The roles of structural and lexical similarity in cross-linguistic influence in the L3 at higher proficiency levels are under researched. This study investigates the L3 Norwegian of such speakers. In alignment with the Linguistic Proximity Model (Westergaard et al., 2017), we assume that L3 structures are initially weak representations, becoming increasingly target-like with further input and use. We investigate what target this represents – the prescriptive rules of the language or movement towards L1-like use from community interaction and input. The properties investigated are the indefinite article and third person and reflexive possessives, by L1 Polish–L2 English and L1 English speakers. These categories provide fertile ground for investigation due to the (dis)similarities with the target language. The methodology consisted of an Acceptability Judgement Task. Results indicated possible structural-similarity based CLI and adherence to grammatical rules in intermediate-to-advanced proficiency L3ers – though this does not necessarily equal L1-like choices.

This empirical study aims to shed light on L3 initial-stage transfer and later development by investigating Q-operations in L1 English–L2 Cantonese and L1 Cantonese–L2 English bilinguals’ L3 Mandarin and L1 English speakers’ L2 Mandarin at low and high proficiency levels. Data from an online cross-modal priming task and an offline acceptability judgement task found that structural similarity determines transfer source selection. Adopting a de-compositional approach to cues of different domains, we have found both facilitative and detrimental transfer effects from Cantonese, with the latter triggered by orthographic and phonological cues. Our data also suggest that detrimental transfer effects can persist at an advanced stage and that L3 development and acquisition results can be affected by various factors such as word frequency and the nature of learning situations.