Are speech error patterns affected by a monitoring bias?

Abstract

This paper reviews the proposal that observed patterns of speech errors are partly a function of language production mechanisms (that create an underlying pattern of errors), and partly a function of self-monitoring mechanisms (that intercept some errors more often than others and thus alter the underlying pattern). It is essential to find out whether such a ‘monitoring bias’ is real, because of the important role it plays in debates about the interactivity or modularity of the language production system. Data are reviewed that support an important precondition of monitoring bias accounts: The monitor is fast enough to intercept errors in the speech plan. Subsequently, a number of criteria are proposed for allowing a monitoring bias explanation, and five speech error patterns are compared with those criteria. I conclude that in no case is there unequivocal evidence for a monitoring bias, but that such an account is very plausible in the case of some speech error patterns (e.g., the lexical bias effect), but very implausible in the case of other patterns (e.g., morphophonological effects on errors of subject-verb number agreement).

Acknowledgment

Els Severens and Timothy Desmet are thanked for their comments on an earlier draft of this manuscript, and the participants of the International Workshop on Language Production in Marseille (September 2004) are thanked for stimulating discussion on many of the issues reviewed here.

Notes

¹That is, in interactive models like Dell (1986), the input from the sublexical to the lexical level is automatically determined solely by the activation of sublexical units and the weights of the connections.

²In particular, Rapp and Goldrick (2000) reported on two individuals with brain damage, whose impairments were consistent with a semantic locus (patient KE) and a post-semantic locus (patient PW). Whereas PW displayed a mixed error bias, KE did not, and this is difficult to reconcile with a monitoring bias account: There is no independent evidence to suggest that KE, but not PW, had a monitoring deficit. However, Roelofs (2004, 2005) argued that his version of the monitoring account is compatible with the patient findings, under the assumptions that (a) KE's semantic errors are errors of concept selection (because of the semantic locus of the impairment) whereas PW's semantic errors are errors of lemma selection (because of the post-semantic locus); (b) the cohort effect in speech perception results, in case of a mixed competitor (calf), in a relatively high activation of the target word (cat) at the lemma level; (c) but as a result of rather weak connections between lemma and concept levels, there is little propagation of this activation to the concept level. The concept CAT(X) would become about as active for a mixed and a semantic-only error. Therefore, KE, who can only detect the error by checking whether the selected concept is the right concept, is equally likely to detect the error calf and dog for cat, while PW, who can detect the error by checking whether the selected lemma is right, is more likely to detect the error dog. Rapp and Goldrick (2004), Footnote 8), criticized this proposal on the basis of plausibility. Because the connections between lemma and concept level are also used for language comprehension, it is unclear why these connections would be so weak that differences in activation levels at the lemma are'washed out' at the concept level lemma.

³A speech-shadowing study in Dutch (Cohen, 1980) showed that participants indeed detected perseverations and anticipations at comparable rates (65% vs. 56%).

⁴Additionally, the effect might follow from a mechanism of post-selection inhibition, which is incorporated in the model of Dell (1986). In order to prevent perseveration, the activation of a just-selected unit would be turned off, making it more difficult to select the unit again. Whether that explanation is sufficient to account for this effect remains to be seen however, because the activation level of the recently selected unit will quickly'bounce back' (because of that unit's connections with highly active units at the lexical level).

Additional information

Notes on contributors

Robert J. Hartsuiker

Correspondence should be addressed to Robert J. Hartsuiker, Department of Experimental Psychology, Ghent University, Henri Dunantlaan 2, 9000 Ghent, Belgium. [email protected]