What does it take to predict RTs in Maltese subliminal language processing? – Traditional experimental predictors vs. predictors based on distributional semantics

In this study, we examine auditory and visual masked priming in Maltese, a Semiticlanguage shaped by close language contact with (Sicilian) Italian and English, bycombining traditional categorical priming conditions with continuous semantic similarityfrom a Maltese BERT model. In both modalities, higher cosine similarity between primesand targets predicted faster responses, revealing gradient semantic effects. Auditorypriming was more sensitive to these effects, especially for phonologically unrelated andmixed-origin items, while visual priming showed weaker, more restricted patterns. Ourresults demonstrate that distributional semantics can complement, but not completelyreplace, categorical quantitative modelling designs by uncovering fine-grained lexicalrelationships, with effects modulated by modality and etymology.