| # Sentence Position Prediction Dataset | |
| ### Data Example | |
| ``` | |
| { | |
| "story": | |
| "为了证明自己看见了这一切，路过银树林时，守望星折了一根小树枝。[MASK]丽娜听到树枝折断的声音时，回头问：“什么声音？”[MASK]“什么声音也没有。”[MASK]她的大姐说，“可能是哪座城堡的塔楼里，猫头鹰在叫唤。”[MASK]她讲话的时候，迈克悄悄地溜到前头，上了楼梯，第一个进了公主们的房间。[MASK]他推开窗户，顺着藤条滑了下去。[MASK]到花园的时候，太阳刚刚开始升起，他要开始工作了。[MASK]这一天，迈克捆扎鲜花的时候，故意把那根银色的树枝扎进了献给小公主的花里。[MASK]不过，她没有告诉姐姐们。", | |
| "sentence": | |
| "丽娜发现银树枝时，吃惊极了。", | |
| "label": | |
| 8 | |
| } | |
| ``` | |
| - "Story" (`str`)：input story，`<MASK>` means the candidate position | |
| - "sentence" (`str`)：the removed sentence | |
| - "label" (`int`): label=$l$ means the $l$-th position is correct. | |
| ### Evaluation | |
| The prediction result should have the same format with `test.jsonl` | |
| ```shell | |
| python eval.py prediction_file test.jsonl | |
| ``` | |
| We use accuracy as the evaluation metric. The output of the script `eval.py` is a dictionary as follows: | |
| ```python | |
| {"accuracy": _} | |
| ``` | |