Untitled

Levels of Processing: A Retrospective Commentary on a Framework for Memory Research

Main Takeaway: The "Levels of Processing" (LoP) framework (Craik & Lockhart, 1972) revolutionized memory research by shifting focus from structural memory stores to procedural encoding operations. Despite conceptual and empirical criticisms over 18 years (since 1972), the core ideas have largely survived, been refined (e.g., elaboration, distinctiveness), and have successfully stimulated a vast data-base for future memory theory construction.

I. Introduction and Overview • Source: Canadian Journal of Psychology, 1990, 44(1), 87-112. • Authors: Robert S. Lockhart and Fergus I.M. Craik, University of Toronto. • Purpose of Article: • Review the influence of LoP on memory research. • Evaluate conceptual and empirical criticisms. • Refine the original formulation of depth of processing through concepts like elaboration and distinctiveness of encoding. • Clarify misconceptions about the original article (e.g., abolishing STM/LTM distinction). • Original Framework (Craik & Lockhart, 1972): Presented a general approach to memory research, widely influential, and subject to critical scrutiny. • Goal: Displace the "structuralist" theory of memory (items in stores, refined by number/properties of stores) with a procedurally oriented one. • Objection to Stores: Variable properties (capacity, coding, forgetting rates) across experiments led to postulating new stores for new findings, deemed unprofitable. • Advocacy: Look directly at relations between encoding operations and memory performance. • Core Proposal: • Cognitive system is hierarchically organized; stimuli processed to different levels of analysis. • Early (shallow) sensory analyses serve as input to later (deeper) semantic analyses. • Deeper, more semantic analyses yield more durable memory records. • Two Further Ideas from 1972 Article: 1. Rehearsal Types: • Type I (Maintenance Rehearsal): Maintains processing at the same level. • Type II (Elaborative Rehearsal): Involves deeper or more extensive processing. • Only Type II should improve memory if performance is a function of the deepest analysis achieved. 2. Primary Memory (PM): Retained but viewed as continued processing activity, not a separate mechanism or structure. II. Foundation of Levels of Processing • Underlying Principle: Memory trace is a by-product or record of normal cognitive processes (comprehension, categorization, discrimination), not a specialized memory-encoding process. • Common Sense Starting Point: Influencing the form of perceptual or conceptual analysis performed on material significantly affects remembering, even when traditional determinants (presentation rates, serial position, test form) are constant. • Absence of a "Memorizing" Process: • No distinct "committing to memory" process alongside other cognitive operations. • Intentional instructions ("try to remember") merely prompt subjects to choose processing strategies. • Different processing forms have differential consequences for subsequent memory, but none is a "basic process of committing to memory." • Immediate Implications: 1. Preference for Incidental Orienting Tasks: • Intentional instructions reduce experimental control by confounding subject's strategy choice and its consequences. • Orienting tasks provide better experimental control, operationally defined independent variables, and a check on processing. • Ecologically valid approach: Laboratory paradigms should capture features important for everyday adaptive cognition. 2. Memory as Retrieval: • The only cognitive operation legitimately called "remembering" is retrieval. • LoP places retrieval firmly at the center of memory research. 3. Memory Structures and Systems: • Should not be defined by retention characteristics (e.g., short-term memory as a store). • If trace durability is determined by initial processing, retention varies continuously; discrete, temporally defined memory systems become unnecessary. • Short-term memory is discussed in relation to working memory, which is considered theoretically different. 4. Memory and General Cognition: • An adequate memory theory must be constructed within a general theory of cognition. • Memory cannot be a self-contained module if its determinants serve general cognitive goals. • Key: Formulate principles linking remembering to the full scope of cognitive processing (captured by depth of processing concept). III. Critical Assessments of Levels of Processing • Reasons for Appeal: Dissatisfaction with structural limitations of memory stores, receptivity to process-oriented conceptualization. • Criticisms (e.g., Baddeley, 1978; Eysenck, 1978; T.O. Nelson, 1977): • Group 1: Conceptual and Methodological Issues: • Circularity: Defining depth by memory outcome (deeper processing = longer-lasting traces, but well-remembered events assumed deeply processed post hoc). • Lack of Independent Index of Depth: Primary criticism, makes falsification difficult. • Value of General Principles: Baddeley (1978) questioned seeking general functional principles, advocating for specific, experimentally identifiable mechanisms (linked to brain structures). • Linear Sequence Assumption: Baddeley questioned the assumption of one linear sequence of levels, noting lack of evidence for further levels within semantic domain. • Unfalsifiability/Vagueness: Ideas deemed not falsifiable and vague by T.O. Nelson (1977). • Group 2: Empirical Issues: • Two Types of Rehearsal: T.O. Nelson (1977) and Baddeley (1978) questioned if maintenance processing strengthens memory. • Long-lasting Sensory Traces: Craik & Lockhart (1972) claimed rapid loss of sensory info; evidence grew for long-lasting sensory traces. • Neuropsychological Studies: Amnesic patients comprehend deeply but have poor memory, seemingly at odds with LoP. IV. Conceptual and Methodological Issues (Response to Criticisms) A. Circularity and the Need for an Independent Index of Depth • Partial Justification: • Qualitative distinctions (semantic vs. phonemic processing) can be made independently of memory effects. Hypothesis that semantic processing > phonemic processing for recall is falsifiable (e.g., Craik & Tulving, 1975). • Circularity arises more with within-domain differences (e.g., "yes/no" answers to semantic questions, or distinct semantic tasks like synonymity vs. self-description) where an independent index is lacking other than post hoc memory performance. • What an Independent Index Would Look Like: • Measures like processing time, effort, or neurophysiological data are possibilities. • Validation Problem: Cannot validate by monotonic relation to memory performance (circularity). • Not Unidimensional: Depth was never envisaged as a simple, unidimensional continuum with a monotonic relation to memory. • End Product, Not Starting Point: A complete definition and quantification of "depth" or "level" is the end product of research within a general theory of cognitive functioning, not its starting point. Original terms were tentative labels. • Self-Reference Orienting Tasks (Rogers, Kuiper, & Kirker, 1977): • High performance relative to other semantic tasks shows complexity of semantic processing/memory relations. • Attributing superiority simply to "deeper processing" is a weak explanation; LoP served as a starting point for deeper analysis. • LoP framework stimulated this data collection, stressing explanation through analysis of meaning-extracting operations. • Example of memory research integrating into broad cognition (e.g., research into self-concept structure). B. Principles and Mechanisms • Baddeley's Criticism (1978): Abandon broad principles for specific mechanisms (e.g., articulatory loop within working memory). • Authors' Response: Both principles and mechanisms are needed. • Mechanisms must be within a larger cognitive architecture embodying general principles. • Earlier STM models failed due to isolation from general cognition (e.g., rehearsal as simple item transfer). • Overemphasizing isolated mechanisms (e.g., Sternberg paradigm scanning) leads to studying the mechanism itself, not its relation to general cognition. • Principles Without Mechanisms: General principles (like depth of processing) guide data gathering but don't constitute a complete theory or mechanism. • "Deep processing" isn't a mechanism directly influencing memory. • Mechanisms (procedural, not structural at cognitive level) linking orienting tasks to retrieval remain to be established. C. Levels: A Fixed Series of Stages? • Baddeley's Criticism (1978): Unreasonable assumption of a single linear sequence. • Authors' Response: • Original statement may have encouraged simplistic interpretation, but authors rapidly corrected this (Lockhart et al., 1976). • Validity of framework doesn't depend on a specific model of sequencing. • Adequate model likely involves complex interactions between top-down and bottom-up processes, with parallel/overlapping processing. • What's important is the pattern of analyses finally achieved, not strict sequencing. V. Empirical Issues (Response to Criticisms) A. Types of Rehearsal • Original Suggestion (1972): Maintenance rehearsal does not increase retention, only elaborative rehearsal does. • Contradiction: Atkinson & Shiffrin (1968) model predicted residence in rehearsal buffer transferred info to LTM. • Empirical Support (early): Craik & Watkins (1973), Woodward, Bjork, & Jongeward (1973) supported maintenance having no effect. • Complications (later): • Woodward et al. (1973) also showed maintenance rehearsal increased recognition. • T.O. Nelson (1977) showed repeated phonemic processing improved recall. • Greene's (1987) Conclusion: • Pure maintenance rehearsal leads to small increases in recall (inconsistent data). • Clear-cut evidence for increased recognition memory. • Possible Explanation: Recall depends on interitem elaboration, recognition on intraitem integration (Mandler, 1979). Maintenance rehearsal has little interitem processing. • Naveh-Benjamin & Jonides (1984) Suggestion: Maintenance rehearsal is two-stage: 1. Assembling items, setting up articulatory program (memory enhanced). 2. Repetitive execution of program (little effort, no further enhancement). • Recent Work (Greene, 1987): Distinguish "repetition of items" (clear memory increase) from "rote maintenance" (weak changes). • Maintenance of Linguistic Items: Enhances acoustic/articulatory properties (Elmes & Bjork, 1975; Geiselman & Bjork, 1980). Retrieval effectiveness depends on qualitative rehearsal type. • Summary: Rehearsal-memory relations are more complex than 1972, but original qualitative distinction was on "right lines." Fruitful research resolved inconsistencies. • Footnote: Naveh-Benjamin & Jonides's second stage of maintenance rehearsal resembles Baddeley's articulatory loop (where learning isn't expected). B. Domains and Levels Within Domains • Baddeley's Criticism (1978): Unclear how levels within phonemic/semantic domains are specified. • Authors' Concession: "Reasonable criticism." • Lockhart et al. (1976) suggested "depth" covered two concepts: 1. Qualitatively different domains (shallow sensory preceding semantic). 2. Further analyses within a domain (here, elaboration better captures the essence). • Further Semantic Processing: • Elaboration: Greater enrichment/elaboration of encoding leads to enhanced memory (Anderson & Reder, 1979; Craik & Tulving, 1975). Operationally defined by number of semantic decisions (Johnson-Laird et al., 1978; Ross, 1981) or effort (Tyler et al., 1979), though qualitative type of encoding is key. • Example: Nouns encoded with core-meaning adjectives recalled better (Hashtroudi, 1983). • Distinctiveness: Key concept for good retention. • Words categorized specifically (e.g., "vegetables") recalled better than generally (e.g., "foods") (Frase & Kamman, 1974). • K. Klein & Saltz (1976) suggested distinctiveness as mechanism for depth effects. • Moscovitch & Craik (1976) linked encoding uniqueness to later retrieval benefit, dependent on retrieval cue compatibility. • Developed by Bransford et al. (1979) and Stein (1978) as precision of encoding, relation to existing knowledge, and cue availability. • D.L. Nelson (1979) argued depth unnecessary if distinctiveness and interactiveness are considered. • Conclusion: Original "depth" too simple for within-domain processing. Many studies stimulated by LoP added richness/precision to memory knowledge. C. Long-Term Retention of Sensory Features • Original Claim (1972): Shallow (sensory) processing leads to transitory memory traces, in line with sensory registers. • Contrary Evidence (since 1972): Sensory information persists for hours, minutes, even months (Kirsner & Dunn, 1985). • Authors' Concession: Original suggestions were too simple. Surface information can persist. • Comments from LoP Perspective: 1. Implicit Use: Some long-term retention of surface form involves implicit use (priming, data-driven, modality-specific), not explicit retrieval (e.g., Kolers, 1979). This conflicts with original statement. 2. Recall of Modality vs. Events: Recalling input modality is different from recalling specific events; modality recall is often choosing between general types (visual/auditory). 3. Modulation/Expertise: Surface form may modulate verbal info (e.g., speaker A vs. B), or be transformed into deep info through expertise (novice wine-tasting vs. expert). "Deep" wasn't restricted to linguistic meaning. 4. Picture Memory: Rapidly processed to deep level, well-retained. 5. Intraub & Nicklos (1985): Better picture memory after physical questions than semantic, explained by enhanced distinctiveness. 6. Emotional Utterances: Excellent verbatim memory (Keenan, MacWhinney, & Mayhew, 1977). Possible explanations: weaker "flashbulb memories," self-referential processing. • Conclusion: More surface information is retained than believed in 1972. A full account is still developing, potentially involving different procedural memory systems or deep nonlinguistic processing. VI. Further Conceptual Developments A. Depth, Elaboration, and Distinctiveness • Depth of Processing: Initial attempt to framework qualitatively different processing forms to memory performance. • Fundamental claim: Analyses organized by degree of meaning extraction. • Never intended as a fixed, well-defined concept; more of a "common sense" statement. • William James: "The art of remembering is the art of thinking ... to connect it with something already there" (James, 1983, p. 87). • Marker Orienting Tasks: Early LoP experiments used these to demonstrate effects, not as definitive definitions of levels. • Misconception: Tendency to treat these tasks (case, rhyme, semantic) as having privileged status, defining discrete levels. • Criticism: D.L. Nelson (1979) argued tasks are "impure representations." This is true (e.g., Stroop effect implies semantic processing even in phonemic tasks), but only relevant if studying specific cognitive operations in isolation. • Justified Use: Marker tasks are justified to represent a wide range of processing levels to examine relative sensitivity of different memory forms (e.g., implicit memory unaffected by task variations). • Criticism: Against treating levels as discrete classes with equivalent processing within each. • Elaboration (Craik & Tulving, 1975): • Discovery: "Yes" responses to orienting questions yielded higher memory than "no" responses. • Interpretation: Compatible (positive) questions enriched/elaborated the encoded trace. Richly elaborated traces support better recall/recognition. • Metaphor: "Spread of encoding." Some suggest "extensiveness of processing" is better than "depth." Others (Anderson & Reder, 1979) argue elaboration is the only necessary concept, as deeper levels offer more scope for it. • Authors' View: Both depth and elaboration are necessary. • Depth: Refers to qualitatively different types of processing (e.g., sensory precede conceptual, require fewer resources). • Elaboration: Refers to richness or extensiveness of processing within a qualitative domain (e.g., proof-reading vs. reading for gist). • Transfer Appropriate Processing (Morris, Bransford, & Franks, 1977): • Argument: Memory performance depends on match between encoding and retrieval operations. Sensory processing can be as good as semantic if retrieval conditions are appropriate. • Authors' Endorsement: Strongly endorse this principle (similar to encoding specificity, repetition of operations). • Caveat: While appropriate processing improves memory for specific types of information ("apple-memory for apple-processing"), when comparing general memory performance (recall/recognition of an event), some types of initial processing are still superior, and these involve deeper, semantic analyses. Example: "semantic-yes" processing yielded .68 recognition, while "rhyme-yes" yielded .40 for appropriate tests. • Distinctiveness: • Tied to elaboration (Craik, 1977; Jacoby & Craik, 1979; Moscovitch & Craik, 1976). • Key determinant of memorability (Murdock, 1960). • Mechanism: A distinctive memory trace "stands out" and is readily retrievable, especially when contrasted with dissimilar encoded events. • Mechanism for Elaboration-Memory Link: • Anderson (1976): Elaboration generates redundant network of encoded info, enhancing retrieval. • Ausubel (1962): Depth/elaboration produce distinctive encodings against well-learned meaningful knowledge. • Why deep elaboration > surface elaboration? Latter's elements are unrelated; meaningful schematic knowledge allows interpretation of stimulus as unitary and reconstruction at retrieval. • "Types of Processing" vs. "Levels/Depth": • Criticism: Should only talk about different types of processing, not hierarchical levels. • Authors' Argument: Still necessary to talk about "depth." • Some analysis of physical features needed before meaning. • Incoming patterns often comprehended linguistically before abstract analyses. • Not all patterns are linguistic (e.g., visual scenes). • Alternative Notion (Lockhart et al., 1976): Analysis proceeds in domains of knowledge, tied to sensory modalities for shallow analyses, and proceeding to deeper levels within that specific domain. • Some deep analyses restricted to domain (wine-tasting expertise). • Others (language) have rich cross-modal connections (e.g., Paivio's dual coding). • Conclusion: "Deeper processing" is a sensible notion, referring to greater involvement of interpretive and implicative processes within the relevant domain. Greater knowledge/expertise enable deeper processing and superior memory. Not restricted to linguistic domain. B. Levels of Processing, Short-Term, and Working Memory • Criticism of STM: Directed at its structural incarnation in "box and arrow" models (e.g., Atkinson & Shiffrin, 1968), which reify memory structures and lead to unanswerable questions (capacity, decay rate). • Capacity of STM: • Problem: Capacity (quantitative) needs appropriate units. Measures (items, chunks, bits) failed to establish universal constant. • Reason: Capacity is dependent on processing skill/resources of subject and material properties (Jenkins, 1979). Varies widely (e.g., Chase & Ericsson, 1981, 1982 subject with digit span of 80 but unchanged letter span). • Redefinition: To maintain constant item capacity, "item" would need to be redefined by processing operations, which is unclear and theoretically weak. • Authors' Argument: Limited capacity of STM is a consequence of limited processing resources, not an invariant structural property. Appropriate units are processing resources, not memory information. • Example: Chase & Ericsson's subject acquired processing skills for deep processing of numbers, not structural capacity increase. Integrates memory with skilled performance and attention theories (Kahneman, 1973). • Rate of Decay or Loss From STM: • Defining characteristic of STM was rapid decay/expulsion (parametric approach, quest for universal constants, e.g., Ebbinghaus, Brown-Peterson paradigm). • Authors' Argument: Possible to produce virtually any forgetting function by varying factors (item exposure, type, subject skills, instructions). • Conclusion: If memory structures defined by forgetting, then there's a continuum of systems. Either discard "memory systems" or find another basis (like Baddeley's working memory). • Working Memory: • Definition: Set of processes/structures for temporary holding, manipulation, and integration of information. • Compatibility: Largely orthogonal to LoP (encoding/LTM relations). Authors have "no quarrel" and believe it's a necessary concept. • Clarification: LoP did not abolish STM/LTM distinction. It argued against the structural notion of memory stores. • PM (1972 paper, p. 676): "Superimposed on this basic memory system there is a second way in which stimuli can be retained by recirculating information at one level of processing... refers to the same concept of maintaining information at one level of processing." Synonymous with James's (1890) PM (items in consciousness). • Differences from Short-Term Store: 1. PM is a processing activity, not a structure. Limited capacity is a processing limitation, not "space." 2. PM is not one "place." Represents deployment of processing resources to various parts (phonological, pictorial, semantic). "Puzzle of what constitutes the short-term code dissolves." (Shiffrin, 1976 had similar view: STM as active subset of LTM). • Baddeley's Working Memory (1974, 1986): • Shares features: processors, not passive stores; deals with various info types. • Differences: Baddeley's appears more structural, located in one place (Central Executive coordinating slave systems like articulatory loop, visuospatial sketchpad). Authors question if Central Executive capacity/prowess can be independently measured, risking similar criticisms (vague, poorly defined) as LoP faced. • Authors' Preference: Flexible PM within LoP aligns more with domain-specific view of working memory (Monsell, 1984; Daneman & Tardif, 1987). • Working memory as a collection of computational abilities intrinsic to specialized subsystems. • Control: Monsell suggests higher-order processors monitor others. Craik & Lockhart equate PM with attention/processing resources. Executive control remains a puzzle. • Summary on STM Functioning: 1. Did not deny STM/LTM distinction; endorsed it, seeing PM as flexible allocation of processing resources. 2. Capacity limitations are processing limitations, not structural. 3. Working memory is compatible; prefer domain-specific view over Baddeley's model. C. Levels of Processing and Retrieval • Criticism: Neglect of retrieval. • Authors' Response: • Partially justified as incompleteness (addressed in Lockhart et al., 1976). • Unfounded if claiming retrieval unimportant; LoP assigns retrieval a dominant role as the only distinctly memory process. • Key Issue: Is depth effect wrong, and superiority of semantic processing an artifact of retrieval conditions (e.g., better matched)? • Transfer Appropriate Processing (TAP): Argument that sensory processing can be remembered as well as semantic if retrieval conditions are appropriate (Morris et al., 1977). • Authors' Acknowledgment: • General principle that memory determined by encoding-retrieval interaction is undoubtedly true (Tulving's encoding specificity). • Crucial Question: Are there main effects of encoding depth even when transfer appropriateness is constant? • Answer: Yes, a main effect of depth exists (as discussed under elaboration). • How this Effect Operates (Two Possibilities for Retrieval): 1. Decontextualization: Deep processing (relative to shallow) sustains retrieval over a wider range of retrieval conditions/contexts. Retrieval becomes (relatively) robust against context changes. Research question: establish how encoding operations affect contextual robustness. 2. Retrieval Boosting (Context-Dependent): Deep processing boosts retrieval, but this boosting is highly dependent on retrieval context. Levels effects are reduced if retrieval conditions mismatch too much. • TAP is correct, but the performance increment for appropriate deep processing > increment for appropriate shallow processing. • Fisher & Craik (1977) support: Semantic processing with semantic cues yielded higher recall than rhyme processing with rhyme cues. • Some deep encoding may make retrieval relatively context-independent. VII. Concluding Comments A. The Complementary and Interactive Roles of Data and Theory • Misunderstanding of LoP: Interpreted narrowly as a hypothetico-deductive theory for immediate experimental test. • Problem with Narrow Theory Testing: Often generates data with little archival value, discarded when theory is discarded, hindering cumulative data-base building. • Authors' View: At current stage of memory understanding: • Major role of theory: Guide data gathering. • Data role: Evaluate theories and determine lifespan. • Theory influence on data: Help determine if data contribute to a cumulative data base that guides subsequent theory. • Theories must leave behind "stepping stones" for better theories. B. The Influence of the Levels of Processing Framework • Baddeley (1978): LoP "rapidly approaching obsolescence" (p. 146). • White (1983): LoP article "undoubtedly has had the greatest influence of any single contribution published in the 1970's" (p. 426). • Beyond Citations: Influence reflects a "spread of effect" beyond traditional memory research. • Wide Range of Applications (non-exhaustive list): • Pharmacological studies (e.g., Malone et al., 1988) • Psychophysiological studies (e.g., Sanquist et al., 1980) • Hemispheric function (e.g., Leong et al., 1985) • Selective attention (e.g., Tsuzuki, 1986) • Cognitive deficits (e.g., Boyd & Ellis, 1986) • Reading and prose comprehension (e.g., Friederici, 1985) • Educational psychology (e.g., Watkins, 1983) • Child development (e.g., Owings & Baumeister, 1979) • Time perception (e.g., Arlin, 1986) • Heuristic Success: Framework stimulated gathering and interpretation of new data. • Evolution of Ideas: Authors acknowledge many ideas changed since 1972, and views were refined due to 17+ years of research and thousands of experiments. This openness to change is core to the framework's spirit. • Survival of Original Ideas: • Memory performance is directly and strongly linked to initial processing. • Adequate memory theory needs analysis of processing operations. • Abandonment of static structural models. • Memory theory must integrate into general cognitive theory. • Conclusion on Success: LoP framework has achieved substantial success in promoting empirical research and theory development that yields a cumulative data base for future theories. file:///private/var/mobile/Containers/Shared/AppGroup/F090F121-F440-4B29-BBB8-F15BF0519A87/share/ComposeTempFiles/43674