James Magnuson,1 Michael K. Tanenhaus,2 & Richard Aslin2
1 Columbia University, 2 University of Rochester
Swinney [1] and Tanenhaus et al. [2] reported a temporal dissociation of top-down and bottom-up information in word recognition. Both found syntactic frames do not immediately prevent homophone activation: at offset, "flower" and "stood" are primed by either "they all bought a rose" or "they all rose." 200 ms later, only the consistent homophone is primed [2]. This suggests word recognition operates initially on the bottom-up signal, and syntactic constraints are integrated later.
This is inconsistent with recent work indicating that (even non-linguistic) constraints are integrated immediately in language processing [3, 4]. However, verb- and noun-biased frames provide weak constraints; expecting one of thousands of nouns rather than one of thousands of verbs doesn't narrow interpretation much. Shillcock and Bard [5] tested this hypothesis with homophones of closed-class words ("wood," "would") and found immediate integration (i.e., a spoken context predicting the closed class form, "would," never primes "timber"). A related perspective suggested by recent modeling work on frequency effects [6] is that late detection of an influence may reflect graded integration rather than a second stage of processing. In the current experiments, we examined whether syntactic biases based on strong pragmatic context could constrain word recognition.
Adjective use depends on the set of possible referents. Suppose someone wants a red bowl from a set of objects. If there is also a blue bowl, a descriptive NP is required (e.g., "hand me the red bowl"). If there is only one bowl, using the adjective approaches infelicity.
An artificial lexicon was used so that overlap, length and frequency could be precisely controlled. Participants learned 9 nouns (referring to shapes) and 9 adjectives (textures). Each artificial word had competitors in the same form class (e.g., noun-noun) and in the other form class (noun-adjective). We constructed contexts in which a descriptive NP (adjective-noun) was required for unambiguous reference (two exemplars of two types, with different textures) or infelicitous (four types).
After a learning phase, we tested whether phonologically overlapping adjectives and nouns competed when the context favored one form class. On each trial, participants were instructed to touch one of the objects. Adjective use conformed to visual constraints. We only found competition between items from the form class consistent with the context, suggesting immediate integration of top-down information. This supports the hypothesis that top-down information is available immediately in language processing, but detecting evidence of top-down integration early depends on the strength of the constraint.
References
[1] Swinney, D. A. (1979). Lexical access during sentence comprehension: (Re)consideration of context effects. J. Verbal Learning and Verbal Behavior, 18, 645-659.
[2] Tanenhaus, M. K., Leiman, J. M. & Seidenberg, M. S. (1979). Evidence for multiple stages in the processing of ambiguous words in syntactic contexts. J. Verbal Learning & Behavior, 18, 427-440.
[3] Sedivy, J. C., Tanenhaus, M. K., Chambers, C. G., & Carlson, G. N. (1999). Achieving incremental interpretation through contextual representation. Cognition, 71, 109-147.
[4] Tanenhaus, M. K., Spivey-Knowlton, M., Eberhard, K., & Sedivy, J. (1995). Integration of visual and linguistic information in spoken language comprehension. Science, 268, 1632-1634.
[5] Shillcock, R. C. & Bard, E. G. (1993). Modularity and the processing of closed-class words. In G.T.M. Altmann & R.C. Shillcock (Eds.), Cognitive Models of Speech Processing, 163-185.
[6] Dahan, D., Magnuson, J. S., & Tanenhaus, M. K. (2001). Time course of frequency effects in spoken-word recognition: Evidence from eye movements. Cognitive Psychology, 42, 317-367.