Phonetic reduction is a common feature of everyday speech. Numerous studies have documented that words, syllables, and other linguistic elements which are more predictable are pronounced with less acoustic prominence than words, syllables, and elements which are less predictable. This phenomenon is referred to as predictability-based phonetic reduction.
Several accounts of this phenomenon exist. The focus of this dissertation is the listener-oriented account, which theorizes that predictability-based phonetic reduction arises from an interaction between the competing forces of conservation of effort and conservation of intelligibility. From this perspective, talkers use the least effort possible to provide the maximum level of comprehension---that is, their speech productions are guided by consideration of the listener's needs. For elements which are predictable, likelihood of comprehension is relatively high, and thus the talker is free to conserve effort and produce the element in a phonetically reduced way. For unpredictable elements, however, likelihood of comprehension is relatively low, and thus the talker must produce the element in a clear manner.
This dissertation presents the results of nine experiments examining different aspects of the listener-oriented account. The first three experiments tested the prediction of this account that individual theory of mind ability is positively correlated with extent of phonetic reduction. Results suggest that no such relationship exists for the variables of lexical frequency, phonological neighborhood density, and second mention reduction. For semantic predictability, however, a negative correlation was observed, such that talkers with poor theory of mind ability had a greater extent of phonetic reduction than talkers with good theory of mind ability. These results fail to support the listener-oriented prediction.
The next three experiments tested the foundational assumption of the listener-oriented account that unreduced speech is easier for the listener to process than reduced speech. This assumption was tested at multiple levels of processing: subjective judgements of speech clarity, speech intelligibility, lexical decision, and semantic acceptability. Results suggest that, at all of these levels of processing, unreduced speech facilitates lexical retrieval relative to reduced speech. This finding is consistent with the assumption of the listener-oriented account. However, significant and systematic individual variation in responses was observed to be partially modulated by theory of mind ability. This unexpected finding is not predicted under a listener-oriented account.
The final three experiments used phonetic corpus analysis to investigate the extent to which predictability-based reduction can be attributed to a single factor, such as listener orientation. Results suggest that at least two and probably three factors are required to adequately model these effects: lexical and contextual factors are distinct, and contextual factors could be further split into discourse-specific and domain-general factors. This result is not consistent with major theory of predictability-based reduction, but can be accounted for under a hybrid model, combining an egocentric model of common ground with an exemplar-dynamic model of the lexicon.
Taken together, these results suggest that the listener-oriented account enjoys limited explanatory adequacy. The results and implications for this study are discussed in terms of our understanding of speech production, individual differences, and speech communication.