# Discursive Circuits: How Do Language Models Understand Discourse Relations?

arXiv: https://arxiv.org/abs/2510.11210 · HTML: https://mlsanity.com/paper/2510.11210/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Placed in Circuits and mechanistic analysis, one of 13 directions in Mechanistic interpretability.
- Builds on 42 of its 83 references that are in the map.
- First posted on arXiv 13 Oct 2025.

## Placed in

- Mechanistic interpretability / [Circuits and mechanistic analysis](https://mlsanity.com/mechinterp/circuits-and-mechanistic-analysis/)

## Builds on (most cited first, 30 of 42)

- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small](https://mlsanity.com/paper/2211.00593/) (2022)
- [Attention is not Explanation](https://mlsanity.com/paper/1902.10186/) (2019)
- [Towards Automated Circuit Discovery for Mechanistic Interpretability](https://mlsanity.com/paper/2304.14997/) (2023)
- [Attention is not not Explanation](https://mlsanity.com/paper/1908.04626/) (2019)
- [Towards Best Practices of Activation Patching in Language Models: Metrics and Methods](https://mlsanity.com/paper/2309.16042/) (2023)
- [Attribution Patching Outperforms Automated Circuit Discovery](https://mlsanity.com/paper/2310.10348/) (2023)
- [Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla](https://mlsanity.com/paper/2307.09458/) (2023)
- [Have Faith in Faithfulness: Going Beyond Circuit Overlap When Finding Model Mechanisms](https://mlsanity.com/paper/2403.17806/) (2024)
- [Towards Faithful Model Explanation in NLP: A Survey](https://mlsanity.com/paper/2209.11326/) (2022)
- [Teach Me to Explain: A Review of Datasets for Explainable Natural Language Processing](https://mlsanity.com/paper/2102.12060/) (2021)
- [Circuit Component Reuse Across Tasks in Transformer Language Models](https://mlsanity.com/paper/2310.08744/) (2023)
- [Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control](https://mlsanity.com/paper/2405.08366/) (2024)
- [On the Similarity of Circuits across Languages: a Case Study on the Subject-verb Agreement Task](https://mlsanity.com/paper/2410.06496/) (2024)
- [DiscoPrompt: Path Prediction Prompt Tuning for Implicit Discourse Relation Recognition](https://mlsanity.com/paper/2305.03973/) (2023)
- [Efficient Automated Circuit Discovery in Transformers using Contextual Decomposition](https://mlsanity.com/paper/2407.00886/) (2024)
- [How Do LLMs Acquire New Knowledge? A Knowledge Circuits Perspective on Continual Pre-Training](https://mlsanity.com/paper/2502.11196/) (2025)
- [pNLP-Mixer: an Efficient all-MLP Architecture for Language](https://mlsanity.com/paper/2202.04350/) (2022)
- [QUDEVAL: The Evaluation of Questions Under Discussion Discourse Parsing](https://mlsanity.com/paper/2310.14520/) (2023)
- [Why Can't Discourse Parsing Generalize? A Thorough Investigation of the Impact of Data Diversity](https://mlsanity.com/paper/2302.06488/) (2023)
- [Design Choices for Crowdsourcing Implicit Discourse Relations: Revealing the Biases Introduced by Task Design](https://mlsanity.com/paper/2304.00815/) (2023)
- [eRST: A Signaled Graph Theory of Discourse Relations and Organization](https://mlsanity.com/paper/2403.13560/) (2024)
- [Discourse Analysis via Questions and Answers: Parsing Dependency Structures of Questions Under Discussion](https://mlsanity.com/paper/2210.05905/) (2022)
- [QUDsim: Quantifying Discourse Similarities in LLM-Generated Text](https://mlsanity.com/paper/2504.09373/) (2025)
- [Annotation-Inspired Implicit Discourse Relation Classification with Auxiliary Discourse Connective Generation](https://mlsanity.com/paper/2306.06480/) (2023)
- [Infusing Hierarchical Guidance into Prompt Tuning: A Parameter-Efficient Framework for Multi-level Implicit Discourse Relation Recognition](https://mlsanity.com/paper/2402.15080/) (2024)
- [Contextualize-then-Aggregate: Circuits for In-Context Learning in Gemma-2 2B](https://mlsanity.com/paper/2504.00132/) (2025)
- [Towards Informative Few-Shot Prompt with Maximum Information Gain for In-Context Learning](https://mlsanity.com/paper/2310.08923/) (2023)
- [Multi-Label Classification for Implicit Discourse Relation Recognition](https://mlsanity.com/paper/2406.04461/) (2024)
- [Towards Unification of Discourse Annotation Frameworks](https://mlsanity.com/paper/2204.07781/) (2022)

## Built on by (most cited first, 4 of 4)

- [Latent Agents: A Post-Training Procedure for Internalized Multi-Agent Debate](https://mlsanity.com/paper/2604.24881/) (2026)
- [Illusions of Confidence? Diagnosing LLM Truthfulness via Neighborhood Consistency](https://mlsanity.com/paper/2601.05905/) (2026)
- [Improving Implicit Discourse Relation Recognition with Natural Language Explanations from LLMs](https://mlsanity.com/paper/2602.21763/) (2026)
- [Not Just an Evaluation: A Case Study of Discourse Particle Sensitivity in Language Models](https://mlsanity.com/paper/2610.13794/) (2026)

## Abstract (arXiv)

Which components in transformer language models are responsible for discourse understanding? We hypothesize that sparse computational graphs, termed as discursive circuits, control how models process discourse relations. Unlike simpler tasks, discourse relations involve longer spans and complex reasoning. To make circuit discovery feasible, we introduce a task called Completion under Discourse Relation (CuDR), where a model completes a discourse given a specified relation. To support this task, we construct a corpus of minimal contrastive pairs tailored for activation patching in circuit discovery. Experiments show that sparse circuits (≈ 0.2% of a full GPT-2 model) recover discourse understanding in the English PDTB-based CuDR task. These circuits generalize well to unseen discourse frameworks such as RST and SDRT. Further analysis shows lower layers capture linguistic features such as lexical semantics and coreference, while upper layers encode discourse-level abstractions. Feature utility is consistent across frameworks (e.g., coreference supports Expansion-like relations).