2.1 VPE

As is well-known, VPE is a special kind of ellipsis in the sense that it is less constrained than other elliptical constructions such as gapping or sluicing. For instance, (i) it may be exophoric (as in example (1), see Miller & Pullum’s 2014 seminal work), (ii) it may be anaphoric or cataphoric (as in examples (2) and (3), respectively), (iii) it may allow for voice mismatches between the antecedent and the ellipsis site (as in example (4), see Kertz 2013; Poppels & Kehler 2019; Gandón-Chapela 2020a; 2020b), and (iv) it may allow for different inflectional features with respect to its antecedent (as in example (5)):

It should also be noted that certain discourse conditions have been shown to play a role in cases of VPE, whose focus may be either auxiliary choice or subject choice. These discourse conditions have been tackled by Kehler (2000; 2002), Kertz (2010; 2013), Miller (2011; 2014), Miller & Pullum (2014), or Miller & Hemforth (2014). Following Kertz (2008), Miller (2011) and Miller & Pullum (2014), we have distinguished two central uses of Post-Auxiliary Ellipsis (PAE henceforth, where VPE would be a subtype of PAE): auxiliary choice and subject choice, which would need to comply with the following conditions (Miller & Pullum 2014: 12):

Type 1: Auxiliary choice

Formal characteristics: The subject of the antecedent is identical with the subject of the PAE construction and the auxiliary is (at least weakly) stressed, signalling a new choice of tense, aspect, modality, or (in the most overwhelmingly frequent case) polarity.

Discourse requirement: A choice between the members of a jointly exhaustive set of alternative situations must be highly salient in the discourse context, and the point of the utterance containing the PAE is strictly limited to selecting one member of that set.

Type 2: Subject choice

Formal characteristics: The subject of the antecedent is distinct from the subject of the PAE construction, and stressed if it is a pronoun.

Discourse requirement: A particular property must be highly salient in the discourse context, and the point of the utterance containing the PAE must be strictly limited to identifying something or someone possessing that property.

These two different types of focus (marked in italics) are exemplified by Miller & Pullum (2014) in instances (6) and (7), respectively:

As can be observed, in (6) the focus is on the auxiliary (doesn’t), marking the change of polarity in the ellipsis site with respect to its antecedent (as the subject of both clauses is the same). In the examples provided in (7), on the other hand, one can observe that the focus is placed on the contrast of subjects between the antecedent and the target of ellipsis (i.e., the boys vs. I and she vs. I). The role played by these discourse conditions is relevant for this paper, since, as mentioned in the literature, more mismatches seem to be allowed in the auxiliary choice condition (see Kertz 2013; Miller & Pullum 2014) between the source and the target of ellipsis. This entails that gender mismatch, the object of study of this paper, could play a (small) role, given that the materials that will be tested are all subject choice, that is, the focus is placed on the contrast between the subjects of the antecedent and the ellipsis site (see section 4.2 for more details).

In the following subsection, we will provide an overview of the two questions that have been traditionally investigated in the ellipsis literature, i.e., the structure and the identity questions.

2.2 Ellipsis and gender (mis)matches

Two of the long-standing debates in the study of ellipsis revolve around trying to decipher whether there is structure at the ellipsis site that needs to be computed (known as the “structure question” in the literature) and determining the type of relation held between the elided material and its antecedent (also known as the “identity question”). In the case of the former, the two possible answers that could be given to this question have crucial consequences for the theory of grammar, since, if the answer is affirmative, the theories of grammars must posit the existence of unpronounced phrases and heads. There is a vast literature on ellipsis dedicated to arguing for some evidence that would support either the structural or the non-structural stances, and this has obviously been done indirectly: if there are effects on the surrounding material which may be attributed to elided elements — the so-called ‘connectivity effects’ —, then one can claim that there is syntax internal to the ellipsis site; if, on the contrary, ‘non connectivity effects’ between the elided material and the surrounding material are found, then there is evidence to posit that there is no syntactic structure internal to the ellipsis site, i.e., ‘what you see is what you get’. In Merchant’s words (2019: 25), “if effects are found which we would otherwise attribute to properties of structure X in similar, nonelliptical, cases, but structure X is, by hypothesis, internal to the ellipsis site, then X exists. If, on the other hand, expected properties are missing, one could conclude that structure X is absent”. Those in favour of a structural approach (Sag 1976; Williams 1977; Fiengo & May 1994; Chung et al. 1995; Kehler 2002; Sag & Hankamer 1984; Merchant 2001; van Craenenbroeck 2010; Aelbrecht 2010; among others) have shown that there exist some connectivity effects between the antecedent and the ellipsis site (locality effects, case matching, preposition- stranding, the distribution of complementizers, infinitivals and predicate answers), while those defending a non-structural approach (known as the direct interpretation) (Keenan 1971; Dalrymple et al. 1991; Hardt 1993; Ginzburg & Sag 2000; Culicover & Jackendoff 2005; among others) have provided some counter-evidence illustrating the absence of unpronounced structure (absence of locality effects, exceptions to the P-stranding generalization, voice and category mismatches, violation of binding principles, lack of case-matching effects, pragmatic antecedents, non-extractable remnants, etc.).³

Concerning the identity question, most of the debates have focused on determining the type of recoverability condition that is at stake in cases of ellipsis, since omitted elements need to be recoverable from the (either linguistic or extralinguistic) context in order for them to be elided. Hankamer & Sag (1976) proposed a distinction between deep (forms such as do it, do this, do that, do so) and surface (e.g., VPE) anaphora putting forth that the former does not require a syntactic parallelism and it can be exophoric, whereas the latter does and therefore it can never be exophoric. However, as shown above, research has revealed that exophora and voice mismatches are indeed possible with surface anaphora such as VPE (see, among others, Webber 1979; Dalrymple et al. 1991; Hardt 1993; Kehler 2000; Jacobson 2003; 2008; 2009; Kertz 2010; Miller & Pullum 2014). Traditionally, several identity relationships between the antecedent and the ellipsis site have been proposed in the literature: semantic (Dalrymple et al. 1991; Hardt 1993; Ginzburg & Sag 2000; Merchant 2001; Merchant 2004; Culicover & Jackendoff 2005; van Craenenbroeck 2010; Aelbrecht 2009; 2010; Thoms 2010; 2013), syntactic (Sag 1976; Williams 1977; Fiengo & May 1994; Chung et al. 1995) or both (Chung 2006; 2013; van Craenenbroeck 2010; Merchant 2013).⁴ In this regard, research has shown that syntactic parallelism seems to play a role in constraining the relation held between the elided material and its antecedent.⁵ For instance, antecedent and ellipsis morphosyntactic mismatches can result in ungrammaticality when be and the auxiliary have are involved, as observed in the examples provided by Lasnik (1999) in (8):

The ungrammaticality of examples (8a) and (8b) stands in contrast with the grammaticality of the matched control examples in (8c) and (8d), which demonstrates that the lack of morphosyntactic isomorphism must be the reason why (8a) and (8b) are judged unacceptable.

However, it must be noted that absence of morphosyntactic parallelism, e.g., gender mismatch, seems to be grammatical with nominal ellipsis (Merchant 2014; Sudo & Spathas 2020; Kučerová et al. 2021; Sprouse et al. 2022) and predicative ellipsis (Lasnik 1999; Depiante & Masullo 2001; Nunes & Zocca 2005; 2009; Bobaljik & Zocca 2011; Merchant 2014; Aparicio et al. 2015) in several languages such as English, Spanish, Brazilian Portuguese, Russian, German or Greek. What seems to be at stake in these languages is that inflectional features may not be relevant for ellipsis computation. For instance, Merchant (2014: 10) claims that the example from Greek illustrated in (9) is grammatical despite containing a gender mismatch between the ellipsis site and its antecedent.

In order to shed some light on this particular issue, Aparicio et al. (2015) tested the effect of gender (mis)matches in Spanish stripping (see example (10)), another similar construction which contains the neuter particle ‘lo’ (i.e., a deep anaphor, see example (11), emphasis in the original), and non-elliptical sentences thanks to two eye-tracking experiments that involved 28 and 21 L1 Spanish participants, respectively. Subjects had to perform a yes-no grammaticality judgement about the sentence they had just read.

Their results showed that, in general, there was a high acceptability of gender mismatches (≥80%). However, participants were highly sensitive to morphological feature mismatch, as there were degraded acceptability judgments and elevated reading times for elliptical mismatched conditions. Interestingly, these authors also found that the ordering of the gender mattered too: the ordering feminine-masculine was judged less acceptable than the ordering masculine-feminine in contexts where there was a mismatch. Aparicio et al.’s (2015) results also pointed to the fact that mismatches were judged more acceptable in cases of deep anaphora as opposed to surface anaphora (stripping) (which is in line with Hankamer & Sag 1976) and that the gender mismatch penalty was completely absent for non-elliptical counterparts (Aparicio et al. 2015: 2). Thus, their conclusion was the following: “Elliptical sentences containing conflicting inflectional phi-features are more costly to process than elliptical sentences without feature mismatch” (Aparicio et al. 2015: 2). That is to say, morphological features would be relevant for ellipsis computation.

Abeillé & Kim (2022), on their part, conducted a corpus-based study on English and French additive fragments, a type of stripping except for the presence of an additive adverb (too/aussi), by analysing data after 1980 from well-edited corpora such as the COCA (Davies 2008) and Frantext (www.frantext.fr). As these authors mentioned, since Hankamer & Sag (1976), stripping had been considered a surface anaphor which was more constrained than deep anaphors. However, they challenged this idea by providing empirical data showing that additive fragments have “much more flexibilities in their uses than previously thought and even allow for non-local antecedents and exophoric uses” (Abeillé & Kim 2022: 498). The following would be an example of exophoric stripping in English, where the context functions as the non-verbal antecedent (Abeillé & Kim 2022: 511):

Crucially for the purposes of the present paper, these authors also pointed to the fact that gender mismatch is possible in French stripping (Abeillé & Kim 2022: 502), as exemplified in (13):

Their analysis of stripping constructions in both languages led the authors to propose a direct interpretation approach (that is, a non-structural one in which ‘what you see is what you get’) along the lines of Ginzburg and Sag (2000), thus defying previous accounts based on syntactic reconstruction and coordination.

Finally, Sprouse et al. (2022) analysed gender mismatches in nominal ellipsis in order to probe the ellipsis test put forward by Bobaljik & Zocca (2011) for the gender asymmetries that seem to arise in English noun pairs such as actor/actress, waiter/waitress, etc. (asymmetric nouns) vs. prince/princess, king/queen, etc. (symmetric nouns), as illustrated in examples (14) and (15) (Sprouse et al. 2022: 349–350):

As can be observed, in (14) these nouns evince an asymmetry under ellipsis due to the fact that the unmarked form (actor) can be the antecedent of a female elided predicate, whereas the marked form (actress) cannot be the antecedent of a male elided predicate. In (15), however, both noun orderings result in unacceptability. As Sprouse et al. (2022: 349) put it, “[t]he underlying idea of the ellipsis test is that the identity requirement on ellipsis can be leveraged to uncover these asymmetries, and crucially, convert those asymmetries into unacceptability”. Thus, these authors administered an acceptability judgement test to 2,612 English native speakers based on 7-point scale task containing sentences such as the ones shown in (14) and (15), as well as their non-elliptical (e.g., John is an actor and Mary is an actor too) and gender-matched counterparts (e.g., John is an actor and Bill is an actor too). The results of their study showed that the expected differences between asymmetric and symmetric nouns were detected and could be accounted for without resorting to ellipsis, that is to say, ellipsis was not a requirement for mismatches to be tolerated (contra Bobaljik & Zocca 2011). It is important to note that, in the case of asymmetric noun pairs, these authors found that, independently of ellipsis, the masculine-feminine ordering (John is an actor and Mary is too and John is an actor and Mary is an actor too) was preferred over the feminine-masculine ordering (Mary is an actress and John is too and Mary is an actress and John is an actress too). This preference for masculine-feminine gender orderings would be in line with Aparicio et al.’s (2015) findings for Spanish stripping mentioned above. In addition, Sprouse et al. (2022) discovered that there was a significant increase in the acceptability of feminine-masculine orderings in elliptical constructions for two of the noun pairs under study (queen/king and widow/widower).

2.3 Reflexive anaphora and gender (mis)matches

In the past three decades, psycholinguistic research has drawn its attention towards the interplay between reflexive anaphora and VPE and the two possible interpretations that may arise: strict or sloppy (Shapiro & Hestvik 1995; Frazier & Clifton 2000; Shapiro et al. 2003; Frazier & Clifton 2006). For instance, Shapiro & Hestvik (1995) run two experiments that studied the on-line processing of coordinated and subordinated VPE containing reflexive anaphora, as illustrated in (16) and (17), respectively (Shapiro & Hestvik 1995: 517):

Their aim was to try to decipher, thanks to a priming technique, if listeners (n = 60 college students) automatically access the strict reading immediately after reading the bare auxiliary by showing reaccess of the subject noun phrase in the antecedent clause. Their findings showed that depending on the type of relation established between the two clauses, different time courses of activation emerged: in coordinated VPE, gap filling of the strict reading was immediate, whereas in subordination, this only occurred “later during the temporal unfolding of the sentence” (Shapiro & Hestvik 1995: 530). As Frazier & Clifton (2006) state, these findings suggest that “the strict interpretation (“the fireman defended the policeman”) is computed, at least temporarily, in sentences like (16), where the strict interpretation is clearly unpreferred”. Interestingly, this has been claimed to occur even with sentences such as (18), where the strict interpretation would not be possible due to the nature of the verb (Shapiro et al. 2003: 6):

Frazier & Clifton (2006), on their part, tested Kehler’s (2000; 2002) claim that a reflexive may receive a strict interpretation only if a causal discourse coherence relation is established (as in example (19a), but not with a resemblance coherence relation (as in (19b)) (Frazier & Clifton 2006: 8). In other words, according to Kehler, Principle A would need to be violated in sentences involving causal relations, but obeyed in those involving resemblance ones.

The prior findings reported by Shapiro & Hestvik (1995) and Shapiro et al. (2003) would suggest that the strict interpretation is computed even in sentences involving a resemblance coherence relation and even for sentences which may not receive a strict reading in the end. Frazier & Clifton (2006) found that participants (n = 48 University of Massachusetts students) responding to a forced-choice task with the two possible interpretations (strict or sloppy) chose strict readings in both causal (53%) and resemblance (48%) sentences, and the differences attested were not statistically significant, which means that “there was no substantial difference in the probability of assigning a strict interpretation to a reflexive in VP ellipsis in cause-effect relations vs. in resemblance relations” (Frazier & Clifton 2006: 8). Therefore, these findings disconfirm Kehler’s (2000; 2002) hypothesis that a strict interpretation would only be licensed in causal relation sentences.

There have also been some studies that have analysed the interplay between English reflexive anaphora and gender (mis)matches. For instance, Runner & Head (2014) focused their study on how the grammatical constraints of the syntactic Binding Theory, that is, the structural constraints on reflexives and pronouns, apply during on-line processing. They used a novel visual world eye-tracking method which manipulates the gender of potential antecedents visually and they found clear evidence that native speakers of English (n = 25) consider gender-matching potential antecedent NPs for reflexives and pronouns that match in gender regardless of whether they are licensed structurally by the Binding Theory. That is to say, in examples such as (20) and (21), where there are inaccessible antecedents in the relative clauses that appear in between brackets, the participants in the study tended to compute Molly as a possible antecedent of drove herself because it matches in gender. However, this was not the case with Darrin, a mismatched inaccessible antecedent (Runner & Head 2014: 8):

Storoshenko (2017), on the other hand, studied the interpretation made by L1 speakers of English (n = 40) of reflexives in three different contexts (object self, by-phrase self and adverbial self, as shown in (22a–c), respectively) and one-anaphora (as in (22d) in cases of VPE with gender (mis)matches. In his work, participants had to perform a contextual felicity task wherein target sentences with VPE were presented together with a short text giving background information about the situation. The subjects taking part in the study were then asked to rate on a 7-point Likert scale the degree to which the target matched the context using sentences such as the ones in (22), which contain gender mismatch conditions:

Interestingly, Storoshenko (2017) found that, among the three reflexive cases explored, participants favoured the sloppy interpretation and that there were no gender mismatch effects. More precisely, in the case of object self sentences, which are the ones that resemble the experimental items in our study, the results showed that in both gender match and mismatch conditions, native speakers of English significantly (p < 0.001) preferred the sloppy reading (6.09 for both match and mismatch) over the strict one (3.71 for match and 4.09 for mismatch). However, in the case of one-anaphora constructions, his findings suggested that native speakers of English do not make a strict/sloppy distinction, except for a slight drop on the sloppy gender mismatch condition (which he attributed to a reduced number of participants systematically rejecting those target sentences, that is to say, assigning ratings of 1–2 while other participants assigned 6–7).

2.4 L2 acquisition of reflexives and pronoun gender agreement

The last section of the literature review will broach L2 studies on the acquisition of reflexive pronouns, pronoun gender agreement and their interplay with learners’ proficiency.

It is well-known that gender mistakes in L2 English pronoun production are very frequent. Antón-Méndez (2010), for instance, observed that Spanish speakers exhibited significantly more he/she confusions than any other type of pronoun errors (e.g., number errors, person errors, omission errors). Gender agreement seems to be especially problematic when English learners need to decide the type of possessive determiner that goes with the following noun, as in (23), where we have “The mother”, which is feminine gender, and then, following, “boy”, which is masculine. Thus, a common mistake would be to say “his boy” when the learner would mean “her boy”.

In addition, some works have shown positive (Apaloo & Cardoso 2022 for Brazilian) and negative (Muñoz Lahoz 1994 for Spanish; White et al. 2007 for French and Catalan/Spanish; Imaz-Aguirre 2015 for Basque/Spanish) L1 transfer effects. What these studies have found is that pronoun gender agreement seems to improve with cross- linguistic awareness for positive transfer, and with target language proficiency for negative transfer.

Also, the findings of L2 studies on gender agreement in L2 reflexives (see, for instance, Felser & Cunnings 2012 for German; Liang et al. 2018 for Chinese) have shown that, when assigning gender, native speakers and learners behave differently because the former have no problems in this regard since they follow principle A of the Binding Theory. However, in the case of L2 learners, their interpretations show that they have not acquired Principle A of the Binding Theory and do not bind the reflexive locally. This appears to be so because they prefer to resort to discourse-based coreference assignment. This is illustrated in example (24) from Felser & Cunnings (2012), where L1 German learners of English tend to bind the pronoun she (underlined) with the reflexive pronoun herself.

In this regard, it would be worth mentioning for the purposes of this paper that, as has been shown in the literature (Thomas 1989; 1991; 1993; Padilla, 1990; Eguren 2013), both Spanish and English languages bind reflexive pronouns locally, unlike other languages such as Chinese or Japanese, where both long-distance and local binding would be possible. Therefore, one may assume that Spanish speakers’ knowledge about English binding conditions would be similar to English native speakers’. Paradoxically, Thomas (1989) found no statistically significant differences between Chinese- and Spanish-speaking groups. What is more, Chinese learners exhibited a more similar performance to English native speakers as compared to Spanish learners, as the Chinese group bound 69% of the reflexives locally, whereas the Spanish one produced 60% of local responses. In other words, Spanish learners of English did bind the reflexives long-distance even if it was not expected either in their L1 or L2.

There have also been some L2 studies that have analysed the effect that target language proficiency may exert on the acquisition of pronouns and ellipsis so as to check whether more advanced learners’ behaviour resembles that of native speakers more than less proficient learners’ behaviour. Research has revealed that proficiency has an impact on the acquisition of various aspects of L2 pronouns, for instance, cliticisation (Scuitti 2020), gender agreement (Dong et al. 2015) and anaphora resolution (Contemori et al. 2019). Interestingly, concerning reflexive pronouns, the effect of this variable has shown that the higher the L2 learners’ proficiency, the better their acquisition of reflexives (Yip & Tang 1998). In other words, higher proficient learners are capable of treating the binding properties of the L2 reflexives as an independent system from their L1.

Finally, with regard to the literature that has broached the interplay between proficiency, ellipsis and the interpretation of reflexives, research has revealed that higher-proficiency learners’ readings resemble more those by native speakers than lower-proficiency learners’ ones (Ying 2005; Epoge 2012; Park 2016; Gallardo-del-Puerto & Gandón-Chapela 2024). As will be briefly described below, Ying (2005), Park (2016) and Gallardo-del-Puerto & Gandón-Chapela (2024) investigated L2 English learners’ readings of reflexives in bare, referential, and non-referential contexts (instantiated in examples (25)–(27), respectively), whereas Epoge (2012) studied only the contrast between bare and referential contexts.

Crucially, examples (25) and (27) are ambiguous as to whether VPE would be interpreted strictly (Bill defended John) or sloppily (Bill defended Bill). However, in the case of (26), the contextual information offered biases the reading towards a strict interpretation.

In Ying’s (2005) study the participants were 50 L1 Chinese university learners of English (intermediate (n = 28), advanced (n = 22)) studying English majors in Shangai and a control group of 20 native speakers of American English. The findings of this work show that in bare contexts all participants, including the native speakers, interpreted the VPE sloppily (ranging from 74.1% to 82.1%). However, intermediate learners were the ones who did so to a larger extent, the differences attested across the two learner groups and between the intermediate learners and the control group being statistically significant. Nevertheless, no statistically significant differences were found between the advanced learners’ group and the control one. Concerning referential contexts, both intermediate and advanced learners favoured the strict interpretation, and so did the native speakers (from 64.9% to 82.1%). This time the differences found among the three participant groups were statistically significant. Finally, in the case of non-referential contexts, neither the learner groups nor the native speakers showed a marked preference for either of the two possible readings, as no statistically significant differences were attested (from 52.5% to 55.8% for the sloppy reading).

Park (2016) replicated Ying’s (2005) study with 26 fourth-year university students in Busan (South Korea) (intermediate (n = 13), advanced (n = 13)) and a control group of 10 native speakers of English. As far as bare contexts are concerned, intermediate and advanced learners of English favoured the sloppy interpretation similarly (91.7% and 88.5%), which was significantly different from that of natives, as they did so to a lesser extent (79.2%). Regarding referential contexts, the control group showed a slight preference for the strict reading (57.5%), the intermediate learners showed a clear preference for the sloppy one (64.1%), and advanced learners chose either interpretation similarly (sloppy 49.4% vs. strict 50.6%). In this case, the interpretations made by intermediate learners were significantly different compared to the two other groups, since no differences were found between the advanced and the control groups. Finally, in the case of non-referential contexts, the sloppy interpretation was the preferred one by all of the groups (ranging from 78.2% to 94.2%), especially by native speakers and advanced learners. One more time, no statistical differences were found between the control and advanced groups, while the intermediate one significantly differed from them.

Gallardo-del-Puerto & Gandón-Chapela (2024), on their part, provided data from 104 Spanish learners of English (A2 (n = 32), B1 (n = 37), and B2 (n = 31) levels according to the Common European Framework of Reference for Languages) from the University of Cantabria (Spain) and a control group of 32 native speakers of British or American English. Their results revealed that participants preferred the sloppy interpretation of reflexives in bare (ranging from 66.4% to 84.4%) and nonreferential (from 58% to 81.2%) contexts, while strict readings were predominant in referential ones (from 70% to 92.2%). They also reported significant differences in the interpretation of learners when compared to native speakers, the differences among the three learner groups studied not being so marked. However, the group with the lowest proficiency level distanced itself most from native speakers. These findings partially confirmed previous research and the authors attributed the discrepancies attested to extraneous variables such as the learners’ L1, the range of the proficiency levels under scrutiny, or the characteristics of the control groups that took part in the studies.

Lastly, Epoge (2012) conducted a similar study with 128 L2 learners of English enrolled in the Department of English at the University of Yaounde I (Cameroon) at Level One (n = 40), Level Two (n = 45) and Level Three (n = 43). In this case, no control group of native speakers was included. His work focused only on two of the contexts mentioned so far: bare and referential. Concerning bare contexts, the higher the proficiency level, the more favoured the sloppy interpretation was (ranging from 53.1% to 68.6%). On the other hand, in referential contexts the opposite tendency was attested since strict interpretations increased as the level of proficiency raised (from 78.3% to 91.9%). It should be noted, however, that in this work no inferential statistics was performed on the data.

What can be gathered from the L2 literature review offered is that one variable that has been unexplored is the effect of gender (mis)match on the readings of English reflexive pronouns (as either strict or sloppy) in cases of VPE (see Storoshenko 2017 for L1 speakers of English). Additionally, the fact that Spanish and English bind the reflexives locally (Thomas 1989; 1991; 1993; Padilla, 1990; Eguren 2013), but Spanish learners of English have been found to disregard their L1 knowledge and sometimes bind the reflexives long-distance (Thomas 1989) would justify a further look into both English native speakers and Spanish learners of English.

4.1 Participants

Fifty-two Spanish-speaking University of Cantabria (Spain) students learning English and 32 native speakers (NS henceforth) of British or American English participated in this study (22 females and 10 males, average age 38.9, and age range 19–71). The non-native participants were students that were taking BA degrees in Early Childhood and Primary Education Teacher Training, as well as MA students of Second Language Learning and Teaching. As in previous research (see Ying 2005; Xiao 2015; Gallardo-del-Puerto & Gandón-Chapela 2024), a placement test was used in order to check the learners’ proficiency levels, i.e., the Quick Placement Test (Oxford University Press). Traditionally, there have been four types of proficiency determinants in the field of applied linguistics: standardised tests scores, course or grade levels, length of formal instruction, and duration of residence in the target language community. However, according to Xiao (2015), since standardised test scores allow for comparison across test takers, they are among the most reliable indicators of proficiency. Therefore, in line with this view, we have decided to use the Quick Placement Text (OUP) in its forty-item version, which measures students’ proficiency up to C1 level of the Common European Framework of Reference for Languages (CEFR). The results of this English proficiency test indicated that our participants’ level of English ranged between B1 and C1 levels. According to the Oxford Placement Test (OPT henceforth) assessing guidelines, a score of 24–30 indicates a B1 level, one of 31–35 a B2 level and one of 36–40 a C1 level. Three different proficiency groups were attested among the participants of this study. The first group was formed by 19 B1-level students (18 females and 1 male) whose average age was 22.21 (age range: 20–28) and OPT mean score was 28.15 (24–30). The second group was formed by 22 B2-level students (14 females and 8 males) whose average age was 22.9 (age range: 20–44) and OPT mean score was 32.59 (31–35). Lastly, there was another group composed by 11 C1-level students (9 females and 2 males) whose average age was 22.45 (age range: 20–29) and OPT mean score was 37.81 (36–40). These data have been gathered in Table 1 for visual convenience:

4.2 Instruments

In this study we have used two kinds of instruments: an English proficiency test measuring one of the independent variables of the study, that is, target-language proficiency (as reported in section 4.1); and one two-alternative (sloppy-strict) forced-choice task to check the impact of gender (mis)match on participants’ interpretations of English reflexive anaphora in cases of VPE. This task contained 10 gender-matched and 10 gender-mismatched sentences as in (28) and (29), and 20 distractors. The direction of the gender mismatch (feminine antecedent-masculine target of ellipsis and vice versa) was also considered, since the gender-mismatched experimental items contained half feminine-masculine and half masculine-feminine directions. Examples (28) and (29)–(30) illustrate gender match and mismatch between the reflexive pronoun that acts as the antecedent and that of the ellipsis site, respectively. As for the direction of the gender mismatches, example (29) displays a masculine-feminine ordering, and (30) represents a feminine-masculine one. It must be noted that the gender (mis)match was indicated only by using gendered names which were known by the learners, as they were proper names of high frequency. As can be observed, the participants had to choose one of the two options given immediately below:

4.3 Data Gathering Procedure

The data were gathered in one session during the students’ regular class time. Firstly, students had to take the Quick Placement Test (OUP), which would measure their English language proficiency level. They responded to the 40 multiple-choice questions in this test in around 30 minutes. Then, participants were asked to fill in a short biographical and linguistic survey (which contained questions regarding their age, gender, years of English learning, and other languages known) and the two-alternative forced-choice task via Google forms (using either their phones or laptops), which they completed in around 10 minutes on average. In the case of the native speakers, they also completed the short biographical and linguistic survey together with the forced-choice task, but were exempt from taking the Quick Placement Test. As in previous research (Ying 2005; Gandón-Chapela & Gallardo-del-Puerto 2019; Gallardo-del-Puerto & Gandón-Chapela 2024), in the forced-choice task participants were asked to choose one of the two interpretations provided which matched their initial interpretation of the underlined part of the sentence. That is to say, the participants were asked to select the answer that came to their minds first, i.e., following their initial intuition and without going back or making any changes after their initial choice.

4.4 Data analysis procedure

Data were quantitatively analysed with the help of the Statistical Package for Social Sciences (SPSS). Concerning descriptive statistics, interpretation preferences (mean scores) and standard deviations were calculated. These figures informed about the central tendency and the dispersion of the data, respectively, as regards the interpretations of English reflexive anaphora in cases of VPE in each of the four research groups (B1, B2, C1 and NS). These figures were calculated for each of the two (mis)matched contexts separately –10 experimental sentences with gender matches and 10 experimental sentences with gender mismatches. Similarly, descriptive statistics are provided for the two gender orderings in mismatched contexts separately –5 experimental sentences with masculine-feminine orderings and 5 experimental sentences with feminine-masculine ones.

As for inferential statistics, the first analyses performed were those pertaining to normality. In that sense, Kolmogorov-Smirnov tests distinguished between those cases where parametric procedures were to be used (since a normal distribution of the samples was found), and those where non-parametric procedures should be employed (as there was at least a sample whose distribution was skewed). (Parametric) T-tests or (non- parametric) Wilcoxon signed-rank tests were alternatively computed to explore differences between the two contexts appearing in the forced-choice task sentences, i.e., gender match vs. gender mismatch, as well as between the two gender orderings, i.e., masculine-feminine vs. feminine-masculine. As for the inquiry into proficiency, (non-parametric) Kruskal Wallis one-way analyses of variance were employed to look into comparisons among the four research groups. Whenever significant differences were discovered, (non-parametric) post-hoc Mann-Whitney tests established all pertaining pairwise comparisons (B1-B2, B1-C1, B1-NS, B2-C1, B2-NS, C1-NS). Statistical probability was considered significant when alpha reached the levels of 0.05(*), 0.01(**) and 0.001(***), respectively. In addition, effect sizes were calculated, with Cohen’s d values above 0.2, 0.5 and 0.8 being considered of a small, medium and large strength of difference, respectively.