Gender Assignment Strategies Among Simultaneous Spanish/English Bilingual Children from Miami, Florida

3.1 Analogical Gender

The analogical criterion^[5] (i.e., when the gender assigned to an English-origin noun is based on the gender of the Spanish translation equivalent) has been found to be an influential factor in mixed nominal constructions. Jake et al. (2002) examined oral production data from 10 Spanish/English adult bilinguals of South American background working in the U.S. or studying in U.S. graduate schools. In their examination of 151 noun phrases (NPs) consisting of gender-marked Spanish determiners and English nouns, they found that analogical gender was the strongest predictor of gender assignment. Whereas analogical gender accounted for 52% (78/151) of English-origin nouns, the phonological criterion only accounted for 23% (35/151) of mixed DPs. Consonant with Jake et al.’s findings, in more recent work, adherence to the analogical criterion has also been found among simultaneous Spanish/English adult bilinguals from Gibraltar (Liceras et al. 2016: 122); El Paso, Texas (Królikowska et al. 2019); and New Mexico (Rodríguez-González et al. 2019).

In the case of children, the application of the analogical criterion is less clear. Previous work suggests that Spanish/English simultaneous bilingual children display a tendency to favor the analogical criterion, but this has been attested only in acceptability judgment data (for relevant discussion, see Bellamy et al. 2018). Liceras et al. (2012) examined judgment data from sequential and simultaneous Spanish/English bilingual children on gender assignment and agreement structures. They found that while both groups showed adherence to the analogical gender, simultaneous bilingual children were less sensitive to it than their sequential bilingual counterparts who more strongly adhered to the gender of the translation equivalent (cited in Liceras et al. 2016). Thus, based on previous findings, it remains inconclusive how simultaneous Spanish/English bilingual children employ the analogical gender in oral production.

It is noteworthy that in studies involving other language pairs, some evidence for the employment of the analogical criterion has also been reported. For example, Radford et al. (2007), in their analysis of longitudinal data from four French/German children (between ages 1;5 and 5;1), found that children relied on the analogical criterion. When children produced mixed DPs with French articles and masculine or feminine German nouns, they generally matched the gender of the French article to the gender of the German noun. At the same time, however, a point we further discuss in Section 3.2, Radford et al. also reported the application of the masculine default gender with neuter German nouns.

3.2 Masculine Default Gender

The employment of a default gender,^[6] which is the masculine in Spanish (Roca 1989), has also been argued to play a predominant role in the gender assignment process in bilingual speech (e.g., Balam 2016a; Otheguy and Lapidus 2003; Valdés Kroff 2016, among others). In his study, Valdés Kroff (2016) found that masculine-marked mixed nominal constructions accounted for 96% of data produced by adult Spanish/English bilinguals from Miami, Florida. Crucially, a total of 34.8% (103/296) of English nouns that were assigned the masculine gender had feminine Spanish translation equivalents. These included cases in which human referents that were transparently feminine were assigned the masculine default gender as in un renaissance woman ‘a renaissance woman’. Similar examples have also been reported for Spanish/English bilinguals in Northern Belize (Balam 2016a) and New York (Otheguy and Lapidus 2003).

This tendency to assign the masculine default gender to English nouns appears to be more pronounced in Spanish/English communities where bilinguals more frequently engage in code-switching (Balam 2016a; Królikowska et al. 2019). Naturalistic speech data have shown that in contexts that are characterized by ubiquitous code-switching, it is not surprising to find that speakers completely neutralize the masculine/feminine gender distinction in bilingual speech (i.e., only masculine-marked mixed nominal constructions are produced), as it has been attested in Northern Belize (Balam 2016a).

In the case of bilingual children, there is less conclusive evidence as to the employment of the masculine default gender in oral production (Liceras et al. 2008, 2016). Importantly, scholarly work on French/German bilingual children offer insight into the use of this strategy. Radford et al. (2007) found that when French-German children mix between French articles and German neuter nouns, children largely employed the masculine default gender. Along these same lines, Eichler et al. (2012: 253) found evidence that children use the masculine gender as the default gender. Eichler et al. analyzed 707 mixed DPs from 15 simultaneous bilingual children (Spanish/German, Italian/German and French/German children between ages 1;0 and 5;4) with different degrees of language dominance. Eichler et al. found that when combining Romance determiners with German neuter nouns in mixed nominal constructions, bilingual children overwhelmingly employed the masculine default gender strategy.

While it is true that the gender system configuration between French/German (two genders vs. three genders) and Spanish/English (two genders vs. no gender) are different, it is nonetheless remarkable that at very early stages of bilingual acquisition the masculine default strategy has been found to be operative among bilingual children with different degrees of language dominance. Based on current research, however, it remains unclear whether in the case of Spanish/English bilingual children, there is any preferential or systematic use of the masculine default gender in bilingual speech.

3.3 Phonological Gender

Lastly, the phonological gender criterion (i.e., when gender is assigned based on the terminal phoneme of an English noun) has been argued to play an influential role in the gender assignment process (Clegg and Waltermire 2009; DuBord 2004; Jake et al. 2002; Parafita Couto et al. 2016; Poplack et al. 1982; Smead 2000). The terminal phoneme (TP) of English nouns is relevant given that Spanish TPs are relatively predictive of gender assignment. Although there are many exceptions (for examples, see Gathercole 2002: 209–210; Roca 1989), the prototypical terminal phonemes –o and –a are predictive of the masculine and feminine gender in Spanish respectively. Nouns with non-canonical TPs (e.g., nouns that end in –d, –e, –u, etc.), which are less predictive, are usually more difficult to acquire and use than those ending with the canonical masculine and feminine morphemes (Brisk 1976; Teschner and Russell 1984, but see Fernández Fuertes et al. 2016).

Of relevance to our analysis is that there are certain Spanish TPs that favor the masculine gender while others favor the feminine gender (Teschner and Russell 1984). TPs that favor the masculine gender include –n, –o, –r, –s, –e, and –l, or NORSEL, whereas TPs that favor the feminine gender include –a, –ad, –ión, –is, or ADIONIS (Clegg and Waltermire 2009; DuBord 2004; Smead 2000; Teschner and Russell 1984). Support for the influence of the phonological criterion was found in a study conducted by Clegg and Waltermire (2009), who examined 204 Spanish/English NPs extracted from spontaneous conversations with 15 Spanish/English adult bilinguals from Northern New Mexico. In contrast to previous work (Jake et al. 2002), Clegg and Waltermire’s variationist analysis showed that the gender associated with the TP of the English noun was more predictive than analogical gender (for relevant discussion, see Valdés Kroff 2016).

We must highlight that an inherent challenge that arises when examining the phonological gender, as aptly pointed out by Clegg and Waltermire, is that the feminine gender morpheme –a is not a productive TP in English lexical items (also see Jake et al. 2002: 82). Contrariwise, masculine-marked morphemes (i.e., NORSEL) are productive; thus, presenting us with an unavoidable confound. When speakers assign the masculine gender to an English noun that has a feminine translation equivalent (e.g., el _masc equirrel ‘la_fem ardilla_fem’), do we attribute this to the masculine default gender, to the masculine phonological gender, or to both? In Section 4, we describe the methodology that we employed to investigate children’s gender assignment patterns in monolingual Spanish versus mixed nominal constructions. We also explain the coding scheme that we used to address the aformentioned confound in order to analyze the use of the masculine default gender in mixed nominal constructions.

4.1 Research Questions

Our study was guided by the following two questions.

1. What are the overall gender assignment patterns in children’s production of monolingual Spanish and mixed nominal constructions?

In line with results from previous research (Gathercole 2002; Montrul and Potowski 2007), we anticipated that two-way bilingual fifth graders would show fewer gender mismatches in monolingual Spanish nominal constructions than second graders in English immersion programs. In Spanish/English nominal constructions, our expectation was that they would primarily be masculine-marked. We also anticipated productive use of the feminine gender, as a result of the employment of the analogical gender criterion in mixed nominal constructions.

2. How do different gender assignment strategies account for the gender that is assigned to English nouns with feminine translation equivalents?

When assigning gender to English nouns with Spanish feminine translation equivalents, we anticipated that children would exhibit a predisposition towards the application of the analogical gender, in line with Liceras et al.’s (2012) previous findings for acceptability judgment data. Alternatively, following Liceras et al. (2008, 2016, we expected that at least some chidren, particularly second graders, would not show a clear preference for a particular gender assignment strategy.

4.2 Data and Analysis

We analyzed narrative data from the CHILDES database (MacWhinney 2000). The oral narratives, elicited using Mayer’s (1969) wordless picture book Frog Where Are You, were collected by Pearson (2002). We examined 1774 monolingual Spanish and 220 Spanish/English nominal constructions (for complete list of mixed nominal constructions, see Appendix A), extracted from 39 Spanish narratives produced by 39 US-born Spanish/English bilingual children from Miami Dade, Florida. Excluded from the dataset of monolingual Spanish constructions were all cases with el/la agua ‘the water’ (n = 48), as agua is an exceptional hermaphroditic noun that is phonologically-conditioned and exhibits variation in terms of gender assignment (Eddington and Hualde 2008: 15).

The narratives for the present analysis were selected via purposeful sampling. Given our interest in gender assignment in bilingual speech, only children who produced at least one mixed nominal construction were selected for the analysis. Note that when telling the Spanish stories to their interlocutors (who knew Spanish and English), the children were in a bilingual mode. They used mainly Spanish with some code-switching to English in these stories. The bilingual nominal constructions overwhelmingly comprised cases in which a gender-marked Spanish determiner co-occurred with an English noun (e.g., un rock ‘a rock’, un reindeer ‘a reindeer’, una owl ‘an owl’, etc.).

As Table 1 shows, the sample included nineteen second graders (ages 7;0–8;0) and twenty fifth graders (ages 10;0–11;0) with different schooling backgrounds. The children across the two instructional programs and grade levels came from homes where the language of the home was either ‘Mostly Spanish’ or ‘Spanish and English equally’ (from birth until age 5). All of them had been in the same educational program since kindergarten and only those children who had experienced the educational program for two years were included in the original study conducted by Pearson (2002) and her colleagues. Note that despite this distinction between the language of the home as ‘Mostly Spanish’ or ‘English and Spanish equally’, by the time children moved to the second grade, they all tended to be English-dominant (see Pearson 2002: 171).

The original sample in our study included twenty second graders, but there was an English immersion second grader (ID11221366) whose data showed exceptional patterns. The gender assignment patterns of this child skewed the data, particularly monolingual Spanish tokens, which mostly comprised gender assignment mismatches in which masculine nouns were assigned the feminine gender (e.g., *la niño ‘the boy’ and *la perro ‘the dog’, 92%, 36/39). Notably, in mixed nominal constructions (n = 9), this child assigned the feminine gender to all English nouns. It is not clear why this child overwhelmingly employed the feminine gender. It may constitute an idiosyncratic case in which a feminine default gender was applied as opposed to the canonical masculine default gender. Given that this child’s gender assignment patterns markedly deviated from those attested among the remaining children, we excluded these data in our regression analyses.

To analyze the data, we extracted all monolingual Spanish nominal constructions and mixed nominal constructions from the same narratives produced by each child. Each nominal construction was then coded for gender on the determiner (masculine vs. feminine); noun gender (i.e., gender of the Spanish noun or translation equivalent in the case of mixed nominal constructions); phonological gender (i.e., gender based on NORSEL and ADIONIS); instructional program (i.e., English immersion vs. two-way bilingual); and grade level (i.e., second vs. fifth graders). The narrative transcripts provided the information on these extra-linguistic factors. Each transcript had a unique ID number, which provided information on the child’s background. We extracted information only from digits 1 (instructional program) and 4 (grade level).

For Spanish nominal constructions, noun gender refers to the gender of the noun as an inherent lexical property (i.e., masculine or feminine). In the case of mixed nominal constructions, we coded noun gender according to the gender of the Spanish translation equivalent. Translation equivalents for English-origin nouns, both masculine- and feminine-marked, were provided by Pearson (2002) in the CHILDES transcripts.^[7] Note that these translation equivalents took into account the child’s monolingual Spanish discourse and dialectal differences in the use of specific lexical items (for a complete list, see Appendix A). In the few cases where equivalents were not provided by Pearson and colleagues, we chose the equivalent whose token frequency was the highest for the particular sub-group of children.

To examine the employment of the masculine default gender, we only analyzed English nouns with feminine translation equivalents. As previously pointed out, a problematic confound with masculine-marked mixed nominal constructions with masculine translation equivalents is that we are unable to differentiate among the application of the analogical masculine gender, the phonological gender, and/or the masculine default gender (e.g., the masculine gender in el _masc hole/el_masc hueco_masc can be accounted for by all three gender assignment stretegies).

Therefore, a more reliable method to evaluate the use of the masculine default strategy is to focus on English nouns with feminine translation equivalents that are nonetheless assigned the masculine gender (e.g., el _masc beehive for la _fem colmena _fem ). In order to identify gender assignment cases that could more likely be attributed to the employment of the masculine default gender, we devised a coding scheme (see Table 2) that distinguishes cases of gender assignment ambiguity or overlap (i.e., cases that could be accounted for either by the masculine default gender or the masculine phonological gender) versus those without any overlap (i.e., cases that could only be accounted for by the masculine default strategy).

As Table 2 shows, we coded for ambiguous cases of the analogical masculine gender, which included masculine-marked mixed nominal constructions that could be accounted for by the analogical masculine gender, the masculine default gender, and/or the phonological gender. Our focus, however, was primarily on cases that had feminine translation equivalents. Recall that only these examples are more likely to cast light on the potential employment of the masculine default gender. For mixed nominal constructions containing English nouns with feminine translation equivalents, we differentiated among gender assignment cases that were ambiguous versus those that were not. This coding scheme was particularly advantageous as it allowed us to identify cases that could more likely be accounted for by the masculine default gender, as well as those that could be accounted for by the analogical criterion.

Cases of the unambiguous analogical feminine gender refer to mixed nominal constructions in which the gender of the article only matched the gender of the translation equivalent (e.g., una _fem owl ‘una _fem lechuza _fem’). On the other hand, cases of the unambiguous feminine gender referred to examples where the gender of the article did not match the gender of the translation equivalent, and the feminine gender was assigned in a ‘default-like’ manner (e.g., la _fem other side ‘el _masc otro _masc lado _masc’). In our coding scheme in Table 2, we have used a question mark symbol to indicate that these highly infrequent feminine-marked mixed nominal constructions suggest that children may still be in the process of acquiring an important code-switching grammatical pattern. In our analysis in Section 5.2, we return to this point. For feminine-marked tokens the phonological gender was not relevant, as feminine TPs (i.e., ADIONIS) were not attested. Consonant with previous findings (Clegg and Waltermire 2009; Jake et al. 2002), our data revealed that only English nouns with masculine TPs (i.e., NORSEL) were produced.^[8]

The ambiguous masculine default gender included cases with gender assignment overlap (e.g., un _masc hole ‘un _masc hueco _masc’), in which the gender of the article could be accounted for either by the analogical masculine gender, the masculine default gender, and/or the phonological gender. Lastly, the unambiguous masculine default gender referred to cases where only the masculine default gender could account for the gender assigned to the English-origin noun (e.g., un _masc rock ‘una _fem piedra _fem’). This coding scheme yielded a granular insight into the employment of the ambiguous versus the unambiguous masculine default gender strategy.

Note that following previous work (Clegg and Waltermire 2009; DuBord 2004; Smead 2000), in order to determine phonological gender, we coded mixed nominal constructions using the following categories: TPs that favor the masculine gender (i.e., NORSEL, excluding –s as morphological marker of plurality); TPs that favor the feminine gender (i.e., ADIONIS); and atypical TPs (i.e., other TPs such as –g in el frog ‘the frog’, –k in un rock, and –f in una cliff ‘a cliff’). Of relevance to our analysis were TPs that favored the masculine gender in particular, as these are the TPs that contribute to a confound in the employment of gender assignment strategies. In the ensuing section, we present our results.

5.1 Gender Assignment in Monolingual Spanish Nominal Constructions

As Figure 1 shows, the majority of monolingual Spanish and code-switched nominal constructions were masculine-marked. It is noteworthy that in the monolingual Spanish constructions, masculine-marked tokens accounted for 70% (1242/1774) of the data. This percentage is higher than those reported for adult Spanish/English bilinguals, which are typically less than 55% (e.g., data from Northern Belize: Balam 2016a; data from New Mexico: Clegg and Waltermire 2009; data from New York: Otheguy and Lapidus 2003).

Figure 1:

Masculine- versus feminine-marked nominal constructions.

The data showed that the masculine/feminine gender distinction was fully operative in Spanish nominal constructions (n = 1774) produced by all children (n = 39). Table 3 below shows the total number of masculine- versus feminine-marked Spanish nominal constructions for each sub-group, followed by their respective proportions in relation to the total number of Spanish nominal constructions produced.

In terms of gender assignment mismatches (i.e., cases when a masculine determiner was combined with a feminine noun, as in el _masc rana _fem ‘the frog’, or when a feminine determiner was combined with a masculine noun, as in ‘la_fem niño_masc’), fifth graders enrolled in a two-way bilingual program generally produced the least number of nominal constructions in which the gender on the article did not correspond with the gender of the noun (see Table 4). The highest rates of gender assignment mismatches were attested among two-way bilingual second graders (4.3%) and English immersion fifth graders (3.9%). Crucially, none of the sub-groups’ mean production of mismatches were above 5%; hence, suggesting successful acquisition of grammatical gender among all groups.

In order to determine whether the two independent factors of grade level (second vs. fifth graders) and instructional program (English immersion vs. two-way bilingual) significantly impacted the likelihood of the children’s (n = 39) assignment of accurate grammatical gender in monolingual Spanish nominal constructions (n = 1774), a fixed-effects logistic bionomial regression analysis was conducted (see Table 5) using Language Variation Suite Toolkit, a web-based application for sociolinguistic analysis, built in R (for details, see Scrivner and Díaz-Campos 2016). The dependent variable in the regression model, namely accuracy consisted of two levels (gender assignment matches vs. mismatches). For the dependent variable, the mismatch option was chosen as the reference value.

As Table 5 illustrates, both grade level (0.467) and instructional program (0.313) do not exert a significant effect on the degree of gender assignment accuracy in monolingual Spanish nominal constructions. Thus, despite differential experiences in terms of schooling, children do not differ in terms of gender assignment accuracy in monolingual Spanish nominal constructions.

Among the four sub-groups of children, as Table 6 shows, there were only minor differences in terms of gender mismatches. Particularly in the speech of two-way second graders and two-way bilingual fifth graders, mismatches mostly comprised cases in which feminine nouns were assigned the masculine gender (e.g., *el ventana ‘the window’, *un piedra ‘a stone’, *el rana ‘the frog’, etc.). Note that overgeneralization of the masculine gender among Spanish/English bilingual children in the U.S. has also been found in previous work on gender assignment (Cuza and Pérez-Tattam 2016; Goebel-Mahrle and Shin 2020). For example, in their analysis of gender mismatches among three groups of US child heritage speakers (ages 5;0–6;0, 7;0–8;0, 9;0–11;0), Goebel-Mahrle and Shin (2020) find that the masculine gender is particularly overgeneralized in the context of direct object clitics that refer to feminine nouns.

Contrariwise, among English immersion fifth graders, gender assignment mismatches were primarily cases in which masculine nouns were assigned the feminine gender (e.g., *la sapo ‘the frog’, *la arbol ‘the tree’, *la niño ‘the boy’, etc.). It must be highlighted, however, that this was an exceptional case, as 90% (10/11) of these mismatches were produced by a single child. Importantly, this English immersion fifth grader’s overuse of the feminine gender only applied to Spanish nominal constructions but not mixed nominal constructions. Thus, overall, the masculine gender was overgeneralized to feminine nouns.

In sum, despite differences in grade level and instructional program, children had largely acquired grammatical gender and did not significantly differ in terms of gender assignment accuracy or the types of mismatches they produced in monolingual Spanish nominal constructions.

5.2 Gender Assignment in Mixed Nominal Constructions

In bilingual speech, as Table 7 illustrates, the use of the masculine gender was also predominant across the four sub-groups of children, whereas the production of feminine-marked mixed nominal constructions was infrequent (i.e., 8.2%, 18/220). No statistical difference was found among groups in terms of their gender assignment patterns. Chi-square tests of independence showed that the distributions of masculine- versus feminine-marked mixed nominal constructions and instructional program were independent for both second graders (χ² = 0.268, df = 1, p = 0.605) and fifth graders (χ² = 1.166, df = 1, p = 0.280). They were also independent when analyzing the data according to children’s enrollment either in English immersion (χ² = 0.291, df = 1, p = 0.589) or two-way bilingual programs (χ² = 1.051, df = 1, p = 0.305). This suggests that there is no association between schooling and children’s production of masculine- versus feminine-marked mixed nominal constructions. Irrespective of the grade and program in which children were enrolled, their gender assignment patterns in bilingual speech were largely similar.

The predominant use of the masculine gender and the infrequent employment of the feminine gender closely resemble the asymmetrical pattern that was found among Spanish/English adult bilinguals from Miami, Florida (Valdés Kroff 2016). However, one way in which our results differ from Valdés Kroff’s findings is that not all feminine-marked mixed nominal constructions (n = 18) were characterized by disfluencies, reformulations, repetitions and/or pauses leading up to the mixed nominal construction, as in (4).

Of the eighteen feminine-marked mixed nominal constructions (see Appendix B), there were eight cases in which there were no apparent disfluencies leading to the mixed nominal construction, as in (5) and (6). Thus, it is not the case that feminine-marked mixed nominal constructions are necessarily instantiations of ‘less planned’ code-switching that are triggered by a failure to retrieve the intended Spanish noun, as Valdés Kroff observes.

It is important to point out that feminine-marked mixed nominal constructions were produced by both second (n = 7) and fifth graders (n = 7) and included children who were enrolled in English immersion (n = 6) and two-way bilingual programs (n = 8). Therefore, these infrequent constructions cannot be associated with a particular schooling experience.

Another noteworthy difference about children’s feminine-marked mixed nominal constructions is that there were attested cases in which ‘a feminine default’ was seemingly employed (e.g., la other side ‘el otro lado’; una log ‘un tronco’). Though highly infrequent (n = 4), these examples in which the feminine-marked article does not match the gender of the translation equivalent seem to be distinctive of children’s oral production. In the data sets of feminine-marked mixed nominal constructions produced by Miami and Northern Belize code-switchers, there are no such examples. In adult data from both Miami (e.g., la cheerleader pesada ‘la animadora pesada’, las sheets ‘las sábanas’, etc.) and Northern Belize (e.g., la nurse ‘la enfermera’, la bell ‘la campana’, etc.), most feminine-marked mixed constructions are accounted for by the analogical feminine gender (Balam 2016a; Valdés Kroff 2016). Thus, the ‘default-like’ use or overextension of the feminine gender in children’s oral production suggests that they may still be in the process of acquiring the nuanced code-switching pattern that whereas English nouns with translation equivalents from both genders can be assigned the masculine gender, the same is not true for the feminine gender.^[9]