Verifying Properties from Different Modalities for Concepts Produces Switching Costs

According to perceptual symbol systems (Barsalou, 1999), sensory-motor simulations underlie the representation of concepts. It follows that sensory-motor phenomena should arise in conceptual processing. Previous studies have shown that switching from one modality to another during perceptual processing incurs a processing cost. If perceptual simulation underlies conceptual processing, then verifying the properties of concepts should exhibit a switching cost as well. For example, verifying a property in the auditory modality (e.g., BLENDER-loud) should be slower after verifying a property in a different modality (e.g., CRANBERRIES-tart) than in the same modality (e.g., LEAVES-rustling). Only words were presented to subjects, and there were no instructions to use imagery. Nevertheless switching modalities incurred a cost, analogous to switching modalities in perception. A second experiment showed that this effect was not due to associative priming between properties in the same modality. These results support the hypothesis that perceptual simulation underlies conceptual processing. Switching costs in property verification 3 Modern psychology relies heavily on the digital computer as a metaphor for human cognition (e.g., Fodor, 1975; Pylyshyn, 1984). According to this view, the software of the mind can be distinguished from the hardware of the body, with mental representations being amodal redescriptions of sensory-motor experience. Increasingly, however, researchers argue that this approach is fundamentally wrong, suggesting instead that interactions between sensory-motor systems and the physical world underlie cognition. For example, Barsalou’s (1999) theory of perceptual symbol systems proposes that conceptual knowledge is grounded in sensory-motor systems. To represent a concept, neural systems partially run as if interacting with an actual instance. For example, to represent the concept CHAIR, neural systems for vision, action, touch, and emotion partially reenact the experience of a chair. Increasingly, behavioral evidence supports this view (e.g., Klatzky, Pellegrino, McKloskey, & Doherty, 1989; Solomon & Barsalou, 2001, 2002; Spivey, Tyler, Richardson, & Young, 2000; Stanfield & Zwaan, 2001; Wu & Barsalou, 2002; Zwaan, Stanfield, & Yaxleu, in press), as does neural evidence (e.g., Martin, 2001; Martin & Chao, 2001; Martin, Ungerleider, & Haxby, 2000; Pulvermüller, 1999). See Barsalou (1999; in press) and Glenberg (1997) for further evidence. Several aspects of sensory-motor simulations are important for the experiments presented shortly. First, simulations are componential, not holistic. Rather than being like a holistic video recording, a simulation contains many small elements of perception— perceptual symbols—organized coherently. Second, perceptual symbols arise on all modalities of experience—vision, audition, smell, taste, touch, action, emotion, introspection, etc. Third, perceptual symbols vary in accessibility. On a given occasion, only those perceptual symbols most active enter a simulation, such that the simulations of a concept vary considerably across occasions. Furthermore—and most importantly for our purposes—the modalities represented in simulations vary as well. On one occasion, the simulation of a concept might focus on how an object looks (e.g., a LEMON is yellow); on another occasion, a simulation might focus on how the object tastes (e.g., a LEMON is sour). Although multiple modalities may typically be represented, one may often be more salient than others. Switching costs in property verification 4 Furthermore, over time, the focus may remain in a single modality, or it may switch from one modality to another. If switching between modalities occurs during conceptual processing, then a phenomenon from the perception literature is relevant. Spence, Nicholls, and Driver (2000) had subjects discriminate whether a signal occurred on the left or the right in any of three modalities monitored simultaneously (i.e., a light in vision, a touch on a finger, a tone in audition). When two consecutive signals occurred on the same modality, processing stayed within a single system. When consecutive signals occurred on different modalities, processing had to switch between systems. Most importantly, Spence et al. found that switching modalities incurred a cost: Detecting a signal was slower when the previous signal was on a different modality than on the same (also see Spence & Driver, 1998). If conceptual processing utilizes sensory-motor systems, then an analogous cost should occur when conceptual processing switches from one modality to another. To investigate this prediction, we used the property verification task. On target trials, subjects verified a property in one of six modalities (vision, audition, taste, smell, touch, action). For example, subjects might verify the auditory property loud for BLENDER. On the previous trial, subjects either verified a property from a different concept on the same modality or on a different modality (e.g., LEAVES-rustling versus CRANBERRIES-tart). Table 1 provides examples of the critical materials. Because the concepts on the two trials were always unassociated, no associative priming between concepts should occur. Also a high ratio of filler trials to critical trials masked the purpose of the experiment (i.e., the number of paired trials on the same modality was relatively small). The key prediction was that having to switch modalities would slow verification time, relative to staying within the same modality, analogous to modality-switching costs in perceptual processing. Experiment 1 also explored whether the stimulus onset asynchrony (SOA) between presentation of the concept and presentation of the property is a factor in switching costs. Perhaps switching costs disappear when properties lag behind concepts, because the concept has longer to activate properties across modalities. Alternatively, switching costs may remain constant across SOAs if subjects do not commit to a dominant modality until receiving the Switching costs in property verification 5 property word. To assess these possibilities, some subjects received the concept and property on each trial simultaneously (SOA=0 ms), whereas others received the concept first, followed by the property 260 ms later (SOA=260 ms). ----------------------------------------------Insert Table 1 about here ----------------------------------------------Experiment 1 Method Subjects and design. Sixty-four volunteers from Emory University participated for course credit. Thirty-two were assigned randomly to each of the two between-subjects conditions for SOA. Same versus different modality was manipulated within subjects, with equal numbers receiving each counterbalanced version of the list. Materials. A set of 100 concept-property items was developed. Each property was more salient on one modality than on the others. We selected 26 properties from vision, 24 from motor actions, 18 from audition, 12 from touch, 12 from taste, and 8 from smell. Because some modalities have more words for properties than others, the number of properties differed across modalities by necessity. From the 100 concept-property items, 50 pairs were formed. Half contained two properties from the same modality; half contained properties from different modalities. The two items forming a same-modality pair were chosen randomly from items on the relevant modality. According to the norms of Nelson, McEvoy, and Schreiber (1999), the properties in these pairs were not associated. One item in each same-modality pair was randomly assigned to be presented first (the context item), and the other to be presented second (the target item). Table 1 presents an example from each modality. The two items comprising a differentmodality pair were chosen randomly from the remaining items. In pairs of both types, if the two concepts exhibited a relation, they were replaced with items having no relation. Two lists were created such that each target had a same modality context in one list but a differentmodality context in the other. Thus each target item appeared with both same-modality and different-modality contexts, counterbalanced across lists. All critical properties were true of their respective concepts. Switching costs in property verification 6 The experimental trials included 150 pairs, with 50 being critical, for a total of 300 trials. The remaining 100 pairs were fillers, designed to mask the nature of the experiment. Within the filler pairs, 50 contained two false items, 25 contained a true item then a false item, and 25 contained a false item then a true item. Thus true and false responses were equally likely overall. Properties in the fillers sometimes referred to a specific modality but also referred to properties that are represented on multiple modalities (e.g., CAMERAcompact, TOY-plastic, MAP-complicated). To ensure that subjects actually verified the properties of concepts (Solomon & Barsalou, 2002), the concept and property in many false items were related (e.g., OVEN-baked, BUFFALO-winged, BUTTERFLY-bird). The critical and filler pairs were randomly intermixed for each subject. All concepts and properties were used only once. The practice trials consisted of 24 true items and 24 false items, similar in nature to the experimental trials. Procedure. Each trial began with a fixation stimulus (* * * * *) two lines above where the concept name would appear. After 500 ms, the fixation stimulus disappeared. In the 0 ms SOA condition, three lines of text appeared aligned vertically, each two lines apart. The first line contained the concept word in upper case; the second line contained the words “can be” in lowercase; the third line contained the property word in upper case. In the 260 ms SOA condition, the concept word appeared for 160 ms, then “can be” was added for 100 ms, then the property name was added. RTs in all cases were measured from the onset of the property word. All lines remained on the screen until the subject made a "true" (?/ key) or "false" (z key) response. The initial instructions stressed that a decision should be based on whether the property was "usually true" of the concept. For example, the pair CARNATION-black could theoretically be true, but black would be a highly unusual property for CARNATION. Therefore, the correct response for such a property was "false". Subjects received feedback for 600 ms after pressing the wrong key (“ERROR”) or after taking 2000 ms or longer to respond (“TOO SLOW”). The next trial began 300 ms after the response, or in the case of feedback, 300 ms after the feedback disappeared. Because subjects responded to each item individually, nothing indicated that items were paired in the underlying design. Also, because Switching costs in property verification 7 only 1 of every 12 trial transitions contained properties from the same modality, it was not obvious that modality switching was of interest. The experiment began with 48 practice trials, followed by the 100 critical and 200 filler trials in a different random order for each subject. After each block of 50 trials, subjects took a brief break and saw the percentage of errors from the previous block. When errors exceeded 15%, subjects were urged to be more accurate. When errors fell below 5%, subjects were complimented. When ready, subjects began the next block. Results and Discussion RTs were removed for target trials on which errors occurred. Target RTs were also removed when subjects erred on the previous context trial, given that an assessment of modality switching assumes that subjects processed both the context and target items correctly. When subjects erred on a context trial, a variety of complicating factors could affect processing on the target trial. Medians for same-modality versus different-modality RTs on target trials were computed for each subject and then averaged. ----------------------------------------------Insert Table 2 about here ----------------------------------------------As Table 2 illustrates, RTs on the target trials were slower when the modality switched from the context trial to the target trial than when modality remained constant, F(1,62)=6.87, p<.05. Although the switching effect was slightly larger in the 0 ms SOA condition than in the 260 ms SOA condition (29 ms vs. 20 ms), the interaction between SOA and switching was not significant, F(1,62)=0.24. No effects occurred for errors, indicating that a speed-accuracy tradeoff was unlikely. The effect of SOA was significant, F(1,62)=56.12, p<.01. Subjects in the 260 ms SOA condition were 270 ms faster than subjects in the 0 ms SOA condition. The near equivalence between the difference in RTs and the difference in SOAs indicates that subjects in the 260 ms SOA condition began task-relevant processing immediately on receiving the concept in isolation. By the time the property arrived, these subjects were further into the necessary processing than the 0 ms SOA subjects. Most importantly, however, the effect of modality switching occurred for both groups. Switching costs in property verification 8 We began with the hypothesis that modality-specific brain areas represent properties in concepts. Based on this assumption, we predicted that switching modalities while verifying properties would incur a processing cost, analogous to the cost incurred while switching modalities in perceptual processing. Unlike perceptual studies, however, the switching costs here occurred while subjects processed linguistic stimuli, not perceptual ones. This suggests that the linguistic stimuli initiated sensory-motor simulations, which behaved similarly to sensory-motor processing. Experiment 2 An alternative explanation remains to be addressed. Perhaps properties across all modalities are stored together in a single system of amodal knowledge. Within this system, amodal symbols that represent properties from the same modality are associated to each other, such that they prime each other when processed sequentially. If so, then these associations could underlie the switching costs in Experiment 1. When a subject verifies two properties from the same modality, associations between their amodal symbols speed processing, relative to properties from different modalities whose symbols are not associated. As already noted, the critical property pairs in Experiment 1 were not associated in the Nelson et al. (1999) norms. Perhaps, however, these norms are not sufficiently sensitive to detect weak associations that link properties from the same modality. This hypothesis can be tested by using highly associated property pairs from the Nelson et al. norms. If nonmeasurable associations speed same-modality pairs whose normed strengths are 0, then even greater priming should occur as associative strength increases. Thus Experiment 2 sampled pairs of properties from the Nelson et al. norms that are highly associated (e.g., spotless-clean; polyester-cheap). These associated properties were then combined with concepts to form pairs of verification trials (e.g., “SHEET can be SPOTLESS”—“AIR can be CLEAN”; “SHIRT can be POLYESTER”—“MEAL can be CHEAP”). If the associative hypothesis is correct, then substantial priming should be found for the second members of these pairs, relative to when the context and target items have unassociated properties (e.g., “SHEET can be SPOTLESS”—“MEAL can be CHEAP”). Switching costs in property verification 9 In contrast, we did not predict an associativeness effect. Many previous studies have found that priming diminishes substantially—and typically disappears—when an unrelated word separates two associated words (Bentin & Feldman, 1990; Dannenbring & Briand, 1982; Joordens & Besner, 1992; Masson, 1995; McNamara, 1992). Given that three words stood between properties on adjacent trials in Experiment 1, it seems unlikely that the first property could have primed the second associatively. For example, in “SHEET can be SPOTLESS” followed by “AIR can be CLEAN”, the three words “AIR can be” lie between “SPOTLESS” and “CLEAN”. We expected that these intervening words would extinguish any possible priming. In addition to the strongly associated properties, we also presented pairs of unassociated properties from the same vs. different modalities (a replication of Experiment 1). First, we wanted to replicate the modality shifting effect in another experiment. Second, we wanted to directly compare this effect with any associative priming effect. Because highly associated properties are necessary for testing the associativeness effect, whereas unassociated properties are necessary for testing the modality-switching effect, different property pairs were used to test the two effects. Method Subjects and design. Eighty-eight volunteers from Emory University participated for course credit. Associated versus unassociated pairs were manipulated within subjects, as were same vs. different modalities, with counterbalanced versions of the list being distributed equally across subjects. Materials. Thirty pairs of associated properties were selected from the Nelson et al. (1999) norms that averaged 23.1% in associative frequency (i.e., how often the second property was produced as an association of the first). This is a very high level of associative strength, with approximately 95% of the words in the norms having a lower first associate (Nelson, personal communication). The first property in a pair was always the cue in the norms, and the second property was always a response. The introduction to this experiment provides examples. Switching costs in property verification 10 Two critical lists were formed from the 30 pairs of associated properties. In one list, 15 of the associated pairs remained intact, and the other 15 were scrambled to form unassociated pairs (as illustrated in the introduction). In the other list, the first 15 pairs were scrambled to form unassociated pairs, whereas the second 15 pairs remained intact to form associated pairs. All critical items were true. An additional 30 pairs of trials were selected from the materials of Experiment 1. Two lists were created so that each list contained 15 pairs from the same modality and 15 from different ones. For the same pairs, the associative strength between properties was 0. Across lists, each concept-property combination occurred in both conditions. An additional set of 120 filler pairs was constructed. Within these fillers, 60 contained 2 false items, 30 contained a true item then a false item, and 30 contained a false item then a true item. Fifteen of the false target items were associatively related to the previous trial, and 15 were in the same modality as the previous trial. Thus, the relation between two consecutive trials was not predictive of the correct response for targets. Subjects could not give a "true" response on the basis of the context item being associated or in the same modality as the target. The remaining 90 filler pairs were unrelated. The practice materials consisted of 48 additional trials that were comparable to the experimental materials. No concept or property was repeated across the materials. Procedure. The 0 ms SOA procedure from Experiment 1 was used here. Results and Discussion As in the previous experiment, target trials were removed either when subjects erred in response, or when they erred on the previous context trial. Median RTs in the relevant conditions were computed for each subject and then averaged. ----------------------------------------------Insert Table 3 about here ----------------------------------------------As Table 3 illustrates, an associative priming effect did not occur in the RTs, F(1,87)=0.016, or in the errors, F(1,87)=0.22. Unassociated properties were not reliably slower than associated properties, indicating that associations between properties did not underlie the switching effect in Experiment 1. Indeed, these two condition differed by only 1 Switching costs in property verification 11 ms. In contrast, the RTs on the different modality trials were 41 ms slower than those on the same modality trials, F(1,87)=9.40, p<.01. The difference between associative priming and modality switching was nearly significant in the two-way interaction, F(1,87)=3.76, p=.056. The error data did not show any significant effect. As these results show, associative strength does not explain the switching costs in Experiment 1. If associations between properties from the same modality had been responsible, an associative effect should have occurred in Experiment 2, given that the property pairs were much more associated than those in Experiment 1. Instead, no associative priming effect was obtained, whereas there was again a reliable effect of modality switching. This leaves modality-specific processing as the best account of the switching costs. After verifying a property, attention rests on its modality. If the subsequent property resides on a different modality, attention must shift, thereby incurring a cost. General Discussion According to perceptual symbols theory (Barsalou, 1999), simulations in sensorymotor areas represent properties during conceptual processing. If the conceptual system rests on sensory-motor systems, then phenomena in perceptual processing should also occur in conceptual processing, at least to some extent. Thus, the presence of switching costs in perceptual processing suggests that analogous switching costs should occur during property verification. The results of Experiments 1 and 2 support this prediction. When subjects verified pairs of properties, they verified the second property faster when it came from the same modality as the first property than when it came from a different modality. Experiment 2 ruled out the alternative hypothesis that associations between properties from the same modality were responsible. When subjects verified a pair of associated properties, no priming occurred, even though the associations between them were considerably stronger than any associations that might have existed between properties from the same modality in Experiments 1 and 2. Thus, switching from one modality-specific brain system to another appears to be the critical factor in these experiments—not associative strength. Recent neuroimaging work on the localization of concepts corroborates this conclusion (e.g., Martin, Switching costs in property verification 12 2001; Martin & Chao, 2001; Martin et al., 2000), as does the literature on lesion-based conceptual deficits (e.g., McRae & Cree, in press; Simmons & Barsalou, 2002). Recent fMRI work in our lab shows that verifying the six types of properties assessed in Experiments 1 and 2 activates the respective modality-specific neural systems (Pecher, Hamann, Simmons, Zeelenberg, & Barsalou, 2002). One issue is the generality of the modality-shifting effect. Within the experiments reported here—and subsequent ones like them—all six modalities generally exhibit trends in the predicted direction (due to the noisiness of the data and the small number of properties on each modality, individual trends are rarely significant). Across four replications of the basic paradigm—an unpublished initial experiment, Experiment 1 (SOA=0), Experiment 1 (SOA=260), and Experiment 2—the mean differences between the same-modality RTs and the different modality RTs were: 37, 28, -7, 65 for vision; 42, -2, -38, 43 for audition; 48, 48, 20, 39 for motor; 86, 59, 104, -18 for smell; 10, 32, 100, 10 for taste; -10, 42, -20, -34 for touch. In a given experiment, not every modality shows a trend, but across experiments, each modality shows one at least once. A related issue is whether modality shifting effects occur for properties that do not come from the six modalities we address, such as the properties of abstract concepts. Barsalou (1999) suggests that abstract concepts draw heavily on introspective experience, such as emotional states and cognitive operations. Wiemer-Hastings, Krug, and Xu (2001) provide evidence for this hypothesis. To the extent that other sorts of properties arise on different modalities of experience, shifting effects should occur between them as well. For example, if emotion and cognitive operations constitute different domains of introspection, they might exhibit shifting effects. This issue awaits further research. Together with other recent evidence, the findings here converge on the conclusion that the conceptual system is grounded in sensory-motor simulation. It is becoming increasingly difficult to argue that the conceptual system is completely modular and amodal. To the contrary, the conceptual system appears to share many mechanisms with perception and action, thereby making it non-modular and modal. Switching costs in property verification13 ReferencesBarsalou, L. W. (1999). Perceptual symbol systems. Behavioral & Brain Sciences, 22, 577-660.Barsalou, L. W. (in press). Situated simulation in the human conceptual system. Languageand Cognitive Processes.Bentin, S., & Feldman, L. B. (1990). The contribution of morphological and semanticrelatedness to repetition priming at short and long lags: Evidence from Hebrew.Quarterly Journal of Experimental Psychology, 42A, 693-711.Dannenbring, G. L. & Briand, K. (1982). Semantic priming and the word repetition effect in alexical decision task. Canadian Journal of Psychology, 36, 435-444.Fodor, J. A. (1975). The language of thought. Cambridge, MA: Harvard University Press.Glenberg, A. M. (1997). What memory is for. Behavioral and Brain Sciences, 20, 1-55.Joordens, S., & Besner, D. (1997). Priming effects that span an intervening unrelated word:Implications for models of memory representation and retrieval. Journal ofExperimental Psychology: Learning, Memory and Cognition, 18, 483-491.Klatzky, R. L., Pellegrino, J. W., McCloskey, B. P., & Doherty, S. (1989). Can you squeeze atomato? The role of motor representations in semantic sensibility judgements. Journalof Memory and Language, 28, 56-77.Martin, A. (2001). Functional neuroimaging of semantic memory. In. R. Cabeza & A.Kingstone (Eds.), Handbook of functional neuroimaging of cognition (pp. 153-186).Cambridge, MA: MIT Press. Martin, A., & Chao, L. (2001). Semantic memory and the brain: structure and process.Current Opinion in Neurobiology, 11, 194-201.Martin, A., Ungerleider, L. G., & Haxby, J. V. (2000). Category-specificity and the brain: Thesensory-motor model of semantic representations of objects. In M. S. Gazzaniga (Ed.),The new cognitive neurosciences (2nd ed., 1023-1036). Cambridge, MA: MIT Press.Masson, M. E. J. (1995). A distributed memory model of semantic priming. Journal ofExperimental Psychology: Learning, Memory, and Cognition, 21, 3-23. Switching costs in property verification14 McNamara, T. P. (1992). Theories of priming: I. Associative distance and lag. Journal ofExperimental Psychology: Learning, Memory, and Cognition, 18, 1173-1190.McRae, K., & Cree, G. S. (in press). Factors underlying category-specific deficits. In E.M.E.Forde & G.W. Humphreys (Eds.), Category specificity in mind and brain. London:Psychology Press.Nelson, D. L. (personal communication). January, 23, 2002.Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (1999). The University of South Floridaword association, rhyme and word fragment norms.http://web.usf.edu/FreeAssociation/ .Pecher, D., Hamann, S. B, Simmons, W. K., Zeelenberg, R., & Barsalou, L. W. (2002). fMRIevidence for modality-specific conceptual processing in six modalities. Manuscript inpreparation.Pulvermüller, F. (1999). Words in the brain's language. Behavioral and Brain Sciences, 22,253-336.Pylyshyn, Z. W. (1984). Computation and cognition. Cambridge, MA: MIT Press.Simmons, K., & Barsalou, L. W. (2002). The similarity-in-topography principle: Reconcilingtheories of conceptual deficits. Invited article to appear in a special issue of CognitiveNeuropsychology.Solomon, K. O., & Barsalou, L. W. (2001). Representing properties locally. CognitivePsychology, 43, 129-169.Solomon, K. O., & Barsalou, L. W. (2002). Perceptual simulation in property verification.Under review. Spence, C., & Driver, J. (1998). Auditory and audiovisual inhibition of return. Perception &Psychophysics, 60, 125-139.Spence, C., Nicholls, M. E. R., & Driver, J. (2000). The cost of expecting events in the wrongsensory modality. Perception & Psychophysics, 63, 330-336.Spivey, M., Tyler, M., Richardson, D., & Young, E. (2000). Eye movements duringcomprehension of spoken scene descriptions. Proceedings of the 22nd AnnualConference of the Cognitive Science Society, (pp.487-492). Mahwah, NJ: Erlbaum. Switching costs in property verification15 Stanfield, R. A., & Zwaan, R. A. (2001). The effect of implied orientation derived fromverbal context on picture recognition. Psychological Science, 12, 153-156.Wiemer-Hastings, K., Krug, J., & Xu, X. (2001). Imagery, context availability, contextualconstraint, and abstractness. Proceedings of the 23rd Annual Conference of theCognitive Science Society, 1134-1139. Mahwah, NJ: Erlbaum.Wu, L., & Barsalou, L. W. (2002). Perceptual simulation in property verification. Underreview.Zwaan, R. A., Stanfield, R. A., & Yaxleu, R. H. (in press). Do language comprehenders routinelyrepresent the shapes of objects? Psychological Science. Switching costs in property verification16 Author NoteThis research was performed while the first author was a research scholar at EmoryUniversity on a TALENT stipend from the Netherlands Organization for Scientific Research(NWO). The second author was supported by a grant from the Foundation for Behavioral andEducational Sciences of the Netherlands Organization for Scientific Research. The thirdauthor was supported by National Science Foundation Grant SBR-9905024, which alsosupported the experiments. We are grateful to Aron Barbey, Sergio Chaigneau, and BrianCornwell for valuable comments made throughout the research. We are also grateful toShurin Hase, Holly Drazin, and Meghan Mitchell for help running the experiments, and toDouglas Nelson for consultation on his association norms. Address correspondence to DianePecher, Erasmus University, Psychology Department, Burgemeester Oudlaan 50, 3062 PARotterdam, The Netherlands, email: [email protected]. Switching costs in property verification17 Footnotes1 Notationally, we will use uppercase italics to represent concepts, lowercase italics torepresent properties, and quotes to represent linguistic forms (words, sentences).2 Although most of the target properties were found in the Nelson et al. norms, not all were.Those properties found did not have their context properties as associations. Thoseproperties not found were comparable, appearing unassociated. Experiment 2 addressedassociativeness directly and found that it was not a factor in these experiments.3 One might argue that the words “can be” actually do not count as full content words, andthus may not decrease priming. Nevertheless there is still one completely unrelatedcontent word between the two properties, namely, the concept for the second one. Thework just cited shows that even one intervening word can dissipate priming between tworelated words.4 An earlier experiment also showed no effect for a large manipulation of associative strengthin a similar design. Thus the absence of an associative effect appears robust.5 Because the critical the items were rotated through a counter-balanced design, itemanalyses are technically not necessary (Raaijmakers, Schrijnemakers, & Gremmen, 1999).For the record, though, in three cases of the four cases where an items test was possible, aneffect was present (unpublished pilot experiment, t(44)=2.28; Experiment 1 (SOA=0),t(49)=2.44, Experiment 1 (SOA=260), t(49)=0.87; Experiment 2, t(29)=3.92). The criticaleffect is generally present across items. Switching costs in property verification18 Table 1.Examples of the Target and Context Trials from the Six Modalities in Experiment 1. Context trialModality Target trialSame modalityDifferent modalityAudition BLENDER-loudLEAVES-rustlingCRANBERRIES-tartVision BABY CLOTHES-pastel HAIR-fairTOAST-warmTaste CUCUMBER-bland BUTTERMILK-sour BIRD EGG-speckledSmell SOAP-perfumedOLD BOOK-mustyTELEVISION-noisyTouch MARBLE-coolPEANUT BUTTER-sticky BED SPRINGS-squeakingMotor FAUCET-turnedROCK-hurledHIGHWAY SIGN-green Note. The context trial immediately preceded the target trial. The concept and property were presentedin a sentence frame stating the possibility that the “CONCEPT can be PROPERTY.” Switching costs in property verification19 Table 2.Mean Reaction Times and Error Rates in Experiment 1 for Verifying Properties on TargetTrials Following a Context Trial Either on the Same Modality or on a Different Modality(Standard Errors Shown in Parenthesis). 0 ms SOA260 ms SOAContext trialRTs (SE) Errors (SE) RTs (SE) Errors (SE) Same modality 1124 (27.8) 5.1 (0.83) 859 (23.3) 5.0 (0.71)Different modality 1153 (28.9) 5.6 (1.38) 879 (24.7) 4.0 (0.76) Switching cost290.520 -1.0 Switching costs in property verification20 Table 3.Mean Reaction Times and Error Rates in Experiment 2 (Standard Errors Shown inParenthesis).___________________________________________________Context trialRTs (SE) Errors (SE)___________________________________________________ Associatively related 1143 (15.4) 9.9 (0.82)Associatively unrelated 1144 (14.6) 10.2 (0.72)