The activation—selection model of meaning: Explaining why the son comes out after the sun

VINCENT R. BROWN 0 0 Hofstra University , Hempstead, New York The activation-selection model (ASM) of determining the meaning of an ambiguous word is unique in that it is able to account for the long-term effects of meaning selection without an explicit mechanism for suppressing the representation of the nonselected meaning. The model assumes that a meaning is selected when a threshold number of attributes associated with that particular meaning are activated. When a meaning is selected, the ASM assumes that the weights of the attributes that are associated with the chosen meaning are increased. This two-phase process (transient activation followed by long-term weight changes) provides a mechanism by which meaning selection at one time can affect meaning selection at a much later time. The ASM can explain the results of the presently reported experiments, in which the meaning selected for a homophone presented in an unbiased context is affected by multiple previous presentations of the homophone in different contexts. In particular, although participants who are initially oriented toward the secondary meaning of a homophone show an increased proportion of dominant responses when next primed by the dominant meaning of the homophone, the proportion of dominant responses decreases to below baseline levels when the homophone is later presented in a neutral context, indicating the lasting influence of the initial secondary meaning context. - Prior to the late 1980s, research on the semantic ambiguity of single words had focused on the process of lexical access. Although various researchers favored different forms of access (exhaustive, context-dependent, ordered; see Simpson, 1984, for a review), the common tendency was to treat each occurrence of an ambiguous word as an independent event. A more recent trend in the ambiguity processing literature has been to examine the effect of the repetition of an ambiguous word in situations in which the appropriate meaning of the ambiguous word varies across occurrences. Changing the contextually appropriate meaning of a homograph across trials in a task can result in large decrements in performance, such as when an individual is required to decide if seal is related to walrus after earlier deciding that seal is related to glue. Maintaining the same meaning context generally facilitates performance, for example, deciding that seal is related to walrus shows a benefit if earlier the participant had decided that seal is related to dolphin (for reviews of this literature see Gorfein, 2001a; Simpson & Kang, 1994). Although Simpson and Kang (1994) have argued that in normal discourse, words are often repeated and therefore the long-term effects of processing each occurrence of an ambiguous word need to be understood, surprisingly little theoretical work has addressed the issue. Simpson and Kellas (1989) reported that once a homograph had been used as a prime for a word related to one of its less common meanings (e.g., bank as a prime for river), a target word related to the dominant meaning of the homograph (e.g., bank followed by money) showed a prolonged naming time in comparison with a neutral baseline, even after a lag of 12 intervening trials. Simpson and Kang (1994) concluded that Processing one meaning of a homograph and responding to that meaning results in the active and specific inhibition of competing meanings (p. 376, italics added), thus extending the idea of transient suppression of the nonselected meaning of an ambiguous word to a longer term inhibitory process that endures over a period of time. From a theoretical perspective, Simpson and Kang (1994) suggested that the observed decrement is a form of negative priming (see, e.g., Tipper, 1985), but did not present a specific model. Simpson and Adamopoulos (2001) reemphasized the inhibition interpretation of negative priming (but cf. Neill & Valdes, 1996, for a noninhibitory explanation of negative priming). Gernsbacher, Robertson, and Werner (2001) reported ation times in what is known as the subordinate-bias effect that for sentence sensibility judgments, there is a large cost (Duffy, Morris, & Rayner, 1988).1 Duffy et al. (2001) emof changing the meaning of a homograph on consecutive ployed the constraint satisfaction architecture of Spivey and sentence trials. Participants are slower and less accurate in Tanenhaus (1998) competitive integration model to credeciding that the sentence She blew out the match made ate a computational version of the reordered access model. sense after deciding that the previous sentence, She won the Two sets of constraints are employed: the balance of the match, was sensible. However, in contrast to the cost found homograph, which is a measure of the relative frequency in the naming task used by Simpson and Kang (1994) over of alternative meanings, and the weight of the semantic 12 intervening trials, Gernsbacher et al. found that after context. These constraints combine to produce a candidate four intervening sentences, there was no cost associated meaning for the homograph. The weights of these proviwith the meaning change. The duration of the decrement sional interpretations feed back to the constraint represenassociated with meaning change appears to vary with the tations, resulting in the re-weighting of both constraints. way that the local meaning context is presented, and per- The re-weighting continues until a threshold is reached haps also with the methodology used to assess the effects. and a meaning is selected. The number of cycles to reach Gernsbacher et al. suggested that their data could not be threshold determines the reading time for each combination accounted for by models that do not involve mutual inhibi- of homograph and context. The reordered access model is tion or suppression between alternative meanings (such as not formulated to account for long-term effects of meaning the episodic-retrieval model of Neill & Valdes, 1996). selection over many intervening trials or long time inter Gernsbacher (1990) proposed a structure-building vals, but we believe it could be modified to do so. Although model as an explanation of the immediate effects of pro- not explicitly stated by the authors, the re-weighted meancessing words in a sentence context. This model includes ing frequency values could be employed in a subsequent active suppression of inappropriate-to-the-context mean- encounter with the homograph, thus producing a long-term ings during initial processing of a homograph, and a signal effect of choosing a particular meaning. (Since the model is transmitted by activated memory cells that promotes designed to account for processing times, it would also have the facilitation of related-to-the-context interpretations. to be modified to account for response proportions.) Gernsbacher and St. John (2001) presented a connection- Another recent model of ambiguity processing, the ist model of meaning selection in which top (...truncated)