The representation of variable tone sandhi in Shanghai Chinese

Introduction. Variation in phonological patterns challenges models of spoken word recognition. One crucial question is how listeners process and represent variant forms in the mental lexicon. It has been shown in behavioral experiments that more frequent forms were responded to more quickly than less frequent forms (e.g., American English word-final t/d deletion, Deelman & Connine, 2001). Ranbom & Connine (2007) further showed that in nt-flapping in English (e.g., [sɛɾɚ] for center), the less frequent form [nt] also had a strong phonological representation in the lexicon, suggesting that both forms of a phonological alternation contribute to lexical access. These studies, however, mainly focused on variation of phonetic reduction processes, which can be interpreted as processes of late phonology (Coetzee & Pater, 2013). The processing of morphosyntactically conditioned phonological alternations, which presumably occur earlier in the derivation, has received considerably less attention. The current study investigates one such alternation — a variable tone sandhi pattern in Shanghai Chinese. The majority of disyllabic verb-noun (V-N) combinations in Shanghai Chinese can undergo a tonal extension tone sandhi, whereby the base tone of the first syllable is spread onto the entire disyllable. But the sandhi applies variably. For example, /tshɑ̃24/ “to sing” + /ku53/ “song” can be realized with either the sandhi form ([tshɑ̃33 ku44]) or the non-sandhi form ([tshɑ̃24 ku53]). In a series of subjective rating experiments, Yan (2016) found that native speakers generally preferred the non-sandhi form over the extension sandhi form in V-N, and that more frequent V-N items preferred the non-sandhi form even more. However, how V-N items with variant sandhi forms are processed during spoken word recognition remains unclear. To this end, we designed an auditorily primed lexical decision experiment in which disyllabic V-N targets were primed by a monosyllable that shared the segments of the first syllable with different tonal conditions. Based on Yan (2016), we expect the prime with the non-sandhi tone to elicit facilitative priming in general, suggesting that the nonsandhi form is represented in the mental lexicon. Moreover, we also predict that the priming effect may be modulated by the familiarity ratings of the V-N items, with the non-sandhi tone prime facilitating the recognition of more frequent V-Ns, and the sandhi tone prime facilitating the recognition of less frequent ones. These results would suggest that both variant forms of a V-N item are represented in the mental lexicon and activated to different degrees depending on frequency of the item. Methods. Thirty-six native Shanghai speakers performed a lexical decision task. Each disyllabic target was preceded by a sandhi tone prime, which shared the same tone with the initial syllable of the extension sandhi, a non-sandhi tone prime, which had the same underlying tone as the first syllable of the non-sandhi form, and an unrelated tone prime, which was not related to the first syllable of the target in tone (see Table 1). Given that the extension sandhi form applies variably in V-N disyllables, the stimuli-recording speaker was asked to produce the disyllables as naturally and comfortably as possible without considering variation. She applied the non-sandhi form to all V-N critical targets, which is consistent with Yan (2016). Hence, the non-sandhi form was used for all the V-N targets in the current study. All primes shared the same segments with the first syllable of the target. After the priming experiment, a familiarity rating task on the disyllabic targets was conducted with the same participants.