Papers › Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension

Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension

2 Feb 2020arXiv:2002.00293archive 2025-07-28

Max Bartolo, Alastair Roberts, Johannes Welbl, Sebastian Riedel, Pontus Stenetorp

Innovations in annotation methodology have been a catalyst for Reading Comprehension (RC) datasets and models. One recent trend to challenge current RC models is to involve a model in the annotation process: humans create questions adversarially, such that the model fails to answer them correctly. In this work we investigate this annotation methodology and apply it in three different settings, collecting a total of 36,000 samples with progressively stronger models in the annotation loop. This allows us to explore questions such as the reproducibility of the adversarial effect, transfer from data collected with varying model-in-the-loop strengths, and generalisation to data collected without a model. We find that training on adversarially collected samples leads to strong generalisation to non-adversarially collected datasets, yet with progressive performance deterioration with increasingly stronger models-in-the-loop. Furthermore, we find that stronger models can still learn from datasets collected with substantially weaker models-in-the-loop. When trained on data collected with a BiDAF model in the loop, RoBERTa achieves 39.9F1 on questions that it cannot answer when trained on SQuAD - only marginally lower than when trained on data collected using RoBERTa itself (41.0F1).

PaperPDFCode

In Syntology View this paper on Syntology: its repositories, every harvested function with whether it ran, its licence and the call to fetch it.

Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

maxbartolo/adversarialQA officialmentioned on GitHubMIT report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Reading Comprehension

Datasets

Introduced by this paper, per the archive.

AdversarialQA

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Reading Comprehension AdversarialQA RoBERTa-Large D(BERT): F1 65.5 #1 of 3 Archive leaderboard report
Reading Comprehension AdversarialQA RoBERTa-Large D(BiDAF): F1 74.1 #1 of 3 Archive leaderboard report
Reading Comprehension AdversarialQA RoBERTa-Large D(RoBERTa): F1 53.4 #1 of 3 Archive leaderboard report
Reading Comprehension AdversarialQA RoBERTa-Large Overall: F1 64.4 #1 of 3 Archive leaderboard report
Reading Comprehension AdversarialQA BERT-Large D(BERT): F1 62.4 #2 of 3 Archive leaderboard report
Reading Comprehension AdversarialQA BERT-Large D(BiDAF): F1 71.3 #2 of 3 Archive leaderboard report
Reading Comprehension AdversarialQA BERT-Large D(RoBERTa): F1 54.4 #2 of 3 Archive leaderboard report
Reading Comprehension AdversarialQA BERT-Large Overall: F1 62.7 #2 of 3 Archive leaderboard report
Reading Comprehension AdversarialQA BiDAF D(BERT): F1 30.2 #3 of 3 Archive leaderboard report
Reading Comprehension AdversarialQA BiDAF D(BiDAF): F1 28.6 #3 of 3 Archive leaderboard report
Reading Comprehension AdversarialQA BiDAF D(RoBERTa): F1 26.7 #3 of 3 Archive leaderboard report
Reading Comprehension AdversarialQA BiDAF Overall: F1 28.5 #3 of 3 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

AdamAttentionAttention DropoutBERTDense ConnectionsDropoutLayer NormalizationLinear LayerLinear Warmup With Linear DecayMulti-Head AttentionResidual ConnectionRoBERTaSoftmaxWeight DecayWordPiece

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections