Pii: S0010-0277(01)00134-2

Caramazza and Costa, 2000 (Cognition 75, B51±B64) report three picture-word interference experiments testing the response set mechanism of the WEAVER11 model of spoken word production. They argue that their ®ndings are problematic for WEAVER11 and that the model's architecture needs to be changed. I show that there is no need to fundamentally modify the model. Instead, the ®ndings of Caramazza and Costa, and all previous ®ndings, are explained by assuming that only a limited number of responses can be kept in short-term memory and that memory improves with response repetition. q 2001 Elsevier Science B.V. All rights reserved. Caramazza and Costa (2000), henceforth C&C, report three picture-word interference experiments testing the WEAVER11 model of spoken word production (e.g. Levelt, Roelofs, & Meyer, 1999; Roelofs, 1992, 1993, 1997). Participants had to name pictured objects while simultaneously trying to ignore superimposed written words. In Experiments 1 and 2, semantic inhibition effects on picture naming latencies were obtained from distractor words that were not part of the response set and with one picture per semantic domain. In Experiment 3, the size of the semantic inhibition effect did not differ between response and non-response distractor words. According to C&C, these ®ndings demand a fundamental modi®cation of WEAVER11. aIt is not obvious that minor changes to the model ± that is, changes that do not alter the fundamental architecture of the model ± would be successful in this regardo (C&C, p. B61). They conclude that aif one were willing to drop the response set principle used in WEAVER11, the new model would have to be able A. Roelofs / Cognition 80 (2001) 283±290 283 Cognition 80 (2001) 283±290 www.elsevier.com/locate/cognit 0010-0277/01/$ see front matter q 2001 Elsevier Science B.V. All rights reserved. PII: S0010-0277(01)00134-2 COGN I T I O N * Tel.: 131-24-3521347; fax: 1 31-24-3521213. E-mail address: [email protected] (A. Roelofs). to account for the data reported here and the various other data that were previously used to support the old WEAVER11 modelo (C&C, p. B61). In this paper, I show that there is no need for a fundamental change of WEAVER11. Instead, the supposedly problematic ®ndings of C&C, and all previous ®ndings that support the model, are explained by assuming that aa response set is only marked in memory when the number of responses is small and can be kept in short-term memoryo (Levelt et al., 1999, p. 62) and that response repetition helps establishing a response set in memory. The mental lexicon is realized in WEAVER11 as a network that is accessed by spreading activation. In word retrieval, activation spreads from a conceptual stratum with concept nodes to a syntactic stratum with lemma nodes. A lemma node is selected when it reaches a critical difference in activation relative to the other lemma nodes. WEAVER11 has been tested on classical ®ndings from the picture-word interference task, in particular, on semantic inhibition effects obtained with picture naming and semantic facilitation effects obtained with picture and word categorizing (i.e. Glaser & DuÈngelhoff, 1984). C&C do not mention the categorization data, but the performance assumptions made for WEAVER11 (and tested by C&C) cannot be understood without reference to these data. Picture naming (e.g. saying adogo to a pictured dog) is slower with semantically related distractor words (e.g. acato) than with unrelated distractor words (e.g. atreeo). However, when the pictured dog has to be categorized by saying aanimalo, then the distractor word acato speeds up the naming response relative to distractor atreeo. A semantic facilitation effect (of over 150 ms) is also observed when the words have to be categorized and the pictures serve as distractors. WEAVER11 explains the semantic inhibition effect with picture naming as follows. In saying adogo to a pictured dog, the semantically related distractor word acato increases the response latency because the picture of a dog activates the distractor lemma cat via the connection between the concept nodes DOG(X) and CAT(X). Such areverse primingo does not happen for semantically unrelated distractors such as atreeo. Consequently, it takes longer before the target lemma dog reaches the critical difference in activation when the distractor is the word acato than when it is the word atreeo. To account for the difference in semantic effect between naming and categorizing, inhibition versus facilitation, it was assumed that response selection can be limited to a restricted set of words. In particular, in a categorization task only hyperonyms such as aanimalo, aplanto, and so forth, are considered for selection. In WEAVER11, this was achieved by marking the permitted responses in memory and applying the response criterion to these marked responses only. Consequently, in saying aanimalo to a pictured dog, distractor acato primes the target lemma animal but cat is not a competitor because it is not a marked response. In contrast, distractor atreeo primes the competitor lemma plant. This causes the semantic facilitation effect in categorization. Computer simulations showed that the model quantitatively ®ts the semantic inhibition and facilitation effects and their exact time course (Roelofs, 1992). Furthermore, semantic facilitation effects have also been obtained from non-response distractors in picture naming in subsequent studies (Roelofs, 1992, 1993), con®rming predictions of the model. A. Roelofs / Cognition 80 (2001) 283±290 284