A 369-item BIG-bench multiple-choice task: infer a calendar date in MM/DD/YYYY from one or two English sentences.
unassessed
| Category | reasoning |
|---|---|
| Subcategory | calendar-date arithmetic from short English context (MM/DD/YYYY) |
| Page status | unknown |
| Metric | multiple_choice_grade |
| Direction | higher_is_better |
| Unit | % |
| Dataset size | 369 |
| Dataset licence | Apache-2.0 |
| Publisher | Google (BIG-bench collaboration) |
date_understanding gives one or two English sentences that name or offset a day, then asks what that day (today, tomorrow, ten days ago, and similar) is in MM/DD/YYYY. Dates may be written in mixed formats, named holidays, or relative phrases, sometimes with distractor numbers such as a clock time. Authors Xinyi Wu and Zijian Wang wrote the items by hand. The skill is date conversion and arithmetic, not year-only document dating. BIG-bench Hard keeps a 250-item sample of the same task.
Multiple choice with preferred metric multiple_choice_grade. append_choices_to_input is false. 311 items have six options and 58 have five; each item has one scored target. Canary GUID embedded in task.json.
No model card in ModelSpec reports this benchmark yet.