CLUE: CMRC 2018 (Simplified Chinese span-extraction reading comprehension)

CLUE's simplified-Chinese span-extraction reading task, adopted wholesale from HFL's separately published CMRC 2018 shared-task dataset.

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
Subcategoryspan-extraction reading comprehension (Simplified Chinese)
Page statusactive
Metricexact match (EM)
Directionhigher_is_better
Unit%
Dataset size16363
Dataset licenceCC-BY-SA-4.0, per the original ymcui/cmrc2018 GitHub repository; not separately re-stated by CLUE for its own bundled copy
PublisherHarbin Institute of Technology - iFLYTEK Joint Laboratory (HFL); shared task organised by the Chinese Information Processing Society's Computational Linguistics Committee (CIPS-CL), sponsored by iFLYTEK

What it measures

CMRC2018 gives a model a Chinese Wikipedia paragraph and a question, and the model must extract the exact answer span from the passage -- SQuAD-style extractive reading comprehension, over roughly 19,071 human-annotated questions. CLUE did not build this task: it adopted CMRC2018 wholesale from a dataset separately published by Yiming Cui and colleagues at the Harbin Institute of Technology-iFLYTEK Joint Laboratory (HFL), whose original shared task keeps its own test answers hidden. Cui also co-authored the CLUE paper, so the adoption carried the original team's direct involvement rather than being a cold reuse.

Task format

Extractive question answering: given a passage and a question, output the answer text span; scored by exact match (and F1, though CLUE's own leaderboard reports EM as the final figure).

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub