The acquisition of a phonological grammar requires the segmentation of an utterance into individual consonants and vowels as a first step, yet it is often taken as a given. I show that segmentation is not a trivial problem by drawing evidence from Mandarin Chinese, where the shortage of morphophonological processes leads to ambiguity in the segmentation of prenuclear glides. I present two language game experiments, in which Mandarin speakers are asked to disassemble syllables in their language, thus revealing their segmental structure. The task, based on fanqie secret languages, involves taking a disyllabic word and swapping its two onsets, in order to form a codeword. What the speaker does with the glide informs us on how they have segmented the sound. A key finding of the experiments is that /j/ is more likely to be treated as an independent segment after non-palatal onsets, compared to palatal ones. Speaker variation, both interspeaker and intraspeaker, is observed. Nevertheless, Mandarin speakers converge on three distinct types.

Two dialects of Mandarin Chinese, Beijing and Liaoning, display different patterns in r-suffixation, a morphophonological process. This paper investigates the low vowel rhymes: /an/, /a/, /aŋ/ in the two dialects. In Beijing, /an/ and /a/ undergo neutralization with the addition of the diminutive /-r/ suffix (Zhang 2000). But in Liaoning, the two rhymes maintain their contrast in the suffixed forms by each employing a different /r/ allophone for the suffix. /an/ selects for the retroflex rhotic [ɻ], while /a/ opts for the non-retroflex, or bunched, [ɹ] in Liaoning. I present a contrast preservation analysis to account for the difference between the two dialects. The analysis is set in the framework of Flemming’s (1995) Dispersion Theory. It employs MinDist constraints and PreserveContrasts-BD, which is a modified version of Flemming’s MaximizeContrasts, adapted to refer to output-output mapping. I argue that the overarching objective for both dialects is to maintain salient contrasts between the suffixed forms, but they differ in whether they prioritize preserving the three-way contrasts of low vowel rhymes or maintaining enough auditory distance between the suffixed rhymes. Liaoning prioritizes preserving the three-way contrasts, ranking PreserveContrasts-BD alongside MinDist, whereas Beijing is less preoccupied with keeping three suffixed forms, ranking PreserveContrasts-BD below MinDist. This is why the contrast between /an/ and /a/ is preserved in Liaoning via /r/ allophony, but neutralized in Beijing. Acoustic evidence for the two allophones of the r-suffix is provided. Segmental processes involving the low vowel and the nasal coda are also discussed.