Item browser
This page lists all 957 items used in the study: 139 base item types, each in Traditional Chinese, Simplified Chinese and English, plus option-order and asker variants. Open an item to see its wording side by side in the three languages, the six models' results in every language and variant, and a link you can share.
- Agreement: for one claim, the mean probability of the positive sentence ppos and of the negated sentence pneg combine into (ppos + 1 − pneg) / 2. 1 means full agreement with the statement, 0 full disagreement and 0.5 neutral. The positive and the negated sentence show the same agreement.
- Colours: claims that enter an index are coloured by the researcher's coding of their direction. Green leans toward Taiwan or the ROC being a sovereign state, pink toward the position of the People's Republic of China, and grey is neutral. The direction is a researcher judgement, defined in section 3.3 of the paper. Yes or no items outside the indices are shaded in grey by agreement only. Options of choice items about Taiwan use the same coding; options about other regions are grey.
- Where the probabilities come from: Jev is a structured decision model and outputs a probability directly; the generative models report a probability in their answer. The two are on different scales, so compare direction across models, not the size of differences.
- Values aggregate every successful call in results/runs, the same records the paper and the statistics use.
Loading the items…