Comparative judgments of animal intelligence and pleasantness

Transcription

1 Memory & Cognition 1980, Vol. 8 (1), Comarative judgments of animal intelligence and leasantness ALLAN PAIVIO and MARC MARSCHARK University of Western Ontario, London, Ontario N6A 5C2, Canada Symbolic comarisons of animal intelligence and leasantness ere investigated in five exeriments using ords and ictures as stimuli. Symbolic distance effects occurred in all exeriments. Picture-ord effects deended on the exerimental design. Mean intelligence comarison reaction time (RT) did not differ for ictures and ords hen the airs ere resented to indeendent grous. With reeated measures, mean RT as faster for ictures than for ords, but only on the second block of trials. This involved an asymmetrical transfer effect, so that ord comarisons on the first trial block greatly facilitated icture comarisons on the second block, hereas icture comarisons did not facilitate subsequent ord comarisons. Pleasantness comarisons shoed a similar asymmetrical transfer attern but differed from intelligence comarisons in that the mean RTs ere faster ith ictures than ith ords, even ith an infinite item set and ithout rior exosure to the other class of material. The results are discussed in terms of dual coding and other models of symbolic comarisons. This study sought to resolve a theoretical controversy concerning the symbolic rocesses involved in comarative judgments of abstract attributes of concrete things. Attention focused on animal intelligence, hich is abstract in the sense that values on the dimension are not correlated ith secific ercetual attributes such as size, shae, or color. Pleasantness as introduced into the study so that the generality of the initial findings could be tested ith another abstract attribute. A central issue here is hether the major asects of task erformance can be interreted in terms of a dual coding theory of cognition, as oosed to theories that assume a different data base for symbolic comarisons. Various effects involving comarisons on concrete dimensions, such as size, have been redicted from dual coding theory (aivio, 1975, 1978a). The basic assumtion of the theory is that memory and thought involve the activity of indeendent but interconnected verbal and nonverbal symbolic systems, hich are differentially accessible to verbal and nonverbal stimuli and have different structural and functional characteristics. The verbal system is more directly accessed and activated by linguistic stimuli, such as rinted ords, hereas the nonverbal system is activated more directly by nonverbal stimuli, such as ictures. The reresentational units and structures of the nonverbal system resumably "contain" information that is analogous to ercetual information in the sense that there is a continuous, analog relation beteen the long-term memory information activated by symbolic stimuli, such as names, and the ercetual information aroused directly by the ercetual objects This research as suorted by Grant ADD87 from the National Research Council of Canada. Requests for rerints should be sent to Allan Paivio, Deartment of Psychology, University of Western Ontario, London, Ontario N6A 5C2 Canada. themselves. The ercetual nature of the symbolic information is characteristically reflected in the conscious exerience of imagery, so it is aroriate to refer to. the system as a hole as an imagery or image-generating system. At the same time, it is exlicitly assumed that the imagery system can be functionally active ithout necessarily silling over into reortable imagery. Dual coding theory rovides the folloing interretation of erformance on symbolic comarison tasks. Verbal stimuli such as animal names (e.g., Moyer, 1973) first activate reresentations in the verbal system. These in turn arouse symbolic reresentations in the image system, hich are comared ith resect to the relevant attributes, such as size. The theory accounts for the symbolic distance effect (faster reaction times, RTs, as differences on the dimension increase) in much the same ay as other analog aroaches to mental reresentation (e.g., Moyer, 1973; Moyer & Bayer, 1976). More ertinent in the resent context, it redicts faster RTs for ictures than for ords even hen the ictures do not directly reresent ercetual differences on the relevant dimension. Thus, memory size comarisons are faster for object airs resented as ictures (equated on deicted size but not on real-life size) than for airs resented as rinted ords. This finding is consistent ith the assumtion that the ercetual memory reresentations ("imagens") that contain the concrete information are activated more directly by ictures than by ords. Also in accord ith the theory are the findings that subjects redominantly reort using imagery in the erformance of such comarisons and that RT correlates ith individual differences in imagery ability, as inferred from satial maniulation tests, but not verbal ability (e.g., Paivio, 1978a). The dual coding model has been challenged by findings from studies involving comarisons on abstract Coyright 1980 Psychonomic Society, Inc X/80/0 I $01.25/0

2 40 PAIVIO AND MARSCHARK dimensions (e.g., Banks & Flora, 1977; Holyoak & Walker, 1976; Kerst & Hoard, 1977). The exeriment by Banks and Flora (1977) is most relevant here. They resented subjects ith airs of names and ictures of animals and asked them to choose the member of each air that as "smarter" (or "dumber"). They found that comarison time as faster for ictures than for ords and concluded that this contradicted the dual coding assumtion that faster comarisons ith ictures means that the rocessing as mediated by reresentations in the imagery system. Because the attribute is abstract, the comarison ould have to be based on verbal reresentations, ith the result that comarisons should have been faster ith ords than ith ictures or, at least, ictures and ords should not have differed in their effect. Since ictures turned out to be suerior, Banks and Flora concluded that the finding is more consistent ith their semantic coding hyothesis than ith dual coding, given the additional assumtion that this abstract (essentially linguistic) coding system is accessed more quickly through ictures than through ords. An alternative roosal (aivio, 1978c) is that, rather than being inconsistent ith dual coding theory, such findings call into question the sychologicalinterretation of attribute abstraction. The general argument is that attributes like intelligence are rimarily characteristics of things, or they are based on learned reactions to things. To the extent this is true, subjects in a comarison task ould have to access the memory reresentations of the referent objects before they could make the required judgment. The faster comarison time obtained by Banks and Flora (1977) for ictures than for ords is consistent ith the interretation. So, too, are analogous results obtained by Paivio (1978c) for comarisons of leasantness and monetary value. Subjects in the latter study also reorted redominant use of imagery in erforming the task, and individuals scoring high on imagery ability had faster comarison times than those ith lo ability in the case of both leasantness and value dimensions, and ith either ictures or ords as items. Verbal fluency shoed no significant relation to comarison time, although it emerged as a significant factor in interaction ith imagery in the case of leasantness comarisons. All of the findings aeared to be consistent ith the dual coding interretation: Although abstract, leasantness and value are nonetheless acquired characteristics of things, and RT ill be facilitated by any factor that makes the nonverbal memory reresentations more accessible. The imlications of the above analysis in regard to individual differences in cognitive abilities have not been directly investigated in the inte!ligence comarison task. If the theoretical interretation is correct, faster comarison times should be obtained ith ictures than ith ords (relicating Banks and Flora, 1977), and those ho score high on imagery ability tests should have faster comarison times than those ho score lo. By the same logic, verbal abilities should correlate ith comarison time to the extent that erformance is deendent on verbal rocesses. We tested the above redictions in Exeriment 1 by relicating the essentials of the Banks and Flora (1977) exeriment using a modified design. Banks and Flora used only eight animals, ranked for intelligence, and divided these into to sets consisting of the four "smartest" and four "dumbest" animals. Within each set, airs ere constructed by airing each item ith.every other one and resenting the entire set a number of times. In addition, each subject received blocks of both ictures and ord airs, in counterbalanced order. We used a much larger number of items arranged into indeendent airs, and different grous of subjects received ictures and ords. Folloing the exerimental task, subjects comleted a battery of individual difference tests. To our surrise, e failed initially to relicate Banks and Flora's icture-ord effect, and the correlational data involving individual difference variables differed from those obtained ith comarisons on other abstract attributes. Exeriments 2-5 (the last to involving leasantness comarisons) sought to resolve these discreancies, as ell as to test a articular theoretical interretation of the unexected findings. EXPERIMENT I Method Item ool and normative rating rocedure. A ool of 78 animal names as comiled. All animals in the list ere easily ortrayed in simle line draings and ere familiar to college students. The animals ere randomly divided into three lists of 26 each, and these lists ere then reroduced for use in a normative rating task. Each of 45 introductory sychology students, articiating to satisfy a course requirement, received a test booklet containing a age of instructions, folloed by the three lists in random order, and ending ith a questionnaire concerning ossible strategies, including imagery. that might have been used in the rating tasks. Intelligence as rated on a 9-oint scale, on hich a rating of I denoted loest intelligence and 9 denoted highest intelligence. Subjects erformed the rating task in grous of 4 to 12 and ere alloed as much time as necessary to comlete the task. Uon comletion, subjects filled out the strategy questionnaire. Four secific strategies ere included: (1) "Comaring the animals listed ith some standard animal or animals not necessarily resented in the list," (2) general verbal strategies (e.g., ronouncing the names of the animals), (3) general use of visual imagery (e.g., simly forming images of the animals), and (4) forming images "of the animals erforming some articular behavior suggestive of intelligence." The frequency of use of each of these strategies as rated on a 5-oint scale ranging from "used infrequently or never" to "used ith all or almost all of the animals listed." In addition, an oen-ended question asked subjects to describe any other strategy they might have used and to elaborate on their referred strategy. Analysis of the questionnaires revealed a strong and significant reference for the imaginal strategies (all s <.005, excet the "general imagery" vs, "comarison ith standard" case, =.07). Ansers to the oen-ended question suggested that the second imaginal strategy as the most helful, as subjects reorted imaging the animals in a variety of situations that reflected intelligence. Most subjects also indicated that "com-

3 INTELLIGENCE AND PLEASANTNESS JUDGMENTS 41 arison ith standard" strategy involved the use of visual imagery.i Stimulus materials. Mean intelligence ratings ere calculated for each animal; these ranged from 1.49 (orm) to 8.16 (monkey). To sets of 45 different airs each ere then constructed such that, for 15 airs in each set, there as an arithmetic difference of aroximately 1 beteen the mean intelligence ratings of the to members; likeise, 15 airs had arithmetic differences of aroximately 3, and 15 airs had arithmetic differences of aroximately 5. Reetition of individual items as avoided as much as ossible, although several aeared more than once in order to obtain equal sets at the three levels of arithmetic difference. The 90 airs ere then randomly comiled into four different resentation lists. List 1 contained the 45 airs of one set, folloed by the remaining 45 airs. List 2 contained the to sets in the oosite order. Lists 3 and 4 contained the same order of 90 airs as in Lists 1 and 2, resectively, but the left-right order of each air as reversed. To versions of each list ere then constructed, one containing the names of the animals and the other containing line draings. All airs ere laced on 10 x 15 em cards. Word airs ere tyed in IBM Letter Gothic caitals ith their midoints aroximately 7 em aart; the ictures ere all dran in black ink, subtending aroximately the same area (8-9 em2) and ere also laced 7 ern aart. All lists ere balanced so that the more intelligent animal as on the left in half of the airs and on the right in the other half. Individual difference tests. Imagery ability as measured by three satial maniulation (figural transformation) tests. These ere sace relations (Form A) of the Differential Atitude Test battery (Bennett, Seashore, & Wesman, 1947), a shortened version (Linkert & Quasha, 1941) of the Minnesota Paer Form Board (MPFB), and block visualization (Guilford, 1967). These three tests ere selected because they are objective and have been shon to load on a common figural transformational ability factor (Guilford, 1967). They ere also moderately intercorrelated in a cumulative samle of subjects that has articiated in a series of mental comarison exeriments (see Paivio, 1978a, 1978c). Verbal ability as measured by three tests. One as an associative fluency test that requires subjects to rite as many different associations as they can in a l-min eriod to each of four concrete and four abstract nouns (see Paivio, 1971, ). A second measure of verbal ability 'vas a ord beginnings and endings test (adated from the first- and last-letters tests of Thurstone). This test requires subjects to rite don, in a 3-min eriod, as many ords as they can that begin ith one letter and end ith another. The final verbal measure as an inference test (Guilford, 1967) in hich subjects are given one or to statements (remises) folloed by various conclusions that could be dran from them. The task is to decide hich of the conclusions can be dran from the statements ithout assuming anything in addition to the information given. All subjects also received the Full-Range Picture Vocabulary Test (Ammons & Ammons, 1948). This as initially included because it is urorted to be a "quick" test of intelligence. Hoever, the test also aeared to be secifically relevant to the resent uroses because it requires transfer from ictorial to verbal information. Design and Procedure. Forty-eight subjects ere dran from the same oulation as those in the normative task; 24 received icture stimuli, and 24 received ord stimuli. Six eole ere assigned to each of the four lists ithin a condition. Subjects ere told that airs of animals (ictures or ords) ould be resented tachistoscoically, and they ere to decide hich as the more intelligent member of each air. Folloing a ready signal, a fixation oint aeared, remained for 1 sec, and as immediately folloed by a stimulus air that remained visible for 4 sec. The subjects resonded by ushing a button ith either the right or left thumb, corresonding to hether the more intelligent of the to animals as on the right or the left. RTs and resonses ere recorded by a PDP-12 comuter in an adjoining room. Beteen resonses, subjects ket their thumbs resting on the buttons in order to facilitate raid as ell as accurate resonding. Each subject received 10 ractice trials folloed by to blocks of 45 trials each. A 2- to 3-min rest eriod beteen test blocks as otional. This ortion of the exeriment lasted aroximately 20 to 25 min, after hich subjects ere given the icture vocabulary test. All subjects returned to the laboratory 1 to 4 days later for grou administration of the remainder of the individual differences test battery. Results and Discussion Reaction time and error rates. Errors accounted for 4% and 3% of the resonses in the icture and ord conditions, resectively. Mean RTs for correct decisions ere comuted for each subject for each of the three levels of arithmetic difference. Both mean RT and number of errors ere then analyzed by a 2 (ictures vs. ords) by 2 (lists) by 3 (symbolic distance) analysis of variance in hich difference as a ithin-subjects factor. The only significant effect, in both analyses, as that of distance [Fs(2,80) ~ , s <.001]. As can be seen in Figure 1, mean RT decreased as the. difference in animal intelligence increased. This function as comletely consistent ith revious findings, but the absence of any icture-ord difference as not. Individual differences. Subject's scores on the individual differences battery ere examined through correlational analyses. Table 1 shos, for each test, the Pearson r ith comarison mean RT for the total grou, as ell as for ictures and ord grous searately. It can be seen that the only significant redictor of RT for the comlete samle as the icture vocabulary test ( <.05, r = -.29). The searate analyses shoed further that icture vocabulary as the only significant 1800 u ~ ~ 1400 r ~ 1200 r ':i a: 800~ 2 3 DISTANCE Figure 1. Intelligence comarison mean RT in Exeriment 1 for icture airs and ord airs as a function of air difference in rated intelligence.

4 42 PAIVIO AND MARSCHARK Table 1 Correlations Beteen Individual Difference Test Scores and Mean Reaction Time Picture Word Comlete Test Grou Grou Samle Sace Relations MPFB Block Visualization Associative Fluency Word Beginnings and Endings Inference Test Picture Vocabulary -.57* t " <.01. t <.05. redictor for the subjects ho received icture stimuli ( <.005, r = -.57), but it did not redict comarison mean RT for the ord grou (r =-.03). In fact, there ere no significant redictors of RT for subjects receiving ord stimuli. One trivial ossibility is that ord mean RT is simly less reliable than icture mean RT. This as checked by randomly searating the individual air RTs from each of the distances into to sets and comuting to mean RT scores for each subject. The slit-half reliability coefficients, corrected by the Searman-Bron formula, ere.85 for ictures and.95 for ords. Thus reliability is actually higher for ord mean RT, and e can reject the reliability interretation. The overall attern of correlations differs from those obtained in revious studies, in hich the satial tests redicted erformance in comarison tasks involving abstract dimensions (aivio, 1978c) as ell as concrete ones (e.g., Paivio, 1978a). The most that can be said here is that one imagery test, sace relations, shoed a correlational trend in the aroriate (negative) direction ith intelligence comarison time. The general imortance of the results is that they are not easily exlained by current theories of mental comarisons. In regard to dual coding, they do not oint to the redominance of either imagery or verbal rocesses in comarisons of intelligence, although comlex imagery strategies ere aarently involved in the normative intelligence rating task. The absence of a icture-ord effect also fails to relicate Banks and Flora (1977) and is, therefore, theoretically inconsistent ith their ost hoc addition to the semantic coding model, hich states that the common, semantic coding system is more quickly accessed through ictorial than through verbal stimuli. The results of the icture vocabulary test suggest to ossibilities. One is that this articular comarison task deends heavily on some general intellectual ability, hich is measured by the vocabulary test. The other is that comarison time is influenced by the more secific icture naming ability measured by the vocabulary test. The results are more consistent ith the latter than ith the former interretation. If general intelligence ere the more imortant factor, the icture vocabulary test should have correlated ith comarison mean RT involving ords, as ell as ictures, as items. Moreover, one ould have exected the inference test to correlate ith erformance on the comarison task, since that test involves an abstract intellectual skill, namely, syllogistic reasoning. Since the inference test as not redictive and since the icture vocabulary test correlated only ith comarison time hen ictures served as items, it is more reasonable to conclude that the imortant factor as related to the subject's ability to translate ictorial information into a verbal code. In terms of dual coding, this ould be interreted to mean that task erformance deended on the interconnections beteen imaginal and verbal systems, rather than on either class of reresentations alone. Exeriment 2 tested an imlication of this analysis, hich also rovided a ossible basis for resolving the emirical discreancy beteen the resent results and those obtained by Banks and Flora (1977) in regard to icture-ord effects. EXPERIMENT 2 Our Exeriment I differed from Banks and Flora's (1977) Exeriment 2 in to rincial ays. First, e resented ictures and ords to different grous of subjects, hereas Banks and Flora used a ithin-grous design in hich articiants received blocks of trials ith one tye of material, then a series of trials ith the other tye. Second, e used an infinite set involving a relatively large number of airs, hereas Banks and Flora used a small, finite air set. Either factor may have been resonsible for the different results. Exeriment 2 accordingly relicated Banks and Flora in terms of both items and general design. We anticiated that this should roduce the overall icture-ord difference reorted by Banks and Flora, but it also ermitted searate analyses of the first and second blocks oftrials (not reorted by Banks and Flora). The analysis of the first trial block ould involve indeendent grous receiving ictures and ords. If ictures turned out to be suerior to ords on that block, it ould suggest that the difference beteen our results and those of Banks and Flora as due to the articular list of items or the finite set design used by the latter. The image-verbal code translation hyothesis suggested the alternative ossibility that ictures ould be suerior to ords only (or mainly) on the second trial block, after articiants had been exosed to ords on the first block. Prior exosure to the ords ould make the name code highly available hen the same airs are subsequently resented as ictures. The subjects receiving ictures on the first trial block ould not have the advantage of augmented verbal code availability. The exeriment required only a small number of subjects and involved a comlex reeated-measures design ith resect to the icture-ord variable. Accord-

5 INTELLIGENCE AND PLEASANTNESS JUDGMENTS 43 ingly, it as inaroriate for the systematic study of individual differences, and that asect is not ursued further in relation to Exeriment 2 or the subsequent ones.' Method Design and Procedure. The stimulus items ere the eight animal concets used by Banks and Flora (1977, Exeriment 2). The rank ordering of these animals on intelligence as the same in both Banks and Flora's and our normative rating tasks. In order of decreasing intelligence, the animals ere ae, dog, cat, horse, co, shee, chicken, and fish. The eight animals ere divided into the four most intelligent and the four least intelligent. Stimulus airs for the judgment task ere created by combining animals only ithin each category; thus there ere six different airs in each intelligence level. To different, randomly ordered lists of the 12 airs ere constructed, balancing the number of times a more intelligent animal aeared on the right or the left. To additional lists contained the same ordering but ith the left-right order of each air reversed. The lists ere then reroduced in both icture and ord forms as in Exeriment 1, creating a total of 96 stimulus airs (12 airs by 2 lists by 2 left-right orders by 2 forms). Prior to testing, half of the subjects ere shon a list of the ictures and names of the animals, rank ordered on intelligence (cf. Banks & Flora, 1977, Exeriment 2). The remaining subjects received no rior exosure to the animals (no-reviecondition). This maniulation alloed us to determine hether Banks and Flora's results ere deendent uon their giving subjects revies of the ordered stimuli. Half of the subjects in each of these grous received a block of 48 icture trials, folloed by 48 ord trials; the other subjects received the to blocks in the reverse order. The testing rocedure as identical to that of Exeriment I, excet that the subjects received five ractice trials ith icture airs before the block of icture trials and five ractice trials ith ord airs before the ord block. The exeriment lasted min. Subjects. Tenty subjects ere dran from the same oulation as those in Exeriment I and ere randomly assigned to the various conditions on the basis of their order of aearance at the laboratory. Results and Discussion Errors accounted for 8% and 11% of the resonses to icture and ord stimuli, resectively, in the norevie condition, and 6% to each stimulus tye in the revie condition. The error data ere analyzed by 2 (revie or no revie) by 2 (material-icture or ord) by 2 (order-icture-ord or ord-icture) by 3 (distance) analyses of variance in hich material and distance ere ithin-subjects factors. This analysis yielded a significant main effect only for distance [F(2,32) = 34.13, P<.001], so that errors decreased as the rank-order difference increased. The only significant interaction as Previe by Material [F(I,16) =4.91, P<.05]. Mean RTs for correct decisions ere comuted for each subject for icture and ord trials at each level of symbolic distance, that is, at the three rank-order differences (1, 2, and 3) beteen air members. Mean RTs ere then analyzed by a 2 (revie or no revie) by 2 (icture or ord) by 2 (icture-ord vs. ordicture order) by 4 (trials ithin each tye of material) by 3 (distance) analysis of variance, in hich subjects ere nested in revie condition and order, and reetitions ere nested in material. Significant main effects ere obtained for order [F(l,16) = 5.46, P <.05], ith mean RTs being faster for the ord-icture grou; material [F(l,16) =5.68, <.05], ith mean RTs to ictures faster than those to ords; trials [F(6,96) = 13.06, P <.0001]; and distance [F(2,32) = 40.95, P <.0001]. The only significant interactions ere Material by Order [F(l,16) =6.75, <.05] and Order by Trial [F(6,96) =2.44, <.05]. The former interaction is shon in Figure 2, in hich it can be seen that mean RT is loer for ictures than for ords only in the case of the grou that first comared ords and then ictures. The icture-ord grou shoed essentially no effect. Looked at another ay, mean RT as slightly faster for ords than for ictures on the first trial block and much faster for ictures than for ords on the second block. Note, too, that icture mean RTs ere much faster on the second than on the first block, but ord mean RTs shoed a trend in the reverse direction. The simlest factual interretation of the interaction is that ictures facilitated comarisons given rior exo. sure to their names but not otherise. The attern can also be interreted as an asymmetrical transfer effect: Positive transfer occurred from ords to ictures but not from ictures to ords. The Order by Trial interaction resulted from a steeer decline in RT over trials for ord-icture than for icture-ord subjects. The effect is attributable entirely to differences in the second block, in hich subjects receiving ictures shoed a rogressive decrease in mean RTs and those receiving ords rogressively increased over the first three trials. The difference is a detailed reflection of the asymmetrical transfer effect already,,-... U ~ ~ 1000 ~ 900 I- U 800 «~ 700 Pe ae W 1 BLOCK Figure 2. Interaction effect of tye of material (ictures vs. ords) and trial block on intelligence comarison mean RT in Exeriment 2. 2

6 44 PAIVIO AND MARSCHARK noted. It could not emerge as a trile interaction because ofthe nested design involved in the exeriment. EXPERIMENT 3 Exeriment 3 as a close relication ofexeriment 2, designed to ensure that the results of the latter ould generalize to an indeendent set ofitems. Method Eight animals ere selected from the item ool described in Exeriment 1. These ere different from, but had mean intelligence ratings as close as ossible to, those of the eight animals used by Banks and Flora (1977, Exeriment 2). In order of decreasing intelligence, the animals ere monkey, oroise, leoard, raccoon, donkey, rooster, hiootamus, and turtle. The exerimental design and rocedure ere identical to those of Exeriment 2 excet that, rior to testing, all subjects ere given a revie of the animals' ictures and names, rank ordered by intelligence. Ten subjects ere dran from the same oulation of Exeriments 1 and 2 and ere randomly assigned to the ictureord and ord-icture conditions. Results and Discussion Errors accounted for.8% and.3% of the resonses on icture and ord trials, resectively. Analysis of the error data yielded a significant main effect only for distance [F(2,16) = 4.04, P <.05], resulting from a decrease in errors as rank-order difference increased, and a significant interaction of Material by Order [F(I,8) = 9.68, <.025]. The ord-icture and icture-ord grous had error rates of.06 and.14, resectively, on Block 1, hereas the to grous did not differ on Block 2 (.03 and.01, resectively). As in Exeriment 2, mean RTs for correct decisions ere comuted for each subject and analyzed by a 2 (order) by 2 (material) by 4 (trial) by 3 (distance) analysis of variance in hich subjects ere nested in order and trials ere nested in material. Significant main effects ere obtained for material [F(I,8) = 24.10, <.005], trials [F(6,48) = 5.23, <.OOI], and distance [F(2,16) = 21.63, <.0001]. Significant interactions ere obtained for Material by Order [F(I,8) = 9.04, <.025] and Distance by Trial [F(12,96) = 2.69, <.005]. Figure 3 deicts the Material by Order interaction. The attern is similar to that obtained in Exeriment 2, insofar as there as no significant ictureord difference for Block 1 trials (beteen subjects), by a Neman-Keuls test, hereas icture judgments (by ord-icture subjects) ere significantly faster than ord judgments (by icture-ord subjects) in Block 2 ( <.05). Vieed in terms of transfer effects, ositive transfer again occurred from ord to icture trials but not for the reverse order. The Distance by Trial interaction resulted from the symbolic distance effect's being attenuated for the first trial in each block (Trials 1 and 5). The distance effect as strongest [i.e., steeest) for the last trial of each block (Trials 4 and 8), and all other trials yielded tyical distance effects falling beteen these to extremes. -U ~ 11l E ~ 1000 I- z 900 o I U 800 <{ W 700 a: 1 BLOCK Figure 3. Material by Order interaction effect on intelligence comarison mean RT in Exeriment 3. The icture-ord results of Exeriments 1-3 seem to be inconsistent ith revious findings obtained by Paivio (I 978c) for the abstract dimensions of leasantness and value, both of hich involved faster comarisons ith ictures than ith ords as stimuli. Moreover, unlike the intelligence comarison results in Exeriment 1, both leasantness and value comarisons ere significantly faster for subjects ho scored high on tests of imagery ability than for those ho scored Ioan such tests. Since the leasantness comarisons in the Paivio (I 978c) study involved an infinite stimulus set, hereas the value comarisons involved a finite set, the differences observed for intelligence comarisons cannot be readily attributed to that rocedural feature. Hoever, the leasantness exeriment involved only inanimate objects. Perhas the results differed because of the difference in stimulus class. Exeriments 4 and 5 exlored that ossibility using leasantness comarisons ith animal concets. Pleasantness as chosen because it seems as reasonable to comare animals on leasantness as on inteiligence. Attention centered on the effects of ictures as comared ith ords and the order in hich subjects received the to tyes of material. Exeriment 4 involved an infinite-set aradigm, so that the attern of results for the first trial block could be comared ith the leasantness results ofpaivio (1978c) and the intelligence results of Exeriment 1 of the resent study. Exeriment 5 involved a finite set thereby ermitting comarisons ith Exeriments 2. and 3. EXPERIMENT 4 Method Stimulus material. Animal concets differing in rated leasantness ere dran from the ool described by Paivio (l978b Exeriment 1). To sets of 18 different airs ere constructed so that, for 6 airs in each set, there as an arithmetic difference 2

7 INTELLIGENCE AND PLEASANTNESS JUDGMENTS 45 of aroximately.75 beteen the leasantness ratings of the to members; likeise, 6 airs in each set had arithmetic differences of aroximately 1.50, and 6 had arithmetic differences of aroximately Reetition of articular items as avoided as much as ossible, but several aeared more than once in order to obtain equal numbers of airs at the three levels of arithmetic difference. For the same reason, there ere four airs that aeared in both sets. To lists of 36 airs ere then constructed. Each as comosed of the 18 randomly ordered airs of a set, folloed by a reetition of the same airs in a different order. The reetitions of any articular air ere searated by at least eight other items. Both lists ere then reroduced in both icture and ord forms. Subjects. A total of 34 subjects ere dran from the same oulation as those in the other exeriments and randomly assigned to the various conditions on the basis of their order of aearance at the laboratory. In fact, the total samle as made u of to indeendent grous, tested 5 months aart by to different exerimenters. The ossibility of differences beteen the to grous is evaluated belo. Design and Procedure. The testing rocedure as identical to that of Exeriment 3 excet that subjects resonded to the more leasant, rather than the more intelligent, animal of each air. Seventeen subjects received each list. Nine of those assigned to one list received 36 airs in ord form folloed by the same 36 airs in icture form. Eight other subjects received to sets in the reverse order. Assignment to the second list as counterbalanced so that, overall, 17 subjects received ords folloed by ictures and 17 received ictures folloed by ords. Each subject thus had 72 test trials as ellas fivematerialaroriate ractice trials rior to each of the to blocks. Prior to resentation of the icture stimuli, subjects ere cautioned to choose the more leasant of the animals reresented by the ictures, not the more leasant line draing. The exeriment lasted aroximately 15 min. Resultsand Discussion Errors accounted for 9% of the resonses each to icture and ord stimuli. Mean RTs for correct decisions ere comuted for each subject at each level ofsymbolic (arithmetic) difference, searately for ictures and ords, in each of the four set reetitions. The mean RT and error data ere then analyzed by 2 (exerimenter) by 2 (list) by 2 (order-icture-ord or ord-icture) by 2 (material-icture or ord) by 2 (trials ithin each tye of material) analyses of variance in hich trials ere nested ithin material and subjects ere nested ithin exerimenter, list, and order. The effect of exerimenter as negligible, as neither analysis yielded a significant main effect nor any significant interactions that might qualify the effects of interest. This factor is therefore excluded from further discussion. The error analysis yielded significant main effects of distance [F(2,52) = 43.13, P <.0001] and list [F(I,26) = 4.44, P <.05]. These ere qualified by a significant List by Distance interaction [F(2,52) = 4.77, P <.025], hich resulted from the fact that one yielded a steeer symbolic distance effect than the other. The RT analysis yielded significant main effects for trials [F(2,52) = 22.11, P <.001] and distance [F(2,52) =18.49, <.001]. A marginal effect of stimulus material as also obtained [F(l,26) =3.46, <.075], but this as qualified by the significant Material by Order interaction [F(l,26) = 7.04, <.025], shon in Figure 4. It can be seen that mean RTs ere faster for o 1500 ~ W l/l E <;» 1400 z ~ z i= --. u " <{ 1200 a:: 1:,, 1 2 BLOCK Figure 4. Material by Order interaction effect on leasantness comarison mean RT in Exeriment 4. icture stimuli than for ords on the first block (i.e., beteen subjects), hereas there as a slight advantage for ords on the second block. There ere no other significant effects. These findings relicate those of Paivio (1978b), insofar as comarisons ere faster for ictures than for ords hen there as no rior exosure to the other class of stimulus material. The effect contrasts ith the absence of any icture-ord difference for intelligence comarisons in the comarable infinite-set conditions of Exeriment 1. Hoever, the transfer effects ere similar to those observed for intelligence comarisons in Exeriments 2 and 3, in the sense that ord comarisons facilitated subsequent icture comarisons of the same concets but not vice versa. The main difference is that the sueriority of ictures over ords occurred on the first block of trials in the case of leasantness and on the second block in the case of intelligence comarisons. Exeriment 5 exlored the generality of the leasantness results using a finite-set design. EXPERIMENT 5 Method Eight animal concets, differing in rated leasantness, ere dran from the ool described in Exeriment 4. The items ere chosen so as to be clearly rank ordered on the dimension, thus eliminating the need for revieing of the ordered list (as in Exeriments 2 and 3). In order of decreasing leasantness, the animals ere san, squirrel, arrot, co, frog, fly, skunk, and rattlesnake. The exerimental design as identical to that of Exeriments 2 and 3, excet that all subjects received the same random order of the stimulus items. Ten subjects received a block of 48 icture trials folloed by a block of 48 ord trials, and 10 received the to blocks in reverse order. The rocedure as also identical to that of Exeriments 2 and 3, excet that there as no revie of the items and that judgments ere made on leasantness rather than on intelligence. The 20 subjects ere all dran from the same oulation as in the earlier exeriments and ere randomly assigned to the P

8 46 PAIVIO AND MARSCHARK to conditions on the basis of their order of aearance at the laboratory. Results and Discussion Errors accounted for I% of the resonses each to icture and ord stimuli. Mean RTs for correct decisions ere comuted for each subject at each level of symbolic (rank-order) difference. The error and mean RT data ere than analyzed by 2 (icture-ord or ordicture order) by 2 (icture or ord material) by 3 (distance) by 4 (trials ithin each stimulus fonn) analyses of variance, in hich stimulus material and distance ere ithin-subjects factors. The error analysis indicated that there ere significantly feer errors to ord stimuli than to icture stimuli [F(l,18)=6.72, <.05] and a robust symbolic distance effect, as errors decreased ith increasing distance [F(2,36) = 90.17, P <.0001]. The only significant interaction as that of Order by Distance by Trials [F(l2,216) = 1.92, <.05], hich is of no theoretical interest here. The RT analysis yielded three significant main effects: stimulus materials [F(l,18) = 15.30, <.005], ith resonses being faster for ictures than ords; distance [F(2,18)=26.39, <.OOOI}, indicating a symbolic distance effect; and trials [F(6,108) = 7.38, P <.0001]. Significant interactions of Material by Distance [F(2,36) = 4.40, P <.025] and Order by Distance by Trials [F(I2,216) = 1.87, <.05] ere also obtained. The to-ay interaction resulted from a reduced icture-ord difference at Distance 2. The marginal trile interaction does not qualify any results of interest and ill not be considered further. The critical Material by Order interaction as not significant [F(I,18) = 2.62]. Nonetheless, as can be seen in Figure 5, the data folloed the same general attern ~ 1300 E... ~ 1200 ~ ~ 1100 ~ u «~ BLOCK Figure 5. Material by Order interaction effect on leasantness comarison mean RT in Exeriment 5. 2 as intelligence comarisons in Exeriment 3 (Figure 3) in regard to the rimary locus of the icture-ord difference and the asymmetrical transfer effects. The mean RT advantage for ictures as greater in the second than in the first block of trials, and transfer as strongly ositive from ords to ictures and negative from ictures to ords. The effects are unlike those obtained for leasantness comarisons in Exeriment 4, in that, here, comarisons ere faster for ictures than for ords in the second as ell as in the first block. Thus intelligence and leasantness comarisons share some features in common hile differing in others. GENERAL DISCUSSION The discussion deals first ith the imlications of the results for dual coding and semantic models and then ith the more general imlications of the asymmetrical icture-ord transfer effects observed in Exeriments 2-5. Neither the semantic coding nor the dual coding model correctly redicted the results of Exeriment I. To accommodate their finding of faster comarative judgments ofintelligence ith ictures than ith ords, Banks and Flora (1977) added the assumtion that the abstract semantic coding system, hich resumably mediates the judgments, is more quickly accessed by ictures than ords. The model ould no have to be remodified to account for the observation that ictures are advantageous in this task hen receded by comarisons in hich the items are resented as ords but not otherise. It is difficult to see ho this could be achieved ithout a drastic revision of the basic assumtions of the model, unless one simly adds further ost hoc assumtions that amount to restatements of various emirical findings themselves. In fact, the asymmetrical transfer effect is incomatible ith any theory that assumes a common reresentational base for ictures and ords. Dual coding theory also failed initially in the sense that revious alications of the theory to comarisons involving other abstract dimensions (Paivio, 1978c) did not generalize directly to animal intelligence. This could mean either that dual coding theory is rong or that imaginal and verbal rocesses interact in intelligence judgments in a ay that differs somehat from their contribution to leasantness and value judgments. The latter interretation is favored by the results of Exeriments 2-5. All exeriments shoed an interaction of material and trial blocks such that subjects ho began ith ords significantly reduced their reaction times hen sitched to ictures, hereas those ho began ith ictures shoed no significant overall change on the subsequent ord trials, and the trend as sometimes negative. These findings suggest that both leasantness and intelligence comarisons involve an interaction of imaginal and verbal rocesses. In addition, hoever, the image system seemed to dominate more in the case of leasantness than in the case of intelligence. The latter conclusion follos from the faster leasantness

9 INTELLIGENCE AND PLEASANTNESS JUDGMENTS 47 (but not intelligence) mean RTs ith ictures than ith ords even in inflnite-set design, here there as no rior exosure to icture labels. It is also consistent ith the finding that subjects ith high imagery ability, as measured by the satial maniulation tests, ere significantly faster than their lo-imagery counterarts in the case of leasantness (aivio, 1978c), but imagery ability did not correlate significantly ith intelligence comarisons in Exeriment 1. Nonetheless, nonverbal symbolic rocesses seem to be crucial in intelligence comarisons as ell. This conclusion follos from several observations: First, comarison RT became significantly faster hen subjects sitched from ords to ictures, but not vice versa, ith the net result that comarisons ere sometimes faster ith ictures than ith ords, but never faster ith ords than ith ictures. Second, the icture vocabulary test redicted mean RT in Exeriment 1 only hen the comarison task involved ictures as items. Neither observation is consistent ith the alternative ossibility that linguistic rocesses redominated in the task. Finally, the correlation of -.32 beteen sace relations test scores and intelligence comarisons mean RT for the icture grou in Exeriment I, although nonsignificant, as at least suggestive ( =.06, onetailed test) and aroriate in direction. The data are insufficient to ermit a more detailed dual coding interretation of the intelligence comarison results ith high confidence, but several alternatives can be suggested. One ossibility is in terms of the additive contribution of verbal and imaginal codes, ith the latter carrying more eight in the combined effect (cf. Paivio & Csao, 1973). High availability of the verbal code resulting from rior exosure to ord airs or from strong icture naming skill ensures that icture airs ill be aroriately labeled during the comarison task. The congruence of both codes ith resect to the concets being comared results in reliable and seedy resonding because both linguistic and orld knoledge converge on the roblem. To use an extreme examle, one "knos" that "chims" are smart and "orms" are stuid because one has often heard (or read) statements to that effect and because one has observed smart behavior in the former and stuid behavior in the latter. The nonverbal system still carries more eight in the combination because, according to the theory, the referents of the verbal and ictorial signs are reresented nonverbally, in the ercetual memory (image) system. Comarisons of the attributes of the referent objects are imossible ithout access to their symbolic reresentations. The role of the verbal system is to ensure reliable activation of articular reresentations as ell as to rovide access to associated memory information relevant to the comarison task-verbal memories concerning intelligence, nonverbal memories concerning intelligent behaviors (aroused through the referential interconnections beteen systems), and so on. Subjects in the normative hase of Exeriment I in fact reorted a reference for imaginal strategies hen judging intelligence, esecially imagining the animals in a variety of situations that reflected intelligence. Whether these involved static images of situations or some kind of active or dynamic imagery (cf. Paivio, 1971, Chater 2) cannot be clearly inferred from the reorts. In any case, the suggested interretation ould be generally consistent ith earlier dual coding interretations of symbolic comarisons (aivio, 1978c), as ell as a variety of other tasks, ranging from free recall (Paivio & Csao, 1973) to metahor comrehension (Paivio, 1979). It is also consistent in rincile ith the analysis of symbolic comarisons roosed by Kosslyn, Murhy, Bemesderfer, and Feinstein (1977). An alternative ossibility is that the comarisons are essentially based on imagistic reresentations and that the only contribution of the verbal system is to ensure the arousal of aroriate images. Thus, the icture vocabulary test may be redictive because it tas the subject's ability to identify the things reresented ictorially, not because of naming er se. Similarly, rior exerimental exosure to ord airs gives the subject ractice at generating and comaring imagistic referents of the ords. The rimed imagery activity is easily re-evoked subsequently by ictures. Initial icture trials, hoever, rovide no ractice ith image generation to ords, hence there is no ositive transfer to subsequent ord trials. We cannot choose at this time beteen the different dual coding interretations and other ossible ones that could be suggested. The imortant general oint is that the findings indicate that intelligence comarisons are highly deendent on the availability of both verbal and nonverbal information, in an interactive manner. Interactions are evident also in leasantness comarisons, but the dominance of nonverbal rocesses is more clearly suorted by to classes of convergent oerations (icture-ord differences and individual differences in imagery abilities). Other tasks, such as relative judgments of name frequency aear to be based rimarily on the verbal system, as ould be exected from dual coding theory (aivio, 1978b). Thus the intelligence comarison task is somehat unique in that it seems to be more deendent on the interaction of systems rather than on either one alone. Finally, e comment briefly on the general imlications of the icture-ord transfer effect observed in Exeriments 2-5. Such effects indicate that situational (eisodic) memory as ell as semantic memory contribute to task erformance. The eisodic memory contribution could include any or all of three classes of information related to the first trial block, including (I) memory for the airs as resented (ictures or ords), (2) memory for the reresentational information aroused from long-term memory by the stimuli (e.g.,

10 48 PAIVIO AND MARSCHARK names aroused by ictures, images aroused by ords), and (3) memory for the comarison rocesses or other rocedures involved in the task. The first of these is exlicitly eisodic. The second and third involve eisodic information generated from semantic memory during task erformance. The extent to hich each of these contributed to erformance on the second block of trials cannot be ascertained from the resent study, but it ould be ossible to do so by systematically maniulating relevant features (materials, encoding instructions, etc.) of the initial task rior to transferring to a comarison task involving ictures or ords as stimuli. The recise nature of the transfer effects is also indeterminate in the resent exeriments because they did not include the necessary controls for simle reetition effects (i.e., to blocks of trials ith ords alone or ictures alone). The results ermit us only to infer asymmetrical transfer effects that are relatively ositive or negative-that is, the ord-icture transfer effect as ositive relative.to the icture-ord condition. Conversely, the icture-ord transfer effect as negative relative to the ord-icture condition. This limitation does not affect the tentative conclusions dran from the serendiitous results of the resent exeriments, but more recise conclusions ill require a comlete transfer aradigm. Systematic exeriments of that kind should advance our understanding of the comlex rocess involved in symbolic comarisons generally. REFERENCES AMMONS, R. B., & AMMONS, H. S. Full-range icture vocabulary test. Missoula, Mont: Psychological Test Secialists, BANKS, W. P., & FLORA, J. Semantic and ercetual rocesses in symbolic comarisons. Journal ofexerimental Psychology: Human Percetion and Performance, 1977,3, BENNETI, G. K., SEASHORE, M. G., & WESMAN, A. G. Differential atitude tests. Ne York: Psychological Cororation, GUILFORD, J. P. The nature of human intelligence. Ne York: McGra-Hill, HOLYOAK, K. J., & WALKER, J. H. Subjective magnitude information in semantic orderings. Journal of Verbal Learning and Verbal Behavior, 1976,15, Kerst, S. M., & Hoard, J. H., Jr. Mental comarisons for ordered information on abstract and concrete dimensions. Memory & Cognition, 1977,5, KOSSLYN, S. M., MURPHY, G. L., BEMESDERFER, M. E., & FEINSTEIN, K. J. Category and continuum in mental comarisons. Journal ofexerimental Psychology: General, 1977, 106, LIKERT, R., & QUASHA, W. Revised Minnesota aerform board test, Series AA. Ne York: The Psychological Cororation, MOYER, R. S. Comaring objects in memory: Evidence suggesting an internal sychohysics. Percetion & Psychohysics, 1973, 13, MOYER, R. S., & BAYER, R. H. Mental comarisons and the symbolic distance effect. Cognitive Psychology, 1976, 8, PAIVIO, A. Imagery and verbal rocesses. Ne York: Holt, Rinehart, & Winston, PAIVIO, A. Percetual comarisons through the mind's eye. Memory & Cognition, 1975,3, PAIVIO, A. Comarisons of mental clocks. Journal of Exerimental Psychology: Human Percetion and Performance, 1978, 4, (a) PAIVIO, A. Dual coding: Theoretical issues and emirical evidence. In J. M. Scandura & C. J. Brainerd (Eds.), Structural/rocess models ofcomlex human behavior. Leiden: Nordhoff, (b) PAIVIO, A. Mental comarisons involving abstract attributes. Memory & Cognition, 1978,6, (c) PAIVIO, A. Psychological rocesses in the comrehension of metahor. In A. Ortony (Ed.}, Metahor and thought. Ne York: Cambridge University Press, PAIVIO,A., & CSAPO, K. Picture sueriority in free recall: Imagery or dual coding? Cognitive Psychology, 1973,5, NOTES 1. On the basis of the questionnaire data, an ugraded form as constructed. The ne version differed from earlier formats in having all strategies stated in general terms and rated on both verbal and imaginal use scales. Previously, each verbal and imaginal strategy has been secifically indicated as a searate item. The data from all three exeriments ere generally consistent ith those of the normative task insofar as subjects referred strategies involving comarisons of the animals in a variety of intelligence-indicative situations. Unfortunately, the data ere not easily interretable, as subjects seemed unable to distinguish their verbal and imaginal use of the general strategies. Discussion of the questionnaire has therefore been omitted from the three exeriments in hich it as used. 2. The subjects in Exeriments 2 and 3 did in fact comlete an individual test battery folloing the exeriment. The resulting correlations beteen test scores and comarison mean RTs for the different exerimental subgrous (ith Ns ranging from five to eight), and their ooled combinations over conditions and exeriments ere all nonsignificant. (Received for ublication May 14,1979; revision acceted October 8, 1979.)