Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension

2 Feb 2020  ·  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).

PDF Abstract


Introduced in the Paper:


Used in the Paper:

SQuAD Natural Questions DROP

Results from the Paper

 Ranked #1 on Reading Comprehension on AdversarialQA (using extra training data)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Reading Comprehension AdversarialQA RoBERTa-Large D(BiDAF): F1 74.1 # 1
D(BERT): F1 65.5 # 1
D(RoBERTa): F1 53.4 # 2
Overall: F1 64.4 # 1
Reading Comprehension AdversarialQA BERT-Large D(BiDAF): F1 71.3 # 2
D(BERT): F1 62.4 # 2
D(RoBERTa): F1 54.4 # 1
Overall: F1 62.7 # 2
Reading Comprehension AdversarialQA BiDAF D(BiDAF): F1 28.6 # 3
D(BERT): F1 30.2 # 3
D(RoBERTa): F1 26.7 # 3
Overall: F1 28.5 # 3