Competition and conceptual convergence in bilinguals

In the process of converting thought to speech, the mind of a bilingual rests on a delicate balance between inhibition and activation of competing language systems. A growing collection of findings from production studies with speakers of diverse language combinations suggest that bilinguals seem to have surprisingly limited control over ‘switching off’ the non-target language system. Evidence for a parallel activation comes for instance from picture naming; bilinguals are faster in naming cognates (e.g., gato in Spanish, gat in Catalan for “cat”) than non-cognates (e.g., mesa in Spanish, taula in Catalan for “table”) (Costa, Caramazza & Sebastián-Gallés, Reference Costa, Caramazza and Sebastián-Gallés2000; Hoshino & Kroll, Reference Hoshino and Kroll2008). The inability to avoid accessing representations from the non-target system points to a shared conceptual substrate for lexical retrieval. Similar co-activation is detectable on the level of grammar. For example, we know that crosslinguistic gender priming effects surface in translation tasks with noun phrases containing gender-marked adjectives. German–Dutch (Lemhöfer, Spalek & Schriefers, Reference Lemhöfer, Spalek and Schriefers2008) as well as Greek–German bilinguals (Salamoura & Williams, Reference Salamoura and Williams2007) were found significantly faster in translating noun phrases with identical grammatical gender in both languages. This indicates that grammatical representations are also in some way contained within the same functional system. But co-activation of L1 and L2 representations does not tell us much about how bilinguals integrate relevant grammatical elements to build larger conceptual frames (e.g., temporal or spatial frames) when organising information in response to complex verbal tasks (Brown & Gullberg, Reference Brown and Gullberg2008, Reference Brown and Gullberg2011; Levelt, Reference Levelt1989). What are the implications of two simultaneously active grammatical representations for the ways in which temporal information is organised for expression? Zooming in on language production patterns in narratives, do sequential bilinguals structure temporal information as in their L1; or as in their L2; do they build a system based partly on L1 and partly on L2 elements; or do they opt for something else?

Language-mediated conceptual change in bilingual speakers has stepped up as a vibrant topic in current SLA and bilingualism research (Athanasopoulos, Reference Athanasopoulos and Pavlenko2011; Cook, Reference Cook and Cook2003; Gentner & Goldin-Meadow, Reference Gentner and Goldin–Meadow2003; Green, Reference Green1998; Pavlenko, Reference Pavlenko and Pavlenko2011). This work is L2 production-focussed, and builds on the classification of conceptual change types proposed by Pavlenko (Reference Pavlenko and Pavlenko2011) and Jarvis and Pavlenko (Reference Jarvis and Pavlenko2008), who distinguish (a) coexistence of L1 and L2 conceptualisation patterns, (b) transfer of L1-based patterns, (c) convergence of L1 and L2 leading to unique or in-between patterns of performance, (d) restructuring, (e) internalisation of new patterns, (f) influence of L2 on L1 performance, and (g) attrition of L1 patterns. The key notions in this work are conceptual restructuring, i.e., “performance of bilinguals that diverges from the L1 pattern and begins to resemble, albeit not necessarily fully, that of the L2 speakers” (Pavlenko, Reference Pavlenko and Pavlenko2011, p. 247), and conceptual convergence, i.e., “a particular type of restructuring, namely convergence between two systems, whereby a unitary conceptual category is created that incorporates both L1 and L2 features” (Jarvis & Pavlenko, Reference Jarvis and Pavlenko2008, p. 164). The analyses draw only on learners’ production in their L2, not in their L1: thus possible L2 influence on L1 patterns remains unexplored.

Active interest in the area of conceptual restructuring and convergence is all the more propelled by the fact that related studies yielded discrepant outcomes. With focus on language production studies, evidence was documented both in favour of conceptual restructuring in the direction of the L2 (e.g., Cadierno, Reference Cadierno, Achard and Niemeier2004; Hohenstein, Eisenberg & Naigles, Reference Hohenstein, Eisenberg and Naigles2006; Wolff & Ventura, Reference Wolff and Ventura2009, for motion events), as well as against it (e.g., Cadierno & Ruiz, Reference Cadierno and Ruiz2006, for voluntary and caused motion; Carroll & Lambert, Reference Carroll, Lambert, Dimroth and Starren2003; Schmiedtová, von Stutterheim & Carroll, Reference Schmiedtová, von Stutterheim, Carroll and Pavlenko2011, for event construal). Convergence in production patterns, i.e., linguistic performance not conforming to either the L1 or the L2 but found to be a hybrid of the source and the target patterns, was reported for language production of late bilinguals in the expression of caused motion (Hendriks, Hickmann & Demagny, Reference Hendriks, Hickmann and Demagny2008), in temporal subordination (Natale, Reference Natale2013), additive and contrastive event linkage (Tomita, Reference Tomita2013), and also event linearization (Vanek, Reference Vanek, Roberts, Ewert, Pawlak and Wrembel2013).

In Hendriks et al. (Reference Hendriks, Hickmann and Demagny2008), video retellings of advanced English learners of French exhibited idiosyncratic non-target-like devices in reference to motion, which was strongly evocative of a converged system of prototypical L1 and L2 patterns. In L1s, English speakers systematically conflated Motion with Manner in the main verb and expressed Path in non-verbal satellites (e.g., He pushes the basket of apples across the street), and French speakers typically conflated Motion with Path in the main verb and encoded Manner in peripheral structures (e.g., Il traverse la route en poussant son panier de pommes “he crosses the street pushing his basket of apples”). Interestingly, learners were found to transform existing Path verbs into satellite-like devices (e.g., Il pousse le panier de pommes au travers la route “he pushes the basket of apples across/through the road” (p. 30), which does not express boundary crossing). This shows that learners’ expressions resemble the target-like expression of Path in verb-like forms, yet they clearly digress from the target pattern, i.e., from expressing Path in main verbs and Cause+Manner in peripheral structures.

Of similarly close relevance to this study, Natale (Reference Natale2013) and Tomita (Reference Tomita2013) identified symptoms of converged concepts in late bilinguals’ narratives. Natale (Reference Natale2013) examined principles of subordination in the film retellings of French learners of Italian. Learners employed atypical subordination devices that conformed neither to the French L1 pattern (most frequently presentatives combined with relative clauses: there is an x which. . .) nor to the Italian L1 pattern (most frequently gerunds, e.g., Allora alza le mani cercando di raccogliere queste gocce “so he lifts his hands trying to collect these drops”, p. 158). Learners typically used temporal subordinate clauses, also for situations where this type of linking played a “minor role” (p. 163). In Tomita (Reference Tomita2013), film retellings of advanced German learners of Japanese exhibited an interesting type of convergence of L1 and L2 principles pertaining to strategies for contrastive event linkage. Whilst German L1 speakers typically preferred to follow a shift-in-TIME principle (P did not do X at time Z. Now, however, P did do X.), Japanese L1 speakers typically opted for a shift-in-ENTITY principle in response to the same stimulus (P did not do X at time Z, and R did do X at time Y. Finally, P also did X.). Learners tended to follow a hybrid set of principles, with (a) a significantly higher frequency of ‘now’-adverbials than in Japanese L1, resembling the German L1 pattern, (b) a comparable frequency of the additive particle –mo “also” to that in the Japanese L1 production, and (c) learner-specific uses of ‘finally’-adverbials which express the target-like meaning of [after a long and eager waiting] as well as “those which do not” (p. 142).

This study aims to contribute to the debate on conceptual change in the L2 in two ways. The first objective is to quantitatively assess the role of language distance in learnersʼ reliance on source language (L1) structuring principles. The second objective is to qualitatively examine learner-specific structuring techniques to identify whether preferences of sequential bilinguals reflect prototypical L1, L2, or other event conceptualisation patterns.

Language structure and temporal conceptualisation

Explorations on reference to time across languages have accumulated extensive support for the idea that preferences in how native speakers organise temporal information for expression are clearly related to grammaticalised form-function connections available in a given language (Berman & Slobin, Reference Berman and Slobin1994; Bohnemeyer & Pederson, Reference Bohnemeyer and Pederson2011; Filipović & Jaszczolt, Reference Filipović and Jaszczolt2012; Gumperz & Levinson, Reference Gumperz, Levinson, Gumperz and Levinson1996; Klein & Li, Reference Klein and Li2009; Li & Shirai, Reference Li and Shirai2000; Lucy, Reference Lucy1992; Stromqvist & Verhoeven, Reference Stromqvist and Verhoeven2004; Talmy, Reference Talmy, Robinson and Ellis2008). Speakers seem naturally sensitised to the form-function connections that their language grammars code obligatorily. This view is driven by the assumption that our mother tongue(s) trains us to pay attention to different aspects of experience thanks to routine marking of certain contrasts. By the same token, the absence of overt marking of specific contrasts in a given language can lead its speakers to show lower sensitivity to particular aspects of experience, or increased sensitivity to other aspects of experience. Formation of cognitive routines (Robinson & Ellis, Reference Robinson and Ellis2008) reflects the interaction of the conceptual substrate with language in the mind of the speaker, through a conventionalised adaptation of responses to communicative goals (Hudson, Reference Hudson2007; Langacker, Reference Langacker1999; MacWhinney, Reference MacWhinney, Robinson and Ellis2008; Tomasello, Reference Tomasello2003).

It is important to highlight that typological similarity of languages does not automatically imply identical performance of their speakers. For instance, film retellings by speakers of Czech (an aspect language) clearly showed that they conceptualise events much like speakers of German (a non-aspect language) rather than showing resemblance to preferences of speakers of Russian (an aspect language) (Schmiedtová & Sahonenko, Reference Schmiedtová, Sahonenko, Grommes and Walter2008). The context of such variation invites further crosslinguistic research on temporal reference to focus not on bipolar aspectual contrasts and categorical differences, but to address a finer level of detail and articulate hypotheses that incorporate the notions of regularity degrees in aspect marking.

Crosslinguistic similarity can be operationalized as the way in which a specific concept is grammaticalised across languages (e.g., von Stutterheim, Nüse & Murcia-Serra, Reference Von Stutterheim, Nüse, Murcia Serra, Behrens, Fabricius–Hansen and Johansson2002), and for this work the concept under scrutiny is ongoingness. The underlying assumption is that preferences in eventFootnote ¹ construal are linked to the availability of grammatical forms that encode the conceptual category of ongoingness. Ongoingness is marked in very different ways across the source and target languages, allowing for the possibility to test hypotheses built on specific degrees of L1-L2 overlap. To briefly sketch out the major contrasts, in English, ongoingness marking is regular across tenses via the V+ing form that denotes imperfectivity. Czech is less systematic in this respect. The imperfectivising suffix –va, as in koupit/kupovat “to buy/to be buying”, is the only pure grammatical marker of ongoingness in Czech. And since the occurrence of simplex perfective verbs (e.g., koupit “to buy”) is infrequent, imperfectivising suffixation via –va is relatively scarce (Schmiedtová, Reference Schmiedtová2004). This means that in relevant contexts the grammar of Czech is likely to highlight ongoingness less than the grammar of English. In comparison, the Hungarian–English overlap is even smaller in terms of highlighting ongoingness. Hungarian simplex verbs are aspectually ambiguous (Csirmaz, Reference Csirmaz, Blaho, Vicente and de Vos2004), and because grammatical aspectual operators are largely absent (they only surface for particle verbs), the expression of ongoingness is achieved typically via lexical means. These language-specific aspectual distinctions present potential difficulties for learners in the process of acquiring a novel system of temporal framing for which new form-meaning connections need to be used. If a higher degree of L1-L2 overlap in grammatical marking of ongoingness implies facilitation in temporal restructuring, we should observe some advantage of Czech–English over Hungarian–English bilinguals in terms of approximation to target-like structuring patterns.

Theoretical underpinnings

Natural language production requires the mind of the speaker to prepare content for expression. This involves transforming information units into a format expressible in the used language. The preparation of content for expression includes three stages of message planning: conceptualisation, formulation and articulation (Levelt, Reference Levelt1989, Reference Levelt, Brown and Hagoort1999). A refined version of Levelt's language production model distinguishes four subprocesses within the conceptualisation stage: namely segmentation, selection, structuring and linearization (Carroll & von Stutterheim, Reference Carroll, von Stutterheim and Ramat2003; Habel & Tappe, Reference Habel, Tape, Klabunde and von Stutterheim1999). Segmentation (partitioning of situations) and selection (choice of components for verbalisation) together constitute macroplanning processes (the level of what to say), while structuring (temporo-spatial anchoring) and linearization (component ordering) represent microplanning processes (the level of how to say it)Footnote ² .

Structuring concerns anchoring several information units in a longer sequence within a particular referential frame (e.g., temporal and spatial anchor). In terms of possible temporal structuring contrasts, choices for the speaker lie between an anaphoric vs. deictic referential anchor (Carroll & von Stutterheim, Reference Carroll, von Stutterheim and Ramat2003). Each of these anchors serves to establish and maintain temporal coherence in a specific way. A deictic anchor entails the presentation of events as ongoing, and in larger chunks of discourse it is typically exhibited by keeping event times identical with the utterance time (Figure 1). An anaphoric anchor is characterised by frequent reference to endpoints and by temporal shifts from one (sub-)event to another (Figure 2). Building on this contrast, it is posited that speakers of languages encoding the concept of ongoingness grammatically within their aspectual system (such as English) will exhibit a high level of sensitivity towards ongoingness in their event conceptualisation processes, and that they will prefer to anchor events deictically. Grammatical marking of ongoingness in the aspectual system will lead to a specific attentional focus in speakers, which will lead them to highlight the ongoing event features in descriptions (e.g., a young man is surfing, the wind is blowing him off the board (Carroll et al., Reference Carroll, von Stutterheim, Nüse, Pechmann and Habel2004, p. 190)). However, speakers of languages with no specific marker for ongoingness (such as Hungarian) or having a specific ongoingness marker only for a small group of verbs (such as Czech) are expected to show lower sensitivity to this concept, since their grammars do not strongly direct their attention to this particular event feature. Lower sensitivity to ongoingness on the level of event linkage surfaces as anaphoric anchoring (an example from German L1: ein kleiner Mann surft auf den Wellen, dann wird er plötzlich von dem Brett geweht “a little man surfs on the waves, then he is suddenly blown off the board” (Carroll et al., Reference Carroll, von Stutterheim, Nüse, Pechmann and Habel2004, p. 190)). These assumptions are in accordance with Slobin's (Reference Slobin, Gumperz and Levinson1996) view that grammar is not a mere formal system but a set of general notions forming a schematic framework for conceptual organisation. If these assumptions are valid, we should find marked crosslinguistic differences in event conceptualisation processes.

Figure 1. Temporal linkage based on deictic anchoring (TT is maintained constant with the TU, events are linked to the TU, individual relations between TSits are left implicit and their temporal boundaries unspecified). TT stands for topic time (i.e., time for which the assertion is made), TSit for situation time (i.e. time for which the situation holds true), and TU for utterance time (i.e. time at which the utterance is made) (Klein, Reference Klein1994, p. 3).

Figure 2. Temporal linkage based on anaphoric shifts (the current TT is located in the post-time of the preceding TSit, event boundaries (T-boundary) are specified to act as anchor points for the next TT, and event chain formation is more typical than linking each TT to the TU separately).

Extending investigation to second language learners, previous findings show that bilinguals are capable of building language systems of their own which are not necessarily fixed in principles either of the L1 or of the L2 referential frame (Jarvis & Pavlenko, Reference Jarvis and Pavlenko2008). Instances of learner-specific phenomena of this type are treated as fully autonomous multicompetence features stemming from language interaction in the bilingual mind (Cook, Reference Cook1992, Reference Cook and Cook2003).

Some prominent theoretical models of the cognitive architecture in bilinguals posit language-specificity on the (sub)lexical level while remaining largely neutral to whether or how the status of conceptual representations for each language varies (e.g., the revised hierarchical model (Kroll & Stewart, Reference Kroll and Stewart1994); the bilingual interactive activation model (Dijkstra & van Heuven, Reference Dijkstra and van Heuven2002)). Other, usefully augmented models, such as the shared distributed asymmetrical model (Dong, Gui & MacWhinney, Reference Dong, Gui and MacWhinney2005) or the dynamic distributed activation model (de Bot, Lowie & Verspoor, Reference De Bot, Lowie and Verspoor2007), build on a converging view of conceptual representations in bilinguals, recognising that concepts contain L1-specific, shared, as well as L2-specific elements. A common denominator of these models is their focus on lexicalised concepts. Approaching conceptualisation from a different angle, this study targets grammaticalised concepts, namely the concept of ongoingness, and models the role of more than one type of grammaticalised knowledge at the stage of conceptualisationFootnote ³ . In doing so, we build on Levelt's language production model (Reference Levelt1989, Reference Levelt, Brown and Hagoort1999), and fully endorse his idea that access to language-specific grammatical information is crucial already at the stage of microplanning (1999, p. 93). This view aligns with Slobin's (Reference Slobin1987, Reference Slobin, Gumperz and Levinson1996) thinking-for-speaking hypothesis, positing language-specificity for cognitive operations already in the speech planning stage. In synchrony with previous suggestions that posit language-specific grammatical knowledge to play a role at the stage of conceptualisation (e.g., Flecken, von Stutterheim & Carroll, Reference Flecken, von Stutterheim and Carroll2013; von Stutterheim & Nüse, Reference Von Stutterheim and Nüse2003), the extension of Levelt's model in this work takes one additional step, and assumes different grammatical knowledge types to compete when bilingual speakers build a temporal conceptual frame during the process of preverbal message generation.

Besides comparing group means for the identification of more general trends in specific bilingual cohorts, current theoretical discourse in SLA points out the importance of considering within-group variation attributable to a host of factors with a potential impact on conceptual restructuring towards the L2, including learning context (Malt & Sloman, Reference Malt and Sloman2003), or length and amount of language exposure (Wolff & Ventura, Reference Wolff and Ventura2009). This study considers one within-group factor, the age of onset of L2 acquisition (AOA). The motivation for examining different AOAs of instructed learners relates to the yet unanswered question why in a school setting “the additional time associated with an early headstart has not been found to provide more substantial long-term proficiency benefits” (Harley, Reference Harley and Med1998, p. 27). One explanation could be that potential age effects on L2 learning examined alongside L2 input have not been sufficiently explored to date (Muñoz, Reference Muñoz2014). In response to this gap, production tests on instructed language learners with different AOAs, but with very similar proficiency levels and a comparably high frequency of L2 exposure outside the classroom, promises to further our understanding about the (lack of) age effect on L2 acquisition in input-limited environments.

Previous research on temporal structuring across languages and learner varieties

Native speakers’ responses in previous studies on temporal structuring were commonly found to conform to two main tendencies. These tendencies surfaced in relating topic time (TT, i.e., time for which the assertion is made) to situation time (TSit, i.e., time for which the situation holds true) (Klein, Reference Klein1994). In their film retellings, speakers of languages with a grammatical marker for ongoingness, including English, Arabic and Spanish, typically inclined to topic time maintenance (i.e., deictic anchoring, illustrated in Figure 1). Topic time maintenance represents event linkage where TT gets ‘pegged’ to the time of utterance (e.g., mum is unwrapping the cake, she is putting it on the table, and she is decorating it with chocolate icing). The temporal anchor for a given TT is the deictic ‘now’ of the event. Event construal characterised by deictic referential anchoring is typically marked with imperfective forms signalling ongoingness. Sequentiality is not in focus, and explicit lexical linkage of temporal relations between propositions is infrequent. Thus, in a context when sequentiality is defocussed in two consecutive or partially overlapping TSits, the listener has to rely to a great extent on contextual cues and general knowledge to decipher the exact temporal relation between individual TSits.

The second main tendency in structuring temporal information is topic time shifting (i.e., anaphoric anchoring). It is a type of event construal dominantly employed in film retellings by speakers of languages without grammatical marking of ongoingness, such as Swedish and German (Bylund, Reference Bylund and Pavlenko2011a; Carroll and von Stutterheim, Reference Carroll, von Stutterheim and Ramat2003; von Stutterheim and Lambert, Reference Von Stutterheim, Lambert and Hendriks2005). Characteristic features of this technique are holistic event construal and explicit specification of temporal relations between TT and TSit of the preceding event (e.g., then the mother unwraps the cake, and after that she puts it on the table, and then she decorates it). In other words, the TT of a given event is located in the posttime of the preceding TSit interval by means of temporal shifters (anaphoric markers such as then, after that, whereupon). When TT is shifted anaphorically, the temporal anchor for a given TT is the preceding TSit (Figure 2).

Findings from previous research on temporal structuring signal an important role for the language-specific category of grammatical aspect in event conceptualisation. Consensus was reached in claiming (a) that languages with no grammatical ongoingness marking direct their speakers to view events holistically and hence they employ TT shift as the preferred event linking technique; and also (b) that languages marking ongoingness grammatically sensitise their speakers to ongoing phases more, resulting in a tendency to maintain TT in event linking.

Research on temporal structuring in L2 has brought more heterogeneous results. Carroll and von Stutterheim (Reference Carroll, von Stutterheim and Ramat2003) found that advanced German learners of English significantly diverged from English temporal structuring patterns and preferred typically German TT shifting. Evidence for clear digressions from target-like event structuring also came from a study by von Stutterheim and Lambert (Reference Von Stutterheim, Lambert and Hendriks2005) testing French and German advanced learners with English L2. Even though highly proficient in terms of formal knowledge, the learners employed interesting techniques with no parallel in the English L1 data. They mixed temporal adverbials (TADVs) expressing punctuality with imperfective structures, as in the ground is suddenly opening (2005, p. 226). At variance with these findings, event structuring of Spanish speakers in L2 Swedish (Bylund, Reference Bylund2011b) was found target-like. Structuring patterns were examined by measuring the frequency with which speakers linked events via TADVs. Quantitative analyses showed that L2 production was characteristic for a high frequency of anaphoric TADVs (typically co-occurring with topic time shifting in Swedish), which was interpreted as evidence for conceptual reorganisation in L2 speakers at the level of microplanning (i.e., at the level of how temporal information for expression is structured). Although different, these findings are not necessarily contradictory. The difference in conceptual reorganisation may be attributed, inter alia, to acquisition context (whilst the participants in Bylund (Reference Bylund2011b) were naturalistic learners, those in von Stutterheim and Lambert (Reference Von Stutterheim, Lambert and Hendriks2005) were instructed learners) or to length of residence in the target language (TL) environment (participants in Bylund (Reference Bylund2011b) had resided in TL setting for over 15 years). These findings together signal that reorganisation of temporal structuring principles may be a very late feature in L2 development, and may only develop under certain circumstances.

With instructed learners in focus, the next undertakings in this line of research need not only to look at whether typological aspectual properties of L1 do or do not influence particular event conceptualisation processes in L2, but to carefully examine how specific degrees of L1-L2 overlap in ongoingness marking impact the L2 production of comparable learner groups. This type of research necessitates analyses of language produced by learners from at least two typologically distinct source languages (SL) acquiring the same TL – which differs from both SLs in the degree of overlap in marking ongoingness. With such design it is not only possible to test how learners from specific L1 backgrounds structure temporal information in discourse, but also to see whether particular structuring choices change as a function of similarity in L1-L2 ongoingness marking.

Research questions

1. (a) Do specific event structuring patterns per L1 group interact with the formal marking of ongoingness in these languages, and if so then how exactly? Which particular form-function mappings are preferred for temporal structuring in each of the L1 groups? Are temporal structuring preferences per group consistent across task types?

2. (b) What are the event structuring patterns that characterise Czech–English and Hungarian–English sequential bilinguals? Do sequential bilinguals adopt target-like principles for organising temporal information in discourse, do they remain rooted in the principles typical of the corresponding L1s, or do they prefer other choices? Can language distance in terms of encoding ongoingness in its aspectual system serve as a reliable predictor of how learners will digress from target-like structuring patterns? Does variation in the age of onset of L2 acquisition affect bilinguals’ structuring patterns?