Abstract
In this study, phonological errors of speech in Taiwan Southern Min (TSM) were observed analyzed to show the factors affecting segmental errors and tone errors on the one hand, and how these errors shed light on the phonological structures of the language on the other hand. The segmental errors observed in the data showed that the contextual errors occurred typically within 7 syllables from the source of errors, that the target syllable usually shared phonetic similarity with the source syllable, and that Markedness hypothesis did not seem to have an effect on the errors. The tonal errors showed that the language-specific tone sandhi phenomenon had strong effect on the errors, that the high-falling tone as a target was most likely affected and the errors most commonly occurred as the level tones, and that the tone sandhi errors occurred mostly in relation to the syntactic structure of the target word. We next tried to see whether the errors may shed light on the theoretical issues. The /iu/-/ui/ issue did not find a conclusive answer, due to the paucity of relevant error tokens. The syllable internal structures were found to be more appropriately right-branching, like the onset-rime model. As for the tone sandhi mechanism, the selection model better conforms to the patterns shown in tone sandhi errors than the rule-derivation model.