2.1 The Interface Hypothesis and some derivative hypotheses

Within the Minimalist framework, extensive research has explored the L2 acquisition of the grammatical modules and their interface with the pragmatic component. Sorace and colleagues proposed the first version of the Interface Hypothesis (IH-1) as stated in (1).

In particular, Sorace & Filiaci (2006: 340) argued that the developmental optionality attested in L2 acquisition may be the result of underspecification at the level of knowledge representation or it may stem from insufficient processing resources to integrate the multiple types of information at the interfaces. A particularly fruitful ground to test the IH-1 is the syntax of pronominal subjects. In the null subject languages, null subjects are syntactically licensed but their distribution is pragmatically determined. Mastery of pronominal subjects in the null subject languages therefore requires both the correct representation of the syntactic licensor (i.e. the correct setting of the null subject parameter) and knowledge of the pragmatic interface conditions that govern the felicitous use of null or overt subject pronouns in context. Indeed, most work in this framework has explored anaphora resolution, i.e. the way in which second (and subsequent) reference to an entity is grammatically encoded in discourse.

Sorace & Filiaci (2006) examined anaphora resolution in bi-clausal sentences in which a null or overt pronoun appears in the subordinate clause. In forward anaphora, in which the main clause precedes the subordinate clause, the processor encounters two referential expressions before the pronoun. In backward anaphora, which has the subordinate clause first, the pronoun is encountered prior to the mention of any other referential expressions. In forward anaphora example (2a) from Italian, the subordinate null pronoun is coreferential with the matrix subject la mamma (‘the mother’), while the overt pronoun lei (‘she’) refers to the object la figlia (‘the daughter’) of the matrix clause. However, the overt pronoun may also be coindexed with the matrix subject. Similarly in backward anaphora (2b). (Examples are taken from Sorace & Filiaci 2006: 352.)

In order to test the IH-1, Sorace & Filiaci (2006) administered a picture verification task involving the interpretation of null/overt pronouns in the context of forward and backward anaphora. The participants were 14 English-speaking near-native learners of Italian and a control group of 20 native speakers of Italian. The results showed that the near-native speakers differed from the natives in their preferences for overt pronouns in forward anaphora, as the L2 learners chose the interpretation in which the pronominal subject of the subordinate clause was coreferent with the NP subject from the matrix clause significantly more often than the natives. On the other hand, no differences were found between the two groups in the choice of referent for the null subject of the subordinate clause. In backward anaphora contexts, the L2 learners also chose the NP subject of the matrix as antecedent for the overt pronoun of the subordinate, while the natives preferred an extralinguistic antecedent. However, both groups preferred to interpret null subjects as coreferent with the NP subject of the matrix clause. Overall, the results of this study indicated that the L2 learners had problems with the interpretation of overt pronouns in relation to their antecedent in both anaphora contexts, although the learners had acquired target-like processing strategies for anaphora resolution of null subjects. Thus, the results were not fully consistent with the IH-1. This appears to be the case in various studies following the IH: while problems are attested with overt pronouns, null pronouns are target-consistent, even though both overt and null pronouns fall under the scope of the IH.

Again in the framework of the IH, Belletti et al. (2007) also studied the acquisition of pronominal reference in L1 English – L2 Italian speakers. Their overall results showed that the near-native learners produced a large number of null subjects, as in the performance of natives, but also showed a significantly higher rate of overt pronominal subjects than the natives, confirming overuse of overt pronouns in near-native discourse contexts. The authors’ interpretation of the facts was that grammatical representation was the source of the difficulty, as the results were consistent across tasks, thus indicating that processing cost was not relevant. The attribution of the problems faced by L2 learners to competence or processing has been a matter of debate from the first formulation of the IH.

In order to examine the production of subjects, Tsimpli & Sorace (2006) collected data from 10–20 minute-long oral interviews from three experimental groups of Russian learners of Greek, one intermediate-level group, one lower-advanced group and one upper-advanced group. The results showed that all experimental groups had a clear preference for null subjects, following native-like patterns. However, the L2 groups occasionally overused 1^st/2^nd person overt pronouns; the type of person had an effect on the L2 performance, as the learners showed less non-native uses of 3^rd than 1^st/2^nd person. A statistical comparison between 1^st/2^nd vs. 3^rd person pronouns revealed significant differences in all groups of learners. Still, the performance of the L2 learners was not always in line with the IH-1, since the L2 groups presented no problems with the distribution of null subjects.

Sorace (2011) introduced a more explicit version of the IH, which we focus on in our study:

Sorace (2011) specified that interface deficits affect information processing more often than representation knowledge at the highest level of ultimate attainment in L2 (Sorace 2011: 9; 2012: 213). For this, amongst other reasons, the IH has also had its detractors. Rothman (2009) and White (2009; 2011a; b), for example, question the proposal that structures at the interface levels present irresolvable instability or optionality in L2 acquisition. In fact, White (2011a: 578) argues that, even if non-target performance is due to representational or processing difficulties, this does not mean that the interface structures are permanently inacquirable. Furthermore, as White (2011a: 587) states, interfaces are not monolithic: it is not the case that all interfaces lead to difficulties and all phenomena at a particular interface are necessarily problematic, so acquisition failure can be avoided. In his version of the IH, Rothman (2009) also argues that interface problems may be eventually overcome, so that both syntactic and interface features can be finally acquired. In his research, the advanced group performed native-like in all contexts examined, confirming full command of the syntax-pragmatics interface, against the predictions of the IH. Still, in Rothman’s view, interface problems may be found at earlier stages with the pattern predicted by the IHs, as the intermediate group he tested presented non-target-like performance, showing that they had not yet acquired the pragmatic conditions regulating null and overt subject distribution.

Looking into the L2 Spanish by L1 English speakers, Lozano (2016) examined topic-continuity contexts and topic-shift contexts and found, as in much of the literature reported, that the expected null pronouns were the most frequent option in topic-continuity contexts, though the L2 advanced learners significantly differed from the native speakers in producing unfelicitous overt pronouns. In an attempt to make the prediction of the IH more specific and empirically adequate, and building on the neogricean principles of Manner and Informativeness, Lozano formulated the Pragmatic Principles Violation Hypothesis (PPVH), stated in (4).

The overt-when-null violation in topic-continuity contexts does not lead to a communicative breakdown (only to redundancy), as the anaphor can be resolved, while the null-when-overt violation leads to a communicative breakdown (ambiguity) in topic-shift contexts, as the anaphor cannot be resolved. Lozano’s stance is that pragmatic principles banning redundancy are violated more often than the principles banning ambiguity, with a differential effect depending on pragmatic context.

2.2 The distribution of subjects in Spanish and Greek

As known from Montalbetti (1986), Hernanz & Brucart (1987) and Fernández-Soriano (1989) for Spanish and Philippaki-Warburton (1987; 1989), Horrocks (1994) and Joseph (1994) for Greek, Spanish and Greek are two null subject languages that display rich verbal inflection, which shows the features of person (1^st, 2^nd, 3^rd) and number (singular, plural), exemplified in (5a) for Spanish and (5b) for Greek.

However, the two languages are not identical in all respects (Roussou & Tsimpli 2006). Syncretism in the inflectional marking of subject person renders verb morphology ambiguous in certain Spanish paradigms (Luján 1999), such as the verb ir (‘go’) in imperfect tense which could lead to the use of an overt pronoun for person disambiguation purposes regardless of the context (6a). In Greek, there is no systematic verb inflectional ambiguity (6b) which would trigger overt subject use to the same extent (Dimitriadis 1996), except for some forms of third person singular and plural (e.g. ine ‘be’). The expression of third person overt pronouns would be possible with deictic/demonstrative interpretation in Greek, differently from Spanish (Manolessou 2001).

According to Roussou (2009) Greek is more overtly productive in agreement than Spanish, since it lacks non-finite constructions unmarked for subject agreement. In (7a) for Greek, the matrix verb matheno (‘learn’) and the subordinate pezo (‘play’) are finite forms that agree in number and person with the matrix subject. On the other hand, Spanish displays non-finite constructions (infinitives), triggering reduced marking for agreement. In (7b) variation between the finite verb aprender (‘learn’) and the infinitive tocar (‘play’) is observed.

Regarding discourse relations, the coreference between null/overt subjects and their antecedents is not always identical in Spanish and Greek. A null pronoun pro is generally licensed by the most prominent referent, the antecedent subject of the structure in both languages (8a, 9a). According to some authors, an overt pronoun él (‘he’) in Spanish (8b) can retrieve both prominent and non-prominent antecedents such as the subject and object without incurring a significant processing penalty (Alonso-Ovalle et al. 2002; Filiaci et al. 2014); see, however, Gelormini-Lezama & Almor (2011; 2014) who showed a higher processing delay of an overt over a null pronoun when the coreferent antecedent (i.e. the matrix subject) is salient in Spanish, to the effect that Spanish is less flexible in the use of overt pronouns than some of the previous literature would lead one to believe. The pronoun aftos (‘he’) in Greek (9b) is preferably associated with the antecedent object in the reference structure (Papadopoulou et al. 2015).

Regarding the acquisition of subjects, it is understudied in the case of L1 Greek/L2 Spanish and L1 Spanish/L2 Greek combinations. Margaza & Bel (2006) is one of the first attempts to examine these two null subject languages (L1 Greek/L2 Spanish), following Bini’s (1993) study for L1 Spanish/L2 Italian. In their results, the intermediate group overused overt pronouns in referential contexts (as in Bini 1993), but the L2 learners achieved target patterns at advanced levels of knowledge, so that the assumed interface vulnerability did not have a permanent effect on the distribution of null subjects.

Georgopoulos (2017) explored anaphora resolution in L2 Spanish and, following Sorace’s (2011; 2012) IH-2, expected that interface deficits would be persistent even at high levels of proficiency. In this case, he examined a written corpus of production data by L1 Greek and L1 English learners at three proficiency levels (intermediate, advanced, upper-advanced). The English learners of Spanish seemed to differ from the Greek participants and the control group, as they employed 3^rd person anaphoric pronouns and noun phrases to a greater extent. On the other hand, the L1 Greek advanced and upper-advanced groups presented a relatively similar distributional pattern to the native control group. The L1 Greek intermediate learners of Spanish produced a few redundant 3^rd person pronominal subjects in same-reference contexts, though they did not show the overexplicit patterns of their L1 English peers (even the L1 English upper-advanced group produced some target-deviant 3^rd person overexplicit subjects). Nevertheless, the IH-2 was not confirmed at higher levels of proficiency, indicating that the linguistic features located at the syntax-pragmatics interface may not be problematic in the L1 Greek-L2 Spanish combination, so that interface deficits may not be permanent (Rothman 2009), depending on the facilitating role of the L1.

Lozano (2018) also examined the distribution of pronominal subjects in the L2 Spanish of Greek learners. As in Lozano (2002), he administered an acceptability judgment task with topic-continuity contexts and contrastive contexts.¹ In the results the intermediate group showed significant divergence from target native-like attainment in topic-continuity contexts, since they alternated equally between null and overt subjects. The rest of the groups (lower and upper-advanced) discriminated between null and overt subjects, but they showed divergence from the Spanish natives by mildly accepting a redundant overt pronoun. In contrastive contexts, all L2 groups discriminated between the felicitous overt pronoun and the unfelicitous null pronoun. The upper-advanced group showed full convergence with the native patterns, showing that anaphora resolution properties were eventually acquirable at very advanced levels in contrastive contexts. Deficits were more persistent at upper levels in topic-continuity contexts, in which the L2 learners showed a preference for redundant overt pronouns. Lozano (2009) had also carried out work on the language combination English-Spanish focusing on phi-features (i.e. the person and number) of the pronominal paradigm and distinguishing between 1^st/2^nd person (corresponding to the speech act participants) and 3^rd person (corresponding to an anaphoric use of the pronoun). Lozano claims that his results show that deficits at the syntax-pragmatics interface are selective and do not affect the whole pronominal paradigm in L2 Spanish, as there was robust mastery of the deictic uses of 1^st and 2^nd person, but vulnerability with the anaphoric uses of 3^rd person singular pronouns.

Recently, Giannakou (2018) studied the distribution of subjects in the L2 Greek of L1 Spanish speakers in an oral production task. The participants were 20 learners of Greek (4 basic, 5 intermediate, 5 advanced and 6 near-native) and 40 monolinguals, 20 Greek and 20 Spanish native speakers. The results showed that all groups produced a high rate of null subjects in topic-continuity contexts, though the L2 group did not achieve the pattern of native Greek speakers, regardless of their competence level, favouring the IH-2. In topic-shift contexts, all groups showed a higher preference for lexical subjects over overt pronouns, and the L2 learners did not present significant differences from the Greek control group. Thus, in this context, the native-like performance of the L2 learners was not in support of the IH-2. The study also showed that the distribution of subjects by native speakers of Spanish and Greek was not identical in various contexts examined (with the Spanish speakers producing more overt pronouns in topic-shift contexts and Greek speakers producing more null subjects in topic-continuity contexts).

3.1 Experimental design

To explore the predictions above, we designed a multiple-choice task in Spanish and the equivalent task in Greek. Each task consists of a total of 30 stimuli, 10 for each condition. 5 fillers were also included concerning the use of articles, adverbs, verbs, etc.³ The conditions combine the variables (i) person (1^st, 2^nd, 3^rd) and (ii) discourse context (topic-introduction or -continuity, topic-shift).

Condition 1 involves the distribution of 1^st and 2^nd person subjects in non-contrastive referential contexts in both Spanish and Greek. Each task contains 5 items with subjects of 1^st person (singular/plural) and 5 items of 2^nd person (singular/plural). Both 1^st and 2^nd person allow the production of null subjects in non-contrastive referential contexts, irrespective of the type of (matrix or subordinate) clause. The independent variable examined is person in a given context, giving rise to two test conditions: the 1^st person, non-contrastive referential context and the 2^nd person, non-contrastive referential context. Under the IH-2 the prediction is that L2ers will select unfelicitous sentences (i.e. with an overt pronoun); likewise, under Lozano’s PPVH, L2ers will tend to select overt pronouns when null pronouns are selected by natives.⁴ All things being equal, no difference between 1^st and 2^nd person pronouns is predicted.

To illustrate, in example (10) for Spanish, a null subject is felicitous in non-contrastive referent-introduction contexts, as the inflection of the verb dar (‘give’) shows the 1^st person singular, without creating referential ambiguity in the matrix clause.

In example (11) from Greek the non-contrastive referent-introduction context also allows the production of 1^st person null subjects due to the inflection of the verb kano (‘do’), which identifies the person and number of the referent in the matrix clause.

Conditions 2 and 3 examine the omission/expression of 3^rd person subjects in various referential contexts and, unlike in most of the literature, in Condition 2 we focus on contexts that require or favour a null pronoun. Condition 2 consists of 10 items, and two contexts are tested, both calling for a null pronoun: 5 items for referent-continuity and 5 items for referent-shift (with one antecedent). The independent variable is context of production of 3^rd person, giving rise to two test conditions: the 3^rd person (singular or plural), referent-continuity context and the 3^rd person (masculine or feminine), referent-shift context (with one antecedent). In referent-continuity contexts, the 3^rd person null subject is examined in coordinate or subordinate clauses. In referent-shift contexts (with one antecedent), the 3^rd person null subject is examined in subordinate clauses. Still, a null subject is felicitous in all the contexts examined, regardless of the type of clause. Under the IH-2, since the selection of a null pronoun is dictated by the pragmatic context, L2ers are expected to fail in their selection; likewise, for Lozano’s (2016) PPVH, L2ers will fail because of their preference for overt pronouns.

To illustrate, in examples (12a) and (12b) for Spanish, the inflection of the verbs hacer (‘do’) and precisar (‘need’) shows the third person singular, and it allows null subjects that maintain the antecedent Rosa in the coordinate clause or recover the more distant referent Ángela in the subordinate clause, respectively (whether coordinated or subordinate, the same kind of subject is licensed). In line with this, the expression of the pronominal subject ella (‘she’) would be redundant in the contexts examined.⁵

In examples (13a) and (13b) from Greek, also, the inflection of the verbs djavazo (‘read’) and parusiazo (‘present’) specifies the third person singular, and allows the omission of subjects that either maintain the adjacent referent Martha in the coordinate clause or recover the distant antecedent Ageliki in the subordinate clause. On the other hand, the overt pronoun afti (‘she’) is redundant in (13a), and can only receive deictic/demonstrative interpretation in (13b).

Condition 3 consists of 10 items that require the expression of subjects in order to recover one of two antecedent referents in topic-shift contexts (or contrastive-focus contexts). The 3^rd person overt pronoun (masculine or feminine) is used in a subordinate clause, recovering an antecedent referent presented in the introductory clause. In this case, verbal inflection is not sufficient to identify the 3^rd person of a more distant referent. As in the two previous conditions, the IH-2 predicts that L2ers will fail to select the felicitous option, due to the involvement of the syntax-pragmatics interface in selecting an overt pronoun. Differently from previous conditions, under Lozano’s PPVH, L2ers are predicted to perform better, possibly in a native-like manner, because errors giving rise to ambiguity are unexpected, and a null pronoun in the contexts tested in Condition 3 would precisely give rise to ambiguity.

To illustrate, in examples (14) from Spanish and (15) from Greek the expression of the pronouns ella (‘she’) and afti (‘she’) in the subordinate clauses is required to avoid the ambiguity between the two referents María/Jorge and Meri/Jorgos, respectively. Having a coordinated subject strongly favours the use of an overt disambiguating subject.

3.2 Subjects

Three groups of native and non-native speakers took part in each task. In the first task the intermediate and advanced groups consisted of Greek L1-Spanish L2 learners and a control group of Spanish natives. At the time of the experiment the intermediate and advanced learners were attending a Spanish language course for four hours a week at Athens’ University Institute of Foreign Languages. Both experimental groups had taken the official examination towards a Diploma in Spanish as Foreign Language (DELE). The intermediate learners had achieved an average score of 83% in the B1 Exam, while the advanced learners had achieved an average score of 88% in the C1 Exam of the Common European Framework of Reference for Foreign Languages. The native speakers of Spanish were living in Madrid and were students at the Universidad Autónoma de Madrid. Table 1 presents information on the native and non-native speakers of Spanish.

In the second task, the intermediate and advanced groups consisted of Spanish L1-Greek L2 learners and a control group of Greek natives. The L2 learners were taking classes in Greek for four hours a week at the School of Greek Language in Madrid. The two groups had taken the official examination for the Certificate of Attainment in Greek Language (CAGL); the intermediate learners had an average score of 81% in the B1 Exam, while the advanced learners had an average score of 86% in the C1 Exam. The native speakers of Greek were university students living in Patras. Table 2 provides details on native and non-native groups of Greek.

3.3 Procedure

The experiment in Spanish was administered in written form to the L2 learners at Athens’ University Institute of Foreign Languages, and to the natives at the Universidad Autónoma de Madrid. The experiment in Greek was administered in written form to the L2 learners at the School of Greek Language in Madrid and to the natives at the University of Patras. All participants were instructed to complete the respective multiple-choice task by selecting one of two options (a null or an overt pronominal subject, other options in the case of distractors) provided in the answer sheet. Participants were given an example with the felicitous option to ensure full comprehension of the task. L2 learners were also given clarifications in their L1 (Greek or Spanish) to make sure they understood the instructions fully. The time limit to complete the task was 30 minutes.

3.4 Coding of the results and statistical analysis

The responses were coded for subject type (overt, null) and for appropriateness in the context given. All felicitous and unfelicitous subjects were averaged for each group in order to make comparisons between the groups. The values of subjects were also averaged for each participant and coded in SPSS to perform non-parametric statistics, the Mann-Whitney and the Wilcoxon tests. Non-parametric statistics were chosen because of the type of data: non-parametric tests are appropriate for categorical, ordinal or rank data (see Warren 2018). The size of the sample also played a role in the preference for non-parametric tests. These tests are designed for small-size samples (n ≤ 30) (see Corder & Foreman 2009). The analyzed samples are random and independent samples that have free distribution or behave heterogeneously (see also Hollander et al. 2014). In our analysis, the Mann-Whitney test measured the significant differences between independent paired groups of L2 learners and native speakers (intermediate-advanced, intermediate-control and advanced-control). The aim of this test was to determine the effect of the independent factor of Group on the dependent factor of Subject distribution (i.e. null/overt subjects) in the conditions examined. The effect of Language on the performance of native and non-native groups was also measured. The interaction of Person and Group, or Context and Group on the selection of Subjects was measured by means of the Wilcoxon test for the three groups of each task.

3.5 Results

Each multiple-choice task yielded a total of 2,700 responses (900 responses from each group). In the first condition of non-contrastive referential contexts, the non-native groups, like native speakers, showed a high preference for the omission of 1^st/2^nd person subjects in both Spanish and Greek. However, in the task for L2 Spanish the intermediate group selected null subjects less often than the other groups, showing higher divergence from native-like patterns. Even the advanced learners of Spanish did not reach the rates of native speakers. On the other hand, in the task for L2 Greek both experimental groups followed the behaviour of the control group, irrespective of their competence level. The frequency and percentage rates and the Standard Deviation of the groups are shown in Table 3 for both languages.

The results are graphically represented in Figure 1. According to the statistical analysis, in the Mann-Whitney test for the Spanish data, the differences between the intermediate and advanced groups (U = 257.000, p = .009), the intermediate and control groups (U = 107.000, p < .003) and the advanced and control groups (U = 263.000, p < .003) were significant with Bonferroni correction. Thus, the effect for Group was significant in all cases. On the other hand, in the Greek data, the differences between the intermediate and advanced groups, the advanced and control groups, and the intermediate and control groups were not found to be statistically significant. The differences between Spanish and Greek native speakers were significant (U = 295.000, p = .002) in the Mann-Whitney test, while the differences between the L2 groups of Spanish and Greek were not.

Figure 1

Subjects of 1^st/2^nd person in non-contrastive referential contexts.

Turning to the results by Person, all experimental and control groups preferred null subjects more often than overt subjects with both 1^st and 2^nd person. However, the non-native groups selected null subjects of 1^st more often than 2^nd person. The advanced groups of both languages showed more native-like patterns, despite some differences between the advanced learners of Spanish and native speakers. Even the control groups did not present indistinguishable patterns in the two languages, as Spanish natives allowed a higher rate of null subjects with both 1^st/2^nd person than Greek natives, who selected a few more overt subjects of 2^nd person. Results for native and non-native groups are shown in Table 4.

In the Mann-Whitney test for the results on Spanish, the differences between the intermediate and advanced groups (U = 313.000, p = .024) and the intermediate and control groups (U = 270.000, p < .003) were significant with Bonferroni correction for 1^st person subjects, but they were not for the advanced and control groups. In the case of 2^nd person, the differences between all paired groups (intermediate-advanced: U = 255.500, p = .006; intermediate-control: U = 122.000, p < .003; advanced-control: U = 280.000, p = .003) were significant. The Wilcoxon test also indicated that there was an interaction of Person and Group for the intermediate (Z = –4.103, p < .001) and advanced learners (Z = –2.491, p = .013), but not for the control speakers. On the other hand, for Greek, the Mann-Whitney test with Bonferroni correction showed that the differences between all paired groups were not significant for 1^st and 2^nd person subjects. The Wilcoxon test indicated that there was an interaction of Person and Group for the intermediate (Z = –2.257, p = .024) and control speakers (Z = –2.070, p = .038) and close to the limit rate (Z = –1.913, p = .056) for the advanced learners. The Mann-Whitney test also showed that the differences between Spanish and Greek native speakers were significant for both 1^st (U = 390.000, p = .040) and 2^nd person (U = 326.000, p = .009). Thus, Language had an effect on the performance of natives. The results for the two languages are graphically represented in Figures 2 and 3.

Figure 2

Subjects in non-contrastive referential contexts in Spanish.

Figure 3

Subjects in non-contrastive referential contexts in Greek.

In unambiguous referent-continuity and -shift contexts of 3^rd person (with one antecedent) (Condition 2), the non-native groups of both L2s also accepted null subjects, like native speakers. However, the intermediate groups did not reach the rates of the natives, showing a more pronounced tendency to the expression of overt subjects. On the other hand, the advanced group of L2 Greek showed native patterns. Native speakers also presented distinguishable patterns, as Spanish natives selected more null subjects than Greek natives, who accepted the expression of 3^rd person pronominal subjects more often. See Table 5 for the overall rates of all groups in both referent-continuity and -shift contexts (with one antecedent) in both languages.

In the Mann-Whitney test for the data in L2 Spanish (see Figure 4), the differences between the intermediate and control groups (U = 186.000, p < .003) and the advanced and control groups (U = 232.500, p = .003) were statistically significant, while the differences between the intermediate and advanced groups were non-significant with Bonferroni correction. Thus, the effect for Group was significant for the intermediate-control and advanced-control groups. As for the data in L2 Greek (again see Figure 4), the statistical differences were close to significance for the intermediate and control groups (U = 293.500, p = .057), while they were not for the advanced and control groups and the intermediate and advanced groups with Bonferroni correction. Thus, the effect for Group was marginally significant for the intermediate-control comparison, but not for the rest of the comparisons. Regarding the Language effect, the differences between the Spanish and Greek native speakers were significant (U = 267.000, p = .005) in the Mann-Whitney test.

Figure 4

Subjects of 3^rd person in unambiguous referent-continuity and -shift contexts.

The interaction of Context and Group was also examined in the case of unambiguous referents of 3^rd person. Although the null subject option is overall the one preferred by native speakers in this condition, all non-native groups preferred 3^rd person null subjects more often in referent-continuity than in shift contexts (with one antecedent). As for referent-shift contexts, all non-native groups presented variation between null and overt subjects, though the intermediate group of Greek selected overt subjects more often than the other groups. The two experimental groups of Spanish performed unlike the native speakers, while the advanced group of Greek performed like the control group. In any case, the two native groups did not exclude overt subjects in the sentences examined, but the Greek native group preferred a higher rate of 3^rd person overt subjects in shift contexts. The frequency and percentage rates of all groups in each context are shown in Table 6.

In the Mann-Whitney test (with Bonferroni correction) for Spanish (see Figure 5), the differences between the intermediate and advanced groups (U = 280.000, p = .003) and the intermediate and control groups (U = 255.000, p < .003) were statistically significant for referent-continuity contexts, but they were not for the advanced and control groups. In the case of referent-shift contexts (with one antecedent), the intermediate and control groups (U = 223.000, p = .003) and the advanced and control groups (U = 234.500, p = .003) presented significant differences, while the intermediate and advanced groups did not. As for Greek (see Figure 6), the differences between intermediate and advanced groups were significant (U = 314.500, p = .036) in referent-continuity contexts, but they were not for the rest of paired comparisons. In referent-shift contexts, the intermediate and control groups showed significant differences (U = 289.000, p = .045), while the other paired groups did not. The Wilcoxon test indicated that there was an interaction between Context and Group for both the intermediate groups (Z = –3.763, p < .001 in Spanish and Z = –3.796, p < .001 in Greek), the advanced groups (Z = –4.587, p < .001 in Spanish and Z = –3.341, p = .001 in Greek) and the control groups (Z = –3.896, p < .001 in Spanish and Z = –4.055, p < .001 in Greek). Regarding the Language effect, the Mann-Whitney test showed that the differences between Spanish and Greek native speakers were significant for both referent-continuity (U = 330.000, p = .003) and -shift contexts (U = 283.000, p = .011).

Figure 5

Subjects of 3^rd person in unambiguous referential contexts in Spanish (with one antecedent).

Figure 6

Subjects of 3^rd person in unambiguous referential contexts in Greek (with one antecedent).

In the results of Condition 3, of contrastive referent-shift contexts, on the other hand, all non-native groups selected a high rate of 3^rd person overt subjects, like the control groups, recovering the more distant of two antecedent referents. However, the intermediate group of Spanish learners preferred overt pronouns less often than the advanced and control groups, while both experimental groups of Greek performed like the control group, though the intermediate group selected a few more null subjects. See Table 7 for the scores and the SD in contrastive referent-shift contexts (with two antecedents) in both L2s.

The Mann-Whitney test (with Bonferroni correction) for the data in L2 Spanish (see Figure 7) showed significant differences between the intermediate and control groups (U = 200.500, p < .003) and the intermediate and advanced groups (U = 276.500, p = .027), but no significant difference between the advanced and control groups. Therefore, the effect for Group was significant for the intermediate-control groups and not for the advanced-control groups. Regarding the data for L2 Greek (in Figure 7), though, none of the differences between the intermediate and advanced groups, the intermediate and control groups and the advanced and control groups reached significance in the Mann-Whitney test. Thus, the effect for Group was non-significant for the comparisons examined. As for the Language effect, the Spanish and Greek native speakers did not present significant differences in their responses.

Figure 7

Subjects of 3^rd person in contrastive referent-shift contexts (with two antecedents).