Analogical Transfer Task / 跨情境类比迁移

1. Task Overview

One solved source story precedes one different-surface target problem. A participant receives either a structurally matched or unmatched source, then writes an initial solution without an analogy prompt and a second solution after an explicit analogy hint. All transfer scores remain null until two blinded human raters agree or an independent adjudicator resolves differences. These are original, unnormed Chinese materials, not a clinical test or exact replication of Gick & Holyoak(1980).

Participant UI uses the neutral name 阅读与问题解决. Naive spontaneous-transfer studies must use controlled neutral/kiosk administration and prevent prior access to this repository, catalog, plot or paradigm-revealing URL. The public catalog/URL cannot preserve blinding by itself. Do not tell participants the analogy hypothesis before the initial response.

2. Task Flow

Block-Level Flow

Trial-Level Flow

Controller Logic

Other Logic

3. Configuration Summary

Settings are from config/config.yaml; the other three profiles are explicit synthetic diagnostic smoke tests.

a. Subject Info

b. Window Settings

c. Stimuli

d. Timing

e. Triggers

Behavioral mock trigger default. No-response trigger is not an unconditional offset marker. No EEG validation or physical timing calibration. RT measures response submission latency including search, typing and editing; source/hint RT measures early advancement. No cross-phase subtraction using coarse global timestamps.

f. Adaptive Controller

4. Methods (for academic publication)

The design adapts Gick and Holyoak(1980), whose experiments examined transfer from a solved source in a semantically distant domain to a target, with varying source relations and analogy instructions. Full author-hosted scanned methods were audited: Table2p317, ExperimentIpp319–320, ExperimentIIpp333–334, ExperimentIVpp341–343 and ExperimentVp344. Here one newly authored Chinese water-town source is selected between participants. The matched source instantiates multiple capacity-limited paths with simultaneous central arrival; the unmatched source solves an ordered inspection problem by sequential movement along one route. Both groups receive the identical light-lock target and identical target windows. No copyrighted classic story, empirical norm or pilot result is supplied.

Participants read the source for up to60s, attempt the target for up to120s without an analogy prompt, reread their source with an explicit hint for up to60s, then attempt the same target for up to120s regardless of their first answer. They submit committed Chinese text withF2. Editors restart blank, but both original response records persist. These timings, universal second attempt, source re-exposure, Chinese text and absence of memory recall/distractors are explicit adaptations, not original experimental parameters. Neutral participant-facing naming reduces but cannot guarantee spontaneous-transfer blinding in a public web task.

Structural transfer requires human evidence of distinct converging paths, safe low contributions, and simultaneous sufficient combined input without violating constraints. The detailed rubric distinguishes other feasible solutions, missing submissions and unrated ambiguity. Runtime scores are withheld. See human coding protocol and tools/human_coding.py for preparation, validation, adjudication and aggregation. A source-arm contrast and the within-person stage change have different interpretations: stage improvement conflates the analogy hint with extra time, repetition and source re-exposure; there is no neutral extra-time or no-source arm. A first-attempt structural solution is consistent with spontaneous transfer but does not prove source use. Materials require independent comprehension, familiarity/difficulty and reliability studies before substantive inference.

Run python main.py human; diagnostics: python main.py qa, python main.py sim --config config/config_scripted_sim.yaml, python main.py sim --config config/config_sampler_sim.yaml. Run the five build gates via task-build scripts. Native backend Unicode tests do not verify operating-system IME; do not claim otherwise. Offline coding requires Python, PyYAML; coding CLI accepts strict nativeCSV and webJSON.