We investigated interaction among the languages of multilingual speakers, and the degree to which processing of one language relies on knowledge of other languages (i.e., cross-linguistic influence, CLI). Specifically, it is as yet unsettled in the literature whether L3 processing is modulated by CLI from the L1, the L2 or both. The current study investigated this issue in morpho-syntactic processing, and tested whether CLI is similar in online and offline measures. The study focused on syntactic processing of English as an L3 among Arabic-Hebrew-English university student trilinguals (n=44). Importantly, both L1 (Arabic) and L2 (Hebrew) of participants are typologically distant from L3 (English), which allows overcoming confounds of previous research. The performance of trilinguals was compared to that of native English monolingual controls (n=37). Participants read grammatical and ungrammatical sentences in English while their eye-movements were recorded. Sentences included morpho-syntactic structures in English which differed from L1, from L2 or from both L1 & L2. Participants also performed grammaticality judgements on each sentence (offline measure).
Results demonstrate that interference in L3 processing can come from either the L1 or the L2, but the patterns of interference varied across measures. In the offline measures, we found significant interference effects from all three critical conditions, whereas online measures showed only interference from either the L1 or L2 individually, specifically in early-stage reading measures. These findings suggest that during L3 processing the multilingual language system is fully interactive, but also suggest that knowledge in L1 and L2 may not interact with L3 processing in the same way. Finally, divergence between online and offline measures suggests that online measures reflect mainly automatic activation, whereas offline judgements might also involve strategic and meta-linguistic decision making.