UC Berkeley

Studies of connected speech have repeatedly shown that the contextual predictability of a word is related to its phonetic duration; more predictable words tend to be produced with shorter duration, when other factors are controlled for (Aylett & Turk, 2004, 2006; Bell et al., 2003, 2009; Gahl, 2008). Speaker-oriented accounts of phonetic variation posit that the probability of a word in its context is related to its accessibility for planning processes (Jurafsky et al., 2001; Bell et al., 2003; Gahl et al., 2012), with greater accessibility giving rise to shorter durations. However, the mechanism by which accessibility relates to phonetic duration has not been fully elaborated.

This dissertation presents data relevant for understanding the connection between lexical accessibility and phonetic duration, and it argues that the relevant mechanism rests at the level of phonological encoding. The motivation and interpretation of the studies assume an interactive spreading activation model of speech production (Dell, 1986, 1988) that includes both sequential encoding of phonological segments (Sevald & Dell, 1994) and cascading activation flow.

The primary hypothesis is that examining the effects of phonological neighborhood structure on the phonetic duration of words and segments can shed light on the connection between lexical activation and duration. Three experiments are undertaken to better explain the relationship between neighborhood structure and phonetic duration. Experiment 1 pursues a novel word learning study, in which preschool children are taught words that create minimal pair relationships with already known words. The results indicate that the specific phonological relationships between the words in children's lexicons do not have an appreciable effect on articulatory duration. However, the significant effect of phonotactic probability on children's articulation suggests that the relatively great importance of practice effects (at either the phonological or motor level) may have overshadowed any influence of lexical activation.

Experiment 2 provides a reanalysis of data from a previous study that found an effect of minimal pair neighbors on phonetic duration in adults' single word productions. Baese-Berk and Goldrick (2009) found that words with a voicing-initial minimal pair neighbor (e.g. "cod", which has a neighbor "god") were produced with longer voice onset time (VOT) than words without such a neighbor (e.g. cop). The reanalysis presented in the dissertation suggests that this effect may not be related to minimal pair status per se, but rather to the number of neighbors in the lexicon that differ in their initial consonant segment. In addition to the findings for VOT, when segmental content is taken into account, words with more total phonological neighbors are produced with shorter rime duration. These findings support the idea that positional overlap between phonologically related neighbors is facilitatory for phonological encoding, with facilitation being reflected in the time to articulate a given segment.

Experiment 3 asks whether the results from the study of adult single word productions can be extended to spontaneous speech. An analysis of the Buckeye Corpus of Conversational Speech (Pitt et al., 2007) indicates that they can; monosyllabic English words beginning with voiceless stop consonants are produced with longer VOT when they have relatively more neighbors differing in the initial consonant. The relationship between total neighborhood density and rime duration is not found to be significant, but the numerical pattern is in the predicted direction.

To account for the observed relationship between positional phonological overlap and shorter phonetic duration, the Articulate As Soon As Possible Principle (AASAPP) is proposed. The AASAPP posits that the articulatory plan for a given segment is initiated and executed as quickly as possible, and that the time course for the production plan is related to the activation level of the target segment at the time of selection. Positional competition between co-activated segments is argued to be associated with longer articulatory duration because it slows the process of phonological encoding, while positional overlap is associated with the facilitation of encoding and therefore, articulation.