Human recall of items in a specified category (for example, breeds of dogs) follows a well-established probabilistic structure in which more familiar items tend to be recalled earlier. This paper proposes a minimal generative technique that reproduces human recall order patterns using probabilistic sampling combined with a real-time deduplication process, and presents a computational model for evaluating this approach. When familiarity is represented as frequency (duplicates), deduplication-based sampling produces human-like recall order, with more familiar items more likely to appear earlier in the output. When averaged across many runs, the model’s ranking converges closely to that predicted by a frequency-weighted recall order expectation. These results suggest that a simple generative process can account for key features of human recall order and provide a useful baseline for modeling and analysis. v19: Changed “simulation” to “model,” reworded the introduction, and corrected typos.
John M. Smith (Sun,) studied this question.