UnNatural Language Inference

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 7329–7346 August 1–6, 2021. ©2021 Association for Computational Linguistics 7329 UnNatural Language Inference Koustuv Sinha1,2,3, Prasanna Parthasarathi1,2, Joelle Pineau1,2,3 and Adina Williams3 1 School of Computer Science, McGill University, Canada 2 Montreal Institute of Learning Algorithms (Mila), Canada 3 Facebook AI Research (FAIR) {koustuv.sinha, prasanna.parthasarathi, jpineau, adinawilliams} @{mail.mcgill.ca, mail.mcgill.ca, cs.mcgill.ca, fb.com} Abstract Recent investigations into the inner-workings of state-of-the-art large-scale pre-trained Transformer-based Natural Language Under- standing (NLU) models indicate that they appear to know humanlike syntax, at least to some extent. We provide novel evidence that complicates this claim: we find that state-of-the-art Natural Language Inference (NLI) models assign the same labels to permuted examples as they do to the original, i.e. they are largely invariant to random word- order permutations. This behavior notably differs from that of humans; we struggle with ungrammatical sentences. To measure the severity of this issue, we propose a suite of metrics and investigate which properties of particular permutations lead models to be word-order invariant. In the MNLI dataset, for example, we find almost all (98.7%) examples contain at least one permutation which elicits the gold label. Models are sometimes even able to assign gold labels to permutations that they originally failed to predict correctly. We provide a comprehensive empirical evaluation of this phenomenon, and further show that this issue exists for both Transformers and pre-Transformer RNN / ConvNet based encoders, as well as across multiple languages (English and Mandarin Chinese). Our code and data are available at https://github.com/facebookresearch/unlu. 1 Introduction Of late, large scale pre-trained Transformer-based (Vaswani et al., 2017) models—such as RoBERTa (Liu et al., 2019), BART (Lewis et al., 2020), and GPT-2 and -3 (Radford et al., 2019; Brown et al., 2020)—have exceeded recurrent neural networks’ performance on many NLU tasks (Wang et al., 2018, 2019). Several papers have even suggested that Transformers pretrained on a language model- ing (LM) objective can capture syntactic informa- Premise Hypothesis Predicted Label Boats in daily use lie within feet of the fashionable bars and restaurants. There are boats close to bars and restaurants. E restaurants and use feet of fashionable lie the in Boats within bars daily . bars restaurants are There and to close boats . E He and his associates weren’t operating at the level of metaphor. He and his associates were operating at the level of the metaphor. C his at and metaphor the of were He operating asso- ciates n’t level . his the and metaphor level the were He at as- sociates operating of . C Table 1: Examples from the MNLI Matched develop- ment set. Both the original example and the permuted one elicit the same classification label (entailment and contradiction respectively) from RoBERTa (large). A simple demo is provided in an associated Google Co- lab notebook. tion (Hewitt and Manning, 2019; Jawahar et al., 2019; Warstadt and Bowman, 2020; Wu et al., 2020), with their …

Content truncated to 500 words