CLUE: C3 (free-form multiple-choice Chinese reading comprehension)

CLUE's multiple-choice reading task, adopted from the separately published C3 dataset spanning Chinese dialogue and mixed-genre text, 2-4 options per question.

unassessed

This page is a discovery lead. Nobody has yet assessed it against the catalogue contract, so it carries no disposition. Absence of evidence here is not evidence of staleness.
Categorycomposite
Subcategoryfree-form multiple-choice reading comprehension (Chinese)
Page statusactive
Metricaccuracy
Directionhigher_is_better
Unit%
Dataset size19577
Dataset licence"C3 dataset is intended for non-commercial research purpose only" (license.txt in the original nlpdata/c3 repository); not separately re-stated by CLUE for its own bundled copy.
PublisherC3 dataset authors (TACL 2020); specific institutional affiliations were not independently confirmed from the sources read for this page

What it measures

C3 gives a model a Chinese document -- either a two-person dialogue transcript or a more formally written mixed-genre passage -- plus a question and two to four labelled answer options, and asks it to pick the correct one. It was the first free-form multiple-choice Chinese reading-comprehension dataset at publication. CLUE adopted it wholesale from a dataset separately published by Kai Sun, Dian Yu, Dong Yu and Claire Cardie (TACL 2020); Sun and Yu also co-authored the CLUE paper, so, as with CMRC2018, the adoption carried the original authors' direct involvement rather than being a cold reuse.

Task format

Multiple-choice question answering (2-4 labelled options, varying per question) over a Chinese document, scored by accuracy.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub