Learning aspect in Italian as additional language. The role of second languages

4.1 Models of additional language learning

Unlike true L2 acquisition where transfer^[6] can only derive from the L1 (Odlin 1989: 27), in Ln learning transfer phenomena can originate in both the L1(s) and the L2(s) (e.g., Cenoz 2001; Ringbom 2001; Williams and Hammarberg 2009). Another difference between true L2 acquisition and Ln acquisition is that they entail qualitatively different processes of language appropriation, as a multilingual speaker, in contrast to a true L2 learner, already has experience of one or several non-native languages (Bardel and Falk 2012; Bardel and Sánchez 2017; Hammarberg 2001; Hammarberg 2018; Hufeisen 2018; Jessner 2003). The multilingual speakers’ experience as language learners provides them with a set of specific competencies. For instance, they have generally a high degree of metalinguistic knowledge, i.e., conscious knowledge about formally learnt languages (Paradis 2009: 39), and crosslinguistic awareness, meaning that they are generally more capable of detecting similarities and differences between any formerly known languages and the Ln than bilingual speakers (Jessner 2006: 116). These competencies can all be strategically exploited in approaching a new language, where knowledge of multiple languages is involved (Jessner 2003).

Models of Ln learning emphasize different factors as responsible for the activation of certain linguistic sources in a multilingual speaker. This section focuses on the role of proficiency, typological proximity and the L2 status factor in Ln learning. However, in the present study, the two latter factors are not investigated separately. Instead, we focus on the interplay between typological proximity and the L2 status. We conclude the section by briefly presenting two recent models of Ln learning, namely the Scalpel Model (Slabakova 2017) and the Linguistic Proximity Model (Westergaard 2021), both models assuming that transfer is not a holistic phenomenon but it occurs property-by-property.

The proficiency factor for transfer entails both previously learnt L2s and the TL. As for the L2s, a high level of proficiency in the L2 is generally assumed to facilitate its transferability into a new language (Bardel and Lindqvist 2007; Sánchez 2020). However, there is also evidence that transfer may come from a low-proficiency L2, especially when the L2 and the TL are typologically similar to each other (see for example Lindqvist and Bardel 2014 for transfer from a weak L2). As for TL proficiency, transfer generally decreases as learners become more proficient in their TL. In fact, learners are likely to consistently rely on previously known languages when they are yet to fully develop their TL knowledge (Williams and Hammarberg 2009).

According to the Typological Primacy Model (Rothman 2011), linguistic proximity is the strongest predictor in Ln learning. In for example Rothman’s study from 2011, there were two groups of L3 learners of Portuguese and Spanish, with L1 English/L2 Spanish and L1 Italian/L2 English, respectively. The results provide evidence for transfer from a typologically similar language, i.e., Italian L1 for the L3 Spanish group and L2 Spanish for the L3 Portuguese group.

The L2 status factor hypothesis (henceforth, L2SFH, Bardel and Falk 2012) posits a strong role of the L2s in Ln syntactic transfer. At the core of the L2SFH is the declarative and procedural model (Paradis 2009), according to which there is an inherent difference between L1 acquisition and second/subsequent language learning. L1 acquisition involves unconscious mechanisms of language appropriation, leading to implicit linguistic competence, namely the ability to use language without knowledge of the actual underlying mechanisms (Paradis 2009: 38). It is supported by the procedural memory system. Second/additional language learning later in life is determined by explicit processes of language appropriation and involves metalinguistic knowledge, i.e., knowledge of explicit facts about language. It is sustained by the declarative memory system. According to the L2SFH, it is more likely for the L2s to be activated in Ln learning than for the L1, since the L2s and the Ln have the same processes of language appropriation (Bardel and Falk 2012). In their study (2007), Bardel and Falk found evidence for the L2 status factor by examining the placement of the negation by L3 learners of Swedish and Dutch, both verb-second (V2) languages. By comparing two mirror-image learner groups with the V2 structure in the L1 and in the L2, they argue for the L2 status factor, as the group with a V2 language as L2 transferred this structure in their L3 to a significantly higher extent than the group with a V2 language as L1.

Finally, there are two recent transfer models that are important to mention, namely the Scalpel Model (henceforth, SM, Slabakova 2017) and the Linguistic Proximity Model (henceforth, LPM, Westergaard 2021). In a nutshell, both models argue against the idea of wholesale transfer, which is maintained by the TPM. As mentioned above, the TPM argues that the entire grammar of a previous language is transferred during the initial stages of Ln learning due to perceived linguistic similarity. Conversely, both the SM and the LPM assume that Ln learning is an incremental property-by-property phenomenon. Specifically, they contend that neither the L1 nor the L2 has a privileged role in Ln learning. Instead, transfer is hybrid, that is, it may come from either the L1 or the L2 depending on the structure taken in consideration. Several factors are deemed as responsible for the selection of a source language. For example, structural similarity between certain properties is considered as fundamental in the LPM, while the SM also points out the importance of other external factors, e.g., input frequency or instruction. Further, both models concur on that transfer may be positive or negative, the latter due to misleading comparisons between a previously acquired language and the TL.

4.2 Transfer in the Ln acquisition of aspect

The L2SFH and the Typological Primacy Model have been adopted in two studies (Eibensteiner 2019; Foote 2009), both dealing with the Ln acquisition of the contrast perfective/imperfective in Romance languages. There are three additional studies (Diaubalick et al. 2020; Salaberry 2005; Vallerossa 2021) adding another relevant layer to the Ln acquisition of aspect, i.e., the role of prototypical associations. The studies by Diaubalick et al. (2020), Foote (2009), Salaberry (2005) and Vallerossa (2021) take knowledge of a Romance L2 into account, whereas the study by Eibensteiner (2019) focuses on the role played by aspectual knowledge in English as L2.

Diaubalick et al. (2020) tested a group of native speakers (n = 149) and three groups of German L1 speakers with no knowledge (n = 15), basic (n = 39) and advanced (n = 19) knowledge of a Romance language as L2. Data was a written 40-item discourse-based forced-choice task with prototypical and non-prototypical combinations. The results showed that the higher the knowledge of the Romance language, the closer the learners’ selection of TA morphology was to the control group for prototypical associations. However, for non-prototypical associations, all learner groups behaved differently from the native speaker group, which is connected to a persisting detrimental effect from the L1, namely German. These results are interpreted in the light of different systems underlying the acquisition of prototypical and non-prototypical aspectual configurations (for a similar argument, see Salaberry 2020). While the acquisition of prototypical meanings is sustained by explicit knowledge and can be influenced by transfer from the L2s in line with the L2SFH, non-prototypical meanings are negatively constrained by the L1 German, which lacks inflectional morphology.

Foote (2009) compared three groups of learners of Romance languages with different language constellations and a control group of native speakers on a morphology test and a sentence conjunction judgment task. In the sentence conjunction task, imperfective Romance tenses (i.e., imperfetto in Italian, imparfait in French, imperfecto in Spanish) made the sentences logical, while perfective Romance tenses (i.e., passato prossimo in Italian, passé composé in French, pretérito indefinido in Spanish) made them illogical. As for lexical aspect, Foote included stative verbs and telic verbs, the latter further divided into accomplishments and achievements. Of the three learner groups, one had English as L1 and was tested in a Romance language they had studied as L2. The remaining groups were two mirror-imaged learner groups of a Romance L3; one group had English as L1 and a Romance language as L2 whereas the other group had a Romance language as L1 and English as L2. The L3 groups completed both tasks in two Romance languages, whereas the L2 group was tested in one Romance language. In line with the Typological Primacy Model (Rothman 2011), the L3 groups performed in a similar way to the native group and were significantly more accurate than the L2 group, whether the Romance language was their native or non-native language. Compared to the L3 groups and the native speaker group, the L2 group was less straightforward in judging illogical sentences (i.e., those sentences including perfective tenses) with telic verbs. They were also insecure when judging both illogical and logical sentences, the latter containing imperfective tenses, with stative verbs. In brief, transfer of knowledge was attested across typologically similar languages.

Salaberry (2005) tested the influence of L2 Spanish on L3 Portuguese among English L1 speakers through a verb-selection task of a comic strip. A distinction was made in the study between English eventive verbs, which can receive perfective and imperfective morphology (Simple Past/Past Progressive), and stative verbs, which instead seldom occur in forms other than the Simple Past. An overall positive influence of L2 Spanish on L3 Portuguese was found among these participants but difficulties were found with associations of Preterit and stative verbs, i.e., non-prototypical associations. As for non-prototypical contexts with stative predicates, L1 transfer overrode transfer from the L2s, as these learners detrimentally relied on the L1 English and established erroneous connections of one form/one meaning as in their L1.

The study by Vallerossa (2021) focused on the interplay between linguistic typology, TL proficiency and prototypes in learning aspectual contrasts in Italian as Ln. Twenty-five undergraduate students were divided into four groups depending on a combination of knowledge of a Romance language as L2 (Romance; non-Romance) and proficiency level in Italian (low-proficiency; high-proficiency). The results from an oral retelling story test showed that both prototypes and knowledge of a Romance L2 interact with TL proficiency. Specifically, knowledge of a Romance L2 had a boosting effect on the emergence of the imperfective tense among low-proficiency learners while, among high-proficiency learners, such knowledge helped learners moving past prototypes with the perfective tense.

Finally, Eibensteiner (2019) investigated the influence of L2 English on L3 Spanish by L1 German speakers, adopting two sentence interpretation tasks of aspectual contrasts, one in Spanish and one in English. The latter allowed for a distinction between low and high English achievers. Significant differences between these groups were found in contexts presenting one form/one meaning relationships between English and Spanish. The German learners of L3 Spanish performed well in both prototypical and non-prototypical contexts, which is explained in the light of a positive L2 influence.

Foote (2009) emphasizes the positive influence of having knowledge of a language that is typologically similar to the TL, whether it is an L1 or an L2. Although this positive influence was evident in her study across all combinations of grammatical and lexical aspect, stative predicates were those generating greater difficulties among students without previous knowledge of a Romance language.

The study by Vallerossa (2021) validates Eibensteiner’s (2019) and Foote’s results (2009) by sustaining a positive influence on both prototypical and non-prototypical associations. Unlike Foote (2009), Eibensteiner (2019) explains this positive influence in the light of the L2 status factor. However, in Eibensteiner’s study (2019), as well as in Vallerossa (2021), the L2 is more similar to the target L3 in its ways of codifying aspect than the L1. Hence, it is not possible to disentangle whether the L2, namely English in Eibensteiner (2019) and a Romance language in Vallerossa (2021), has a privileged role due to its relative closeness to the TL as regards aspect, its status as a L2, or a combination of both factors.

The studies by Salaberry (2005) and Diaubalick et al. (2020) are useful to understand the differential role exerted by the L1 and the L2, in relation to aspectual combinations. In their studies, knowing a Romance language as L2 resulted in positive transfer, but only with prototypical associations. Instead, an L1 codifying aspect in a way different from the TL manifested its detrimental effect in the case of non-prototypical associations. Following their argumentation, there are two developmental patterns; one pattern regarding prototypical associations, where knowledge of additional languages may usefully be exploited, and one pattern involving non-prototypical associations, which are persistently constrained by the L1 (cf. Salaberry 2020: 60).

In summary, the aforementioned studies indicate some type of influence from previously known languages and contribute to the understanding of different dimensions involved in the learning of aspect in an Ln. However, there is no agreement on whether transfer is holistic or limited to certain types of associations (i.e., property-by-property), and on the factors that are responsible for transfer (i.e., the status of a language as L1/L2 or typological proximity).

5.1 Research questions

We address the following two research questions:

6.1 Participants

The participants in this study (N = 42) were undergraduate learners of Italian Ln with previous knowledge of L2 English (n = 34) and a control group of Italian native speakers (n = 8). There were 23 female and 11 male participants. Their age ranged from 20 to 76 (M = 42.6; SD = 18.74). Five participants who responded to an advertisement on social media were studying or had studied Italian privately or through educational associations (Studieförbund). The remaining twenty-nine participants were recruited from seven undergraduate courses of Italian. Although there is no official comparability between each undergraduate course and proficiency level according to the Common European Framework for Languages (Council of Europe 2001), the most advanced undergraduate course (i.e., the Bachelor thesis course in Italian) roughly corresponds to the advanced level C1 (see also Vallerossa 2021). The distribution of these participants was as follows: nine participants were recruited from the first preparatory course of Italian, three from the second preparatory course, six from Italian I, four from Italian II, four from Italian III, two from Italian IV and one from the Bachelor thesis course in Italian (Kandidat). Given the variation in study background and current study level of Italian, a C-test in Italian (see Section 6.2) served as a general measure of the participants’ level of proficiency. All the Ln learners of Italian had advanced knowledge of Swedish, and knowledge of a Romance L2 for the Romance group. Twenty-six participants had Swedish as L1, four participants had two L1s, one of which was Swedish, and four participants had another L1 but were highly proficient in Swedish.^[7] The learner group was divided into two subgroups: a non-Romance group (n = 12), consisting of learners of Italian L3 with previous knowledge of English as L2, and a Romance group (n = 22), including learners of Italian L4 with knowledge of English and another Romance language as L2s (either French or Spanish). The labels L3/L4, however, only refer to the languages controlled for in the present study and are not to be considered as absolute indicators of all the languages known by the participants. In fact, many participants reported previous knowledge of other languages^[8] in a sociolinguistic questionnaire (Ancient Greek, Arabic, Chinese, Bosnian, Croatian, Serbian, Dutch, German, and Russian). The participants chose which group they wanted to belong to after having received oral and written information about the selection criteria for each of the two learner profiles (Romance/non-Romance). Although the non-Romance group participants decided to be classified as such, several participants in this group reported some basic knowledge of another Romance L2 from studies in school.^[9] This general multilingual background is not surprising as Italian in Sweden, as has already been mentioned, is almost never approached as the chronologically third language, but it is generally preceded by studies in several other foreign languages (Bardel 2006).

6.2 Tests

To measure the learners’ knowledge of Italian aspect, an Italian interpretation test (IIT), inspired by the SPLLOC project (Domínguez et al. 2013), was conducted. The participants were presented with an aspectual situation, here referred to as ‘task’. Each task consisted of two parts: an aspectual context written in English^[10] and an Italian target sentence. Aspect, as expressed in the context, could be perfective, imperfective progressive or imperfective habitual. In relation to every context, the participants were presented with an Italian target sentence summarizing the context, but substituting the verb with a blank. The missing verb was presented to the participants in three different tenses, here referred to as ‘items’. The participants had to judge how appropriately, on a four point Likert-scale (completely/rather (in)appropriate), the three different verb tenses would describe the context. The option “I do not know” was provided with the indication of not overusing it, as it would determine the exclusion from the analysis (see Figure 1 below).

Figure 1:

Example of an aspectual situation from the Italian interpretation test. The English sentence expresses a habitual aspectual context, and the participant is asked to use an atelic activity verb (mangiare ‘to eat’) and judge the acceptability of the verb in three different tenses: passato prossimo, imperfetto and the periphrastic construction.

As for lexical aspect, the verb predicate was telic or atelic. The atelic predicates were either stative predicates or activities when inserted in a perfective or habitual aspectual context.^[11] In a nutshell, every task, consisting of a context and a target sentence, had three items, every item specifying a particular configuration of tense (passato prossimo, imperfetto, periphrasis), grammatical aspect (perfective, imperfective progressive, imperfective habitual) and lexical aspect (telic, activity, stative).

In total, the test consisted of 16 tasks, each one of these consisting of three items. The test also included four filler tasks that were removed from the analysis. Further, four items expressing another past tense form (i.e., the passato remoto) were also removed from the analysis. That left 32 items for the analysis. Two versions of the test were created and subsequently balanced so that some learners received one version of the IIT and others received the other version.

In Table 1, we report the 21 combinations of aspect, tense and lexical aspect tested. Some combinations were expected to yield high acceptance rates while other low. Those combinations where high acceptance rates were expected were: passato prossimo in perfective contexts, imperfetto in habitual and progressive contexts, and the periphrastic construction in progressive contexts. Instead, passato prossimo in habitual and progressive contexts, imperfetto in perfective contexts, and the periphrastic construction in perfective and habitual contexts were expected to yield negative responses. Generally, in this type of data, the native speakers’ interpretations are considered as a reference point against which to compare the learners’ interpretations.

Aspectual knowledge in English was measured by an English interpretation test (EIT) similar to the one in Italian, which was inspired by Eibensteiner (2019). The EIT tested knowledge of the English aspectual contrast between perfective and progressive aspect, as expressed by the contrast Simple Past/Past Progressive. There were 15 tasks in the EIT. Each task consisted of a context in Swedish and two target sentences, one with the Simple Past and one with the Progressive Past. Only one condition rendered each sentence acceptable. The sentences tested five perfective one-time contexts (three telic; two atelic) with the Simple Past providing a correct interpretation and seven progressive contexts (three telic, four atelic) with the Progressive Past providing a correct interpretation. Further, there were three distractor tasks.

In order to ensure that the learner groups did not differ as for proficiency level in Italian, they completed a test inspired by the C-test (henceforth referred to as C-test [Klein-Braley 1985, 1997]) in Italian. The C-test, inspired by Bardel et al. (work in progress) and Ågren et al. (2021), consisted of two short texts containing 31 blanks each for a total of 62 blanks. All the tests were performed online through the platform ‘itslearning’ for the C-test and Google forms for the interpretation tests.

6.3 Group divisions

For the analysis, the learner group (N = 34) was divided into learners with low aspectual knowledge (LAK) in English (n = 15) and learners with high aspectual knowledge (HAK) in English (n = 14). The sample was divided into these two levels based on the median of the EIT (Mdn = 18).^[12] As mentioned above (see Section 6.2), the learner groups completed a C-test as a measure of proficiency in Italian. As shown in Table 2, no significant difference was found between these groups regarding proficiency in Italian. Hence, the groups had comparable knowledge of Italian.

6.4 Procedure

The study was cross-sectional. The non-Romance group and the Romance group completed the C-test in Italian, the EIT and the IIT. While data from the non-Romance group were collected during one session, data from the Romance group were gathered in two sessions.^[13] The native speaker group only completed the IIT.

First, one of the researchers contacted Italian instructors at a Swedish university and briefly presented the project to undergraduate students from different levels. The participants who showed interest were subsequently contacted by the researcher with detailed written information and then gave their informed written consent. Finally, the same researcher met each participant online, either alone or in small groups. During the meeting, the participants performed the tests in the presence of the researcher. There was no time limit for the tests. Time of completion was approximately 1 h per participant.

8.1 Knowledge of a Romance language as L2

Our first research question addressed differences between learners of Italian as Ln, with and without knowledge of another Romance language as L2 and Italian native speakers regarding the interpretation of the semantic meanings entailed by tempo-aspectual morphology in relation to lexical aspect. This was tested with the IIT. The results showed significant differences for three combinations. Two of these differences were with habitual aspect, namely the combinations habitual.passatoprossimo.activity and habitual.passatoprossimo.telic. These combinations were expected to yield negative responses.

Starting with the combination habitual.passatoprossimo.activity, it was accepted by the native speaker group to a significantly higher extent than the non-Romance group. A similar outcome revealing uncertainty by native speakers has been attested in previous research (Foote 2009: 110). This result helps understanding the complexity of judging aspectual contrasts, which is not straightforward for native speakers either.^[14]

The case of the combinations habitual.passatoprossimo.telic is different because it was incorrectly accepted by the non-Romance group to a higher extent than the Romance group and the native speaker group. With this combination, there is congruence at the level of the predicate between lexical and grammatical aspect, and a clash with the overarching habitual perspective. Although learners in the non-Romance and the Romance group had comparable knowledge of Italian (cf. Section 6.3), additional knowledge of a Romance language seems, in our case, to offer an advantage in the realm of aspect. The non-Romance group’s judgments tended to be based on congruence at the level of the predicate, i.e., telic verbs drawing perfective morphology. The Romance group, similarly to the native speakers, tended instead to rely more consistently on the aspectual contrast at a sentential level. If compared to the postulates of the LAH (Andersen 1993), the Romance group seems more mature, aspectually speaking, than the non-Romance group; their judgments are less reliant on the predicate and instead tend to account for the complex aspectual constellation of the sentence (see also Rosi 2008 for a similar outcome regarding intermediate and advanced learners of L2 Italian). Also, the advantage of the Romance group is manifested with habitual aspect, which may be indicative of positive transfer from other Romance languages, which codify habitual aspect morphologically as Italian.

Finally, a significant difference was found for the combination perfective.imperfetto.stative between the native speakers and both learner groups. Hence, no L2-related differences connected to Romance languages were found and both learner groups behaved similarly. Crucially, neither knowledge of a Romance language as L2 nor English aspectual knowledge seems to influence the learning of these associations (see Table 4). This suggests that learning these associations might be influenced by factors other than L2 knowledge. Arguably, it could be due to a negative influence exerted by Swedish, which is the L1 of the majority of the learners. If so, our results would align with those of Salaberry (2005, 2020 and Diaubalick et al. (2020). We believe that the differential path of acquisition proposed by Salaberry (2005, 2020 and Diaubalick et al. (2020) might operate for multilingual Swedish-speaking learners of Italian as well. In this case, both Swedish and the L2 English would contribute to negative transfer. In fact, like German, Swedish also lacks inflectional morphology, which complicates the acquisition of non-prototypical contexts (Diaubalick et al. 2020). Moreover, high aspectual knowledge of English does not play any role when it comes to non-prototypical meanings, since dynamic and stative predicates are differentially marked in English. While dynamic predicates occur in both Simple Past and Past Progressive, stative predicates are usually inflected in the Simple Past, which, according to Salaberry (2005), would make learners more conservative in employing stative verbs with one tense form. Further, the consistent preference of imperfective morphology with stative predicates reflects semantic principles (Congruence Principle: Andersen 1993) and was also found in studies on Italian as a TL (cf. Banfi and Bernini 2003; Giacalone Ramat 1990). This study, however, differs from Salaberry (2005) and Diaubalick et al. (2020) in such a way that it does not seek to distinguish between L1 and L2 influence.

What we found evidence for is that knowledge of a Romance language L2 does not seem to play a role for these associations, suggesting that both learner groups detrimentally rely on the aspectual representations characteristic of other previously known languages lacking aspectual contrasts. Nonetheless, we did not control for level of aspectual knowledge in the learners’ Romance language. Learners might not have grasped this particular aspectual contrast in their Romance L2 in the first place. Arguably, their knowledge of a Romance L2 may not suffice for transfer into an Ln in the case of this particular combination.

8.2 Aspectual knowledge of English

Our second research question concerned whether differences could be found between learners of Italian as Ln, with low aspectual knowledge in English as L2 and high aspectual knowledge in English as L2, regarding the interpretation of the semantic meanings entailed by tempo-aspectual morphology in relation to lexical aspect. This was investigated through differences in results from the IIT that could be traced back to aspectual knowledge in English as L2. Six significant differences were found, two in relation to habitual aspect and four in relation to progressive aspect. In all the cases, high aspectual knowledge in English translated into responses more similar to those provided by the native speakers.

As for habitual aspect, HAK students were more prone to correctly reject the combinations habitual.passatoprossimo.telic and accept the combination habitual.imperfetto.activity to a significantly higher extent than the LAK group. Although English does not encode habitual aspect morphologically, it has a device marking habitual aspect (i.e., ‘to use to’), which arguably bootstrapped the correct remapping, for the HAK group, of habitual aspect onto the target imperfetto, and not onto the passato prossimo (see Eibensteiner 2019: 80 for a similar argument regarding the English periphrasis). However, it could be the case that the existence of the construction with bruka in Swedish could be deemed as responsible for this outcome as well. In this case, there would be positive transfer from both Swedish and English.

Further, four significant differences were found with progressive aspect; here, HAK learners outperformed LAK learners in both accepting the imperfetto and rejecting the passato prossimo. As mentioned in Section 6.2, having in-depth knowledge of the English aspectual system corresponded, in the present study, to having knowledge of perfective and progressive aspect in English, as expressed by the contrast Simple Past/Past Progressive. Our data indicate that having aspectual knowledge in English results in being able to grasp the notion of progressive aspect and attaching it correctly onto a target form, in our case the Italian imperfetto. This finding can be connected to studies focusing on crosslinguistic (dis)similarities between the L1 and the TL. More specifically, it echoes the results from Labeau (2005), who argues that the existence of a periphrastic construction in the learners’ L1 (i.e., English, as in Howard 2005) facilitates the acquisition of the progressive meaning entailed by the imparfait whereas its absence would complicate it as it does for learners with Swedish as L1 (Kihlstedt 1998). Our data corroborate this hypothesis, but we have found that such an effect does not only derive from having an L1 that realizes progressive aspect morphologically, but may also originate in an L2, as long as learners have grasped this contrast in their L2.