mixed-effects linear regression with the normalized, rescaled COG value as the continuous dependent variable and speaker and lexeme as random effects.

Variable	Factor	Coefficient	Tokens	Rescaled COG (Hz)
City (p = 0.001)	Madrid	63.838	793	4101.1
	Malaga	−5.442	695	4032.8
	Cordoba	−14.479	717	4024.1
	Seville	−43.917	970	3976.8
Range (Hz)				124.3
Preceding context (p < 0.001)	Central vowel	35.486	79	4110.5
	Alveolar	8.322	235	4106.0
	Front vowel	1.517	1078	4043.9
	Back vowel	−45.326	1783	4009.4
Range (Hz)				101.1
Realization coding (p < 0.001)	[ts]	26.8	868	4081.4
	[tʃ]	−26.8	2307	4011.7
Range (Hz)				69.7
Following vowel (p = 0.010)	Front	10.21	502	4073.0
	Central	8.4	1036	4050.1
	Back	−18.61	1637	4005.7
Range (Hz)				67.3
Tonicity (p = 0.001)	Tonic	19.406	543	4077.4
	Atonic	−19.406	2632	4021.2
Range (Hz)				56.2
Speech context (p < 0.001)	Scripted	16.916	1140	4047.9
	Female Unscripted	−0.303	988	4037.7
	Male Unscripted	−16.612	1047	4005.6
Range (Hz)				42.3
Gender (p = 0.021)	Men	20.921	1521	4041.7
	Women	-20.921	1654	4020.8
Range (Hz)				21
Interaction: Speaker Gender * Speech context (p = 0.001)	Male:Scripted	18.081	593	4050.0
	Female:Scripted	13.735	547	4045.6
	Female:Female Interloc	4.345	542	4042.0
	Male:Male Interloc	−4.345	482	4039.9
	Male:Female Interloc	−13.735	446	4032.5
	Female:Male Interloc	−18.081	565	3976.4
Range (Hz)				73.6
Scaled normalized following vowel F2 (p = 0.015)	Continuous	Coefficient
	1	0.026
n = 3175; df = 20; Log-likelihood = −21,322; AIC = 42,683; R2 Fixed = 0.105; R2 Total = 0.148

The final model in Table 6, a logistic regression of realization types, compared productions based on auditory coding reinforced with visible acoustic measurements taken in the waveform and spectrogram. This model included eight factors, including five linguistic and three extralinguistic. As opposed to the model in Table 5, this one had rescaled, normalized COG of frication as an independent variable and the coded realization type as the dependent variable. Variables in this model are organized by their range in factor weights, reflecting the overall magnitude of effect. Speaker gender is first; while male politicians produced higher normalized COGs in affricate frication periods overall, this model shows that female politicians produced more affricates coded as the alveolar [ts]. This disparity merits perceptual analysis to determine if the trend holds among native speakers of Andalusian varieties of Spanish as well—it is unclear whether this is the result of gendered perceptions of language or perhaps the presence of an additional acoustic cue not under consideration in this study that may further inform classification.

Several of the remaining variables have similar trends to those identified in the linear regression model in Table 5. For the variables with the second- and third-highest magnitudes of effect, the preceding and following environment, back vowels and central vowels disfavored [ts] in the preceding context, while back vowels alone disfavored it in the following context. Two non-significant main effects included due to their participation in a significant interaction were city and speech context, the latter approaching significance with a p-value of 0.085. We see that speakers from southwestern and southeastern Andalusia produced less [ts], while those in Cordoba and Madrid, as well as those in unscripted interviews with women and in scripted speeches, favored [ts]. For the remaining continuous independent variables, affricates with a higher COG, a greater percentage of the segment occupied by frication, and a raised F1 in the following vowel all favored [ts] production.

Table 6. Mixed-effects logistic regression of affricate realization type with the alveolar production [ts] as the application value and the speaker and lexeme as random effects.

Variable	Factor	Coefficient	Tokens	% [ts]	Factor Weight
Gender (p = 0.001)	Female	0.827	1654	37.7%	0.696
	Male	−0.827	1521	16.1%	0.304
Range					39.2
Preceding context (p < 0.001)	Consonant	0.663	235	49.4%	0.660
	Front vowel	0.329	1078	30.5%	0.581
	Central vowel	−0.34	79	29.1%	0.416
	Back vowel	−0.652	1783	22.4%	0.343
Range					31.7
Following vowel (p < 0.001)	Central	0.335	1036	35.5%	0.583
	Front	0.097	502	29.3%	0.524
	Back	−0.432	1637	21.6%	0.394
Range					18.9
City (p = 0.270) †	Cordoba	0.515	717	38.4%	0.626
	Madrid	0.452	793	37.3%	0.611
	Seville	−0.362	970	18.9%	0.411
	Malaga	−0.605	695	16.4%	0.353
Range					27.3
Speech context (p = 0.085) †	Unscripted female	0.146	988	30.2%	0.536
	Scripted speech	0.036	1140	26.8%	0.509
	Unscripted male	−0.182	1047	25.3%	0.455
Range					8.1
Speaker Gender * Speech context	Female:Scripted	0.172	547	0.397	0.543
(p = 0.037)	Female:Female Interloc	0.14	542	0.382	0.535
	Female:Male Interloc	0.031	565	0.352	0.508
	Male:Female Interloc	−0.031	446	0.204	0.492
	Male:Scripted	−0.14	593	0.148	0.465
	Male:Male Interloc	−0.172	482	0.137	0.457
Range (Hz)					8.6
Scaled COG of frication (p < 0.001)	Continuous
	+1	0.001
Percent frication (p < 0.001)	Continuous
	+1	3.362
Normalized F1 of following vowel	Continuous
(p = 0.016)	+1	0.017
City * Following Context (p = 0.007)
n = 3175; df = 25; Log-likelihood = −1391; AIC = 2832; R2 fixed = 0.262; R2 = 0.485

† Note that the two main effects, city and speech context, are included in the table in italics, as they do not have a significant p-value, because they form part of a significant interaction (listed at the bottom of the table).

Table 6. Mixed-effects logistic regression of affricate realization type with the alveolar production [ts] as the application value and the speaker and lexeme as random effects.

Variable	Factor	Coefficient	Tokens	% [ts]	Factor Weight
Gender (p = 0.001)	Female	0.827	1654	37.7%	0.696
	Male	−0.827	1521	16.1%	0.304
Range					39.2
Preceding context (p < 0.001)	Consonant	0.663	235	49.4%	0.660
	Front vowel	0.329	1078	30.5%	0.581
	Central vowel	−0.34	79	29.1%	0.416
	Back vowel	−0.652	1783	22.4%	0.343
Range					31.7
Following vowel (p < 0.001)	Central	0.335	1036	35.5%	0.583
	Front	0.097	502	29.3%	0.524
	Back	−0.432	1637	21.6%	0.394
Range					18.9
City (p = 0.270) †	Cordoba	0.515	717	38.4%	0.626
	Madrid	0.452	793	37.3%	0.611
	Seville	−0.362	970	18.9%	0.411
	Malaga	−0.605	695	16.4%	0.353
Range					27.3
Speech context (p = 0.085) †	Unscripted female	0.146	988	30.2%	0.536
	Scripted speech	0.036	1140	26.8%	0.509
	Unscripted male	−0.182	1047	25.3%	0.455
Range					8.1
Speaker Gender * Speech context	Female:Scripted	0.172	547	0.397	0.543
(p = 0.037)	Female:Female Interloc	0.14	542	0.382	0.535
	Female:Male Interloc	0.031	565	0.352	0.508
	Male:Female Interloc	−0.031	446	0.204	0.492
	Male:Scripted	−0.14	593	0.148	0.465
	Male:Male Interloc	−0.172	482	0.137	0.457
Range (Hz)					8.6
Scaled COG of frication (p < 0.001)	Continuous
	+1	0.001
Percent frication (p < 0.001)	Continuous
	+1	3.362
Normalized F1 of following vowel	Continuous
(p = 0.016)	+1	0.017
City * Following Context (p = 0.007)
n = 3175; df = 25; Log-likelihood = −1391; AIC = 2832; R2 fixed = 0.262; R2 = 0.485

† Note that the two main effects, city and speech context, are included in the table in italics, as they do not have a significant p-value, because they form part of a significant interaction (listed at the bottom of the table).

Two additional interactions were identified as significant predictors of variation. The first paired politician gender with the speech context. The correlation is demonstrated in Figure 8. There is a general tendency for unscripted speech with male interviewers to have the lowest rates of [ts], while unscripted speech with female interlocutors has a consistently higher rate, and female politicians are most likely to use [ts] in all contexts. The second interaction involved politicians’ city and the following context, provided in Figure 9. While it is the case that a following /i/ and /e/ favor [ts] production in Madrid, the pattern in the Andalusian cities is for a following /a/ to favor [ts], even in Cordoba, where rates of the fronted variant are comparable to those in Madrid. This phenomenon is generally more frequent in the two cities closest to central Spain, even as there appears to be different fronting patterns based on linguistic context that are differentiated depending on whether the city is part of Andalusia.

5. Discussion

5.1. Phonetic Variation in the Affricate

This study shows variable production of the Spanish affricate based on quantitative differences that surpass the level of perceptibility described by other research on segment frequency measures (e.g., Labov et al. 2013; Herrero de Haro 2017a; Gordon et al. 2002). While the values of F1 and F2 at the 20% mark of vowels following the affricate did not show marked differences based on realization types, both the center of gravity of the frication period of the affricate and the percent of frication in the segment followed trends identified in previous studies for phonetically perceptual and even phonemic distinctions between alveolar and post-alveolar fricative segments. This includes Recasens and Espinosa’s (2007) findings for Catalan, where the duration of both the frication period and whole segment of affricate [ts] were found to be longer than [tʃ] in both Majorcan and Valencian. It also includes Gordon et al. (2002), who found the alveolar fricative to generally be the longer of the two segments across an array of typologically distinct languages. A comparison of Gordon and colleagues’ findings with those for the frication period of affricates in the current study is shown in

Table 7. Comparison of sibilant fricative results from Gordon et al. (2002) and the current ones for allophones of the Peninsular Spanish affricate.

Language	/s/ Duration	COG	/ʃ/ Duration	COG
1. Aleut	361.8	5219	-	-
2. Apache	172.3	5461	175.9	4859
3. Chickasaw	123.6	5163	112.8	4679
4. Gaelic	130.4	4884	110.7	4396
5. Hupa	276.4	4797	217.7	4440
6. Montana Salish	171.8	4601	178.9	4134
7. Toda	198.3	4529	239.6	4704
Average	204.9	4950.6	172.6	4535.3
st.dev.	79.4	317.5	48.3	239.2
Peninsular Spanish	[ts] duration	COG	[tʃ] duration	COG
8. Frication portion	61.8	4783.4	54.5	3978.4
8.1. Total for affricate	140.3		127.4
Overall average	187.1	4929.7	155.7	4455.8
st.dev.	94.1	323.0	65.7	318.7

. As they do not normalize sibilant fricatives, the raw COG values for the current study are used in this table.

The ordering for both duration and COG in Spanish is consistent with the language tendencies identified by Gordon and colleagues. For the duration of the frication period of the Spanish affricate, it is important to note that this represents only around half of the overall segment length (i.e., in addition to the occlusion period of the affricate, Almeida (2019)), and it should not be taken for granted that affricate frication necessarily maps onto trends in fully fricative segments. However, by examining the duration, as well as Table 4 with its frication percentage model, similarities can be observed between these frication portions and the fricative segments examined in other languages. Additionally, while the COG of frication in the realization coded as the alveolar affricate [ts] is well within the range of other sibilant fricative productions of /s/, the COG for the post-alveolar production is lower than any other value found by Gordon and colleagues. In part, as these are non-normalized results, this could stem from the difference between data recorded in a laboratory and the naturalistic televised audio used in the present study. However, it could also point to a difference in sibilant quality of the normative Peninsular post-alveolar affricate compared to the prototypical production of /ʃ/ found in these other languages. In general, though, the average Spanish alveolar production of [ts] is longer than the post-alveolar production [tʃ] and has an elevated COG of its frication period, similar to the distinctions made in other languages where this difference is phonemic in the sibilant fricative system.

These results offer insight into this novel phenomenon, while at the same time providing baseline numbers for typical affricate productions in naturalistic speech in Peninsular Spanish. While the affricate is often treated as unvaryingly post-alveolar among middle-aged and younger speakers, these results suggest that this is not fully true at the phonetic level, at least among these political actors from both Andalusia and Madrid. Although this phenomenon does not directly affect the phonemic system of Spanish (although see the discussion by Vida-Castro and Villena-Ponsoda (2016) about the overlap in perception between [ts] as /tʃ/ and aspiration in the /st/ cluster—also discussed below at the end of Section 5.3), it is nevertheless expected to reach a level of perceptibility for listeners, offering a context for social stratification between the alveolar and post-alveolar forms. Further research is needed, particularly among members outside the political speech community,2 to determine the extent to which this is an ongoing shift in the phonetics of Peninsular Spanish.

5.2. Affricate Variation in Peninsular Political Speech

The next focus in this study was to identify variables influencing allophonic selection in the political speech of Madrid and Andalusia. As shown in Figure 3, speakers’ articulatory tracts tended to scale their affricate productions alongside those of /s/, such that individuals with productions of the fricative with higher COG tended to also produce affricates with higher COG. Normalization reduced these differences, particularly those related to gender and individual differences in articulatory tract size. Additionally, descriptive differences across realization type showed some degree of variation across all four social variables: city, political affiliation, gender, and speech context. However, through the mixed-effects models, specific linguistic and extralinguistic variables were found to significantly explain the differences in the quality and type of affricate that speakers produced, as laid out in

Table 8. Significant variables and tendencies from the mixed-effects regression models.

#	Variable Name	Frication Percent (Favor Higher % Frication)	Center of Gravity (Favor Higher COG)	Realization Coding (Favor [ts])
Linguistic Variables
1	Frication Percent		-	Greater percent
2	Center of Gravity	-		Higher COG
3	Realization Type	[ts]	[ts]
4	Preceding Context	Consonants, back and central vowels	Consonants, front and central vowels	Consonants and front vowels
5	Following Context	Front vowels	Front and central vowels	Front and central vowels
6	Lexical Frequency	More frequent	-	-
7	Following Vowel F2 (normalized and scaled)	-	Higher F2	-
8	Following Vowel F1 (normalized and scaled)	Higher F1	-	Higher F1
9	Tonicity	-	Tonic syllables	-
Extralinguistic Variables
10	City	-	Madrid	Cordoba and Madrid
11	Political Party	PSOE politicians	-	-
12	Speech Context	Male and female unscripted	Scripted and female unscripted	Scripted and female unscripted
13	Gender	-	Men	Women

. In the following paragraphs, significant variables are discussed, and contrasts are made across the three models to describe what trends duration, COG, and the categorical realization coding identify for the analysis.

Two linguistic variables with consistently high magnitudes of effect were the surrounding context, both preceding and following. In general, the Spanish affricate /tʃ/ most often occurs intervocalically (e.g., mucho ‘a lot’), although there are also cases with a typically alveolar preceding consonant (e.g., ancho ‘wide’). With respect to the following context, there was a general trend across all three models in which front vowels favor production of an affricate with a longer frication portion, a higher COG, and coding as [ts]. Additionally, for COG and realization type, the central vowel [a] favored these productions. Given the consistency and direction of effect across models, this suggests a coarticulatory effect of the following linguistic environment, where following [u] or [o] is less likely to permit tongue advancement to the alveolar ridge for the affricate.

Meanwhile, the preceding context offers a slightly more complicated picture. While, for the duration-adjacent measure of frication percent in the segment, greater rates of frication correlated with consonants, back, and central vowels, both increased COG and realization coding of [ts] were favored in the context of preceding consonants and front vowels (as well as central vowels for COG). As Gordon et al. (2002) discussed, duration-related measures were the least reliable in distinguishing between the place of frication production. In only three of the six typologically distinct languages for which the duration of /ʃ/ in comparison with /s/ was considered (i.e., Apache, Montana Salish, and Toda) did /s/ have a shorter duration of the two segments, even though the overall average showed it to have a longer one. The difference identified in the current study may reflect the more uncertain role of duration in distinguishing between sibilant production location. For the most part, it seems that preceding consonants and central vowels favor a more alveolar production of the affricate, while increased COG and realization coding as [ts] additionally favored preceding front vowels, which may point to a similar effect as seen for the following context.

The next pair of linguistic variables significant across all three models was the formant height of the following vowel. The first formant was significant for both frication percent and realization coding, with higher F1 values favoring coding as [ts] and a longer period of frication. This change in formant value would correlate with a general lowering in vowel height following a more alveolar production, possibly reflecting the prototypically concave apico-dental production of the Spanish alveolar sibilant, which lowers the dorsum of the tongue and turns the tip upward (Núñez-Méndez 2022). This tongue movement may result in a co-articulatory lowering of the following vowel, in part due to the pressure being imposed by the coarticulatory resistance of the adjacent /s/. Alternately, the second formant was only found to be significant for the COG model, with higher values of F2 (i.e., more fronted productions) being favored alongside higher values of frication COG. This finding would make articulatory sense, as a raised tongue gesture in the affricate would be expected to have an effect on the following sound.

Context alone does not influence production: tonicity was also found to play a role in the continuous COG model. Instances of the affricate in tonic syllables tended to have a higher COG than those in atonic syllables, with an average difference of 56.2 Hz. This suggests that more fronted productions would favor those contexts with greater emphasis, aligning with previous discussion of this phenomenon as a possible stylistic marker growing in salience (Pollock 2023a). It is important to note, however, that the average difference in normalized COG between tonic and atonic contexts does not quite reach the perceptibility threshold of 60 Hz described in previous vowel studies (e.g., Labov et al. 2013; Herrero de Haro 2017a), which may suggest incipient change still underway surrounding this form.

Lexical frequency was also found to play a role in variable production based on the frication period. For this variable, more frequent lexical items within the corpus favored production with longer duration periods, correlated to some extent with [ts] production. There were slightly over 200 unique lexemes across the 3175 tokens, with only 4 having more than 100 tokens (i.e., mucho/a(s) ‘a lot’, hecho ‘done’, dicho ‘said’, and marcha ‘proceeding’), and 19 having 10 or more occurrences. In many cases, it has been determined that language change spreads first from high-frequency items to those that are less frequent, as argued by Bybee (2002, 2010), particularly in the type of retiming phenomena being tracked by the percent frication measure (e.g., Bybee 1998, p. 422; Díaz-Campos et al. 2023)3. When combined with tonicity, the results for lexical frequency indicate that the variation targeted in this study may in fact represent the first stages of a change in progress.

The final three linguistic variables selected as conditioning variation were considered as a way to connect the mixed-effects models. These included the realization type, the COG of frication in the affricate, and the percent frication (i.e., the dependent variables) being treated as independent variables in the other models. Each of the three was selected as a significant predictor of variation at least once elsewhere. When realization type was the dependent variable, tokens coded as [ts] favored production with a higher COG and with a longer frication period, while in the continuous models for COG and frication duration, realization type was selected as significant predictors of variation. The relationship between realization type and the two quantitative continuous factors suggests that, even if COG and percent frication do not fully map onto each other, both can be seen to play a role in perception of the affricate.

Four social variables remain to be discussed. First, the city of the politician was significantly predictive of COG, as well as in an interaction with the following context for realization type. Speakers from Madrid were most likely to produce variants with a higher COG (i.e., correlating with the alveolar production), while those from all three cities in Andalusia were less likely to have increased COGs. Furthermore, the range of production differences by city was over 120 Hz, which should fall well within the range of human perceptibility (Labov et al. 2013; Herrero de Haro 2017a; Gordon et al. 2002). Based on realization type, both speakers from Cordoba and Madrid had a nonsignificant tendency to produce [ts], but there was also a significant difference in production norms, with speakers from Cordoba, Malaga, and Seville all producing the most [ts] after [a], while those in Madrid instead produced it at higher rates after [i] and [u]. This difference in production norms could, as a reviewer points out, be related to the fact that affricate production in Madrid is incipient but the most salient, due to the highest degree of fronting, and that contexts preceding high, front vowels are a frequent context for affricate changes, from which change can spread.

Next, for speaker gender, based on the continuous analysis of COG, men favored the production of a higher normalized COG for affricates than women overall (22 Hz), although this difference falls below the expected threshold of perception. However, for realization type, this tendency is reversed, with women identified as producing around twice as many alveolar tokens as men (i.e., 37.7% vs. 16.1%). This trend merits further discussion. First, there is relatively little work that describes means of normalizing sibilant production in the articulatory space of multiple speakers. In Toda’s (2007, p. 828) examination of normalized /s/ and /ʃ/ in two languages, the author finds that distinctness was easier to identify in Japanese than in French, potentially as a result of the allophonic relationship between them in French, which may lead to greater variability in production norms. Meanwhile, Gordon et al. (2002) chose to look at raw COG values across the seven languages they considered, presenting individual speakers by gender without attempting to compare norms quantitatively.

While the method of normalization used in the current article has aimed to reduce differences in individual speech, across social factors, and within audio recordings, it is unclear the extent to which the normalization and rescaling processes can erase secondary markers of variation used in perception. Admittedly, female articulatory spaces can be smaller than those of men, at times leading to sibilants with a higher COG, as seen in Figure 3,4 but the high rate of classification of female affricates, such as [ts], suggests that there are audible differences between the alveolar and post-alveolar productions that are not captured in the continuous models, suggesting additional acoustic correlates that may be at work. This requires further analysis using perceptual instruments to determine the extent to which native speakers of the Andalusian and northern central Peninsular Spanish (NCPS) varieties perceive these differences, and to begin identifying how these various correlates and potential others work together to influence classification.

When examining speech context, there appears to exist a sociocultural environment reminiscent of Flores’s (2014) study of Chilean radio speech. In her analysis of linguistic phenomena including an alveolar affricate [ts], she found that men tended to use a greater degree of vernacular forms with other men, while women used more prestige forms with other women. Meanwhile, scripted speech tended to carry the most careful and normative productions. In the current study, in the interaction for realization type between speaker gender and speech context, both male and female politicians used higher rates of [ts] with female interviewers, while only women used similar rates in scripted speeches (Figure 8). Meanwhile, in the interaction shown in Figure 7 between speaker gender and scripted speech, there is a marked dip in COG for female politicians when producing /tʃ/ with a male interlocutor for both variants, but particularly the post-alveolar affricate. This suggests a strategy of backing among female speakers to accommodate to affricate production by male colleagues.

Finally, for the model analyzing frication duration, there is another deviation from the COG and realization type models—scripted speeches have the shortest frication period, while those in unscripted interviews are longer. This may suggest an unrelated phenomenon associated with register rather than sibilant place, where frication production is slightly more evenly produced alongside the occlusion period of the affricate in more professional scripted settings. In any case, considering the social tendencies Flores identified (i.e., that same-gender dyads lead to increased rates of vernacular or prestige forms, depending on group gender), the results for COG and realization type suggest an association of more alveolar-type affricate productions with female speech, and a move away from these norms in conversations with male interviewers. Men use [ts] more with female interlocutors than in scripted speeches, while female politicians use it in both scripted speeches and with female interlocutors, but produce a considerably more backed affricate with male interlocutors, suggesting that the alveolar variant is associated with both professional prestige and feminine identity.

The final social factor under consideration was the political party of the speaker. This factor was selected as a significant predictor of variation for frication period, with the largest range in production across all the variables in the model in Table 4. Socialist politicians in the PSOE produced a frication period that occupied 53.5% of their affricates, on average, while PP politicians’ frication periods were closer to 48.4%. As this tendency is not reflected in the other two models, this may provide yet more evidence of the finding of Gordon et al. (2002); namely, that duration-related measures are subject to greater variability in the production of the sibilants [s] and [ʃ]. Unlike phenomena related to liquid neutralization and opacity, referenced by Pollock (2023a), as well as intervocalic /d/ elision, described by Cruz-Ortiz (2022), Pollock and Wheeler (2022), Pollock and Wheeler (forthcoming) and others, which offer political actors of certain affiliations resources to distinguish themselves from their peers and craft an identity rooted in regional identification, the affricate is not so well-integrated into the indexical norms of Andalusian politics. While greater variation does seem to exist in cities like Madrid, these differences are not sufficiently consistent to fall outside the realm of random chance, as illustrated in Figure 6. However, there is evidence to suggest that this may be changing.

5.3. A Labovian Sociolinguistic Marker

The final question addressed by this study involved the role of the Spanish affricate in the Peninsular speech community, and whether it has acquired social stratification to the point that it could be considered a sociolinguistic marker, rather than an indicator. Based on the description by Labov (1972), this would mean that there is variability not only at the linguistic level, but also the extralinguistic. Given the tendencies described in the three mixed-effects regression models, it is clear that the linguistic environment, tonicity, and COG do not on their own account for all variability in use—instead, variable use also differs based on speaker gender, speech context, political affiliation, and city, as well as lexical frequency and tonicity (which may serve as initial signals of incipient change).

The contexts in which the realizations coded as alveolar, as well as productions with higher centers of gravity and longer frication periods, are used—by women, in scripted speeches, in interviews with women, as well as in Madrid and in tonic syllables—are traditionally associated with contexts of prestige and attention paid to speech (e.g., Flores 2014; Labov 1972). This comes alongside the fact that these speakers are major politicians at the regional and national level, often with considerable education and a cultivated public persona, all of which are aspects of identity associated with prestige. Finally, it is noteworthy that none of these individuals produced the deaffricated post-alveolar fricative [ʃ], which is frequently associated with rural and working-class Andalusian speech. Only the post-alveolar and alveolar affricates were identified across three dozen hours of scripted discourse and interviews, suggesting that even if politicians may use certain regional forms to index working-class identity (e.g., Hernández-Campoy and Cutillas-Espinosa 2013; Pollock 2023a), the spirantized allophone of the affricate is too stigmatized to serve as a resource for identity construction among these politicians.

Interestingly, as there was only a slight connection between political party and frication duration, and a general tendency for the alveolar affricate [ts] to be favored by Madrid politicians, this phenomenon may still be in the early stages of acquiring social meaning and becoming a sociolinguistic marker. Given its increased use preceding front vowels, as well as its appearance in tonic syllables, the phenomenon may have initially emerged from coarticulatory effects, particularly in female speech. However, based on the social and frequency-related variables that now condition duration, COG, and perception, this seems to be undergoing indexicalization as speakers become more aware of it and it takes on social meaning (Silverstein 2003).

As the Spanish sibilant inventory contains only a single affricate, variability in its production is not fully surprising. However, should alveolar production of the Spanish affricate acquire further social meaning and currency in Peninsular Spanish, there is also the opportunity for the neutralization of phonological contrast for some Andalusian speakers. As described by Torreira (2006), Ruch (2012), Del Saz (2019, 2023), and Vida-Castro and Villena-Ponsoda (2016), among others (also see Footnote 1), there is a separate process whereby the alveolar fricative /s/ undergoes elision in /st/ clusters, yielding not only productions with post-aspiration [t^h], but also those with affrication [ts]. Perceptual work by the aforementioned authors has also identified comparisons between [ts] resulting from /st/ and /tʃ/ segments. In varieties where both this phenomenon and the fronting of the affricate coexist, further examination is necessary to determine if productions of [ts] differ in minimal pairs like hecho ‘fact’ [etso] and esto ‘this’ [etso]. In order to avoid the loss of contrast, the Andalusian sibilant affricate inventory may develop additional complexity to allow for distinction through duration, COG, or some other measure.

6. Conclusions

In addition to offering insight into the phonetic, linguistic, and social outlook of the Spanish affricate, this study also opens the door to future research and methodologies to better conceptualize the phenomenon of Peninsular affricate fronting. More perceptual work is necessary not just to measure speaker attitude and awareness of the difference between the alveolar and post-alveolar variant, but also to determine the perceptual threshold of differences in segments, as a means of better reinforcing and refining the development of a normalization procedure for sibilant consonants. This, in turn, will help determine the exact status of the phenomenon in Andalusia and Madrid, in addition to offering a concrete understanding of the differentiation that native Spanish speakers carry out between adjacent fricative sounds. While it may not be the case that speakers have a completely overt awareness of the social distinctions between these two allophones, there does seem to be an awareness of differences in affricate production that maps onto some form of prestigious, feminine identity. This should be examined further, particularly outside of the realm of political speech, to understand how widespread this phenomenon may be in Peninsular varieties of Spanish.

Overall, this analysis identifies systematic differences in production norms for affricates by Peninsular politicians and determines that both linguistic and social variables condition usage. As a result, the Spanish alveolar affricate has the characteristics necessary to be described as an incipient Labovian marker that shows initial signs of social stratification. There is consistent variability in production across individuals, stratification across social groups, and signs that the alveolar production is associated with female, emphatic, and prestige speech particularly in NCPS, with growing use in Andalusian Spanish as well.