The L2 acquisition of agreement: comparing the inter language of Dutch, English, French and Swedish-speaking learners of Spanish

Abstract

Much of current generative research into non-native language (L2) acquisition of morphosyntax has focused on L1 transfer and access to Universal Grammar. Subject-Verb agreement (1) has figured more prominently than nominal agreement ((1 )-(2)) in this debate, but empirical findings remain inconclusive. For instance, Hawkins & Franceschina (2004) conclude that UG features (e.g. [GENDER]) not realised in the L1 cannot be acquired, whereas White et al. (2001) argue the opposite. The present study examines the acquisition of nominal and verbal agreement marking in L2 Spanish through acceptability judgement, comprehension and production tasks carried out amongst adult L2 acquirers matched for at least two levels of proficiency, with L1s which vary in terms of the realisation of nominal and/or verbal agreement: I demonstrate that the fact that L2ers can produce or recognise agreeing morphological markers is not sufficient to ascribe to them knowledge of syntactic agreement (and hence of the relevant functional features). The experiments address this issue by examining (non)agreement in non-contiguous ('long' distance) contexts with a complex sentential subject consisting of a head noun and an intervener (as illustrated in (3a-b) for nominal gender marking). Such test items systematically contrast contexts where the head noun and intervener have matching (3a) versus opposite (3b) agreement features L2ers at lower proficiency level perform significantly better at contexts with matching than opposite gender agreement features, suggesting that they rely more on linear word order and hence general cognitive learning strategies. The most advanced L2ers, however, demonstrate native-like 'long' distanceagreement in all contexts, suggesting (hierarchical) structure dependency and hence acquisition that is specific to Language (contra Hawkins & Chan's (1997) Failed Functional Features Hypothesis, but supporting access to UG as defined by Schwartz & Sprouse's (1996) Full Transfer/Full Access Theory). The data also reveal that not all types of morphosyntactic agreement are equally acquirable. For all L2ers regardless of their L1, nominal and verbal [NUMBER] are less problematic than [PERSON] and [GENDER]. These L2A findings differ from the results of studies into the L 1A of Spanish agreement morphology. L1 children master gender agreement before they start producing nominal number agreement (Marrero & Aguirre 2003, Hernandez Pina 1984) and produce distinctions between different verbal persons (1st and 3rd)before plural verb forms emerge (Bel2002, Grinstead 2000, López Ornat 1997).The L2ers' L1 does play a role, however, in the initial stages of L2A, particularly in the field of L2 morphology. Problems with remapping syntactic features onto surface morphology cause difficulties for L2ers whose L1 operates a different morphological system to L2 Spanish. L 1 French speakers, for instance, have fewer problems with the acquisition of separate morphemes for nominal gender and nominal number agreement in L2 Spanish than Dutch and Swedish L2ers whose L1 uses a portmanteau morpheme to realise both features. These problems in the field of 'morphological competence' (Lardiere 2005) appear more relevant than issues of syntactic transfer as predicted by Schwartz & Sprouse (1996). Indeed, L 1 English learners of Spanish do not seem to experience more problems building up a morphosyntactic system for nominal agreement from scratch than the Swedish and Dutch L2ers who need to 'remap' (i.e. disentangle and reassemble - Lardiere 2005) syntactic features to agreement morphemes. The finding that mapping problems between syntactic features and lexical forms prevent some L2ers from producing concording agreement morphology is also confirmed by the discrepancy between L2ers' ability to interpret and judge agreement marking, as reflected in the acceptability judgement and comprehension tasks, and the L2ers' more limited ability to produce agreement marking. Moreover, the least marked features often act as defaults, as demonstrated by the overgeneralization of [+MASC], [+3P] and [+SG] markings.