# Assessing Phrasal Representation and Composition in Transformers

arXiv: https://arxiv.org/abs/2010.03763 · HTML: https://mlsanity.com/paper/2010.03763/

## Key facts

- Cited by 41 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 8 Oct 2020.

## Built on by (most cited first, 30 of 41)

- [Dynabench: Rethinking Benchmarking in NLP](https://mlsanity.com/paper/2104.14337/) (2021)
- [Phrase-BERT: Improved Phrase Embeddings from BERT with an Application to Corpus Exploration](https://mlsanity.com/paper/2109.06304/) (2021)
- [Break It Down: Evidence for Structural Compositionality in Neural Networks](https://mlsanity.com/paper/2301.10884/) (2023)
- [Characterizing Intrinsic Compositionality in Transformers with Tree Projections](https://mlsanity.com/paper/2211.01288/) (2022)
- [It's not Rocket Science : Interpreting Figurative Language in Narratives](https://mlsanity.com/paper/2109.00087/) (2021)
- [AStitchInLanguageModels: Dataset and Methods for the Exploration of Idiomaticity in Pre-Trained Language Models](https://mlsanity.com/paper/2109.04413/) (2021)
- [Does Vision-and-Language Pretraining Improve Lexical Grounding?](https://mlsanity.com/paper/2109.10246/) (2021)
- [Neural reality of argument structure constructions](https://mlsanity.com/paper/2202.12246/) (2022)
- [Complexity Control Facilitates Reasoning-Based Compositional Generalization in Transformers](https://mlsanity.com/paper/2501.08537/) (2025)
- [Initialization is Critical to Whether Transformers Fit Composite Functions by Reasoning or Memorizing](https://mlsanity.com/paper/2405.05409/) (2024)
- [Uncovering Intermediate Variables in Transformers using Circuit Probing](https://mlsanity.com/paper/2311.04354/) (2023)
- [Modeling structure-building in the brain with CCG parsing and large language models](https://mlsanity.com/paper/2210.16147/) (2022)
- [Scope Ambiguities in Large Language Models](https://mlsanity.com/paper/2404.04332/) (2024)
- [Exploring the Representation of Word Meanings in Context: A Case Study on Homonymy and Synonymy](https://mlsanity.com/paper/2106.13553/) (2021)
- [Interpretation of Black Box NLP Models: A Survey](https://mlsanity.com/paper/2203.17081/) (2022)
- [Are Representations Built from the Ground Up? An Empirical Examination of Local Composition in Language Models](https://mlsanity.com/paper/2210.03575/) (2022)
- [Semantics of Multiword Expressions in Transformer-Based Models: A Survey](https://mlsanity.com/paper/2401.15393/) (2024)
- [Automatic Evaluation and Analysis of Idioms in Neural Machine Translation](https://mlsanity.com/paper/2210.04545/) (2022)
- [Can Language Models Be Specific? How?](https://mlsanity.com/paper/2210.05159/) (2022)
- [Transparency Helps Reveal When Language Models Learn Meaning](https://mlsanity.com/paper/2210.07468/) (2022)
- [On the Interplay Between Fine-tuning and Composition in Transformers](https://mlsanity.com/paper/2105.14668/) (2021)
- [Unbalanced Optimal Transport for Unbalanced Word Alignment](https://mlsanity.com/paper/2306.04116/) (2023)
- [Bridging Continuous and Discrete Spaces: Interpretable Sentence Representation Learning via Compositional Operations](https://mlsanity.com/paper/2305.14599/) (2023)
- [Hidden Schema Networks](https://mlsanity.com/paper/2207.03777/) (2022)
- [Learning High-Quality and General-Purpose Phrase Representations](https://mlsanity.com/paper/2401.10407/) (2024)
- [Interpretability of Language Models via Task Spaces](https://mlsanity.com/paper/2406.06441/) (2024)
- [Interpreting token compositionality in LLMs: A robustness analysis](https://mlsanity.com/paper/2410.12924/) (2024)
- [Sample Efficient Approaches for Idiomaticity Detection](https://mlsanity.com/paper/2205.11306/) (2022)
- [Using Paraphrases to Study Properties of Contextual Embeddings](https://mlsanity.com/paper/2207.05553/) (2022)
- [PiC: A Phrase-in-Context Dataset for Phrase Understanding and Semantic Search](https://mlsanity.com/paper/2207.09068/) (2022)

## Abstract (arXiv)

Deep transformer models have pushed performance on NLP tasks to new limits, suggesting sophisticated treatment of complex linguistic inputs, such as phrases. However, we have limited understanding of how these models handle representation of phrases, and whether this reflects sophisticated composition of phrase meaning like that done by humans. In this paper, we present systematic analysis of phrasal representations in state-of-the-art pre-trained transformers. We use tests leveraging human judgments of phrase similarity and meaning shift, and compare results before and after control of word overlap, to tease apart lexical effects versus composition effects. We find that phrase representation in these models relies heavily on word content, with little evidence of nuanced composition. We also identify variations in phrase representation quality across models, layers, and representation types, and make corresponding recommendations for usage of representations from these models.