Abstract: Bilinguals’ languages are influenced both by language proficiency and linguistic similarity. Three models generated from the Reversed Hierarchical Model have been proposed to explain this phenomenon: Word-Association Model, Concept Mediation Model, and Multiple Access Model. Importantly, previous studies were conducted with bilinguals and only their alphabetic language ability was tested. We aimed to investigate whether the same findings could be applied to multilinguals on their ideography language ability. Moreover, most previous studies applied explicit memory paradigm, which could suffer undesired strategy-use problem. Using implicit memory paradigm, the present study examined a group of Tibetan-Mandarin-English trilinguals to explore their multi-language association with a specific aim to test the linguistic association models.
A 2×2 mixed design was adopted, with test-word types (learned and un-learned test-word / test-word translation equivalents) as the within-subject variable and linguistic relation between learning and test (the same or different language) as the between-subject variable. Linguistic relations were counterbalanced within subjects. All the materials were rated by homogeneous participants before the test. The subjects were native speakers of Tibetan, had been exposed to Mandarin since early childhood (all subjects had lived in the mainland for 5 years or more), and had studied English in school for 5 years. They were asked to self-rate their proficiency in three languages on a 7-point scale, which showed significant proficiency differences between Tibetan and English, and between Mandarin and English. There was no difference between Tibetan and Mandarin ability for these subjects.
In experiment 1, the main effect of test-word types and the interaction between two variables were significant while the main effect of linguistic relation was non-significant, indicating that the Tibetan translation equivalent words were not activated in Mandarin semantic access. In experiment 2, both two main effects and their interaction were significant, suggesting the representations of English words and Mandarin translation equivalent words were activated, whereas the representation of Tibetan translation equivalent words was not activated in the English semantic access. In experiment 3, the two main effects and the interaction were significant, indicating no activation of English translation equivalent words in the Mandarin semantic access.
Based on above findings, we discussed the similarity and difference between the linguistic association models for trilingual and bilinguals and reached the following conclusions: (1) The association models between the two directly-associated languages for Tibetan-Mandarin-English trilinguals are similar to those for bilinguals, while a learning medium is required for building association between the two indirectly-associated languages; (2) The linguistic association models for trilinguals are affected by language proficiency, linguistic similarity and learning medium.

Key words: Tibetan-Mandarin-English trilingual, linguistic association model