Skip to content

Think you have solved question answering? try ARC, the AI2 reasoning challenge

Authors: Peter Clark; Isaac Cowhey; Oren Etzioni; Tushar Khot; Ashish Sabharwal; Carissa Schoenick; Oyvind Tafjord
Year: 2018
Venue: arXiv preprint

type: dataset collection: sok-references status: auto

Links: arXiv:1803.05457

Summary

Auto-extracted summary (unreviewed)

Presents the AI2 Reasoning Challenge (ARC): 7,787 natural grade-school science questions split into a Challenge Set (2,590 questions that both a retrieval-based and a PMI word co-occurrence solver answer incorrectly) and an Easy Set (5,197), together with a 14M-sentence ARC Corpus and three neural baselines (DecompAttn, BiDAF, DGEM). None of the baselines, including models strong on SQuAD and SNLI, significantly outperform random guessing on the Challenge Set.

Abstract

We present a new question set, text corpus, and baselines assembled to encourage AI research in advanced question answering. Together, these constitute the AI2 Reasoning Challenge (ARC), which requires far more powerful knowledge and reasoning than previous challenges such as SQuAD or SNLI. The ARC question set is partitioned into a Challenge Set and an Easy Set, where the Challenge Set contains only questions answered incorrectly by both a retrieval-based algorithm and a word co-occurence algorithm. The dataset contains only natural, grade-school science questions (authored for human tests), and is the largest public-domain set of this kind (7,787 questions). We test several baselines on the Challenge Set, including leading neural models from the SQuAD and SNLI tasks, and find that none are able to significantly outperform a random baseline, reflecting the difficult nature of this task. We are also releasing the ARC Corpus, a corpus of 14M science sentences relevant to the task, and implementations of the three neural baseline models tested. Can your model perform better? We pose ARC as a challenge to the community.
Source: arxiv

Extracted details

Field Value Source Confidence
contribution ARC question set of 7,787 public-domain science exam questions llm 0.90
contribution Challenge/Easy partition defined by failure of retrieval and co-occurrence solvers llm 0.90
contribution ARC Corpus of 14M science-related sentences (1.4 GB) llm 0.90
contribution Baseline implementations (DecompAttn, BiDAF, DGEM) and a public leaderboard llm 0.90
contribution Taxonomy of knowledge and reasoning types appearing in Challenge questions llm 0.90
models DecompAttn llm 0.90
models BiDAF llm 0.90
models DGEM llm 0.90
models retrieval-based (IR) solver llm 0.90
models PMI word co-occurrence solver llm 0.90
datasets ARC llm 0.90
datasets ARC Corpus llm 0.90
tasks multiple-choice question answering llm 0.90
tasks science question answering llm 0.90
tasks reasoning llm 0.90
code url https://github.com/allenai/arc-solvers (URL found in the full text) regex 0.50
code url https://github.com/allenai/arc-solvers llm 0.90
limitations Knowledge and reasoning type categorization is approximate and based on a subjective sample of 100 questions llm 0.90
limitations Only about 75% of sampled corpus documents were judged science-relevant llm 0.90
key result dataset size: 7,787 questions (Challenge 2,590; Easy 5,197) (ARC question set) llm 0.90
key result corpus size: 14M sentences (1.4 GB) (ARC Corpus) llm 0.90
key result baseline accuracy: not significantly above random (all tested baselines on the Challenge Set) llm 0.90

SoK reference [49]

P. Clark, I. Cowhey, O. Etzioni, T. Khot, A. Sabharwal, C. Schoenick, and O. Tafjord, “Think you have solved question answering? try ARC, the AI2 reasoning challenge,” arXiv:1803.05457v1, 2018.

Listed on the reference page of the site.

Keyword mentions

Counts of vocabulary terms in the full text (first page of occurrence in brackets).

  • model: CNN 1 [p.2]
  • dataset: ARC 62 [p.1], SQuAD 8 [p.1]
  • task: Question answering 7 [p.1]
  • topic: Open source 2 [p.7]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/sok-references/049_think-you-have-solved-question-answering-try-arc-the-ai2-rea.pdf sok-references 10 283.8 KB b24b722c93f3

Database key: think-you-have-solved-question-answering-try-arc-the-ai2-reasoning · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set think-you-have-solved-question-answering-try-arc-the-ai2-reasoning COLUMN VALUE