Progressive aspect across temporalities: variation between synthetic and analytic forms in L1 and L2 Spanish

2.1 Tense and aspect morphology

Across the languages of the world, temporality can be encoded by different means, such as tense, or the location of an event on a timeline (Binnick 2009; Comrie 1985). Tense has a deictic center (the present), relative to which events are placed. The present tense, for instance, typically refers to states and processes which hold at reference time, even if they began prior to it or extend into the future. There is a difference in tense in the examples in (2), given that the event referred to is situated in present (2a) or past time (2b).

In the examples in (2), no distinction is made regarding the internal temporal structure of the event, commonly referred to as grammatical aspect (Comrie 1976; Smith 1997). Aspectual distinctions specify the boundedness of an event independently of its location on the timeline. The main opposition is that between perfective aspect, used for bounded events, and imperfective aspect, used for unbounded (i.e., open or repetitive) events. Two subcategories exist within imperfective aspect; namely, habitual aspect, which refers to events occurring repeatedly over time, and continuous aspect. The latter concerns an event in progress at a given reference time and is further divided into progressive aspect (susceptible to change) and nonprogressive aspect (not susceptible to change). Further updates to traditional aspectual classes (Comrie 1976) in the light of grammaticalization theory (Bybee et al. 1994) subsume continuous and non-progressive under progressive, leaving the latter as the main opposition to the habitual domain.

In addition to grammatical aspect, there are also nongrammatical aspectual distinctions based on the semantics of verbs and their arguments. In this sense, four aspectual classes are commonly used to differentiate predicates (lexical aspect or Aktionsart), which can be defined based on three semantic features (Smith 1997; Vendler 1967; see Table 1). Punctuality is a feature of events that are instant in nature and therefore lack duration. Telicity characterizes events with a specific endpoint, whereas dynamicity typifies events marked by change and a lack of stativity. Stative predicates differ from other classes because they are not punctual, have no specific endpoint, and continue over time. Activities differ from states in their dynamicity. Accomplishments and achievements differ from each other because the latter class refers to events that are completed instantly. Importantly, aspectual classes are described as being compositional (Verkuyl 1993), meaning that the immediate linguistic context impacts their interpretation, in a process referred to as aspectual coercion (De Swart 1998). A predicate containing an inflected form of escribir ‘to write’ by itself, for instance, is an activity but escribir una carta ‘write a letter’ is an accomplishment because it contains an inherent endpoint.

2.2 The acquisition of tense and aspect in Spanish

Aspectual classes and their semantic features are at the core of the AH, one of the most influential and frequently tested hypotheses in research on tense and aspect acquisition (Bardovi-Harlig and Comajoan-Colomé 2020). The AH originates from a study on the acquisition of Spanish as an L2 by English-speaking children. Andersen (1991) observed that such learners tended to mark pastness via the preterit before the imperfect, and even when the imperfect emerged, it was primarily associated with certain aspectual classes (e.g., states), unveiling a novel developmental sequence. The main tenet of the AH is hence that lexical aspectual classes have an influence on the acquisition of tense-aspect morphology, particularly at earlier stages (Andersen 2002; Andersen and Shirai 1994, 1996). The hypothesis suggests prototypical associations between aspectual classes and morphological markers (e.g., telic predicates and the preterit) to which learners are sensitive, but associations gradually weaken as learners gain experience.

Whereas research testing the AH has shown disagreement regarding the order and robustness of stages in the developmental sequence, most scholars agree that lexical aspect plays a role – limited or broad – in the acquisition of tense-aspect morphology. Of particular interest for the current study are the third stage of a commonly cited developmental sequence proposed by Andersen and Shirai (1996): progressive markers (in languages that utilize such devices) first emerge with activities, followed by accomplishments and achievements (i.e., dynamic predicates); and the fourth stage, proposing that such markers “are not incorrectly overextended to statives” (p. 533) because of their semantic features (i.e., a lack of punctuality, telicity, and especially dynamicity).

Spanish marks both tense and aspect (Montrul and Salaberry 2003), which is demonstrated by its extensive network of morphological markers. With respect to related learnability difficulties, most research in Spanish has centered on the range of past-time forms (Howard and Leclercq 2017), primarily the preterit and imperfect opposition (Domínguez et al. 2013, 2017; González and Quintana Hernández 2018; Montrul and Slabakova 2003; Salaberry 2011). English-Spanish bilinguals tend to overextend the preterit to contexts where the imperfect is expected, as English lacks an overt imperfective marker, whereas Spanish past verb forms can select both a perfective (trabajé ‘I worked’) and an imperfective marker (trabajaba ‘I was working/I used to work’).

A similar mismatch in form-meaning mappings exists in progressive and habitual domains. The English simple present tense prescriptively does not allow a progressive reading (Cowper 1998; Huddleston and Pullum 2002); both Spanish forms (i.e., synthetic and analytic), conversely, allow for progressive and habitual readings. However, as compared to perfective aspect, research on the acquisition of progressive and habitual aspect lags behind. An exception is a recent line of inquiry exploring variability between present indicative and present progressive forms in these aspectual domains from a variationist perspective (Fafulas 2010, 2015; Geeslin and Fafulas 2012, 2022; Kanwit et al. 2015), which is addressed in Section 2.4, following a review of prior research on the alternation between the forms among speakers of Spanish as an L1.

2.3 Progressive aspect in L1 Spanish variationist research

The imperfective aspectual domain has been a productive research topic in Spanish, with numerous studies focusing on the present progressive and present indicative opposition. It is generally agreed that the forms are not always semantically interchangeable, since the present indicative has a wider range of meanings (Alarcos Llorach 1994; Yllera Fernández 1999). However, as is standard, variants are studied in contexts of functional equivalence, i.e., when fulfilling the same grammatical function (Sankoff and Thibault 1981; Schwenter and Torres Cacoullos 2010). Namely, the forms are compared when referring to events in progress, and, increasingly, to habitual events. With respect to linguistic predictors, the predicate’s lexical aspect and the semantic value of co-occurring adverbs have been shown to be key factors conditioning variation between synthetic and analytic forms in the present tense (Cortés-Torres 2005; Fafulas 2012; Márquez Martínez 2009; Mayberry 2011).

Contact with English is purported to lead to an increase in the use of the present progressive in bilingual contexts (Klein 1980), resulting from the transfer of pragmatic features in the direction of English, as in the use of the present progressive with futurate readings among English-Spanish bilinguals in Los Angeles (Sánchez-Muñoz 2004). However, other studies have found no statistically significant difference between bilingual and monolingual groups (Cortés-Torres 2005; Fafulas 2012). Independently of contact with English, Torres Cacoullos (2012) argues that the predominant estar ‘to be’ + V-ndo progressive construction continues to grammaticalize; i.e., show gradual change over time between functional meanings and the different grammatical markers that express them (Bybee et al. 1994). In this case, estar + V-ndo has changed from a purely locative meaning (3a) to the progressive meaning it currently holds (3b) and is now expanding to habitual contexts (3c), where the present indicative form is prescriptively expected but estar + V-ndo is nevertheless acceptable, particularly in Latin American varieties. Such development supports the cross-linguistic tendency of periphrastic forms to increasingly occupy the terrain formerly held by morphological forms (Hopper and Traugott 2003). Although bilingualism might accelerate the change, contact with English by itself cannot explain shifts seen over time, which are not limited to bilingual contexts. Indeed, in an oral production analysis, Fafulas (2021) found similar patterns of progressive usage among Spanish-Yagua bilinguals and English-speaking learners of Spanish in the US, despite Yagua’s lack of an overt progressive marker, which might indicate that it is contact itself (with English or otherwise) that accelerates changes relative to monolingual Spanish.

In a diachronic analysis, Torres Cacoullos (2015) also presents evidence that the development of the Spanish progressive can be explained by a gradual loss of analyzability through usage-based chunking (Bybee 2010); i.e., the storing and processing of a string as a single unit resulting from frequent repetition. As evidence, Torres Cacoullos (2015) cites a continuously decreasing occurrence of intervening words (e.g., clitics) between estar and the present participle in 13th–20th century data. A significant structural priming effect for related estar constructions (e.g., locatives) was also found in 13th–17th century data, but it no longer held in the 20th century. Torres Cacoullos (2015) contends that such trends, all indicative of decreasing analyzability, show that gradual diachronic changes have led to synchronic variability. Most recently, Fuchs et al. (2020) used sentence acceptability data from speakers of Rioplatense, Iberian, and Mexican Altiplano Spanish as evidence that estar + -ndo is part of a widely documented cyclical and cross-dialectal grammaticalization path, where progressive constructions undergo syntactic generalization and become imperfective markers (Bybee et al. 1994). Fuchs et al. (2020) also reported that the distribution of present indicative and present progressive forms was subject to contextual modulation. In particular, ambiguous simple present forms were less likely to be accepted when the speaker and their interlocutor did not have shared perceptual access to the event in progress at reference time. Nevertheless, this difference did not hold in Mexican Altiplano Spanish, where the simple present was disfavored across categories, pointing to greater advancement for this variety in the progressive-to-imperfective grammaticalization path.

2.4 L2 variation and developmental stages

Recent studies have also analyzed variability in the present tense in L2 Spanish, adapting the methods and variables used in L1 research. In written preference tasks examining ongoing contexts, lexical aspect (particularly dynamic predicates: activities, accomplishments, and achievements) and the co-occurrence of temporal adverbs (notably those denoting immediacy) have been the strongest predictors of selection of the present progressive (Fafulas 2010, 2012; Geeslin and Fafulas 2022; Kanwit et al. 2015). In terms of learner level, Fafulas (2010) tested intermediate, advanced, and near-native learners and found that selection of the present progressive increased with course level, with near-native speakers patterning with the L1 Spanish baseline. Geeslin and Fafulas (2022) reported similar results in another cross-sectional study examining both ongoing and frequentative contexts. Lexical aspect and co-occurring adverbs were significant factors individually, and their interaction was also significant at higher levels, providing further evidence that learners may rely on different cues for selection across developmental stages.

Using a more open-ended simultaneous video narration task, Fafulas (2015) expanded the scope of variation to alternative progressive constructions (e.g., andar ‘go around’ + -ndo), with lexical aspect a significant predictor at all levels except the lowest considered (i.e., 4th semester). Other predictors gained significance at higher levels (i.e., adverbial type, temporal reference), revealing previously undocumented developmental stages (e.g., a lack of significance for lexical aspect among intermediate learners) hidden in previous work that used the same task to assess only advanced learners (Geeslin and Fafulas 2012).

Existing L2 studies of this contrast have not included usage-based factors as predictors, although lexical frequency may help account for how exposure to a particular linguistic structure in the input might shape mental representations (Blumenthal-Dramé 2012; Bybee 2006, 2008). Indeed, recent L2 variationist research has found factors such as lexical frequency to have direct (Solon et al. 2018) or mediating effects (Linford and Shin 2013; Linford et al. 2016) on variable structures such as intervocalic /d/ reduction and subject pronoun expression. Usage-based approaches, in addition, can operationalize frequency in different ways, including not only token frequency but also the frequency of selection of a variant in a collocational pairing, as in Kanwit and Geeslin’s (2020) finding that copula interpretation in frequent collocations was more susceptible to lexical effects for L1 Spanish speakers than for advanced learners. Similarly, a recent corpus study by Minnillo et al. (2022) examined L1 and L2 Spanish written production data and found evidence that certain verbs were significantly more likely to co-occur with the preterit or the imperfect. Moreover, correlations between the preterit and the verb forms likely to co-occur with it were modestly stronger than those of the imperfect, although differences became less prominent for highly frequent forms such as fue ‘was/went’ in the preterit and tenía ‘had/used to have’ in the imperfect.

Generative research has provided further valuable insights into the acquisition of the semantic values of the present indicative and the present progressive. Cuza and López Otero (2016) examined the distribution of the forms in both ongoing and habitual contexts in L2 learners, heritage speakers, and L1 Spanish speakers. The results of a sentence completion task indicated that L2 learners and heritage speakers overused the present indicative in ongoing contexts. This finding was also reflected in the results of a forced preference task (i.e., with only two choices), although heritage speakers approximated the L1 Spanish baseline more in selection than sentence completion. Moreover, all groups chose the present indicative over 50% of the time. López Otero and Cuza (2020) is the only recent study (regardless of theoretical orientation) to examine the distribution of past progressive forms in tandem with their present counterparts. Data came from semi-spontaneous oral narratives by English-speaking and Portuguese-speaking heritage speakers of Spanish from the US and Brazil, respectively. Although synthetic forms (present indicative and imperfect) were predominant in both present and past-time items, a follow-up analysis restricted to contexts with activity and accomplishment (i.e., dynamic) verbs revealed a significant effect for group. Spanish-English bilinguals overextended synthetic forms across contexts, whereas Portuguese-Spanish bilinguals did so with analytic forms, in line with the form-meaning mappings of their dominant languages. In terms of rates of overall response patterns (i.e., independently of linguistic predictors), the studies reviewed above found that synthetic forms were used or selected more frequently than analytic forms, with more prominent differences in production tasks, rather than preference, completion, or acceptability tasks.

In summary, variation between synthetic and analytic forms presents difficulty for English-speaking learners of Spanish because the two languages differ with regard to progressive aspect expression, with the English simple present tense prescriptively not allowing an ongoing reading (Cowper 1998; Huddleston and Pullum 2002). English encodes progressiveness primarily through one construction (be + V-ing), hence fostering the creation of a one-to-one form-meaning mapping in learners’ interlanguage (i.e., between an analytic form and an event-in-progress reading, even if the reading is also viable for a synthetic form in Spanish). As revealed in both L1 and L2 research, variation is also conditioned by numerous other linguistic factors to which learners must become sensitive. Prior variationist L2 studies have reported developmental stages (Fafulas 2010, 2015), as learners tend to overgeneralize at first, and develop gradual sensitivity to linguistic factors (primarily lexical aspect and adverbial type) as they gain proficiency. Thus, we see evidence of a transition from one-to-one to multifunctional form-meaning connections in learners’ TMA systems (Andersen 1984), worth examining in depth through progressive constructions as a lens for the study of multiple interlanguage form-meaning associations across temporalities (Bardovi-Harlig 2017).

4.1 Participants

Five groups participated in the current study. To account for different learner levels that may inform developmental stages, four L1 English learner groups were recruited: 2nd, 3rd, 4th, and 5th semester learners. All were adult students enrolled in a university in the northeastern United States. The second-semester course is the second half of the elementary Spanish-language sequence. Recruitment started at this level because the learners had already been introduced to progressive constructions. The 3rd and 4th semester courses are the first and second half of the intermediate sequence, respectively. Students in the 5th semester were enrolled in a conversation or advanced grammar course. In accordance with national proficiency guidelines (ACTFL 2012), students having completed the elementary language sequence should have skills corresponding to the novice high level, whereas completion of the intermediate and 5th semester courses should correspond to the intermediate mid-high and advanced low-mid levels, respectively.

A fifth participant group consisted of L1 Spanish speakers to serve as a baseline, as is customary in L2 variationist research (Geeslin 2003; Geeslin et al. 2018). These participants were Spanish-English bilinguals so that the target was that of speakers able to use both languages effectively rather than a monolingual ideal (Cook 2008; Ortega 2013). All L1 speakers were either instructors in the Spanish-language sequence or graduate students in another discipline from the same university speech community, who therefore reflected the cross-dialectal input provided in the classroom. Participants in this group originated from Spain (n = 9), Peru (n = 5), Mexico (n = 2), Colombia (n = 2), Venezuela (n = 2), and Cuba (n = 1). They had resided in the United States for an average of 4.68 years. Each group’s characteristics are reported in Table 2, including scores on a 25-item grammar test described in the next section.

4.2 Data elicitation

Participants completed three self-paced tasks in the following order: a written preference task, a grammar test, and a background questionnaire.^[1] The background questionnaire targeted sociodemographic data and information regarding participants’ L2 experience. It included questions related to enrollment, prior language instruction, and study abroad location and duration (when applicable). Participants in the baseline group completed it in Spanish and learners completed it in English. In the case of L2 Spanish groups, only learners instructed on Spanish (who did not grow up exposed to it) and who were currently enrolled in courses in the Spanish-language sequence were included in the study.^[2] The 25-item grammar test, completed in Spanish, was administered to confirm that all participants were grouped appropriately (i.e., that a given score was not markedly different from the group average), as employed in prior studies (Linford 2014).^[3] Each item on the test listed three possible responses for a multiple-choice question out of which only one response was prescriptively accurate. The grammatical structures tested included hypothetical clauses, presentational haber ‘there be’, and the use of prepositions and pronouns.

The preference task was also completed in Spanish and contained 30 contextualized items which told a story about a day in the life of a college student. Synthetic and analytic response options were automatically randomized to avoid order bias. This type of instrument has proven to be successful in studies of L2 variation of this and other variable structures (Geeslin 2003; Kanwit et al. 2015), as it provides the same number of equivalent contexts to the participants, while allowing the researcher to manipulate a range of independent variables. A sample item can be found after the description of variables in Section 4.3.

4.3 Data coding

Our dependent variable is the participant’s selection of (i) the synthetic form, (ii) the analytic form, or (iii) that both are equally possible in a particular context. The latter option was offered as in prior L2 variationist research on progressive aspect (e.g., Fafulas 2010; Kanwit et al. 2015) to avoid forcing a choice by the participant and to explore contexts in which the more innovative (i.e., analytic) form might be gaining ground, following the direction of language change, especially since these variants have received little L2 attention across temporalities. Four independent variables were manipulated: dynamicity, co-occurring adverbs, temporality, and lexical frequency.

Two categories were considered for each predicate’s dynamicity [dynamic or non-dynamic], the main lexical-aspectual opposition in previous studies on progressive expression (Fafulas 2012, 2015; Kanwit et al. 2015), as well as for the preterit-imperfect opposition in past-time expression (Domínguez et al. 2013, 2017; González and Quintana Hernández 2018; Salaberry 2011). Similarly, two categories were considered for the co-occurring adverb variable, with the item containing an adverb of [immediacy] or [none]. One of four temporal adverbials was included in items for the former category: ahora ‘now’, ahorita ‘(right) now’, ya ‘now’, and en ese momento ‘at that time’. The preceding coding scheme for dynamicity and co-occurring adverbs kept the instrument at an appropriate length for a variety of course levels and allowed for the incorporation of temporality and lexical frequency as novel predictors, along with the completion of the grammar test and background questionnaire. Tense was manipulated in the temporality variable,^[4] such that [past time] items contained an imperfect or imperfect progressive form;^[5] [present time] items contained a present indicative or present progressive form; and [future time] items included a synthetic future or future progressive form.^[6] Finally, lexical frequency was coded using rankings from the Frequency Dictionary of Spanish (Davies and Hayward Davies 2018), based on the 2-billion-word Corpus del español (Davies 2016–), as a measure of global, cross-dialectal frequency. More frequent lexemes rank more highly (i.e., show a smaller value in the frequency ranking). Correspondingly, two opposing frequency bands were created, so that verb lexemes fell either in the [high] or [low] range.^[7] For the 24 verbs included in the target items and their corresponding ranking in Davies and Hayward Davies (2018), see Table 3. Most verb lexemes were not cognates, although to ensure learner comprehension of some of the lower frequency verbs, some cognates were necessary (e.g., residir ‘reside’). To control for any item effects, each verb was used only once, and the individual item was entered as a random effect in modeling.

To avoid introducing additional sources of variation, all grammatical subjects were animate, and all verb forms contained positive polarity. In the preambles preceding responses we abstained from using present participles or the experimental verb (as listed in Table 3) provided in the responses in order to avoid priming a particular option. Factorial design was used to create the instrument: there was one combination of all categories of the four independent variables described in Table 4 (2 × 2 × 3 × 2 = 24). Although the use of three different temporalities helped obscure the purpose of the study, six additional items targeting different variable structures were also included to serve as distractors (e.g., the copula contrast), yielding the 30-item total.

Instructions prompted participants to choose the sentence they would be more likely to say, as in the sample item (4). This example, with almorzar ‘to have lunch’, was coded as [+dynamic lexical aspect], [no adverb], [present time], and [low frequency verb]. Bolding has been provided presently for the reader.

5.1 Distribution of the forms

The overall rates of selection of the variants by participant group are summarized in Table 5. No group differed drastically from the others, given that all selected synthetic forms in over 50% of contexts, with rates ranging from 50.27% for 4th semester learners to 61.24% for 5th semester learners. Groups were also similar in their selection of analytic forms, which L1 Spanish participants selected the most (34.92%) and 5th semester learners the least (30.68%). The “both equally” option was the least frequently selected, with 4th semester learners selecting it the most (16.80%) and 5th semester the only group to do so at rates below 10% (8.08%).

With respect to group differences, there were slight decreases in the selection of synthetic forms from the 2nd semester to the 4th semester. Likewise, 4th and 5th semester groups differed the most in terms of overall rates of selection for all three response options (Figure 1).

Figure 1:

Relative rates of selection by participant group.

For inferential statistical analyses, “both equally” responses were combined with analytic responses to determine which contexts permitted acceptance of the analytic form and where it might be gaining ground on the synthetic variant. This decision was preferred rather than excluding a sizeable, informative subset of the data. Such a combination also aligns with the direction of language change attested in studies of L1 Spanish variation (Fuchs et al. 2020; Torres Cacoullos 2012, 2015).

An independent one-way analysis of variance (ANOVA) examined the mean proportion of analytic forms allowed (dependent variable), with participant group as the between-subjects factor (independent variable). No significant effect for participant group was observed in the model, F (4, 133) = 2.29, η ² = 0.06, p > 0.05. The lack of significant differences between groups (Figure 2), which showed a uniform trend when examining exclusively raw averages, indicates that the participant’s group alone cannot explain response patterns. To tease apart developmental patterns and find whether there are expected differences according to a range of linguistic variables, we conducted an analysis by variable context on the unaveraged data.

Figure 2:

Boxplots of mean analytic forms allowed by group. Note. Boxes show the interquartile range representing 50% of the data, horizontal bars represent medians, whisker extend to 1.5 times the interquartile range. Dots stand for data points falling above or below 1.5 interquartile ranges.

5.2 Mixed-effects models

We now consider the results of mixed-effects regression models meant to determine how each of the independent variables conditioned selection. For this purpose, we used GLMs with a logit link function. This choice of model responds to the binary dependent variable and also minimizes possible effects of associations between data points, as they can be clustered (Verbeke et al. 2018). Thus, the independent linguistic variables were considered as fixed effects,^[8] whereas the participant and target item were considered as random effects to account for the influence of variation at the individual participant level and any additional variability that could be attributed to the experimental item, respectively. We also tested for any significant interactions between the independent variables tested.

Separate regressions were run for each group to examine differences in predictor significance and effect sizes across groups, which would be obscured in an analysis that considered all levels together and assumed the same underlying grammar. We report the results of the best-fit model for each group, which contains only variables and interactions that significantly aided the model’s predictive power (as indicated by AIC values and ANOVA comparisons between the models). For the models reported, variation inflation factors revealed no multicollinearity between the predictors, and validation with bootstrapping indicated that no models were overfitted. Table 6 summarizes the significant and non-significant linguistic predictors and interactions in the regression analyses run for each participant group, as indicated by p-values. Reference levels are specified first in the subscript, followed by predicted levels. A hyphen indicates that the comparison was not included in the best-fit model for that group.

Of the four linguistic predictors examined, L1 Spanish speakers were only conditioned independently by lexical aspect, although the interaction between dynamicity and temporality was also significant for this group. The 2nd semester group was not significantly influenced by any of the fixed predictors independently, although the interaction between lexical frequency and co-occurring adverb yielded significance. The 3rd semester group was significantly conditioned by dynamicity, lexical frequency, and temporality, in addition to a significant interaction between dynamicity and both levels of the temporality variable. A similar pattern emerged at 4th semester and 5th semester levels, when dynamicity and lexical frequency, along with an interaction between dynamicity and temporality (_{present → future}), were significantly predictive of selection. Although this overview provides a first glance at developmental patterns, consideration of the effects and directions of effect conveys important information about shifts in learner grammars, as addressed with a focus on each of the four linguistic independent variables in Sections 5.2.1–5.2.4. Positive estimates (β) indicate a higher likelihood of analytic form allowance, whereas negative values indicate disfavoring of analytic forms (i.e., favoring of the synthetic response). Larger values indicate stronger effects. For reference, full models and exact values are provided in the Appendix.

5.2.1 Dynamicity

L1 Spanish speakers were significantly conditioned by dynamicity, which was the most predictive independent variable in the group’s model (β = 5.49, p < 0.001). Despite 2nd semester learners’ slight preference for analytic forms in dynamic contexts, dynamicity had no significant effect for this group. Conversely, the difference between synthetic and analytic selection according to dynamicity increased for 3rd semester learners (β = 1.38, p < 0.01), and noticeably so for 4th semester (β = 2.52, p < 0.001) and 5th semester learners (β = 2.58, p < 0.001), for whom this predictor was highly significant with an estimate closer to the L1 Spanish baseline. In this sense, the consistent increase in estimates across learner groups highlights the importance of dynamicity as a predictor (Figure 3).

Figure 3:

Percentage of analytic forms allowed according to dynamicity.

5.2.2 Temporality

Temporality, and specifically the comparison between present-time and past-time items, was a significant predictor for the 3rd semester group (β = 0.99, p < 0.05). For this same group, the interaction of dynamic verbs with past-time contexts (β = −1.29, p < 0.05) and with future-time contexts was significant (β = −1.77, p < 0.01), whereas only the later interaction was significant for the 4th semester (β = −2.42, p < 0.01), 5th semester (β = −2.68, p < 0.01), and L1 Spanish groups (β = −7.21, p < 0.001). Regarding overarching patterns of selection, learners in the 3rd semester chose analytic forms most frequently in past-time contexts. This directionality also held for learners in the 4th semester, but in their case present-time items did not differ significantly from the other temporalities. In the case of 5th semester leaners and L1 Spanish speakers, there was a shift in preference, as analytic forms were more frequently selected in present-time contexts for both groups, followed by past-time contexts. As can be seen in the bar charts (Figure 4), learners started with a lack of distinction for temporality and progressed into a pattern of variation where analytic forms were most frequently selected in past-time contexts, until learners reached the 5th semester, when the present became the most frequent context for analytic forms, with the future on the opposite end.

Figure 4:

Percentage of analytic forms allowed according to temporality.

5.2.3 Lexical frequency

Analytic forms were significantly more likely to be selected in contexts that included a verb in the low frequency range for participants in the 3rd semester (β = 0.65, p < 0.01), 4th semester (β = 0.91, p < 0.01), and 5th semester (β = 1.06, p < 0.05). This effect was also attested in the 2nd semester group but only in co-occurrence with an adverb of immediacy, as indicated by the interaction term (β = 0.90, p < 0.001). Analogous to dynamicity, estimates for lexical frequency increased with learner level. Unlike dynamicity, however, frequency was not significant for the L1 Spanish group. Nevertheless, learners matched the directionality of L1 Spanish preference by frequency at all levels (Figure 5).

Figure 5:

Percentage of analytic forms allowed according to lexical frequency.

5.2.4 Co-occurring adverb

The presence of a co-occurring adverb was, overall, the least predictive factor. Although all groups selected analytic forms slightly more when items contained an adverb of immediacy (Figure 6), the distinction between these items and those with no adverb was not significant for any group. It yielded significance in one interaction, with lexical frequency, for the 2nd semester learners.

Figure 6:

Percentage of analytic forms allowed according to co-occurring adverb.