CoDial: Interpretable Task-Oriented Dialogue Systems Through Dialogue Flow Alignment
Radin Shayanfar, Chu Fei Luo, Rohan Bhambhoria, Samuel Dahan, Xiaodan Zhu
Read paperResearch
A research consortium for legal AI.
Our researchers come from law, computer science, operations management, and dispute resolution — and our work spans foundational research on legal language models, applied research with industry partners, and access-to-justice deployment.
Areas of focus
NLP and legal language models
Training, fine-tuning, and evaluating large language models on legal tasks across multiple jurisdictions and legal traditions. Lead: Xiaodan Zhu and team.
Legal reasoning evaluation
Frameworks for measuring whether a model can perform issue analysis, rule recall, and structured legal reasoning — beyond surface-level accuracy. Reference: CoDial.
Dispute resolution analytics
Predictive and descriptive analytics over negotiation, mediation, and settlement data — the founding research thesis of the lab. See also: Deel Lab.
Access to justice
Empirical research on whether AI tools materially close the gap for unrepresented litigants, low-income workers, and small businesses. In collaboration with the A2J Lab.
AI safety
Evaluation frameworks, audit trails, and policy recommendations for deploying AI in high-stakes regulated contexts.
Featured research
All publications
| Title | Authors | Actions | |||
|---|---|---|---|---|---|
| CoDial: Interpretable Task-Oriented Dialogue Systems Through Dialogue Flow Alignment | Radin Shayanfar, Chu Fei Luo, Rohan Bhambhoria, Samuel Dahan, Xiaodan Zhu | Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics | 2026 | NLP |
Page 1 of 1 · 1 publication
Open source
Code, datasets, and benchmarks.
The legal profession deserves AI it can audit, replicate, and extend. Below: code, datasets, and benchmarks the lab has released for the broader research community.
Code
Reasoning platform for encoding legal and regulatory expertise into structured flows on top of any AI model.
Public AI-powered legal tools for workers — classification, termination compensation, reasonable notice.
Datasets
- LegalQA
Manually annotated legal Q&A across Canadian, US, and French law.
- Worker Classification Corpus
Multi-jurisdictional labour case dataset.
- OJ Document Library
Document corpus released alongside OpenJustice.
Benchmarks
- OpenJustice Evaluation Suite
Benchmarks for legal reasoning across issue analysis, rule recall, narrative construction diversity, and bias risk.
- Benchmark-athon results
Community evaluation results from the lab's annual benchmark-athon.