This study investigates the role of phonology in spontaneous imitation in Seoul Korean speakers' imitation of aspirated stops by comparing the primary and non-primary cues. Seoul Korean aspirated stops are differentiated from stops of other phonation types by at least two distinct acoustic properties, stop VOT and f0 of the post-stop vowel, with the latter being the primary cue. In the imitation experiment, Seoul Korean speakers heard and shadowed model speech that contained aspirated stops

more »

... pulated by either raising post-stop f0 or lengthening VOT. Their realization of these properties in /tʰ/, /t/, and /t*/ productions were compared before, during, and after exposure. Although both high f0 and long VOT induced imitative changes in post-shadowing productions, the results revealed that exposure to an enhanced non-primary cue (long VOT) also influences the production of the primary cue for aspirated stops (post-stop f0). However, an enhanced primary cue (high f0) does not have similar effects on the non-primary cue. These results provide evidence that spontaneous imitation is not strictly tied to individual phonetic properties but it is rather phonological in that abstract categories are involved in the process of imitation. Kwon: The role of native phonology in spontaneous imitation Art. 10, page 2 of 24 in direct opposition to each other, but they focus on different aspects of spontaneous imitation. While the exemplar account claims that the nature of the memory system gives rise to observed patterns of imitation with regard to the effects of lexical frequency, recency, and amount of exposure, the direct realist account claims that the intrinsic link between speech perception and production at the articulatory level explains the rapid and direct imitation. Many researchers have investigated the role of various social factors in spontaneous imitation, and reported that non-grammatical or social factors, such as gender, conversational role, or attractiveness of the model speaker influence both the likelihood of imitation (i.e., whether the model speech is imitated in the subsequent productions or not) and the imitation fidelity (i.e., how similar the subsequent production becomes to the model speech) (e.g., Pardo, 2006; Babel, 2012; Babel, McGuire, Walters, & Nicholls, 2014) . While these social aspects of imitation are an intriguing topic that has been intensively explored, the present study focuses on the role of phonology in imitation and asks the following question: When speakers converge to a model speaker, that is, when the imitation is triggered, how does the imitators' phonological knowledge influence the pattern of imitation (i.e., how the model is imitated), as well as the likelihood and the degree of imitation? Previous research on the role of phonology in spontaneous imitation have suggested that changes in certain phonetic properties are not imitated when they are detrimental to phonological contrast or phonologically irrelevant. For example, Nielsen (2011) reports that English speakers imitate lengthened VOT for English voiceless /p/, but not shortened VOT. This suggests that the imitation is phonologically selective: Imitation is attenuated by the presence of a phonological boundary. In addition, Mitterer and Ernestus (2008) report that duration of pre-voicing of Dutch voiced stops is not imitated, attributing the lack of imitation to the phonological irrelevance of the property. According to Mitterer and Ernestus, in Dutch, presence versus absence of pre-voicing, but not its temporal extent, is phonologically relevant and, consequently, speakers do not imitate longer versus shorter pre-voicing. Mitterer and Ernestus also examine whether Dutch speakers imitate different variants of Dutch /r/ (alveolar and uvular trills) and further claim that imitation occurs on an abstract phonological level. The two variants of Dutch /r/ are different articulatorily but equivalent phonologically, and the speakers hardly deviate from their habitual articulation to imitate the variant they have heard. Moreover, the response latency does not increase due to a gestural mismatch between the model speech and participants' response. Based on these findings, Mitterer and Ernestus claim that phonetic details,