University of Pennsylvania Working Papers in Linguistics


Most first language (L1) attrition research focuses on syntactic and morphological deterioration in environments where L1 ‘attriters’ rarely have contact with their L1, such as immigrants. There is no study on L1 attrition in tones and in contexts where L1 can still be often heard. This study examines this attrition type in Taiwan, where the attriters cannot speak their L1 Taiwanese fluently and have become L2-Mandarin dominant after five years old. This study investigates L1 attriters’ Taiwanese tonal system by evaluating tonal attrition hypotheses based on four of the six L1 attrition proposals in the literature. The data in this study are composed of natural speech provided by 10 L1 Taiwanese attriters, 6 older L1 Taiwanese non-attriters, and 5 younger L1-Taiwanese L2-Mandarin bilinguals. The participants performed a film retell and a story-telling task. The results indicate that Taiwanese tone sandhi is so ingrained in the attriters’ phonology that the attriters are still capable of accurately performing tone sandhi (approximately 90% accurate). Although the attriters have become L2 Mandarin dominant, L2 interference is not observed. Given these findings, the L1 Taiwanese tonal attrition hypotheses referencing L2, such as the simplification hypotheses, cannot account for the attriters’ system. Rather, attrition hypotheses referencing L1 acquisition, such as the threshold hypothesis, are more successful at accounting for the attriters’ tonal system. There is still a larger question as to why the complex Taiwanese tone sandhi system is so well maintained in light of the presence of the competing dominant L2 Mandarin tonal system. Perhaps it is because the Taiwanese system is so complex. Tone sandhi interacts with the syntax, the semantics, the morphology and even the segmental phonology of the language. If the complexity was learned before the onset of massive exposure to Mandarin, then we can revise slightly the threshold hypothesis by saying complicated structures that are learned well are unlikely to attrite.