| A corpus of 1053 speech errors made by Mandarin speakers is examined for evidence regarding a number of controversial issues in Mandarin phonology. The data support the following claims: (1) The similarity of consonants as defined by the number of shared features effects the frequency with which two segments are mutually involved in syntagmatic errors. Affricates in Mandarin are not as unitary as affricates in English, as they can be split in errors. The dentals and palatals come from the same underlying source, but the dental/retroflex (and probably velar) contrast is neutralized on the surface in palatal environments. (2) Vowel features are affected hierarchically in errors: the feature [round] is violated most often, and the features [high] and [tense] are violated least often, which can be explained in part by Lin's (1989) underspecification model. Both syntagmatic phonological errors and paradigmatic lexical substitution errors support the five vowel system /i, y, u, e, a/ proposed by Lin (1989). (3) Data from paradigmatic substitutions suggest that there is no syllable structure represented in underlying form, and the onset-rhyme analysis best accounts for Mandarin syllable structure in the surface form. In surface representation, syllables in Mandarin can have from one to 3 X-slots, partially following Chung's (1989) analysis for Kejia, in that syllables may have a branching rhyme, nucleus or X-slot under the onset or nucleus. However, two branching X-slots may not co-occur, and a branching nucleus and rhyme may not co-occur. (4) In underlying representation, tones are stored with phonological representations and may be linked to rhymes. During phonological processing, tones are unlinked from segments, as segments and tones participate in errors separately. After all errors occur, tones become linked to segments, and tone sandhi applies to the resultant string. Tone does not function as a prosodic organizer of the syntagmatic string, as stress does in English; rather tone has a lexical function. Mandarin tones are underlyingly unitary and are not made up of a sequence of level tones; no tone blends or tone spreading errors were observed. (5) Nonlinear phonological theories, particularly those with a clear account of underlying representations, best explain these findings. |