# When classifying grammatical role, BERT doesn't care about word order... except when it matters

arXiv: https://arxiv.org/abs/2203.06204 · HTML: https://mlsanity.com/paper/2203.06204/

## Key facts

- Cited by 24 papers in the map (data through 14 Oct 2026).
- Builds on 15 of its 40 references that are in the map.
- First posted on arXiv 11 Mar 2022.

## Builds on (most cited first, 15 of 15)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Probing Classifiers: Promises, Shortcomings, and Advances](https://mlsanity.com/paper/2102.12452/) (2021)
- [Designing and Interpreting Probes with Control Tasks](https://mlsanity.com/paper/1909.03368/) (2019)
- [Information-Theoretic Probing with Minimum Description Length](https://mlsanity.com/paper/2003.12298/) (2020)
- [Colorless green recurrent networks dream hierarchically](https://mlsanity.com/paper/1803.11138/) (2018)
- [Information-Theoretic Probing for Linguistic Structure](https://mlsanity.com/paper/2004.03061/) (2020)
- [Masked Language Modeling and the Distributional Hypothesis: Order Word Matters Pre-training for Little](https://mlsanity.com/paper/2104.06644/) (2021)
- [Out of Order: How Important Is The Sequential Order of Words in a Sentence in Natural Language Understanding Tasks?](https://mlsanity.com/paper/2012.15180/) (2020)
- [Conditional probing: measuring usable information beyond a baseline](https://mlsanity.com/paper/2109.09234/) (2021)
- [What Context Features Can Transformer Language Models Use?](https://mlsanity.com/paper/2106.08367/) (2021)
- [BERT & Family Eat Word Salad: Experiments with Text Understanding](https://mlsanity.com/paper/2101.03453/) (2021)
- [Deep Subjecthood: Higher-Order Grammatical Features in Multilingual BERT](https://mlsanity.com/paper/2101.11043/) (2021)
- [Probing Linguistic Systematicity](https://mlsanity.com/paper/2005.04315/) (2020)
- [Do Syntactic Probes Probe Syntax? Experiments with Jabberwocky Probing](https://mlsanity.com/paper/2106.02559/) (2021)

## Built on by (most cited first, 24 of 24)

- [Revisiting the Role of Language Priors in Vision-Language Models](https://mlsanity.com/paper/2306.01879/) (2023)
- [Why is Winoground Hard? Investigating Failures in Visuolinguistic Compositionality](https://mlsanity.com/paper/2211.00768/) (2022)
- [Mission: Impossible Language Models](https://mlsanity.com/paper/2401.06416/) (2024)
- [Log Probabilities Are a Reliable Estimate of Semantic Plausibility in Base and Instruction-Tuned Language Models](https://mlsanity.com/paper/2403.14859/) (2024)
- [Language models and brains align due to more than next-word prediction and word-level information](https://mlsanity.com/paper/2212.00596/) (2022)
- [A Psycholinguistic Evaluation of Language Models' Sensitivity to Argument Roles](https://mlsanity.com/paper/2410.16139/) (2024)
- [A Method for Studying Semantic Construal in Grammatical Constructions with Interpretable Contextual Embedding Spaces](https://mlsanity.com/paper/2305.18598/) (2023)
- [Large GPT-like Models are Bad Babies: A Closer Look at the Relationship between Linguistic Competence and Psycholinguistic Measures](https://mlsanity.com/paper/2311.04547/) (2023)
- [Dissecting Multimodality in VideoQA Transformer Models by Impairing Modality Fusion](https://mlsanity.com/paper/2306.08889/) (2023)
- [Linguistic Interpretability of Transformer-based Language Models: a systematic review](https://mlsanity.com/paper/2504.08001/) (2025)
- [The Role of Linguistic Priors in Measuring Compositional Generalization of Vision-Language Models](https://mlsanity.com/paper/2310.02777/) (2023)
- [Multilingual Nonce Dependency Treebanks: Understanding how Language Models represent and process syntactic structure](https://mlsanity.com/paper/2311.07497/) (2023)
- [Learning Interpretable Representations Leads to Semantically Faithful EEG-to-Text Generation](https://mlsanity.com/paper/2505.17099/) (2025)
- [Syntactic Substitutability as Unsupervised Dependency Syntax](https://mlsanity.com/paper/2211.16031/) (2022)
- [Vocabulary embeddings organize linguistic structure early in language model training](https://mlsanity.com/paper/2510.07613/) (2025)
- [Humans and transformer LMs: Abstraction drives language learning](https://mlsanity.com/paper/2603.17475/) (2026)
- [Local Structure Matters Most in Most Languages](https://mlsanity.com/paper/2211.05025/) (2022)
- [Word Order's Impacts: Insights from Reordering and Generation Analysis](https://mlsanity.com/paper/2403.11473/) (2024)
- [LLMs Learn Constructions That Humans Do Not Know](https://mlsanity.com/paper/2508.16837/) (2025)
- [The Grammar of Transformers: A Systematic Review of Interpretability Research on Syntactic Knowledge in Language Models](https://mlsanity.com/paper/2601.19926/) (2026)
- [Dynamic Syntax Mapping: A New Approach to Unsupervised Syntax Parsing](https://mlsanity.com/paper/2312.14966/) (2023)
- [Do Large Language Models know who did what to whom?](https://mlsanity.com/paper/2504.16884/) (2025)
- [A Typologically Grounded Evaluation Framework for Word Order and Morphology Sensitivity in Multilingual Masked LMs](https://mlsanity.com/paper/2603.00432/) (2026)
- [Vocabulary shapes cross-lingual variation of word-order learnability in language models](https://mlsanity.com/paper/2603.19427/) (2026)

## Abstract (arXiv)

Because meaning can often be inferred from lexical semantics alone, word order is often a redundant cue in natural language. For example, the words chopped, chef, and onion are more likely used to convey "The chef chopped the onion," not "The onion chopped the chef." Recent work has shown large language models to be surprisingly word order invariant, but crucially has largely considered natural prototypical inputs, where compositional meaning mostly matches lexical expectations. To overcome this confound, we probe grammatical role representation in English BERT and GPT-2, on instances where lexical expectations are not sufficient, and word order knowledge is necessary for correct classification. Such non-prototypical instances are naturally occurring English sentences with inanimate subjects or animate objects, or sentences where we systematically swap the arguments to make sentences like "The onion chopped the chef". We find that, while early layer embeddings are largely lexical, word order is in fact crucial in defining the later-layer representations of words in semantically non-prototypical positions. Our experiments isolate the effect of word order on the contextualization process, and highlight how models use context in the uncommon, but critical, instances where it matters.