HackerPapers

arXiv:1902.01007cs.CL

Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inference

R. Thomas McCoy, Ellie Pavlick, Tal Linzen

Abstract

A machine learning system can score well on a given test set by relying on heuristics that are effective for frequent example types but break down in more challenging cases. We study this issue within natural language inference (NLI), the task of determining whether one sentence entails another. We hypothesize that statistical NLI models may adopt three fallible syntactic heuristics: the lexical overlap heuristic, the subsequence heuristic, and the constituent heuristic. To determine whether models have adopted these heuristics, we introduce a controlled evaluation set called HANS (Heuristic Analysis for NLI Systems), which contains many examples where the heuristics fail. We find that models trained on MNLI, including BERT, a state-of-the-art model, perform very poorly on HANS, suggesting that they have indeed adopted these heuristics. We conclude that there is substantial room for improvement in NLI systems, and that the HANS dataset can motivate and measure progress in this area

arXiv comments
Camera-ready for ACL 2019
Journal reference
DOI
Subjects
cs.CLComputation and Language

On Hacker News

Linked by 4 identifiable commenters in 5 comments across 5 threads, from Jul 2019 to Dec 2021. Found in 1 HN submission from Mar 2021 to Mar 2021. Open submission discussions.

Comments linking this paper, per year.

Discussed in comments5 threads

  1. A cartel of influential datasets are dominating machine learning research 1 comment
  2. Dense Vectors: Capturing Meaning with Code 1 comment

    • > But the later transformer models have a good grasp of grammar AFAIK that is an open question. They don't even have a grasp of negation beyond superficial heuristics, as seen in arXiv:1902.01007.

      dunefoxView on HNParent context

  3. Machine learning won't solve natural language understanding 1 comment

    • There's the HANS dataset which is used to evaluate NLI models to check if they pick up surface heuristics: https://huggingface.co/datasets/hans https://arxiv.org/abs/1902.01007

      fishmasterView on HNParent context

  4. Why GPT-3 Matters 1 comment

    • >> The Aggregate performance graph shows how GPT-3 does translation and other tasks without ever learning to do that. But it also shows it's not very good on any of those tasks. In any case, machine translation, despite its great popularity as a natural language processing task…

      YeGoblynQueenneView on HNParent context

  5. Microsoft is investing $1B in OpenAI 1 comment

    • What exactly does that 96% mean, though? It means that on some fixed dataset you're achieving 96% accuracy. I'm baffled by this stupidity of claiming results (even high-profile researchers do this) based on datasets with models that are nowhere near as robust as the actual…

      Alas1View on HNParent context

Submitted to Hacker News

Connected through shared HN threads, not inferred similarity or a recommendation.