Abstract

In the present study, we examined how translation equivalents are linked in the bilingual mental lexicon. Some bilingual models assume that translation equivalents are linked through their shared concept (Multilink, Dijkstra et al., 2018). Other models assume the two words are also connected through a form-level paired association (RHM, Kroll & Stewart, 1994). We used the masked translation priming paradigm to examine the extent to which each link ties translation equivalents together. It has been well established that L2 targets are recognized faster when they are primed by their L1 translation equivalents than by unrelated L1 words, and the facilitation effect is assumed to occur because primes-target pairs share underlying links across languages. Taking advantage of the Japanese language, we teased apart facilitation via conceptual link and via form-level paired association by manipulating visual familiarity of Japanese translation primes. Our critical primes were either visually familiar Kanji primes (e.g., 真珠) or visually unfamiliar Katakana-transcribed primes (e.g., シンジュ), paired with English targets (e.g., PEARL). With Kanji primes, both form-level and conceptual links are intact. With Katakana-transcribed primes, form-level association is lost, but they still retain a conceptual link by relying on a phonological-to-conceptual route. In Experiment 1, we demonstrated that Katakana-transcribed primes can facilitate their base Kanji word targets, confirming that these primes can facilitate targets when underlying links exist. In Experiment 2, using Kanji primes, we observed a significant translation priming effect, replicating previous studies (Nakayama et al., 2012, 2013). Experiment 3 was our critical test, using Katakana-transcribed primes. Critically, Katakana-transcribed primes produced no priming effect. The lack of translation priming suggests that form-level paired association, as assumed in RHM, is a critical link that ties translation equivalents, challenging Multilink’s assumption that translation equivalents are linked only via shared concepts.