FREQUENCY AND AUTOMATICITY 161 RESEARCH ON SPOKEN LANGUAGE PROCESSING Progress Report No. 28 (2007) Indiana University Frequency of Use Leads to Automaticity of Production: Evidence from Repair in Conversation 1 Vsevolod Kapatsinski Speech Research Laboratory Department of Psychological and Brain Sciences Indiana University Bloomington, Indiana 47405 1 I would like to thank the NIH for financial support through Training Grant DC-00012 and Research Grant DC-00111 and to Adam Albright, Tessa Bent, Joan Bybee, Dan Jurafsky and the audiences at HDLS VII and the Workshop on Gradience and Frequency Effects in Phonology for helpful comments.KAPATSINSKI 162 Frequency of Use Leads to Automaticity of Production: Evidence from Repair in Conversation Abstract: Investigation of spontaneous replacement repairs found in the Switchboard Corpus (Godfrey et al., 1992) shows that low-frequency repaired words are more likely to be interrupted prior to replacement than high-frequency words are. These results provide novel empirical support to the hypothesis that the production of high-frequency words is more automatic than the production of low-frequency words (Bybee, 2002; Logan, 1982). The relationship between the effects of frequency on interruptibility is argued to be partially mediated by the effect of frequency on duration. In addition to testing the link between frequency and automaticity, the present paper reports that replaced words tend to be more frequent than the words that replace them, providing support for the hypothesis that high-frequency words are easier to access in word production, which has been criticized on the basis of not observing this frequency asymmetry in semantic substitution errors (Garrett, 2001). Finally, whether a word is interrupted is found to depend strongly on the length of the word, with long to-be-replaced words being more likely to be interrupted than produced completely. Thus, while speakers prefer to produce constituents with a continuous delivery (Clark & Wasow, 1998), the drive to produce a continuous constituent competes with the drive to interrupt as soon as possible (Main Interruption Rule, Levelt, 1983, 1989). Introduction Theoretical Background Bybee (2002) suggests that the production of high-frequency words and phrases is more automated than the production of low-frequency words and phrases. Under this hypothesis, high-frequency words are more cohesive than low-frequency words: the parts forming a high-frequency word are more tightly linked together than the parts forming a low-frequency word. Previous evidence for a link between cohesion and frequency has come from studies showing that high-frequency words are more likely to undergo reductive sound change (Bybee, 2002; Hooper, 1976). Mowrey and Pagliuca (1995; Pagliuca & Mowrey, 1987) go as far as claiming that all internally-motivated regular sound changes in progress that have been attested can be explained by an increase in gestural compression. Bybee (2001: 79-83) and Phillips (2001) suggest that there are other sources of sound change but that Mowrey and Pagliuca’s claim holds for sound changes that involve lexical diffusion from high-frequency to low-frequency words. An increase in the temporal overlap between successive gestures and temporal compression of the sequence of articulatory goals corresponding to a word is expected to result from automatization of word production (Bybee, 2002). Assuming that in a sequence of articulatory goals, a goal gains control of articulation when it is activated sufficiently, and that activation spreads from earlier goals to later ones, a goal will receive control of articulation earlier when it is strongly connected to the preceding goal. Thus, the preceding goal is less likely to be completely reached when the following goal is highly predictable in the context. In addition, when the gestures called for by successive goals do not interfere with each other, which could cause undershoot, articulatory overlap between gestures implementing successive goals is more likely in a high-frequency sequence. Under this account, a high-frequency word is a more cohesive unit than a low-frequency word.FREQUENCY AND AUTOMATICITY 163 However, the finding that reductive sound changes start in high-frequency words has also been interpreted as indicating that speakers do not expend as much articulatory effort in such words because of their high contextual predictability for the listener (e.g., Bybee 2002: 269; Gregory et al., 2000; Lindblom, 1990). Fowler (1988) shows that words that have already been mentioned in the course of the conversation are shorter than words that are mentioned for the first time (see also Fowler and Housum, 1987) but only if the two tokens are co-referential. Words are not shortened if a homonym has recently been pronounced but are shortened if preceded by a synonym. Fowler (1988: 317) writes that “production of a homophone of a target… is not sufficient to yield shortening… even though the word’s articulatory routine has recently been used. Apparently the shortening reflects the talker’s estimate that a listener has other information available to help identify the word”. Gregory et al. (2000) support this interpretation by showing that semantic relatedness to the discourse topic influences word duration even when repetition is controlled: words related to the discourse topic are shorter than unrelated words. Under this alternative interpretation, word frequency does not directly influence gestural compression, automaticity of production, or word cohesion. Rather, frequency is simply one of the factors that influences contextual predictability, which serves as a constraint on how much reduction the speaker thinks s/he can get away with. In the present paper, we investigate a hitherto untested prediction of the hypothesis that the production of high-frequency words is more automatic than the production of low-frequency words. As Anderson (2000: 99) puts it, “automaticity occurs when practice eliminates most of the need for central cognition”, which leads to the behavior becoming relatively impervious to cognitive