Surviving truncation : informativity at the interface of morphology and phonology

When disyllabic words of Chinese are compounded, truncation often applies to yield a disyllabic output that draws one syllable from each of the contributing words, e.g., xiàdiÄ“ + fúdù → [InlineEquation not available: see fulltext.] + [InlineEquation not available: see fulltext.] → diÄ“fú. A substantial portion of new words enter Chinese via this process, and all combinations of four underlying syllables are attested in disyllabic neologisms. Variation in which syllable of the base escapes truncation has been described as arbitrary, but our analysis uncovers a clear pattern. Informative syllables survive. We fit a logistic regression model to Chinese truncation patterns using two indices of informativity, constituent family size and frequency ratio. Both of these factors were significant predictors. To further explore the nature of informativity-based constraints on surface forms, we analysed the truncation probabilities predicted by the model for each base in the neologism corpus. As truncation probabilities increased so too did model accuracy. We interpret this result as evidence for a constraint regulating the degree of uncertainty in base-truncation mappings.