Abstract: ionist Positions Abstractionist positions view implicit memory as reflecting modification of the state of abstract lexical, semantic, or procedural knowledge structures; byionist positions view implicit memory as reflecting modification of the state of abstract lexical, semantic, or procedural knowledge structures; by contrast, explicit memory is assumed to depend on formation and retrieval of memory traces representing specific experiences. Abstractionists are often neuroscientifically oriented, using brain lesion data to constrain their theories. Of particular interest are findings that amnesic patients are selectively im paired on direct tests of memory but show normal learning as measured by some indirect tests (reviewed below). These deficits are ascribed to an impairment of the system responsible for memory of specific experiences. In neuroscience, modularity of systems is the rule rather than the exception: If one accepts a straightforward relationship between brain systems and cogni tive systems, the hypothesis of multiple memory systems is a logical exten sion of current knowledge (Cohen 1984; Oakley 1983; for discussion, see Cohen 1 985; Olton 1985; Schacter 1986) . Tu1ving's (1972) heuristic distinction between episodic and semantic forms of memory was later developed into a multiple-system theory (Schacter & Tulving 1982a, b; Tulving 1983, 1984a) . Episodic memory "deals with unique, concrete, personal, temporally dated events," while semantic mem ory "involves general, abstract, timeless knowledge that a person shares with others" ( 1986, p. 307). In recent versions of the theory (Tulving 1984b,c; 486 RICHARDSON-KLA VEHN & BJORK 1985a,b,c; 1986) episodic memory is viewed as a specialized subsystem of semantic memory, with both systems embedded within a procedural memory, an arrangement Tulving terms monohierarchical. This position was designed to facilitate conceptualization of Tulving's (1983) hypothesis of the phylogenetic evolution of episodic from semantic memory, and to account for recent arguments that whereas indirect memory measures reveal evidence of memory early in human ontogeny, the capacity to perfonn direct memory tasks first emerges only at 8-9 months of age (Schacter & Moscovitch 1984) . There are a number of other multiple-system formulations that are some what analogous to the episodic-semantic distinction (e.g. Halgren 1984; Johnson 1983 ; Oakley 1983 ; O'Keefe & Nadel 1978; Olton et al 1979; Schacter & Moscovitch 1984; Warrington & Weiskrantz 1982) . One of these deserves special note: Morton's (1969, 1970, 1979, 1981) mUltisystem theory differs from Tulving's in that conceptual and factual knowledge, as well as personal or episodic memories, are dealt with by the same system (the cognitive system); a separate system contains abstract representations for words (logogens) that are responsible for lexical access. Other theorists regard the distinction between procedural and declarative (or propositional) memories (accepted by Tulving) as sufficient to explain the observed dissociations between direct and indirect measures (e.g. Baddeley 1984; Cohen 1984; McKoon et al 1986; Squire & Cohen 1984) . The pro cedural/declarative distinction was originally formulated by workers in artifi cial intelligence as a distinction between types of knowledge (e .g . Barr & Feigenbaum 1981; Winograd 1975) , but it has been extended into a multiple system viewpoint (Cohen 1984; Squire & Cohen 1984) . Procedural memory involves "reorganization or other modification of existing processing struc tures or procedures ," whereas declarative memory "represents explicitly new data structures derived from the operation of any process or procedure" (Cohen 1984, pp. 96-97). Although procedural memory can be revealed only when a task reengages prior processing operations, it is abstract in the sense that it does not record the specific prior events that caused those processing operations to be modified. The declarative system is considered to be respon sible for conscious access to facts and past experiences; it is necessary for performance of direct memory tests, and is impaired in amnesia. An approach analogous to the procedural/declarative distinction is proposed by Mishkin (Mishkin et al 1984; Mishkin & Petri 1984) , who distinguishes between a memory system and a habit system. The distinction between activation and elaboration (Graf & Mandler 1984; Mandler 1980; Mandler et al 1986) is a process-oriented viewpoint; it differs from other abstractionist positions in being neutral with respect to the issue of memory systems. Activation of a preexisting mental representation "strength ens the relations among its components and increases its accessibility" (Graf MEASURES OF MEMORY 487 & Mandler 1 984, p. 553); elaborative processing is necessary in order to retain new relationships and relate stimuli to the context in which they were presented. Activation alone is sufficient to result in processing facilitation that is revealed in indirect memory tests, whereas elaboration is necessary for direct tests of memory . A similar concept of trace activation has been proposed by Diamond & Rozin (1 984; see also Mortensen 1980) . Nonabstractionist Positions Nonabstractionists are unified by their disagreement with the necessity to distinguish abstract representations from memory traces that preserve in formation from specific experiences. They are typically mainstream cognitive psychologists who concern themselves primarily with the behavior of normal human subjects. Kolers (e.g. 1979, 1 985; Kolers & Roediger 1984; Kolers & Smythe 1 984) has attacked the distinction between procedural and declarative (or proposi tional) knowledge, arguing that "statements or declarations . . . do not fail of procedural representation" (Kolers & Roediger 1984, p. 437). When a subject displays knowledge, he/she is assumed to be engaging in a form of skilled performance. Knowledge is regarded as being specific to the processes by which that knowledge is acquired: Rather than offering a "unitary" theory in opposition to multisystem approaches, Kolers suggests that mentation con sists of a multiplicity of processes whose properties are poorly correlated. Memory is revealed to the extent that processing operations at study and test overlap (the principle of transfer-appropriate processing, Bransford et al 1 979) . Dissociations between direct and indirect measures of retention are to be expected when members of the two classes of task make different process ing requirements , and not otherwise [see also Moscovitch et al ( 1986) for a similar viewpoint]. Jacoby ( 1983b) , Blaxton ( 1 985), and Roediger & Blaxton ( 1 987a,b) have used the terms conceptually driven and data-driven as a taxonomy of the processing demands of memory tests. Direct memory tests typically involve more conceptually driven than data-driven processing because the subject uses associative information to reconstruct the study episode mentally; in direct memory tests typically involve more data-driven than conceptually driven processes because the subject focusses on external stimuli (e.g. a fragment of a word) at test. Dissociations between data-driven and con ceptually driven memory tests would be expected as a function of the type of information (semantic-associative vs perceptual) encoded in a prior episode. Jacoby ( 1982, 1983a,b, 1984, 1 987; Jacoby & Brooks 1984; Jacoby & Dallas 198 1 ; Jacoby & Witherspoon 1982) argues that implicit and explicit memory are reflections of "different aspects of memory for whole prior processing episodes" ( 1 983a, p. 21) . The aware and unaware aspects of 488 RICHARDSON-KLA VEHN & BJORK memory for episodes are assumed to result from differences in information provided by test cues, and possibly accompanying differences in retrieval processes (see Whittlesea 1 987 for a similar position).