| Models of spoken word production disagree over whether the process involves feedback from phonological to lexical levels, or whether it is feed-forward only. To address this question, the experiments here investigated the underlying cause of the lexical bias effect—the tendency for phonological speech errors to be more likely when the outcomes are real words rather than nonwords. This lexical effect on phonological processing can be explained either by feedback, (Dell, 1986); or in models where feedback is prohibited, by a prearticulatory lexical editor that selectively aborts nonword output (e.g. Levelt, Roelofs & Meyer, 1999). The present experiments specifically compare these alternative explanations. Experiment 1 elicited phonological speech errors using the Slip procedure (Baars & Motley, 1974), and showed that the lexical bias effect in exchange errors is asymmetrical, with the first word showing a greater lexical bias than the second. This is not predicted by a lexical editor, but is predicted by feedback if exchanges are incremental, where the first part of the error precipitates the second. Two further experiments provided support for this incremental model of exchanges, by showing that more exchanges were created by priming the anticipatory part of the exchange than by priming the perseveration. A fourth and final experiment attempted to replicate the primary piece of experimental evidence for a lexical editor—Baars, Motley and MacKay (1975) Experiment 2, which showed the disappearance of the lexical bias effect when speakers were producing only nonwords, and should therefore not have been editing on the basis of lexicality. However, Experiment 4 did demonstrate significant, and equal lexical biases in nonword and word contexts, contra Baars, Motley and MacKay (1975) and the predictions of a lexical editor. These experiments provide the most precise tests to date of these alternative explanations of the lexical bias effect. The results argue strongly for feedback, rather than a lexical editor mechanism underlying lexical bias in speech errors, and therefore necessitate the inclusion of feedback from phonological to lexical levels in models of language production. |
