A key issue in models of lexical production is whether a speaker's model of hearer knowledge has effects on production. Some researchers claim that lexical production is sensitive to egocentric, or speaker-only knowledge (Bard et al., 2000; Brown & Dell, 1997; Horton & Keysar 1996). Others claim that production is additionally influenced by speakers' assumptions of what is given to the hearer (Clark & Wilkes-Gibbs, 1986; Fowler & Houssum, 1987; Lieberman, 1963). It has been shown that word length is influenced by the number of times it has been said in a conversation: Words get shorter upon subsequent use (Bard, Lowe, & Altmann, 1989; Fowler, 1988; Fowler & Houssum, 1987). Speaker-based models claim that this shortening is a byproduct of speaker internal processes, such as lexical priming. Alternatively, hearer-based models claim that this effect is a result of the speaker's assumptions concerning hearer-knowledge (Bolinger 1972; Fowler & Houssum 1987).
Exploiting the fact that words become shorter upon subsequent mention, Bard et al. (2000) found that speakers continue to attenuate words even in the presence of a new hearer, concluding that speakers do not make use of hearer knowledge during articulation. Although they provide evidence of speaker-based shortening, the Bard et al. results do not rule out an additional effect of hearer knowledge. Would the presence of an old hearer produce even more attenuation?
In a study of word duration in conversation, we extend Bard et al.'s (2000) results by demonstrating that speakers do shorten words even more in the presence of an old hearer. In a study in which participants were asked to repeat a short story, we compared the first use of a referring expression in Narration 1 to the first use of the same referring expression in Narration 2 and varied the discourse status of the hearer. In the repeated hearer condition, speakers told the story to the same hearer. In the new hearer condition, speakers told the story to a new hearer. An analysis of variance indicates a significant interaction between narration and condition, F(1, 21) = 5.67, MSE = .001, p < .05, reflecting the fact that the first uses of referring expressions in Narration 2 were shorter than those in Narration 1 for the repeated hearers; t(21) = 2.56, p < .05, but not for the new hearers.
These results indicate that the speakers assumptions concerning hearer
knowledge play a role in production. Speakers have a model of hearer
knowledge that affects the durations of words. These results provide
evidence that the reduction of predictable words has a communicative
function: The more information the hearer has about a word, the safer it
is to reduce, likely because the hearer has other sources of information
by which to identify it (Fowler, 1988; Jesperson, 1922; Lieberman, 1963).