Cognition Implicitify Research Team

Overgeneral Memory: Williams's AMT and What Personality Schemas Do to Recall

Most people, asked to remember a time they felt happy, will produce a specific event: a particular afternoon, a particular room, a particular other person. They will not say "I was happy on holidays" or "I used to be happy when I was younger." Those second kinds of answers — vague, summary, category-level — turn out to be diagnostically interesting. They mark a real and replicable phenomenon, and a sizable fraction of the cognitive psychology of depression and trauma over the last forty years has been organized around explaining it.

What Williams and Broadbent actually noticed

In the early 1980s, J. Mark G. Williams and Keith Broadbent were running a fairly simple paradigm: read participants a positively or negatively valenced cue word ("happy", "sorry", "successful", "lonely") and ask them to retrieve a specific personal memory associated with it. Specific was operationalized strictly: the memory had to refer to an event that occurred at a particular place and on a particular day, lasting less than 24 hours. Anything more general — "whenever I'm with my family", "in school", "I usually feel that way around my brother" — was scored as categoric, a summary across many similar episodes. Anything spanning a longer interval — "during my twenties", "the year I lived in Boston" — was scored as extended.

The 1986 paper Autobiographical memory in suicide attempters (Williams & Broadbent, Journal of Abnormal Psychology, 95, 144–149) reported the finding that defined the field. Suicide attempters, compared to controls, retrieved markedly more categoric memories — and the deficit was not limited to negative cue words. They were slow and overgeneral on positive cues too. The pattern was not "depressive bias toward negative content." It was a more interesting failure of the retrieval process itself.

The finding generalized. Over the next decade, overgeneral memory (OGM) was documented in major depression, in PTSD, in complicated bereavement, and in survivors of childhood abuse who did not currently meet diagnostic criteria for any disorder. Williams's Cognitive Psychology and Emotional Disorders (1988, second edition with Watts, MacLeod, and Mathews 1997) made the case that OGM was not a state-marker that came and went with mood; it was closer to a stable cognitive style that predicted future depression rather than merely accompanying current depression.

The CaR-FA-X model: why retrieval gets stuck at the categoric level

Williams's mature theoretical statement — Autobiographical memory specificity and emotional disorder (Williams, Barnhofer, Crane, Hermans, Raes, Watkins, & Dalgleish, 2007, Psychological Bulletin, 133, 122–148) — proposed that overgeneral retrieval is the joint product of three mechanisms, summarized in the CaR-FA-X acronym.

Capture and Rumination (CaR). When a cue activates a self-referent concern, the retrieval process is captured at the higher, conceptual layer of the autobiographical knowledge base — what Conway and Pleydell-Pearce (2000) called general events and lifetime periods — and recycled there ruminatively, rather than descending to the specific episodic detail.
Functional Avoidance (FA). Stopping retrieval at the categoric level reduces the affective intensity of the memory. For people with histories of trauma or repeated affective injury, this is reinforcing: an overgeneral memory hurts less than the specific one would.
Reduced Executive control (X). Descending from the conceptual layer to a specific episode requires working-memory and executive resources. When those are taxed — by depression, by acute distress, by ongoing stress — retrieval stalls before specificity is reached.

The strength of this account is that it predicts something the older "negative-bias" account did not: OGM should appear across cue valences, in non-depressed populations with histories of trauma, and should be reducible by interventions that build executive control or interrupt rumination. All three predictions have held up reasonably well. Memory specificity training (MEST) and certain mindfulness-based interventions modestly improve specificity (Raes, Williams, & Hermans, 2009; Hitchcock, Werner-Seidler, Blackwell, & Dalgleish, 2017), and overgeneral retrieval prospectively predicts depressive episodes in initially non-depressed at-risk samples (Sumner, Griffith, & Mineka, 2010).

What the AMT became as a clinical instrument

The administration that emerged from this program — usually called the Autobiographical Memory Test (AMT) — is austere. Ten to twenty cue words, balanced for valence (positive, negative, sometimes neutral). One memory per cue, with a 30- or 60-second response window. Responses scored by trained raters into the four-category Williams & Broadbent taxonomy (specific, categoric, extended, semantic-associate / omission). The primary outcome is the proportion of responses meeting the strict specific criterion.

Inter-rater reliability for the AMT is consistently high (κ values in the .80s and above), and test-retest reliability is adequate over weeks (Williams et al., 2007). Convergent validity with other indices of overgeneral retrieval is strong; discriminant validity from depressive content bias measures is one of the points the literature has worked hardest to establish, and is now reasonably well-supported.

The core limitation is bandwidth. Standard AMT cues are valenced but not theoretically targeted. They tell you whether retrieval is overgeneral in general; they do not tell you whether the impairment is concentrated around any particular set of self-relevant concerns. For a measure that began as a probe of cognitive vulnerability to depression, that was the right design choice. For a measure used to characterize personality, it is a missed opportunity.

Where personality schemas come in

The bridge from AMT-style retrieval to personality assessment runs through the schema concept. The schema-therapy tradition (Young, Klosko, & Weishaar, 2003) and the broader cognitive personality literature converge on a basic claim: people organize self-relevant experience around a small number of recurrent thematic patterns — abandonment, defectiveness, mistrust, subjugation, entitlement, and so on. These patterns are not just beliefs; they are retrieval frames. When a stimulus activates a schema, the autobiographical knowledge base preferentially returns memories that fit the frame.

Two empirical observations make this assessment-relevant. First, schema activation appears to intensify the overgeneral memory effect: cued retrieval slows down and gets vaguer when the cue resonates with an active schema (Spinhoven et al., 2006; Sumner, 2012). Second, the pattern is selective. A person organized around abandonment-and-mistrust may retrieve specifically and fluently to neutral and achievement-coded cues, but produce categoric, extended, or semantic responses to relational-loss cues. The selective nature of the impairment is informative in a way that a global specificity ratio cannot be.

This is the design idea behind a schema-congruent AMT. Instead of generic positive and negative cues, the test selects cue words organized by personality domain — Avoidant, Dependent, Narcissistic, Antisocial, Compulsive, Histrionic, Borderline, in the Millon-derived taxonomy used by the AIMS — and matches each schema-congruent cue to a control word equated on lexical properties (frequency, length, lexical-decision response time) using the English Lexicon Project database. Specificity is then computed both globally and as a differential: the gap between specificity to neutral controls and specificity to schema-congruent cues. A positive differential indicates schema-driven retrieval impairment; the size of the differential, broken down by domain, indicates which schemas are doing the work.

The CCRT-style coding of the content of any specific interpersonal memory the participant does retrieve is a natural extension of the same logic. Where the AMT asks "did the person reach a specific episode," CCRT scoring asks "if so, what was it about." The two layers of analysis are independent and complementary.

What it does and does not buy you

A schema-congruent AMT-derivative is not a clinical diagnosis. It does not tell you whether the participant meets criteria for any disorder, and it is not a replacement for a structured interview or for self-report personality measures. What it does is give you a behavioral, performance-based read on a question that self-report has trouble answering directly: which thematic content interferes with the participant's ability to descend from general beliefs about themselves to the specific episodes those beliefs are supposed to summarize.

That is a modest claim, and it is the right one to make. Overgeneral memory is one of the more reproducible findings in clinical cognitive psychology, but it is not a master key. It tells you something about how a person's autobiographical knowledge base is organized and how readily it can be queried. Combined with self-report personality measures, with attachment instruments, and with implicit motive coding, it is a useful piece of a multimethod profile — and it is a piece that no questionnaire can substitute for.