BRE CSc FACILITATION IN RECOGNIZING PAIRS OF WORDS. EVIDENCE OF A DEPENDENCE BETWEEN RETRIEVAL OPERATIONS’ Ball Telephone Laboratories, Murray HU, New Jersey DAVID B. MEYER? xD ROGER W. SCHVANEVELDT Unversity of Colorado ‘Two experiments are reprted in which Ss were prosented two stings of letters simuitanecusly, with one string displayed visually above the other, In Exp. 1S fesponded "yes" if bth strings were words, otherwise respanding "no." In Exp, 1, 5s responded "samo" if tho Two strings Were either both words or both noawords, otherwise responding "differen." "Yor" responses and "samo" espouses were faster for pais of commanly associated words than for pairs of Uuaassociated words, "Same" responses were slowest for pairs of nonwords. "No? responses were faster when the top string in the disp ‘whereas "different responses were faster when the top siting was word, TI results of both experiments support e retrieval model involving a dependence betteen separate successive decisions about whether exch ofthe two rings is ‘word, Possible mechanisms that underlie dependence are discussed, Several investigators recently have studied how Ss decide that a string of letter is a word (Landauer & Freedman, 1968; Meyer & Ellis, 1970; Rubenstein, Garfield, & Millikan, 1970). ‘They typically have presonted a single string on a trial, measuring reaction time (RT) of the dexical decision as a function of the string’s meaning, familiarity, otc. In onc such experiment, RT varied inversely with word frequency (Rubenstein et al, 1970). When word frequency was gontrolled, lexical decisions were faster for homographs (ic., words having two or more meanings) than for nonhomographs. ‘To explain these results, Rubenstein ct al. proposed that word frequency affects the order of examining stored words in long-term memory and that more replicas of homographs than of nonhomographs are stored in long-term memory. Tn another experiment, Meyer and Ellis (1970) measured both the time taken to “This paper ies roport from work begun is ‘dependently by the two authors at Bell Telepht oratories andthe State University of New York at Stony Brook, respenively. We thank 8. Stemberg, 7 Lande, nd Alene Pook hl bu ‘comateals, A. 8. Coriell for preparing the apparsts, ‘nd G- Elis and B, Kunz forsuaning ‘aeequoste for reprints should be Fen to Dovid E. ‘Meyer, Boll Telephone Laboratories, 600 Monatain Avenue, Mareay Hil, Nevr Jersey 09974, 2 devide that a string of letters (e.g, HOUSE) is a word and the time taken fo decide that it belongs to a prespecified semantic category. When the category was rela- tively small (¢.., BUILDINGS), the latter type of semantic decision was significantly faster than the former loxical decision. However, when the catezor was relatively large (c.g., STRUCTURES), the semantic decision was slightly slower than the lexical decision. To explain these and othor ‘results, Meyer and Ellis suggested that the semantic decision may have involved searching through stored words in the semantic catcgory and that the lexical decision did not entail a search of this kind among the set of all words in memory. ‘The present paner provides further data about the effect of meaning on lexical de- cisions. To deal with this problem, we have extended the lexical-decision task by, simultaneously presenting two strings of letters for S to judge. The stimulus may involve either a pair of words, a pair of nonwords, or a word and a nonword. In ‘one task, Sis instructed to respond "yes" if both strings aro words, and otherwise to respond "no." In a second task, the instructions require S to respond "same" if the two strings are either both words and both nonwords, and otherwise to respond. "different." In each task, RT for pairs of208 DAVIDE. MEYER AND ROGER W. SCHVANEVELDT words is measured as a fimetion of the associative relation between the two words. The two tasks together are designed to give information about the nature and the invariance of underlying retrieval opera- tions. One of their advantages is that the relation between words can be varied while keeping the overt response constant, We reasoned that the response might involve separate, successive decisions about each of the two words. By varying the degree of association between the words, wo then hoped it would be possible to test for a dependence between memory-accessing components of the two decisions. Experiment I reports the results of such variation in the context of the ‘yeseno task. EXPERIMENT I Method Subjects, — The Ss were 12 high school students ‘who served as paid volnteers. Stimul. The following test stimuli were used: 48 pais of associated words, 0, BREAD SUTTER and'Nunse-poctor,selocted from the Connecticut Froe Association Norms (Bousfoll, Coben, Whitmarsh, 1961); 48 pair of unassocitod words, 2g, HRUAD-DOCTOR and NURSE-BUTTER, formed by fndomly iaterchanging the response’ tes, ber freon the 48 pair of ntcisied words ao that tere ‘wore 20 obvious sasociatone within the resling es: 48 pairs of nomwords, and 96 pairs involving E'vord and e nonvword, Within each pair of ass0- lated words, the Second member was ether tho fiat orsecond most frequent fro associate given in response to the frst member. Within each pair of Unasvocited words, the second member Was never to fir or sepond most frequent free associate of tho frst member. The median length of stings in the palms of ‘sssociated words nd pais of tnassocited words was 3 letters and ranged fom 3 107 levers; the median word frequency was 59 por nin, and ranged from 1,747 to less than 1 pot zillion (Kucera & Francs, 1967).A seperate sf of 536 words tas used for tio pairs involving s word tnd a nonword, These words wore simile tothe ‘srocated words in toms of fequency, length, and Semantic. classification, Nonvwords "sere ‘constricted. from ‘common words, eg. MARK, feplacing at lest ome leter with another lotr Vowels wero. used. to. replace vowels, and ‘consonants, The rosultng stings of levers, eg. MARS, were Fronouneeable and were equal i average leogth 10 {ho words paired with them. A majoriy of the ‘onwords differed by only single letter from some Enalsh word, and the diferences were not syste matically associated with any ne fetter positon. Tn edtin tothe test stimuli 24 pars of words, 8 raits of nenwords, and 16 pers involving & word and a nonword were consinicted as practice Stimuli. Degreo of association was not” varied Systematically among th pairs of practice words. Apparatus. — Tho stinnli were generated on @ Stromberg Cavoas” SCH060 “graphics ssc, photographeden 16-mm.movie film and presented fa. rearprojestion soreen by a Perceptual Dovelopmeat Laboratories” Mark ILL Pereeptoscope. The Ss responded via a panel having finger koys for the right and left hands. Reaction ime was measured to. the nearest smilisocond by counting the eyoles of a 1,000-H2. ‘oscillator. Procedure and design. — The $5 were run in- dividually during one session involving a series of ‘iscroto RT tls. The S was seated in front ofthe darkened sereen throughout the session. At the beginning of each trial, the word READY was pre- sented briefly au a waming signal on the sereet, A. smuall fixation box, which subtended epproximate ‘ial angles of 3°40" horizontally and (°50" vert cally, then appeared daring a 1-sec, foreperiod. Following the foreperiod, the stimulus. was di layed horizontally ia (white) capital letters ia the middle of the box, with one siting of lettors entered above the other. If both strings were words, § pressed a key labeled "yes" with his Hight index finger, otherwise pressing 2 “no" key with the left index finger. Reaction time was measured from simolas-onnct 0. tho rexponso, which termingted tho stimulus display. During en ‘pproximato 2-sec. intorval when the screen was blank after each tal, $ was informed of whether Bis reeponto had been corect. "The session lasted about 45 min. and included a short instruction period and two Blocks of 24 practice trials, followed by four blocks of 24 rest trials. After cack block, S was informed of his sean RT and total mimbor of errors forthe block, ‘hile he rosiad for sbout2 min. This feedback was Intended to encourage fast md zccurato responses. ‘Tofinther motivate good performance, § was given $3 atthe stat af the seston end then penalized 1g for each | sec. in mean RT on each trial block, and 3¢ foreach error. Whatever money remained atthe ‘session served as $s payment for the uring the two practice trial blocks. During the test tial Blocks, each was shown 16 paits of nonwords, 32 pairs involving 2 word and 2 nonword, 24 pairs of associated words, and 24 pairs of unassoeiuted words from the total set of {ost stimuli, Half of the practice trials and test trials therefore required "yes" responses. Proseatation of the tet stimuli was balanced, so that each individual stimulus of « given typo was prosented equally often across So; e.g, each pair ‘Of associetod words was presented a fotal of six times across.$5, while each pair of nomvords was presented a total of four times. No S saw any String of letters more than once. In displaying both the pairs of atsociated words and the paits of runsssosintod words, the top string (e.g, BREAD) ‘was always a stimulus term from the norms of Bousfield etal. (1961), while the bottom string:FACILITATION IN WORD RECOGNITION 29 TABLE 1 MEAN RRACTION Toss (RTS) 0¥ CORRECT RESPONSES AND MEAN PERCENT ERRORS DTH YES-NOTASK “Type of stimuas paie stresponee | Proportion | Mean RT | Mean Comet response | Srerals | see 08s Topsting Bottom sing ond ‘assoiated word se 855 63 fend ‘Maaasociated weed ie ot 3 cond snonvond no ios | 276 laoivcord word no Sat 18 fnoncord Monod to Bat 26 (e.g. BUTTER) was always a response tom. For stimuli containing a least onenonword, cach sting ‘was assigned equally offen across Ss to the top and bottom display positions. There were thus five types of stimuli, which are listed in Table 1 togelher with. their relive frequencies "of occurrence, Relative frequencies of these types ‘are balanced within tial blosks to equal their relative frequencies in the fotel set of test simul ‘The above set of constraints on presentation was used to construct sbi {eststimuli each, » Subject to these eenstrain's wo random osders of simuls. presentation were ‘obtained for each list, Bach S was then readonly assigned One of the lists presented in one of the ‘orders, so that each list in each order was used for exactly one 8. Results and Discussion Reaction time and error data from the test trials were subjected to Ss x Treatments analyses of variance (Winer, 1962). Prior to analysis, an arc-sine transformation was applied to each S's error rates. ‘The reported standard errors and F ratios were computed using error terms derived from the Ss * Treatments interactions. ‘Tables 1 summarizes mean RTs of correct responses and mean percent errors averaged over Ss. "Yes" responses averaged 85 + 19 msec, faster for pairs of associated words than for pairs unassociated words, F (1, 11) = 20.6, p < 001, Not respons 1 pais involving & word and a nonword averaged 18. + 14 mseo. faster when the nonword was displayed above the word, F (, 11) = 171.7, p <.001. "No" responses for pairs of “nionwords' were not significantly faster 20: 14 msec.) than "no" responses {or alts whereanonword, was displayed above a word, (1, 11)=2.0, p> The error rates for pairs of unassociated words versus pairs of associated words did not differ significantly, F (1, 11) =2.1, p>.10. The error rate for pairs involving a word and a nonword was significantly greater when the word was displayed above the nonword, F'(1, 11) = 18.9, p < 005. The error rate for pairs of nonwords ‘was significantly less than that for pairs where a nonword was displayed above a word, (11) = 5.5, p <.05. Error rates were relatively low except for pairs where a word was displayed above a nonword. A possible reason for this exception is, considered “in later discussion. The pattern of errors s that a speed-acturacy.tnde-oft dif not cause the observed differences in mean RTs; ic., mean error rates tended to correlate positively with mean RTs lhe, ggauls of Exp. I suggest that degree of association it power factor ‘ing lexical decisions in the yes-no task. For example, the effect of association appears to be on the order of two or three times larger than the average effect of homography reported by Rubenstein et al. (1970). This offect of association occurred consistently across ‘Ss, and 11 of the 12.Ss showed it in excess ©f of'30 msce. In Exp. If, nother group of Ss performed the samé-different task to further study the generality of the effect. EXPERIMENT I Method ‘Subjects. "The Ss were 12 high school students ‘who served as paid voluntcers. They had not been "Exp. 1, bat were drawn from the same popalation.230 TABI DAVID E. MEYER AND ROGER W. SCHVANEVELDT LE 2 MEAN REACTION TIMES (RT'S) OF CORRECT RuSPONSHS AND MAN PERCENT ERRORS IN THE, SAME-DIFFERENT TASK ‘Tepe of sinwlos pair i Proportion | MeanRT | Means Comectrcepcase | “fils | mse.) | eons ‘Top sting Bottom string ‘word asociated word same Bs 1035 2 word unaszociated word | same 125 i 37 nonword | nonword same 25 1357 89 word onward Uiffereat 1318 us aonword word different 1386 2p Stimull. — The same sot of tost stimuli was used as in Exp. I In addition, 16 pairs of words, 16 pairs of noawords, and 32 pairs involving & word and & nonword were constructed practice stimull, Most of these practice stimuli hao had been used in Exp. [. Apparatus. — The seme ep in Exp. 1 Procedure and design. — The proveduse and sign were similar 10 those used in Exp. 1, except for the following modifications. The S rested a "same" key with his right index finger Ifthe stimulus involved either twe words or two noawords, otherwise pressing, a “different” koy ‘with the Ioft index finger. The complete session lasted about | ir, and included a short instruction period, two blocks of 32 practice trials, and six blocks of 32 tost trials. [two lists of 192 tost stimuli each wero constructod. For each list, to random orders of presentation were obtained, fubject fo constraints like those used in Exp. L Baok of theso List * Order combinations was ‘of the Ss. During the test tral yas presented 48 pairs of nonwords, 96 pairs involving a word and a onward, 24 pairs of associated words, and 24 atus was used 2s Ps {ost stimuli, Half of the trials therefore required “same” responses. Because the seme-different sk was somowhat more difficult then the yos-no task, each S was given $3.50 at tho start of the session. Results The results were analyzed in the same way as Exp. I. Table 2 summarizes moan RTs of correct responses and mean percent errors averaged over Ss. "Same" responses averaged 117+ 18 msce, faster for pairs of associated words than for pairs of unassociated words, F (1, 11) = 42.6, p < 001. "Same" responses avoraged 18529 nsec. slower for pairs of nonwords than for pairs of unassociated words, F (1, 11) = 40.7, p < 001. "Different" responses averaged 68 4 25 msec. faster when the word was displayed above the nonword, F (1, 11)=73,p <025. The error rate for pairs of associated words was significantly less than the error rate for unassociated words, F (1, 11) = 16.6, p-<.01. The difference between error rates for pairs of unassociated words and airs of nonwords was not significant, (1, 11) < 10. For pairs involving a word and & nonword, the error rate did not depend significantly on whether the word was displayed above or below the non-word, F G,11)<1.0. ‘A comparison of mean RTs in the yes-no task (Exp. 1) versus mean RTs in the same-different task reyealed the following: “Yes” responses to pairs of words averaged 216 & 68 msec. fester than "same" responses to pairs of words, F (1, 22) = 10.2, p <.01. The effect of association on "same" responses to pairs of words did not diffor significantly from its effect on " responses, F (1, 22) = 1.4, p > 20. "No" responses to pairs involving a word and a nonword averaged 357 + 74 msec. faster than "different" responses, F (1, 22)=23.6, P<.001. For pairs involving a'word and a nonword, the effect of the word's display position ‘on RT interacted significantly with the task, F (1,22)= 76.4, p < 001. DIscUssION As a framework for explaining our results, ‘we tentatively propose a model involving t%0 separate, successive decisions. According to this model, stimulus processing, typically be- ‘gins with the top siring of letters in the display. ‘The first decision is whether the top string is & word and the second is whether theFACILITATION IN WORD RECOGNITION 231 botiom string is @ word. | Ifthe fst desision ismegative in the yes-no task, we presume that processing terminates without the. second decision and S responds "no." Otherwise, both decisions are made end Ss response corresponds to the second decision's outcome. Tt is assumed that in the samecdifierent task, both decisions are normally made regardless of the outcome of the first, Afterboth devisions, their outcomes, ‘are compared. Tf the outcomes match, feaponds, sume"; otherwise, he responds "different ‘Now let us consider the RTs and error rates of yes-no,_ responses. The serial-decision model is Why “no” responses fare fastor when the top string is a nonword. ‘This happens because only the frst decision is ‘made, wheroas both deoisions are made when ‘the top siting is a word. The model also explains why "no" responses are about equally fast for pairs where only the top siting is a nonword, as compared to pairs where both strings are nonwords; ie, for either kind of pair, only the first decision is ordinarily made.’ An occasional reversal order of stimulus processing, beginning with the bottom rather than top siting, might eccount for the slightly faster responses io pairs of nonwords. ‘The relatively high error rate for pairs involving a word above « nonword suggests processes, ling "yes" responses sometimes terminate prematurely alr the first decision. In theso oases, 8 may feel that discovering ¢ word in the top. posgn ig sufficient evidence for responding "yes," Without making the second decision. “This behavior would not be too reasonable, given, tho relative frequenoies of tho various fypes of stimuli. Such premature termination of stimulus processing, together with an occasional reversal in, the processing order, ‘would also explain why "no" uses ‘were most accurate for. pairs ofronwords. The RTs fiom the same-different task do not provide direst evidence for testing the proposed serial-decision model because both lexical decisions are assumed to be made before all same-different responses. However, the relative invariance ofthe association effect across yos-no and same-different. tasks suggests that similar processes occur in both tasks. An additional operation, — which compares the outcomes of the to lexical decisions for a match, would explain. why responses were somewhat slower in the same-differont task than in the yes-no task. Several factors in the present experiments ‘may have induced Ss to process the strings of letters serially. For example, $5. were en- couraged to perform with high accuracy and ‘were allowed to move their eyes in examining the stimulus display. Under other ciroum- stances, eg., with brief stimulus presentation and/or a more relaxed error eriterion, Ss might ‘process two or more words in parallel. If the serial-decision model is valid for the present experiments, then one can use the yyes-no daa to estimate the time taken in de- biding that a string of letiers is a word, In particular, let Zn represent the mean RT Fespond "20" to’ nonword displayed above « word. Let Taq represent. the mean RT to respond "no" to a nonword displayed below a ‘word. Then, with certain assumptions (ef. Sternberg, 1969), the difference Tuy — Tye is & ‘measure of the mean time to decide that the top string is a word. From the results of Exp. I, ‘an estimate ofthis difference is 183: 14 msec, ‘An oceasional reversal in the order of stimalus processing would make this difference an “underestimate of the true mean, ‘One can also estimate approximately how much time is required to compare the outcomes of the two decisions before same-different responses. For exemple, suppose the mean RT of "yes" responses (Exp. 2 Jr sbicied from he meen RT of “a responses to pairs of words (Exp. n with certain ossomplions, the dierence of 216 £ 68 msco. is an estimate of the Comparison. time when the two decisions ‘match. On the ether hand, suppose the mean RT of “no” responses toa word displayed above a nonword is subtracted from the corresponding mean RT for “different” responses. Then the difference of 231 + 76 see. is an estimate of the comparison time ‘Wien the to detsons donot math ‘What kind of operation occurs during each ofthe two proposed deisions? One possibly is that visual and/or accoustic features of a string of Jetters are used to computo an "address" in memory (Normen, 1969; Schiffrin & Atkinson, 1969), A’ lexical decision about @ siring might then involve accessing and checking some part of the contents of the siring’s computed memory location (of, Rubenstein et al., 1970). Given this model, memory locations would be computed. for both words and nonwords, although the contents of non-word locations might differ qualitatively from those of word locations. In essence, we are therefore suggesting that both words and non-232 words may have locations "reserved" for them in long-term memory. ‘The effect of association on RT does not necessarily imply that tho meaning of a word is retrieved to make a lexical decision, To understand why, consider the following elab- oration of the serial-decision model, which may explain the effect. First, suppose thet Jong-term memory is organized semantically, i.e. that there is a structure in which the lo cations of two associated words are oloser than those of two unassociated words. Evie dnce from other studies of semantic memory suggests that this assumption is not total unreasonable (Collins & Quillian, 1969; Meyer, 1970). Let Ly and Lo denote the ‘memory locations examined in the first and second” decisions, respectively. Seoond, suppose that the time taken to make the second decision depends on where Lis relative to Ly In particular, lot us assume thatthe time taken sccossing information for the second desision ‘aries direotly with the "distance® botween 1 and 1. Then responses to pais of associated ‘words would be faster than those to pairs of “unassociated words, This follows because the proximity of associated words in the memory structure permits fasior accessing _ of information for the second decision, The. argument holds even if the accessed informs- tion is (@) sufficient only to determine whether a string is a word and (b) does not include aspects of its meaning four sooond assumption above is correct, then any retrieval operation Ra that is required sufficiently soon after another operation Ry will generally depend on R,, This would mean that human long-term memory, like many bulksstorege deviees, lacks the property known in the computer literature as random access (ef. MeCormick, 1959, p. 103). Re- cently, Meyer (1971) has oollesied data in other tasks that are consistent with this notion. ‘There are several ways in which this de- plsnee between retseval oprations might realized, One possibility is that retrieving information fiom a particular memory locati produces a passive "spread of excitation” to other nearby locations, facilitating Tater re~ trieval from thet (Collins & Quillian, 1970; Warren, 1970). A. second speculative possi bility is that retrieving information from long-ferm memory is like retrieving informa- tion from a magnetic tape or disk. In this Tater model, fecilitation of retrieval would secur Because (a) information can be "read out" of only one location daring any given DAVID E. MEYER AND ROGER W. SCHVANEVELDT instant, (6) time is required to "shift readout from one focation to another, and (c) shifting time, inereases with the distance botween cations. ‘The present data do not provide a direct test betwen this location-shifting model and the spreading-exeitation model. However, the location-shifting model may oxplain’ one ‘result that is difficult to account forin terms of spreading excitation. In particular, consider the following argument about the finding thet "different" responses were faster when @ word vas displayed save '¢ neaword, We pe. viously have at i processing normal begin witha decision about the fop sng and then proceeds to a decision about the bottom ‘one, Let us now assume that memory is oF ined by familiarity as well as by meaning, frequently examined locations in one "sector" and infrequently examined locations jn another sector. Recently, Swanson and Wickens (1970) have collected date ‘supporting a similar assumption that Oldfield (1966) has made. Suppose further that before each tial, a Tocation is preselected in, the sector where familiar words are stored, which ‘would be optimal under most eircumstances (Gf. Oldfield, 1966). Then the response to « word displayed above a nonword would require only one major shift between memory locations in the familiar and unfamiliar sectors. This shift would occur after the first decision, changing readout from the familiar to the unfamiliar sector.> In contrast, the response to anonword displayed above a word ‘would require two major ., one from the familiar tothe unfamiliar seotor before the first decision and one returning tothe familiar soctor before the second decision. This would ‘make "different" responses slower when the nonword is displayed above the word. More ‘over, the assumption that the starting location is in the familiar sector fits with the finding that lexical decisions are generally faster for familiar than for unfamiliar _ words (Rubenstein ot al, 1970); Le., « major shift between locations is required to access potential information about an unfamiliar word, whereas such a shift would not be required for a furnilier one Viowpoit of revival, a omword tt hglotmay betel os very ontmiir word Wows looston ix examined infrequently.FACILITATION IN WORD RECOGNITION 233 The effect of association on “same” responses 0 paits of words Gxp. I) is also relevant fo a recent finding by Schaeffer and Wallace (1969). In their study, Ss. were presented With a par of words and required to respond “sume” if both words belonged to the semantic category LIVING. THINGS or if both belonged to the category NONLIVING THINGS. Otherwise, Ss_ responded “dit. ferent." Reaction time of "same" responses varied inversely with the semantig similarity ‘ofthe words in the pair ¢., "samo" responses to astimulus like TULI-PaNSY were Taster than "same" responses to « stimulus like TULA. “In contrast, Schaeffer ‘and Wallace (1970) found thatthe RT of "dif ferent” responses varied diroctly with semantic similarity. ‘They attributed the ‘effets of similarity on both "same" and "different" responses toa process that ompares the mesnings of the! words in a stimulus. ‘The effects of association in Exp. Land IL possibly could have been caused by such a comparison process, rather than by the retrieval mechanisms discussed above. However, if the "meaning" of a word is represented by the semantic celegories to which it belongs, then there seemingly is a difference botween the same-differcnt task of Exp, Tl and the one studied by Scheofior and Walia (1369, 1970). Lagically, Exp id not require S' to compare the meanings o the tems ina stimulus; Fe, Se did not have to judge whether both strings belonged to the semantic category, 0-8. THINGS. Instead, Exp. Tt only required comparing the items lexical status. Moreover, inparison of meanings would have been impossible for those pairs involving at least ‘one noaword, since the nonword would. have rho meaning in the usual senso. Ono might therefore argue that Ss did not compare the mosnings of tems in Exp. TL. The argument isroinforoed bythe fue that Exp. 1 Gjes-n0 sa "whieh oily inate comparing the stings in any "tay, prodiced mn effect ofasoviation like the ons Oerved in Exp Th ‘Our reasoning sopgcats,frthemoce, that the findings of Schneier and 1970) may not ave resulted #0 Comparison of word meanings. Rater, thir findings could have been cansed at least in part by a retioval process te an ~ posed. This point i supported by the magn {des ‘of the amulary eleats they observed, Which averaged 176 msco, for fctiating "Same" responses (Schacter & Wallace, 1969) and 120 mseo. for inhibiting "different re sponses (Schaeffer & Wallace, 1970), In par- ticular, consider the following detsiled argu- ment. Suppose that judgments in thoir task involved two components: an initial retrieval proves similar to the one we have proposed, Which might be necessary to access: word meanings, and a process that compares word meanings’ (ef. Schaeffer & Wallace, 1970), ‘Suppose further that our exporiments required ‘only the first proves. One might then expeot ‘that whenever both ofthese processes are used in *same" judgments, they would both be th- litated by semantic similarity. However, ‘when thoy are used in “different” judgments, similarity would inhibit the comparison process while still fecilitating the retrioval brocess. This would explain why the effet of association on "same" responses in Exp. (G17 msec.) was less than the effect of sinuilarity on "samo" responses in the study by Schaeffer and Wallace (1969). Moreover, it ‘would also explain their finding that semantic “different” responses less ted "same" responses Unfortunately, the argument is partially ‘weakened by atleast one fat; ie. thir results for "same" versus "different® rosponses were otlnined in Separate experiments. using somewhat different semantic categories an text words. Regardless of whether spreading excitation, location shifting, comparison of meanings, of some other process is involved, the effects of «assooiation appear limited neiter to semantic ssmor to same-different judgments. At resent we do not have ways to test all the possible explanations of these effects. How- ‘ever, procedures like the ones wo have de- soribed may provide a way to study relations baween retrieval gperatons that are tee orally contiguous. We may therefore be eble {0 eam more about both te naire OF ine dividual memory processes and how they affect one another. ‘REFERENCES BousFIELD, W. A, Conen, B. H, ‘Warnarst, G.A,"& Kincatn, W. D. The Connecticut free associational norms. Tech. Rep. ‘No.35) Sioas, Coun: University of Connectiont, 1961. (COLLINS, A. M,, & QUILLAN, M. R. Retroval time ‘Couns, A.M, ‘lrieval from’ semantio memory: The effect ‘epeating part of an inforonco. in A. F. Senders (Ed) Attention and performance Uf. Amsteréant, Nosti:Holland Publishing Company, 1970,234 Kucera, H., & FRANGS, W. N. Computational anabsis of present-day American English. Provi- sence, RL: Brown University Press, 1967, LaxaUin, T. K., & Fesspainx, J. , Information retrival from Tong-term memory: Category size ‘and recognition time, Joumal of Verbal Leaming ‘and Verbal Behsto, 1968,7,291-295. MeConack, E. M. Digital computer primer. New “York: MeGraw-Hill, 1959. \MaYaR, D. E.On the representation and rotrioval of stored semantic information. Cognitive Psy- cholo 1970, 1,242-300, ‘Mevax, D. B. Duel memory-search of related and ‘unrelated semantic categories. Papor prosontod atthe mestiag of the Eastem Psychological ASSo- ciation, Now York, Apil L971 Mavue, D.E, & ELS, G. B. Panllel processes in ‘word recognition. Papor presented atthe meeting of the Paychonomic Society, San Antonio, No- vember 1970. NoRMAN, D. A. Comments on the information strveture of momery. In A. F. Sandere (Ed), Attention and peformance Il. Amsterdam: ‘Nocth-follaad Publishing Company, 1970. OLAELD, RC. Things, words and the brain, ‘Quarterly Jounal of Experimental Paychology, 1966, 18,340353, DAVID B. MEYER AND ROGER W. SCHVANEVELDT Rupeystes, H, GARAEID, 1, & MIEKA LA Homographic entries in tho intomal lexicon. Joumal of Verbal Leaming ane Verbal Behavior, 1970, 9, 487-494 ScHARFTE, B., & WALLACE, R. Semantic similarity and the comparison of word memings. Joumal of Experimental Psychology, 1969, 82, 43-346, Scunuem, B., & WALLACE, R. The comparisen of word memings. Joumal of Experimental P5y- chology, 1970, 86, 14-152. Scimeram, RM, & ATKDSON, R. C, Storago and retrieval processes ia long-term memory. Psy chological Review, 1969, 76, 179-193, ‘Swanson, J. My, & WICKINS, D. D. Preprovessing 1 the basis of frequency of oocurencs. Quarterly Jounal of Experimental Poychology, 19, 22 378383, STDBERO, S. Memory sanning: Mental process revealed by reactionstime experiments. American Scent, 1969,57, 421-457. WARREN, R. E. Sinus encoding end memory. Unpublished doctoral dissriation, University of regen, 1970. Was, B. J. Statistical principles in experimental design. New York: MeGra-Hil, 1962. (Received Febmary 6, 1971)