1'; ".42" ‘0' '4II:""' :W' ' In -"‘)‘..' l :, 1", 1.. .1.E..'."';L "_ w ' ' "'fi'144'f'1‘1‘44g'fi”fig-2"1":ku "r"" '4 I984 I." -. - I " 1"¢"4."~' ~, v'»‘ IMO} I, Q .' ; “I y ‘I ”J; 11ml" ” ' ":5 h" . $3 5'. ‘ ’6 Km{-‘&.. ' 1"" ' ~"l- ",L'? $11.5 ' ‘ #4"? “Yb-'5‘ "3‘ . . $3 “444* 4, 41‘ 45ng gb' I“ {($91}; 3434.1 ward“ m5 I I. '. 4" if?“ I “491. W H " fig)” 1:1" 4 1‘4”” “’4 (:2 Y_ ’1 'EJ Win-154$ 4346‘! '3‘, $3.??? ‘43? fix“ ., ,I " " """P1""' ' ""111"? """"'"4" ‘1- " “".' fly .4“ 414.4 “Vt v ' I""“J°";I1m""‘ 4“ . 34 4‘. ,' «MI-5,4111 ~ - u 'II'llgl'; traEI4i“" i":(\' "4...! 3:1” I '(\ ‘. ya *fil') :4" firflg H "I! w 39431;“ 4 . 4 {0.5% 1 :2: EH?" "“1 I' VH4“ ‘33? 4'. 13 ', I} “1),,3II56‘fIY‘II 1 a .- ~ L”: r '_ < ' .2 ~ ‘ . ‘ ¢'_' ‘ ._ $2“: .. «.— ~ :- fl _:~‘ ~-_--.: : :.r‘: $1433? I II) 444' ‘‘‘‘‘‘ "“ ' ‘ .1- .1“) $494 '4 r ,. }' '1;:lj:.‘ W4" "‘ é' _ 9'" ”I1." ‘Jl‘ 'fiy'n'etl‘l Iris'l'ln.5'* '1’?" ‘)""|'K 4 I ,'|‘ 1:? 'fl.:') £s4‘éEJ ;\f| . .‘ " a t, ‘1 .4 144' 4:". ‘L 2.7:), 1' 1.1I ‘m H. :1'1'4': gill ' 4" ' J""'l-,l."!;i"4:r:? ' HIT-'6' ' " '1" h 1" U '14 1" :' RM lb "1' '4':- "1' '1‘! i“ A . '"M Y ‘ ‘ “1’11 >9 . 1.1431131"? 4 ‘4'." \Sfil' '44} ‘ id" ih'u’x- 3.4:“ q“. I I -’r - I, . l .4}' , ‘ 4' '1 ,1... "1.34.“ 4 fr , 4,. II, "'.".I11'1 I I ‘ ""4";,':I"" '4 "(' “'1' , '4‘” .l {5 1' wk M" .. "1 I" "1.5"" " ‘4' H4". " ' ."1'*"'- H ."},'I" I.I‘Y'JL'":‘; . " I I" '1. I" "'I‘u' '3: ' (41 i" ' I.‘ ' 1 2." "4".‘4'4‘H .. . ’l' '-‘\'. 371‘“ ””3"", I. "2,421.“. til-Q "'1"'"""‘ Ikivgjb"; ‘ 28g k: ‘ 0 II. -' II' "1'44 - H 1|""';.'.‘-" -I-,‘ "'4‘4'11u .. '”.1-'14'-;.;,‘:I';'_:':.I- 0"". 1;“. 4 . ‘2 " ' "'44,”:“49' 1 ‘42.; '1: M '1 . ”4" 'I v.1 "'0". .I1‘I"'. . .1“ .44.: “I.“ er :ll' '5‘ "Jg’? in.“ 5:“ ‘13:, Indy”: '1' '. )'|":'I.'41’" l11|-"' Il-f'W‘I 41!. ' .'I “LR" ”it": ..' '1' n :' .l‘" ‘. ""1" _:.I 4" :II”, -‘|.‘ 1' 1"". HI Inf“ 6| ' YI" (Yih.""::_l1ul.'" W!- '."‘ I'. I1I.I’|'.::I I .‘II‘f" | J .4?» ”I . ', I 1‘ I" ,. ‘ ‘ ‘5‘” , [1"I'1-a I 'h .. ' I . ;ll ’ .I ll"; 3‘}: ‘ u-I " .z'f‘I'J' 0'4"?“ II 4“ h It. {H lino; ',"lf_"'||i9l II . I' - "4| 'I‘I . "l”|‘1":'1‘ II. n. (‘3 . "', .rK1I'I1.'!IV|4IJ JG." 'I‘ H'H'lll'LII II “'14:." I". l I" § 1.1-, 4| I, ‘ .“;;;_'f. .. ., 1' ' II 1. l‘i' I ‘1‘ 1' n I, '4 . .‘ My I'p‘ol' f| U ‘l‘(%h N; JV. ‘1'" II)" .I‘ ‘0 ‘l ”'3' ‘ 1' 7511' ...l "145W “I '.“'I"" "II-I4“ :CQR'L',“ I”, ~11 '1 II M I: .1: '. "34"" "‘4: 4 ' -éLJL " ”4.5.4344 I1. "'1. (2,311, 4:4... bls'l".14HRL-J4 " ”WW/WM This is to certify that the dissertation entitled THE EFFECTS OF ATTENTION AND AWARENESS IN PRIMING STUDIES OF THE EPISODIC AND SEMANTIC MEMORY DISTINCTION presented by Dale Dagenbach has been accepted towards fulfillment of the requirements for Ph.D. degree in Psychology ' M ' f : Thomas H."°é§'f»’fi“°’ Date 1/4/85 um”.-- AL...” A ' ’- 1m 0-12771 MSU LIBRARIES RETURNING MATERIALS: Place in book drop to remove this checkout from your record. FINES will be charged if book is returned after the date stamped below. THE EFFECTS OF ATTENTION AND AWARENESS IN PRIMING STUDIES OF THE EPISODIC AND SEMANTIC MEMORY DISTINCTION BY Dale Dagenbach A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Psychology 1985 ABSTRACT THE EFFECTS OF ATTENTION AND AWARENESS IN PRIMING STUDIES OF THE DISTINCTION BETWEEN EPISODIC AND SEMANTIC MEMORY BY Dale Dagenbach The functional independence of episodic and semantic memory was tested by examining the nature of episodic and semantic priming effects in a lexical decision and a recognition task. Two manipulations designed to isolate strategic from automatic components of priming effects were used in order to determine whether any effects obtained were due to inherent relationships between episodic and semantic information or were a consequence of subjects' intentional efforts to relate that information. Subjects first memorized a list of paired associates for a recall test, after which they made either lexical decisions or episodic recognition judgments for targets preceded by primes that were episodically, semantically, or episodically and semantically related, or unrelated to the targets. The primes were presented at either a 200 msec or 2900 msec stimulus onset asynchrony from the target, and above or below the threshold for conscious detection. The stimulus onset asynchrony and threshold manipulations provided two tests of whether any effects obtained were automatic or mediated by strategic factors. The strength of episodic associations present also was varied by including an overlearned condition wherein subjects intensively studied episodically related pairs. This was included in order to provide evidence as to whether any differences between episodic and semantic priming effects were entirely a consequence of the greater strength of semantic associations. The results generally indicated that under conditions precluding strategic factors in priming, semantic decisions were facilitated only when a semantic relationship between items was present, and episodic decisions were facilitated only when an episodic association was present. These results, in conjunction with several dissociative priming effects between tasks, wherein episodic or semantic primes produced different effects in the lexical decision and recognition tasks, are seen as providing support for a distinction between episodic and semantic memory. Overlearned episodic associations gave some indication of producing priming effects similar to semantic associations, but it is argued that this was due to a change in semantic memory rather than a change in the episodic association. However, there also were indications of interference with automatic semantic priming as a consequence of episodic activation, especially in the subthreshold priming conditions, suggesting that there may be a more intricate relationship between episodic and semantic memory than one would expect from completely separate systems. ACKNOWLEDGEMENTS The research described herein has benefitted greatly from the profound advice and assistance of Tom Carr. A large measure of gratitude is due to him for his contributions to this study, and for serving as a mentor throughout my graduate career. Also due thanks are the other members of my dissertation committee: Ellen Strommen, Gordon Wood, and Rose Zacks. Few dissertations have been blessed with such careful reading by such insightful reviewers. ii TABLE OF CONTENTS Page LIST OF TABLES ...................................... v INTRODUCTION ........................................ 1 THE EPISODIC AND SEMANTIC MEMORY DISTINCTION ........ 4 Related Distinctions ............................. 11 Multiple Trace vs. Strength Theory ............. 11 Personal Memory, Generic Memory, and Skills .... 12 Summary of the Distinctions ...................... l3 Empirical Evidence ............................... l7 Priming Studies ................................ 17 Perceptual Recognition and Recognition Memory .. 36 Sentence Recognition and Verification .......... 47 Recognition Failure of Recallable Words ........ 51 Summary and Discussion of the Empirical Evidence . 54 OVERVIEW OF THE CURRENT STUDY ....................... 6% Theoretical PrediCtionS O O O O O I O I O O O O O O O O O O O O O O O O O O 64 Priming of Lexical Decisions ................... 65 Priming Of Recognition O O O O O O O O O O O O O O O O O O O O O O O O O 66 METHOD 0.00.00.00.00.0...0.0.0....OOOOOOOCOOOOOOOOOOO 68 Subjects ......................................... 68 Stimulus Materials ............................... 68 Prime-Target Relatedness and Control Conditions 68 Study List Construction ........................ 73 Test List Construction ......................... 76 Materials in Subsequent Conditions ............. 78 Pattern Mask ................................... 89 Procedures ....................................... 80 Session I ...................................... 80 Session II ..................................... 81 iii RESULTS AND DISCUSSION O...O0.0...OOOOOOOOOOOOOOOOOOO Lexical Decision Task ............................ Suprathreshold Long SOA ........................ Suprathreshold Short SOA ....................... Subthreshold Long SOA .......................... Subthreshold Short SOA ......................... Summary of the Lexical Decision Results ........ Episodic Recognition Task ........................ Suprathreshold Long SOA ........................ Suprathreshold Short SOA ....................... Subthreshold Long SOA .......................... Subthreshold Short SOA ......................... Summary of the Recognition Results ............. Cued Recall ...................................... GENERAL DISCUSSION .................................. Transfer Effects ................................. Dissociative Priming Effects ..................... The Relationship between Episodic and Semantic Memory 0..00......00......OOOOOOOOOCOOOOOOOOO0.0.0 REFERENCES .0.0......0....O0......OOOOOOCOOOOOOOOOOOO APPENDIXA00.00....OOOOOOOOOOOCOOOOOOOOOO0.00.0.0... Sample Set of Stimulus Materials ................. iv 88 88 92 95 99 109 111 113 113 117 119 123 124 125 126 126 131 133 136 144 144 10. 11. LIST OF TABLES sumary Of Priming StUdieS oooooooooooooooooooooo Summary of Perceptual Recognition and Recognition studies 00....O0.0.0...0.0.000...OOOOOOOOOOCOOOOO Prime-Target Relationships ...................... Mean RT (msec) and Percent Error for Lexical DeCiSions: word Targets OOOOOOOOOOOOOOOOOOOOOOOO Mean RT (msec) and Percent Error for Lexical Decisions: Nonword Targets ..................... Mean Lexical Decision RT (msec) and Percent Error in Subthreshold Conditions after Subject Deletion: Word Targets ......................... Mean Lexical Decision RT (msec) and Percent Error in Subthreshold Conditions after Subject Deletion: Nonword Targets OOOOOOOOOOOOOOOOOOOOOO Mean RT (msec) and Percent Error for Recognition: word Targets 0.0...OOOOOOOOOOOOOOOOOOOOOOOOOOOOOO Mean RT (msec) and Percent Error for Recognition: Nonword Targets .0.0.0.0....OOOOOOOOOOOOOOOOOOOOO Mean Recognition RT (msec) and Percent Error in Subthreshold Conditions after Subject Deletion: word Targets .0..0...OOOOOOOOOOOOOOOOOOOOOOOOOOOO Mean Recognition RT (msec) and Percent Error in Subthreshold Conditions after Subject Deletion: Nonword Targets 0.0.0.000...OOOOOOOCOOOOOOOOOOOOO Page 35 48 70 89 91 102 103 114 116 120 121 INTRODUCTION Many current accounts of human memory distinguish among various domains in one way or another. For example, recent influential theories have suggested distinctions between memory of a transient nature and memory that is permanent (Atkinson & Shiffrin, 1968), memory for propositional information and memory for analogical information (Pavio, 1971), memory for procedural information and memory for propositional information (Anderson, 1976), etc... In some instances these taxonomies represent an effort to parse an enormously complex puzzle into more tractable research problems, and as such are not likely to engender much debate. However, in other cases these distinctions are meant to imply the existence of functionally distinct.domains of memory, with real psychological, and presumably physical, differences between them. Theories proposing distinctions of this type are usually highly controversial. The present study tested one such theory proposing that episodic memory, or memory for autobiographical events, is functionally distinct from semantic memory, or memory for impersonal facts. Theories arguing for distinct domains in memory often generate extensive empirical testing aimed at demonstrating exactly how the proposed forms of memory differ from each other, or that they, in fact, do not. 1 2 This is not an easily achieved goal since distinct forms of memory might result from maintaining information in discrete stores, having different forms of representation, requiring different procedures for encoding or retrieval, or the various possible interactions of these factors. Conclusively demonstrating the presence or absence of these characteristics in human memory has proved to be an immensely difficult task that typically requires a multitude of studies examining the issue from different perspectives. The present study provides one such perspective for the episodic/semantic distinction. The present study addressed the question of whether or not episodic and semantic memory are functionally distinct systems by examining how episodic and semantic judgments for specific items were affected by the prior presentation of episodically related and semantically related information. Under at least some conditions, finding effects of semantic information on episodic judgments, and of episodic information on semantic judgments, can be seen as evidence against distinct systems. The present study further examined whether the effects obtained were due to conscious expectations about the relationships between items, or were an inherent consequence of automatic spreading activation in memory and therefore potentially more informative regarding the cognitive architecture. This was accomplished by varying the amount of time elapsing between the presentation of 3 the related information and the subsequent judgment, and by manipulating the extent to which the preceding related information was consciously accessible. THE EPISODIC AND SEMANTIC MEMORY DISTINCTION One recurring and intuitively appealing distinction between domains of memory focuses on the difference between one's recollection of an actual event in one's life, and one's memory for facts and impersonal information such as the meanings of words. Bergson (1911) drew this distinction in contrasting what he called "pure memory" and "habit memory". Pure memory was the remembrance of one's own past experiences as specific events, whereas habit reflected adaptation to one's environment, as might be revealed by adjustments of perceptual and motor mechanisms. Reiff and Scheerer (1959) drew on Bergson's thinking in making a similar distinction based on whether or not there was an autobiographical index involved in a memory. Tulving (1972) brought these ideas to the attention of psychologists in the form of the episodic/semantic distinction. In this, he distinguished between memory for events and the temporal interactions among those events (episodic memory), and the memory that was necessary for the use of language and other symbolic knowledge, including knowing the meanings of language components and the relations between them (semantic memory). According to Tulving's distinction, statements about episodic knowledge referred to personally remembered unique experiences, whereas statements about semantic memory did 4 5 not, akin to Reiff and Scheerer's (1959) distinction based on the autobiographical vs. impersonal nature of a memory. Tulving further suggested that most existing research had studied episodic rather than semantic memory functioning. Tulving's (1972) proposal specified a number of ways in which the two memory systems might differ, thus phrasing the distinction in terms appropriate for empirical tests. One suggested difference involved the nature of the information that one would expect to find stored: Episodic memory would contain information about events, including their temporal and spatial coordinates, whereas semantic memory would contain facts, and would not include information about the context of occurrence. Moreover, episodic memory would contain information having an autobiographical reference, whereas information in semantic memory would have a cognitive reference. Tulving (1972) further suggested that the conditions and consequences of retrieval in each system might differ. For example, semantic memory would be capable of deriving information that never had been explicity stored because its conceptual organization permitted inferences to be made, whereas episodic memory by definition could only retrieve what had been stored. Retrieval from either system would change episodic memory by adding a new episode, whereas it would have no substantial effect on the more stable semantic memory system. Also, information stored in semantic memory would be less vulnerable to 6 interference or erasure than information stored in episodic memory since the former is more likely to be overlearned. Finally, Tulving suggested that episodic and semantic memory would likely differ with respect to the degree to which each depended on the other. The episodic/semantic distinction received a great deal of attention, some favorable and some critical. Tulving (1983) himself has characterized his 1972 theory as inchoate, and suggests that many criticisms of it were due to misunderstandings of what was intended. He now indicates that there were several fundamental inadequacies in the 1972 formulation, including its failure to sufficiently elaborate the similarities of the two memory systems, its failure to clearly indicate that it was not meant to be an exhaustive taxonomy of human memory, the choice of the term "semantic memory", which implied too limited of a domain, and most importantly, its failure to clearly define the inherent relationship between episodic and semantic memory. Tulving's (1983) current explanation of the episodic/semantic distinction varies most noticeably from the 1972 formulation with respect to the last issue mentioned above. Tulving now distinguishes between the remembrance of a personal episode and the knowledge of its semantic contents; the former is supported by episodic memory, but the latter is supported by semantic memory. In accordance with this modification, Tulving now argues, 7 perhaps with some exaggeration, that most existing memory research has studied semantic memory and that virtually nothing is known about episodic memory. In his reworking of the episodic/semantic distinction, Tulving (1983) has again enumerated ways in which the two systems might be expected to differ from each other. These differences are speculative, many are extensions of those suggested in the 1972 theory, and most are meant to be quantitative rather than qualitative--they refer to differences in the degree to which each system will be characterized by some attribute rather than exclusive differences. They will be described in some detail because they form one basis by which a test of the episodic/semantic distinction must proceed. First, Tulving (1983) suggests that the characteristics of information storage in episodic and semantic memory differ in several respects. One difference of this nature involves the source of the information: Episodic memory may store relatively unprocessed sensory information, whereas the semantic system requires that information first be understood and comprehended. The units of information in each system also differ: Episodic memory stores episodes or events, with discrete beginnings and endings, whereas semantic memory contains different, as yet unspecified, units of information (e.g., features, facts, ideas, concepts, propositions, etc...). The organization of information in 8 each memory system is distinguishable: Episodic memories are organized according to some temporal scheme, whereas semantic memory has a conceptual organization. The reference of information in each system is also different: Episodic memories necessarily have a personal referent, whereas semantic memories refer to the external world in ways that may or may not include information about the rememberer. Finally, Tulving suggests that unspecified elements present in episodic memory give rise to a subjective sense of veridicality upon retrieval that is lacking when information in semantic memory is retrieved--although one believes that a semantic fact is true, retrieval of it is not usually accompanied by a feeling of belief according to Tulving. Tulving (1983) also suggests that there are differences in the operations of each system. These overlap to some extent with the previously described differences in contents. Registration of information into episodic memory can be more direct than in semantic memory; the episodic system can record directly perceptible attributes of events, whereas the semantic system registers the products of analyses of information, usually in symbolic form. The episodic system can respond to queries about the temporal succession of stored information because it includes temporal context information in the memory trace, whereas the conceptually organized semantic system cannot. Conversely, the 9 episodic system has limited inferential capabilities; it is generally incapable of extracting implicit information from events, whereas the semantic system is adept at doing so. Affect may play a greater role in determining operations in the episodic system since affective information is more likely to be included in an episodic memory trace. Context dependency also may distinguish the systems; it is often suggested that a critical difference is that episodic memory preserves information about the context of encoding, whereas semantic memory information is decontextualized. Also, information in the episodic system is more vulnerable to transformation or loss as a consequence of subsequent operations. Episodic and semantic memory also may be distinguished in terms of accessing characteristics. Although not always the case, accessing episodic information typically requires conscious effort, whereas accessing semantic information is typically automatic and effortless. There is a difference in the language typically used to address each system: Questions requiring episodic information have the form "What did you do at time A in place B2", whereas those requiring semantic information have the form "What is X?". Similarly, reports following retrieval of information from each system may be distinguishable on a linguistic basis: We tend to "remember" episodic information, whereas we "know" semantic information (Tulving, 1983). The 10 consequences of retrieval from each system may differ: Episodic memories are more subject to recoding, and are more likely to show greater accessibility as a result of past retrieval. Finally, Tulving (1983) notes that the two memory systems probably vary in their developmental history, although not necessarily in the way that one would predict. Critics of the episodic/semantic distinction have noted that episodic events logically have to precede the formation and organization of semantic memory (e.g., Schank, 1975). While this is true, the episodic memory system per se does not necessarily have to precede the semantic memory system. Tulving argues that the evidence suggests the contrary; children learn semantic information before there is any evidence of episodic remembering in Tulving's sense of the term. In addition to providing an expanded list of possible distinctions between episodic and semantic memory, Tulving (1983) has tried to clarify what he means by the distinction. It is not intended to be merely a heuristic, or to indicate just a difference between the type of information stored in each, or to represent opposite ends of some continuum. Instead, Tulving means the distinction to imply functionally distinct systems. Tulving further argues that there is more to the distinction than just the presence or absence of personal reference or contextual information, but since he fails to expand on that line of ll reasoning, other operational definitions are difficult to determine. Related Distinctions Since Tulving (1972) suggested the episodic/semantic distinction, a number of similar distinctions have been proposed by other theorists. Some of these are sufficiently similar to the episodic/semantic distinction to warrant a brief description since tests of them may be pertinent to that distinction. Multiple Trace vs. Strength Theory A common operational definition of the distinction between episodic and semantic memory notes that in episodic memory a new memory trace must be established for each event, whereas information processing in semantic memory typically yields activation of representations already existing in memory rather than the creation of a new trace. Thus, demonstrating that subjects were accessing unique traces in episodic tasks and some unitized representations in semantic tasks would lend support to the episodic/semantic distinction. This operational definition of the episodic/semantic distinction leads into the debate over multiple trace vs. strength theory accounts of the effects of repetition in various memory tasks. Hintzman (1976, 1978) has argued for a multiple trace explanation of the effects of repetition, but also suggests that multiple trace theory may specifically characterize the effects of repetition on 12 episodic memory, whereas strength theory may describe the effects of repetition on semantic memory. Similarly, Feustel, Shiffrin, and Salasoo (1983) attribute repetition effects in perceptual recognition tasks to unique episodic traces formed for each presentation of an item, but further suggest the existence of a unitized representation or logogen that influences processing of lexical stimuli, and thus maintain a weak form of episodic/semantic distinction. Jacoby (1983a) has suggested that one way of reformulating the episodic/semantic distinction is to view semantic memory organization as an emergent property of amalgamations of episodic memory encodings. Personal Memory, Generic Memoryyiand Skills Brewer and Pani (1983) distinguish between three memory systems in their "botany" of memory: personal memory, generic memory, and skills. In their model, personal memory operates in the recollection of a particular incident from an individual's past and thus corresponds closely to episodic memory. Generic memory consists of semantic and perceptual memory for generalized information and thus corresponds closely to semantic memory, although it is an expansion of Tulving's original use of the term. Finally, skills are the ability to carry out practiced motor or cognitive acts. Brewer and Pani (1983) suggest a reformulation of episodic memory that corresponds partly to Tulving's (Tulving, 1983). In their scheme, some subjective 13 experience of an autobiographical component of a memory is the prerequisite for classification of that memory as a "personal memory". Memory of the content of an autobiographical episode is distinguished from an experiential remembering of the episode. Thus, if one's phenomenal experience of the memory is best described as "I remember that I ...", rather than "I remember (doing, being, etc...) ...", the memory is classified as generic rather than personal since it is for the content of the episode rather than for the episode per se. They argue that traditional verbal learning experiments, such as remembering word pairs, do not require episodic knowledge, but rather require rote linguistic skills or generic memory, contrary to Tulving's (1972) original suggestion. Summary of the Distinctions Several possible bases for distinguishing between domains of memory are suggested in the above accounts. Those comprising qualitative distinctions will be focused on in this discussion because of the inherent difficulties in testing distinctions of a quantitative nature, wherein one domain would be characterized as having more or less of some attribute than the other. Adopting this strategy yields several differences between domains to consider. First, the critical element of episodic memory for both Tulving (1983) and Brewer and Pani (1983) is subjective; memories having a personal referent or an autobiographical component are somehow different from 14 those without such a component. Personal or episodic memory is experienced as events from one's own past, whereas semantic or generic memory is not. In Brewer and Pani's (1983) model, there is a vaguely defined experiential quality to personal memory that is lacking from other memories; not only does it contain information from a unique episode, but accessing it brings about some degree of reexperiencing of the event. "Since such reexperiencing presumably does not accompany the retrieval of all information from unique episodes, Brewer & Pani's formulation of personal memory seems to describe a subset of episodic memory as Tulving (1983) defines it. The subjective, autobiographical aspect of the episodic/semantic distinction appears to be the critical factor in these models. However, it also may be the least studied. The obvious test for whether a memory is episodic or personal according to the autobiographical criterion would be to make use of the phenomenal experience associated with the memory, but the problems associated with such introspection have led researchers to try to define episodic or personal memories in ways that are more amenable to empirical test. Attempting to operationalize the episodic/semantic distinction without resorting to introspection points to the other key elements: To demonstrate retrieval of an episodic memory, one wants to show that a unique memory trace has been accessed that contains information about a 15 specific event in the subject's past. There are two sources of evidence that one might use to support such a claim. First, one might examine the consequences of accessing that trace on the availability of other information in memory: If the trace is unique and not stored in some organized semantic system, accessing it ought not necessarily activate other traces that are semantically but not episodically related to that event. On the other hand, if the trace is in semantic memory, one would expect automatic activation of semantically related information according to most models of semantic memory. Secondly, the presence of information about encoding context in the memory trace has been used as a basis for arguing that a unique trace is being accessed rather than some unitized representation containing information about the event. This argument reflects the assumption that unitized representations in semantic memory by definition would not include such information. In this vein, Tulving (1983, p. 55) operationalizes the episodic/semantic distinction as follows, "If successful performance on the task is not possible in the absence of retained information from a particular episode, the task is classified as an 'episodic' one; if successful performance on the task is possible in the absence of information from any particular episode, the task is classified as a 'semantic' one." Such operational definitions are not completely l6 satisfactory since the interrelationships of these components are likely to be complex. As Johnson (1983) points out, memory with an autobiographical reference generally may include information about the encoding context, but this does not always have to be the case--one can experience a face as being familiar without being able to place it in some context, or one can have a sense of deja vu for an experience without being able to retrieve a previous context. Similarly, information about the context at encoding conceivably could be contained in a propositional format of the type, "I remember that I saw A at time and place 8.", which would then make it part of episodic memory according to Tulving's formulation, but not part of personal memory as defined by Brewer and Pani (1983). Tests of the episodic/semantic distinction and other similar distinctions must consider the extent to which they provide specific information about each of these operational definitions since it is possible that these describe independent factors rather than various manifestations of one underlying phenomenon. Such tests must specify clearly whether they have used the presence or absence of a subjective sense of reexperiencing upon retrieval of a memory, information about encoding context, or examined the activation of other information as a means of testing the distinction. Evidence about one does not necessarily reflect on the others, and no general law for l7 determining their relationships is immediately apparent. The point is that experiments using one criterion need to consider the extent to which the procedures fulfill the other criteria. Indeed, one of the benefits of such tests may be their ability to tell us whether these factors are or are not intrinsically related. Empirical Evidence Tulving (1983) is probably correct in surmising that no single experiment will provide a critical test of the episodic/semantic distinction. Instead, the pattern of many such studies will serve to guide eventual decisions about its status, and presumably will serve to guide revisions in the distinction as needed. Fortunately, the requisite data base is at least partially complete. There are several studies that claim to provide a direct test of the episodic/semantic distinction, and many others that have provided relevant information. Priming Studies Many experimental studies of the episodic/semantic distinction have employed priming techniques to address the issue. Priming refers to the effect that the presentation of one item has on the speed or accuracy with which judgments can be made about another. Such effects can result from both semantic relatedness of items for semantic judgments (Meyer & Schvaneveldt, 1971), and episodic relatedness of items for episodic judgments (Ratcliff & McKoon, 1981). 18 Current theories suggest that there are automatic and strategic components of priming (e.g., Neely, 1977; Posner & Snyder, 1975a, 1975b). The automatic component is manifested at very short prime-target stimulus onset asynchronies (SOAs--e.g., 0-250 msec), quickly reaches asymptote, and, according to most studies, produces facilitation from related primes as revealed in faster, more accurate responses, but does not generate inhibition from unrelated primes. The strategic component of priming develops more slowly (e.g., 500-2000 msec prime-target SOAs), is slow to reach asymptote, and can result in inhibition when the prime is not related to the target, as well as facilitation when the prime is related to the target. Most studies support this characterization of priming effects, although some discrepancies have been noted; for example, inhibition has been reported at short prime-target SOAs in some cases (Antos, 1979). In general, priming experiments addressing the episodic/semantic issue search for priming effects, or the lack of such effects, as a result of semantic priming in episodic tasks, and episodic priming in semantic tasks. Experiments of this type have been referred to as "transfer" studies (Tulving, 1983). Evidence of episodic priming effects in a semantic task, or of semantic priming effects in an episodic task, as indicated by changes in the speed or accuracy of the episodic or semantic decisions, is seen as weighing against the l9 episodic/semantic distinction, whereas the failure to find these effects supports the distinction. This is especially true when the priming effects appear to be generated by the automatic component of the priming process. In addition to looking for transfer effects, one can also test the episodic/semantic distinction by looking for dissociative priming effects, wherein episodic priming or semantic priming produces different effects in an episodic vs. a semantic task. If one obtains dissociative priming effects-~different effects from the same kind of priming in the different tasks-—one has evidence in support of the episodic/semantic distinction, whereas parallel operating characteristics weigh against it. Of course, the validity of priming tests depends in part upon whether the tasks are truly episodic or semantic, and on whether the relationship between the prime and target is truly episodic or semantic. Priming studies typically have used some form of recognition judgment about whether an item occurred previously in a specific context at a specific time as an episodic task. Examples of semantic tasks used to test semantic memory include verification of assertions (Tigers have stripes) and lexical decisions (Is "glump" a real word?). A semantic prime in either kind of task would be an item with some preexperimental association to the target, based on similarity in meaning or high associative frequency, 20 whereas an episodic prime would be one whose relationship to the target was based solely on cooccurrence with the target at the time of encoding, such as preceding the target in a list of words memorized during the experiment. Hermann & McLaughlin (1973) reported one of the first such tests of the episodic/semantic distinction in which they manipulated the episodic and semantic relationships between items in a recognition task. Subjects in their experiment studied exemplars of several categories, and then were given a recognition task in which they had to indicate if neither, one, or both of the words in a two-word display were old (studied) or new (unstudied). Semantic relationships between the items in this task did have an effect on the episodic judgments, but only in conjunction with episodic relatedness: Responses to both old and new items were faster if both items were members of one of the categories represented in the studied materials than if they were unrelated, but when old or new items related solely on the basis of preexperimental association (category relationships not present in the study list) were presented, no benefit of relatedness was found. These results were interpreted as support for the episodic/semantic distinction since no priming effect due solely to semantic relatedness was found in an episodic task. Using the same design, Hermann and Harwood (1980) focused on the results when neither word in the two-word 21 recognition task had been presented previously. Some unstudied pairs were semantically primed--both items were members of one category that was not present in the studied material. Other unstudied pairs were episodically and semantically primed--both items were unstudied exemplars of one of the studied categories. Finally, still other unstudied pairs were neither episodically nor semantically primed. Hermann and Harwood reasoned that if only semantic priming occurred, one would expect equal facilitation for pairs that were semantically related, and pairs that were episodically and semantically related, compared to pairs that were unrelated. If episodic, but not semantic, priming occurred, one would expect effects for the pairs that were both episodically and sematically related, but not for those pairs that were only semantically related. Finally, if both episodic and semantic priming occurred, one would expect effects in both the semantically related, and semantically and episodically related conditions, but greater facilitation should occur in the latter. Semantic priming effects were found only when the words were also members of one of the categories that had been represented in the study materials, in agreement with what one would expect if, in fact, only episodic priming occurred. No semantic priming was found when both words were members of a category that had not been present in the studied material. Arguing that this null result was 22 only one of many failures to find semantic priming of episodic decisions in repeated experiments, Hermann & Harwood (1980) concluded that episodic and semantic memory were distinct systems. Other early evidence of a possible difference between episodic and semantic memory based on priming effects in a semantic task was reported by Neely (1977). Neely found that semantic priming of lexical decisions could occur under conditions appropriate for automatic or strategic priming, but episodic priming apparently required the investment of conscious attention, and only occurred at the long prime-target SOAs associated with strategic effects. This seems quite plausible and fits nicely with current views suggesting that strong associations, such as one would expect for semantically related items, should produce both automatic and strategic priming, whereas weak associations, such as one might expect for episodically related items, would not generate automatic priming effects. However, Neely's finding must be qualified by the fact that episodic priming was defined somewhat uniquely in this study--it was based on an attention shift procedure wherein seeing the name of one particular category as a prime warned subjects to expect an exemplar from a different category as the target. Moreover, although a distinction between episodic and semantic priming effects such as Neely drew seems reasonable, a subsequent study by Ratcliff and McKoon 23 (1981) indicates that automatic episodic priming also can be obtained, at least in an episodic task. In their study, recognition judgments for items from studied sentences were facilitated when preceded by another item from the same sentence even when the items were separated by a short prime-target SOA. In contrast, priming with a word from a sentence different from the one in which the target word had appeared yielded inhibition only at longer SOAs. Manipulating the proportion of related primes, which is known to affect the strategic component of priming but not the automatic component, had no effect on the facilitation they obtained. Thus episodic priming within an episodic task behaved similarly to semantic priming within a semantic task. Further studies by Neely and Durgunoglu (1985) described below have replicated this. McKoon and Ratcliff (1979) also used a priming procedure to performed a more comprehensive test of the episodic/semantic distinction. They examined the effects of episodic and semantic associations between items on the latency and accuracy of decisions about those items in a lexical decision (semantic) task and an item recognition (episodic) task. In the lexical decision task, subjects memorized short lists of word pairs for a possible cued recall test. Each study list was followed by a list of words and nonwords requiring lexical decisions. Semantic priming of 24 words due to a semantically related word preceding the target in the test list produced significant facilitation, as one might expect. Episodic priming due to the first member of a word pair preceding the second in the test list also yielded significant and approximately equal facilitation. In the recognition task, subjects again memorized short lists of word pairs. Each study list was followed by a test list in which subjects had to indicate whether the items had been present in the study list. Semantically primed words were those preceded in the test list by a semantically related target, and episodically primed words were those preceded in the test list by the same item that had preceded them in the study list. Episodic priming produced a large amount of facilitation, and episodic and semantic priming together yielded even more. Semantic priming by itself did not significantly facilitate response latency, but did increase the error rate significantly. Based on these results, McKoon and Ratcliff (1979) concluded that there was little merit to the episodic/semantic distinction: They had found effects of episodic information in a semantic task, and effects of semantic information in an episodic task, as one would expect if episodic and semantic memory were not distinct systems. Moreover, they claimed that these priming effects were automatic, and thus eliminated possible 25 strategic explanations of the results. However, there are two possible problems with their experiments, and consequently with their conclusions. First, McKoon and Ratcliff (1979) erred in a procedural manipulation that leaves one of their claims in question. Their findings could be reconciled easily with the episodic/semantic distinction by assuming that these effects were strategic in nature--given the prime, the subject intentionally anticipated the episodically related target in the lexical decision task. McKoon and Ratcliff argued that this was not possible because they used a very short prime-target stimulus onset asynchrony (SOA) of 150-300 msec. In this range of SOAs, one would find automatic but not strategic priming according to current theories (Posner & Snyder, 1975a,b; Neely, 1977). But McKoon and Ratcliff confused SOA with interstimulus interval (181), or the duration between the end of the prime and the beginning of the target. Their prime was presented for 500 msec, and together with the 150-300 msec ISI, created actual SOAs of 650 to 800 msec that may have been long enough to allow strategic effects to occur. Secondly, the effect of semantic priming in the episodic task was only for error rate and not for response latency, contrary to how these results have often been portrayed. Finding increased error rates in a recognition task following semantic priming does not necessarily support a unitary model of memory. A number of models of 26 recognition suggest that ease of identification is a separate process that may lead to a positive recognition judgment regardless of whether the original memory is accessed (e.g., Mandler, 1980; Jacoby and Dallas, 1981; Jacoby, 1983). Therefore, subjects in McKoon and Ratcliff's (1983) experiment might have had to actively suppress semantic information in order to prevent false recognition judgments. The greater rejection of semantically primed targets might have been a side effect of this active suppression. A subsequent experiment by Carroll & Kirsner (1982) that properly controlled for prime-target stimulus onset asynchrony failed to find effects of episodic priming in a semantic task. Carroll and Kirsner examined episodic and semantic priming in recognition and lexical decision tasks using two-word displays, wherein the stimulus onset asynchrony is essentially zero. In their study, an initial block of lexical decision trials served as the study condition for a subsequent test condition wherein subjects made either more lexical decisions or recognition judgments. The variables assessed included the effects of semantic relatedness at test, semantic relatedness at encoding (in the study block), the number of prior presentations of the items in the pair, and the episodic relatedness of the pair (as established in the study block). For lexical decisions, there was no benefit from 27 presenting semantically associated words in the study block if that semantic relationship was not preserved in the second block. An effect of episodic relatedness was found in the second block of trials, but only for semantically related pairs. Semantic relatedness at test yielded significant facilitation as one would expect, but repetitions of semantically related pairs did not increase the amount of facilitation obtained. Also, no benefits of repetition were found when the test items were two nonwords. In the recognition memory experiment, subjects were instructed to identify the word pairs in the second block as "old" if both items had appeared in the first block, or "new" if either or both items had not appeared in the first block. Clear benefits from episodic relatedness (presentation of the same items paired together in both the first and second blocks) were found for both accuracy and response latency. Semantic relatedness at encoding yielded significantly greater accuracy, but no decrease in response latency. Repetitions of episodically related pairs significantly increased accuracy and decreased response time. There was no effect of semantic relatedness at test, and no interaction between semantic relatedness at test and repetitions. Thus, Carroll and Kirsner (1982) found differential effects of semantic status and of episodic status on recognition and lexical decision, and differential 28 consequences of repetition for these effects. Carroll and Kirsner interpreted their results as indicating that there were two forms of recognition memory--episodic memory and perceptual memory, paralleling the conclusion of another study of this phenomenon (Jacoby & Dallas, 1983). Overall, Carroll and Kirsner's (1982) results also are compatible with those predicted by the episodic/semantic distinction: Semantic relatedness between items facilitated recognition only when the items were also episodically related, and episodic relatedness facilitated lexical decisions only when the items were also semantically related. Neely and Durgunoglu (1985) and Durgunoglu and Neely (1985) also have recently conducted a series of priming investigations similar in purpose to and roughly concurrent with the present study in which they sought to determine the factors that would account for the discrepancies between McKoon and Ratcliff's (1979) and Carroll & Kirsner's (1982) findings. Neely & Durgunoglu (1985) performed two experiments, but because one of these was contaminated by a procedural problem according to their own analysis, only the results from the other will be discussed here. In that experiment, subjects studied lists consisting of 12 word-word and word-nonword pairs, each of which was presented twice, for a recall test. After each recall test, they had to make either episodic recognition judgments based on whether the target was in 29 the study list, or lexical decisions. In either case, a prime requiring no response appeared either 150 msec or 950 msec before the target item--the test trials were not blocked by prime-target SOA. The prime-target relationships were varied so that target words and nonwords were primed by the same semantically unrelated word they were paired with in the study list (episodic priming), by a different semantically unrelated word from the study list (unrelated intralist), or by a neutral prime (xxx). Studied word targets were also primed by semantically related words from the same study list, but which they had not been paired with (semantic intralist). Finally, the test lists also contained a condition wherein the target was semantically primed, but neither the prime nor the target had appeared in the study list (semantic extralist). This procedure allowed Neely and Durgunoglu to obtain two measures of priming. Priming effects could be determined by comparing related to unrelated conditions, or by comparing both related and unrelated conditions to the neutral condition. Because it allows more direct comparison with other results described, and because there may be problems associated with the use of "xxx" as a neutral prime (de Groot, Thomassen, & Hudson, 1982; den Heyer, Briand, & Dannenbring, 1983), their findings will be described in terms of related compared to unrelated conditions unless otherwise noted. 30 In the episodic task, Neely and Durgunoglu (1985) found that recognition of word and nonword targets was significantly facilitated at both SOAs by episodic priming, replicating Ratcliff and McKoon (1981). On the other hand, semantic intralist priming of studied word targets produced significant inhibition at both SOAs, and semantic extralist priming of unstudied word targets produced significant inhibition at the long SOA. For lexical decisions, Neely and Durgunoglu (1985) found no episodic priming for word or nonword targets at either SOA. However, they also found no evidence of semantic intralist priming, and semantic extralist priming produced facilitation only in comparison to the neutral prime, but not the unrelated prime. Neely and Durgunoglu's (1985) results offer some support for the episodic/semantic distinction stemming from the dissociative priming effects obtained: Episodic information influenced episodic but not semantic judgments; semantic intralist priming of studied targets produced significant inhibition at both SOAs for recognition, and no significant effects for lexical decisions; and semantic extralist priming yielded significant inhibition for episodic judgments, and insignificant facilitation compared to the neutral prime for lexical decisions. However, the implications of these findings may be overshadowed by their failure to replicate the nearly ubiquitous finding of semantic priming of 31 semantic decisions. One might view the failure to obtain semantic intralist priming of lexical decisions at the long SOA as a consequence of subjects expecting the episodically related target following the prime, and therefore responding slower when the target was a semantically related word: Having studied "cat-light", the subject expected to see "light" after "cat“, and therefore was slower to respond to "dog". But according to current models of priming (e.g, Posner & Snyder, 1975a,b), the effects of such conscious expectations are not manifested at short SOAs, and therefore the short SOA trials in Neely and Durgunglu's experiment should have produced semantic priming. Indeed, this is what has been reported in other experiments (Neely, 1977). Moreover, Neely and Durgunoglu also failed to obtain semantic priming of nonstudied target words at either SOA. In this condition, there was no prior expectation of a different target following the prime. Durgunoglu and Neely (1985) performed a subsequent set of experiments that elaborated the preceding findings. Once again, their primary goal was to explain the differences in findings between studies such as McKoon and Ratcliff's (1979) that provide evidence of episodic priming of semantic tasks, and those such as Carroll & Kirsner's (1982) and Neely and Durgonoglu's (1985) that fail to find such effects. The same basic procedure was used as in their previous study : A list of 32 paired-associates was studied, followed by a set of test trials with targets primed by episodically related and semantically related items. The same priming conditions were used, but, in this case, only in a lexical decision task, and with the systematic manipulation of several variables that had differed in the above studies. In one experiment, Durgunoglu and Neely varied between subjects whether prime-target SOAs (150 vs. 950 msec) were blocked or random, and whether the semantic primes appeared in the same study list (semantic intralist condition), but with a different item as in Carroll & Kirsner's (1982) study, or were not seen before the test list (semantic extalist condition) as in McKoon and Ratcliff's (1979) study. Priming from episodically related and semantically related primes was measured relative to a neutral and an unrelated prime. As before, the discussion will focus on the priming relative to the unrelated prime unless otherwise noted. Episodic priming of word targets failed to produce facilitation of lexical decisions at either SOA for those subjects who received semantic intralist primes and for whom the prime-target SOAs were mixed. However, episodic priming of word targets did produce significant facilitation at the long SOA for each of the other groups of subjects--those for whom the semantic primes were from the same list (semantic intralist), but for whom the prime-target SOAs in the test trials were blocked; those 33 for whom the prime-target SOAs in the test trials were mixed, but for whom the semantic primes came from outside the study list (semantic extralist); and those for whom the semantic primes were from outside the study list and the prime-target SOAs were blocked. Thus, episodic priming of lexical decisions at the long SOA was eliminated by the combination of mixed SOAs and semantic intralist primes, but not by either factor in isolation. More importantly, semantic priming once again produced incongruous effects. Semantic priming of studied word targets by primes not in the study list (semantic extralist) failed to produce significant facilitation in any case. The semantic intralist primes for studied word targets produced inhibition compared to the neutral condition, but not compared to the unrelated condition. Extralist semantic priming of unstudied word targets also failed to produce significant facilitation for any group. By running a group of subjects using only the materials from the test lists, with no study lists, Durgunoglu and Neely (1985) were able to obtain facilitatory semantic priming at both SOAs using the materials from the "studied word" conditions and at the long SOA for the materials from the "nonstudied word" conditions, and thus demonstrated that the absence of semantic priming was not an artifact of the particular materials used. Durgunoglu and Neely therefore concluded that something about having studied paired associates prior to a lexical decision task 34 interfered with the semantic priming mechanisms, although exactly what factor caused this effect in their research but not in others remains to be explained. These results suggest some constraints on the assumption that facilitatory semantic priming in a semantic task will be automatic and ubiquitous. Under at least some conditions, semantic priming of semantic decisions does not appear to occur. Episodic study of relationships may be one such condition, suggesting some interaction between episodic and semantic memory. It should be noted, however, that Durgunoglu and Neely's (1985) interpretation of these findings is not the only possible one. In their studies, the locus of the effect is unclear: It may be that semantic priming is occurring to a full extent, but that biases or interference in the response stage induced by the episodic primes are masking it. Further research is needed to determine just what aspect of paired associate learning might affect subsequent semantic priming in a semantic task. In another experiment, Durgunoglu and Neely (1985) obtained episodic priming of lexical decisions at a short prime-target SOA, apparently demonstrating a transfer effect due to automatic activation, a finding that should seriously weaken the episodic/semantic distinction. However, this was accomplished by having the subjects study all of the word targets and none of the nonword 35 Table 1 Summary of Priming Studies Episodic Priming in Semantic Task yes no McKoon & Ratcliff (1979) Carroll & Kirsner (1982) Durgunoglu & Neely (1985) Neely & Durgunoglu (1985) Semantic Priming in Episodic Task yes no McKoon & Ratcliff (1979) Hermann & McLaughlin (1973) Neely et al. (1983) Hermann & Harwood (1980) Neely & Durgunoglu (1985) Carroll & Kirsner (1982) Properties of Episodic and Semantic Priming same different Ratcliff & McKoon (1981) Neely (1977) Durgunoglu & Neely (1985) Neely & Durgunoglu (1985) Durgunoglu & Neely (1985) 36 targets, and by having no extralist semantic primes. Under these conditions, wherein study status was perfectly confounded with lexicality, episodic relations were sufficient to generate facilitation at the short SOA. Durgunoglu and Neely note that by using these constraints, they made it possible for subjects to solve the "semantic" task using episodic information--in other words, the task was no longer a test of semantic memory. The results of the priming studies of the episodic/semantic distinction that have been reviewed are summarized in Table 1. An attempt to reconcile these disparate conclusions will be made after the other pertinent literature has been reviewed. Perceptual Recognition and Recognition Memory A second series of studies that is germane to determining the merits of the episodic/semantic distinction are those ones examining the difference between perceptual recognition and recognition memory. In general, the application of this literature to the episodic/semantic distinction revolves around the issue of whether perceptual recognition entails accessing a unitized representation, such as one would expect to find in semantic memory, or whether it requires accessing of individual memory traces. Perceptual recognition can be viewed as a form of semantic memory task, in contrast to recognition, which is an episodic memory task, although this classification is not accepted by everyone (see 37 Tulving, 1983). If the classification of perceptual recognition as a semantic memory task is appropriate, than if one can demonstrate that perceptual recognition is based on accessing individual memory traces representing particular experiences rather than unitized representations, one has potential evidence against the existence of a distinct semantic memory system. Jacoby and Dallas (1981) first reported a dissociation between the effects of various manipulations on recognition and perceptual recognition, where the latter was measured by the ability to identify briefly exposed words. Using an incidental learning procedure, they examined the effects of item meaningfulness and the extent of processing on recognition memory and perceptual recognition. Using an intentional learning procedure, they examined the effects of word frequency, study time, item repetition, retention interval, and perceptual similarity between study and test items on these same measures. The incidental learning manipulations affected recognition memory--those items receiving "deeper" processing were more likely to be recognized. However, the same manipulations had no effect on the likelihood of perceptual recognition, which benefitted from a previous exposure, but was independent of recognition memory performance. To explain these findings, Jacoby and Dallas (1981) invoked the notion of a functionally distinct 38 perceptual memory component of semantic memory. Their results favored the idea that perceptual memory and autobiographical memory are independent, and to the extent that autobiographical memory and perceptual memory are analogous to episodic and semantic memory, supported the episodic/semantic distinction. However, not all manipulations in Jacoby and Dallas's (1981) study yielded evidence of independence. Repetitions of a word improved both recognition memory and perceptual recognition performance, while changing the modality of presentation so that study items were heard and test items were seen decreased the benefit from repetition for both recognition and perceptual recognition. These parallel effects slightly weaken the case for the episodic/semantic distinction. Another finding in Jacoby and Dallas's (1981) study may be even more problematic for the episodic/semantic distinction. They reported that the benefits accruing from a single prior exposure of a word in the perceptual recognition task persisted for 24 hours. This is incompatible with most current theories proposing an independent semantic memory system with some form of unitary representations since activation of the unitary representation from a previous exposure would not be expected to endure for this period of time. The repetition effect itself is not problematic--a straightforward interpretation of repetition effects on 39 perceptual recognition from the framework of a model with unitized representations in semantic memory such as logogens (Morton, 1979) would argue that presentation of a word during study might activate the semantic memory representation, and this activation would persist for a brief period, making subsequent perceptual recognition easier. However, the persistence of the repetition effect in perceptual recognition contradicts most assumptions about the expected duration of activation in semantic memory. If a single exposure causes such enduring activation in semantic memory, then in the course of reading any lengthy text, most of semantic memory would be activated. This would seem to obviate the utility of the concept of activation. Of course, this does not necessarily fatally damage logogen or other semantic memory models--it may be that current assumptions about the duration of activation are wrong, and that semantic activation does persist over great lengths of time. For example, one might have a logogen system in which activation was relative, with most items being activated to some extent, and in which there was a systematic relationship between frequency and recency of prior occurrence in terms of the amount of activation (see, e.g., Scarborough, Cortese, & Scarborough, 1977). Jacoby and Witherspoon (1982) extended the perceptual recognition research by examining dissociations of memory and awareness. They found that the presentation of 40 homophones in questions biased their subsequent spelling for both normal subjects and Korsakoff amnesics, thus revealing an apparent change in a semantic memory task (spelling) as a consequence of a single episodic presentation. Both normal subjects and Korsakoff amnesics showed a dissociation between this effect and recognition of the homophones. However, manipulating the relationship between aspects of the presentation form of items, such as typeface, and the form during later perceptual recognition revealed a great deal of specificity in the effect, leading Jacoby and Witherspoon to argue that the perceptual recognition effect reflected memory for individual instances rather than the activation of a central semantic memory representation. This contradicts the predictions of the episodic/semantic distinction if perceptual recognition is seen as a form of semantic task. Prompted mainly by the difficulty inherent in reconciling the persistent benefit in perceptual recognition accruing from a single exposure with any form of semantic memory model containing unitized representations, Jacoby (1983b) now argues that perceptual recognition and recognition memory tasks show independence only to the extent that they require different aspects of memory. According to Jacoby, any apparent independece between perceptual recognition and recognition memory thus reflects the different demands of the tasks and the different cues provided in each, rather than a difference 41 in the domain of memory that is accessed. Jacoby and Brooks (1984) describe a series of studies that elaborates on this conclusion. In these studies, they manipulated the amount of attention and the types of processing given to an item during encoding, and then examined the consequences of these manipulations on subsequent perceptual recognition. Jacoby and Brooks concluded that an item must be attended if subsequent perceptual recognition is to benfit from prior exposure, and that subsequent benefit will be influenced by surface characteristics such as modality of presentation or typeface to the extent that these characteristics are attended to during the initial phase. Because of the many parallel effects of manipulations of this type for perceptual recognition and recognition memory, Jacoby and Brooks argue that a unitary memory system supports both perception and recognition. It is worth noting at this point that the above interpretations of these findings are not the only possible ones, even though they are reasonable and deserving of consideration. For example, Jacoby (1983b) discounts differences in performance in perceptual recognition and recognition memory tasks as reflecting the different demands of the tasks rather than being a consequence of independent memory systems. In contrast, Johnson (1983) also argues that different task demands may differentially elicit the information stored in memory 42 subsystems, but in her theory this does not imply that the memory subsystems are not independent. Feustel, Shiffrin, & Salasoo (1983) recently reported data relevant to both the perceptual memory literature and to the literature on priming effects in episodic and semantic memory. In different experiments, the signal-to-noise ratio of masked words and nonwords was either set at a predetermined level for single item-mask exposures and accuracy was measured; or the signal-to-noise ratio in repeated short exposures of item-mask pairs was gradually increased until a desired level of accuracy was attained; or the signal-to-noise ratio of the item-mask pairs was increased at a constant rate until unambiguous identification occurred, in which case latency was the dependent measure. In each task, the primary factor of interest was the effect of previous exposures on performance. For the first two tasks, word and nonword exposure durations were set independently so that approximately 50% of each could be identified. In these tasks, the repetition of items yielded greater accuracy in perceptual recognition for both words and nonwords. Moreover, this improvement was not eliminated by changing from uppercase to lowercase letters for either words or nonwords. Feustel et a1. (1983) argue that these data indicate that some more abstract code than a direct physical copy of the item is being accessed since the typeface change does not 43 alter the repetition effect. Considered in isolation, the notion of an abstract code is in keeping with the predictions of a logogen model of semantic memory. However, because this effect is obtained for nonwords as well as words, Feustel et al. argue that facilitation due to repetition is a consequence of accessing unique episodic traces, and therefore argue against the standard logogen model explanation of repetition effects as a consequence of the activation of a unitized representation in semantic memory. To reconcile Feustel et al.'s (1983) data with existing logogen models, one would have to accept the idea that a single presentation of a nonword is sufficient to result in the formation of a logogen. This might be possible even though it contradicts the notion that semantic memory representations are the product of numerous exposures to an item, but Feustel et a1. argue that other aspects of their data fail to support such an explanation-—for example, they obtained an interaction between task and lexicality, wherein the first task was more difficult than the second for words, and the second task was more difficult than the first for nonwords, suggesting that some aspect of processing was different for words and nonwords. If nonwords had established a logogen that operated identically to that of words, then one would not expect this difference. Feustel et al. therefore argue that the repetition effect is due to 44 episodic memory traces. Of course, one might instead argue that this difference could be seen as a function of some as yet undetermined aspect of a newly established logogen, such as strength, number of connections to other logogens, or susceptibility to interference, rather than a consequence of the absence of a logogen for a nonword. Although Feustel et a1. (1983) argue that the repetition effect is due to episodic information, they do not completely abandon the notion of an episodic/semantic distinction. They also found a consistent advantage for words compared to nonwords, and to explain this advantage, they suggest that there are unitized representations available for words that facilitate the automatic identification of near-threshold information. This argument preserves a weak form of episodic/semantic distinction. Feustel et a1. (1983) also compared perceptual recognition in these tasks with recognition memory in an attempt to determine whether these tasks would show independent effects of various manipulations. When items were presented that differed in one or two letters from preceding items, false alarms in recognition judgments increased and identification latencies in perceptual‘ recognition decreased, indicating some common mechanisms between the tasks. This is consistent with their interpretation of the repetition effect as resulting from unique episodic traces. However, some independence was 45 also found; recognition memory in this task showed a significant decrement due to the lag between preceding presentations and the test item, whereas the identification measure showed no such decrement. If identification requires access to the same representations as recognition, this apparent independence is puzzling. Feustel et al. suggest that this independence was due to differences in the cues that could be used to access the episodic representation for identification vs. recognition judgments: The contextual cues that are used in recognition change over time, whereas the cues that yield identification are temporally invariant features such as the physical characteristics of the stimulus. The change of the recognition cues means that a lag between presentations will result in a decrement, whereas the same will not be true for unchanging identification cues. One might object to this line of reasoning by noting that the real world physical characteristics of stimuli such as words do change to some extent with changes in typeface, etc..., but such changes still do not affect identification according to their own data. Feustel et a1. (1983) argue that the overall pattern of results they obtained demonstrates that the repetition effect in perceptual identification is the result of accessing previously established episodic representations, thus concurring with Jacoby and Witherspoon (1982), and Jacoby and Brooks (1984). However, they also argue that 46 their results indicate the existence of some form of logogen or semantic code for words, thus maintaining a weak for of episodic/semantic distinction. A model developed to account for their results indicates that perceptual recognition could result from a combination of system responses to both episodic and semantic traces, either in a race between the two or in some form of synergistic interaction. A somewhat different form of perceptual recognition study has looked at differences in priming effects in word fragment completion (e.g., _e_d_l_m, etc...) and recognition memory. In such studies, word fragment completion is viewed as another example of a perceptual recognition task, although this may be problematic since word fragment completion logically involves a very high degree of attended strategic processing that may not be needed in normal perceptual recognition. Tulving, Schacter, & Stark (1982) had subjects study a list of words, and then administered a recognition test and a word completion test. Word completion benefitted significantly from previous exposure to the word in the experiment, even after an interval of seven days. However, the priming effects for word fragments were essentially uncorrelated with performance on the recognition memory task; that is, the likelihood of a word being completed following previous exposure was independent of whether it could be recognized. In addressing these results, Tulving (1983) 47 speculates that yet another memory system, different from both episodic and semantic memory, may support priming effects in perceptual recognition. He notes that the information causing the facilitation in perceptual recognition is not episodic, since it does not facilitate recognition, and yet the persistence of the effect contradicts most current versions of semantic memory functioning, and therefore speculates that these kinds of priming effects may be due to activation in a distinct procedural system. Graf, Squire, and Mandler (1984) report that amnesics perform as well as normals in the word completion task, while showing significant deficits on free recall, recognition, and cued recall, thus suggesting that perceptual recognition requires processes that are independent of those in recognition memory. Graf et al. also argue that perceptual recognition is based on procedural memory, and interpret their findings as demonstrating that procedural learning skills are spared in amnesics. Sentence Recognition and Verification Shoben, Wescourt, and Smith (1978) reported results supporting the episodic/semantic distinction based on studies of sentence verification, wherein subjects judged the truthfulness of simple sentences (e.g., A tiger has stripes), and of sentence recognition. Within each task, they varied the semantic relatedness of the subject and Table 2 48 Summary of Perceptual Recognition and Recoggition Studies Study Jacoby & Dallas (1981) Jacoby & Witherspoon (1982) Jacoby & Brooks (1984) Tulving et al. (1982) Graf et a1. (1984) Feustel et al. (1984) Conclusions Depth of processing affected recognition memory but not perceptual recognition. Repetition aided perceptual recognition and recognition memory. A change in typeface hindered perceptual recognition and recognition memory Prior presentation aided perceptual recognition even after 24 hours. Presentation of homophones in context affected subsequent spelling by normals and amnesics. Changing the typeface eliminated the spelling effect. Features of stimuli attended during encoding affect perceptual recognition tasks to the extent that the tasks require knowledge of those features. Word fragment completion benefitted from prior exposure even after 7 days. Performance in fragment completion is not correlated with recognition memory. Study aids fragment completion, but not recall or recognition in amnesics. Repetition benefits perceptual recognition of both words and nonwords. The benefit from repetition is not eliminated by a change in typeface. 49 predicate of the sentences, and the number of propositions learned about any particular concept (fanning-~e.g., A tiger has stripes. A tiger has thighs. A tiger has ears, etc...). Semantic relatedness would be expected to affect a semantic memory task such as sentence verification on a priori grounds, but not necessarily an episodic one, whereas fanning would be expected to influence an episodic task, but not semantic memory, according to most versions of the episodic/semantic distinction. In the sentence verification task, Shoben et al. (1978) found that subject-predicate semantic relatedness resulted in faster response latencies, but fanning had no effect. In the recognition task, they found an effect due to fanning when the fanning manipulation was applied to both the subject and predicate terms, but none due to semantic relatedness. These results were slightly compromised by a weak effect of semantic relatedness at the first appearance of any particular subject term in the sentences, but, overall, their results supported the existence of independent episodic and semantic memory systems. Shoben et a1. suggested that semantic representations might be viewed as relatively decomposed, consisting of the meaning components of individual terms, whereas episodic memory representations might have a more "surface-like format". This is consistent with other arguments that semantic memory involves more abstract codes than episodic memory (e.g., Feustel et al., 1983; 50 Tulving, 1983). However, other experiments using sentence verification and recognition paradigms have arrived at the opposite conclusion. McCloskey and Santee (1981) argued that Shoben et al.'s (1978) interpretation of the presence of fanning effects in recognition and their absence in verification as support for an episodic/semantic distinction relied on certain assumptions about the nature of the search through associative memory that may not be true. Specifically, they argued that if one assumed a parallel rather than serial search through memory, the effects in Shoben et al.'s study could be explained without recourse to the idea of different domains in memory. McCloskey and Santee (1981) further argued that the relatedness effect in verification, but not recognition, obtained by Shoben et a1. (1978) can be understood by postulating that verification, in comparison to recognition, requires an additional computational step of feature matching that uses semantic information. They reported evidence to this effect: Rejection of new sentences in a recognition task was slower when the predicate was semantically related to the predicate in the study sentence than when the predicates were semantically unrelated--having seen "The gambler as a sword", subjects were slower to reject "The gambler has a dagger" than "The gambler has a dog". McCloskey and Santee (1981) argued 51 that this demonstrated that semantic information was preserved in an episodic trace, and therefore concluded that episodic and semantic memory were not functionally distinct systems. Anderson and Ross (1980) also used a sentence vefification task to produce evidence against an episodic/semantic distinction. They had subjects study sentences containing various classes of assertions regarding a concept, and then make true/false judgments for these category relationships embodied in other sentences. Their results provided weak evidence of effects of episodic study on the semantic task: Learning additional information about a concept increased response latencies for sentences about that same concept in some instances, and learning spurious information (e.g., A tiger has wings) about a concept increased the error rate. Thus information was apparently transferred from episodic memory to semantic memory, which, they claim, is in contradiction to the predictions of the episodic/semantic distinction. Recognition Failure of Recallable Words As noted earlier with respect to studies of priming, comparisons of the operating characteristics of the epsiodic and semantic memory systems also may be used as evidence about the episodic/semantic distinction. Finding different operating characteristics supports the distinction, and finding similar operating characteristics 52 tends to weigh against it. Muter (1978) devised such a test in which he sought to determine whether the rate of recognition failure of recallable items in semantic memory would correspond to the rate established for episodic memory. Subjects were first asked to identify famous surnames in a list of names (recognition test), and then asked to recall the surname after being given information about the character and the first, or first and middle, name (e.g., U.S. author of Call of the Wild, and The Sea ngf, Jack ). Subjects failed to recognize recallable surnames in this task at the same predictable rate found for other stimuli in episodic memory tasks (Tulving & Wiseman, 1975). Muter therefore argued that the concept of a unique episodic trace was not necessary to explain recognition failure of recallable words in episodic memory, since the same thing occurred in semantic memory, and that it would be more parsimonious to abandon any qualitative distinction between episodic and semantic traces. Neely and Payne (1983) recently replicated Muter's (1978) research using a more rigorous design. They argue that only direct comparisons of variables' effects on episodic and semantic memory performance in experiments that equate all variables other than the instructions necessary to specify the type of judgment can reveal useful information about the episodic/semantic distinction. They devised a variation of Muter's (1978) 53 experiment that met this stipulation. Subjects in both the episodic and semantic task conditions studied a list containing 24 word pairs, 12 famous first name-surname pairs, and 12 nonfamous first name-surname pairs. All subjects received, in order, an episodic recognition test for the word pairs, an episodic or semantic recognition test for the names (depending on the condition), an episodic cued recall test for the word pairs, and an episodic or a semantic cued recall test for the surnames. The results confirmed Muter's (1978) finding that the Tulving-Wiseman function (Tulving and Wiseman, 1975) accurately predicted recognition failure of famous names in a semantic test as well as recognition failure in an episodic test, and regardless of whether the names had been seen recently or not. These results mitigate against the episodic/semantic distinction, although Neely and Payne (1983) caution that even with separate systems, some parallels in operating characteristics might be expected, and suggest that other research holding conditions constant across tasks is needed to further evaluate the distinction. Semantic Activation and Episodic Storage Balota (1983) studied the effects of subthreshold semantic primes on lexical decisions and on subsequent recognition. In this experiment, some primes were presented for a brief amount of time and followed by a pattern mask to prevent conscious perception. First a it. 54 lexical decision task using subthreshold and suprathreshold semantic priming was given. A number of the targets in this task were homophones. In a subsequent recognition test, these target homophones were either semantically and episodically primed as a result of being paired with the same word that served as the prime in the lexical decision task, or just semantically primed by being paired with a word related to their alternate meaning not elicited by the prime in the lexical decision task. Significant facilitation was obtained from the suprathreshold and the subthreshold primes in the lexical decision task, in keeping with other reports (e.g. Marcel, 1983a, 1983b; Fowler, Slade, Wolford, & Tassinary, 1981). Subsequent recognition of homophones was faster when they were paired with the suprathreshold prime used in the lexical decision task than when they were paired with a prime related to their other meaning. However, the same effect was not obtained when they were paired with the subthreshold primes. Thus, for subthreshold primes, activation seemed to have taken place in semantic memory, as manifested in the facilitation in the lexical decision task, but it was not sufficient to result in storage in episodic memory. Summary and Discussion of the Empirical Evidence The results from empirical tests of the episodic/semantic distinction are indeed mixed. One finds instances of apparent effects of episodic information on 55 semantic judgments (McKoon & Ratcliff, 1979; Anderson & Ross, 1980; McCloskey & Santee, 1981; Feustel, Shiffrin, & Salasoo, 1983; Durgunoglu & Neely, 1985) and other instances in which no such effect is found (Shoben, Wescourt, & Smith, 1978; Carroll & Kirsner, 1982; Neely & Durgunoglu, 1985). One likewise finds instances of semantic information affecting episodic judgments (McKoon & Ratcliff, 1979; McCloskey and Santee, 1981; Feustel, Shiffrin, & Salasoo, 1983; Neely and Durgunoglu, 1985), and instances where it does not (Shoben, Wescourt, & Smith, 1978; Hermann & Harwood, 1980; Hermann & McLaughlin, 1973). These differences in results make the resolution of the episodic/semantic debate using the existing data base uncertain, although the majority of evidence might seem to argue against the existence of distinct memory systems. There are several probable sources of these discrepancies in findings. Neely and Payne's (1983) argument regarding the need for equivalence of conditions across episodic and semantic memory tests is well-founded, and the failure of most studies to consider this may account for some of the discrepancies. However, even the studies that conform to their stricture, such as their own, Carroll and Kirsner's (1982), and Neely and Durgunoglu's (1985), arrive at opposite conclusions regarding the episodic/semantic issue. More importantly, it seems that not all so—called 56 tests of the episodic/semantic distinction actually test that distinction, or, if they do, do so in a way that does not allow meaningful interpretation of their findings. For example, in priming studies it may be necessary to distinguish between semantic priming as found in lexical decision tasks, and the priming effects found in perceptual recognition tasks--as noted earlier, some researchers have suggested that perceptual identification, and the repetition effect in perceptual recognition, is supported by a procedural memory system or some other memory system (Graf et al., 1984; Tulving et al., 1982; but see also Jacoby & Brooks, 1984; Feustel et al., 1983), rather than being directly related to semantic memory. Thus, tests of perceptual recognition compared to recognition memory do not directly address the episodic/semantic distinction if this is true. Durgunoglu and Neely's (1985) intentional demonstration of automatic episodic priming of a semantic decision under circumstances that allowed the subjects to make the semantic decision based entirely on episodic information illustrates another way in which an apparent test of the episodic/semantic distinction may fail to actually test it. Further consideration of these studies also suggests that rigorous thought must be given to what constitutes proof or disproof of the episodic/semantic distinction even if the tasks used do tap into the appropriate 57 systems. First, it is often tacitly assumed that the episodic/semantic distinction must be absolute in the sense that no interaction between the systems should be found in a task. This seems patently foolish; for example, awareness of semantic information at the time of an episodic occurrence logically should result in its inclusion in the memory for the incident--an episodic encoding may be presumed to contain some meaning, and semantic priming of episodic decisions may reflect this. Second, the type of priming effect obtained from transfer studies often is not considered. Whether an effect is facilitatory or inhibitory should be an important consideration since facilitative priming effects potentially are more informative with respect to the organization of memory than inhibitory priming effects: Consider the case wherein one finds episodic priming producing inhibition in a semantic task. This inhibition could be a consequence of some interfering effect within the same memory system, but it could also reflect the disruption of retrieval or response processes that serve two distinct systems. To use a crude analogy, if a subject performing a semantic memory task was struck with a hammer, his response in the semantic task might be sufficiently disrupted so that semantic priming effects would be masked. Yet, no reasonable person would argue that this demonstrates that semantic memory and the hammer sensing part of the mind are the same. In the case 58 of episodic priming in a semantic task, the episodic prime may automatically activate its episodic memory component, as evidence suggests that it should (Ratcliff & McKoon, 1981; Neely & Durgunoglu, 1985), and trigger a recognition response. This recognition respone in turn might impede retrieval from semantic memory even if episodic and semantic traces do not coexist within the same memory system. Third, if semantic priming is found in an episodic task, it does not necessarily indicate that semantic activation primed the recognition of the target item by activating the existing episodic trace; one could instead suggest that semantic priming facilitated the initial perceptual encoding of the word, after which a match was made between the product of this encoding process and the episodic trace containing the word. Thus, semantic facilitation might be found even with a distinct episodic memory. Or, given a long prime-target SOA, conscious expectations based on a semantic prime could lead subjects to prepare to respond to related items in episodic memory. The point is that without some way of distinguishing between the various possible sources of a priming effect, the demonstration of semantic priming in recognition does not constitute strong evidence against the episodic/semantic distinction. Fortunately, these arguments are not as easily applied to the situation wherein one tries to explain 59 episodic facilitation of semantic decisions. If a very long SOA is used, and subjects can consciously generate some expectation about a target based on awareness of episodic associations, then episodic priming effects in a semantic task can be explained without damaging the credibility of the episodic/semantic distinction. However, using short prime-target SOA's or other techniques to avoid conscious search processes in a truly semantic task would allow one to overcome this problem. One can therefore predict that under such conditions, if the tasks used actually require semantic information, one should not find any evidence of facilitation due to episodic priming of the semantic decision if the episodic/semantic distinction is valid. OVERVIEW OF THE CURRENT STUDY The current study addressed the episodic/semantic distinction in light of some of the issues raised above. In particular, it addressed the question of whether conscious or strategic mediation is required to obtain episodic priming of semantic judgments, and vice versa. This was done using a procedure roughly analogous to that of McKoon and Ratcliff (l979)--priming effects due to episodic and semantic relationships were examined in lexical decision and item recognition judgments following memorization of short study lists. However, the current study differed from McKoon and Ratcliff's in several important respects. First, the ISI/SOA confusion found in that study was avoided. A comparison of priming effects obtained with short and long prime-target SOAs was included in the current study to allow a time course investigation of the strategic and automatic components of priming effects obtained under these conditions (Neely, 1977; Posner & Snyder, 1975a, 1975b). Second, an additional assessment of the source of priming effects in these tasks was obtained by presenting some primes below the threshold required for conscious detection. It was argued previously that various tests of the episodic/semantic distinction potentially were confounded by subjects' active efforts to relate episodic 60 61 and semantic information. One ideal way of eliminating that potential confound is to bypass subjects' conscious strategies altogether by eliminating their conscious knowledge about the prime. Although the idea is counterintuitive, several studies suggest that primes presented below the threshold required for conscious awareness can produce significant effects on judgments for subsequent items. Specifically, a number of studies have reported that pattern-masked words, presented under conditions that preclude conscious perception of them, facilitate semantic judgments for related words (Balota, 1983; Dagenbach & Carr, 1982, 1984; Fowler et al., 1981; Marcel, 1983a, 1983b). Thus, "nurse" might facilitate a lexical decision for "doctor" even when no conscious information about "nurse" was available. Presumably, any priming effects obtained from such subthreshold primes would be free of strategic considerations, making this a promising technique for investigating questions such as whether memory systems are functionally distinct. The promise of this technique is partly offset by an array of problems having to do determining whether a prime is really subthreshold. Critics of this research have suggested that the primes in the above studies were not truly subthreshold; that is, that subjects were capable of extracting some consciously available information about the primes, such as might be manifested by higher than chance performance on a forced-choice recognition task 62 following presentation of the subthreshold item, even though the subjects claimed to have seen nothing (Merikle, 1982; Cheesman & Merikle, in press). Cheesman and Merikle (in press) have elaborated this notion and suggest that there is a "subjective" threshold at which subjects claim to have no knowledge about a pattern-masked word, and at which one can obtain semantic priming effects that are qualitatively different from those at_full threshold, and a more stringent "objective" threshold at which subjects have no conscious knowledge at all about the prime, as measured by performance on a forced-choice recognition task, and at which all semantic priming effects are eliminated. The debate over this is ongoing. In the present study, it is likely that the primes were, in fact, presented below the subjective threshold, but above the objective threshold. Rather than representing a problem, one might argue that this is the perfect procedure for presenting primes in a study of the distinctiveness of memory systems: Semantic priming effects exist in this region, but subjects are not consciously aware of the primes and therefore cannot intentionally use their knowledge of the prime to prepare for the target. Jointly using a prime threshold and a prime-target SOA manipulation also provided an opportunity to replicate earlier reports indicating that these might not produce parallel effects. Theoretically, subthreshold priming effects should reflect the operation of automatic 63 processing, but Fowler et a1. (1981) and Balota (1983) both report that facilitation from subthreshold primes is found only at long SOAs. These reports suggest that subthreshold priming may involve a different set of processes than those responsible for facilitation from visible primes at short prime-target SOAs. Finally, the current study included a comparison of overlearned vs. newly learned episodic associations. This manipulation addressed the issue of whether any apparent differences between episodic and semantic priming obtained might be eliminated by increasing the strength of the relationship between episodically associated items. Many proposed differences between episodic and semantic memory appear to hinge on characteristics that one would logically expect to be affected by the strength of association, such as automaticity of access. Given the standard argument that semantic memory representations and associations between items reflect multiple prior occurrences, one might expect to find differences of this nature in priming studies. The manipulation of the strength of association between episodic items in the present study was included to allow some insight into whether this was the only source of any obtained differences between episodic and semantic priming. Subjects were tested in two sessions. During the first session, they were given a set of materials to memorize to an overlearned level. In the second session, 64 they were assigned to either the lexical decision or episodic recognition task, and then received short study lists of word-word and word-nonword pairs to memorize for a possible cued recall test. Some pairs in these short study lists were from the previously memorized set and some were new. Each study list was followed by a series of the appropriate lexical decision or item recognition judgments, wherein subjects indicated whether the target was a real word, or whether it had been present in the preceding study list. Studied word targets were primed by episodically related, semantically related, episodically and semantically related, or unrelated words, and studied nonword targets were primed by episodically related or unrelated words. The short study lists and test trials were administered in four blocks reflecting the conditions under which the prime was presented: 1) Suprathreshold prime with a long prime-target SOA, 2) subthreshold prime with a long prime-target SOA, 3) suprathreshold prime with a short prime-target SOA, and 4) subthreshold prime with a short prime-target SOA. The order of the prime presentation conditions was counterbalanced between subjects using a latin square design. Theoretical Predictions A number of predictions regarding the pattern of priming effects that might be obtained in each of the above conditions were generated prior to the experiment. The predictions were based on considerations of what 65 patterns of between system priming (episodic priming of semantic judgments, and semantic priming of episodic judgments) one might expect to find if episodic and semantic memory were distinct, and strategic factors were responsible for prior demonstrations of such priming. Priming of Lexical Decisions At the long prime-target SOA with suprathreshold primes, one would predict that episodic, semantic, and episodic and semantic priming would facilitate lexical decisions. Episodic priming's effectiveness would be contingent on having sufficient time for the strategic component of priming to function. Deviations from this pattern would be surprising, although Neely and Durgunoglu‘s (1985) research suggests that episodic study might interfere with the semantic priming. At the short prime-target SOA with suprathreshold primes, one would expect semantic priming, but not episodic. If episodic priming occurred for newly learned associations, it would be hard to reconcile with the notion of separate memory systems. If episodic priming of only overlearned associations occurred, it would suggest that obtaining priming at short SOAs is partly a function of the strength of association, and fit nicely into a model wherein the difference between episodic and semantic memory is due to semantically associated items having occurred together sufficiently often in the past to allow automatic activation of one following activation of the 66 other. Finding no effects from overlearned episodic associations would suggest that either the degree of overlearning was insufficient, or that entry of information into semantic memory requires a different set of processes than those used in the current experiment's study condition. Predictions about the effects of subthreshold priming must be more cautious given the disputed nature of the effect in the first place. To the extent that such effects occur, they should reflect semantic priming rather than episodic priming. Previous reports (Balota, 1983; Fowler et al., 1981) suggest that they might be found only at the long SOA. Because it is still not clear what factors cause subthreshold priming to be found in some cases, and not others, or to produce inhibition in some cases rather than facilitation (Dagenbach & Carr, 1985), and because episodic study is present, which interferes with even suprathreshold semantic priming in some cases, it seems reasonable to expect a less clean pattern of data from this procedure. To the extent that priming effects are found, they should be from semantic rather than episodic relations if the episodic/semantic distinction is valid. Priming of Recognition Priming from episodic and semantic relationships might be expected with suprathreshold primes at the long prime-target SOA since there is sufficient time for 67 strategic mediation. Facilitation from suprathreshold episodic primes at the short SOA would replicate Ratcliff and McKoon's (1981) finding of automatic episodic priming in an episodic task. Facilitation from semantic priming might indicate either an interaction between the systems, or some effect on perceptual encoding. While it therefore would not be a decisive test, it would weigh against the episodic/semantic distinction. The cautions about subthreshold priming of lexical decisions also apply to subthresold priming of recognition judgments. In the case of recognition, subthreshold facilitation would represent a novel and perhaps surprising finding. If from episodic associations, it might indicate parallel operating characteristics in episodic and semantic memory. If from semantic associations, it would tend to weigh against the episodic/semantic distinction, although it again might be the case that the semantic prime's effect was on perceptual encoding, and therfore this would not be a critical test. METHOD Subjects The subjects were 64 undergraduates recruited from introductory psychology classes at Michigan State University who received course credit in return for their participation. All subjects were required to have normal or corrected-to—normal vision and hearing, and to be native English speakers. Subjects were alternately assigned to the episodic recognition or the lexical decision task. Stimulus Materials The same materials were used in both the episodic recognition and lexical decision tasks. A set of stimuli sufficient for use in one prime presentation condition (i.e., suprathreshold long SOA, suprathreshold short SOA, subthreshold long SOA, or subthreshold short SOA) was generated initially. This set contained 16 study lists consisting of word-word and word-nonword pairs, and 16 test lists consisting of word primes and word and nonword targets. The study lists each contained five pairs of stimuli, and the test lists each contained 6 prime-target pairs. Prime-Target Relatedness and Control Conditions The study was designed so as to contain six word target conditions and two nonword target conditions of special interest with respect to the episodic/semantic 68 69 distinction, and four word target and three nonword target control conditions (see Table 3). The word target conditions of interest with respect to the episodic/semantic distinction were those in which the prime had an episodic or semantic relationship to the target in the test set (conditions 1-6 in Table 3): episodic related, episodic unrelated, semantic related, semantic unrelated, episodic and semantic related, and episodic and semantic unrelated. In the episodic related, episodic unrelated, semantic related, and semantic unrelated conditions, the subject studied a pair of semantically unrelated words in the study list (coat—pepper). The second member of that pair then served as a target during the subsequent test list. It was primed by the same word it had been paired with in the study list (coat-pepper) in the episodic related condition, by a different, but semantically related, word (salt-pepper) in the semantic related condition, and by a different unrelated word (rain-pepper) in the episodic unrelated and semantic unrelated conditions. In the episodic and semantic related, and episodic and semantic unrelated conditions, a pair of semantically related words (bread-butter) appeared in the study list. Again, the second member of that pair served as the target for the subsequent recognition or lexical decision judgments. It was primed by the first member of the study list pair in the episodic and semantic related condition 70 Table 3 Prime-Target Relationships Prime-Target Relatedness Conditions: Word Targets Condition Study List Test List 1. episodic related coat pepper coat pepper 2. episodic unrelated wine queen rain queen 3. semantic related day woman man woman 4. semantic unrelated jail light school light 5. episodic and semantic bread butter bread butter related 6. episodic and semantic blossom flower black flower unrelated Prime-Target Relatedness Conditions: Nonword Targets Condition Study List 7. episodic related 8. episodic unrelated purse lince model vade Test List purse lince gate vade Control Conditions: Word Targets Condition Study list Test List 9. jail light jail view 10. camera gretch camera short 11. city town city score 12. color blue Control Conditions: Nonword Targets Condition Study List Test List 13. jail light jail spang 14. guide masset guide wabby 15. doctor nurse doctor frim 71 (bread-butter), and by a different unrelated word (blossom-butter) in the episodic and semantic unrelated condition. For the two nonword target conditions pertinent to the episodic/semantic distinction (conditions 7-8 in Table 3), the subject studied a word-nonword pair in the study list (purse-lince), and then the nonword served as the target in the subsequent test trial. It was primed by the first member of the study list pair (purse-lince) in the episodic related condition, and by a different word (gate-lince) in the episodic unrelated condition. The control conditions provided the lures for the recognition task. These conditions also appeared in the lexical decision task to ensure the equivalence of tasks in all respects other than the decision that was required. In these conditions, the subject studied an unrelated word pair, and then the first member of that pair primed a new word (condition 9) or nonword (condition 13) target that was not in the study list. Or the subject studied a word-nonword pair, and then the first member of that pair primed a new word (condition 10) or nonword (condition 14) target in the test list. Or the subject studied a semantically related pair of words in the study list, and the first word from that pair primed a new unrelated word (condition 11) or nonword (condition 15) target in the test list. Finally, there was a control condition in which a 72 pair of semantically related words appeared in the test list even though neither word had appeared in the study list (condition 12 in Table 3). This condition was included in order to increase the proportion of items that were semantically primed in the test lists. A further subdivision of the materials in the conditions described in Table 3 was made on the basis of whether the study list pairs were overlearned or newly learned. In the case of overlearned material, the subject saw the study list pairs during the first phase of the experiment, wherein an attempt was made to have the subject form extremely strong episodic bonds between stimuli through studying them extensively. In contrast, the subject saw study list pairs for newly learned material only in the study lists of the second phase, and presumably formed weaker episodic associations. One half of the items representing each condition in Table 3, except condition 12, were from the overlearned material, and one half were from the newly learned material. The assignment of materials to the overlearned and newly learned conditions was counterbalanced between subjects. Conditions 1-8 in Table 3 were represented four times at each learning level within a given prime presentation block of trials. Control conditions 9-11 and 13-15 each were represented two times at each learning level within the prime presentation condition. Finally, condition 12, in which neither the prime nor target was studied before 73 the test list, was represented eight times within the prime presentation condition. Study List Construction To generate the items used in the study lists for conditions 1-6 at one learning level, 24 highly associated word pairs from the Palermo and Jenkins (1964) norms were selected. The words in these pairs were of medium to high frequency and ranged in length from 3‘to 8 letters. The first item of each selected pair was designated as the semantically related prime (salt), and the second item was designated as the target (pepper). A semantically unrelated prime (coat) of approximately equal length and frequency as the semantically related prime was then selected from the Kucera and Francis (1967) norms. This yielded 24 triplets consisting of the semantic unrelated item (coat), the semantic related prime (salt) and the target (pepper). Then four triplets were randomly assigned to each one of conditions 1-6 in Table 3. To generate the items used in the episodic related and episodic unrelated conditions with nonword targets (conditions 7 and 8), four pairs of unrelated words of approximately equal length and frequency as those in the preceding conditions were selected, and then letters were substituted in the second word of each pair to produce 8 pronouncable nonwords. Each pair appeared once in both the episodic related and episodic unrelated nonword conditions within a prime presentation condition, so that 74 a subject might see purse-lince in a study list, and then see lince primed by purse in the subsequent test list, and later would see purse-lince in another study list, but with lince primed by a different word in the test list. The items representing the control conditions in the study lists were derived from those generated for conditions 1-8 in one of two ways. First, in some cases the study list items for the control conditions were repetitions of those used in the unrelated conditions 6 and 8. For example, the study list items in control conditions 11 and 15, in which the subject studied semantically related words, were repetitions of the pairs used to represent the episodic and semantic unrelated condition (condition 6) in other study lists. The study list items for control conditions 10 and 14, in which the subject studied a word-nonword pair, were repetitions of the pairs representing the episodic unrelated nonword target condition (condition 8) in other study lists. In the other control conditions, a single presentation of the study list pair for the semantic unrelated condition served as the study material for the control condition as well as for that condition. This was the case for conditions 9 and 13, in which the subject studied an unrelated word pair, and then the first member of that pair primed a new word (condition 9) or nonword (condition 13). Thus, the subject studied a semantically unrelated word pair (day—woman), and in the subsequent 75 test list, the second member of that pair was primed by a semantically related word (condition 3: man-woman), and in the same test list, the first member of that pair primed a new word (condition 9: day-view) or nonword (condition 13: day-spang). The same procedures were then used again to generate a second set of materials to be used at the other level of learning for each of the conditions described above. As noted earlier, the materials assigned to the overlearned and newly learned levels were counterbalanced between subjects. The items from both learning levels were then randomly assigned to the study lists with three constraints: The first constraint had to do with the placement of pairs from different conditions within each five pair study list: The items of most interest, the prime-target relatedness conditions with word targets (conditions 1-6 in Table 3), were assigned to the second, third, or fourth position in these lists to lessen primacy and recency effects. The other pairs were assigned to the first and last positions in the study lists. The second constraint had to do with the order of appearance of items: First, the materials were assigned to the study lists so that one representative of each condition at each learning level appeared before a second representative of any condition appeared, except for condition 12, which appeared twice. Finally, for pairs that appeared more 76 than once in a prime presentation condition block, such as the word-nonword pairs in conditions 7 and 8, the second appearance of the pair occurred at least 4 study lists away from its previous occurence. An example of a complete set of study lists is available in Appendix A. Test List Construction The identity of most test list primes and targets was determined by the constraints of the conditions in the study lists. Thus, if the study list contained materials representing the episodic related, semantic related, or episodic and semantic related word target conditions, or the episodic related nonword target condition, the target in the test list was the second item in the study pair and the prime was the appropriate related word for that particular condition, as already described. For the corresponding unrelated conditions, the target item was again the second member of the pair. The primes for these unrelated conditions were obtained by using the first member from a study list pair of stimuli assigned to the same unrelated condition, but appearing elsewhere. Thus, the prime in the episodic unrelated condition was the first word from another pair of stimuli assigned to the episodic unrelated condition appearing in a different study list, and the prime for the semantic unrelated condition was the first word from a different study list pair assigned to the semantic unrelated condition. 77 Selection of items for primes in the unrelated conditions followed the constraint that the study list in which they appeared was at least 4 lists away from the test list in which they were the prime. For most control conditions (conditions 9-11 and 13-15 in Table 3), the prime was the first word from the study list pair. The word or nonword targets were generated by selecting words of approximately equal length and frequency as those in other parts of the experiment, and substituting letters in them to form pronouncable nonwords where appropriate. For the final control condition (condition 12 in Table 3), 8 more pairs of semantically associated words were selected. These pairs appeared only in the test sequences, and not in the study lists. The test list sequences following each study list consisted of the materials appropriate for the 5 conditions embedded in the study list and for 1 additional control condition. These were randomly assigned to the six positions in each test sequence. This arrangement of materials meant that within each prime presentation condition in the recognition task, 64 items required "yes" responses, indicating that they had appeared in the preceding study list, and 32 required "no" responses. Correspondingly, in the lexical decision task, 68 of the 96 trials contained word targets requiring "yes" responses, and 28 required "no" responses. The prime was 78 a valid clue (either episodic, or semantic, or both), on one third of the trials in both tasks. An example of a complete set of test lists also is available in Appendix A. Materials in Subsequent Conditions The above procedure provided the materials for one prime presentation condition. The same materials were reused in three subsequent prime presentation condition blocks following appropriate reassignment of the materials to different prime-target relatedness and control conditions. To generate the second set of materials, items in the episodic related word target condition were reassigned to the semantic related word target condition, and vice versa. Similarly, those in the episodic unrelated word target condition were reassigned to the semantic unrelated word target condition, and vice versa. Items in the episodic and semantic related condition were reassigned to the episodic and semantic unrelated condition, and vice versa, and those in the episodic related nonword target condition were reassigned to the episodic unrelated nonword target condition, and vice. versa. The control conditions were then generated using the same procedures as before, and new study list and test lists were generated as before. To generate the third set of materials, partial reassignment of the materials in the first set was used, and then the materials in the study lists and test lists 79 were presented in reverse order. In this case, items in the episodic related word target condition in the first set of materials were reassigned to the episodic unrelated word target condition, and vice versa, and those in the semantic related word target condition were reassigned to the semantic unrelated word target condition, and vice versa. This procedure was followed instead of using complete rearrangement to economize on the number of slides necessary for the experiment and to lessen the amount of time and memorization required of subjects. In these cases, the study lists were as before, but appeared in reverse order. The primes preceding the target items were changed to fit the demands of the prime relatedness conditions where appropriate, using the same protocol as before. In all of these reassignments, materials in the overlearned condition for any subject remained in that condition, and those in the newly learned condition remained in that condition. The fourth set of materials was then generated by applying this same procedure to those in the second prime presentation condition. The entire set of materials thus generated is displayed in Appendix A. The assignment of these four sets of materials to the prime presentation condition blocks was counterbalanced between subjects, so that each set of material appeared in each prime presentation condition an equal number of times. 80 Pattern Mask A pattern mask was used in subthreshold priming conditions that consisted of partially overlapping letters in the same typeface as the other stimuli arranged in random orientations inside a rectangular space. Procedures Session I In the initial phase, subjects memorized a small set of stimulus pairs in order to form overlearned episodic associations for use in the subsequent session. They were given a list containing 4 word-nonword pairs, 16 unrelated word-word pairs, and 8 semantically related word-word pairs that would serve as the overlearned condition materials in the study lists of the second part of the experiment. They were instructed to memorize these pairs until they could correctly produce the second item (target) of each pair following presentation of the first item (cue) for the entire list 10 times in succession without error. The study packets provided included randomized lists of cues to be used in the recall task, and subjects were required to write down their responses. Memorization and completion of 10 successive lists without an error required approximately 35 mins. on the average. Most subjects made a mistake on the first attempt, and responded correctly thereafter. No subject required more than 14 attempts to fulfill the memorization requirement. Due to scheduling constraints, six subjects proceeded 81 to the next phase of the experiment immediately following the first part. The remainder were scheduled to return 1-3 days later. Those subjects who returned at a later date were given a copy of the study list to review just prior to arriving at the lab. Session II In this part of the experiment, subjects were alternately assigned to either the lexical decision or episodic recognition task. Each subject received a series of short study lists that were to be memorized, with each study list followed by a test list consisting of primes not requiring a response and target items requiring the appropriate judgment (recognition or lexical decision). The trials were blocked according to the 4 prime presentation conditions varied within subjects (suprathreshold long SOA, suprathreshold short SOA, subthreshold long SOA and subthreshold short SOA-~these conditions are described in detail below). The order in which the prime presentation conditions were presented to each subject was counterbalanced between subjects using a latin square design, with each of the four orders administered to eight subjects. Subjects were tested individually in a very dimly illuminated room, and received a 30 minute rest break between the second and third conditions. Prior to each of the prime presentation condition blocks of trials, a retest for the overlearned material 82 from the first session was administered. The subject was required to respond correctly to each item in two cued recall lists analogous to those administered in the first session. No subject required more than 3 attempts to correctly remember all pairs on two lists in succession. Within each prime presentation condition, the subject saw 16 study lists and 16 test lists. Each study list was followed by a test list consisting of six of the appropriate test trials (lexical decision or recognition) in which a prime requiring no response appeared prior to each target. The subject was instructed to memorize the study lists for a possible cued recall test that actually occurred on approximately one third of the trials, and then to respond in the subsequent test trials (lexical decision or recognition) as quickly as possible while remaining accurate. The response was made by pressing one of two response keys located directly in front of the subject. Subjects assigned to the lexical decision task were instructed to indicate whether or not the target item was a real English word, and subjects assigned to the item recognition task were instructed to indicate whether or not the target item had appeared in the immediately preceding study list. It was emphasized that the decision was to be made for the target item only, and not the combined prime-target pair. The subject was instructed to read the prime when possible, and to try to use it to aid 83 in responding as quickly as possible. The various prime-target relationships possible were then illustrated, followed by actual testing. The prime's exposure duration and the interval between the prime and the target varied with the prime presentation condition, as described below. Prime Presentation Conditions Suprathreshold Long SOA. In this condition, the prime was shown for 500 msec, followed by a dark field for 1500 msec, and then the target for 500 msec. Both the prime and the target were clearly visible. Subthreshold longiSOA. In this condition, a detection threshold for seeing briefly presented pattern masked words was determined prior to the study list and test trial presentations. This threshold was defined as the level at which the subject's ability to distinguish between real words and blank fields preceding a pattern mask fell to chance. The lexical decision or item recognition trials were then administered with the prime presented at that threshold. A dark field followed the prime-mask presentation for 1900 msec, and then the target appeared for 500 msec. The threshold setting procedure worked as follows: The subject first determined his or her dominant eye by binocular and monocular alignment of the index finger with a visual stimulus. The subject was then given a pair of Polaroid III glasses to wear in which the lenses were 84 polarized filters. Other polarized filters placed in front of the projectors were then arranged in such a manner as to allow dichoptic presentation, with the initial presentation (word or blank field) seen only by the nondominant eye, and the pattern mask seen only by the dominant eye. The subject was instructed to indicate whether a word or a blank field had been present prior to the pattern mask on each trial. The subject was further told that the trials would be administered in sets of eight, and that within each set four words and four blank fields would be shown in a random order, and instructed to try to guess accordingly. The administration of the trials then commmenced. A word or a blank field was shown to the nondominant eye for 15 msec, followed by a blank field for 80 msec, and then a pattern mask was shown to the dominant eye for 50 msec. A dark field then followed for 1900 msec, and then the binocularly presented word "answer" appeared for 500 msec, at which time the subject gave a response. After 8 such trials, if the subject had gotten 5 or more correct, the interstimulus interval (ISI) between the initial word or blank field and the pattern mask was shortened and a new series of trials was begun. The ISI in this procedure was initially 80 msec, and was reduced in intervals to 60, S0, 40, 30, 20, 15, 10, 8, 6, 4, 2, 0 msec, stopping when the subject got 4 or fewer correct over 8 trials. If the subject was still getting 5 or more 85 correct at an ISI of 0, the presentation time of the initial slide containing the word or blank was shortened in intervals to 10, 8, 6, 4,, and 2 msec. This extra manipulation was required in several cases. On the other hand, if the subject had gotten 4 or fewer correct on the initial series of trials with the ISI of 80, the ISI was lengthened to 100, 120, and 140 msec until performance exceeded chance, and then decreased in the same intervals until performance fell to chance. When four or fewer trials were answered correctly (chance criterion) in a block of 8 trials, a more extensive test was given. A block of 20 trials using that ISI was administered. If 11 or fewer of these were correct, this was considered to be equal to chance, whereas if 12 or more were correct, the threshold setting procedure was reimplemented at one lower interval. After meeting the chance criterion on 20 presence-absence judgments, subjects then had to try to identify 20 words presented under the same conditions. They'were informed that a different word would be present on each trial, and instructed to try to identify each of them. They were required to guess at the word's identitity even if they claimed to have seen nothing at all. If a subject correctly identified any word, the preceding threshold setting procedures were reimplemented at a lower increment again until all the requirements for chance performance were met. 86 The study list and test trials were then administered with the prime and pattern mask in the test trials presented at the threshold determined above, followed by a blank field for 1900 msec, and then the binocularly presented target for 500 msec. The subject was instructed to pay attention to the initial flash containing the prime, and then to respond as quickly as possible to the target. Subjects were informed that there would be a prime present on each trial, and also asked to indicate if they felt that they actually saw the prime on any trial by naming it after their lexical decision or recognition judgment had been made. Following 96 lexical decision or recognition trials, a retest of the threshold using 20 presence—absence judgments was administered. Suprathreshold short SOA. In this condition, the prime was presented for 100 msec, followed by a 100 msec blank interval, and then the target item presented for 500 msec. Subthreshold short SOA. In this condition, the prime presentation threshold was determined as in the subthreshold long SOA condition, except that the word "answer" in the detection trials appeared approximately 200 msec after the prime onset. The threshold thus established was used in the subsequent trials, wherein everything was the same as in the subthreshold long SOA condition except that the target appeared approximately 200 msec after the prime's onset. 87 Apparatus Stimuli were presented using a Gerbrands 3 channel projection tachistoscope, and response times were recorded to the nearest msec using a Gerbrands timer. The materials were projected onto a round screen designed to minimize the reflection of light from it that was located 8 feet in front of the subject. Slides containing stimuli were prepared by photographing typed words and nonwords against a white background. The study list materials were photographed in pairs so that the members of each pair appeared adjacent to one another on one slide and on the screen when projected. The test trial items (primes and targets) were photographed individually so that only one item appeared per slide. These items subtended less than 2 degrees of visual angle when projected onto the screen. In the subthreshold conditions, the Polaroid III system of filters and glasses for 3-dimensional presentation was used to obtain dichoptic presentation. Polarizing filters were placed in front of the projectors containing the priming stimuli and the pattern mask, and the subject was given Polaroid III filter glasses to wear. This procedure allowed presentation of the prime to one eye and the pattern mask to the other eye during the subthreshold priming conditions following the appropriate placement of the filters in front of the projectors. RESULTS AND DI SCUSSION The data from the lexical decision and recognition tasks were analyzed separately. The results from each task will be presented separately at first to aid in exposition. Lexical Decision Task Median response latencies were calculated for correct responses in the six word target and two nonword target conditions addressing the episodic/semantic distinction (episodic related, episodic unrelated, semantic related, semantic unrelated, episodic and semantic related, episodic and semantic unrelated in the case of words, and episodic related and episodic unrelated for nonwords). Response times greater than 2000 msec and less than 200 msec were deleted from these analyses, following McKoon and Ratcliff's (1979) procedure for removing extreme data points. The means of the median response latencies for word targets are presented in Table 4, and the means of the median response latencies for nonword targets are presented in Table 5. Because a specific set of orthogonal comparisons among the means in the priming conditions was inherent in the design of the experiment, planned comparisons were used to analyze those effects pertinent to the episodic/semantic distinction. Unless otherwise noted, these comparisons are between the related and unrelated 88 Table 4 Mean RT (msec) and Percent Error for Lexical Decisions: Word Targets Suprathreshold Long SOA Newly learned E 549 ( 8 ) s 575 (2.6) ES 518 (2.1) E 685 (8.8) E 605 (2.8) "E 586 ( 8 ) Net effect +56*(0.8) 30 (0.2) +68**(2.l) Overlearned E 517 (3.1) S 553 (0.8) ES 543 ( 0 ) E 588 (3.4) 8 606 (1.8) 55 622 (2.1) Net effect +7l**(0.3) +53**(1.0) +89**(2.l) Suprathreshold Short SOA Newly learned E 579 (8.3) S 542 (1.8) ES 522 (2.3) E 585 (3.1) 8'565 (1.6) 88 576 (3.1) Net effect +6 (5.2) +23* (0.2) +54**(0.8) Overlearned E 531 ( 8 ) s 533 (2.3) as 528 (2.3) E 561 (3.6) E 572 (1.6) ‘8 561 (2.6) Net effect +38 (3.6) +39**(0.7) +4l**(0.3) Related Conditions E Episodic S Semantic ES Episodic and Semantic * p. < .05 ** p. < .01 Episodic S'Semantic ES Episodic and Semantic Unrelated Conditions Table 4 (cont‘d) 90 Subthreshold Long SOA Newly learned E 633 (4.2) s 682 (6.5) 83 589 (5.9) E 596 (3.9) E 686 (3.4) 58 622 (1.6) Net effect -37* (8.3) +4 (3.1) +33 (4.3) Overlearned E 598 (3.1) s 563 (8.8) ES 685 (2.3) E 597 ( 8 ) § 688 (3.9) ES 685 (2.6) Net effect -1 (3.1) +37* (3.1) 0 (0.3) Subthreshold Short SOA Newly learned E 582 (3.6) s 592 (2.6) ES 572 (3.4) E 574 (2.8) s 598 (3.1) 88 612 (8.8) Net effect +8 (8.8) +2 (8.5) +40* (2.6) Overlearned E 576 (1.8) s 591 ( 8 ) as 604 (0.8) E 628 (1.6) s 574 (1.0) 88 576 (2.3) Net effect +44* (8.2) -17 (1.0) -28 (1.6) Related Conditions E Episodic S Semantic ES Episodic and Semantic * p. < .05 ** p. < .01 Unrelated Conditions Episodic ‘E S Semantic ES Episodic and Semantic 91 Table 5 Mean RT (msec) and Percent Error for Lexical Decisions: Nonword Targets Suprathreshold Long SOA Newly learned E 699 (10.2) Overlearned E 617 (15.1) E 674 ( 8.1) E 717 ( 8.1) Net effect -25 (2.1) +100**( 7.0) Suprathreshold Short SOA Newly learned E 727 (15.6) Overlearned E 741 (11.7) ml ml 681 (16.7) 688 (9.6) Net effect -46 ( 1.1) -53* (2.1) Subthreshold Long SOA Newly learned E 678 (18.4) Overlearned E 623 (15.1) MI E 686 (15.1) 697 (18.1) Net effect +8 ( 3.3) +74**( 5.0) Subthreshold Short SOA Newly learned E 702 (14.3) Overlearned E 733 (15.1) E 674 (12.4) E 711 (13.0) Net effect -28 ( 1.9) -22 ( 2.1) E Episodic Related E. Episodic Unrelated * p. < .05 ** p. < .01 92 members of a single prime type condition at a single learning level (for example, episodic related vs. episodic unrelated for words in the newly learned condition), and facilitation or inhibition refers to the difference in response times between the related and unrelated conditions. Also, unless otherwise noted, separate error terms for each comparison were used (Keppel, 1982, chapters 17 and 19). Similar comparisons based on error rates were performed; in these, the percent correct scores for each condition were determined, and then contrasted as above. The results of these comparisons will be described for each prime presentation condition separately to facilitate exposition. Suprathreshold Long SOA Word Targets Related suprathreshold primes separated from target words by a long time interval (2000 msec prime-target SOA) produced facilitation compared to unrelated primes in the same condition irrespective of prime type. For newly learned targets, episodic priming and combined episodic and semantic priming produced signficant facilitation (56 msec, F (1,31) = 6.77, p. < .02, and 68 msec, F (1,31) = 22.16, p. < .0001). Semantic priming produced 29 msec of facilitation, but this did not quite attain a conventional level of statistical significance (F (1,31) = 3.29, p. .08). For overlearned targets, all types of priming 93 produced significant facilitation. Specifically, episodic priming yielded 72 msec of facilitation, (F (1,31) = 17.16, p. < .001), semantic priming yielded 54 msec of facilitation (F (1,31) = 21.54, p. < .001), and combined episodic and semantic priming yielded 79 msec of facilitation (F (1,31) = 9.25, p. < .01). Error rates were generally low in this task, and no significant differences in accuracy between the related and unrelated conditions in either the newly learned or overlearned material were present. Thus, in this condition wherein the prime could be seen clearly and there was adequate time for strategic mediation following the prime, virtually all types of priming yielded significant facilitation. The failure of semantic priming for newly learned word targets to attain significance was somewhat surprising. It parallels Neely and Durgunoglu's (1985) failure to find semantic priming of lexical decisions following paired associate learning, but given the amount of facilitation obtained in the present study, in conjunction with the significant facilitation from semantic priming for overlearned targets, it seems more likely that this represents a type 2 error. More importantly for the episodic/semantic distinction, episodic primes in this semantic task produced very large facilitatory effects. This evidence in and of itself would seem to weigh against the 94 episodic/semantic distinction, but, as noted in the introduction to this study, priming effects obtained in this condition do not constitute a critical test because of the uncertainty over their source: Strategic factors plausibly could underlie the episodic priming effects obtained. Nonword Targets The comparison of response latencies for newly learned nonword targets showed insignificant inhibition due to episodic relatedness. Neely and Durgunoglu (1985), and Durgunoglu and Neely (1985) also found inhibition from episodic priming of nonword targets. This inhibition might be understood as reflecting some form of response competition: The nonword was familiar and appeared in conjunction with the same item it was paired with in the study list, and therefore might have automatically elicited an affirmative recognition response that conflicted with the negative response required in the lexical decision task. Feustel et a1. (1983) have suggested that something of this nature underlies slower response latencies for repeated presentations of nonwords in a lexical decision task. On the other hand, episodic relatedness between the prime and target produced substantial facilitation when the prime-target relationships were overlearned (100 msec, F (1,31) = 17.18, p. < .001). This indicates that the combination of intensive study and a long prime-target SOA 95 are sufficient to allow subjects to prepare the "no" response. In that sense, this condition might be thought of as analagous to Neely's (1977) attention-switch condition. Suprathreshold Short SOA Conclusions about the interrelatedness of episodic and semantic memory suggested by the above findings must be tempered by the results obtained from suprathreshold priming with a short prime-target SOA. In this condition, in which facilitatory priming effects from strategic processes presumably were eliminated, the pattern of priming results was considerably different. Word Targets Episodic priming of newly learned word targets produced only 6 msec of facilitation (n.s.), whereas semantic priming and combined episodic and semantic priming both yielded significant facilitation (23 msec, F (1,31) = 4.79, p. < .05, and 55 msec, F (1,31) = 9.11, p. < .01). No differences in accuracy between the conditions were significant. These findings support the arguments developed earlier suggesting that episodic priming in a semantic task might be eliminated by properly controlling for strategic mediation. Essentially the same pattern was present for overlearned materials, but with one interesting difference. Semantic priming, and combined episodic and semantic priming, again produced significant facilitation 96 (39 msec, F (1,31) = 15.54, p. < .001, and 41 msec, F (1,31) = 8.05, p. < .01, respectively). Episodic priming produced 29 msec of facilitation, but this failed to attain statistical significance (F (1,31) = 2.62, p. > .10). However, the analyses of error rates in this condition did reveal a significant facilitatory effect from episodic priming: Episodically primed targets were labelled as words more accurately than those primed by unrelated words (F (1,31) = 4.33, p. < .05). No other differences in accuracy were significant. Thus, the strong facilitatory effects of episodic priming within a semantic task obtained at long prime-target SOAs were eliminated for the newly learned material, presumably as a consequence of using a short prime-target SOA that precluded strategic priming effects. In the overlearned condition, episodic priming yielded nonsignificant facilitation of response latency and significant facilitation of accuracy. The facilitation from overlearned episodic primes might be seen as evidence that.part or all of the difference between episodic and semantic priming depends on the strength of the association between items. However, there is a problem for that interpretation stemming from reports that automatic episodic priming of recognition judgments can occur (Ratcliff & McKoon, 1981; Neely & Durgunoglu, 1985). This finding also was replicated in the recognition task of the present study, as will be discussed later. If 97 automatic episodic priming can occur, and does for the same set of materials under the same conditions in a different task, then it is unclear why overlearning the association should change it or make it better or more effective. The existing episodic association is sufficient to activate the episodic representation of the target without overlearning. One could understand episodic priming's failure to produce activation in the semantic task by arguing, as Jacoby and Brooks (1984) might, that the tasks tap different aspects of information, but then that would not explain why overlearning the association produces an effect. Therefore, it may be necessary to speculate that overlearning has resulted in a change in the semantic memory representation or activation, instead of, or in addition to, a change in the episodic association. This change in semantic memory might consist of the construction of a new link between the existing items. Overall, the present data support the notion that strategic mediation is needed to obtain episodic priming effects in a semantic task. To the extent that the short prime-target SOA precludes this, one fails to find priming from episodic associations. This is especially true if one accepts the argument that the effects of overlearning are on semantic memory rather than on the episodic association. 98 Nonword Targets The comparisons of response latencies for the nonword conditions indicated nearly significant inhibition for newly learned pairs resulting from episodic relatedness (46 msec, F (1,31) = 3.63, p. = .07), and significant inhibition from episodic relatedness for overlearned pairs (53 msec, F (1,31) = 4.55, p. < .05), replicating previous research (Neely & Durgunoglu, 1985; Durgunoglu & Neely, 1985). No significant differences in accuracy were present. As noted ealier, this inhibition may be another consequence of the same mechanism that causes repeated nonwords to be rejected more slowly than new nonwords in a lexical decision task. One view of that phenomenon is that normally an item is labelled a nonword following an unsuccessful memory search for it. However, in the case of a previously seen nonword, the memory search yields a positive trace and a competing response (Feustel et al., 1983). The present data suggest that preserving the studied prime-target relationship contributes to the strength of that competing response, since the targets themselves were studied equally in the episodic related and episodic unrelated conditions. The inhibition found in this condition also stands in contrast to the facilitation found at the long SOA, reaffirming the earlier suggestion that the long SOA condition for nonwords may resemble Neely's (1977) attention-shift 99 condition. Subthreshold Long SOA The subthreshold priming results provide a different perspective on the above issues. First, however, they require a bit of introduction. As an investigative technique, subthreshold priming is plagued by several problems. Among these are the difficulties inherent in defining a threshold, deciding if subjects are initially below that threshold, and the possibility that even if subjects start out below theshold, the threshold may change during the task (see Dagenbach & Carr, 1985, for a discussion). In pilot work for this research that examined the phenomenon of subthreshold priming, the prime and the pattern mask were presented binocularly--the subject saw both with both eyes. Very few subjects showed any evidence of a threshold change during subsequent trials when this procedure was used. Moreover, the semantic priming effects obtained appeared to have resulted from those subjects who did not go above chance on a threshold retest given at the end of the experiment (Dagenbach & Carr, 1982, 1984, 1985). In the present experiment, dichoptic presentation of the prime and pattern mask was used in an attempt to heighten the priming effect. This procedure proved to be much more susceptible to threshold shifts during the course of the lexical decision and recognition task trials with subthreshold primes. In the lexical decision condition of 100 the present study, fourteen of the thirty-two subjects exceeded chance performance in the threshold retest. Moreover, several subjects were able to identify one or more of the subthreshold primes, something which never occurred when binocular presentation was used. Thus, there is a question of whether all the subjects in this study were truly below the threshold of conscious awareness at all times. Because of this question, the data were analyzed for all the subjects, and then reanalyzed after deleting those subjects who gave any evidence of having conscious knowledge about the primes, either by exceeding chance performance on detection trials during the threshold retest, or by naming one of the primes at any point during any of the trials. The results of the analyses of the data after deleting subjects evidencing knowledge of the prime are in Tables 6 and 7. Since priming below conscious awareness seems to be qualitatively different from priming above threshold (Cheesman & Merikle, in press; Marcel, 1983a, 1983b), this manipulation might be expected to result in a different set of outcomes. In the subthreshold priming condition with a long prime-target SOA, the mean prime-mask interstimulus interval (ISI) at which performance in the detection task fell to chance was 19 msec for those subjects with prime-mask ISIs greater than zero. Seven of the 32 subjects in this condition were performing above chance 101 with prime-mask ISIs of zero. For these subjects, the prime-mask ISI was zero, and the prime exposure duration of 15 msec was shortened until chance performance was reached-~the mean prime exposure duration for these seven subjects was 4 msec. Word Targets For newly learned targets, planned comparisons showed a significant inhibitory effect due to episodic priming from primes that could not be detected: Newly learned targets preceded by episodically related primes were responded to 37 msec slower than those preceded by episodically unrelated primes (F (1,31) = 8.56, p. < .01). However, this effect disappeared after deleting the data of those subjects showing some knowledge of the prime (-15 msec), suggesting it may have been due to some conscious mechanism. Semantic priming of newly learned targets in this condition produced only 4 msec of facilitation (n.s.), while combined episodic and semantic primes produced 33 msec of facilitation (n.s., F (1,31) = 2.41, p. > .10). For overlearned word targets, neither episodic priming nor combined episodic and semantic priming had a significant effect, but semantic priming produced significant facilitation (37 msec, F (1,30) = 4.43, p. < .05). Deleting subjects showing some knowledge of the primes did not eliminate this effect (53 msec, F (1,16) = 4.44, p. < .05). Comparisons of error rates failed to Table 6 102 Mean Lexical Decision RT (msec) and Percent Error in Subthreshold Conditions after Subject Deletion: Word Targets Subthreshold Long SOA Newly learned E 657 (6.4) s 639 (5.4) as 638 (2.9) E 642 (4.6) E 638 (1.9) EE 677 (2.9) Net effect -15 (1.8) - 9 (3.5) +39 ( 8 ) Overlearned E 649 (2.9) s 681 ( 8 ) as 644 ( 8 ) E 633 ( 8 ) E 654 (1.5) EE 653 ( 8 ) Net effect -16 (2.9) +53*(1.5) +9 ( 8 ) Subthreshold Short SOA Newly learned E 597 (4.6) s 689 (3.2) ES 586 (3.2) E 591 (1.8) E 604 ( 8 ) EE 653 ( 8 ) Net effect +6 (2.8) -5 (3.2) +67 (3.2) Overlearned E 613 ( 0 ) S 598 ( 0 ) ES 618 ( 0 ) E 652 (2.8) E 576 (1.8) EE 588 (2.8) Net effect +39 (2.8) -22 (1.8) -30 (2.8) Related Conditions E Episodic S Semantic ES Episodic and Semantic * p. < .05 Unrelated Conditions E Episodic ‘S Semantic ES Episodic and Semantic 103 Table 7 Mean Lexical Decision RT (msec) and Percent Error in Subthreshold Conditions after Subject Deletion: Nonword Targets Subthreshold Long SOA Newly learned E 784 (21.5) Overlearned E 646 (12.8) E 698 (15.2) E 747 (18.8) Net effect -6 ( 6.3) +lgl*( 2.9) ‘4 Subthreshold Short SOA Newly Learned E 748 (15.4) Overlearned E 775 (15.7) E 699 ( 9.6) E 746 (13.6) Net effect -49 ( 5.8) -29 ( 2.1) E Episodic Related Episodic Unrelated ** p. < .01 104 reveal any significant differences. Thus, in this condition two subthreshold priming effects were obtained. The inhibitory effect from episodic priming was eliminated by subject deletion, suggesting some role of conscious perception. The fact that this effect was inhibitory rather than facilitatory is of some importance since one can argue that inhibitory effects are less informative about the organization of memory that facilitatory effects: .Inhibition plausibly could arise from activation in a distinct memory system if that activation served to capture mental resources or misdirect the search process regardless of the subject's efforts to direct the search elsewhere. Therefore, the inhibition obtained would not have posed an insurmountable problem for the episodic/semantic distinction even if it were not eliminated by subject deletion. The facilitation from subthreshold semantic priming effect of overlearned word targets was expected based on the prior research on subthreshold priming, and replicates those claims. Finding semantic priming but not episodic priming under these conditions provides some support for the episodic/semantic distinction. However, that support is qualified because subthreshold priming logically also should have been found for newly learned word targets, and might have been expected in the combined episodic and semantic conditions. Given the vagaries of subthreshold priming, one alternative is to argue that to the extent 105 that effects are present, they support the episodic/semantic distinction, and simply ignore the failure to find priming effects in the other conditions. However, it may be useful to consider why subthreshold priming was not found in those conditions. There are two possible explanations of why subthreshold priming was found only for semantic priming of overlearned materials. First, subthreshold priming is not a particularly robust phenomenon, and any interfering conditions may be sufficient to disrupt it. One set of interfering conditions may have arisen as a consequence of the relative activation of the primes and targets in each condition: The failure to obtain facilitation from combined episodic and semantic priming of any targets, might reflect different activation of the prime due to prior study, resulting in different degrees of conscious access during the subthreshold trials. In those conditions, the prime had been studied just prior to the lexical decision trial, and therefore was likely to be at a higher level of activation in memory. This actually may have interfered with the facilitatory effect of the prime--pilot research for the present study has suggested that in subthreshold priming having more knowledge about the prime can hinder semantic facilitation (Dagenbach & Carr, 1982; 1984). In the semantic priming conditions, the target was likely to have been in a higher state of activation in the overlearned condition, and thus may have 106 been more sensitive to the weak effect of the subthreshold prime. Alternatively, these failures to obtain facilitation may be a consequence of interference in semantic priming due to the episodic relations present. This would be similar to the elimination of semantic priming following paired associate learning described by Durgunoglu & Neely (1985). In the present study, the results could be interpreted as an indication of episodic activation of the target impeding the subthreshold semantic priming effect. By this argument, no facilitation is obtained in the combined episodic and semantic conditions because episodic activation is occurring as well as semantic activation due to the subthreshold prime, and this episodic activation interferes with the semantic priming effect. A preview of the data from the recognition task may help support this argument. That data shows that subthreshold primes can activate episodic associations as well as semantic ones: Subthreshold episodic priming of newly learned targets at the long SOA produced a trend towards greater accuracy of recognition (10.7 vs. 20.4 % errors, F (1, 17) = 3.28, p. = .08), and subthreshold combined episodic and semantic priming significantly inhibited response latencies for overlearned targets. The absence of semantic priming for newly learned targets again might be seen as a type 2 error, with the overlearned targets being more sensitive to the effects of the subthreshold prime. It also might 107 be seen as some as yet not understood consequence of the subject devoting more effort or different kinds of processing to the encoding of the newly learned pairs, since the overlearned pairs were already highly familiar. Such extra study might have resulted in greater episodic activation of the target and therefore caused interference. This idea contradicts the previous suggestion that overlearned targets are more activated and therefore more sensitive to the subthreshold prime, but has the advantage of explaining the lack of effect from semantic priming of newly learned items using the same mechanism that might explain the lack of priming in the combined episodic and semantic conditions. Further research is needed to choose between these alternatives. Neither the first nor the second argument is particularly compelling, but no other interpretation of this pattern of results is immediately apparent, and those suggested above do a fair job of accounting for the majority of the effects obtained. More research will be required before they can be fully understood. At present, however, it seems likely that episodic activation underlies the failure to obtain priming in the episodic and semantic related conditions, which suggests a more intimate relationship between the episodic and semantic memory representations of the target than a straightforward episodic/semantic distinction might predict. This idea will be elaborated upon in the general 108 discussion. Nonword Targets Subthreshold episodic priming of nonword targets had no effect on response latencies for newly learned targets, but resulted in significant facilitation for overlearned items (74 msec, F (1,31) = 8.81, p. < .01). Deleting subjects showing some knowledge of any subthreshold prime's identity failed to eliminate the significant facilitation for overlearned pairs from episodic priming (101 msec, F (1,17) = 9.66, p. < .01). Accuracy did not differ significantly in either learning condition. Thus, the episodic relationship established between overlearned word-nonword pairs was sufficient to produce subthreshold priming. Exactly why an overlearned episodic association should prime a nonword response, but not a word response, is not clear. One possibility is that in the case of words, existing representations are present in semantic memory that may work counter to the effect when trying to link two normally unrelated items. On the other hand, for a word-nonword pair, the nonword presumably has no initial representation that has to be overcome. 109 Subthreshold Short SOA The mean prime-pattern mask ISI for subjects with prime-mask ISIs greater than zero was 31 msec in this condition. Eight subjects performed above chance at a prime-pattern mask ISI of zero, and therefore required a shortened prime exposure duration; for these subjects, the mean prime exposure duration was 5 msec. Twelve subjects exceeded chance performance on the threshold retest or named a prime during testing, and were deleted for the second set of analyses. Among the remainder, the mean prime-mask ISI was 28 msec for those with prime-mask ISIs greater than zero, and the mean prime exposure duration was 4 msec for the three subjects for whom the prime-mask ISI was zero. Word Targets For newly learned targets, neither episodic nor semantic priming produced a significant effect, but combined episodic and semantic priming yielded significant facilitation (40 msec, F (1,31) = 5.95, p. < .05). This may have been a type 1 error; there was some evidence of a speed-accuracy tradeoff (3.4% in the episodic and semantic related condition, vs. 0.8% in the episodic and semantic unrelated condition, F (1,31) = 3.14, p < .10). Deleting subjects evidencing knowledge about the primes did not eliminate this effect (67 msec, F (1,18) = 8.07, p. < .02), but the hint of a speed-accuracy tradeoff persisted (3.2% vs 0%). For overlearned targets, neither semantic, 110 nor combined episodic and semantic, priming produced a significant effect. Episodic priming resulted in 44 msec of facilitation (F (1,31) = 8.31, p. < .001) with no indication of a speed-accuracy tradeoff, but deleting subjects as before also eliminated this effect. Finding such facilitation from episodic priming even with partial conscious knowledge of the prime may seem contradictory since the prime-target SOA was short. However, the facilitation in this condition then parallels the result obtained in the suprathreshold short SOA condition, wherein substantial nonsignificant facilitation of response latency and significant facilitation of accuracy from episodic priming was obtained for overlearned targets. Therefore, the present effect may represent a repetition of that earlier finding, again perhaps due to the construction of a semantic link. Overall, the only substantial effect of subthreshold priming at a short prime-target SOA was from episodic and semantic priming of newly learned word targets, and the significance of this effect was minimized by the possibility of a speed-accuracy tradeoff. This is in keeping with other reports that have failed to obtain subthreshold priming effects at short prime-target SOAs (Balota, 1983; Fowler et al., 1981). This suggests that activation from subthreshold semantic priming takes some time to accrue, which has been seen as contradictory since subthreshold priming is presumably automatic, and 111 automatic priming can occur with a short prime-target SOA. However, it might be understood as reflecting very slow buildup of activation following a very weak priming stimulus. Some evidence of such slow buildup was present in the current study, wherein subjects attempting to guess the identities of words presented at threshold occasionally came up with items in the list, but two or three trials after that item had occurred. This could have been due to chance, but the opposite pattern wherein items were guessed before they appeared in an identification trial never was obtained. Nonword Targets Episodic priming of nonword targets did not significantly affect response latencies or accuracy for either newly learned or overlearned material, again in keeping with the notion that subthreshold priming requires a long prime-target SOA. Summary of the Lexical Decision Results Overall, the comparisons indicate that semantic priming of word targets can occur with or without a long prime-target SOA, and with or without conscious knowledge of the prime. In contrast, episodic priming of lexical decisions for word targets appears to require strategic mediation, thus supporting the rationale developed earlier regarding the nature of episodic effects in a semantic task. This is certainly the case if one accepts the argument that overlearning of episodic associations for 112 word targets may work to allow the construction of semantic memory links rather than changing the nature of the episodic ones, since these are already capable of producing automatic priming in other tasks. In this respect, the current study supports the episodic/semantic distinction. However, the negative consequences of episodic study for subthreshold priming pose possible problems for that distinction. While it would be dangerous to argue completely on the basis of the absence of a subthreshold priming effect, other studies have reported similar interference with suprathreshold semantic priming due to episodic study (Neely & Durgunoglu, 1985; Durgunoglu & Neely, 1985). At the least, this suggests that the executive processes controlling search and response are equally attracted to episodic and semantic activations while performing a semantic task. This could be due to episodic activations in a separate system creating some distraction that masks the semantic priming effect. Alternatively, the interference from episodic activation may reflect a more intrinsic relationship between episodic and semantic memory than a model with completely distinct systems would suggest. The results also indicate that episodic priming of nonword targets in a lexical decision task can produce substantial facilitation if a long prime-target SOA is present. Both suprathreshold and subthreshold episodic 113 priming of overlearned nonword targets yielded significant facilitation at the long SOA, indicating that the effect is not strategic in the sense of being a consciously generated strategy for linking the material. Episodic Recognition Task The recognition task data were analyzed in a similar manner as the lexical decision data. Median response times and accuracy scores were computed for each condition of interest, and planned comparisons were used to examine priming effects. The means of these medians are presented in Tables 8 and 9. As before, the data from each prime presentation condition will be discussed separately to aid in exposition. Suprathreshold Long SOA Word Targets Substantial and significant facilitaton of response time was obtained for newly learned word targets from episodic priming (141 msec, F (1,29) = 22.92, p. < .0001) and combined episodic and semantic priming (220 msec, F (1,30) = 59.12, p. < .0001), but not from semantic priming (20 msec, F < 1). No differences in accuracy reached statistical significance, but there was a trend towards more accurate identification of episodically primed newly learned targets (F (1,31) = 3.77, p. < .10). For overlearned targets, episodic priming and combined episodic and semantic priming both generated a large amount of facilitation of response latencies (174 114 Table 8 Mean RT (msec) and Percent Error for Recognition: Word Targets Suprathreshold Long SOA Newly learned E 562 (11.4) s 641 (21.9) ES 511 (10.4) E 783 (21.8) E 661 (14.8) E“ 731 (15.3) Net effect +l4l**(l0.4) +28 ( 7.1) 220**( 4.9) Overlearned E 581 (13.8) s 665 (17.4) ES 528 (11.7) E 755 (28.6) E 728 (16.4) EE 739 (38.7) Net effect +174**( 6.8) +63 ( 1.8) +211**(l9.0)** Suprathreshold Short SOA Newly learned E 601 ( 9.4) S 645 (19.3) ES 557 ( 9.3) E 661 (15.1) E 668 (12.8) EE 693 (11.7) Net effect +60**( 5.7)* +15 ( 6.5) +136**( 2.4) Overlearned E 593 ( 5.5) S 671 (20.6) ES 568 ( 6.5) E 675 (21.6) E 669 (17.9) EE 738 (18.8) Net effect +82**(l6.1)** -2 ( 2.7) 162**(12.3)** Related Conditions E Episodic E S Semantic ‘S ES Episodic and Semantic ES Episodic Episodic Semantic Unrelated Conditions and Semantic 115 Table 8 (cont'd) Subthreshold Long SOA Newly learned E 672 (11.4) s 723 (15.9) ES 696 (16.6) E 688 (15.4) E 693 (17.2) EE 728 (19.8) Net effect +8 ( 4.8) -38 ( 1.3) +24 ( 3.2) Overlearned E 739 (21.9) s 697 (14.1) ES 746 (14.6) E 693 (28.8) E 698 (9.9) 'E‘ 722 (12.5) Net effect -46 ( 1.9) +1 ( 4.2) -24 ( 2.1) Subthreshold Short SOA Newly learned E 643 (15.1) S 644 (11.7) ES 624 (9.4) E 638 (17.2) 'E 681 (14.3) EE 657 (9.4) Net effect -5 ( 2.1) +37 ( 2.6) +33 ( 8 ) Overlearned E 663 (11.9) S 627 (14.8) ES 645 (12.3) E 786 (23.2) E 655 (16.9) EE 636 (21.3) Net effect +43 (ll.3)* +28 ( 2.1) —9 ( 9.8) Related Conditions E Episodic E S Semantic S ES Episodic and Semantic ES * p. < .05 Episodic Semantic Episodic Unrelated Conditions and Semantic 116 Table 9 Mean RT (msec) and Percent Error for Recognition: Nonword Targets Suprathreshold Long SOA Newly learned E 575 ( 4.2) Overlearned E 637 (11.7) E 558 ( 5.2) ‘E 727 (15.1) Net effect +62**(7.5)* +169**(9.9)** Suprathreshold Short SOA Newly learned E 632 ( 8.3) Overlearned ml 631 (11.7) E 691 ( 8.9) MI 698 (13.0) Net effect -1 ( 3.4) +7 ( 4.1) Subthreshold Long SOA Newly learned E 623 ( 5.5) Overlearned E'649 ( 8.9) Net effect +26 ( 3.4) E 576 (12.5) E 697 (13.5) +121**(1.0) Subthreshold Short SOA Newly learned E 659 ( 7.8) Overlearned E 632 ( 7.8) E 694 (13.5) E 695 (13.3) Net effect -27 ( 0 ) +1 ( 0.2) E Episodic Related E Episodic Unrelated * p. < .05 ** p. < .01 117 msec, F (1,29) = 19.35, p. < .01; 211 msec, F (1,30) = 36.87, p. < .0001, respectively). Semantic priming also facilitated response latencies in this condition, but this effect failed to attain significance (62 msec, F (1,31) = 2.76, p. = .10). Combined episodic and semantic priming of overlearned targets also significantly affected accuracy (F (1,31) = 15.1, p. < .001). Thus, in the recognition task, only primes containing episodic information had an effect; semantic priming failed to produce significant effects even under these conditions wherein the subject was free to use all possible information. Nonword Targets Episodic priming of nonword targets significantly decreased response latency for newly learned items (62 msec, F (1,31) = 12.14, p. < .002), and for overlearned items (169 msec, F (1,31) = 23.25, p. < .0001), but did not significantly affect accuracy. Given the effectiveness of episodic priming for recognition of words, the nonword results might be expected. Suprathreshold Short SOA Word Targets Priming at the short prime-target SOA generated essentially the same pattern as in the previous condition. Episodic priming of newly learned word targets produced 59 msec of facilitation (F (1,31) = 6.51, p. < .01), and combined episodic and semantic priming generated 136 mSec 118 of facilitation ( F (1,31) = 49.6, p. < .0001), whereas semantic priming failed to produce a significant effect (15 msec, n.s.). For overlearned targets, the same pattern pertained; episodic priming and combined episodic and semantic priming generated significant facilitation (83 msec, F (1,30) = 8.38, p. < .01; and 161 msec, F (1,31) = 28.5, p. < .0001, respectively), and semantic priming failed to produce any effect. Analyses of the error rates showed significant effects from episodic priming of newly learned targets (F (1,31) = 4.00, p. = .05) and overlearned targets (F (1,31) = 7.17, p. < .02), and for combined episodic and semantic priming of overlearned targets (F (1,31) = 7.94, p. < .01). In this condition, facilitation only occurs when there is an episodic relationship between the prime and the target. Because the prime-target SOA in this condition is short, these effects presumably reflect an automatic process, and thus support Ratcliff and McKoon's (1981) finding of automatic episodic priming in an episodic task. They also mirror the results in the lexical decision task using suprathreshold primes and a short prime-target SOA, wherein only semantic, and combined episodic and semantic, priming produced an effect. Nonword Targets Interestingly, for nonword targets episodic priming at the short SOA did not significantly affect either 119 either response latency or accuracy. It is not clear why this should be the case since episodic priming of word targets at this SOA is pervasive, and episodic priming of nonword targets is obtained at a long prime-target SOA. Moreover, Neely and Durgunoglu (1985) report having obtained significant facilitation in a similar condition. The obvious interpretation is that it represents a type 2 error in the present study. Subthreshold Long SOA Among all subjects with a prime-mask ISI greater than zero, the mean prime-mask ISI was 24 msec. For the 14 subjects with prime-mask 1813 of zero, the mean prime exposure duration was 5 msec. Twelve subjects either exceeded chance in the threshold retest or identified a prime during the recognition trials, and data from them was deleted for the second set of analyses. Among the remainder, the mean prime-mask ISI for those with prime-mask ISIs greater than zero was 36 msec, and the mean prime exposure duration for those with prime-mask ISIs equal to zero was 7 msec. The data from the analyses following subject deletion are presented in Tables 10 and 11. Word Targets Subthreshold priming with long prime-target SOAs in the recognition task failed to significantly affect response time or accuracy for word targets in the initial analysis regardless of the type of prime or whether the 120 Table 10 Mean Recognition RT (msec) and Percent Error in Subthreshold Conditions after Subject Deletion: Word Targets Subthreshold Long SOA Newly learned E 649 (10.7) S 689 (15.8) ES 672 (16.2) ml ml 656 (20.4) 660 (12.0) ES 717 (23.1) Net effect +7 ( 9.7) -29 ( 3.8) +45 ( 6.9) Overlearned E 768 (22.7) S 637 (11.6) ES 765 (14.8) MI “I 724 (16.6) 671 ( 6.4) ES 671 (14.8) Net effect -44 ( 6.1) +34 ( 5.2) -94* ( 0 ) Subthreshold Short SOA Newly learned E 614 (13.2) S 600 (11.5) ES 591 (11.8) ml ml 614 (16.2) 625 (18.4) ES 636 (13.6) Net effect 0 ( 3.0) +25 ( 6.9) 45 ( 1.8) Overlearned E 648 (11.8) S 602 (14.0) ES 636 (11.8) E 639 (29.4) E 632 (14.9) ES 618 (26.3) Net effect -9 (17.6)** +30 ( 0.9) —18 (14.5)* Related Conditions Unrelated Conditions E Episodic E Episodic S Semantic S. Semantic ES Episodic and Semantic ES Episodic and Semantic * p. < .05 ** p. < .01 121 Table 11 Mean Recognition RT (msec) and Percent Error in Subthreshold Conditions after Subject Deletion: Nonword Targets Subthreshold Long SOA Newly learned E 656 ( 6.9) Overlearned E 587 (12.9) E 681 ( 6.9) E 702 (16.7) Net effect +25 ( 0 ) +115*( 3.8) Subthreshold Short SOA Newly learned E 649 ( 7.9) Overlearned E 664 (17.5) E 621 ( 7.0) E 642 (15.0) Net effect -28 ( 8.9) -22 ( 2.5) E Episodic Related E Episodic Unrelated * p. < .01 122 study list pairs were overlearned or newly learned. Deleting subjects who had gone over threshold, as indicated by either correctly identifying any primes during testing or getting more than 12/20 correct in the threshold retest, changed this pattern somewhat. After deletion of those subjects, there was nearly significant facilitation of accuracy from episodic priming of newly learned targets (F (1,17) = 3.28, p. = .08), and significant inhibition of response latencies for overlearned targets as a result of combined episodic and semantic priming (93 msec, F (1,17) = 4.70, p. < .05). Unfortunately, the present data allow no insight into why this effect is inhibitory. Nonword Targets Episodic priming of nonword targets did not affect newly learned pairs, but did yield substantial facilitation for overlearned targets (122 msec, F (1,31) = 20.18, p. < .001). The results were essentially unchanged after subjects were deleted--there was still significant facilitation due to episodic priming of overlearned targets, but no other effects were present. The results in the recognition task parallel the findings in the lexical decision task, suggesting a possible common component. 123 Subthreshold Short SOA The mean prime-mask ISI for subjects with prime-mask ISIs greater than zero in this condition was 31 msec, and the mean prime exposure duration for the eight subjects with prime-mask ISIs of zero was 4 msec. Eight subjects either exceeded chance on the threshold retest or identified a prime, and were deleted from the second analysis. Among the remainder, the mean prime mask ISI was 21 msec for those with prime-mask ISIs greater than zero. The mean prime exposure duration for those six with prime ISIs equal to zero was 5 msec. Word Targets Subthreshold priming in this condition failed to significantly affect response latencies for word targets, although the effect of semantic priming for newly learned items approached significance (36 msec, F (1,31) = 3.73, p. = .06). However, this effect was eliminated following subject deletion. Episodic priming increased accuracy significantly in the overlearned material (F (l, 31) = 4.80, p. < .05), and this effect was not eliminated by deleting subjects (F (1,18) = 5.34, p. < .01). Combined episodic and semantic priming also yielded significant facilitation of accuracy in the overlearned material after subject deletion (F (1,18) = 4.27, p. < .05). This increased accuracy represents further evidence of subthreshold episodic priming of an episodic decision. 124 Nonword Targets Episodic priming of nonwords failed to produce a significant effect on accuracy or response time. Deleting subjects who exceeded chance in the threshold retest did not substantially change any of these results. Summary of the Recognition Task Data The planned comparisons of the recognition task data indicate that episodic priming of recognition judgments is pervasive, but that semantic priming of such judgments fails to produce significant effects. Even in the suprathreshold long SOA condition, semantic priming failed to produce a significant effect, although the amount of facilitation for overlearned words was large (63 msec). Moreover, the strong effects of episodic priming persisted at the suprathreshold short SOA, replicating other reports of automatic episodic priming of episodic judgments. In that same condition, even the nonsignificant semantic priming effects were virtually eliminated, suggesting that these resulted from some strategic factor. As in the case of the lexical decision task, the data from the subthreshold conditions are less straightforward. Subthreshold combined episodic and semantic priming at a long SOA produced inhibition for overlearned targets, suggesting that some form of processing was occuring for the unseen words that influenced recognition judgments. This also was suggested by the facilitation in accuracy obtained from episodic and combined episodic and semantic 125 priming of overlearned word targets in the subthreshold short SOA condition, and by the substantial facilitation obtained for overlearned nonword targets in the long SOA condition. Thus, it appears that words presented outside of conscious awareness can influence episodic as well as semantic decisions, but in the case of an episodic decision, an episodic relationship between items is required. Most importantly, no evidence of between system priming without strategic mediation was obtained. Cued Recall Following approximately one third of the study lists in the recognition and lexical decision tasks, cued-recall tests were administered. The accuracy of recall on these tests was analyzed to determine if subjects were in fact memorizing the study list items to a similar extent in both tasks. Subjects in the lexical decision task recalled 86% of the items in the cued recall tests correctly, whereas subjects in the recognition task recalled 89.5% of the items correctly. The difference in cued recall performance between the tasks was not significant. GENERAL DISCUSSION First, the priming effects obtained in the present experiment will be reviewed, and their relationship to the episodic/semantic distinction will be considered. It was noted earlier that the episodic/semantic distinction can be tested by looking for transfer effects and by looking for dissociative effects, whereby episodic or semantic primes produce different effects in episodic and semantic tasks. The transfer effects obtained in the present study will be discussed first, followed by a review of the dissociative effects. The purpose of this review is to determine what model of the relationship between episodic and semantic memory can best accomodate the results obtained. Transfer Effects The results from suprathreshold priming of lexical decisions suggest that semantic priming of a semantic decision requires either the existence of a semantic association between the prime and target, or sufficient time to intentionally go from an episodic activation to a semantic activation for the same item. Episodic priming of response latencies for newly learned words occurred only in the suprathreshold long prime-target SOA condition, wherein strategic effects in priming would be expected, and not under conditions wherein strategic effects would not be expected. The failure to obtain 126 127 episodic priming of lexical decisions with a short prime-target SOA suggests that McKoon and Ratcliff's (1979) results showing episodic priming of semantic decisions were an artifact of the ISI/SOA confusion in that study. This finding is in substantial agreement with the conclusions of Carroll and Kirsner (1982) and Neely and Durgunoglu (1985). The facilitation of accuracy by episodic priming for overlearned word targets at the short SOA was consistent with the notion that episodic priming might occur under conditions yielding automatic priming if the associations between the episodically related items could be made strong enough. By itself, this might be understood in terms of episodic associations being weaker and requiring a greater amount of time for activation to occur; with practice, the activation of the target following encoding of the prime becomes more automatic and thus faster. However, because other evidence suggests that episodic priming can occur in an episodic task under identical conditions, it is suggested that the effect of overlearning is on semantic memory rather than the episodic association. The results from subthreshold priming in the lexical decision task are more varied, yet comprehensible and revealing with certain assumptions. The first assumption is that subthreshold semantic priming takes time to accrue, and therefore will not occur at the short SOA. If 128 one dismisses the one significant effect in that condition due to the speed-accuracy tradeoff present, then no subthreshold semantic priming effects were found at the short SOA, replicating other research (Balota, 1983; Fowler et al., 1981). Subthreshold semantic priming of overlearned targets was obtained at the long SOA, whereas no subthreshold episodic effects were significant after deleting subjects with some knowledge about the prime. This provides further evidence of a difference between episodic and semantic priming, and is consistent the the conclusions derived from the suprathreshold conditions. However, additional assumptions are necessary in order to understand the failure to obtain subthreshold semantic priming of newly learned targets, and subthreshold combined episodic and semantic priming of any targets, at the long SOA. Two possible accountings of those failures were suggested based on consideration of either the relative activation of the primes and targets due to prior study, or the possibility that interference arises from episodic activation of the target. The second possibility deserves some further consideration. The notion that episodic activation interferes with subthreshold semantic priming sounds farfetched at first, but may fit in with some other aspects of what has been reported about subthreshold semantic priming and offer some insight into the the nature of episodic and semantic memory. Specifically, 129 Marcel (1983a, b) has argued that subthreshold semantic priming activates all possible meanings of an item, but upon access to conscious awareness, the item is assigned one specific interpretation. Subthreshold priming is thus qualitatively different from suprathreshold priming. If subthreshold semantic priming's effect is on some unbound form of a word, then perhaps it is not capable of producing an effect on particular instantiations of words. Episodic activation may entail the creation or assignment of a particular instantiation, and therefore interfere with the subthreshold priming effect. This reasoning suggests the possibility of an inherent relationship between the episodic and semantic memory representations of the same item that goes against the spirit of the episodic/semantic distinction. In sum, the results from the suprathreshold conditions in the lexical decision task are straightforward and suggest that episodic priming of a semantic decision requires strategic mediation, or at least a different form of episodic association than one would normally develop. The results from the subthreshold conditions are less consistent, and therefore must be interpreted with some caution. To the extent that effects were present, they suggest that semantic priming of a semantic task can be obtained under conditions wherein episodic priming cannot, and thus also offer support for the episodic/semantic distinction. However the absence of 130 some expected subthreshold priming results also suggests the possibility of episodic interference with semantic priming, indicating at a minimum some shared output between episodic and semantic memory systems, or the possibility of a more involved interrelationship. The results from the recognition task were less complicated. Substantial facilitation of recognition judgments from episodic priming was obtained in both suprathreshold conditions from episodic and combined episodic and semantic priming. There was no effect from semantic priming by itself, replicating other failures to find semantic priming of episodic judgments (Hermann & Harwood, 1980; Carroll & Kirsner, 1982). The SOA manipulation provided further evidence of automatic episodic priming within an episodic task, replicating other reports (Ratcliff & McKoon, 1981; Neely & Durgunoglu, 1985). Subthreshold priming effects for recognition were obtained in three conditions: Combined episodic and semantic primes in the overlearned material increased response latency at the long prime-target SOA, and episodic, and combined episodic and semantic, primes in the short SOA condition increased accuracy of recognition. The reason for the inhibition of response latencies is unclear; it seems to stem from the episodic component of the combined prime since episodic priming produces subtantial although not statistically significant 131 inhibition, and semantic priming produced substantial but not significant facilitation, under the same prime presentation conditions. The facilitation of accuracy for overlearned materials at the short prime-target SOA suggests that subthreshold episodic activation can occur at a faster speed than subthreshold semantic priming. Because these effects are all in overlearned conditions, wherein semantic pathways also may exist according to the argument developed earlier, it might be asked whether these are not semantic effects rather than episodic. However, subthreshold semantic priming does not work this quickly according to the lexical decision data, and semantic priming itself does not produce a significant effect in this task. In sum, the recognition task data support the notion that conscious mediation is needed for between system priming effects, even though their failure to do so would not have been surprising--as was noted earlier, semantic facilitation could have resulted from increased fluency of perceptual recognition in this task. The subthreshold priming data also provide evidence of a dissociation between episodic activation and conscious awareness. Dissociative Priming Effects For word targets, with suprathreshold primes at a long SOA, episodic, semantic, and episodic and semantic priming all produce facilitation in episodic and semantic tasks, suggesting parallel operating characteristics. In 132 the suprathreshold short SOA condition, episodic priming produced facilitation in the episodic task, but not the semantic task. Semantic priming generated facilitation in the semantic task, but not the episodic task, thus showing a dissociative effect. The subthreshold conditions also yielded evidence of dissociative effects. At the long SOA, semantic priming produced facilitation in the semantic task and no effect in the episodic task. Episodic priming produced no effects in the semantic task, but did produce an effect in the episodic task. At the short SOA, semantic priming had no effects in either task, and episodic priming produced facilitation in the episodic task, but not the semantic task. Finally, the nonword conditions also yielded some evidence of dissociative effects, although the importance of these effects is minimized by the fact that the fact that nonwords required different responses in the two tasks. Suprathreshold episodic priming produced facilitation for overlearned nonwords at the long SOA in both tasks, but suprathreshold priming of newly learned nonwords produced facilitation in the episodic task and nonsignificant inhibition in the semantic task. At the short SOA, suprathreshold priming produced no effects in the episodic task, and inhibition in the semantic task. Subthreshold priming produced facilitation for overlearned targets at the long SOA in both tasks, and no significant 133 effects at the short SOA. Overall, the dissociative effects are quite striking, and can be seen as support for the episodic/semantic distinction. Of course, one would want to weigh many such pieces of evidence before concluding that the distinction is justified. Moreover, one might also demand a version of the episodic/semantic distinction that can predict these effects. As it presently stands, the episodic/semantic distinction does not do so. W The Relationship between Episodic and Semantic Memory The dissociations between episodic and semantic priming reported above and elsewhere are not easily accounted for by models wherein episodic and semantic memory are not distinct systems, although, as Neely and Durgunoglu (1985) have argued, more evidence of such dissociations in other episodic and semantic tasks would be required before some ultimate judgment could be made. On the other hand, the above pattern of results also is fairly complicated for a straightforward episodic/semantic distinction with separate memory systems to account for. Specifically, such a model might have difficulty accounting for episodic interference with semantic priming, both from the subthreshold data in the present task, and from suprathreshold data such as Neely & Durgunoglu (1985) report. To do so, one needs some arrangement whereby the output from one system can interfere with the output from the other. It is possible 134 that there are separate memory systems with some shared executive retrieval functions that allow the patterns of interactions observed. Another possibility, perhaps more plausible, is that some compromise model will prove to be an effective resolution of this apparent impasse. One possible compromise model that might account for some of the wide array of findings regarding the episodic/semantic distinction is one in which episodic and semantic memory share the same content nodes in memory, but have different independent network associations linking them, with automatic spreading activation occurring in each network. In this case, although they would rely on the same content nodes, episodic vs. semantic activation might result in qualitatively different forms: Episodic activation of a node under some conditions might transform it in a way that hindered further semantic activation without attended or intentional processing as long as the episodic activation persisted, unless some semantic feature was intentionally made part of the episodic encoding. On the other hand, semantic activation would either facilitate or have no effect on episodic activation. Also, the individual's retrieval efforts might focus on one system or the other, making activation along the networks of the attended system more likely. More research is required before any ultimate judgment of the merits of the episodic/semantic 135 distinction will become possible. However, the consistent finding of some interactions of the systems and some dissociations suggests that looking for ways of integrating episodic and semantic memory while at the same time preserving the independence of episodic and semantic relationships may be necessary. REFERENCES REFERENCES Anderson, J. R. (1976). Language, memory, and thought. Hillsdale, N.J.: Erlbaum. Anderson, J. R., & Ross, B. H. (1980). Evidence against a semantic-episodic distinction. Journal of Experimental Psychology: Human Learning and Memory, 6, 41-466. Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The Psychology of Learning and Motivation (Vol. 2, pp. 89-195). New York: Academic Press. Balota, D. (1983). Automatic semantic activation and episodic memory encoding. Journal of Verbal Learning and Bergson, H. (1911). Matter and memory. London: Allen & Unwin. Brewer, W. F., & Pani, J. R. (1983). The structure of human memory. In G. H. Bower (Ed.): The psychology of learning and motivation, (Vol. 17, pp. 1—38). New York: Academic Press. Carroll, M., & Kirsner, K. (1982). Context and repetion effects in lexical decision and recognition memory. Journal of Verbal Learning and Verbal Behavior, 21, 55-69. 136 137 Cheesman, J., & Merikle, P. M. (in press). Word recognition and consciousness. In D. Besner, T. G. Waller, & G. E. MacKinnon (Eds.), Reading research: Advances in theory and practice, (Vol. 5). New York: Academic Press. Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11, 671-684. Dagenbach, D., & Carr, T. H. (1982, November). On semantic activation from unidentified and undetected words. Paper presented at the meeting of the Psychonomic Society, Minneapolis, MN. Dagenbach, D., & Carr, T. H. (1984, May). Now you see it, now you don't: Strategic effects in subthreshold semantic priming. Paper presented at the meeting of the Midwestern Psychological Association, Chicago, IL. Dagenbach, D., & Carr, T. H. (1985). Awareness, attention, and automaticity in perceptual encoding: Conscious influences on unconscious perception. Unpublished manuscript. DeGroot, A., Thomassen, A., 8 Hudson, P. (1982). Associative facilitation of word recognition as measured from a neutral prime. MemoryAand Cognition, 10, 358-370. denHeyer, K., Briand, K., & Dannenbring, G. L. (1983). Strategic factors in a lexical decision task: Evidence for automatic and attention-driven processes. Memory and Cognition, 11, 374-381. 138 Durgunoglu, A. Y., & Neely, J. H. (1985). Episodic and semantic priming in a lexical decision task following paired associate learning. Manuscript submitted for publication. Feustel, T. C., Shiffrin, R. M., & Salasoo, A. (1983). Episodic and lexical contributions to the repetition effect in word identification. Journal of Experimental Psychology: General, 112, 309—346. Fowler, C. A., Wolford, G., Slade, R., & Tassinary, L. (1981). Lexical access with and without awareness. Journal of Experimental Psychology: General, 110, 341-362. Graff, P., Squire, L. R., & Mandler, G. (1984). The information that amnesic patients do not forget. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 164-178. Hermann, D. J., & Harwood, J. R. (1980). More evidence of the existence of separate semantic and episodic stores in long-term memory. Journal of Experimental Psychology: Human Learning and Memory, 6, 467-478. Hermann, D. J., & McLaughlin, J. P. (1973). Effects of experimental and preexperimental organization on recogntion: Evidence for two storage systems in long-term memory. Journal of Experimental Psychology, 99, 174-179. 139 Hintzman, D. L. (1976). Repetition and memory. In G. H. Bower (Ed.), The psychology of learning and motivation, (Vol. 10, pp. 47-91). New York: Academic Press. Hintzman, D. L. The psychology of learning and memory. San Francisco: Freeman. Jacoby, L. L. (1983a). Remembering the data: Ananlyzing interactive processes in reading. Journal of Verbal Learning and Verbal Behavior, 23, 485-508. Jacoby, L. L. (1983b). Perceptual enhancement: Persistent effects of an experience. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9, 21-38. Jacoby, L. L., & Brooks, L. R. Nonanalytic cognition: Memory, perception, and concept learning. (1984). In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory, (Vol. 18, pp. 1-47). New York: Academic Press. Jacoby, L. L., & Dallas, M. (1981). On the relationship between autobiographical memory and perceptual learning. Journal of Experimental Psychology: General, 110, 306-340. Jacoby, L. L., & Witherspoon, D. (1982). Remembering without awareness. Canadian Journal of Psychology, 36, 300-324. Johnson, M. A. (1983). A multiple-entry, modular memory system. In G. H. Bower (Ed.), The psychology of learning and motivation, (Vol. 17, pp. 81-123). New York: Academic Press. 140 Keppel, G. (1982). Design and analysis: A researcher's handbook (2nd ed.). Englewood Cliffs, NJ: Prentice Hall. Kucera, H., & Francis, W. 1967. Computational analysis of present-day American English. Providence, RI: Brown University Press Mandler, G. (1980). Recognizing: The judgment of previous occurence. Psychological Review, 81, 252-271. Marcel, A. J. (19833). Conscious and unconscious perception: Experiments on visual masking and word recognition. Cognitive Psychology, 15, 197-237. Marcel, A. J. (1983b). Conscious and unconscious perception: An approach to the relations between phenomenal experience and perceptual processes. Cognitive Psychology, 15, 238-300. McCloskey, M., & Santee, J. (1981). Are semantic memory and episodic memory distinct systems? Journal of Experimental Psychology: Human Learning and Memory, 1, 66-71 0 McKoon, G., & Ratcliff, R. (1979). Priming in episodic and semantic memory. Journal of Verbal Learning and Verbal Behavior, 18, 463-480. Merikle, P. M. (1982). Unconscious perception revisited. Perception and Psychophysics, 31, 298-301. Meyer, D. E., & Schvaneveldt, R. W. (1971). Facilitation in recognizing pairs of words: Evidence of a dependence between retrieval operations. Journal of Experimental Psychology, 20, 227-234. 141 Morton, J. (1979). Facilitation in word recognition: Experimets casuing change in the logogen model. In P. A. Kolers, M. E. Wrolstal, & H. Bouma (Eds.), Processing visible language I (PP. 259-268). New York: Plenum Press. Muter, P. (1978). Recognition failure of recallable words in semantic memory. Memory and Cognition, 6, 9-12. Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibitionless spreading activation and limited-capacity attention. Journal of Experimental Psychology: General, 106, 226-254. Neely, J. H., & Durgunoglu, A. Y. (1985). Dissociative episodic and semantic priming effects in episodic recognition and lexical decision tasks. Journal of Memory and Language, 24, 466-489. Nolan, K. A., & Caramazza, A. (1982). Unconscious perception of meaning: A failure to replicate. Bulletin 2; the Psychonomic SOciety, 20, 23-26. Paivio, A. (1971). Imagery and Verbal Processes. New York: Holt. Palermo, D. S., & Jenkins, J. J. (1964). Word association norms: Grade school through college. Minneapolis, MN: University of Minnesota Press. Posner, M. I., & Snyder, C. R. R. (19753). Attention and cognitive control. In R. L. Solso (Ed.), Information processing and cognition: The Loyola symposium (pp. 142 Posner, M. I., & Snyder, C. R. R. (1975b). Facilitation and inhibition in the processing of signals. In P. M. A. Rabbitt (Ed.), Attention and performance V (PP. 669-682). New York: Academic Press. Purcell, D. C., Stewart, A. L., & Stanovich, K. E. (1983). Another look at semantic priming without awareness. Perception and Psychophysics, 33, 65-71. Ratcliff, R., & McKoon, G. (1981). Automatic and strategic priming in recognition. Journal of Verbal Learning and Verbal Behavior, 33, 204-215. Reiff, R., & Scheerer, M. (1959). Memory and hypnotic age regression. New York: International Universities Press. Schank, R. (1975). The structure of episodes in memory. In D. G. Bobrow & A. M. Collins (Eds.), Representation and understanding. New York: Academic Press. Shoben, E. J., Wescourt, K. T., & Smith, E. E. (1978). Sentence verification, sentence recognition, and the semantic-episodic distinction. Journal of Experimental Psychology: Human Learning and Memory, 3, 304-317. Tulving, E. (1972). Episodic and semantic memory. In E. Tuvling & W. Donaldson (Eds.), Organization of memory (pp. 381-403). New York: Academic Press. Tulving, E. (1983). Elements of episodic memory. London: Oxford Press. 143 Tulving, E., Schacter, D. L., & Stark, H. A. (1982). Priming effects in word-fragment completion are independent of recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 3, 336-342. Tulving, E., & Wiseman, S. (1975). Relation between recognition and recognition failure of recallable words. Bulletin of the Psychonomic Society, 3, 79-82. APPENDIX APPENDIX A Sample Set of Stimulus Materials Note: The letters and "NW“ refer to unstudied words and nonwords used as filler items to create control conditions. The conditions identified by number below are described in detail in Table 3. Study List cover blash moon star likely bear tree birch numbers letters scissors cut white black army soft money chair crowd prend park tofe track cherry wine queen paper opera gate busket model vade metal brass horror thread gun sleep purse lince grow masset day woman blood round bread butter purse lince Set 1 Test List cover blash tree NW moon star cloth letters long bear bird robin scissors W grow prend army soft color blue foot black carpet chair puzzle opera park NW king queen money NW fruit cherry gate busket gun sleep purse W metal brass cloud thread likely W gate vade fruit apple square round model lince crowd masset bread butter rain woman 144 Condition 7 underlearned 15 overlearned 5 underlearned 6 overlearned 4 overlearned ll underlearned 8 underlearned l underlearned 6 underlearned 4 underlearned 2 overlearned l4 underlearned 3 overlearned 13 underlearned 3 underlearned 7 overlearned overlearned overlearned overlearned underlearned overlearned overlearned ...I mmNU'IQH ...a wmooooww underlearned overlearned underlearned overlearned overlearned park tofe angry arson picnic rug floor purple cloth linen cover blash city town scissors cut bring sit doctor nurse camera gretch clothing sweater jail light coffee orchid model vade camera gretch memory bicycle brush water rain chisel gate busket crowd prend train grass youth tango hold short model vade numbers letters spider web tree birch swing fast moon star 145 park tofe cloth W brush arson floor purple music rock bird rug scissors cut cover W salt sit numbers town picnic W doctor NW coffee orchid fish sweater bring NW lamp light camera gretch model NW gate W horror water purse gretch paper chisel vehicle bicycle flower rose model vade train grass animal short crowd NW table tango tool saw numbers NW moon W city birch youth NW rapid fast spider web underlearned overlearned underlearned overlearned underlearned underlearned underlearned overlearned overlearned underlearned underlearned underlearned underlearned overlearned overlearned overlearned overlearned overlearned underlearned overlearned overlearned overlearned overlearned overlearned overlearned underlearned underlearned overlearned underlearned overlearned underlearned underlearned underlearned cover blash dinner sword blossom flower foot shoe grow masset grow masset cloth linen coat pepper doctor nurse park tofe city town avenue order school home neck eagle purse lince gate busket supper beetle speak talk puzzle nail crowd prend spider web cloud golf fish cod dentist smoke camera gretch cloth linen dinner sword moon star bread butter doctor nurse gate busket likely bear wine queen city town model vade 146 dinner sword white shoe hold W blossom flower grow masset park blash stand pepper cover tofe tree linen doctor nurse grow W coat NW dance eagle city W school home purse lince command order weapon knife day nail cloth flannel speak talk camera busket crowd prend supper beetle tobacco smoke spider NW clothing cod neck W angry golf camera NW Set 2 cloth W spider star doctor NW hard sword speak butter dinner W likely bear gate busket jail queen model NW flower rose city town mxlU‘WO‘l-J H P‘ Ht- NL0\HHF4h uoswn03ma> H l-‘QmUiNN r- H brounmtnu) ll 15 Hr- mto¢~N~JPJ underlearned underlearned overlearned overlearned underlearned underlearned overlearned underlearned overlearned underlearned underlearned overlearned underlearned overlearned overlearned overlearned overlearned overlearned overlearned overlearned underlearned underlearned underlearned underlearned underlearned underlearned underlearned overlearned overlearned underlearned underlearned underlearned overlearned underlearned overlearned overlearned overlearned overlearned overlearned camera gretch paper opera horror thread white black cover blash park tofe picnic rug dentist smoke floor purple- grow masset moon star numbers letters bring sit gun sleep cover blash purse lince scissors cut neck eagle puzzle nail tree birch gate busket fish cod swing fast avenue order crowd prend camera gretch metal brass coffee orchid cloud golf grow masset numbers letters army soft rain chisel brush water purse lince 147 purse gretch needle thread cover W music opera white black bird robin picnic rug bed purple crowd tofe grow masset floor NW track smoke bring sit weapon knife numbers letters cover blash color sleep moon W tree NW doctor cut gun W purse lince hammer nail neck eagle park prend model busket fish cod wine order color blue blood fast blossom brass insect orchid camera W sport golf coffee NW grow NW numbers W ocean water cloth flannel weapon soft camera lince rain chisel wanomm H NNNU’ICDQ 10 13 14 11 12 overlearned underlearned overlearned overlearned underlearned underlearned overlearned underlearned underlearned overlearned underlearned overlearned overlearned underlearned overlearned overlearned overlearned underlearned overlearned overlearned underlearned underlearned underlearned overlearned underlearned overlearned underlearned overlearned underlearned overlearned overlearned underlearned underlearned overlearned underlearned underlearned overlearned overlearned gate busket hold short train grass money chair crowd prend park tofe doctor nurse speak talk clothing sweater spider web grow masset cloth linen rack cherry jail light model vade park tofe coat pepper foot shoe angry arson purse lince city town tree birch blood round blossom flower crowd prend cover blash youth tango memory bicycle supper beetle model vade scissors cut spider web bicycle woman school home camera gretch 148 gate NW house grass crowd W money chair train NW hold short metal talk army W spider NW scissors nurse clothing sweater park tofe cover masset model vade cloth linen memory light tool saw swing cherry fruit apple foot shoe purse W park NW crime arson coat pepper crowd prend dentist round city NW tree birch bread flower music rock avenue bicycle grow blash weapon beetle youth tango gate vade supper NW green home scissors W man woman school W moon web camera gretch 6 9 6 15 5 7 H moor-4.5mm [...a \lO‘WWl—‘Ih overlearned underlearned underlearned underlearned underlearned underlearned overlearned underlearned underlearned underlearned underlearned underlearned underlearned overlearned overlearned overlearned underlearned underlearned overlearned underlearned underlearned underlearned underlearned overlearned overlearned overlearned overlearned underlearned underlearned underlearned overlearned underlearned overlearned underlearned overlearned overlearned underlearned overlearned camera gretch dentist smoke fish cod cloud golf spider web crowd prend puzzle nail speak talk supper beetle gate busket purse lince neck eagle school home avenue order city town park tofe doctor nurse coat pepper grow masset cloth linen grow masset foot shoe blossom flower dinner sword cover blash moon star swing fast tree birch spider web spider web model vade hold short youth tango train grass crowd prend 149 Set 3 camera NW cloud golf neck W fish cod spider NW rapid smoke dinner beetle crowd prend camera busket metal talk cloth flannel puzzle nail weapon knife vehicle order purse lince gun home city W bird eagle coat NW grow W spider nurse cloth linen cover tofe salt pepper park NW grow masset bread flower hold W foot shoe army sword scissors web fruit fast youth NW tree birch moon W numbers NW tool saw dance tango crowd prend long short school grass model vade )- H ranomcnxnu hLDUHOFJb HF‘ H H (noxstu (bramrumuo woo NU‘IWGQQ 13 11 15 12 14 overlearned overlearned underlearned underlearned underlearned underlearned underlearned underlearned overlearned overlearned underlearned overlearned overlearned overlearned overlearned underlearned overlearned underlearned underlearned overlearned underlearned overlearned underlearned underlearned overlearned overlearned underlearned underlearned underlearned underlearned overlearned overlearned underlearned overlearned underlearned underlearned overlearned overlearned overlearned gate busket rain chisel brush water memory bicycle camera gretch model vade coffee orchid jail light clothing sweater camera gretch doctor nurse bring sit scissors cut city town cover blash cloth linene floor purple picnic rug angry arson park tofe purse lince bread butter blood round day woman grow masset purse lince gun sleep horror thread metal brass model vade gate busket paper opera wine queen track cherry park tofe 150 flower rose command bicycle rain chisel purse gretch brush water gate W model NW camera gretch king light bring NW 12 4 1 8 1 10 14 7 4 13 clothing sweater 5 supper orchid doctor NW picnic W city town stand sit cover W moon cut carpet rug music rock train purple angry arson cloth W park tofe day woman speak butter crowd masset model lince tobacco round fruit apple gate vade likely W horror thread blossom brass purse W floor sleep gate busket square cherry money NW lamp queen park NW paper opera 2 ...a NfimmmH H NQONHWCD F3 )- Naawbq overlearned overlearned overlearned underlearned overlearned overlearned overlearned overlearned overlearned underlearned underlearned overlearned underlearned overlearned overlearned underlearned underlearned underlearned overlearned underlearned overlearned underlearned overlearned overlearned underlearned overlearned underlearned overlearned overlearned underlearned overlearned overlearned overlearned overlearned overlearned underlearned overlearned underlearned overlearned numbers letters tree birch likely bear moon star cover blash camera gretch school home day woman spider web scissors cut model vade supper beetle memory bicycle youth tango cover blash crowd prend blossom flower blood round tree birch city town purse lince angry arson foot shoe coat pepper park tofe model vade jail light track cherry cloth linen grow masset spider web clothing sweater speak talk doctor nurse park tofe 151 bird robin animal bear numbers letters doctor star tree NW cover blash Set 4 camera gretch spider web school W rain woman scissors W house home supper NW gate vade picnic tango insect beetle grow NW memory bicycle music rock blossom flower numbers birch city NW blood round crowd prend bring pepper ocean arson park NW purse W fish shoe fruit apple track cherry tool saw jail light tree linen model vade cover masset park tofe white sweater spider NW doctor nurse army W speak talk H \JU‘O‘U‘Uohxl H thxouwxl H Hoounooow mNmI-‘NH H U1\OUIUIO\\) overlearned overlearned overlearned underlearned overlearned underlearned overlearned underlearned overlearned overlearned underlearned overlearned underlearned overlearned underlearned underlearned underlearned overlearned overlearned overlearned overlearned underlearned underlearned overlearned underlearned underlearned overlearned underlearned underlearned overlearned overlearned underlearned overlearned underlearned underlearned underlearned underlearned underlearned overlearned crowd prend money chair train grass hold short gate busket purse lince brush water rain chisel army soft numbers letters grow masset cloud golf coffee orchid metal brass camera gretch crowd prend avenue order swing fast fish cod gate busket tree birch puzzle nail neck eagle scissors cut purse lince cover blash gun sleep bring sit numbers letters moon star grow masset floor purple dentist smoke picnic rug park tofe 152 likely short train NW youth chair crowd W green grass gate NW music chisel camera lince hard soft cloth flannel sport water numbers W grow NW carpet NW needle golf camera W flower orchid metal brass swing fast color blue avenue order foot cod model busket park prend money eagle man nail purse lince gun W scissors cut tree NW man W bed sleep cover blash city letters weapon knife hold sit dentist smoke floor NW grow masset crowd tofe color purple neck rug mmO‘t-‘NH H r- H underutuhd tflUHO~tho wwooxle underlearned underlearned underlearned underlearned underlearned overlearned overlearned overlearned underlearned underlearned overlearned underlearned underlearned overlearned overlearned overlearned overlearned underlearned overlearned underlearned overlearned underlearned underlearned underlearned overlearned overlearned underlearned overlearned overlearned overlearned underlearned overlearned overlearned underlearned overlearned overlearned underlearned overlearned underlearned cover blash white black horror thread paper opera camera gretch model vade city town wine queen likely bear gate busket doctor nurse bread butter moon star dinner sword cloth linen 153 bird robin clothing black hammer opera cover W army thread purse gretch cloth town flower rose model NW wine queen gate busket coat bear dinner W bread butter weapon sword doctor NW moon star cloth W P4P -4Hubndm H I—‘U'IU'IUJU'ID H underlearned overlearned overlearned underlearned overlearned overlearned overlearned overlearned overlearned overlearned underlearned overlearned underlearned underlearned underlearned overlearned