Direction of Priming and Phonetic Prototypicality in VOT Specificity Effects Danyuan Ho 1 & James Sneed German 2 1 Division of Linguistics & Multilingual Studies, Nanyang Technological University, Singapore 2 Aix-Marseille University, CNRS, LPL UMR 7039, 13100, Aix-en-Provence, France hoda003@e.ntu.edu.sg, james.german@lpl-aix.fr ABSTRACT While previous studies have observed that the specificity effect of voice-onset time (VOT) is mediated by VOT length [1], the role of the direction of mismatch at prime versus test has not been directly explored. This study addresses this issue through a long-term repetition priming experiment that simultaneously manipulated both VOT length at test (unmodified vs. reduced) and the VOT match status (matched vs. mismatched). The results show that having an unmodified VOT and matching the VOT of the study prime were both significantly correlated with shorter reaction times at test, though unmodified VOTs were identified faster overall regardless of match status. These findings corroborate the importance of the role played by fine-grained phonetic information in word representations, and we argue that the dominance of VOT length can be explained if the malleability of word-level representations depends on the density of speech experiences across the phonetic space. Keywords: voice onset time; specificity; phonetic detail; learning 1. INTRODUCTION Specificity refers to the phenomenon whereby retrieval is more effective when information present at retrieval aligns with the input stimuli that gave rise to existing representations [2]. In Hintzman, Block and Inskeep [3], this was shown to extend to visual word recognition, in that words were better recognised when presented in a constant typography during test and study as compared to words with different typographies, suggesting that surface details associated to the font type is retained in the orthographic representation of words. Goldinger [4] examined the effect of speaker voice on spoken word identification and found that words produced by the same speakers are more readily recognised than those produced by different speakers, leading the author to conclude that extra-linguistic information is preserved in the representation of spoken words. McLennan, Luce & Charles-Luce [5] explored the effect of allophonic variability on sublexical ambiguity resolution and found that lexical items with matching allophonic details (e.g., [ɾ]  [ɾ] or [t]  [t] as opposed to [ɾ]  [t] or [t]  [ɾ]) resulted in better priming that their unmatched counterparts. Ju and Luce [1] further explored the role of subphonemic detail by studying how words with artificially reduced VOTs affected lexical decision during a long-term repetition priming experiment. Participants listened to primes in various levels of reduction (namely: intact, -1/3, - 2/3) where half of the items at test matched those at prime in terms of VOT and the other half did not match. It was found that the specificity effect of VOT was affected by the degree of VOT modification since only the intact and -2/3 reduced tokens showed a priming advantage for their corresponding targets. In a second experiment within the same study, it was found that VOT length also affects the processing speed of the word tokens, with intact VOT being the fastest and the -2/3 VOT with the slowest processing time. Since VOT categories are both language and dialect-specific, one aim of the present study is to corroborate the key finding of Ju and Luce [1] in a novel linguistic context, namely Singaporean English. The study also seeks, however, to directly assess whether and how the direction of mismatch interacts with VOT length at test. Understanding how the size of the specificity effect varies across these conditions is important for understanding the structure of the word-level representations themselves. More importantly, being able to estimate the robustness of the VOT specificity effect under various conditions can be useful for constructing more sophisticated follow-up studies that are specifically designed to assess the role of para-linguistic or non-linguistic factors in the word- level encoding of phonetic detail. 2. METHODOLOGY 2.1. Participants 28 Singaporeans took part in the study. The participants received a remuneration of S$10 for