Productive and perceptual constraints on speech-error correction

Two incentives underlie the present study on speech-error detection and correction. First, this area of research has up to now almost completely been approached through experimental techniques. Since it is not all clear whether speakers' detection and correction behaviour is identical inside and outside the laboratory, a comparison is made between experimental and naturalistic data. While the experimental materials are taken from the literature, the naturalistic findings are based upon the analysis of a corpus of more than 6,000 German slips of the tongue. It is shown that the same trends emerge in both data sets, thereby confirming the ecological validity of the experimental, and the reliability of the naturalistic, results. Secondly, the question arises as to the reasons for error correction and its occasional failure. Two working hypotheses are explored. Speakers fail to correct their errors because they have not detected them or because they assume that the error does not interfere with the listener's decoding process. The former reason is understood as a productive, the latter as a perceptual, constraint on the correction of self-produced errors. The empirical analysis discloses a large overlap between the effects of perceptual and productive constraints. However, whereas perceptual constraints can be subsumed under productive ones, the reverse is not possible. On the basis of this outcome it is argued that productive constraints are primary, and perceptual constraints secondary, reference points for error correction. Although the empirical data do not require the postulation of perceptual constraints, it is suggested that both speaker- and listener-based aspects form part of a highly integrated processing system.