# UnNatural Language Inference

arXiv: https://arxiv.org/abs/2101.00010 · HTML: https://mlsanity.com/paper/2101.00010/

## Key facts

- Cited by 69 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 30 Dec 2020.

## Built on by (most cited first, 30 of 69)

- [Winoground: Probing Vision and Language Models for Visio-Linguistic Compositionality](https://mlsanity.com/paper/2204.03162/) (2022)
- [Do Prompt-Based Models Really Understand the Meaning of their Prompts?](https://mlsanity.com/paper/2109.01247/) (2021)
- [When Does Pretraining Help? Assessing Self-Supervised Learning for Law and the CaseHOLD Dataset](https://mlsanity.com/paper/2104.08671/) (2021)
- [Masked Language Modeling and the Distributional Hypothesis: Order Word Matters Pre-training for Little](https://mlsanity.com/paper/2104.06644/) (2021)
- [Post-hoc Interpretability for Neural NLP: A Survey](https://mlsanity.com/paper/2108.04840/) (2021)
- [System 2 Attention (is something you might need too)](https://mlsanity.com/paper/2311.11829/) (2023)
- [Premise Order Matters in Reasoning with Large Language Models](https://mlsanity.com/paper/2402.08939/) (2024)
- [Reasoning Like Program Executors](https://mlsanity.com/paper/2201.11473/) (2022)
- [Structural Persistence in Language Models: Priming as a Window into Abstract Language Representations](https://mlsanity.com/paper/2109.14989/) (2021)
- [Why is Winoground Hard? Investigating Failures in Visuolinguistic Compositionality](https://mlsanity.com/paper/2211.00768/) (2022)
- [Word Order Does Matter (And Shuffled Language Models Know It)](https://mlsanity.com/paper/2203.10995/) (2022)
- [Does Putting a Linguist in the Loop Improve NLU Data Collection?](https://mlsanity.com/paper/2104.07179/) (2021)
- [Pretraining with Artificial Language: Studying Transferable Knowledge in Language Models](https://mlsanity.com/paper/2203.10326/) (2022)
- [Neural reality of argument structure constructions](https://mlsanity.com/paper/2202.12246/) (2022)
- [Model Internals-based Answer Attribution for Trustworthy Retrieval-Augmented Generation](https://mlsanity.com/paper/2406.13663/) (2024)
- [Rissanen Data Analysis: Examining Dataset Characteristics via Description Length](https://mlsanity.com/paper/2103.03872/) (2021)
- [Can Linguistic Knowledge Improve Multimodal Alignment in Vision-Language Pretraining?](https://mlsanity.com/paper/2308.12898/) (2023)
- [Improving Model Evaluation using SMART Filtering of Benchmark Datasets](https://mlsanity.com/paper/2410.20245/) (2024)
- [Tokenization Falling Short: On Subword Robustness in Large Language Models](https://mlsanity.com/paper/2406.11687/) (2024)
- [A Fistful of Words: Learning Transferable Visual Models from Bag-of-Words Supervision](https://mlsanity.com/paper/2112.13884/) (2021)
- [Variational Cross-Graph Reasoning and Adaptive Structured Semantics Learning for Compositional Temporal Grounding](https://mlsanity.com/paper/2301.09071/) (2023)
- [Unnatural Error Correction: GPT-4 Can Almost Perfectly Handle Unnatural Scrambled Text](https://mlsanity.com/paper/2311.18805/) (2023)
- [Limits for Learning with Language Models](https://mlsanity.com/paper/2306.12213/) (2023)
- [TASA: Deceiving Question Answering Models by Twin Answer Sentences Attack](https://mlsanity.com/paper/2210.15221/) (2022)
- [Compositional Evaluation on Japanese Textual Entailment and Similarity](https://mlsanity.com/paper/2208.04826/) (2022)
- [Language models and brains align due to more than next-word prediction and word-level information](https://mlsanity.com/paper/2212.00596/) (2022)
- [Semantic Sensitivities and Inconsistent Predictions: Measuring the Fragility of NLI Models](https://mlsanity.com/paper/2401.14440/) (2024)
- [How to Make the Most of LLMs' Grammatical Knowledge for Acceptability Judgments](https://mlsanity.com/paper/2408.09639/) (2024)
- [Kernel-Whitening: Overcome Dataset Bias with Isotropic Sentence Embedding](https://mlsanity.com/paper/2210.07547/) (2022)
- [Mitigating Reversal Curse in Large Language Models via Semantic-aware Permutation Training](https://mlsanity.com/paper/2403.00758/) (2024)

## Abstract (arXiv)

Recent investigations into the inner-workings of state-of-the-art large-scale pre-trained Transformer-based Natural Language Understanding (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.