Long-term semantic priming of propositions representing general knowledge

Journal of Memory and Language 79-80 (2015) 30–52

Contents lists available at ScienceDirect

Journal of Memory and Language

journal homepage: www.elsevier .com/locate / jml

Regular Paper

Long-term semantic priming of propositions representinggeneral knowledge

http://dx.doi.org/10.1016/j.jml.2014.11.0020749-596X/� 2014 Elsevier Inc. All rights reserved.

⇑ Corresponding author at: University of Utah, Department of Educa-tional Psychology, 1705 E. Central Campus Dr., Rm 327, Salt Lake City, UT84112-9255, United States. Fax: +1 (801) 581 5566.

E-mail address: [email protected] (D.J. Woltz).

Dan J. Woltz ⇑, Linda J. Sorensen, Timothy C. Indahl, Adrienne F. SplinterUniversity of Utah, United States

a r t i c l e i n f o

Article history:Received 28 April 2014revision received 14 November 2014

Keywords:Semantic primingLong-term priming

a b s t r a c t

Six experiments investigated semantic priming of individual propositions that reflect com-mon knowledge. Participants performed a sentence-completion task in which each sen-tence expressed a single proposition, and the same or similar proposition was expressedin prime-target sentence pairs that shared no words. The primary goal was to evaluatethe degree to which long-lasting facilitation in primed target sentences reflected memoryfor abstract proposition meaning versus memory for the operations used to construct prop-osition meaning and select a response. Priming effects persisted over delays exceeding15 min, were not attributable to explicit memory processes, did not depend on eitherrepeated syntactic structure or response decision, and were greater than facilitation attrib-utable to priming individual word meanings. In total, the evidence provides initial supportfor operation-independent, long-term semantic priming of propositions outside the con-text of connected discourse. This extends the empirical evidence for long-lasting facilita-tion that theories of semantic priming must explain.

� 2014 Elsevier Inc. All rights reserved.

Introduction

Early demonstrations of semantic priming by Meyerand colleagues (Meyer & Schvaneveldt, 1971; Meyer,Schvaneveldt, & Ruddy, 1975), and the majority of subse-quent semantic priming research, has relied on two rela-tively simple verbal tasks: lexical decision and wordnaming. In some respects, the term semantic primingbecame synonymous with facilitation effects in these tasks(see McNamara, 2005). Theory and evidence from theselines of research established semantic priming as a short-lived facilitation on semantically related words.

However, there is growing evidence that tasks requiringgreater degrees of semantic processing produce non-stra-tegic priming effects that persist beyond what theories

based on the simpler tasks can explain (Becker,Moscovitch, Behrmann, & Joordens, 1997; Joordens &Becker, 1997; Tse & Neely, 2005, 2007; Was, 2010;Woltz, 2010; Woltz & Was, 2007). It appears that semanticpriming has different durations and perhaps differentunderlying mechanisms with different task demands. Sofar, the evidence for longer-term semantic priming hascome from tasks that have increased semantic processingdemands, but these tasks all assess priming with respectto individual words. In popular theoretical views of knowl-edge representation, the fundamental unit of knowledgethat has a truth value is the proposition (e.g., Anderson,1983, 1993; Kintsch, 1974). In this research we extendedthe range of complexity in semantic priming tasks toinclude the processing of propositions representing gen-eral knowledge. We investigated relatively long-termsemantic priming in expressions of isolated propositions,and evaluated alternative explanations to the premise thatthe persistent facilitation is attributable to increased mem-

http://crossmark.crossref.org/dialog/?doi=10.1016/j.jml.2014.11.002&domain=pdf

http://dx.doi.org/10.1016/j.jml.2014.11.002

mailto:[email protected]

http://dx.doi.org/10.1016/j.jml.2014.11.002

http://www.sciencedirect.com/science/journal/0749596X

http://www.elsevier.com/locate/jml

D.J. Woltz et al. / Journal of Memory and Language 79-80 (2015) 30–52 31

ory strength in abstract representations of semanticknowledge. The alternative explanations tested correspondto those debated in recent research on long-term semanticpriming in words.

The idea of semantically mediated access to propositionmemory is hardly new; memory for propositions encodedduring text processing has received considerable attentionin reading research. However, we suggest that suchresearch cannot easily isolate the role of abstract proposi-tion memory from the broader meaning context in whichthe propositions are embedded. Furthermore, much of thisresearch has investigated memory for newly formed epi-sodic propositions in discourse. Our purpose was to inves-tigate the nature of semantic priming in isolatedpropositions that represent general knowledge, with thegoal of extending the emerging theoretical understandingof long-term semantic priming.

Evidence and explanations of long-term semantic priming

As noted, most theories of semantic priming cannotexplain evidence for longer-term semantic priming effectsin word processing tasks. This differs from evidence andtheory pertaining to long-lasting repetition priming effectsthat may depend on memory for perceptual or encodingoperations (Kolers, 1973, 1976; Tulving & Schacter,1990). Temporary semantic priming effects can beexplained by spreading activation processes (McNamara,1992; Neely & Kahan, 2001), compound cue theories(Dosher & Rosedale, 1989; Ratcliff & McKoon, 1988) anddistributed network memory models (Masson, 1995;Plaut & Booth, 2000). These theoretical explanations positmemory processes that are displaced quickly by subse-quent processing, so none predicts that semantic primingeffects would be detected after several intervening trials(for discussion of this see Becker et al., 1997, andMcNamara, 2005). Yet, Joordens and Becker (1997)reported priming effects in lexical decisions that spannedas many as eight intervening trials. Tse and Neely (2007)reported similar evidence within a lexical decision task.

Joordens and Becker (1997) suggested that, comparedto previous research, it was the increased semantic pro-cessing demanded by their version of the lexical decisiontask that produced the longer-term semantic primingeffects. Consistent with this idea, Becker et al. (1997)reported semantic priming effects that persisted over 20unrelated trials in a task that required animacy decisionsof category exemplars. Tse and Neely (2005, 2007) foundlexical decision priming after a 30-s unrelated task. Theirpriming manipulation consisted of studying multiplesemantically related words rather than a single lexicaldecision trial. Woltz and Was (2007) and Was (2010)reported long-term priming effects in category exemplardecisions. Participants showed persistent facilitation indeciding if two exemplars were members of the same cat-egory when they had made previous decisions about dif-ferent exemplars from the same category. Finally, Woltz(2010) reported long-lasting semantic priming effects ina task that required participants to decide if two wordswere synonyms. The facilitation showed a slow, gradualdecline but was still evident after 14 unrelated trials.

The evidence linking longer-term semantic priming totask complexity suggests that proposition priming wouldbe relatively long lasting given the inherent semantic com-plexity of encoding propositions versus individual words. Ifso, it is important to consider whether the current theoret-ical view that accounts for long-term word-level primingcould explain persistent proposition priming. Becker et al.(1997) proposed a connectionist network model that wasunique in positing persistent rather than temporarychanges in activated network connections from processinga semantically related prime word. Each target word is pre-sumed to have a partly overlapping network representa-tion with semantically related words. The time requiredfor target words to settle to their network activation pat-terns is reduced when previous prime-word processinghas strengthened a subset of the target stimuliconnections.

Other connectionist networks account for temporarysemantic priming (e.g., Masson, 1995; Plaut & Booth,2000), but the network proposed by Becker et al. (1997)uniquely explains long-term semantic priming effects bythe impact of stimulus inputs on connection weights inthe network. Other connectionist models view networkunits as somewhat similar to switches: Meaning represen-tations are comprised of patterns of units being turned onor off. Priming, then, is the result of a target network statebeing similar to that created by a previous prime word.Such a network only allows for temporary activation statesbecause unrelated stimuli change the pattern of unitsturned on (Masson, 1995). In contrast, Becker describedthe depth of attractor basins for target words resultingfrom persistent weight changes in network connectionsfrom prior activation. In this way, facilitation due to over-lapping patterns of activation between prime and targetstimuli can be longer lasting and resistant to interveningstimuli.

The Becker et al. (1997) distributed network model’sability to account for long-term semantic priming at theword level suggests that a similar approach could be usedto explain long-term priming in semantically related prop-ositions. Sharkey (1990) discussed the representation ofproposition units in memory within a connectionist modelof text comprehension. This model describes a propositionlevel within a connectionist network in which there areexisting patterns for different proposition units. Theseproposition patterns are activated based on the text beingread, such that one proposition pattern is ultimately acti-vated. A model such as this, but one that incorporatedthe Becker et al. (1997) incremental learning mechanismwith attractor basins, could presumably account for persis-tent semantic priming in both word and propositionrepresentations.

A different explanation for long-term semantic primingeffects at the word level also has applicability to primingeffects at the proposition level. The unique feature of thisexplanation is that it attributes performance facilitationto repeated operations performed on the semantic contentof stimuli, rather than to strength changes in the semanticrepresentations themselves. Evidence supporting this viewcomes primarily from experiments using tasks requiringcategory exemplar decisions (Hughes & Whittlesea, 2003;

32 D.J. Woltz et al. / Journal of Memory and Language 79-80 (2015) 30–52

Woltz & Was, 2006, 2007). Task demands differed acrossexperiments reported by these researchers, but a commonfinding was that participants were facilitated in makingdecisions about either category exemplar membership orcommon features of categories when previous trialsrequired comparable decisions with different stimuli fromthe same category. Priming effects in these tasks persistedlonger than those found in lexical decision tasks; however,the observed facilitation could not be attributed to activa-tion or strengthening of abstract or generalized memoryfor a category. Instead, facilitation was evidenced onlywhen a specific category operation was repeated fromthe prime to the target event. For example, Woltz andWas (2007) found persistent priming if the prime eventrequired a decision as to whether words represented fea-tures of the same category (e.g., feather beak), and the sub-sequent target required another feature decision from thesame category (e.g., wing talon). There was no facilitation,however, from feature comparison primes on target trialsthat required a decision about whether exemplars repre-sented the same category (e.g., eagle canary). The same pat-tern of priming effects was found for exemplar categorymembership primes: They facilitated subsequent exemplarcomparisons but not subsequent feature comparisons.Both forms of category decision in this experiment pre-sumably depended on activating the category bird, but thisdid not result in persistent facilitation across the differentdecision operations.

Both Hughes and Whittlesea (2003) and Woltz and Was(2006, 2007) interpreted such findings in terms of opera-tion-specific memory for semantic operations. Persistent,semantically mediated facilitation was not attributable torepeated exposure to a semantic category but to the repe-tition of specific decision processes related to the category.This explanation for persistent semantic priming corre-sponds to earlier operation-based accounts of performancefacilitation (Kolers, 1973, 1976; Kolers & Roediger, 1984). Itis distinct in its attribution of facilitation to a proceduralmemory for the mental operations employed in each trial,rather than strengthened semantic memory of trialcontent.

An operation-based account of semantic priming ispotentially important to the investigation of propositionpriming. Even with different lexical expressions of essen-tially the same proposition in prime and target trials, var-ious processing operations required to construct andevaluate proposition meaning would likely be repeated.In this research we attempted to differentiate primingeffects that reflect memory for the abstract meaning ofpropositions versus those that reflect memory for the pro-cessing operations used to construct and evaluate themeaning.

Overview of experiments

In all experiments, we used a sentence-completion taskin which each target sentence could be primed by a seman-tically similar expression. Each prime-target pair wasintended to express a single proposition, and each expres-sion consisted of a transitive verb with a subject and object(e.g., consumers purchase products and shoppers buy

merchandise). No words were repeated across the semanti-cally related sentence pairs, and each word was used onlyonce in the entire stimulus set. The proposition expressedin each target sentence was designed to be unrelated tothat of all other sentences used in the task, other than ifit was primed by a semantically related sentence. The gen-eral independence of proposition content allowed us toevaluate priming effects attributable to processing a singleprior proposition expression, uninfluenced by contextualpriming that presumably occurs when sentences areembedded in coherent passages. We evaluated facilitationeffects with an average prime-target lag of at least 10intervening trials and, in some cases, with an interveningtask that lasted approximately 15 min.

In each experiment, one word of each displayed sen-tence was omitted, and participants were instructed tochoose the better of two alternative words to fill in theblank. For example, the incomplete sentence consumerspurchase _______ had products and insects as forced-choicealternatives. Sentences could be presented in active voice,as in the previous example, or in passive voice (e.g.,_______ are purchased by consumers). Different experimentsmanipulated various features of the task, but in all casespriming effects were assessed by an increased rate ofselecting the correct alternative in a target sentence dueto the earlier presentation of a semantically related trial.

Using this experimental task, we tested whether anypriming effects observed were dependent on repeatedoperations for encoding and evaluating the propositions.To the degree to which this was the case, facilitation wouldbe attributable to memory for prior mental operationsrather than abstract representation of semantic content.We also evaluated both the role of individual word primingand the potential influence of explicit, episodic memory forprime trial processing as alternative explanations to long-term priming of abstract proposition meaning. Experi-ments 1 and 2 evaluated the role of syntax consistencyfrom prime to target trials. Experiment 3 manipulatedwhether the subject or object was to be completed inprime and target sentences in order to evaluate whethersome or all of the observed facilitation was dependent onthe consistency of sentence-completion decision pro-cesses. Experiment 4 investigated the potential role of rec-ognition memory for prime trials in the facilitation oftarget trials. Experiment 5 evaluated the possible role ofnegative priming of the semantic content of incorrectalternatives rejected in prime trials. Finally, Experiment 6contrasted facilitation effects in target-sentence comple-tions due to priming individual word meanings versuspriming the entire proposition. This experiment providedan initial evaluation of a complex issue: whether long-termproposition priming effects could likely be explained byindividual word priming effects.

Given the expectation that priming effects would bedistributed over both response time (RT) and errors, wetested the hypothesized outcomes with a transformationthat combined these two dependent measures into a rateof correct responding. The use of this combined measureto assess cognitive performance is less familiar in the liter-ature than the use of RT, but such approaches have beenused by other researchers (e.g., Singley & Anderson,


1989; Townsend & Ashby, 1978, 1983). The primary moti-vation for using the combined measure here is that priorsemantic priming research with moderately complex tasksfound facilitation effects distributed across both RT anderrors (Becker et al., 1997; Joordens & Becker, 1997; Tse& Neely, 2005; Was, 2010; Was & Woltz, 2007; Woltz,2010; Woltz & Was, 2006, 2007). Moreover, evidencereported by Woltz (2010) demonstrated that the magni-tude of semantic priming effects was underestimated inanalyses using RT compared to analyses using an index ofcorrect response rate. For nine priming effects evaluatedwith respect to both correct response rate and RT, the aver-age partial g2 was .34 for the former and .22 for the latter.

The correct response rate transformation divided eachparticipant’s proportion correct for a given design cell bythe average RT for all trials of the cell, thus producing anindex of the rate of correct responding (i.e., the numberof correct responses per total time of all responses). Theadvantages of this transformation are (1) it incorporatesboth error and latency variance, which generally resultsin a more powerful test of priming effects while reducingpartially redundant analyses; (2) as a reciprocal transfor-mation of RT, it tends to approximate the normal distribu-tion better than RT; and (3) it has the potential to adjust fordiffering speed-accuracy tradeoffs (Sorensen & Woltz,2014). The first of these advantages is particularly relevantto the current experiments because null results wouldhave theoretical importance. The use of the correctresponse rate index reduces the likelihood of Type II errorsof interpretation because of its positive impact on the mag-nitude of priming effects. Consequently, the primaryhypothesis tests were conducted using this dependent var-iable. However, because most readers have greater famil-iarity with RT and error measures, and because themetric properties of the new measure are not well under-stood, ancillary analyses with the more traditional depen-dent measures are also included parenthetically. Forsimilar reasons, statistical tests with items as the unit ofanalysis are reported.

Experiment 1

Two issues were addressed in the first experiment. Ofprimary importance, we tested the existence of long-termsemantic priming effects in the completion of sentencesthat represent isolated propositions. Facilitation was con-sidered to be long term if it persisted over several seman-tically unrelated intervening trials. The second issuepertained to the abstractness of the expected semanticpriming effects. If facilitation was reduced by the use ofdifferent syntaxes in prime and target sentences, thenfacilitation was assumed to be partly or entirely dependenton the repetition of prior comprehension operations dic-tated by sentence structure. Little or no effect of syntaxchange on target facilitation would be consistent with apriming explanation that refers only to the strengtheningof abstract proposition meaning during prime trial perfor-mance. We manipulated syntax consistency betweenprime and target trials by varying active and passive voiceexpressions of the propositions.

There seems to be a long-standing general consensusthat memory for the surface form of verbal expressions isshort-lived compared to the semantic content (e.g.,Anderson, 1974; Gernsbacher, 1985; Sachs, 1967, 1974).However, there is also evidence that the syntactic structureof previously processed expressions can influence subse-quent processing. Most notably, there is a large literatureregarding syntactic or structural priming on subsequentsentence production (see Pickering & Ferreira, 2008, for areview). Most of this evidence demonstrates the abstract-ness and relative persistence of memory underlying thisform of priming. Furthermore, some evidence supports aview that structural priming in sentence productionreflects procedural rather than declarative memory pro-cesses (Ferreira, Bock, Wilson, & Cohen, 2008). This proce-dural interpretation of persistent structural priming wouldbe consistent with an operation-based or procedural mem-ory explanation of persistent proposition priming whichpredicts facilitation due to memory for the syntax-depen-dent encoding operations.

However, it is important to note the substantial differ-ences between the structural priming literature and thecurrent questions and methodology. First, we investigatedpriming effects on comprehension rather than production.Although some structural priming research has used tasksassessing sentence comprehension (e.g., Branigan,Pickering, & McLean, 2005; Ledoux, Traxler, & Swaab,2007; Traxler, 2008; Traxler & Tooley, 2008), the majorityof this research has focused on sentence production. Sec-ond, in contrast to most structural priming research, wewere not interested in facilitation attributable to abstractmemory representations of syntactic structure (i.e., facili-tation due to common sentence structure that generalizesacross different semantic content). Instead, we were inter-ested in content-specific facilitation attributable to syntax.Our question was, do prime-target facilitation effects in theevaluation of semantically related proposition expressionsdepend on the repetition of syntactic structure?

One way to conceptualize the content-specific facilita-tion due to repeated sentence syntax is in terms of theinformation structure of sentences and the correspondingfocus of attention during meaning construction. Active-voice expressions of a transitive relation emphasize theAgent, or doer of the action. Passive-voice expressionsplace emphasis on the Patient, or receiver of the action. Ifthere is greater semantic priming in semantically relatedpropositions expressed with the same voice, this couldreflect memory for semantic operations associated withthe unique information structure of each voice. Such aneffect would be consistent with operation-based explana-tions of long-term semantic priming.

In this experiment, we presented prime–target sen-tence pairs that were designed to be expressions of a singleproposition, and they varied in syntax with respect toactive and passive voice. Half of all prime trials expressedpropositions in active voice and half expressed them inpassive voice, with the voice of target trials varied orthog-onally to that of the corresponding prime trials. If facilita-tion only occurred, or occurred with greater magnitude,when the voice of prime and target sentences matched,then any semantic priming effects observed would reflect,


in part or whole, memory for syntax-specific operations forconstructing the shared proposition meaning. To investi-gate relatively persistent rather than temporary facilita-tion, prime trials were separated from target trials by anaverage of 10 intervening unrelated sentence completions.

Method

ParticipantsA total of 96 undergraduate students (80% female) par-

ticipated in this experiment for credit in a college of educa-tion course. Median age of the sample was 21 (range 18–55).

ApparatusParticipants performed the experimental task on PCs

with SVGA monitors and standard keyboards. Program-ming of all tasks was completed with E-Prime� software(Schneider, Eschman, & Zuccolotto, 2002).

MaterialsA set of 96 experimental proposition expressions was

constructed such that each had two predicate objectchoices (valid and invalid). For each of these 96 expres-sions, a semantically parallel expression was generatedsuch that corresponding words in the two expressionswere synonyms (e.g., mules carry bundles/inspirations andburros haul packs/motivations). One expression was desig-nated as a prime sentence and the other was designatedas a target sentence. Moreover, each proposition expres-sion was constructed in both active voice, as in the currentexample, and passive voice (e.g., bundles/inspirations arecarried by mules and packs/motivations are hauled by bur-ros). Finally, an additional set of 96 expressions was cre-ated which were similar in format to the experimentalstimuli, except they did not have the correspondingsemantically related expressions required of the prime-tar-get pairs. These stimuli served as filler trials to be insertedbetween prime and target trials and as warm-up trials atthe beginning of each trial block. Filler propositions wereexpressed in active and passive voice, and they shared nowords with the experimental stimuli.

ProcedureEach trial began with a row of asterisks presented for

1 s. After a 1 s blank screen, a three- to six-word sentencewas presented with one word missing (e.g., mules carry_______). The predicate object alternatives (e.g., bundles;inspirations) were displayed below the sentence, one abovethe other. They were presented in a centered vertical arraybecause the blank in each sentence could appear for thefirst or last word, depending on whether the propositionwas expressed in passive or active voice, and we did notwant a horizontal array of alternatives to bias responseselection. The 2-choice completion format was used ratherthan a yes–no fact verification because previous evidencefrom word meaning evaluation tasks indicated that primetrials requiring a no responses produced little or no long-term facilitation (Woltz, 2010). The 2-choice format whichprecluded a negative response allowed for more efficientuse of a limited pool of appropriate stimuli.

Participants were instructed to rest their index fingerson the T and B keys and to select their answer by pressingT for top word or B for bottom word. The position of thecorrect response was randomly determined on each trial.To ensure that participants read each sentence fully beforemaking their choice, a stimulus onset asynchrony (SOA) of1 s was implemented between the appearance of the sen-tence and the response options. There was a 2 s delaybetween each response and the beginning of the next trial.No feedback was provided on individual trial responses.

Sentences were presented in blocks of 20 trials. Eachblock began with four filler trials as warm up: two passive-and two active-voice sentences in random order. Followingthe initial filler trials, there were four prime trials. Twowere active-voice primes, one corresponding in meaningto a subsequent active-voice target and one correspondingin meaning to a later passive-voice target. The other twoprimes were passive voice, one corresponding in meaningto a later passive-voice target and one corresponding inmeaning to a later active-voice target. The four prime trialswere presented in random order. These were followed byfour new filler trials, two active- and two passive-voicesentences, randomly ordered. Finally, there were eight tar-get trials in random order. There were two active- and twopassive-voice target trials corresponding in meaning to theprime trials as described. There were two active- and twopassive-voice unprimed trials that were from the sameexperimental stimulus set, but their prime trials had notbeen presented previously. Given the balance of active-voice and passive-voice sentences throughout the blockand the randomized order of these within-block segments,we assumed that any general syntactic priming of the twosyntax structures would be equivalent across the semanticpriming conditions. Utilizing this sequence of trials consis-tently across blocks, the average lag from prime to relatedtarget trial was 10 unrelated trials.

For purposes of counterbalancing stimulus assignmentto trial condition, the 96 experimental propositions weredivided into eight sets of 12. The eight stimulus sets wereassigned with equal frequency to the trial conditions rep-resenting a 2 � 2 � 2 crossing of the following task vari-ables: passive versus active target voice, same versusdifferent prime voice, and primed versus unprimed target.However, note that for unprimed target trials, voice of theprime trial was irrelevant because the prime trial was notpresented. Therefore, although a 2 � 2 � 2 factorial designwas used for stimulus assignment and counterbalancing,the design for data analysis of target trial performancewas reduced to a 2 (active versus passive target voice) � 3(unprimed versus primed by same voice versus primed bydifferent voice) factorial.

Participants performed the experimental task in a single1 h session. They participated in groups of one to six sub-jects, with each participant seated in a computer carrelseparated by sound-deadening panels.

The session began with computer administered instruc-tions and a block of 12 practice trials. Participants wereinstructed to respond as quickly as possible without sacri-ficing accuracy. Practice trials were followed by the 12experimental blocks, with summary feedback of percent-age correct and average RT displayed following each block.

Target Trial VoiceActive Passive

Res

pons

e S

peed

(cor

rect

tria

ls/m

in)

0

40

42

44

46

48

50

52

54

56

58UnprimedPrimed Different VoicePrimed Same Voice

Fig. 1. Mean correct response rate (correct trials/min) for target trials ofExperiment 1 by sentence voice and priming condition. Error barsrepresent 95% confidence intervals for within-subject designs (Loftus &Masson, 1994).


The summary block feedback was self-paced, allowing par-ticipants to take a brief rest following each block. Partici-pants were reminded to respond as quickly as possiblewithout sacrificing accuracy after each summary blockfeedback.

Results

Responses to prime trials in the active voice were fasteron average (M = 1214 ms, SD = 480) than responses to pas-sive voice primes (M = 1326 ms, SD = 648). There were alsoslightly fewer errors in active-voice primes (M = 4.5%,SD = 5.5) compared to passive-voice primes (M = 5.5%,SD = 6.4). These trends presumably reflect the longerlength and additional processing complexity of passive-voice sentences.

Table 1 presents mean response latency for correctresponses and percentage errors for the target trials bypriming condition. As can be seen in this table, responsesto passive-voice target trials again took longer and weremore error prone than responses to active-voice targets.More importantly, there was a trend in trials of both voicesfor facilitation due to priming, and this trend was seen inboth error and latency data. Of interest, the magnitude offacilitation in both dependent measures did not appear tobe markedly greater when prime and target trials were ofthe same voice.

Fig. 1 presents means of the rate of correct respondingfor target trials by condition. An analysis of variance testedtwo orthogonal, within-subject contrasts corresponding tothe means shown in Fig. 1. The first contrast tested the over-all priming effect by comparing unprimed performancewith that of both primed conditions. The second contrasttested the difference in performance between the same-voice and different-voice primed conditions. There was amain effect for target voice, F1(1,88) = 38.36, p 6 .001, par-tial g2 = .30 [response latency F1(1,88) = 33.01, p < .001,partial g2 = .27; percentage errors F1(1,88) = 7.04, p = .009,partial g2 = .07], with passive-voice responses being slowerthan active-voice responses. In addition, there was a sub-stantial overall priming effect indicated by the contrast ofunprimed trials and both priming conditions, F1(1,88) =8 2.64, p 6 .001, partial g2 = .48 [response latencyF1(1,88) = 35.33, p < .001, partial g2 = .29; percentage errorsF1(1,88) = 13.72, p < .001, partial g2 = .14]. However, as canbe seen in Fig. 1, there was no benefit of priming in the samevoice as indicated by the nonsignificant contrast betweenthe same- and different-voice prime conditions,F1(1,88) < 1 [response latency F1(1,88) < 1; percentage

Table 1Mean response time for correct responses and percentage error in target trials of

Trial type Response errors (%)

Active voice Passive voice

M SD M

Unprimed 6.8 7.6 7.6Primed-Same 4.6 7.4 6.7Primed-Different 4.6 7.3 5.3

Note. Primed-Same and Primed-Different refer to priming by trials of the same

errors F1(1,88) = 1.23, p = .27]. Given the relevance of thisnull finding for understanding priming processes in thistask, a Bayesian posterior probability of the null hypothesisgiven the data (PH0|D) was estimated using a method devel-oped by Wagenmakers (2007) and described by Masson(2011). The estimated value was relatively high,PH0|D = .91, supporting the conclusion that memory for syn-tax-specific processing is unlikely to have affected priming.Finally, none of the interactions was statistically significant:The overall priming effect did not differ for active and pas-sive targets, and there was no interaction of target voiceand type of prime (consistent versus inconsistent voice),F1 < 1 in both cases.

Analysis of the data by items rather than participantsyielded similar results, although the effects were smaller.There was a main effect for target voice, F2(1,95) = 27.48,p 6 .001, partial g2 = .22 [response latency F2(1,95) = 19.62, p < .001, partial g2 = .17; percentage errorsF2(1,95) = 5.20, p = .025, partial g2 = .05], with passive-voice presentation being slower. In addition, there was asubstantial overall priming effect indicated by the contrastof unprimed presentation and both forms of priming,F2(1,95) = 31.36, p 6 .001, partial g2 = .25 [response latencyF2(1,95) = 22.82, p < .001, partial g2 = .19; percentage

Experiment 1 by sentence voice and priming condition.

Response latency (ms)


SD M SD M SD

7.8 1396 552 1591 8048.8 1313 604 1427 7167.8 1316 575 1439 693

and different voice, respectively, compared to the target trial voice.


errors F2(1,95) = 13.98, p < .001, partial g2 = .13]. However,as with the analysis by participants, there was no benefit ofpriming in the same versus different voice of the targettrial, F2(1,95) < 1 [response latency F2(1,95) < 1; percent-age errors F2(1,95) = 1.94, p = .17].

Discussion

Semantic priming of proposition content persisted foran average of 10 unrelated sentence evaluations. Therewas an average increase in correct response rate of over10% attributable to previous exposure to a semanticallyequivalent sentence-completion trial. There were noshared words in prime and target expressions, so facilita-tion presumably reflected the impact of abstract ratherthan lexically specific representations. The persistence ofthis form of semantic priming is greater than what hasbeen reported from simpler priming tasks that requireword recognition or naming (see McNamara, 2005) and,as expected, is consistent with evidence for longer-termsemantic priming reported in tasks with greater semanticcomplexity (Becker et al., 1997; Joordens & Becker, 1997;Tse & Neely, 2007; Was, 2010; Woltz, 2010; Woltz &Was, 2007).

In contrast to the relatively robust overall primingeffect spanning multiple unrelated trials, there was noeffect of prime and target sentences being presented inthe same versus different voice. Compared to previousdemonstrations of persistent syntax priming effects, thecurrent lack of a prime–target voice consistency effectlikely has several contributing factors. First, longer-termsyntax priming seems most readily demonstrated in tasksthat require the production rather than comprehension ofa primed syntactic structure (e.g., Bock, 1986; Bock &Griffin, 2000; Pickering & Ferreira, 2008). Second, corre-sponding prime and target trials of this experiment wereseparated by an average of 10 trials that contained bothactive- and passive-voice sentences. Consequently, a singlesyntactic structure was not activated prior to a target sen-tence; presumably abstract representations of both activeand passive structures were highly activated. Any syntax-dependent facilitation in target trials would require mem-ory representation for prime trial syntax to be content-spe-cific rather than abstract and independent of semanticcontent. An operation-based explanation of long-termsemantic priming predicts that target trial facilitationshould depend on repetition of prime trial comprehensionprocesses associated with sentence syntax and informationstructure. Given the results, the observed facilitation waslikely due to the memory for abstract meaning rather thanfor prior operations used to construct the meaning.

Experiment 2

The average lag of 10 trials between prime and targettrials in Experiment 1 corresponds roughly to delays thateliminated syntax memory effects under some previoustask manipulations (Anderson, 1974; Sachs, 1967). Thequestion addressed in Experiment 2 is whether there areimmediate syntax consistency effects in sentence-comple-

tion priming that fade over multiple intervening trials, orwhether the memory representations that underlie thesepriming effects contain no information about syntacticstructure or the comprehension operations dictated bythe structure. The latter case predicts no syntax consis-tency effect even when there is no intervening trialbetween primes and targets. To test this, and to further testthe persistence of the semantic priming effect, we variedthe prime-target lag between zero and approximately 15intervening trials.

Method

This experiment utilized the same stimulus materialsand general method as Experiment 1. Only task differenceswill be described here.



ProcedureThe task structure was similar to that of Experiment 1,

but a modification was made in order to vary short andlong prime-target lags. Each of six blocks contained 32 tri-als of the form used in Experiment 1. The first four trials ofeach block were warm-up trials. These were followed byfour prime trials for the long-lag condition (one for eachcell in the crossing of prime voice and target voice). Thena series of 16 trials contained four Lag 0 prime-target pairs(one pair for each cell in the crossing of prime and targetvoice), randomly mixed with another eight trials compris-ing four unprimed targets (two active and two passive) andfour filler trials (two active and two passive). The filler tri-als were included to reduce the predictability of semanti-cally related pairs. The final eight trials containedrandomly ordered target trials representing the long-termlag condition. There were two active- and two passive-voice target trials corresponding to the initial prime trialsof the block. There were two active- and two passive-voiceunprimed trials.

A final modification from Experiment 1 was the elimi-nation of the SOA between the sentence and responsealternatives. The SOA was included in the first experimentto ensure that participants read the sentence prior to read-ing the response alternatives. This delay was removed herein case it allowed time for episodic recall of related priorsentences. If little or no persistent facilitation was foundin Experiment 2 after removing the SOA, it would suggestthat facilitation in the first experiment could have reflectedexplicit recall or other expectancy-based processing.

Results

As in the previous experiment, responses to active-voice prime trials were faster (M = 2103 ms, SD = 635) thanthose to passive-voice primes (M = 2284 ms, SD = 728).There were also slightly fewer errors in active-voice primes(M = 3.0%, SD = 4.3) compared to passive-voice primes


(M = 3.4%, SD = 4.4). Note that average RTs were close to 1 slonger than those reported in Experiment 1. This approxi-mated the SOA that was eliminated in this experiment.

Table 2 presents target trial mean response latency forcorrect responses and percentage of errors by trial condi-tion. As in Experiment 1, a trend was evident in both targetvoices for facilitation from priming. However, primingmagnitude in response latency appeared to depend onvoice consistency for Lag 0 but not for Lag 15. There wasan average latency facilitation of approximately 400 msfor same voice and just over 200 ms facilitation for differ-ent voice priming. At Lag 15, average latency savingsapproximated 150 ms for both voice consistency condi-tions. As in Experiment 1, statistical analyses were per-formed on the correct response rate measure thatcombined error and latency variables.

Fig. 2 presents means of the rate of correct responding fortarget trials by condition. There was a large main effect fortarget voice, F1(1,88) = 131.51, p 6 .001, partial g2 = .60[response latency F1(1,88) = 118.10, p < .001, partial




M SD M S

Lag 0Unprimed 5.3 7.9 5.5Primed-Same 4.2 9.4 3.7Primed-Different 3.0 6.0 3.4

Lag 15Unprimed 4.6 6.7 5.6Primed-Same 4.6 7.9 5.3 1Primed-Different 3.2 7.4 5.1

Note. Primed-Same and Primed-Different refer to priming by trials of the same

Lag 0 Priming


Res

pons

e S

peed

(cor

rect

tria

ls/m

in)

0

22

24

26

28

30

32


Fig. 2. Mean correct response rate (correct trials/min) for target trials of Experimerepresent 95% confidence intervals for within-subject designs (Loftus & Masson

g2 = .57; percentage errors F1(1,88) = 2.20, p = .14]. Therealso was a large main effect for overall priming (unprimedversus both same- and different-voice primes),F1(1,88) = 136.52, p 6 .001, partial g2 = .61 [responselatency F1(1,88) = 95.36, MSep < .001, partial g2 = .52; per-centage errors F1(1,88) = 5.92, p = .017, partial g2 = .06]. Asin the previous experiment, there was no main effect forthe contrast of same- versus different-voice priming condi-tions, although the effect approached statistical signifi-cance, F1(1,88) = 3.31, p = .07 [response latency F1(1,88) =11.53, p = .001, partial g2 = .12; percentage errorsF1(1,88) = 2.06, p = .16]. Of primary importance were inter-active effects of Lag with the priming condition contrasts.Lag had a significant moderating effect on overall priming,F1(1,88) = 19.68, p 6 .001, partial g2 = .18 [response latencyF1(1,88) = 12.85, p = .001, partial g2 = .13; percentage errorsF1(1,88) = 2.40, p = .13], with greater facilitation at Lag 0. Lagalso interacted with the contrast of same- versus different-voice primes, F1(1,88) = 6.31, p = .014, partial g2 = .07[response latency F1(1,88) = 10.59, p = .002, partial

Experiment 2 by sentence voice, prime–target lag and priming condition.



D M SD M SD

7.3 2384 877 2692 9408.1 2043 603 2235 7416.8 2191 917 2496 1125

8.1 2407 858 2582 7881.0 2229 859 2500 10699.2 2165 736 2538 971

and different voice, respectively, compared to the target trial voice.

Lag 15 Priming


Res

pons

e S

peed

(cor

rect

tria

ls/m

in)

0

22

24

26

28

30

32


nt 2 by sentence voice, prime–target lag and priming condition. Error bars, 1994).


g2 = .11; percentage errors F1(1,88) < 1]. As can be seen inFig. 2, same-voice primes produced greater facilitation thandifferent-voice primes at Lag 0 but not at Lag 15.

Results were similar when items rather than partici-pants were analyzed. There was a large main effect for tar-get voice, F2(1,95) = 72.80, p 6 .001, partial g2 = .43[response latency F2(1,95) = 71.22, p < .001, partialg2 = .43; percentage errors F2(1,95) = 1.83, p = .18]. Therewas also a large main effect for overall priming (unprimedversus both same- and different-voice primes),F2(1,95) = 130.89, p 6 .001, partial g2 = .58 [responselatency F2(1,95) = 60.00, p < .001, partial g2 = .39; percent-age errors F2(1,95) = 7.55, p = .007, partial g2 = .07]. Therewas no main effect for the contrast of same- versus differ-ent-voice priming conditions, F2(1,95) < 1 [responselatency F2(1,95) = 2.24, p = .138, partial; percentage errorsF2(1,95) = 1.75, p = .19]. Of primary importance were inter-active effects of Lag with the priming condition contrasts.As with the analysis by participants, Lag had a significantmoderating effect on overall priming, F2(1,95) = 13.44,p 6 .001, partial g2 = .12 [response latency F2(1,95) = 9.39,p = .003, partial g2 = .09; percentage errors F2(1,95) = 1.26,p = .27], with greater facilitation at Lag 0. Lag also inter-acted with the contrast of same- versus different-voiceprimes, F2(1,95) = 4.47, p = .037, partial g2 = .05 [responselatency F2(1,95) = 4.54, p = .036, partial g2 = .05; percent-age errors F2(1,95) < 1], with same-voice primes producinggreater facilitation than different-voice primes at Lag 0 butnot at Lag 15.

An additional analysis was conducted with participantsas the object of measurement to evaluate whether themagnitude of priming was reduced from early to later trialblocks. This addressed a general issue of whether perfor-mance facilitation from prior exposure to related semanticcontent declines over trial blocks due to proactive interfer-ence or some form of semantic saturation. A further pur-pose of this analysis will become clear in Experiment 4,in which we altered the task between early and late blocks.It was important for the later experiment to know if prim-ing magnitude changed between early and late blocks inthis experiment where blocks had equivalent taskdemands.

The same within-subject contrasts were evaluated withthe inclusion of tests of the interaction with first and sec-ond halves of the trial blocks, with each half consisting ofthree blocks. The main effect of task-half was significant,F1(1,88) = 17.16, p 6 .001, partial g2 = .16 [response latencyF1(1,88) = 3.88, p = .052; percentage errors F1(1,88) =16.49, p < .001, partial g2 = .16]. Mean correct response rateslowed from 28.10 (SD = 7.32) to 26.82 (SD = 7.72) correcttrials/min across the two halves. However, of primaryimportance, neither priming contrast had a significantinteraction with task half, F1 < 1 [response latency F1 < 1for both contrasts; percentage errors F1 < 1 for overallpriming effect, F1(1,88) = 1.38, p = .24 for voice consistencyeffect in both cases]. This was true regardless of lag. There-fore, although overall correct response rate declined acrosstask halves, probably due to fatigue, neither the overallpriming effect nor the impact of prime-target voice consis-tency differed.

Discussion

When the voice of a target-trial sentence was consistentwith the voice of a semantically related prime sentence,facilitation above and beyond semantic facilitation wasevident, but only when no unrelated trial was presentedbetween them. Thus, memory representations reflectingsentence voice can contribute to facilitation in sentencecompletion, but this contribution is temporary comparedto more persistent, semantically mediated facilitation. Itis unclear from the current evidence whether the syntaxconsistency effect would survive one intervening trial,when the contents of working memory would be replacedby an unrelated trial. However, it is clear from both Exper-iments 1 and 2 that the effect cannot survive 10 or moreunrelated processing events.

Two additional observations regarding the longer-lagpriming effects are worth noting. First, the magnitude ofparticipants’ priming effects at an average lag of 15 trialswas only slightly smaller than that for priming effects seenin Experiment 1 with an average lag of 10 trials. Primed tri-als in this experiment showed an increase in correctresponse rate of approximately 7.5%, compared to theapproximately 10% rate increase reported in Experiment1. This could be attributed to any of the task differencesbetween the experiments, but the increase in number ofunrelated events separating prime and target trials is alikely contributor.

Second, the elimination of the SOA between the sen-tence and the response options, present in Experiment 1,did not eliminate the priming effects. An SOA was usedin Experiment 1 to ensure that participants fully compre-hended each sentence prior to evaluating response choices.However, it could have potentially provided time duringtarget trials, prior to the opportunity to respond, to recalla related prime trial and utilize explicit processing strate-gies. The current finding of roughly comparable primingeffects when the SOA was eliminated renders this explana-tion less likely.

Experiment 3

The first two experiments supported the conclusionthat the relatively long-lasting proposition priming effectsdid not reflect memory for comprehension processes spe-cific to sentence syntax. However, even when syntax dif-fered between prime and target trials in theseexperiments, the missing word identification processalways involved the second argument to the predicate,the sentence object. For example, a passive voice target(e.g., _________ is bought by shoppers) has the same predi-cate argument omitted as a corresponding active voiceprime (e.g., consumers purchase ________). It is possible thatmemory for a content-specific selection operation (e.g.,selecting the predicate object representing the conceptmerchandise) played a role in the observed facilitationeffects.

Experiment 3 varied orthogonally in prime and targettrials the missing proposition argument that was to be


identified (i.e., subject versus object). If memory processesunderlying the observed semantic priming effects repre-sent only the meaning of the entire proposition, then itshould not matter if the proposition argument completedin the target sentence corresponds with or is different fromthe argument completed in the prime sentence. Alterna-tively, if the priming effects reflect, in part or whole, mem-ory for prior selection and decision operations, then themagnitude of facilitation should depend on the consistencyof argument identification across prime and target trials.

Method



MaterialsThe stimulus set from the previous experiments was

modified such that foils were created for the original sub-ject terms. The foils were generated in a manner similar tothose for the predicate objects. For example, the proposi-tion represented by sentences mules carry bundles and bur-ros haul packs had foils for the sentences’ objects as before(inspirations and motivations), and now it also had foils forthe sentences’ subjects (corpses and cadavers). This allowedeach proposition to be presented with either the subject orobject missing, and this could be varied across prime andtarget sentences.

ProcedureThe structure of this experiment resembled that of

Experiment 1. There were 12 blocks of 20 trials. Each blockbegan with four warm-up filler trials, followed by fourprime trials. Four filler trials separated the prime trialsfrom subsequent target trials. There were eight target tri-als, four primed and four unprimed. Thus, the average lagbetween prime and target propositions was 10 unrelatedtrials. As in Experiment 2, there was no SOA between prop-osition and response options. All other task details wereidentical to Experiment 1 except the manipulation of sen-tence-completion format to be described next.

In this experiment, all propositions were expressed inthe active voice. However, half the prime trials requiredparticipants to select one of two options to complete theomitted sentence subject, and half required participants



Subject completion Object completion

M SD M SD

Unprimed 5.8 6.1 4.7 5.3Primed-Same 4.0 5.8 4.0 6.7Primed-Different 5.2 8.2 4.0 5.2

Note. Primed-Same and Primed-Different refer to priming by trials that requircompared to the target trial voice.

to complete the omitted predicate object. This was crossedwith the same manipulation in target trials.

Results

In prime trials, the mean response latency for the sub-ject-completion format (M = 2062 ms, SD = 582) did notdiffer from that for the object-completion format(M = 2074 ms, SD = 527), F1(1,88) < 1. Percentage errorwas also equivalent for subject-completion (M = 3.1%,SD = 4.6) and object-completion prime trials (M = 3.6%,SD = 4.4), F1(1,88) = 1.51, p = .22.

Table 3 presents mean response latency for correctresponses and percentage of errors for target trials by trialcondition. As in the previous experiments, there were cleartrends for primed trials to have shorter latency and fewererrors compared to unprimed trials. Possible differencesbetween primed trials with the same versus different sen-tence-completion format were less evident.

Fig. 3 presents means of the rate of correct respondingfor target trials by condition. There was a small main effectfor the sentence-completion format in target trials,F1(1,88) = 4.27, p = .04, partial g2 = .05 [response latencyF1(1,88) = 4.06, p = .047, partial g2 = .04; percentage errorsF1(1,88) = 3.89, p = .052], with completion of the predicateobject being slightly faster compared to completion of thesubject. As in previous experiments, there was a large maineffect for the contrast of primed and unprimed trials,F1(1,88) = 78.42, p 6 .001, partial g2 = .47 [response latencyF1(1,88) = 38.47, p < .001, partial g2 = .30; percentageerrors F1(1,88) = 4.52, p = .036, partial g2 = .05]. However,as is evident in Fig. 3, there was no difference between tar-get trials primed with the same versus different sentence-completion format, F1(1,88) < 1 [response latencyF1(1,88) < 1; percentage errors F1(1,88) < 1], and theBayesian posterior probability associated with the nullhypothesis was estimated to be PH0|D = .89. Neither ofthe priming contrasts interacted with target sentence for-mat, F1 < 1 in both cases.

When items were the unit of analysis, there was nomain effect for the sentence-completion format,F2(1,95) < 1 [response latency F2(1,95) < 1; percentageerrors F2(1,95) = 1.84, p = .18]. There was a substantialmain effect of primed versus unprimed trials,F2(1,95) = 29.42, p 6 .001, partial g2 = .24 [response latencyF2(1,95) = 16.22, p < .001, partial g2 = .15; percentageerrors F2(1,95) = 2.92, p = .090]. However, as in the analysisby participants, there was no priming difference between

Experiment 3 by sentence completion type and priming condition.


Subject completion Object completion

M SD M SD

2397 725 2427 6622248 609 2233 6382205 594 2294 740

ed the completion of the same or different sentence term, respectively,

Sentence Completion Type

Subject Object

Res

pons

e S

peed

(cor

rect

tria

ls/m

in)

0

22

24

26

28

30

32UnprimedPrimed: Subject CompletionPrimed: Object Completion

Fig. 3. Mean correct response rate (correct trials/min) for target trials ofExperiment 3 by sentence completion type and priming condition. Errorbars represent 95% confidence intervals for within-subject designs (Loftus& Masson, 1994).


target items primed with the same versus different sen-tence-completion format, F2(1,95) < 1 [response latencyF2(1,95) < 1; percentage errors F2(1,95) = 1.91, p = .17].

Discussion

The current finding suggests that memory for priordecision processes in sentence completion plays little orno role in the observed proposition priming effects. Therewas substantial and equivalent facilitation in target trialswith both subject- and object-completion format, and inneither case did the magnitude of facilitation differ as afunction of prime trial format. This outcome is consistentwith the view that persistent semantic priming effects inthis task reflect strength changes to abstract memory forthe shared meaning of prime and target sentences.

Experiment 4

The previous experiments supported a conclusion thatthe sentence-completion priming effects were not attribut-able to memory for processing operations in that they wereunaffected by changes in sentence syntax and the missingproposition argument. This experiment addressed twoadditional questions. First, are these priming effects dura-ble enough to persist over a span of at least 15 min ratherthan 10–15 intervening trials? Whether semantic primingof propositions representing prior knowledge survives overa longer time frame has implications for distributed net-work models attempting to account for the phenomenathrough strength increments and subsequent decay inthe network. To test this issue, we inserted a filler taskbetween prime and target trials that contained differentperformance goals. Second, are facilitation effects in target

trials associated with conscious recognition of the relatedpriming events, such that the facilitation might representthe benefit of deliberate, rather than implicit, memory pro-cesses? To test this issue, some of the target trials requiredparticipants, after responding with the missing word, toindicate whether the sentence content was similar inmeaning to that of a trial seen earlier.

There are obvious obstacles to accurately determiningthe role of recognition memory in priming tasks thatdemand the processing complexity of sentence comple-tion. Readers may lose the ability to recognize the surfaceform of sentences with sufficient time, but their ability torecall proposition content has been demonstrated overseveral months (Kintsch & van Dijk, 1978). Therefore, rec-ognition accuracy for the meaning of previously con-structed propositions under the current task conditionswas expected to be well above chance. The challenge wasto determine whether above-chance recognition of previ-ously constructed proposition meaning could be responsi-ble for the observed facilitation effects.

A primary difficulty in evaluating the role of explicitmemory in this task is that facilitation in primed trialscan be associated with accurate recognition judgmentsfor two reasons. Facilitation can reflect processing benefitsof recognizing previous content, or accurate recognitioncan be based on the perceived fluency of processing duringthe target event. Jacoby and colleagues (Jacoby & Dallas,1981; Jacoby & Whitehouse, 1989) provided evidence thatold-new recognition judgments were influenced by the flu-ency with which the trials were performed. Although per-formance-fluency effects on recognition are oftenassociated with perceptual processes, there is evidencethat these effects extend to conceptual processes (Luo,1993; Masson & MacLeod, 1997). It is conceivable, there-fore, that participants’ perceived fluency in performingprimed compared to unprimed target trials increases theiraccuracy in making old-new judgments.

Acknowledging this issue, we evaluated the possiblerole of recognition memory in target sentence facilitationin several ways. First, we tested whether sentence-comple-tion rate of correct responding changed as a function ofrequiring participants to also produce recognition judg-ments of the semantic content. We reasoned that if recog-nition memory processes were instrumental in theobserved performance facilitation effects, then partici-pants’ anticipation of a subsequent old-new decisionshould not reduce the magnitude of priming; indeed, itmight be expected to increase it by adding an explicit goalto remember the previous trials. If, instead, the previouslyobserved facilitation effects only reflect automatic recruit-ment of abstract semantic memory for the propositions,then the additional goal of episodic recognition, which pre-sumably demands slower, attention-demanding processes,should reduce the magnitude of priming.

Second, we tested whether the magnitude of primingwas a function of accurate recognition of prior semanticcontent. If recognition underlies the priming effects, thenprimed but not unprimed trial performance should be bet-ter when there is perceived familiarity. On the other hand,if both primed and unprimed trial performance levels arebetter for trials perceived as familiar, then it is likely that


recognition judgments are influenced by sentence-comple-tion performance rather than the reverse.

Method



ProcedureThe materials and task were similar to those of Experi-

ment 2 which contained only predicate-object comple-tions, but with the following differences. First, ratherthan contrasting prime-target lags of 0 and 15, the primesand targets were presented in different trial blocks sepa-rated by an unrelated filler task. This tested the persistenceof the priming effects over a much longer lag that wasfilled with different task goals and content. Second, therewere both active- and passive-voice target trials, and eachprimed target trial was presented in a different voice fromthe corresponding prime trial. Consistency of voice acrossprime and target trials had no effect on longer-term facili-tation in previous experiments, so this element no longerseemed important to manipulate. Different voice target tri-als were used exclusively to eliminate the slight possibilitythat trial recognition decisions would partly reflect surfacerather than semantic features of previous trials. Third, halfof the target trials required participants to judge whetherthe trial content was semantically related to an earliertrial.

Following the same task instructions presented inExperiments 1–3, the first three trial blocks of the sessionpresented prime trials. Each block began with fourwarm-up trials followed by 16 prime trials. Across thethree blocks, a total of 48 prime sentences were presented,half of which were in each voice. Participants were notinformed that they would later be tested for recognitionof trial content.

Following the prime blocks, participants performed afiller task that required the identification of words fromscrambled letters. Each trial presented a string of letterscentered in the upper third of the computer display (e.g.,ducer). Below these letters were two words flanked to theleft and right (ducks. . .crude). Participants were instructedto press either a left or right key (z or /) to indicate whichword was the correct unscrambling of the letters. Incorrectalternatives (e.g., ducks) were words that differed from theletter string by a single letter. There were four blocks of 24trials, with average RT and percentage correct presented tothe participant at the conclusion of each block.

Following the scrambled word task, participants wereinstructed to perform another set of sentence-completiontrials similar to those performed at the beginning of thesession. No mention was made regarding the similarity ofsome sentences to those in earlier trials. There were threesuch trial blocks that will be referred to as completion-onlytrials. Each block began with four warm-up trials, followedby eight primed and eight unprimed target trials in ran-dom order. Half the trials in each prime condition were

presented in passive voice and half were in active voice,and primed targets were always presented in a differentvoice from the corresponding prime trial.

Following the completion-only target blocks, partici-pants were instructed that an additional element wouldbe added to the task. After each sentence-completionresponse, they were to indicate whether they had solvedan item with similar meaning earlier in the session. Anexample was provided (TV stations broadcast __________;advertisements or dirty clothes). Accompanying this exam-ple was an item with similar meaning that they had seenearlier during initial task instructions (Networksair_______; commercials or laundry). They were instructedto respond that they had performed a similar item in casessuch as this.

Three blocks of these trials were presented and will bereferred to as completion-plus-recognition trials. Other thanthe recognition component, the structure of these trialswas identical to the completion-only trials. As in all previ-ous sentence-completion blocks, accuracy and RT feedbackwas provided at the conclusion of each block. No feedbackwas provided on recognition performance.

It should be noted that the confound between targettrial condition (completion-only and completion-plus-rec-ognition) and task order was intentional for two reasons.First, if recognition probes were included in the initial tar-get trials, this could have biased participants to focusattention on explicit memory for previous trials in subse-quent completion-only trials. It is likely that after threeblocks of performing sentence-completion trials and indi-cating whether they were related in meaning to earlier tri-als, participants would continue to think aboutrelationships between completion-only trial content andearlier trials. Second, prior evidence of the time course oflong-term semantic priming effects over intervening trialssuggests a power function rate of decline (Woltz, 2010). Ifthis form of forgetting holds in the current task over longerdelays produced by an intervening task, then loss of prim-ing would be minimal across the two target trial conditionsof this experiment. Consistent with this, the analysisreported in Experiment 2 of priming effects over first andsecond halves of the target blocks showed no reliable dif-ference. Experiment 2 was singled out for this analysisbecause its trial conditions matched those of the currentexperiment most closely.

There were eight target trial conditions created by thecrossing of three 2-level task facets (primed versusunprimed, completion-only versus completion-plus-recog-nition, and active versus passive target voice). Assignmentof stimulus sets to these conditions was counterbalancedacross participants such that each stimulus was presentedwith equal frequency in each condition.

Results

Response data for prime trials were similar to thosereported for Experiments 2 and 3. Responses were fasterfor prime trials in active voice (M = 2083 ms, SD = 542)than in passive voice (M = 2290 ms, SD = 709). Also, therewere slightly fewer errors in active-voice primes(M = 3.1%, SD = 3.8) compared to passive-voice primes


(M = 3.9%, SD = 4.2). The scrambled word task that sepa-rated the end of the prime trials from the beginning ofthe target trials took participants an average of approxi-mately 14 min (M = 13.74 min, SD = 1.63).

Table 4 presents mean RTs for correct responses andpercentage of errors by trial condition for target trials.Clear trends for facilitation from priming were evident inresponse latency only for one set of conditions. As seenin Table 4 for completion-only target trials, there was anoverall average priming effect of approximately 100 mscompared to little or no consistent priming for comple-tion-plus-recognition trials. Evidence for facilitation inerror data was more variable. As before, statistical analyseswere performed on the computed variable, rate of correctresponding.

Recognition dataMean accuracy of recognizing old versus new semantic

content of propositions in the final three trial blocks waswell above chance. For passive-voice target trials, meanold-new classification accuracy was 71.9% (SD = 25.0) forprimed and 73.4% (SD = 21.2) for unprimed (meand0 = 1.33, SD = .83). For active-voice targets, mean accuracywas 74.1% (SD = 23.2) for primed and 73.1% (SD = 19.7) forunprimed (mean d0 = 1.42, SD = .76). Differences due to tar-get voice or primed versus unprimed conditions were notstatistically significant, F1 < 1 in both cases.

Correct response rate dataFig. 4 presents the means for rate of correct responding

to the sentence-completion component of target trials bycondition. There was a large main effect for completion-only versus completion-plus-recognition, F1(1,112) =138.62, p < .001, partial g2 = .55 [response latencyF1(1,112) = 144.49, p < .001, partial g2 = .56; percentageerrors F1(1,112) < 1]. Participants were slower to completethe sentences when they knew that a recognition probewould follow. Consistent with previous experiments, therewas also a large main effect of target voice,F1(1,112) = 119.59, p < .001, partial g2 = .52 [responselatency F1(1,112) = 79.92, p < .001, partial g2 = .42; per-centage errors F1(1,112) = 4.90, p = .029, partial g2 = .03].Also, even with the longer prime-target lag compared toprevious experiments, there was a main effect for priming,F1(1,112) = 25.55, p < .001, partial g2 = .19 [responselatency F1(1,112) = 4.89, p = .029, partial g2 = .04; percentageerrors F1(1,112) = 1.73, p = .19]. Of primary importance,

Table 4Mean response time for correct responses and percentage error in target trials of E



M SD M SD

Completion-onlyUnprimed 5.2 7.8 6.6 9.5Primed 5.4 8.4 4.9 7.1

Completion-plus-recognitionUnprimed 4.7 7.6 5.7 7.6Primed 4.1 7.5 5.6 7.7

this priming effect was moderated by the demand for arecognition judgment following performance, F1(1,112) =8.78, p = .004, partial g2 = .07 [response latencyF1(1,112) = 4.57, p = .035, partial g2 = .04; percentageerrors F1(1,112) < 1]. As is evident in Fig. 4, there was apriming effect in both target voice conditions when partic-ipants only performed the sentence completions, but therewas little or no priming effect in both completion-plus-recognition voice conditions as suggested by the substan-tial overlap in confidence intervals. In a post hoc analysis,the main effect for priming in completion-plus-recognitiontrials was not statistically significant, F1(1,112) = 1.74,p = .19 [response latency F1(1,112) < 1; percentage errorsF1(1,112) < 1].

The analysis conducted on correct response rate meansfor items rather than participants yielded similar results.There was a large main effect for completion-onlyversus completion-plus-recognition, F2(1,95) = 292.39.62,p < .001, partial g2 = .76 [response latency F2(1,95) =281.45, p < .001, partial g2 = .75; percentage errorsF2(1,95) = 2.06, p = .16], with substantially slower meancorrect response rate when the recognition probe wouldfollow. There was a large main effect of target voice,F2(1,95) = 107.86, p < .001, partial g2 = .53 [responselatency F2(1,95) = 90.85, p < .001, partial g2 = .49; percent-age errors F2(1,95) = 4.57, p = .035, partial g2 = .05], and asmall main effect for priming, F2(1,95) = 6.34, p < .013, par-tial g2 = .06 [response latency F2(1,95) = 2.93, p = .090; per-centage errors F2(1,95) = 1.64, p = .203]. Again, and ofprimary importance, the priming effect was moderatedby the demand for a recognition judgment following per-formance, F2(1,95) = 10.02, p = .002, partial g2 = .10[response latency F2(1,95) = 4.41, p = .038, partial g2 = .04;percentage errors F2(1,95) < 1]. There was a significantpriming effect in the completion-only trials,F2(1,95) = 13.31, p < .001, partial g2 = .12 [response latencyF2(1,95) = 13.20, p < .001, partial g2 = .12; percentageerrors F2(1,95) = 1.24, p = .269], but not in completion-plus-recognition trials, F2(1,95) < 1 [response latencyF2(1,95) < 1; percentage errors F2(1,95) < 1].

The final test of whether priming effects could be attrib-uted to explicit recognition was performed with partici-pants as the unit of observation. This analysis evaluatedpriming effects broken down by recognition response. Ifexplicit recognition of prior content was instrumental inthe facilitation effects seen in sentence-completion perfor-mance, there should be a larger priming effect when par-

xperiment 4 by sentence voice, recognition demand and priming condition.



M SD M SD

2232 674 2512 9702155 680 2392 855

2600 738 2984 10482649 904 2945 1148

Completion-only (no recognition)

Target Trial Voice

Res

pons

e S

peed

(cor

rect

tria

ls/m

in)

0

20

22

24

26

28

30UnprimedPrimed

Completion-plus-recognition

Target Trial VoiceActive Passive Active Passive

Res

pons

e S

peed

(cor

rect

tria

ls/m

in)

0

20

22

24

26

28

30

UnprimedPrimed

Fig. 4. Mean correct response rate (correct trials/min) for target trials of Experiment 4 by sentence voice, recognition demand, and priming condition. Errorbars represent 95% confidence intervals for within-subject designs (Loftus & Masson, 1994).


ticipants correctly identified a trial as old content. Table 5presents mean RTs for correct responses and percentage oferrors for completion-plus-recognition target trials by rec-ognition response. The sample size was greatly reduced forthis analysis (n = 64), because many participants had miss-ing data for trial conditions broken down by recognitionresponse (a participant could not be included if they hadeither all or no recognition errors for any trial condition).As seen in Table 5, there were no consistent trends forfacilitation when participants indicated the trial contentwas similar to that of a previously seen trial. Of equalimportance, there was a strong trend for lower mean RTand fewer errors on trials that were perceived as old con-tent, regardless of whether they were primed (old) orunprimed (new). This trend was relatively consistent forboth active- and passive-voice trials.

The analysis of rate of correct responding revealed amain effect of voice, F1(1,56) = 7.55, p = .008, partialg2 = .12 [response latency F1(1,56) = 19.24, p < .001, partialg2 = .26; percentage errors F1(1,56) < 1]. The priming effectapproached statistical significance, F1(1,56) = 3.65, p = .061[response latency F1(1,56) = 2.59, p = .11; percentageerrors F1(1,56) < 1]; consistent with trends seen in the RTand error data of Table 5, performance on unprimed trialswas slightly better than on primed trials. The two most

Table 5Mean response time for correct responses and percentage error in completion-plusvoice, and priming condition (N = 64).



M SD M SD

Identified as oldUnprimed 3.8 10.5 3.8 12.1Primed 2.4 5.4 3.7 6.4

Identified as newUnprimed 6.6 11.2 7.2 9.7Primed 9.4 16.1 9.1 16.1

important findings were (1) a relatively large differencein rate of correct responding when participants identifiedthe trial as old (M = 23.13, SD = 6.69) versus new(M = 20.32, SD = 6.74), F1(1,56) = 28.48, p < .001, partialg2 = .34 [response latency F1(1,56) = 18.17, p < .001, partialg2 = .25; percentage errors F1(1,56) = 13.00, p = .001, par-tial g2 = .19], and (2) no interaction between this effectand the primed versus unprimed manipulation,F1(1,56) < 1 [response latency F1(1,56) < 1; percentageerrors F1(1,56) = 3.22, p = .078].

Discussion

Facilitation effects were smaller than those seen in theprevious experiments, even for the completion-only condi-tion, which was most similar in task design. This reductionin magnitude is likely attributable to the much longerdelay between prime and target trials. The previous exper-iments used average lags of 10–15 trials, whichapproached a 1 min delay for the typical participant. Incontrast, Experiment 4’s minimum lag for any prime-targetpair was approximately 14 min (i.e., the minimum lagwould occur when by chance the final prime trial was asso-ciated with the first target trial), and for many primed tri-als the lag was 30 min or more. Despite this increased lag

-recognition target trials of Experiment 4 by recognition response, sentence



M SD M SD

2642 801 2863 11062734 917 2850 847

2787 855 3216 12873029 1578 3318 1379


time, there was still statistically significant facilitation inthe completion-only trials.

Several findings from this experiment suggest that theobserved facilitation does not depend on prime trial recol-lection and may be disrupted by it. First, a priming effectwas observed when there was no subsequent old-new rec-ognition judgment required, but there was little or nopriming effect when the recognition judgment wasrequired following each sentence completion. The comple-tion-plus-recognition trial blocks came after the comple-tion-only blocks in order to avoid a carry-over effect oftrying to recall earlier related trials. Given the length ofdelay between prime and target trials, and a presumedpower function form of forgetting, it is unlikely that theslight increase in delay between conditions could accountfor this difference. Nevertheless, the priming differencecould conceivably reflect proactive interference of the firstthree blocks on the later blocks. However, an analysis ofthe priming difference between first and second halves ofthe target blocks in Experiment 2, which was most similarto the current experiment in trial format, revealed no dif-ference. If proactive interference were operating, it wouldbe expected in that experiment to the same extent. It ismore plausible that the lack of priming in completion-plus-recognition trials is attributable to participants’ atten-tion being focused on recalling previous trial content. Thus,rather than being responsible for primed trial facilitation, itis likely that recognition memory processes interfered withthe implicit semantic memory processes responsible forfacilitation.

Second, there was clear evidence that participants rec-ognized some target trials as related to previous prime tri-als, but there was no evidence that perceiving a trial asfamiliar facilitated primed trials exclusively, as would beexpected if recognition was instrumental in sentence-com-pletion priming effects. Instead, better sentence-comple-tion performance was associated with perceivedfamiliarity with the content regardless of whether the trialcontent had been seen before or not. This suggests that rec-ognition judgments were influenced by performance flu-ency rather than the reverse, which is consistent withresearch using other tasks (Jacoby & Dallas, 1981; Jacoby& Whitehouse, 1989; Luo, 1993; Masson & MacLeod,1997).

1 We thank Michael Masson for pointing out this possible influence.

Experiment 5

Although the evidence suggests that participants do notrely on recognition memory to respond more quickly toprimed targets, it is possible that yet another form of mem-ory other than semantic memory for propositions contrib-utes to the facilitation. In all prior experiments, theincorrect alternative (foil) in each primed target was a syn-onym of the foil from the prime trial. This will be referredto as the standard foil target trial configuration. For exam-ple, mules carry bundles/inspirations was used to prime bur-ros haul packs/motivations. In trials of this format,facilitation in selecting the correct alternative could reflect,in part or whole, a form of inhibition or negative primingfor the semantic content of the rejected alternative rather

than positive priming for the proposition with its correctalternative. That is, the choice of packs may be facilitatedbecause a synonym (bundles) was a positive choice in aprevious trial, and motivations was a synonym of a previ-ously rejected choice (inspirations). This explanation makesno reference to memory for proposition meaning, onlymemory for the semantic content of response alternativeselection and rejection.1

This explanation was tested by adding a new conditionto target trials that replaced the standard foil term withone that was a synonym of the positive response from adifferent prime trial. This target trial configuration willbe referred to as the positive foil condition. In this condi-tion, both alternatives were synonyms of words represent-ing a positive response in prime trials, but only onecompleted the target sentence correctly. If memory forprevious response selection plays a role in the observedfacilitation, then priming effects should be eliminated orreduced in the positive foil condition compared to thestandard foil condition.

To create the new positive foil condition, each stimulusset (a proposition plus correct and incorrect alternatives)was paired with other stimulus sets such that the correctalternative of one could serve as an appropriate foil forthe other. The previous example mules carry bundles/inspi-rations and burros haul packs/motivations was paired withstimulus sets such as vaccinations prevent epidemics/insur-gents and immunizations deter outbreaks/terrorists. Whenthe burros sentence and its pairing were assigned to thepositive foil condition, the target trial would be burros haulpacks/outbreaks (rather than the standard foil target of bur-ros haul packs/motivation). Note that this target trial wouldhave two prime trials: mules carry bundles/inspirations andvaccinations prevent epidemics/insurgents, ensuring thatboth alternatives in the target trial (packs, outbreaks) weresemantically related to prior prime trial positive responses(bundles and epidemics). The positive foil prime trial (e.g.,vaccinations prevent epidemics/insurgents) would only serveto prime the foil in the new target trial condition; its cor-responding target stimulus (immunizations deter out-breaks/terrorists) would not be presented as a target trial.

Method



MaterialsFour additional trial stimuli were added to the 96 prop-

osition pairs used in previous experiments so that thestimuli could be divided into five groups of 20 for counter-balancing. The 20 proposition sets within the five groupswere ordered such that each set had a correct alternativethat could serve as the incorrect alternative for any corre-spondingly ordered set from the other groups (i.e., they


Res

pons

e S

peed

(cor

rect

tria

ls/m

in)

0

22

24

26

28

30

32UnprimedPrimed - Standard Foil Primed - Positive Foil



were appropriate incorrect alternatives because they werenot a plausible correct completion of any of the corre-spondingly ordered sets).

ProcedureThe task was similar to those of Experiment 4 which

contained only predicate object completions and intro-duced a minimum 14 min lag between primes and targettrials. However, this experiment had no recognition trialsand included a target trial condition with positive foils.

There were five blocks of prime trials, each with 16 tri-als: four warm-up trials and 12 primes. Of the 12 prime tri-als per block, four were proposition primes for thestandard foil target trials, four were proposition primesfor the positive foil target trials, and four were primes thatwould not prime a semantically related target trial butwhose correct response would serve to prime the foil in apositive foil trial. Half the trials of each type were pre-sented in active voice and half were presented in passivevoice.

There were five blocks of target trials, each with 20 tri-als: four warm-up trials followed by 16 target trials. Half ofthe 16 target trials per block were unprimed and half wereprimed. Of the eight primed trials, half had the standardfoil configuration and half had the positive foil configura-tion. The former condition was identical to the foil proce-dure in all previous experiments. The voice of eachprimed target trial, regardless of condition, differed fromthat of its proposition prime trial. As in Experiment 4, thetarget trials were separated from the prime trials by thescrambled word filler task.

The five stimulus groups were assigned to trial condi-tions in a counterbalanced fashion across participants suchthat each stimulus occurred with equal frequency in eachcondition. For each participant, two of the groups wereassigned to the unprimed condition, one was assigned tothe standard foil primed condition, one was assigned tothe positive foil primed condition, and one was assignedto the prime trials that provided foils for the positive foilcondition.

Results

As in previous experiments, responses were faster forprime trials in active voice (M = 1980 ms, SD = 476) thanin passive voice (M = 2240 ms, SD = 655). There wereroughly equivalent errors in active-voice primes




M SD M S

Unprimed 6.3 6.9 7.2 6Primed Std. Foil 5.1 7.7 6.4 9Primed Pos. Foil 4.8 6.7 6.4 8

Note. Primed Std. Foil (standard foil) trial type refers to target trials that had sentPrimed Pos. Foil (positive foil) trial type refers to target trials that had sentenceshad a foil that had been the correct response option for an unrelated prime tria

(M = 2.2%, SD = 3.5) and passive-voice primes (M = 2.1%,SD = 2.8). The scrambled word task took participantsapproximately 14 min (M = 13.77 min, SD = 1.51).

Table 6 presents mean response latency for correctresponses and percentage of errors for target trials by trialcondition. As in previous experiments, there were cleartrends for primed trials to have shorter RT and fewer errorscompared to unprimed trials. Differences between the twoprimed trial conditions were less clear.

Fig. 5 presents means of the rate of correct respondingfor target trials by condition. There was a main effect forthe voice of target trials, F1(1,80) = 34.63, p < .001, partialg2 = .30 [response latency F1(1,80) = 25.73, p < .001, partialg2 = .24; percentage errors F1(1,80) = 4.00, p = .049, partialg2 = .05], with correct response rate being faster inactive-voice compared to passive-voice trials. As in previ-ous experiments, there was a relatively large main effectof both primed conditions compared to unprimed trials,F1(1,80) = 24.15, p 6 .001, partial g2 = .23 [response latencyF1(1,80) = 15.10, p < .001, partial g2 = .16; percentageerrors F1(1,80) = 4.19, p = .044, partial g2 = .05]. The overall

Experiment 5 by sentence voice and priming condition.



D M SD M SD

.7 2219 624 2434 764

.0 2158 657 2337 770

.3 2183 726 2328 759

ences and both response options semantically related to their prime trial.and correct response option semantically related to their prime trial but

l.


priming effect magnitude was somewhat smaller thanwhat was found in Experiments 1–3 that had shorter lags,but it was similar to that found in Experiment 4 that con-tained the prime-target lag used in this experiment. Of pri-mary importance, as is evident in Fig. 5, there was noappreciable difference between target trials containingthe two different foil configurations, F1 < 1 [responselatency F1(1,80) < 1; percentage errors F1(1,80) < 1], withthe Bayesian posterior probability associated with the nullhypothesis estimated to be PH0|D = .90. Neither of thepriming contrasts interacted with target voice, F1 < 1 inboth cases.

The analysis by items rather than participants producedsimilar results. Correct response rate was faster for active-compared to passive-voice presentation, F2(1,99) = 38.92,p < .001, partial g2 = .28 [response latency F2(1,99) =31.80, p < .001, partial g2 = .24; percentage errors F2(1,99)= 4.90, p = .029, partial g2 = .05]. There was a main effectof both primed conditions compared to unprimed trials,F2(1,99) = 12.31, p 6 .001, partial g2 = .11 [response latencyF2(1,99) = 5.23, p < .024, partial g2 = .05; percentage errorsF2(1,99) = 3.46, p = .066]. As when participants were theunit of analysis, there was no significant differencebetween target trials containing the two foil configura-tions, F2(1,99) < 1 [response latency F2(1,99) < 1; percent-age errors F2(1,99) < 1], and neither of the primingcontrasts interacted with target voice, F2 < 1 in both cases.

Discussion

This experiment evaluated the potential role of memoryfor semantic content of response selections in the primingeffects observed in the previous experiments. In particular,priming effects seen in the previous experiments couldhave been enhanced by negative priming of the foils (i.e.,semantic content of target foils had been rejected in corre-sponding prime trials). No evidence was found to supportthis alternative explanation, and the most viable explana-tion across all five experiments is that facilitation inprimed target trials stems from an increment in strengthto abstract semantic representations of sentence content.

Experiment 6

Our intention in all experiments was to investigatesemantic priming of memory representations for proposi-tion meaning. However, the long-term priming effectsobserved could reflect word-level semantic priming. Thatis, facilitation in identifying a missing word in the sentenceburros haul packs due to prior processing of mules carrybundles could reflect the priming of semantically relatedindividual words, not the shared proposition. Activatingthe meaning of burros could be more efficient becausethe meaning of mules was activated earlier. Similarly, hauland packs could be processed more quickly because of pre-vious processing of carry and bundles.

A word-level priming explanation of persistent sen-tence-completion facilitation seems unlikely if one refersto the small and short-lived semantic priming effects seenin lexical decision and word naming tasks. However, Woltz

(2010) found that semantic priming effects in individualword meaning comparisons approximated a 7% increasein correct response rate with 14 unrelated trials betweenprimes and targets. Furthermore, priming magnitudeshowed a negatively accelerated decline over lags from 1to 14 trials, suggesting the likelihood of longer-lastingfacilitation.

Experiment 6 evaluated whether priming of completepropositions produces additional facilitation to robustindividual word priming. The target trials were equivalentto those used in previous experiments, and they were par-ticularly close in form to those of Experiments 2 in whichthere were either active- or passive-voice expressions withno SOA between the sentence and the response alterna-tives. New to this experiment, there were two distincttypes of prime trials. Half of the primed target trials wereprimed by an active–voice, semantically related sentence.This corresponds exactly to a subset of priming conditionsof previous experiments. The other half were primed bythree separate word-meaning comparison trials that con-tained words semantically related to those in the targetsentence. Each word-meaning comparison presented theprime word that would have appeared in the propositionprime, along with a synonym and unrelated foil. Ratherthan select a word to complete a sentence as in the propo-sition prime trials, participants selected a word thatmatched another word in meaning. Because the threeword-prime trials for a target sentence were never pre-sented contiguously, we assumed they primed only indi-vidual word meanings and not the propositions.

The goal of this approach was to evaluate whether per-sistent semantic priming of individual words could fullyaccount for the sentence-completion priming seen in theprevious experiments. This does not assess the degree towhich prime sentences primed individual words, proposi-tions, or both. Arguably, our method produces strongerword level priming than exposure to a proposition expres-sion does because the former presents two synonyms ofeach target word and demands an evaluation of the word’smeaning; the latter prime condition presents only one syn-onym of each target word and demands an evaluation ofthe proposition’s meaning. We view our approach as astrong test of the hypothesis that there is propositionpriming beyond word priming. Under the presumed strongpriming of individual words, if proposition priming effectswere larger than word priming effects, this would supporta conclusion that at least some semantic priming occurs atthe proposition level of meaning.

Method



MaterialsIn order to create word-meaning comparison prime tri-

als, stimuli used in previous experiments were modifiedsuch that each word in prime propositions now had a syn-


onym that was not used in the target proposition. Forexample, the sentence mules carry bundles, which was usedto prime burros haul packs, had three new associated syn-onyms donkey, to lug, and bags. Because some verbs couldbe viewed as nouns (e.g., lug as a synonym for carry), allword-meaning comparisons involving verbs were pre-sented in the infinitive form (e.g., to lug and to carry). Eachword-meaning comparison prime trial also required a foil.Foils for prime sentence subjects and objects were alreadycreated for previous experiments. New foils were gener-ated for sentence verbs (e.g., to munch was used as the foilfor to carry). Note that the synonyms and foils were neverwords that appeared in target sentences. To create thisstimulus pool, some original proposition pairs from previ-ous experiments were replaced, and the final number usedwas 90 rather than 96.

ProcedureThere were five blocks of prime trials that contained

two different formats and content. Participants were pro-vided with initial instructions and practice on each trialtype before the priming blocks. Priming blocks 2 and 4contained proposition primes like those used in previousexperiments (e.g., mules carry ______, with bundles andinspirations as the response options). There were 17 trialsin each of these two blocks, consisting of two warm-up fil-ler trials and 15 primes, yielding a total of 30 propositionsprimed for later target blocks. Priming blocks 1, 3, and 5contained 2-choice word-meaning comparison trials. Foreach eventual target sentence assigned to this primingcondition, there were three meaning-comparison trials,one for the subject, verb, and object of what would bethe prime sentence if the entire proposition was primed.The three word-comparison trials for any given proposi-tion always occurred in different blocks, so there was nochance they could appear in sequence. One third of the tri-als in each priming block contained subject, verb, andobject primes, respectively, in a random order. Followingtwo warm-up filler trials, there were 30 trials per block.As a result, after the three word-comparison blocks, eachindividual word of 30 target propositions was primed bya word-meaning comparison.

The format of the word-comparison prime trials resem-bled that of the proposition prime trials in most respects.Each word-comparison trial began with a 1 s orientingcue consisting of asterisks where the first word wouldappear. Following a 500 ms blank screen, one of the stim-ulus words from a priming proposition appeared centeredin the top half of the display in the same position that sen-tences appeared in the other priming condition. This wordremained visible for 1 s before two alternatives appearedin the lower half of the display in the same positions wheresentence completion alternatives had appeared in theother condition. This delay was imposed to ensure thatthe meaning of each stimulus word from the propositionwas fully activated before the alternatives appeared (seeSabol & DeRosa, 1976). All three words remained visibleuntil a response key was pressed. As in the propositionpriming condition, participants were instructed to pressthe T or B key, indicating whether the synonym of the ini-tial word was the top or bottom alternative. The position of

the correct response was randomly determined on eachtrial. A 1500 ms blank screen separated each response fromthe beginning of the next trial.

Following the five prime blocks, participants performedfour blocks of the scrambled word task. This task was iden-tical to that described for Experiments 4 and 5.

The intervening task was followed by five blocks of tar-get trials. All target trials were sentence completions, as inprevious experiments. Each block began with four warm-up filler trials followed by 18 target trials of three typesthat were randomly ordered: six unprimed propositions,six propositions primed by three word-meaning compari-sons, and six propositions primed by a single sentencecompletion. Assignment of stimuli to trial type was coun-terbalanced across participants so that each of the 90 tar-get propositions appeared with equal frequency in eachtrial condition.

Results

Prime trials, which always presented active-voice sen-tences for completion, had similar mean RT for correctresponses (M = 2082 ms, SD = 553) and percentage oferrors (M = 3.1%, SD = 3.2) compared to this condition inprevious experiments. For prime trials that presentedword-meaning comparisons, mean error rate was roughlycomparable to that of the sentence primes (M = 3.9%,SD = 2.7). However, mean RT was shorter (M = 1304 ms,SD = 430). This may partly reflect the simpler task of com-paring word meaning rather than completing a sentence,but it also reflects the 1 s SOA between the first wordand the response alternatives, which allowed some prepro-cessing of the first word. RTs were measured from theonset of the response alternatives. The average time takenfor the intervening scrambled word task was approxi-mately 15 min (M = 14.89 min, SD = 1.49).

Table 7 presents mean response errors and latency fortarget trials by trial condition. Unlike previous experi-ments, there was an unusually low error rate for unprimedactive-voice trials. This resulted in a negative primingtrend in errors for the active-voice condition. However,positive priming trends were seen in the passive-voiceerror rates and in response latency for both voice condi-tions. Of primary interest, when priming trends were posi-tive, they existed both for targets primed by word-meaning evaluations and sentence completions.

Fig. 6 presents means of the rate of correct respondingby condition. As in previous experiments, there was a largemain effect for target voice, F1(1,90) = 156.47, p 6 .001,partial g2 = .64 [response latency F1(1,90) = 110.11,p < .001, partial g2 = .55; percentage errors F1(1,90) = 6.93,p = .01, partial g2 = .07]. There was also a significant overallpriming effect in the comparison of unprimed to bothpriming conditions combined, F1(1,90) = 11.33, p 6 .001,partial g2 = .11 [response latency F1(1,90) = 18.62,p < .001, partial g2 = .17; percentage errors F1(1,90) < 1].Of primary interest, the contrast between the two primingconditions produced a small but statistically significanteffect, F1(1,90) = 4.79, p = .03, partial g2 = .05 [responselatency F1(1,90) = 1.96, p = .16; percentage errorsF1(1,90) < 1]. As seen in Fig. 6, participants had a greater

Table 7Mean response time for correct responses and percentage error in target trials of Experiment 6 by sentence voice and priming condition.

Trial type Response errors (%) Response latency (ms)

Active voice Passive voice Active voice Passive voice

M SD M SD M SD M SD

Unprimed 3.6 4.7 6.6 7.5 2392 738 2571 774Primed-Words 4.9 5.6 4.8 6.0 2262 601 2536 717Primed-Proposition 4.6 5.5 5.1 6.5 2231 661 2506 755


Res

pons

e S

peed

(cor

rect

tria

ls/m

in)

0

22

24

26

28

30

32UnprimedWord Meanings PrimedProposition Primed



correct response rate when primed by complete proposi-tions compared to individual word meanings. This effectdid not interact with target voice, F1 < 1. Finally, a posthoc analysis of target trials primed by word meaningscompared to unprimed targets revealed a small but statis-tically significant priming effect, F1(1,90) = 3.99, p = .049,partial g2 = .04 [response latency F1(1,90) = 11.48,p = .001, partial g2 = .11; percentage errors F1(1,90) < 1].

In the analysis by items, the effects were not as largeand reliable as those found in the participant analyses.There was a large main effect for target voice,F2(1,89) = 96.28, p 6 .001, partial g2 = .52 [response latencyF2(1,89) = 79.66, p < .004, partial g2 = .47; percentageerrors F2(1,89) = 5.75, p = .019, partial g2 = .06]. Further-more, the contrast of unprimed correct response rate tothat of both priming conditions combined revealed a smalloverall priming effect, F2(1,89) = 4.00, p = .048, partialg2 = .04 [response latency F2(1,89) = 8.99, p < .001, partialg2 = .09; percentage errors F2(1,89) < 1]. However, the con-trast between the two priming conditions was not statisti-cally significant, F2(1,89) = 1.56, p = .22 [response latencyF2(1,89) < 1; percentage errors F2(1,89) < 1]. Finally, targettrials primed by word meanings compared to unprimed

targets was not a statistically significant effect,F2(1,89) = 1.49, p = .226 [response latency F2(1,89) = 4.63,p = .034, partial g2 = .05; percentage errors F2(1,89) < 1].

Discussion

The results offer only tentative support for two conclu-sions because the effects were small, and they did notreach statistical significance when data were analyzed foritem means or when RT and error data were analyzed sep-arately. Despite this inconsistency, the outcome of the pri-mary hypothesis tests favor the following interpretations.First, priming of individual word meanings, unconnectedto expressions of a proposition, appeared to have a smallfacilitating impact on subsequent proposition evaluationin the form of sentence completion. This word-primingeffect persisted beyond the typical duration of word-levelsemantic priming effects: The unrelated intervening taskbetween prime and target trials rendered a minimumdelay of approximately 15 min, and for many trials thelag would approximate 30 min. This offers further supportfor the existence of long-term word-level semantic prim-ing effects when priming tasks demand greater semanticprocessing than is typical in lexical decision or word nam-ing tasks (Becker et al., 1997; Joordens & Becker, 1997;Was, 2010; Woltz, 2010).

The second conclusion supported by the primary statis-tical test is that not all of the facilitation in primed targetsentences could be explained by word-level priming. Therewas additional facilitation attributable to the prior pro-cessing of semantically related sentences rather than theindividual words that comprised the sentences. However,as noted, this effect was small in magnitude and was notsignificant in the ancillary methods of analysis. Regardingthis inconsistency, the secondary statistical tests, whichwere included for readers who are more familiar withthem, lacked power in all previous experiments comparedto the primary tests (i.e., the effect size estimates werenotably smaller in the large majority of cases for item anal-yses and participant analyses using RT and error data sep-arately). Therefore it is not surprising that small,statistically significant effects in the primary analysis werefound to be nonsignificant in the less powerful analyses.Consequently, we believe the evidence supports a conclu-sion that there is likely to be a small amount of additionalfacilitation from proposition, compared to individual wordfacilitation produced by the current word-priming method.

Related to the interpretation of this small effect, wechose to address the possible contribution of individualword priming effects by a method designed to maximize


word-meaning facilitation. Using this method, the wordprime trials likely enhanced overall word-meaning activa-tion relative to that from proposition primes for severalreasons. First, the word-comparison prime trials exposedparticipants to an additional synonym of the target trialword meaning compared to the proposition prime trials.In addition, we constructed the word-meaning task toensure that all information relevant to the subsequentmeaning comparison was activated by presenting each ofthese words for 1 s before the response choices were avail-able (see Sabol & DeRosa, 1976). As a result of this method,the summed average time of processing the three wordprimes (including both SOA and RT) was more than threetimes greater than that of corresponding sentence-comple-tion primes. Finally, each word prime trial required anexplicit decision and response regarding word meaning.These task attributes make it likely that the word-meaningprime condition overestimated the contribution of word-level priming in the sentence-completion prime condition.We designed the experiment in this manner, not to evalu-ate the relative contributions, but to provide a strong testof the possibility that there is proposition-level primingbeyond word-level priming. We believe the results, evena small effect in the primary analysis, favor a conclusionthat there is abstract proposition-level semantic primingin addition to word-level priming.

We acknowledge the complexity of determining word-level contributions to persistent proposition primingeffects. This complexity arises because there may be nomethod for indisputably producing the same level of indi-vidual word activation as that produced by the propositionprime sentences. Individual word meanings could beprimed with simpler tasks such as lexical decision, but thismight underestimate the word-level contribution fromsentence primes. Alternatively, participants could beprimed by unrelated proposition expressions containingthe words of target propositions. However, such primingwould produce associations with propositions other thanthe target one, and this could reduce target trial primingdue to semantic competition or interference.2 Finally,rather than priming individual words, one could evaluatethe level of facilitation in meaning-comparison target trialsprimed by proposition expressions. This would evaluatewhether the sentence-completion primes produce individ-ual word meaning facilitation that can be measured in a taskother than sentence completion. In our view, none of thesemethods can answer unequivocally the question of thedegree to which word-level priming contributes to proposi-tion priming effects. Our approach of maximizing thesemantic priming of individual words suggests that thatthe sentence-completion priming of proposition observedin these experiments cannot be explained fully by word-level priming. However, we believe a variety of methods willbe needed in future research to develop a fuller understand-ing of the relative contributions of word-level and proposi-tion-level memory representations.

2 We thank an anonymous reviewer for pointing out a weakness in ourargument against this method: If interference from other propositionassociations affected priming, this would in itself support a view thatlexical priming alone cannot explain the current findings.

General discussion

The six experiments reported here tested the existenceof long-term semantic priming for single propositionexpressions representing general knowledge. A vastamount of research has been devoted to understandingshort-lived semantic priming effects for words processedin isolation (see McNamara, 2005). Despite this focus onword-level meaning, concepts central to the explanationof these effects, such as spreading activation, have beeninvoked to explain cognitive processes that are more inte-grative and complex than word identification. Thisincludes the automatic activation of propositions throughpreviously established associations. For example, in theConstruction-Integration (CI) model of discourse process-ing (Kintsch, 1988, 1998), text is first parsed into proposi-tions, then constructed into propositional networks whilestill in working memory. The resultant networks are thensupplemented by, and integrated with, additional proposi-tional representations presumed to be passively availablein long-term memory (Kintsch, 1998). Similarly, Kintsch,Patel, and Ericsson’s (1999) application of Ericsson andKintsch’s (1995) theory of long-term working memory(LTWM) describes the result of reading comprehension asthe formation of new nodes in memory. These nodes them-selves take the form of propositions which have beenderived from the text and are linked in a complex manner.The process by which linking takes places is automatic andoccurs as textual features elicit appropriate processingstrategies, which in turn create new networks of memorynodes (i.e., propositions). In this manner, the body ofLTWM is formed. Pertinent to the current experiments, justas words in a text automatically generate a LTWM, propo-sitions also ‘‘activate knowledge from their (immediate)semantic neighborhood’’ (p. 10), though for an unspecifiedlength of time.

Other conceptualizations of working memory alsoinclude activated long-term memory representations thatare available for cognitive processing but are outside thefocus of attention (Anderson, 1993; Cowan, 1995, 1999;Just & Carpenter, 2002; Oberauer, 2002). Of interest here,these available but unattended memory representationscan be activated through pre-existing associative linkswith information that is in attentional focus. That is, a formof semantic or associative priming is assumed to play a rolein the availability of some long-term memory elements inworking memory. For these models to explain complexcognitive task performance such as problem solving andcomprehension, this active information cannot berestricted to representations of individual word meaning.

Memory for proposition content has received consider-able attention over many decades both in empirical inves-tigation and theory (e.g., Foltz, 2003; Kintsch, 1974, 1988,1998; Ratcliff & McKoon, 1978). The current research dif-fered from much of the prior work in its investigation ofimplicit rather than explicit access to propositional contentwith the use of a priming paradigm. Furthermore, weinvestigated semantic priming effects attributable to mem-ory for a single isolated proposition that represented priorknowledge rather than new episodic information pre-


sented in the context of discourse. The goal was to isolatecharacteristics of implicit, semantically mediated access toindividual proposition representations of prior knowledge.

These experiments produced evidence supporting threeprimary conclusions. First, facilitation in the sentence com-pletion task due to prior processing of semantically relatedexpressions persisted beyond what has been consideredtemporary semantic priming. In three experiments, facili-tation effects were evident after prime-target delays span-ning 15–30 min.

Second, the facilitation appeared to reflect strengthen-ing of abstract meaning representations rather than mem-ory for encoding, decision, and response processesemployed in prime trials. Several experiments demon-strated that the magnitude of persistent priming was unaf-fected by altering syntax consistency between prime andtarget sentences. One experiment found priming magni-tude to be unaffected by altering whether the subject orobject of the proposition predicate was omitted acrossprime and target sentence completions. Another experi-ment found that memory for prior selection and rejectionof response alternatives did not contribute to the primingeffects. These findings are consistent with the idea thatthe memory responsible for facilitation over this timeframe represented abstract meaning rather than opera-tion-specific features of the prime trials.

Third, the possible role of explicit recall of prime trialcontent in the facilitation effects was investigated.Although this issue is difficult to evaluate with certainty,the evidence supported a conclusion that deliberate recol-lection of prime trial content during target trial perfor-mance was not responsible for facilitation and, morelikely, the engagement of explicit memory processesreduced performance facilitation.

Evidence from the final experiment provided qualifiedsupport for a conclusion about the contribution of individualword priming in the observed sentence-completion facilita-tion effects. The evidence favors an interpretation thatmemory changes to individual word meaning representa-tions may contribute to long-term proposition primingeffects, but it is unlikely to fully account for them. We viewthis experiment as providing initial evidence regarding atheoretically important issue that will demand much moreresearch to resolve. The limited goal of Experiment 6 wasto test whether robust priming of individual word meaningscould account for proposition priming effects. Despite anattempt to maximize the magnitude of word-level priming,the evidence suggested that sentence-completion facilita-tion was greater when the prime events allowed for propo-sitions to be constructed compared to when the componentwords were optimally primed.

Although the focus of this research was on semanticpriming of propositions, this final experiment also pro-vided evidence regarding long-term semantic priming atthe level of word meaning. The evidence suggested word-level priming at prime-target lags that exceeded thoseinvestigated in earlier work. Furthermore, an importantfeature of the current word-level semantic priming effectis that it was obtained within target trials that demandedthe evaluation of sentences rather than individual words.Prior evidence for long-term semantic priming of word

meaning used prime and target trials that both requiredparticipants to make decisions about individual words(e.g., Woltz, 2010). If long-term semantic priming effectsare found only when prime and target tasks are highly sim-ilar in their processing demands, it would indicate a levelof specificity rather than generality in the availability ofstrength increments in semantic memory representations.In contrast to this, the current evidence suggests that thestrengthening of semantic representations for individualword meanings can be recruited after a substantial delayto facilitate sentence completion when related words areinvolved. This not only speaks to the generality of long-term semantic priming across prime-target task demands,but it is inconsistent with alternative explanations thatlong-term semantic priming is either operation-based oris based on episodic retrieval of prior processing events.

In total, the current findings support a view that long-term semantic priming occurs both for word and proposi-tion meaning, and that these two forms of priming share acore attribute. Woltz (2010) found that word-level seman-tic priming effects that persisted over multiple interveningtrials did not reflect memory for prime trial cognitive oper-ations other than the retrieval of word meaning. This isconsistent with the current findings that prime-sentencesyntax and sentence-completion operations do not affectthe magnitude of proposition priming. In both cases, theevidence supports the view that facilitation reflectsstrength changes to abstract representations of meaningrather than operation-specific memory for prime trial pro-cessing. However, it should be noted that word meaningretrieval and proposition construction independent of syn-tax could in themselves be considered operations thatunderlie the observed priming effects. These operationsare so intimately tied to memory for the semantic contentthat they are not readily separated.

It should also be reiterated that we have assumed thememory processes instrumental in the observed primingeffects are pre-existing representations of propositionsrepresenting general knowledge. The assumption of prop-ositions as a fundamental building block of more complexknowledge structures is common in cognitive theory, espe-cially discourse processing theories. However, there arealternative views of the memory processes and structuresthat could underlie these priming effects. Application ofprinciples from Logan’s (1988) instance theory would pro-pose that each prime event creates a new memory repre-sentation that could be responsible for target facilitation.Alternatively, the facilitation effects could be attributedto new associative links between existing conceptual rep-resentations of the arguments within each proposition.This account assumes new links between existing knowl-edge structures. Our findings cannot distinguish betweenthese various views of how prime trial processing affectedsemantic memory representation to produced facilitationin target trials. However, the evidence does support theconclusion that it is the abstract representations of seman-tic content that are changed, not memory for the cognitiveoperations performed on or with the knowledgerepresentations.

Our conclusions have implications for the operation-based explanation of long-term semantic priming that


appears to be appropriate for some prior evidence. In dis-cussing early evidence for long-term semantic priming,McNamara (2005) suggested that existing evidence at thattime could reflect different memory processes than thoseresponsible for temporary priming: Long-term primingcould reflect memory for prior operations rather thanabstract semantic memory. This also was the interpreta-tion of subsequent long-lasting semantic priming effectsreported by Woltz and Was (2007) and Was (2010). Ofinterest, the research supporting operation-based accountsutilized priming tasks that required category membershipevaluations (also see Hughes & Whittlesea, 2003). In thesetasks, when different category evaluation operations wererequired on prime and target trials, priming was elimi-nated. This suggested that facilitation could not be attrib-uted to broad activation of the semantic categories butrather to the specific operations and type of informationused to activate the categories. Such evidence for long-term semantic priming does not challenge existing modelsof short-term priming at the word level because itaccounts for long-term priming effects with memory rep-resentations other than those for abstract meaning. How-ever, the current evidence, and that reported by Woltz(2010), suggests that some long-term semantic primingeffects cannot be explained by memory for prime trialoperations. At least for some stimulus content and taskdemands, both short- and long-term semantic primingeffects appear to reflect increased availability of abstractmeaning representations. Furthermore, these memory rep-resentations reflect meaning at the level of words andpropositions.

A comprehensive theory of semantic priming must becapable of accounting for a wider variety of evidence thanthat accumulated over several decades of research usinglexical decision and word naming tasks. Memory processesdescribed in spreading activation and compound cue theo-ries of temporary semantic priming (Dosher & Rosedale,1989; McNamara, 1992; Ratcliff & McKoon, 1988) are inad-equate for explaining longer-lasting effects (see Hughes &Whittlesea, 2003; McNamara, 2005). As noted earlier, theonly model thus far to account for persistent semanticpriming effects is the distributed network model proposedby Becker et al. (1997). In this model, changes to the net-work due to prime word processing are assumed to reflectslow-decaying strength increments rather than temporaryactivation which can be disrupted by subsequent process-ing. Specifically, the Becker et al. model describes long-term semantic priming as a function of the deepening ofattractor basins. Each stimulus word consists of a patternof attractor basins that deepens as a function of use.Semantic priming effects are due to shared attractor basinsthat reflect overlapping semantic features between primeand target words. Because this model proposed a learningrather than temporary activation mechanism in the net-work, this account of priming predicts facilitation that per-sists over multiple intervening events.

However, the Becker et al. (1997) model was proposedto account for semantic priming effects at the word level.The current evidence suggests that such a model mustaccount for priming at both word and proposition levels.This is a plausible extension given that other distributed

network models account for the representation of bothwords and propositions. As described earlier, Sharkey(1990) proposed a distributed network model thatincluded a process by which existing propositional unitsare selected based on the words in the proposition beingpresented. The most active propositional unit is thenassociated with the stimulus at hand, based on lexicalfeatures.

Given Sharkey’s (1990) model, it is plausible that prop-ositions could be represented across patterns of attractorbasins in a manner similar to the way the Becker et al.(1997) model represented words. That is, simple ideas suchas a wedding can be held in a church would share attractorbasins with semantically related ideas expressed with dif-ferent words and syntax: a cathedral can accommodate amarriage. In the same way that semantically related wordsare subject to priming due to strengthening of these sharedattractor basins, semantically related propositional unitscould exhibit priming due to their shared attractor basins.Given current and previous findings, the challenge for thisor any other model of semantic priming would be toaccount for both short- and long-term semantic primingeffects, possibly but not necessarily with a single mecha-nism, and to predict facilitation effects at both word andproposition levels of meaning.

References

Anderson, J. R. (1974). Verbatim and propositional representation ofsentences in immediate and long-term memory. Journal of VerbalLearning and Verbal Behavior, 13, 149–162.

Anderson, J. R. (1983). The architecture of cognition. Cambridge, MA:Harvard University Press.

Anderson, J. R. (1993). Rules of the mind. Hillsdale, NJ: Erlbaum.Becker, S., Moscovitch, M., Behrmann, M., & Joordens, S. (1997). Long-term

semantic priming: A computational account and empirical evidence.Journal of Experimental Psychology: Learning, Memory, and Cognition,23, 1059–1082.

Bock, J. K. (1986). Meaning, sound, and syntax: Lexical priming insentence production. Journal of Experimental Psychology: Learning,Memory, and Cognition, 12, 575–586.

Bock, J. K., & Griffin, Z. M. (2000). The persistence of structural priming:Transient activation or implicit learning? Journal of ExperimentalPsychology: General, 129, 177–192.

Branigan, H. P., Pickering, M. J., & McLean, J. F. (2005). Primingprepositional-phrase attachment during comprehension. Journal ofExperimental Psychology: Learning, Memory, and Cognition, 31,468–481.

Cowan, N. (1995). Attention and memory: An integrated framework. NewYork: Oxford University Press.

Cowan, N. (1999). An embedded-processes model of working memory. InA. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms ofactive maintenance and executive control (pp. 62–101). New York:Cambridge University Press.

Dosher, B. A., & Rosedale, G. (1989). Integrated retrieval cues as amechanism for priming in retrieval from memory. Journal ofExperimental Psychology: General, 118, 191–211.

Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory.Psychological Review, 102, 211–245.

Ferreira, V. S., Bock, K., Wilson, M. P., & Cohen, N. J. (2008). Memory forsyntax despite amnesia. Psychological Science, 19, 940–946.

Foltz, P. W. (2003). Quantitative cognitive models of text and discourseprocessing. In A. C. Graesser, M. A. Gernsbacher, & S. R. Goldman(Eds.), Handbook of discourse processes (pp. 487–523). Mahwah, NJ,US: Lawrence Erlbaum Associates Publishers.

Gernsbacher, M. S. (1985). Surface information loss in comprehension.Cognitive Psychology, 17, 324–363.

Hughes, A. D., & Whittlesea, B. W. A. (2003). Long-term semantic transfer:An overlapping-operations account. Memory & Cognition, 31,401–411.

http://refhub.elsevier.com/S0749-596X(14)00150-8/h0005











































Jacoby, L. L., & Dallas, M. (1981). On the relationship betweenautobiographical memory and perceptual learning. Journal ofExperimental Psychology: General, 110, 306–340.

Jacoby, L. L., & Whitehouse, K. (1989). An illusion of memory: Falserecognition influenced by unconscious perception. Journal ofExperimental Psychology: General, 118, 126–135.

Joordens, S., & Becker, S. (1997). The long and short of semantic primingeffects in lexical decision. Journal of Experimental Psychology: Learning,Memory, and Cognition, 23, 1083–1105. http://dx.doi.org/10.1037/0278-7393.23.5.1083.

Just, M. A., & Carpenter, P. A. (2002). A capacity theory of comprehension:Individual differences in working memory. In T. A. Polk (Ed.), Cognitivemodeling (pp. 131–177). Cambridge, MA: MIT Press.

Kintsch, W. (1974). The representation of meaning in memory. Hillsdale, NJ:Erlbaum.

Kintsch, W. (1988). The role of knowledge in discourse comprehension:A construction-integration model. Psychological Review, 95,163–182.

Kintsch, W. (1998). Comprehension: A paradigm for cognition. New York,NY, US: Cambridge University Press.

Kintsch, W., Patel, V. L., & Ericsson, K. A. (1999). The role of long-termworking memory in text comprehension. Psychologia: An InternationalJournal of Psychology in the Orient, 42, 186–198.

Kintsch, W., & van Dijk, T. A. (1978). Toward a model of textcomprehension and production. Psychological Review, 85, 363–394.

Kolers, P. A. (1973). Remembering operations. Memory & Cognition, 1,347–355.

Kolers, P. A. (1976). Reading a year later. Journal of ExperimentalPsychology: Human Learning and Memory, 2, 554–565.

Kolers, P. A., & Roediger, H. L. (1984). Procedures of mind. Journal of VerbalLearning and Verbal Behavior, 23, 425–449.

Ledoux, K., Traxler, M. J., & Swaab, T. Y. (2007). Syntactic priming incomprehension: Evidence from event-related potentials. PsychologicalScience, 18, 135–143.

Loftus, G. R., & Masson, M. E. J. (1994). Using confidence intervals inwithin-subject designs. Psychonomic Bulletin & Review, 1, 476–490.

Logan, G. D. (1988). Toward an instance theory of automatization.Psychological Review, 95, 492–527.

Luo, C. R. (1993). Enhanced feeling of recognition: Effects of identifyingand manipulating test items on recognition memory. Journal ofExperimental Psychology: Learning, Memory, and Cognition, 19,405–413.

Masson, M. E. J. (1995). A distributed memory model of semantic priming.Journal of Experimental Psychology: Learning, Memory, and Cognition,21, 3–23.

Masson, M. E. J. (2011). A tutorial on a practical Bayesian alternative tonull-hypothesis significance testing. Behavior Research Methods, 43,679–690.

Masson, M. E. J., & MacLeod, C. M. (1997). Episodic enhancement ofprocessing fluency. In D. L. Medin (Ed.). The psychology of learning andmotivation: Advances in research and theory (Vol. 37, pp. 155–210). SanDiego, CA, US: Academic Press.

McNamara, T. P. (1992). Priming and constraints it places on theories ofmemory and retrieval. Psychological Review, 99, 650–662.

McNamara, T. P. (2005). Semantic priming: Perspectives from memory andword recognition. New York, NY, US: Psychology Press.

Meyer, D. E., & Schvaneveldt, R. W. (1971). Facilitation in recognizingpairs of words: Evidence of a dependence between retrievaloperations. Journal of Experimental Psychology, 90, 227–234.

Meyer, D. E., Schvaneveldt, R. W., & Ruddy, M. G. (1975). Loci ofcontextual effects on visual word recognition. In P. M. A. Rabbit & S.Dornic (Eds.). Attention and performance V (Vol. V, pp. 98–118). NewYork: Academic Press.

Neely, J. H., & Kahan, T. A. (2001). Is semantic activation automatic? Acritical re-evaluation. In H. L. Roediger, III, J. S. Nairne, I. Neath, & A. M.Surprenant (Eds.), The nature of remembering: Essays in honor of RobertG. Crowder (pp. 69–93). Washington, DC, US: American PsychologicalAssociation.

Oberauer, K. (2002). Access to information in working memory: Exploringthe focus of attention. Journal of Experimental Psychology: Learning,Memory, and Cognition, 28, 411–421.

Pickering, M. J., & Ferreira, V. S. (2008). Structural priming: A criticalreview. Psychological Bulletin, 134, 427–459.

Plaut, D. C., & Booth, J. R. (2000). Individual and developmentaldifferences in semantic priming: Empirical and computationalsupport for a single-mechanism account of lexical processing.Psychological Review, 107, 786–823.

Ratcliff, R., & McKoon, G. (1978). Priming in item recognition: Evidence forthe propositional structure of sentences. Journal of Verbal Learning andVerbal Behavior, 17, 403–417.

Ratcliff, R., & McKoon, G. (1988). A retrieval theory of priming in memory.Psychological Review, 95, 385–408.

Sabol, M. A., & DeRosa, D. V. (1976). Semantic encoding of isolated words.Journal of Experimental Psychology: Human Learning and Memory, 2,58–68.

Sachs, J. S. (1967). Recognition memory for syntactic and semantic aspectsof connected discourse. Perception & Psychophysics, 2, 437–442.

Sachs, J. S. (1974). Memory in reading and listening to discourse. Memory& Cognition, 2(1-A), 95–100.

Schneider, W., Eschman, A., & Zuccolotto, A. (2002). E-Prime user’s guide(Version1.1). Pittsburgh: Psychology Software Tools Inc..

Sharkey, N. E. (1990). A connectionist model of text comprehension. In G.B. Balota, B. F. d’Arcais, & K. Rayner (Eds.), Comprehension processes inreading (pp. 487–514). Hillsdale, NJ: Erlbaum.

Singley, M. K., & Anderson, J. R. (1989). The transfer of cognitive skill.Cambridge, MA: Harvard University Press.

Sorensen, L. J., & Woltz, D. J. (2014). Transforming response time and errordata to address tradeoffs, submitted for publication.

Townsend, J. T., & Ashby, F. G. (1978). Methods of modeling capacity insimple processing systems. In N. Castellan Jr., & F. Restle (Eds.),Cognitive theory (Vol. III, pp. 199–239).

Townsend, J. T., & Ashby, F. G. (1983). The stochastic modeling ofelementary psychological processes. Cambridge, UK: CambridgeUniversity Press.

Traxler, M. J. (2008). Lexically independent priming in online sentencecomprehension. Psychonomic Bulletin & Review, 15, 149–155.

Traxler, M. J., & Tooley, K. M. (2008). Priming in sentence comprehension:Strategic or syntactic? Language and Cognitive Processes, 23, 609–645.

Tse, C.-S., & Neely, J. H. (2005). Assessing activation without sourcemonitoring in the DRM false memory paradigm. Journal of Memoryand Language, 53, 532–550.

Tse, C.-S., & Neely, J. H. (2007). Semantic and repetition priming effects forDeese/Roediger-McDermott (DRM) critical items and associatesproduced by DRM and unrelated study lists. Memory & Cognition,35, 1047–1066.

Tulving, E., & Schacter, D. L. (1990). Priming and human memory systems.Science, 247, 301–306.

Wagenmakers, E.-J. (2007). A practical solution to the pervasive problemsof p values. Psychonomic Bulletin & Review, 14, 779–804.

Was, C. A. (2010). The persistence of content-specific memory operations:Priming effects following a 24-h delay. Psychonomic Bulletin & Review,17, 362–368.

Was, C. A., & Woltz, D. J. (2007). Reexamining the relationship betweenworking memory and comprehension: The role of available long-termmemory. Journal of Memory and Language, 56, 86–102.

Woltz, D. J. (2010). Long-term semantic priming of word meaning. Journalof Experimental Psychology: Learning, Memory, and Cognition, 36,1510–1528.

Woltz, D. J., & Was, C. A. (2006). Availability of related long-term memoryduring and after attention focus in working memory. Memory &Cognition, 34, 668–684.

Woltz, D. J., & Was, C. A. (2007). Available but unattended conceptualinformation in working memory: Temporarily active semanticcontent or persistent memory for prior operations? Journal ofExperimental Psychology: Learning, Memory, and Cognition, 33,155–168.







http://dx.doi.org/10.1037/0278-7393.23.5.1083

http://dx.doi.org/10.1037/0278-7393.23.5.1083


























































































































Documents

Long-term semantic priming of propositions representing general knowledge