2.1 Third-person clitics in Spanish and Brazilian Portuguese

Spanish third-person accusative clitics (e.g., lo, la, los, las) are unstressed accusative pronouns that change syntactically and phonologically depending on the verb to which they are attached (Givón 1978; Campos 1986; Clements 1994; 2006; Sánchez 1999; Borgonovo et al. 2006; Leonetti 2011; Cuza et al. 2013b; Zdrojewski & Sánchez 2014). In this language, third-person accusative clitics are definiteness-driven, meaning the clitics are used depending on whether the preceding determiner phrase (DP) is known or identifiable by the speaker or hearer regardless of its specificity (Givón 1978; Campos 1986; Clements 1994, 2006; Sánchez 1999; Leonetti 2011). That is, accusative clitics referring to definite objects are overt (Example 1) while those referring to indefinite objects are null (Example 2) (Campos 1986; Clements 1994; 2006; Sánchez 1999; Borgonovo et al. 2006; Zyzik 2008; Leonetti 2008, 2011; Cuza et al. 2013b; Nardelli & Lobo 2017). Some varieties of Spanish, nevertheless, feature null objects as a result of language contact, including Spanish varieties from Ecuador, Paraguay, Peru, and the Basque Country in Spain (Landa 1995; Sainzmaza-Lecanda & Schwenter 2017; Schwenter 2006).

Additionally, depending on the nature of the verbal host, clitics can be attached as inflexions of their verbal host (known as enclisis) or they can also be independent lexical units placed in front of their verb (known as proclisis) (Pineda & Meza 2005). Proclisis, which is notably more frequent, occurs with finite verbs and negative imperatives (Example 3a-b), while infinitives, gerunds and affirmative imperatives have enclisis (Example 4a-b).

In BP, third-person accusative pronouns are not governed solely by definiteness. They are sensitive to animacy, meaning they are used more frequently when the object in question is human or animate (Schwenter & Silva 2002; 2003; Kato & Raposo 2007; Kato et al. 2009; Sainzmaza-Lecanda & Schwenter 2017). Although BP and Spanish share a significant amount of lexicon and similar syntax, the use of overt and null clitics is determined by different lexical and semantic features in either language. Examples 5 and 6 illustrate the contrast between Spanish and BP clitic use with regard to animacy, which is the focus of the present study.

In Example 5, the accusative pronoun is overt in both languages; the DP, Maria, is animate, making the accusative pronoun overt in BP, and definite, making the clitic overt in Spanish as well. The DP in Example 6, foto da viagem ‘picture from the trip,’ is inanimate, so BP is more likely to present a null clitic, but because the phrase is definite, the clitic is overt in Spanish. In other words, in BP, null clitics commonly refer to inanimate referents in the discourse, yet overt accusative pronouns can refer to inanimate referents as well (Duarte 1989; Cyrino 1994; Creus & Menuzzi 2004).

The marking of a direct object in BP can appear in two positions in relation to the verb: proclisis and enclisis, both of which are rare in spoken language and are more common in written formal registers, which are acquired in educational settings (Duarte et al. 2005; Galves et al. 2005; Kato et al. 2009; Montrul et al. 2011; Washington 2015). Despite both forms being rare in BP, proclisis (7a) is more likely to occur in BP than enclisis (7b) (De Carvalho 1989; Nunes 2015). Examples 7a and b illustrate proclitics and enclitics, respectively:

BP also presents strong pronouns (pronome tônico) (e.g., ele, ela, eles, elas) as an alternative to clitics (Simões 2010). In fact, third-person clitics, including both proclitics and enclitics, have been claimed to not belong to the BP paradigm anymore (Corrêa 1991; Kato 1999; Galves 2002; Galves et al. 2005), and while they still appear in written texts, they do not occur as frequently as they do in European Portuguese, especially o/a (the third person accusative) and lhe (the third person dative) (Galves et al. 2005). Cyrino et al. (2000) explain that because of third person clitics diminishing in use, animate antecedents are typically referred to with strong pronouns (e.g., ele, ela), and this can be seen in examples 8 and 9 taken from Galves et al. (2005: 155):

In sum, clitics in Spanish are overt when the DP is definite and may be null when the DP is indefinite. In BP, the distribution of accusative pronouns is sensitive to animacy, meaning that when the preceding DP is animate, the accusative pronoun is usually overt while when the phrase is inanimate, the accusative pronoun is more likely to take a null form. Also, overt clitics in BP can occur in two positions, proclisis and enclisis, which are rare in BP, particularly enclisis, and occur mostly in formal written contexts. Additionally, third-person clitics are leaving the BP paradigm and are being replaced by strong pronouns.

2.2 L2 acquisition of Spanish clitics

There have been extensive studies on the acquisition of clitics in Spanish (e.g, Duffield & White 1999; Borgonovo et al. 2006; Zyzik 2008; Montrul 2010; Cuza et al. 2013b; Iverson & Rothman 2015; Nardelli & Lobo 2017; Shin et al. 2017; 2019, amongst others), which include those of Spanish clitic acquisition across various first language backgrounds such as Spanish heritage speaking natives of BP. L1 English-speaking learners of Spanish (Zyzik 2008), European Portuguese (EP) native speakers who learned Spanish as children (Nardelli & Lobo 2017), and Chinese-speaking learners of Spanish (Cuza et al. 2013b). Cuza et al. (2013b) found that Chinese-speaking learners of Spanish accept and produce null clitics in Spanish due to cross-linguistic influence. Similarly, Zyzik (2008) found that English-speaking learners of Spanish with low proficiency accept ungrammatical null clitics in Spanish. Collectively, these studies point to an effect of proficiency in the target-like use and acceptability of clitics in Spanish.

Previous research on clitic acquisition focuses mostly on EP (Costa & Lobo 2009; Costa et al. 2008; 2009; 2015; Madeira & Xavier 2009; Nardelli & Lobo 2017, among others). However, there have been few studies on BP accusative pronoun acquisition such as Kato et al. (2009) who investigated the acquisition of BP accusative pronouns among native speakers and found that strong pronouns were typically used in place of third-person clitics, but when the DP was inanimate, they were null. Also, Montrul et al. (2011) examined the acquisition of BP accusative pronouns by L1 English L2 Spanish L3 BP speakers, and found there to be transfer effects from Spanish, which is likely because of the similarity between the languages’ structures and the cross-linguistic tendencies in L3 acquisition. In another study, Parma (2017), which investigated BP accusative pronoun acquisition among L3 BP learners who spoke English and Spanish before, found there to be no significant transfer, but a tendency for speakers to use more null clitics in inanimate and indefinite contexts, which the author traces back to Spanish.

There have, however, been several studies from Brazil, Spain, and Chile that research clitic acquisition by native BP speakers learning Spanish as adults (e.g., González & Barros 1994; González 1998; 1999; Sebold 2002; Simões 2010; 2015; Dos Santos 2011; Iverson & Rothman 2015; Yokota 2016; Cerda-Oñate et al. 2017; Dutra 2019; Lozado & Unternbäumen 2019). For instance, Simões (2010) looked at the effects of age and education level on the acquisition of third-person accusative clitics and found that participants ages 12–18 had a more permeable BP grammar, and participants over the age of 50 had a less permeable grammar, with their non-native intuitions leading them to accept more overt clitics. In a more recent study, Simões (2015) found a generational effect and noted that there was a coexistence of the BP grammar in the Spanish of L2 learners, but that this grammar was more permeable in younger learners than in older learners. Typically, prepubescent learners are less prone to fossilization than older learners, which means that grammatical errors are easier to correct in younger learners (Moskovsky & Ratcheva 2014).

Rothman & Iverson (2013) investigated restrictions on object drop in L2 Spanish among native speakers of BP in simple clauses and island environments. In this study, learners showed knowledge of the semantic restriction of definiteness in simple clauses but had variability in their knowledge of syntactically complex structures. For instance, they respected the syntactic constraints in only one condition (DP islands), which shows that these learners know that some phonetically absent objects in Spanish are possible. However, unlike the control group of native speakers, the L2 learners were not able to distinguish between grammatical and ungrammatical dropped objects in the other conditions (subject-CP islands and adjunct islands) because they relied only on definiteness (the L1 ruling feature) to determine grammaticality. Furthermore, Yokota (2016) found that in the early stages of Spanish acquisition, BP speakers displayed a strong tendency to use the strong pronoun in their L2. This is likely the case because in BP, the strong pronoun is what is typically used, so they are experiencing a negative transfer from their L1, which can be corrected through formal education (Yokota 2016).

Other studies have found evidence of transfer between the participants’ BP and Spanish, which affects the production of personal pronouns, such as the use of ello or elle instead of the masculine third-person pronoun él (González 1999; Cerda-Oñate et al. 2017; Lozado & Unternbäumen 2019). This transfer can be positive in constructions that are the same in BP and Spanish, but negative in cases with similar but not identical use of pronouns such as unstressed accusative pronouns, clitic position, and null objects (Cerda-Oñate et al. 2017). This finding was corroborated by González (1999) who found a high degree of fossilization in the interlanguage of Brazilians learning Spanish. The fossilization of interlanguage, as suggested by Dutra (2019), can be combated by conducting an instructional intervention in learning third-person clitics in Spanish. Results from Dutra (2019) showed that this intervention produced durable effects and progress in college students.

In addition to the aforementioned factors, the proficiency level of Spanish learners has a significant influence. One study by Sebold (2002) focused on the acquisition of overt clitics and their placement among beginner Spanish learners who have studied the language for six months, intermediate learners studying for a year, and advanced learners who have studied for at least three years. Results showed that participants had trouble with overt clitics as they used non-target null clitics, yet they did not struggle with the position of overt clitics, particularly as the language level increased. Similarly, Dos Santos (2011) concluded that null clitics show influence from BP into Spanish, but as the Spanish level increases, these errors are corrected. The author noted the need to further study how this process works.

In sum, previous studies on the acquisition of Spanish clitics by BP native speakers learning Spanish as an L2 have found that while there tends to be a coexistence of BP grammar in the Spanish of L2 learners (i.e., the tendency for speakers in early stages of acquisition to use strong pronouns in Spanish where clitics are typically used), younger speakers between 12 and 18 years of age have a more permeable BP grammar than speakers over the age of 50, likely due to less effects of fossilization (Simões, 2010; 2015; Yakota, 2016). An effective way to correct this interlanguage is through formal education in the form of an instructional intervention explicitly teaching third-person clitics in Spanish (González 1999; Yakota 2016; Dutra 2019). Although there is a high level of transfer from the BP accusative pronoun system, it is typical for participants to correct their Spanish clitic errors as their language level increases (Sebold 2002; Dos Santos 2011).

4.1 Participants

The study comprised 97 adult participants: 72 L1 BP-L2 Spanish speakers ((54 female and 20 male; mean age = 33.84, sd = 9.87) who were all born and raised in Brazil, and lived in this country when data collection occurred) and 23 native Spanish speakers from Mexico. The L1 BP participants had learned Spanish in Brazil after puberty through formal education, and they spoke varieties of BP from all the regions in Brazil, with the Southeastern (n = 37) and Northeastern (n = 17) regions being the most represented, followed by the South (n = 10), Central-West (n = 2), and North (n = 1), while seven participants did not disclose what variety of BP they spoke. Participant recruitment occurred through a recruitment flyer that was shared on the website of the Consejería de Educación en Brasil, which is the institution that represents the Spanish Ministry of Education and Professional Training in Brazil. All the BP-speaking participants were attending or had graduated college at the time of testing. All the participants lived in urban areas of Brazil when data collection happened. Some of them had not received exposure to Spanish in a Spanish-speaking country (n = 22) whereas most of them had either studied, worked or visited at least one Spanish-speaking country (n = 52), including Argentina (n = 28), Spain (n = 26), Uruguay (n = 19), Chile (n = 18), Peru (n = 8), Bolivia (n = 7), Mexico (n = 6), Paraguay (n = 6), Colombia (n = 5), Dominican Republic (n = 2), Cuba (n = 1) and Ecuador (n = 1). The participants who reported visiting Bolivia, Ecuador, Paraguay and Peru also reported visiting other Spanish-speaking countries, indicating that they had been exposed to several varieties of native Spanish outside the classroom. Therefore, we do not expect regional variation in the use of null clitics among Spanish speakers to play a role in the L2 speakers’ knowledge of clitic distribution.

The comparison group in this study consisted of 23 native Spanish speakers from East-Central and Northeastern Mexico. This group comprised 13 females and 10 males (mean age = 42.87, SD = 14.59). They were all born and raised in Mexico and were residing in the country at the time of data collection. Recruitment for this group involved posting on social media and using the snowball approach, in which participants help recruit additional potential participants. Among these native speakers, 20 had earned college degrees, while three had only completed high school. This group encompassed a diverse range of professions, including teachers, engineers, psychologists, hotel managers, and others.

4.2 Data collection instruments and procedures

Before conducting the study, the participants read and signed a consent form that was previously approved by the IRB and informed them of the length of the study, the procedures, and confidentiality of their data, as well as the compensation for their time (a R$50 gift card for a store in Brazil or a $200 MXN gift card for a store in Mexico, which is the equivalent of a $10 USD gift card). All the gift cards were electronic codes that were distributed to the participants via email upon completion of the study.

The study had two main parts: First, the participants met with one of the researchers on Zoom to sign the consent form and to complete a language background questionnaire (LEAP-Q; Marian et al. 2007), as well as the Prueba Léxica de Español y Portugués/Prova Léxica do Espanhol e Português (PLEP; López Otero & Jimenez), which is a lexical knowledge test that was used as a proxy for general language proficiency. The PLEP is a picture naming task with 58 non-cognate items that are organized based on the lexical frequency. This task was administered first in Spanish and then in BP (for the L2 speakers only). Lexical frequency was controlled following the frequency data available in the Corpus del Español (Davies 2018a) and Corpus do Português (Davies 2018b). If participants struggled to recognize a picture, they were given semantic or phonemic cues. Correct naming before or after cues was marked as correct. If, even after receiving cues, participants failed to name the object in the picture correctly, the item was marked as incorrect. The scores for each language are reported in Table 1, which shows that the participants’ lexical knowledge in BP is stronger and less variable than in Spanish, the L2. Appendix A contains the full version of the PLEP.

Measuring language proficiency using productive lexical knowledge as a proxy is congruent with previous literature that has suggested that lexical range, sophistication, and richness in oral production tasks correlate with language dominance in bilingual populations (Daller et al. 2003; Pienemann et al. 2011; Gollan et al. 2012; Treffers-Daller & Korybski 2015). We believe that, in this case, increased vocabulary knowledge reflects more experience with the target language. Similarly, using the PLEP as a proxy for overall proficiency is congruent with research showing the power of picture-naming tasks (such as the Multilingual Naming Test, Gollan et al. 2012) in predicting language dominance (García & Gollan 2022; Sheng, Lu & Gollan 2014).

Following their completion of the PLEP via Zoom, participants were provided with a Qualtrics link and their unique participant code to proceed with the second phase of the study. The Qualtrics link directed them to a modified version of the DELE test (Cuza et al. 2013b), which assesses morphosyntax and lexicon, both of which these two languages share significantly. Table 1 summarizes the PLEP and DELE scores among the L2 speakers. Please note that the Spanish native speakers, who did not complete the DELE or the PLEP in BP, performed at ceiling in the Spanish PLEP (range = 56–58/58; M = 57.78; SD = 0.60).

The participants also completed, on Qualtrics, two experimental tasks: a written Elicited production Task (EPT) and an Acceptability Judgment Task (AJT), which were completed in Spanish first, and then, the bilingual participants completed the tasks in BP. The experimental tasks did not include test items examining indefinite antecedents; therefore, null clitics, although grammatical under a different condition, were not expected to be produced in the Spanish tasks. The AJT also did not contain items examining indefinite antecedents.

The EPT was used to examine the participants’ use of clitics/accusative pronouns (or other forms such as strong pronouns) in their two languages based on the animacy of the object. This task required the participants to fill in the blanks with the appropriate use of clitics/accusative pronouns, and the desired verbs were given in parentheses. The condition animacy had two levels (animate vs. inanimate) based on the animacy of the direct object. All direct objects were feminine and singular. This task had 32 experimental items (64 distractors per language), distributed in two conditions (k = 8, 16 per language). Due to the high number of cognates between Spanish and BP and for consistency purposes, we avoided non-cognates in these controlled tasks. Table 2 provides sample items per condition in both languages and Appendix B includes all EPT test items.

The AJT examined the participants’ receptive knowledge of the clitic system. The AJT included 32 experimental items (64 distractors per language) distributed in four conditions (k = 4; 16 in each language). The participants were presented with sentences featuring grammatical and ungrammatical sentences with both animate and inanimate objects in each language. The participants were asked to rate each sentence using a Likert scale from 1 to 5 (1 – muy extraño ‘very odd’, 2 – extraño ‘odd’, 3 – ni bien ni mal ‘neither good nor bad’, 4 – bien ‘good’, 5 – muy bien ‘very good’). Every time a participant rated an item 1 or 2, they were also asked to explain why, in order to confirm that they had rejected the item due to the phenomenon under examination. The items were counterbalanced and pseudorandomized. Tables 3 and 4 show the distribution of items according to conditions in both Spanish and BP, respectively. Please note that, given the variability regarding clitic realization in BP, we do not label any type of clitic as ungrammatical in the BP version of the AJT. Appendix B includes all AJT test items.

4.3 Data analysis

For this study, only Spanish language data were analyzed with the use of inferential statistics, which were performed with R (R Core Team 2021). The BP data were analyzed using descriptive statistics only as its main purpose is to serve as an L1 baseline for the L2 learners’ acquisition of clitics in Spanish.

The data from the EPT were analyzed using two generalized linear mixed effects models (GLMM) with the help of the glmer function from the lme4 package (Bates et al. 2015) in R. These two GLMMs examined the participants’ responses from two perspectives: by analyzing between-group differences existing between the native and the L2 speakers and by exploring within-group differences among the L2 speakers, with particular attention to the effects of their productive vocabulary knowledge, as measured with the PLEP. In both GLMMS, the participants’ response (1 = proclitic vs. 0 = null clitic; other responses were discarded) was the dependent variable while condition (0.5 = animate antecedent vs. –0.5 = inanimate antecedent) was an independent variable. Additionally, the first GLMM, which examined between-group differences, included group (0.5 = Native vs. –0.5 = L2) as an independent variable whereas the second GLMM, which examined within-group differences among the L2 speakers, included PLEP scores (continuous variable) as an independent variable. PLEP scores were standardized prior to performing the analysis. Furthermore, the second GLMM included interactions between condition and PLEP scores. No other fixed effects structures were tested in any GLMM. Both GLMMs included random intercepts for each subject as well as for each item.

The AJT data were analyzed using two ordinal regression models (ORM) with the clmm function from the ordinal package (Christensen 2022) in R in order to determine the effects of grammaticality, animacy, and productive vocabulary knowledge on the participants’ receptive grammatical knowledge. In both ORMs, response (1 through 5) was the dependent variable while grammaticality (0.5 = grammatical vs. –0.5 = ungrammatical), and condition (0.5 = animate antecedent vs. –0.5 = inanimate antecedent) were independent variables. As in the EPT analysis, the two ORMs examined the AJT data from two perspectives: by analyzing between-group differences and within-group effects. The first ORM, which examined between-group differences, included group (0.5 = Native vs. –0.5 = L2) as an additional independent variable whereas the second ORM, which examined within-group differences among the L2 speakers, included PLEP scores (continuous variable) as an additional independent variable. Both ORMs tested for all possible interactions between their independent variables and included random intercepts for each subject as well as for each item. Appendices C and D include the code used to perform the GLMMs and ORMs as well as their outputs, respectively.

5.1 Brazilian Portuguese results

The BP EPT collected 1184 responses. The participants produced mostly proclitics (n = 673; 56.84%), particularly referring to animate antecedents (n = 393; 66.39% vs. n = 280; 47.30%). Null clitics were the second most produced response (n = 338; 28.55%), especially in contexts featuring inanimate referents (n = 240; 40.54% vs. n = 98; 16.55%). The participants’ responses also included enclitics (n = 61; 5.15%) as well as strong pronouns (n = 18; 1.52%), but with lower frequency. A total of 94 responses were discarded (7.94%) due to different reasons: the use of a periphrasis instead of the verb provided for the test item (e.g., Mário queria visitar a sua cidade natal e poder visitar finalmente o ano passado ‘Mário wanted to visit his birth city and to be able to finally visit last year’ where the main verb to be used was visitar ‘visit’) or different interpretation of the argument structure of the main verb (e.g., Carlos salvou a grande árvore e se abraçaram por vários minutos ‘Carlos saved the big tree and they hugged for several minutes’). Additionally, the responses of one participant were discarded as they used Spanish to complete this task. Table 5 provides a distribution of the participants’ response types across conditions (i.e., animate and inanimate).

These results indicate that proclitics are the most frequent response across conditions and within each condition. Nevertheless, null clitics account for over 40% of the responses in the inanimate object condition. Overall, proclitics and null clitics represent over 85% of the participants’ responses. Strong pronouns were rarely produced (18 items or 1.52% of the time). Out of these 18 items, 13 were produced by a male speaker from Espirito Santo, 4 by a female speaker from Ceara, and 1 by a female speaker from Rio Grande do Sul. Given the low number of participants using strong/tonic pronouns and their different origins, we do not believe that regional variation played a role in the control experiment.

The BP AJT data also included 1184 responses across grammaticality and animacy conditions, 41 (3.46%) of which were discarded as the participants had rejected them for reasons unrelated to the phenomena under examination in this study. Table 6 shows the BP-speaking L2 Spanish speakers’ acceptability data distribution across the contexts under examination in their L1.

The BP AJT data shows variability in their responses. It also shows, nevertheless, that null clitics referring to animate objects receive balanced ratings and the highest number of rejections across conditions: 125 (26+99) out of 296 ratings were rejections and 127 (61+66) were acceptances. On the other hand, null clitics referring to inanimate objects received the highest amount of acceptance ratings across the four contexts under examination: 190 (92+98) out of 296 ratings were acceptances. These findings are consistent with previous literature describing that in BP null clitics are preferred to refer to inanimate objects and usually avoided to refer to animate objects (Kato et al. 2009).

5.2 Spanish language results

The Spanish EPT gathered a total of 1552 responses: 1184 of these responses were produced by the 74 L2 group while 368 were given by the native Spanish-speaking group. Most of the L2 group responses were instances of proclitics (n = 771; 65.54%), particularly referring to animate antecedents (n = 419; 70.78% vs. n = 352; 59.46%). On the other hand, the majority of null clitics were used to refer to inanimate antecedents (n = 198; 33.45% vs. n = 118; 19.93%). Discarded responses amounted to 97 of the EPT Spanish responses (8.19%): 55 (9.29%) of these were produced in the animate object condition while 42 (7.09%) belong to the inanimate object condition. These discarded responses included cases of use of a non-finite verb form (e.g., José cargó a la bebé Mariana y colocarla en la cuna ‘José took the baby Mariana and place her in the crib’), of a periphrasis instead of the verb provided for the test item (e.g., Juan conoció a Maria en Internet en 2012 y fue verla en persona en 2014 ‘Juan met María on the Internet in 2012 and went see her in person in 2014’ where the main verb to be used was ver ‘to see’) or of a different main verb (e.g., Luis llamó a Clara la semana pasada y ella fue en su casa hoy ‘Luis called Clara last week and she went in his house today’ where the main and only verb to be used was visitar ‘visit’). The native speakers produced proclitics across the board: out of 368 responses, 329 (89.04%) were proclitics and only 6 (1.63%) were null clitics. The remaining 33 (8.97%) responses were discarded due to reasons similar to the ones affecting the production of the L2 speakers. Table 7 shows the distribution of response types across conditions and groups in the Spanish EPT.

The first GLMM (GLMM1) analyzing the Spanish EPT data, which focused on between-group differences, revealed that the native speaker group produced more proclitics than the L2 participants (β = 4.58, SE = 0.89, z = 5.14, p < 0.01). Predicted probabilities also indicate this contrast: the predicted probabilities for native speakers to produce proclitics are 0.99 for both inanimate and animate referrents. On the other hand, the predicted probabilities for L2 speakers to produce proclitics is 0.74 and 0.91 for inanimate and animate referrents, respectively.

Additionally, the second GLMM (GLMM2), which explored within-group effects among the L2 speakers, revealed that L2 speakers were less likely to produce proclitics when referring to inanimate objects (β = 1.31, SE = 0.38, z = 3.40, p < 0.01) and that higher PLEP scores led to more use of proclitics (β = 1.60, SE = 0.34, z = 4.65, p < 0.01). The model failed to find an interaction between these two main effects (β = 0.21, SE = 0.24, z = 0.89, p = 0.38), which indicates that the L2 learners produce more proclitics as their productive vocabulary increases regardless of the animacy features of the object to which those proclitics refer. This effect is also noticeable in the model’s predicted probabilities: these remain below 0.60 for both inanimate and animate conditions on the lower end of the PLEP scores (at –2 and –1.5 from the standardized PLEP scores) versus values above 0.97 on the higher end of the PLEP scores (1.5 and 2 from the standardized PLEP scores). Indeed, predicted probabilities signal that the L2 speakers’ development of productive knowledge of the distribution of clitics under examination plateaus after their standardized PLEP scores reaches 0.5 as they remain above 0.9 whereas a steeper increase takes place on the lower half range of the standardized PLEP scores. Appendix D includes the outputs of the GLMMs and predicted probabilities discussed above.

The Spanish AJT also collected 1552 responses (1184 from the L2 group and 368 from the native Spanish-speaking group). Ten responses from the L2 speakers were discarded as they were cases of rejections due to reasons unrelated to the scope of the study. Table 8 below shows the distribution of response types across conditions and groups in the Spanish EPT.

The first ORM (ORM1) analyzing the Spanish AJT data focused on between-group differences and found that, in general, grammatical test items received higher ratings than ungrammatical test items (β = 4.22, SE = 0.14, t = 23.00, p < 0.01) and native speakers gave overall lower ratings (β = –1.14, SE = 0.43, t = –2.64, p < 0.01). No other main effects were significant. However, ORM1 found one significant two-way interaction between grammaticality and group (β = 2.00, SE = 0.26, t = 7.58, p < 0.01), suggesting that the native speakers were more likely to reject ungrammatical items than the L2 speakers. These effects are reflected in predicted probabilities. For instance, the predicted probabilities for native speakers to rate ungrammatical test items as ‘1’ or ‘2’ were 0.27 and 0.58, respectively, for both inanimate and animate objects whereas, for L2 speakers, the predicted probabilities were 0.03 and 0.28 for inanimate objects and 0.06 and 0.42 for animate objects.

The second ORM (ORM2), which explored within-group effects among the L2 speakers, found grammaticality effects (β = 3.33, SE = 0.17, t = 19.12, p < 0.01), which indicates that grammatical test items received higher ratings than ungrammatical ones. Moreover, ORM2 found a significant two-way interaction between grammaticality and PLEP scores (β = 1.09, SE = 0.13, t = 8.34, p < 0.01), which highlights that L2 speakers with higher PLEP scores were more likely to reject ungrammatical test items. A look at the predicted probabilities shows that L2 speakers on the lower end of the PLEP scores (at –2 in standardized PLEP scores) rarely rated ungrammatical test items as ‘1’ or ‘2’ (0.004 and 0.6, respectively, for inanimate objects and 0.008 and 0.10 for animate objects) whereas L2 speakers on the higher end of the PLEP scores (at 2 in standardized PLEP scores) consistently rated ungrammatical test items as ‘1’ or ‘2’ (0.16 and 0.58, respectively, for inanimate objects and 0.29 and 0.57 for animate objects). Appendix D includes the outputs of the ORMs as well as the predicted probabilities discussed above.

5.3 Summary of the results

The BP EPT data show that participants produced mostly proclitics and null clitics and that these were not distributed evenly across animacy conditions: proclitics were the most frequent strategy used in both animacy conditions, yet more instances of null clitics were produced when referring to inanimate antecedents than when the antecedents were animate. This relationship between null clitics and inanimate objects is also observed in the BP AJT data: null clitics referring to animate objects received low ratings whereas null clitics referring to inanimate objects received high ratings. Overall, the BP results indicate that null clitics in BP are mostly used to refer to inanimate objects.

Regarding the Spanish data, we can conclude that the L2 participants have acquired that proclitics are the grammatical clitic form in combination with finite verbs: no enclitics or strong pronouns were produced. Nevertheless, the native group baseline allows us to determine that, overall, the L2 speakers are still in the process of acquiring the distribution of proclitics in Spanish, particularly as opposed to null clitics. The L2 speakers with higher PLEP scores produce more proclitics and abandon the use of null clitics regardless of the animacy features of the object to which the clitics refer. Similarly, the L2 speakers with higher PLEP scores were more likely to reject ungrammatical test items. Overall, PLEP scores predicted the L2 speakers’ productive and receptive grammatical knowledge.