In general, discussion of crosslinguistic influence has focussed almost exclusively on the role of L1 in L2 production, both in the form of crosslinguistic influence on the learner's interlanguage and in the form of language switches to the L1 during L2 production. As yet, there has been little work done on the influence of a learner's other previously learned L2s in the acquisition of a new language (L3). The few studies that have been been carried out on the role of L2 in L3 production however show that L2 does play an important role in L3 acquisition. This paper presents the results of ongoing research on non-adapted language switches, using data from a 2 year longitudinal case study of an adult learner of L3 Swedish with L1 English and L2 German, both of which languages are related to Swedish. Our study is based on a corpus of 844 nonadapted language switches. We identified four main types of non-adapted language switch, 3 of which had pragmatic purpose, namely: i) EDIT (marking self-repair, beginning of turntake etc.), ii) META (used for asides, to comment on L3 performance or ask for help) and iii) INSERT (use of non-L3 items to overcome lexical problems in L3), and the last of which we refer to as Without Identified Pragmatic Purpose (WIPP switch; cf. "non-intentional switches", Poulisse & Bongaerts, 1994). We found that while both L1 English and L2 German occurred in EDIT and INSERT functions, it was only L1 English that occurred in META function and almost only L2 German that occurred in WIPP switches. Of these WIPP switches, the majority were function words. We also noticed that a number of the English utterances used in INSERT function appeared to show German influence, although this was not the case when English was used in META function. Our results show that in this case study of L3 acquisition, L1 and L2 play different roles. We suggest that L2 German, which was shown to be the non-L3 language predominantly used to supply material for lexical construction attempts in the L3 (Williams & Hammarberg, 1994), is activated in parallel to the L3 interlanguage, underlying L3 production and even L1 production. We refer to this as the DEFAULT SUPPLIER role. L1 English, on the other hand, is more seldom used in lexical construction attempts in the L3 or activated in parallel but rather is kept separate from the L3 and used largely with a metalinguistic function. We refer to this as the INSTRUMENTAL role. We propose a developmental model of L3 production based on de Bot's (1992) model of bilingual speech production but involving role assignment to the background languages, such that only the language which has been assigned the role of DEFAULT SUPPLIER is regularly activated in parallel to the L3 interlanguage. Over time, these two roles are largely taken over by the L3 itself. We suggest that these roles also exist in L2 acquisition but since there is only one background language, this takes over both roles and overt distinctions between them collapse.
