Abstract
Aims/Objectives/Purpose/Research Questions:
This study explored Spanish heritage speakers’ (HSs’) and second language learners’ (L2Ls’) mood production and selection. It explored whether proficiency, asymmetries between production and receptive knowledge, and lexical frequency influence HSs and L2Ls to the same degree in their command of mood.
Design/Methodology/Approach:
Thirty-four adult HSs and 26 L2Ls completed production and selection tasks as well as a proficiency measure in Spanish.
Data and Analysis:
All responses were submitted to a binomial logistic regression evaluating mood use as a function of expected mood, group, proficiency, verbal lexical frequency, and task. Individual analyses complement statistical findings.
Findings and Conclusions:
HSs and L2Ls used the indicative more consistently than the subjunctive. The two groups showed similar production tendencies, but L2Ls selected the expected moods more consistently than HSs. Proficiency modulated individual differences in production and selection. Lexical frequency of the subordinate verb did not modulate subjunctive production or selection.
Originality:
This paper is among the first to evaluate relationships between group-level, between-speaker, and within-speaker differences in both HSs and L2Ls side-by-side.
Significance/Implications:
This study demonstrates the importance of evaluating between-speaker and within-speaker effects in bilingual morphosyntax. Findings argue for a dissociation between productive and receptive command. It appears that L2Ls show more asymmetrical knowledge of mood contrasts than HSs when a metalinguistic task is involved.
Introduction
Comparisons of heritage speakers (HSs) and second language learners (L2Ls) have been frequent in bilingualism research because they provide insights into how certain constellations of factors, including but not limited to age of acquisition, instructional experiences, patterns of exposure and use, literacy experience, and metalinguistic knowledge, influence the acquisition of different areas of grammar. Such comparisons aim to identify if and under what circumstances HSs’ and L2Ls’ linguistic profiles are similar to one another. While many studies concentrate on group-level comparisons only (i.e., HSs’ versus L2Ls’ knowledge of a given linguistic property), it is also useful to consider whether variables that account for between-speaker differences such as proficiency, as well as for within-speaker variation such as productive-receptive asymmetries and lexical frequency, affect both groups of bilinguals equally.
On one hand, HSs acquire their heritage language (HL) from birth in situations where a social majority language is more represented in institutional and educational settings. Despite early exposure as native speakers, HSs often do not receive access to schooling in this language (e.g., Rothman, 2009). As a case in point, HSs of Spanish in the United States begin to experience a shift in dominance to English, the social majority language, at the beginning of the school period (Barnett et al., 2007; Castilla-Earls et al., 2019; Hiebert & Rojas, 2021). HSs predominantly use their HL with family and community members rather than in classroom settings, and their linguistic knowledge is often manifested in implicit knowledge and oral production rather than explicit abilities to recognize and use features of the language (Bowles, 2011a, 2011b). Since HSs have reduced input in the HL and often do not have access to this input through schooling, they may not be as familiar with less-common grammatical structures or written contexts of the language as other populations of native speakers or even L2Ls, who oftentimes receive the majority or all of their exposure in the classroom.
On the other hand, L2Ls develop competence in an additional language after having already mastered (or begun acquiring) their first. Although these speakers do not have the same degree of early, naturalistic exposure that HSs receive, they often approach the acquisition task with established literacy skills and frequently receive (varying degrees of) form-focused instruction. L2Ls may lack the social network of speakers of their second language that HSs often have, and their exposure to this language frequently can be restricted to classroom settings that may not have frequent opportunities for producing the language. These bilinguals often experience a more explicit learning process that might facilitate the acquisition of certain components of language that are not highly frequent in already limited input. HSs and L2Ls therefore differ along multiple dimensions given their distinct developmental profiles, although there is considerable variation within each group, and these terms refer to generalizations about these dimensions rather than impermeable categories (i.e., it is possible for L2Ls to have rich social networks or for HSs to have extensive formal instruction).
Comparisons of HSs and L2Ls, however frequent, often remain at the group level, which could potentially obfuscate additional layers of complexity within which these groups may differ from each other. In an analysis with child and adult HSs, Thane (2025) argues that studies on HL acquisition should explore group-level (i.e., HSs versus L2Ls), between-speaker (i.e., proficiency level), and within-speaker variables (i.e., differential performance in production versus comprehension and the lexical frequency of inflected verbs) to fully account for the gradient morphosyntactic systems that bilinguals often exhibit. The present study applies this approach to the study of heritage and second language acquisition of the Spanish subjunctive, as these areas are often not addressed in studies comparing both populations. The aim is to broaden our understanding of how between-speaker effects such as proficiency and within-speaker effects such as productive-receptive asymmetries and lexical frequency may affect adult HSs and L2Ls to different degrees. Therefore, the emphasis of the present study is not on testing which factors distinguish between HSs and L2Ls, but rather, which variables influence between-speaker and within-speaker variation in each of these groups and to what degree.
To this end, research has concentrated on two sources of within-speaker variation, that is, why bilinguals use a particular structure in some contexts but not others. One source of within-speaker variation is asymmetrical performance between production and receptive knowledge. HSs and L2Ls are similar in that they exhibit greater variability in their production of morphosyntactic structures than can be observed in their receptive knowledge (e.g., Herschensohn et al., 2005; Perez-Cortes et al., 2019; Prévost & White, 2000; Putnam & Sánchez, 2013; White et al., 2004). For bilinguals, the challenge lies in acquiring and subsequently mapping syntactic functions onto morphological forms, especially when the social majority language does so differently from the HL or L2 (e.g., Lardiere, 2008, 2009; Putnam & Sánchez, 2013). Perez-Cortes et al. (2019) argue that for HSs, the mapping process is more taxing under the online pressures of linguistic production, which begins at the conceptual level (Chater et al., 2016), than it is when processing and interpreting linguistic information that others produce. Therefore, following these researchers, bilinguals differentially access linguistic features in production compared to comprehension and interpretation. Research on adult L2Ls has advanced similar claims when comparing production to interpretation of inflection (e.g., Herschensohn et al., 2005; Prévost & White, 2000; White et al., 2004). An open question is if HSs and L2Ls show the same degree of asymmetry between productive and receptive knowledge, which is taken up in the present study.
Second, Putnam and Sánchez (2013) and Perez-Cortes et al. (2019) both argue that HSs may alternate between producing or interpreting a given structure in some contexts but not others, which may be due to the frequency of lexical items onto which morphology is assembled. Recent evidence has shown that lexical (lemmatic) frequency influences HSs’ production of subjunctive mood (Giancaspro, 2020), gender agreement (Hur et al., 2020), differential object marking (Hur, 2020), and preterit aspect with state verbs (Thane, 2024b). Beyond the realm of inflection, López Otero (2022) found that verb frequency affects HSs’ word order preferences, as more frequent items favored noncanonical verb–subject production and positive acceptability judgments. Research has also shown that the frequency of individual items also extends to L2Ls’ production tendencies of differential object marking (Hur, 2022) and preterit aspect (Thane, 2023b). However, like with asymmetries between productive and receptive knowledge, it is unclear whether this variable affects HSs more than, less than, or as much as it does L2Ls.
Both the asymmetry between productive and receptive knowledge and lexical frequency comprise within-speaker effects that represent viable accounts for why bilinguals produce or interpret inflectional morphology in some instances but not others. However, few to no studies have aimed to explore whether within-speaker variables are equally impactful on HSs and L2Ls. An untested possibility is that L2Ls experience greater asymmetries between productive and receptive tasks because, in classroom settings, especially at lower proficiency levels, they may have minimal opportunities for output. Moreover, L2Ls’ exposure could be limited to frequent lexical items found in textbooks and pedagogical materials. This learning context could thus lead to greater productive-receptive asymmetries or more sensitivity to lexical frequency effects than HSs.
The Spanish subjunctive, reviewed in the following section, is a useful area of the inflectional system with which to explore whether proficiency, productive-receptive asymmetries, and lexical frequency impact HSs and L2Ls in similar ways. There are multiple learnability challenges for acquiring the subjunctive, as discussed below. These learnability challenges have led to frequent studies on this structure, but of particular relevance here are findings from Montrul and Perpiñán (2011), who provide a notable counterexample to research discussed above that frequently demonstrates an advantage for HSs over L2Ls in the acquisition of morphosyntax. On written sentence conjunction and morphology recognition tasks, low- and intermediate-proficiency HSs showed stronger command of the subjunctive than proficiency-matched L2Ls, yet at the advanced level, L2Ls had an advantage over HSs. 1 However, it remains unclear if these findings also would influence the production of mood where HSs often have an advantage over L2Ls (Bowles, 2011a, 2011b). Moreover, an unexplored question is if these two groups are equally sensitive to within-speaker variation like productive-receptive asymmetries and lexical frequency in the acquisition of the contrast between the indicative and subjunctive moods.
The Spanish Subjunctive in Bilingual Development
The subjunctive mood is an ideal area of the Spanish grammatical system with which to explore the relationships between group-level (i.e., HSs versus L2Ls), between-speaker (i.e., proficiency), and within-speaker (i.e., productive-receptive asymmetries and lexical frequency) variables. It is one of three moods in Spanish, alongside the indicative and imperative (Seco, 1990), only used with between 7% and 10% of inflected verbs (Biber et al., 2006; Kanwit & Geeslin, 2018). This makes it far less frequent than the indicative. Fábregas (2014) argues that the subjunctive is a single set of inflections mapped onto multiple syntactic representations with differing degrees of semantic transparency. Nearly all instances of the subjunctive occur in the subordinate clause with a distinct subject from the main clause (i.e., the disjoint reference effect, Kempchinsky, 2009), such that its acquisition is contingent upon developing the ability to subordinate (Collentine, 1997). In addition, with most verbs, subjunctive inflections differ in a single vowel from those of the more-frequent indicative mood, so the morphophonological contrast between moods is not highly perceptually salient. Specifically, Spanish verbs with /ar/ infinitives have the thematic vowel /a/ in the indicative and /e/ in the subjunctive, while verbs with /er/ and /ir/ infinitives have the thematic vowel /e/ in the indicative and /a/ in the subjunctive. Person/number inflections are then added to the thematic vowel (e.g., for the Spanish infinitive hablar, ‘to talk’, the third person plural of the indicative is hablan and the third person plural of the subjunctive is hablen). Therefore, the subjunctive involves syntactic and morphophonological nuances and is infrequent, which could make its acquisition complex in contexts of bilingual development. In fact, monolingual children are still in the process of mastering the volitional subjunctive through ages four and five (Blake, 1983; Dracos et al., 2019), later than most areas of the Spanish inflectional system.
The volitional subjunctive is one such syntactic context that denotes desire, obligation, or influence following lexical items such as the verb querer (to want) in sentences such as quiero que vengan (literal translation: I want that they come). In such instances, Kempchinsky (2009) argues that these lexical items act as governors in the main clause that carry out lexical selection of the subjunctive in the subordinate clause, where it is purportedly obligatory. In contrast, English does not allow for subordinate clauses following many volitional verbs such as want, and instead requires infinitival control structures (e.g., I want them to come). Therefore, for HSs and L2Ls for whom English is the majority language, acquiring the volitional subjunctive requires perceiving and acquiring inflections that in many instances differ subtly from the indicative, compiling a mental list of lexical items that select this mood, and carrying out lexical selection for a subjunctive verb in the subordinate clause. This may be particularly challenging for lower-proficiency bilinguals who are more dominant in English: crosslinguistic influence could lead them to transfer infinitival control structures and reduced HL or second language input could lead them to use the more-frequent indicative inflections, particularly in production (Perez-Cortes et al., 2019).
Subjunctive Mood in Heritage Speakers
There is ample evidence demonstrating that HSs exhibit a wide range of possible outcomes in acquiring the subjunctive, pointing to roles for proficiency and patterns of exposure. Longitudinal studies demonstrate that young bilingual children show attrition of the subjunctive concurrent with increases in exposure to English (Anderson, 2001; Merino, 1983; Silva-Corvalán, 2014). Recent cross-sectional work has not found a clear effect for age in children during the school period (Dracos & Requena, 2022), but Thane (2025) did report a greater command of the subjunctive in adolescents relative to younger peers. Studies evaluating child and adult HSs have shown that proficiency (Dracos & Requena, 2022; Giancaspro, 2019a; Montrul, 2009; Montrul & Perpiñán, 2011; Perez-Cortes, 2016; Thane, 2025) and patterns of exposure and use (Dracos & Requena, 2022; Perez-Cortes, 2016; Solano-Escobar, 2025; Thane, 2025) affect HSs’ command of the volitional subjunctive.
Research on the subjunctive has also shown within-speaker variation. First, Perez-Cortes et al. (2019) reported that HSs with lower proficiency and less-frequent HL use were more likely to recognize the subjunctive on receptive tasks than to produce it. Thane (2025) also reported greater recognition of subjunctive on a receptive task than production in both adolescent and adult HSs. In addition to productive-receptive asymmetries, the lexical frequency of subordinate verbs modulates production and receptive knowledge of the subjunctive (Giancaspro, 2020). Moreover, López-Beltrán and Dussias (2024) and Perez-Cortes (2022) found that the frequency of lexical items in the main clause also influences whether HSs produce the subjunctive in the subordinate clause. Lexical frequency thus comprises a useful within-speaker variable that, like asymmetries between productive and receptive knowledge, can account for why HSs use a given structure sometimes but not others.
It should be noted that the review of research thus far has cited challenges in acquiring the subjunctive mood, but, as a reviewer points out, this does not fully address the acquisition of the indicative/subjunctive contrast. Crucially, research has positioned the indicative as a default (McCarthy, 2008), such that it spreads into contexts where the subjunctive is expected, but not vice versa. However, there is some past evidence from HSs documenting bidirectional overextension, whereby the subjunctive also extends into expectedly indicative contexts (Giancaspro, 2019b; Perez-Cortes, 2023). Therefore, when possible, it is prudent to consider the acquisition of the indicative/subjunctive contrast rather than of just the subjunctive.
Second Language Acquisition of Subjunctive
Research on L2Ls’ acquisition of the subjunctive has been largely consistent with that on HSs. It is imprudent to analyze the immense body of work on the second language acquisition of the Spanish subjunctive, including many studies oriented toward pedagogical interventions, yet across studies, the most deterministic predictor of subjunctive mood use has been morphosyntactic proficiency (e.g., Borgonovo et al., 2015; Iverson et al., 2008; Montrul & Perpiñán, 2011; Perez-Cortes, 2016). Considerably less attention has been paid to within-speaker variation in L2Ls’ acquisition of the subjunctive, both with respect to asymmetries between productive and receptive knowledge and to lexical frequency. Lustres (2018) reported a higher selection of subjunctive on a preference task than its production by L2Ls using methods highly similar to the ones reported here, but his data evaluated later-acquired subjunctive contexts. In contrast, in a study on the best pedagogical approaches for teaching the subjunctive, Gallego and Pozzi (2018) demonstrated that low-proficiency L2Ls’ production was relatively consistent with its recognition on highly metalinguistic tasks. Therefore, task asymmetries remain less clear for this population. Moreover, Gudmestad (2014) demonstrated that the frequency of individual matrix governors (both verbs and impersonal expressions) influenced L2Ls’ acquisition of the subjunctive. These results differ subtly from Giancaspro’s (2020), as they targeted the main clause verb or expression that selects subjunctive rather than the subordinate verb onto which mood inflections are assembled. Nonetheless, they invite the possibility of lexical effects in L2Ls’ mood knowledge as well as a comparison with HSs’ sensitivity to frequency.
Measurement of the Subjunctive and Metalinguistic Knowledge
The role of metalinguistic knowledge cannot be underestimated in comparing research on the HL and second language acquisition of the subjunctive. It is worthwhile to mention that the subjunctive is often the object of explicit instruction in high beginner- and intermediate-level Spanish language classes and textbooks. Some open-source Spanish for heritage learners textbooks also incorporate the Spanish subjunctive and provide instructions concerning its use and distribution (e.g., Foulis & Alex, n.d.). Since this grammatical structure is frequently a component of textbook-based instruction and is often targeted in intermediate composition courses, L2Ls in particular may have frequent experience in identifying and recognizing the subjunctive explicitly. Indeed, Correa (2011b) tested L2Ls’ metalinguistic knowledge through the identification of linguistic terminology and grammaticality judgments, and reported that higher metalinguistic awareness correlated with higher use of the subjunctive in production and on receptive selection tasks. Importantly, the same correlation between metalinguistic performance and subjunctive production and selection did not emerge for HSs, despite this group showing overall stronger subjunctive knowledge than L2Ls (Correa, 2011a). Following Correa (2011b), in the present study, a binary selection task is used as the receptive measure in order to reduce the metalinguistic burden of measuring sentences and their grammaticality explicitly, although it should be recognized that this task still remains metalinguistic in nature. There is an undeniable role for metalinguistic knowledge in the recognition of the subjunctive here, since this structure receives considerable attention in language coursework, which could be a limitation of this study.
Summary
To summarize, the acquisition of the subjunctive mood has attracted researchers’ attention because this structure is infrequent in the input, syntactically complex, and not highly salient. Although proficiency effects are well-documented in studies comparing HSs with L2Ls, it is unclear whether one group (HSs or L2Ls) is more influenced by between-speaker differences in proficiency than the other. In the same vein, the lexical frequency of the subordinate verb appears to modulate adult HSs’ knowledge of the subjunctive, a within-speaker variable that has not yet been explored in L2Ls’ acquisition of this structure. Finally, it remains unclear whether L2Ls show greater differences between productive and receptive knowledge than HSs. For these reasons, it is possible that these two groups differ in the degree of between-speaker (i.e., proficiency) and within-speaker (i.e., asymmetries between productive and receptive knowledge as well as lexical frequency) variation that influences their command of the indicative/subjunctive contrast. Importantly, metalinguistic knowledge is likely an intervening factor, as it has been demonstrated to affect L2Ls’, but not HSs’, subjunctive use.
The Study
Based upon the research reviewed, four research questions were proposed:
1. Do HSs and L2Ls exhibit similar rates of production and selection of indicative and volitional subjunctive?
Since previous research has reported an advantage in HSs’ production of morphosyntactic structures over L2Ls’ (Bowles, 2011a, 2011b), it was anticipated that these bilinguals would distinguish between the indicative and subjunctive more consistently than proficiency-matched L2Ls in oral production. However, given past findings using receptive tasks concerning subjunctive mood knowledge (Montrul & Perpiñán, 2011), it was predicted that L2Ls would select more subjunctive than HSs at higher proficiency levels, but HSs would do so more than L2Ls at lower proficiency levels. Such a finding would demonstrate that group-level differences (i.e., comparisons of HSs versus L2Ls) interact with within-speaker variables (i.e., productive versus receptive knowledge) at different proficiency levels.
2. Does proficiency modulate the ability to distinguish between indicative and subjunctive?
It was anticipated that HSs and L2Ls with higher proficiency levels would produce and select more subjunctive mood, in line with previous findings (e.g., Giancaspro, 2019a; Iverson et al., 2008; Montrul, 2009). As stated in hypothesis #1, it was anticipated that advanced proficiency L2Ls would select more subjunctive than advanced HSs, but among less-proficient bilinguals, HSs would recognize the subjunctive more than L2Ls, given the metalinguistic nature of the task. Indicative use was hypothesized to reach ceiling.
3. Do HSs and L2Ls distinguish between the subjunctive and indicative moods in a more categorical fashion than they produce them?
In line with existing research in heritage and second language acquisition (e.g., Herschensohn et al., 2005; Prévost & White, 2000; Putnam & Sánchez, 2013; White et al., 2004), it was anticipated that both groups would select the subjunctive on a receptive preference task more than they would produce it. Previous research has not yet explored whether HSs or L2Ls would show differing degrees of asymmetrical performance on tasks. However, L2Ls may have more disparities between input and output opportunities in classroom settings and, potentially, a smaller community with which to use Spanish. Moreover, they may deploy more metalinguistic knowledge in selection, given their classroom experience. Therefore, this group may exhibit greater asymmetries between productive and receptive domains than HSs.
4. Does the lexical frequency of subordinate verbs influence HSs’ and L2Ls’ production and selection of subjunctive mood?
On the basis of limited previous research (e.g., Giancaspro, 2020 for HSs and Gudmestad, 2014 for L2Ls but with matrix verbs), it was predicted that both HSs and L2Ls would show frequency effects in their production and selection of the volitional subjunctive. Specifically, it was anticipated that subordinate verbs with higher lemmatic frequencies would result in more frequent production and selection of the subjunctive, given their greater presence in speakers’ input. Lexical effects were expected to be greater for L2Ls, given their lower exposure to Spanish, which could limit their access to less-frequent lexical items.
Participants
Thirty-four adult HSs and 26 adult L2Ls participated in the present study. Participants in both groups predominantly lived and worked in the region surrounding a large public research university in New Jersey. The participants had not reported attending dual language bilingual education programs, but many were enrolled in undergraduate Spanish courses at this university. To capture the full range of the bilingual continuum, proficiency was measured using the 50-question Diploma de Español como Lengua Extranjera (DELE) assessment described at greater length below. Participants’ DELE scores were incorporated into statistical modeling as a continuous variable; however, to be consistent with previous studies (e.g., Iverson et al., 2008; Montrul & Perpiñán, 2011; Montrul & Slabakova, 2003), for descriptive analysis, participants were also subdivided into proficiency categories as summarized in Table 1. In line with these studies, participants with a score below 30 were classified as beginner, between 30 and 39 as intermediate, and above 40 as advanced.
Number of Participants in Each Proficiency Category by Group.
The average age and proficiency level for each group are summarized in Table 2. There were no statistically significant differences at the p < .05 level between groups in proficiency (t = 1.76, p = .084) or age (t = –0.26, p = .793). However, on a 30-point scale evaluating use of Spanish in six contexts, HSs reported using Spanish on average twice as often as L2Ls with similar standard deviations. 2 The difference in use among participant groups was significant (t = 5.69, p < .001). Importantly, therefore, these groups were matched for proficiency level but not patterns of use, which underscores the likelihood that L2Ls often do not have the same social network with which to use Spanish when compared to HSs. It should be recognized that there are a number of other uncontrolled variables such as metalinguistic knowledge that also could account for differences between groups. As a result, proficiency was maintained as the variable for analyzing between-speaker differences because it serves as a more objective measure of overall linguistic ability that is held constant across participant groups with different patterns of use, exposure, and metalinguistic skills. Consequently, it is adopted here as the heuristic for overall command of Spanish in addressing between-speaker (individual) differences in rates of subjunctive production and selection.
Participant Characteristics by Group.
The HSs represented 13 dialects of Spanish, which maximally reflects the diversity of Spanish in the United States. Faulkner (2021) reported categorical use of the volitional subjunctive across multiple dialects beyond the United States, and multiple authors have reported ceiling-level production of this structure by Spanish-dominant bilinguals in the United States using multiple methodologies (Giancaspro, 2020; Martillo Viner, 2016; Thane, 2025). Furthermore, in Thane (2025), a group of bilingual adults who grew up in Spanish-speaking countries completed the same tasks as in the present study and produced and selected the subjunctive categorically. These findings argue against the likelihood of dialectal differences in HSs’ input from Spanish-dominant participants, although subjunctive omission is feasible in input from other HSs or L2Ls with whom these participants communicate.
Methods
Participants signed a consent form for participation in compliance with Institutional Review Board procedures (protocol: Pro20211001902) and then completed a brief questionnaire (see Thane, 2025), followed by three asynchronous tasks. First, participants carried out the aforementioned DELE proficiency measure adapted by the Modern Language Association, which has been used in extensive previous research (e.g., Duffield & White, 1999; Giancaspro & Sánchez, 2021; Iverson et al., 2008; Montrul & Perpiñán, 2011). This written proficiency measure involved 50 multiple-choice questions, 30 for vocabulary and 20 for morphosyntax. The number of expected answers on the DELE served as each participant’s proficiency score.
Participants also completed two tasks tapping into the indicative and volitional subjunctive in subordinate clauses following the verbs creer (eliciting indicative) and querer (eliciting subjunctive). Both tasks involved the same eight disyllabic –ar verbs with regular morphology (i.e., no changes to verb stems in formation of the subjunctive, such as lleve ‘to carry/to take’ from the verb llevar) to test the subjunctive. 3 These verbs were selected along a continuum of lexical frequency in the Davies Corpus del español. This lemmatized corpus contains 2,000,000,000 words and has been shown to correlate with U.S. Spanish HSs’ self-ratings of lexical frequency (Thane, 2023c), which argues for its vitality as an estimation of the distribution of forms in participants’ input. In addition, it is the same corpus used by Giancaspro (2020), enabling consistency in how frequency is operationalized across studies addressing the acquisition of the subjunctive. Verbs are listed from most to least frequent, according to lemmatic frequencies in the Davies corpus, in Table 3. The indicative verbs’ frequencies were not controlled, as this was not a component of the research questions, but they were also regular, disyllabic –ar verbs. Both tasks were presented in a common communicative context about a mother’s desires for her two twin daughters to take care of their younger brother while at sleepaway camp. The first task tapped participants’ production of the subjunctive and the second their receptive knowledge. The receptive task was referenced as selection as it required choosing between two binary alternatives.
Lemmatic Frequencies of Verbs Selected for Tasks.
The production task contained 47 items, of which eight concentrated on the volitional subjunctive. The task included five indicative items following the verb creer (to believe), which selects the indicative in affirmative sentences similarly to how querer selects the subjunctive, as well as 34 fillers. It followed a sentence completion format, whereby participants read a preamble and then needed to complete a sentence prompt. This sentence prompt contained the verb querer as well as the complementizer que and was missing the target verb whose infinitive appeared in parentheses, as in Figure 1. 4 Indicative items followed a highly similar structure, but the prompts used the verb creer (to believe) that elicits the indicative in place of querer, which elicits the subjunctive. Participants were instructed to record their response orally with any form of the word in parentheses, as well as any additional words that they felt were missing.
Sample of item on production task.
The written selection task contained 55 items, eight addressing the subjunctive following querer, six following the indicative creer, and 41 serving as distractors. The subjunctive items required participants to read prompts and to choose which of two sentences they felt sounded best to answer the question ¿Qué quiere la mamá? (What does the mom want?). The indicative items were similar but followed the question ¿qué cree la mamá? (What does the mom believe?). The two sentences between which participants needed to choose were identical except in the use of either the subjunctive or the indicative in the subordinate clause, as in Figure 2.
Sample of item on selection task.
It is prudent to recognize that the number of items in each task was not counterbalanced between the subjunctive and the indicative. Furthermore, the overall total number of items in the tasks is nonstandard. The stimuli were so designed because children completed an abbreviated set of stimuli, concentrating only on mood (see Thane, 2023a, 2024a, 2025). More attention was paid to the subjunctive in those projects than to the indicative, resulting in a smaller number of distractors. Future research may wish to more appropriately counterbalance the indicative and subjunctive items in the tasks; a transcript of these tasks is available in Thane (2023a) and on the public repository shared below.
Results
All data analysis took place in RStudio (R Core Team, 2017) using the emmeans (Lenth, 2021), lme4 (Bates et al., 2015), lmerTest (Kuznetsova et al., 2017), and tidyverse (Wickham et al., 2019) packages. The patchwork (Pedersen, 2024) package was used for data visualization and the here (Müller, 2020) package enabled reproducibility through relative paths. All anonymized data and code were stored on a GitHub repository (https://github.com/pthane/DLI-Morphosyntax-2023) in the IJB HS vs. L2L Subjunctive folder of the Manuscripts subdirectory.
All responses were scored in a binary fashion, where production or selection of the expected mood was awarded a score of 1. All other responses, including infinitives, the unexpected mood (indicative following querer or subjunctive following creer), or past tense forms, were awarded a score of 0. Responses where participants’ recording was not saved or in which participants produced forms that did not include the prompt were left unscored. Mood use score was the dependent variable for all analyses. A summary of the average and distribution of indicative and subjunctive production and selection rates by group and task is shown in Figure 3 below. Moreover, Figure 4 provides a visualization of the role of DELE proficiency on between-speaker differences. Broadly speaking, regarding the subjunctive, these figures suggest that HSs and L2Ls performed nearly identically at the group level in production, with an advantage for L2Ls in selection. This advantage was subtle, although there was less individual variation in the L2L group. Both groups selected the subjunctive more than they produced it, but this asymmetry appeared greater for L2Ls. Moreover, while there was higher use of the indicative in the expected contexts than of the subjunctive, it should be noted that both groups of bilinguals produced and selected the subjunctive in expectedly indicative contexts with the verb creer, although this tendency did not appear to be modulated by proficiency. This finding implies that variability in participants’ mood contrasts is not unidirectional but rather consists of two-way substitutions: indicative for subjunctive but also, to a lesser degree, subjunctive for indicative.
Statistical summary of subjunctive production and selection by mood, group, and task.
Production and selection rates by group, mood, and task as a function of DELE proficiency.
Interpretation of Proficiency Effects
To establish consistency with previous studies and to locate the effect of proficiency on subjunctive production and selection more easily within a specific range, Figures 5 and 6 offer a statistical summary of indicative and subjunctive production and selection, respectively, with three proficiency bands as spelled out in Table 1. These findings provide greater descriptive detail about the nature of proficiency effects and their impact on subjunctive production and selection. Figure 5 points to overextension of the subjunctive to indicative contexts in both groups and across both tasks, with the exception of the advanced proficiency L2Ls, who did not select the indicative. The statistical summary in Figure 5 suggests that this variability is caused primarily by individual participants rather than representing a general trend across HSs and L2Ls.
Average and distribution of indicative responses by group, proficiency category, and task.
Average and distribution of subjunctive responses by group, proficiency category, and task.
Regarding the subjunctive, the 13 bilinguals (HSs and L2Ls combined) with advanced proficiency produced and selected the subjunctive in all but three sentences. However, at the beginner proficiency level, L2Ls produced and selected more subjunctive sentences than HSs. At intermediate proficiency, HSs produced more subjunctive than L2Ls, but L2Ls selected it more than HSs. While the present study explored continuous data, these descriptive findings provide a necessary nuance that at the highest-proficiency levels, all HSs and L2Ls produced and selected the subjunctive at ceiling, yet L2Ls showed greater sensitivity to the mood contrasts on these relatively metalinguistic tasks at lower proficiency levels.
Figure 7 represents HSs’ and L2Ls’ production and selection of subjunctive by individual verb, arranged from left to right in order of greatest to lowest lemmatic frequencies in the Davies corpus. Descriptive findings do not reveal any clear frequency pattern in production nor in the selection of subjunctive for either group on either task.
Production and selection of subjunctive by group and verb, arranged from most to least frequent lemmatic frequency.
Multivariate Analysis
A generalized linear mixed effects binomial logistic regression model was prepared with mood use score as the dependent variable, participant and item as random effects, and five predictors: expected mood (indicative as reference level, subjunctive), group (HSs as reference level, L2Ls), proficiency (number of anticipated responses on the DELE measure), task (production as reference level, selection), and lexical frequency (lemmatic frequency of each verb in the Davies corpus) as predictors. Two-way interactions between the group and all other variables were included to address the research questions and their predictions. In line with Ortega’s (2020) recommendations, proficiency and lexical frequency were included as continuous predictors. These two variables were standardized prior to analysis to align them on a consistent scale. Due to the large number of predictors, the final model was run using the bobyqa optimizer in RStudio, which optimizes results through quadratic approximation. This resulted in a model achieving convergence whose results differed extremely minimally from the original one without the optimizer; readers are encouraged to observe this similarity using the publicly available and reproducible code cited above. The results of this model are given in Table 4. Results demonstrate a main effect for the subjunctive mood (used significantly less frequently than the indicative), proficiency (higher proficiency leads to stronger command of the indicative/subjunctive contrast), the selection task (more selection than production of the expected moods), and the L2L group by selection task interaction.
Results of Binomial Logistic Regression.
Italicized values indicate statistical significance at the .05 level.
To appropriately address the group by task interaction, pairwise comparisons were conducted. The contrast between groups was not significant at the p <.05 level for the production task (β = –0.11, SE = 0.34, p = .748); however, in consistency with Figure 3, L2Ls selected significantly more subjunctive than HSs (β = –1.21, SE = 0.14, p ⩽ .001). Therefore, participants were more likely to use the indicative than the subjunctive; both groups selected the expected moods on the preference task more than they produced them, target mood use increased with proficiency, and L2Ls showed a greater asymmetry between production and selection than HSs.
Individual Performance
Finally, to better determine the role of individual performance, Figure 8 represents the number of indicative and subjunctive forms that each participant produced and selected. 43/60 (71.6%) participants (24 L2Ls and 19 HSs) used the indicative categorically and 22/60 (36.7%) participants (13 HSs and 9 L2Ls) used the subjunctive categorically. In line with the greater number of participants who used the indicative categorically, there were fewer participants with low selection rates of indicative than subjunctive, implying less individual variation in the acquisition and mastery of this mood. Only one L2L categorically omitted the subjunctive; there was no categorical omission of the indicative by any participant and no categorical omission of either mood by any individual HS. These findings also reinforced that participants seldom produced the subjunctive more than they selected it. To this end, there were two L2Ls who selected the subjunctive in 7/8 instances without producing it, yet no participants displayed the opposite trend of categorical production of subjunctive with no selection. While there are a greater number of L2Ls who show this pattern, it is important to note that two Spanish HSs show similar trends, producing one subjunctive sentence and selecting seven. Therefore, it appears that many HSs’ and L2Ls’ mood systems are characterized by variation between the subjunctive and indicative in the expected subjunctive context, and some bilinguals also experience extension of the subjunctive into indicative contexts.
Individual rates of production and selection of the subjunctive mood.
Discussion
The present study aimed to address whether HSs and L2Ls show similar degrees of between-speaker variation due to proficiency and/or the same extent of within-speaker variation due to productive/receptive asymmetries and lexical frequency. To do so, this study exploited the nuances of the Spanish indicative/subjunctive contrast through production and selection tasks. This project sought to contribute additional evidence to frameworks that emphasize between-speaker variability, asymmetries between productive and receptive knowledge, and lexical frequency effects (e.g., Perez-Cortes et al., 2019; Putnam & Sánchez, 2013). This approach enables us to form a more intricate understanding of bilinguals’ morphosyntactic systems, and whether HSs and L2Ls, who differ from one another along many dimensions, are equally influenced by the same between-speaker and within-speaker variables.
Before turning to a discussion of the research questions, it is useful to point out the unexpected bidirectional overextension of mood, where bilinguals overextended the subjunctive into indicative contexts. Following Figure 5, even high-proficiency HSs demonstrated this tendency, despite showing minimal to no extension of the indicative into subjunctive contexts. HSs and L2Ls were overall more likely to use the expected mood in contexts with the verb creer, which elicits the indicative, than with querer, which elicits the subjunctive. While these trends are not addressed through a research question, as there was not a counterbalanced number of indicative stimuli to fully explore this topic, it is important to note that they constitute additional evidence of bidirectional substitution of indicative for subjunctive and subjunctive for indicative (see also Giancaspro, 2019b; Perez-Cortes, 2023).
The first research question addressed group-level differences between HSs and L2Ls in the production and selection of the indicative and volitional subjunctive. In line with research that has demonstrated an overall advantage for HSs in production of morphosyntactic structures, it was anticipated that HSs would produce more subjunctive than L2Ls, but L2Ls would select this structure on a more metalinguistic task than HSs, while indicative knowledge would be at ceiling. Overall, HSs and L2Ls produced nearly identical rates of both moods, contra the hypothesis, but L2Ls selected the indicative and subjunctive in the expected contexts significantly more than HSs, in line with the hypothesis. It should be acknowledged that metalinguistic knowledge quite plausibly plays a considerable role in this result because L2Ls may have recognized the subjunctive handily on the selection task as a result of their classroom experiences. This could be attributable to previous research demonstrating more explicit knowledge in L2Ls over HSs (Bowles, 2011a, 2011b). Indeed, research has tied the acquisition of subjunctive to metalinguistic knowledge (Correa, 2011b), while HS’ performance does not correlate with this variable (Correa, 2011a). Consequently, it would be highly beneficial to build upon the results of the present study with more implicit tasks tapping processing, such as self-paced reading or eye-tracking measures.
The second research question investigated between-speaker differences in the production and selection of the subjunctive as a function of proficiency. As anticipated, there was a positive correlation between proficiency and the distinction between moods in production and selection: for every standard deviation increase in DELE proficiency (equating to 9/50 points), participants were 3.2 times more likely to distinguish between the indicative and subjunctive in the expected way. With regards to the subjunctive, the descriptive analysis in Figure 6 provides a nuanced account: advanced proficiency bilinguals in both groups were categorical in using the subjunctive in the expected contexts; however, at lower proficiency levels, L2Ls produced and selected more subjunctive than HSs. These findings exemplify the interconnectedness between group-level variables (i.e., comparisons of HSs and L2Ls) and between-speaker proficiency effects, as higher production or selection rates for L2Ls than HSs depended on their proficiency level.
The third research question evaluated asymmetries between production and receptive knowledge of the indicative-subjunctive contrast in HSs and L2Ls tested through a sentence completion production task and a receptive selection task. In line with past research in heritage and second language acquisition (e.g., Herschensohn et al., 2005; Perez-Cortes et al., 2019; Prévost & White, 2000; Putnam & Sánchez, 2013; White et al., 2004), it was anticipated that HSs and L2Ls would distinguish between the indicative and subjunctive more consistently on the selection task than they would do so in production. The findings here supported this hypothesis, given the overall effect favoring the selection task as well as the L2L group by selection task interaction. While both groups produced nearly identical rates of indicative and subjunctive, the L2Ls selected more instances of indicative and subjunctive morphology in the expected contexts than HSs, indicating an overall greater asymmetry between production and receptive knowledge for this group.
These findings must be interpreted with caution, because many Spanish language programs incorporate preference tasks as an assessment of speakers’ knowledge of the subjunctive. Therefore, it is likely that the L2Ls had familiarity with this type of task while HSs did not, which could considerably inflate the former group’s ability to identify each mood. Future work integrating online psycholinguistic methods such as those employed by Fernández Cuenca and colleagues (e.g., Fernández Cuenca & Jegerski, 2023; Giancaspro & Fernández Cuenca, 2025) and Cameron (2017) could be a better fit, as these methods are more implicit and less open to metalinguistic knowledge. Consequently, it is crucial to point out that the higher rates of mood distinction in selection here need not be interpreted as an advantage for L2Ls over HSs in natural conversation. Without this point of clarification, readers may inadvertently interpret these results in a way that could frame HSs’ linguistic abilities as deficient, which is expressly contrary to the aims of this article or area of research.
It is important to consider the descriptive data in Figure 6 and their implications for interconnectedness between the research questions presented thus far. These data support a relationship between groups, proficiency categories and task asymmetries: at the beginner level, L2Ls produced and selected more subjunctive than HSs, at the intermediate level, HSs produced more subjunctive than L2Ls but L2Ls selected it more than HSs, and at the advanced level, both groups produced and selected the subjunctive nearly categorically (but HSs were less categorical with the indicative than L2Ls in production). These descriptive findings are likely related to L2Ls’ metalinguistic awareness on the selection task, as these bilinguals frequently receive disproportionately high amounts of instruction on the subjunctive in class, while low-proficiency HSs without frequent exposure to their HL do not. Therefore, metalinguistic awareness may be a bigger mediator in performance on the receptive task at lower proficiency levels, yet at advanced proficiency levels, participants’ robust knowledge of mood overcame any metalinguistic effect. These findings underscore the importance of exploring the interplay between group-level (i.e., HSS versus L2Ls), between-speaker (i.e., proficiency range), and within-speaker (i.e., task asymmetries) variables.
This study joins a number of recent projects in HL acquisition demonstrating stronger receptive knowledge than what can be observed in production (e.g., Giancaspro & Sánchez, 2021; Perez-Cortes et al., 2019; Sherkina-Lieber, 2015; Thane, 2024a, 2025), which aligns with earlier work in L2 acquisition revealing a similar trend (e.g., Herschensohn et al., 2005; Prévost & White, 2000; White et al., 2004). Recent theorizing in HL acquisition has proposed that with infrequent HL activation, HSs’ form-function mappings can remain persistently optional (Lohndal & Putnam, 2021, 2024), which may also be true for L2Ls. This claim is supported by the individual performance data presented here for both groups of speakers. Specifically, HSs and L2Ls (with the exception of one participant) appear to have the syntax of subjunctive mood necessary to produce or select it multiple times, even if they do not consistently map this representation onto morphology for production or comprehension. Such an understanding of bilinguals’ grammars duly moves away from a binary notion of “having” or “not having” a structure in their repertoire and toward more intricate questions about what additional factors lead to within-speaker differences.
The fourth and final research question addressed one of these possible factors, the role of lexical frequency in the production and selection of subjunctive mood. Contra the hypotheses and previous research, no frequency effect emerged in the descriptive or inferential statistics (Giancaspro, 2020). A possible source of the differences in the results of this study and Giancaspro’s (2020) is that frequency was a continuous predictor here, but in the latter study, it was a categorical variable that involved dividing verbs into high- and low-frequency categories. It is possible that a greater number of lexical items than the eight verbs included here are necessary to observe frequency effects with continuous data. The absence of a frequency effect may not be altogether surprising when considering recent findings that have shown that morphological regularity could be more impactful on production and processing of inflectional morphology than lexical frequency in bilinguals (e.g., Fernández Cuenca & Jegerski, 2023; Thane et al., 2025). The present study used regular –ar verbs rather than irregular ones, which could have underestimated bilinguals’ mood production and selection tendencies. Therefore, future research should control for morphological regularity and test both types of verbs.
It is important to discuss some of the limitations of the present study not taken up in the discussion thus far beyond the most considerable one, the potential confound between metalinguistic awareness and receptive knowledge in the selection task, which has been discussed throughout this manuscript. An additional, related limitation is that HSs and L2Ls may possess wide individual variation in metalinguistic knowledge, which, at least in the case of the latter group, correlates with use of the subjunctive (Correa, 2011b). Therefore, future work should address individual levels of metalinguistic knowledge within groups: some HSs do receive bilingual schooling and may have developed metalinguistic awareness (although none of the participants reported attending immersion programs), while some L2Ls are purely naturalistic learners of Spanish with no previous instruction. This individual variation in the degree of metalinguistic awareness within each participant group can become buried in group-level analyses.
An additional limitation is that selection tasks in general reduce participants’ possible tendencies to the binary (i.e., indicative versus subjunctive), while production tasks, even controlled ones that follow a sentence completion format, are less restrictive. Alternative receptive tasks such as grammaticality judgments provide slightly greater gradience in possible responses. This gradience is helpful insomuch that it is less restrictive than the binary; however, the very act of rating sentences according to degrees of grammaticality is unnatural and arguably skews even more metalinguistic than selection tasks where one option purportedly sounds more natural than the other. Perhaps it is for this reason that Correa (2011a, 2011b) used grammaticality judgments to identify metalinguistic awareness skills but selection tasks highly similar in design to the one used here to measure subjunctive knowledge. Ideally, research can adopt online methodologies such as self-paced reading or eye-tracking to more definitively observe participants’ in-the-moment behaviors as they process linguistic information, as stated previously.
In the spirit of the insightful comments advanced by an anonymous reviewer, it is prudent to consider where we stand as a field with regards to comparisons between HSs and L2Ls more generally. As stated from the onset, the comparison of these populations is far from novel. Thane (2023b) states that many theories of HL acquisition are derived from second language research, and these theories have evolved to innovate and explore within-speaker variation in heritage populations. It is now possible to test the degree to which these same within-speaker variables can be applied to different groups of bilinguals, and whether they are equally predictive for these populations, as in the present study. Nevertheless, HSs and L2Ls themselves differ along many different dimensions. Therefore, while it is common practice to approach these comparisons with the presupposition that HSs and L2Ls exist as ecologically distinct groups that differ in a constellation of factors, it might next be fruitful to find innovative approaches that aim to isolate which individual factors can truly differentiate between them in the acquisition of morphosyntax. Such an approach may even obviate the need for distinguishing between HSs and L2Ls as different categories of bilinguals, and could enable us to reconceptualize these groups as benchmarks along the bilingual continuum. While this study is not positioned to address some of these larger and broader questions in the field of bilingualism, despite its multiple limitations, it aims to make a humble contribution to our understanding of the relationship between group-level and within-speaker variables with HSs and L2Ls side-by-side.
Footnotes
Acknowledgements
The author would like to thank the anonymous reviewer and editorial team for their time dedicated to reviewing this manuscript.
Consent to participate
Consent for participation was obtained through written means prior to enrollment in the study. Publication is permitted per Institutional Review Board approval (proposal Pro20211001902).
Funding
The author received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data availability and AI statement
All data and code and available for review on an online GitHub repository: 
. Please note that ChatGPT was used to check and polish coding available on this repository. All writing and proofreading was carried out by the author, except for a final review of proofs by ChatGPT, which resolved one typo.