调整熟悉化程序后，成人错误信念 anticipatory-looking 任务出现稳定的信念一致预期注视

Humans and other animals predict others’ actions from mental states, even when those states misrepresent reality. The anticipatory-looking false-belief (ALFB) task assesses belief-congruent action prediction without verbal responses, yet its reliability remains debated. Across six experiments, we tested 302 adults using false-belief events embedded in rich nonverbal narratives. We progressively refined familiarization while keeping the critical false-belief induction and test events unchanged. Reliable belief-congruent anticipatory looking emerged after familiarization was restructured, possibly because this reduced reliance on spatial heuristics and encouraged perspective taking. Explicit verbal prediction questions strengthened performance but were unnecessary. The refined procedure was confirmed within the original laboratory and independently replicated in a second laboratory. These findings indicate that belief-congruent action prediction is sensitive to task procedures, especially the predictive structure of familiarization. Our study provides a foundation for evaluating spontaneous theory of mind across development and species.