‘Aquí toman mucho sopa’: Linguistic variables as predictors of non-standard gender agreement production in Basque Spanish

1 Introduction

The language contact situation gives rise to specific linguistic features that may not be present in non-contact varieties of the same language. This is, for example, the case of the language contact between Basque (Euskara) and Spanish in the Basque Autonomous Community (BAC): the Spanish spoken in this territory (henceforth BAC Spanish^[1]) presents certain linguistic features that may not be found in non-contact varieties of Spanish. Among others, one such feature is the production of non-standard gender agreement (N-SGA): these are gender agreement (GA^[2]) cases that do not adhere to the Spanish prescriptive rules, different from standard gender agreement (SGA) cases that do adhere to Spanish prescriptive rules. One example of N-SGA is Un montón de caseríos estaban quedándose viejas ‘A lot of farms [masc.] were becoming old [fem.]’ where the expected SGA would be Un montón de caseríos estaban quedándose viejos ‘A lot of farms [masc.] were becoming old [masc.]’ (example from Basterretxea Santiso 2022, 21). The language contact situation between Basque and Spanish has been found to be the reason for N-SGA production: while Spanish contains grammatical gender (Real Academia Española [RAE] 2016), Basque does not have this feature (Euskaltzaindia 2002, Zubiri and Zubiri 2012).

N-SGA production is not exclusive to BAC Spanish, and it has been extensively studied in language contact situations involving other languages (e.g., English and Spanish, Italian and German, Arabic heritage speakers, or Spanish and Dutch), but mainly through language acquisition lenses and their theoretical frameworks (e.g., Albirini et al. 2013, Avelino Sierra 2021, Balam et al. 2021, Grey et al. 2015, Issa et al. 2020). Thus, the need for variationist sociolinguistic studies exploring N-SGA production in language contact situations is evident. This is also particular to BAC Spanish, as no previous empirical study can be found to delimit the population producing N-SGA or the sociolinguistic factors that impact its production. A previous non-empirical study has associated N-SGA in BAC Spanish with elder native speakers of Basque without much access to education (Fernández Ulloa 1997). A possible reason to assign N-SGA to this population may be that elder BAC inhabitants are believed to produce more linguistic features associated with Basque for historical limitations in education (Camus Bergareche 2021). However, as argued by Basterretxea Santiso (2022), Fernández Ulloa’s (1997) affirmation is not based on an empirical study as he demonstrated that N-SGA in BAC Spanish is also present among young Basque native speakers with access to education. Therefore, we are unaware of the population that produces N-SGA in BAC Spanish neither of the linguistic contexts where N-SGA may be present in this community. By conducting a synchronic variationist study, the aim of this article is not to report potential GA ‘errors’ in BAC Spanish, but rather to investigate the underlying linguistic reasons why N-SGA is produced in language contact situations in general (previous investigations have reported some contradictory results) and in BAC Spanish in particular.

2 Background literature

3 Research question

Taking into consideration the information presented above, this article aims to answer the following research questions:

1. Is the production of N-SGA limited to elder Basque native speakers with limited educational opportunities as suggested by Fernández Ulloa (1997)?

2. To what extent do speakers of BAC Spanish produce N-SGA based on the following list of linguistic variables?

1. Controllers’ gender;

2. Inflectional form;

3. Familiarity with the agreeing words;

4. Distance between targets;

5. Noun animacy;

6. Number of the controller;

7. Type of agreement;

8. Type of clause.

Based on what scholars analyzing GA have previously found, the following are the anticipated hypotheses:

1. As suggested by Palacios (2021), and opposing the generalization made by Fernández Ulloa (1997), N-SGA is expected not to be exclusive to elder Basque native speakers without much access to education, but a widely distributed linguistic feature present among all participants.

2. More N-SGA will be produced when the controller’s ending is non-canonical (e.g., Alarcón 2011, Bianchi 2012, Tarova et al. 2023). When speakers are less familiar with the words, more N-SGA will be produced (Delgado 2018, Hur et al. 2021, Husein 2021). Similarly, as the distance between targets increases, the likelihood of producing more N-SGA will increase (e.g., Díaz Barajas and Orozco 2019, Goebel-Mahrle and Shin 2020, Lipski 2015). More N-SGA is also expected when the controller is an inanimate noun due to its gender being acquired later (e.g., Andersen 1984, Casado et al. 2021, Finnemann 1992). Additionally, more N-SGA is anticipated with singular controllers (e.g., Albirini et al. 2013, Di Pisa and Marinis 2021), with clitics (Goebel-Mahrle and Shin 2020, Torres Sánchez 2021), and in relative clauses.

In order to answer the research questions of the present article, a synchronic variationist study that considers the effect of Basque into Spanish has been conducted. This will contribute to the understanding of N-SGA production in language contact situations and the functioning of GA systems in these situations.

4 Methodology

Participants have been recruited in the three provinces that form part of the BAC where Basque is in contact with Spanish: Bizkaia, Gipuzkoa, and Araba. As mentioned earlier, Basque is also in contact with Spanish in Nafarroa and it is an official language both in the BAC and Nafarroa. Nevertheless, Basque and Spanish are in contact to a lesser extent in Nafarroa (5–6%; Altuna Zumeta et al. 2022) and the sociolinguistic situation is more complex than in the BAC (Basque is only co-official in Nafarroa in the areas where Basque is spoken). Consequently, no participants were gathered in Nafarroa for the present study.

The final sample for the analysis is formed of 73 participants that were contacted through the researcher’s personal contacts, cultural associations, and groups across the three territories in the BAC, and other institutions such as Euskaltegiak (centers for adults to learn Basque). In addition, the snowball sampling method was also employed as participants used their contacts to help the researcher gather more data (Milroy and Milroy 1985). With these participants, a randomly filled judgment sample was created (Hoffman 2014, Meyerhoff et al. 2015, Schilling 2013, Silva-Corvalán and Enrique-Arias 2017). The background information of the 73 participants can be observed in Table 1: all are 18 years or older (age range: 18–90; age median: 41), and have lived in the BAC since their childhood. Something to note is the absence of monolingual Basque or Spanish speakers in the analysis. As it was introduced in Section 2, monolingual Basque speakers are practically non-existent nowadays. Similarly, the number of Spanish monolingual speakers has decreased importantly in the last 30 years, mainly among younger speakers due to the implementation of Basque in education (Gómez Seibane 2020). That explains why no participant in this project identified as being a Spanish monolingual speaker, and vice versa.

Following scholars who have urged for the necessity of analyzing spontaneous and authentic speech in language contact situations (Palacios 2021), and as part of a bigger research project, individual sociolinguistic interviews were conducted (based on Labov’s thematic framework, 1972). Due to geographical distance and COVID-19 pandemic, most of the interviews were conducted online.^[14] The interviews lasted at least 45 min and a maximum of 60 min. A total of 59 h and 40 min of recordings were collected, of which 47 h, and 10 min are analyzed in this article.^[15]

In terms of statistical analysis, the type of GA produced by the participants is the dependent variable: this is any linguistic item that displays GA in Spanish. For this, automatic orthographic transcriptions obtained through YouTube were reviewed and edited by the research team members and re-read to identify GA tokens. This dependent categorical variable is subdivided into three categories: SGA or instances in which standard Spanish rules are followed as in (5), N-SGA or GA instances that do not follow Spanish standard rules as in (6), and NA or GA instances that do not allow to determine whether it is SGA or N-SGA as in (7) (non-canonical adjectives).

Regarding independent variables, linguistic predictors previously found to be important in N-SGA production in language contact situations have been included in the analysis, and Table 2 presents a summary list.

The first linguistic predictor is the Gender of the principal noun (controller), subdivided into masculine or feminine. Since masculine is considered to be the default gender in Spanish (Harris 1991), it is expected that more N-SGA will be produced when the controllers’ gender is feminine as they will overuse the masculine gender (e.g., Tarova et al. 2023).

The second linguistic predictor considered in the analysis is the controllers’ Inflectional form. This follows previous studies that have revealed that more N-SGA is produced when the ending of the controller is non-canonical for its gender (e.g., Hur et al. 2021, Tarova et al. 2023). This predictor is subdivided between nouns with canonical endings (-o for masculine and -a for feminine) and nouns with non-canonical endings (nouns that do not end in -o for masculine and nouns that do not end in -a for feminine).

The third linguistic predictor is the Frequency of the words that agree in gender. In other words, this would be the proxy for speakers’ familiarity with the words they are using to produce GA. In order to use this variable, a corpus internal measure of frequency was created in R (R Core Team, 2022): a list of the 50 most repeated words across participants, after excluding Spanish stopwords (e.g., además ‘in addition,’ ahora ‘now,’ aquí ‘here,’ tú ‘you,’ en ‘in’). The 50 most repeated words are considered as familiar or frequent for the speakers. This includes not only the singular or plural versions of the words, but also variations in morphology (e.g., diminutives; viaje, viajes, viajecito ‘trip,’ año, años, añitos ‘year,’ or cosa, cosita, cosilla ‘thing’).

Another linguistic predictor that is considered is the Distance between the agreeing elements. Some scholars have coded distance in terms of linearity or the number of words between the targets (e.g., Paquet 2018, Pinta 2022), while others have coded it in terms of syntactic distance (e.g., Díaz Barajas and Orozco 2019). In this article, Distance is coded in terms of linearity and adjacency between targets, following Goebel-Mahrle and Shin (2020) and Pinta (2022). Therefore, Distance is divided between (i) immediately adjacent agreement, (ii) non-adjacent but in the same clause, (iii) non-adjacent but in different clause, and (iv) absent controller.

The fifth linguistic variable is nouns’ Animacy: animate vs inanimate nouns. The consideration of this linguistic predictor is based on the fact that inanimate nouns’ gender is typically later acquired when compared to the acquisition of animate nouns’ gender in Spanish (Andersen 1984, Casado et al. 2021, Fernández-García 1999, Finnemann 1992, Hernández-Pina 1984).

The next linguistic predictor is the controllers’ Number: singular vs plural. Scholars have previously found opposing results based on this linguistic variable in terms of N-SGA production. For instance, Di Pisa and Marinis (2021) found no difference based on Number when producing GA in Italian. In this sense, Gonzalez et al. (2019) and Gudmestad and Edmonds (2021) for Spanish, and Albirini et al. (2013) for Arabic, have reported contrasting findings.

The second to last linguistic variable is Type of agreement, or in other words, word class (e.g., Goebel-Mahrle and Shin 2020, Martinez-Nieto and Restrepo 2022, Valdés Kroff et al. 2019). This independent predictor is subdivided into (i) object clitic, (ii) determiner, and (iii) adjective.

The last linguistic variable is Type of clause (similar to, for instance, Torres Sánchez 2021). This variable is subdivided between (i) NP (nominal phrase), (ii) PA (Predicate Adjective), and (iii) relative clause.^[16]

5 Results

Before conducting descriptive and statistical analyses, NA GA instances (744) were excluded because these are word usages that have the same form for both genders with non-canonical adjectives as in el pueblo [masc.] es grande ‘the town [masc.] is big’ or la ciudad [fem.] es grande ‘the city [fem.] is big.’ In addition, following previous scholars who had made the same decision (e.g., Alarcón 2011, Paquet 2018, Pinta 2022, Trawick and Bero 2021), GA instances with animate controllers were excluded from the final analysis because, different from inanimate nouns,^[17] the gender of animate nouns is determined by their semantic information or social gender identity (Harris 1991). Consistent with this fact, only four instances of N-SGA with animate controllers (0.61%) were found in the interviews, illustrated by examples (8) and (9). On the contrary, 650 N-SGA instances (99.39%) were produced with inanimate controllers. It is also notable that no participant employed non-binary GA forms with animate controllers (e.g., -e, -x, and -@).

Then, after the aforementioned exclusions, Table 3 presents the final GA numbers.

The results reveal that, in general, participants exhibited a low frequency of N-SGA production. However, all participants demonstrated this type of instance in their speech, as introduced in Figure 1.^[18] This means that no participant was excluded from the N-SGA analysis.

Figure 1

GA production by each participant.

Before conducting statistical analyses, and in order to test for possible collinearity between independent linguistic variables introduced in this article, a collinearity test with the performance package (Lüdecke et al. 2021) was conducted in R (R Core Team 2022). Results of this test in Table 4 demonstrate a weak correlation^[19] between the linguistic variables, with an exception: Distance and Type of clause. This is a logic result as descriptive statistics below show that SGA in NP is more common when the distance is shorter, followed by PA, and then, relative clauses. Considering that Distance shows a more comprehensive breakdown with four subcategories, Type of clause variable is omitted from further statistical analyses.

Taking the results from the collinearity test into consideration, stepwise binary logistic regressions were performed to statistically assess the impact of the linguistic predictors: Gender of the controller, Inflectional form, Frequency of the controller, Distance between targets, Number of the controller, and Type of agreement. The analysis included random effects for participants, as in the study by Goebel-Mahrle and Shin (2020), because this enables to explore the impact of the independent variables in the dependent variable (type of GA). Interactions between the dependent variable and independent variables were manually added following a stepwise procedure (Crawley 2007). Then, the independent variables that improved the model according to ANOVAs (<0.05) were kept for the final model.

Table 5 presents the results of the final model,^[20] in which, previously, a factor identified as non-significant was excluded: Frequency. According to Ferguson’s (2009) criteria, a strong effect was found for this model (R² = 0.513), which indicates that a large amount of the variation is accounted for by this model.

Following the results in Table 5, the ANOVAs obtained suggest that there is a main effect for the Distance between the agreeing elements (X ²[3] = 657.22, p ≤ 2.2 × 10⁻¹⁶), and the controllers’ Gender (X ²[1] = 259.48, p ≤ 2.2 × 10⁻¹⁶). Indeed, these are the two variables that most affect variation. Table 5 and results from the ANOVAs also indicate that there is a main effect for Number (X ²[1] = 54.91, p ≤ 1.259 × 10⁻¹³). Finally, the two linguistic independent significant variables that affect N-SGA production are Type of agreement (X ²[2] = 6.59, p ≤ 0.037), and Inflectional form (X ²[1] = 4.91, p ≤ 0.026).

In what follows, significant linguistic predictors’ effects are explored in more depth to better understand their effect on N-SGA production, following the order in which they have been shown to impact N-SGA production. As it was aforementioned, the first one is Distance. According to descriptive statistics in Table 6, higher rates of N-SGA occur when the controller is absent in the sentence, as in (10), followed by when the controller is non-adjacent and in a different clause, as in (11). Then, N-SGA production is almost non-existent when the controller is immediately next to the targets, as in (12). This is similar to when the controller is not immediately next to the targets, but in the same clause, as in (13).

After descriptive statistics, post-hoc pairwise comparisons were conducted to further explore the differences between the distance groups. These results are presented in Table 7: there are statistical differences in terms of N-SGA production between the different distances. This indicates a direct correlation between distance and the quantity of N-SGA produced: as the distance increases, N-SGA instances also increase significantly.

The second most affecting independent predictor was controllers’ Gender, as shown by descriptive statistics in Table 8. These results indicate that there is a stronger prevalence of N-SGA when the controller is feminine, as in (14), while N-SGA with masculine controllers is nearly non-existent, as in (15).

The third most significant independent factor that emerged according to the ANOVAs was controllers’ number. Results for this independent variable are presented in Table 9: there is a statistically higher number of N-SGA cases with singular controllers, as in (16), compared to plural controllers, as in (17).

Type of agreement was found to be the second-to-last most influential predictor for N-SGA production. The results for this variable in Table 10 indicate that more N-SGA instances are found when the target is a direct object clitic (lo, la, los, and las), both when the clitic is attached to the verb as in pagarla ‘to pay it,’ or positioned before the verb, as in (18). This happened mainly with masculine forms lo and los. Then, N-SGA instances are less common with adjectives, as in (19), and even less frequent with determiners, as in (20).

In this case, post-hoc pairwise comparisons in Table 11 reveal a significant difference between adjectives and determiners, and another significant difference between clitics and determiners. However, there is an absence of statistical difference between adjectives and clitics.

Inflectional form emerged as the last significant predictor for N-SGA production. Results for this variable in Table 12 indicate that there is a notably higher frequency of N-SGA instances with non-canonical controllers, as in (21), when compared to canonical controllers, as in (22).

In sum, logistic regressions and ANOVAs informed that Distance, Gender of the controller, Number, Type of agreement, and Inflectional form are the significant linguistic factors that predict N-SGA production in BAC Spanish. Conversely, Frequency of the controllers did not emerge as a significant predictor.

6 Discussion

The first research question of this article aimed to delimit the presence of N-SGA production in the BAC Spanish population. According to the results, and confirming the hypothesis (following Palacios 2021, N-SGA would be present among all BAC Spanish speakers and not only among elder Basque native speakers with limited access to education, as suggested by Fernández Ulloa 1997) N-SGA has been found among the three different generations that were included in this research. However, the presence of N-SGA is relatively limited (N-SGA = 6.43%), consistent with previous studies analyzing N-SGA in language contact situations. To introduce a case, Torres Sánchez (2021) found that the N-SGA production rate was 5.27% in the contact between Spanish and O’dam in Mexico, which is 1.16 points lower than the overall result obtained in this article. The reason for the low N-SGA production rate in BAC Spanish might be the transparent nature of gender marking in this language, where almost all masculine nouns end in -o and most feminine nouns in -a (Teschner and Russell 1984). However, the low rate does not imply that these are random cases of variance. On the contrary, the results support the principal hypothesis: linguistic variables systematically influence N-SGA production in BAC Spanish.

Indeed, the second research question was interested in exploring the extent to which BAC Spanish speakers produce N-SGA in Spanish depending on the linguistic variables considered in the analysis. Specifically, one of the hypotheses was that more N-SGA would be produced if the controller is an inanimate noun due to its gender being acquired later in Spanish. This was confirmed by the results as N-SGA in BAC Spanish is nearly absent when the controller is an animate noun. The initial hypothesis is further confirmed since almost all the considered linguistic variables were found to be significant predictors: controllers’ Gender, Inflectional form, Distance, Number, and Type of agreement.

Regarding controllers’ Gender, when the controller’s gender is feminine, significantly more N-SGA is produced, which aligns with previous research showing masculine as the default gender in Spanish (e.g., Balam et al. 2021, Boers et al. 2020). Therefore, similar to Spanish learners (e.g., Sanz 2016, Tarova et al. 2023), BAC Spanish speakers tend to overgeneralize the use of masculine gender. Nevertheless, this overgeneralization is relevant for inanimate nouns but not animate ones, which resembles other languages where the feminine is mainly relevant for animate nouns (e.g., Diyari/Dieri language: an Australian Aboriginal language spoken in the north of South Australia) (Austin 1981, Corbett 1991, Kramer 2015).

Furthermore, this article anticipated that more N-SGA would be produced when the controller’s ending was non-canonical. This was confirmed as in BAC Spanish, more N-SGA instances have been found when the controller is non-canonical. This means that canonical controllers are seen as transparent indicators of gender when compared to non-canonical controllers. This article supports previous studies that found Inflectional form to be a significant predictor for N-SGA production in language contact situations (e.g., Alarcón 2011, Balam et al. 2021, Boers et al. 2020, Pfaff 1979).

It was also hypothesized that as the distance between targets increases, the likelihood of producing more N-SGA increases. In this case, this hypothesis was confirmed as well: when the distance between the targets is bigger, more N-SGA is produced (e.g., Díaz Barajas and Orozco 2019, Goebel-Mahrle and Shin 2020, Lipski 2015, Pinta 2022). This might be explained by accessibility: when the controller is more accessible in the memory of the speakers, there is a lower extent of N-SGA production.

The hypothesis regarding Number of the controllers was also confirmed: this was found to be a significant factor and more N-SGA instances were identified with singular nouns (e.g., Albirini et al. 2013, Gonzalez et al. 2019, Gudmestad and Edmonds 2021). This might be due to the varying number of controllers in each grammatical category: 6,999 singular controllers and 3,104 plural controllers.

Type of agreement was also significant, and the hypothesis was once again confirmed since the highest N-SGA rates were produced when the target was a clitic. This is similar to previous studies that found a big use of the masculine clitic lo ‘it’ to refer to a feminine controller (e.g., Goebel-Mahrle and Shin 2020, Torres Sánchez 2021). Finally, Frequency was not found to be significant, which might be related to the difficulty in accurately measuring word frequency in spontaneous speech (Goebel-Mahrle and Shin 2020) or for participants only using frequent words or words that are well-known to them.

On another note, and following other recent findings (e.g., Hunt Gardner and Kostadinova 2024), the results in this article support the inclusion of online interviews into language contact analysis as no notable disparities have been observed between online and in-person interviews. However, the fact that not all participants may be familiarized with online platforms may be considered a possible limitation in conducting interviews.

7 Conclusions

The analysis has shown that, following Labov (1966), linguistic variation is not random with regard to N-SGA in language contact situations with one of the languages involved not producing GA. Indeed, contrary to Fernández Ulloa (1997), N-SGA in BAC Spanish would be present among all speakers of the aforementioned Spanish variety. These results imply the existence of a specific GA system in BAC Spanish influenced by the language contact situation between Basque and Spanish: feminine is mainly relevant for animate controllers, and the masculine default gender is more prevalent for inanimate controllers. In addition, this article shows evidence to support the notion of the masculine gender functioning as default in BAC Spanish, consistent with other non-contact Spanish varieties. Therefore, N-SGA instances in BAC Spanish are not isolated cases, but form part of a particular GA system that is influenced by certain linguistic predictors (Gender and Inflectional form of the controllers, the Distance between the target elements, Number, and Type of agreement). This conclusion aligns with those of Palacios (2021) and Ramírez Cruz (2009), who have considered N-SGA production in other language contact situations and support the existence of a local GA system. This does not imply, however, that the GA system in BAC Spanish is at risk of losing its system affected by Basque.

Finally, considering future perspectives, social variables should also be considered as possible predictors for N-SGA (e.g., speakers’ origin and province of residence, level of education, bilingual profile) as this article only examined the impact of linguistic variables. In addition, the same linguistic phenomenon should be explored in Nafarroa and Iparraldea: this will enable us to delimit the existence of N-SGA production in the Basque Country with different linguistic situations.