ReacTOD: Bounded Neuro-Symbolic Agentic NLU for Zero-Shot Dialogue State Tracking
arXiv cs.CL·Yanjun Lin, Zimo Xiao, Kartik Natarajan, Mahesh Sankaranarayanan, Niraj Nawanit, Rakshit Parashar, Austin Zhang, Karthik Konaraddi, Rishita Mote, Wei Niu
17h agoQuick Take
ReacTOD introduces a neuro-symbolic NLU architecture for improved zero-shot dialogue state tracking.
Key Points
- Achieves 52.71% joint goal accuracy on MultiWOZ 2.1.
- Implements a self-correcting ReAct loop for enhanced accuracy.
- Cross-benchmark generalization without task-specific training data.
📖 Reader Mode
~2 min readAbstract:Task-oriented dialogue systems -- handling transactions, reservations, and service requests -- require predictable behavior, yet the moderately-sized LLMs needed for practical latency are prone to hallucination and format errors that cascade into incorrect actions (e.g., a hotel booked for the wrong date). We propose ReacTOD, a bounded neuro-symbolic architecture that reformulates NLU as discrete tool calls within a self-correcting ReAct loop governed by deterministic validation. A bounded ReAct loop enables iterative self-correction, improving accuracy by up to 9.3 percentage points over single-pass inference on MultiWOZ. A symbolic validator enforces action compliance, schema conformance, and coreference consistency on every dialogue state update, achieving a 93.1% self-correction rate on intercepted errors and producing structured execution traces. Incremental state prediction and on-demand history retrieval keep prompts compact, empirically improving instruction adherence in parameter-constrained models. On MultiWOZ 2.1, ReacTOD achieves a new zero-shot state-of-the-art: gpt-oss-20B reaches 52.71% joint goal accuracy, surpassing the previous best by 14 percentage points, while Qwen3-8B achieves 47.34% with only 8B parameters. On the Schema-Guided Dialogue (SGD) benchmark, ReacTOD with Claude-Opus-4.6 achieves 80.68% JGA under fully end-to-end evaluation with predicted domains, and Qwen3-32B reaches 64.09% -- demonstrating cross-benchmark generalization without task-specific training data.
| Comments: | Accepted at TrustNLP Workshop at ACL 2026 |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.19077 [cs.CL] |
| (or arXiv:2605.19077v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.19077 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yanjun Lin [view email]
[v1]
Mon, 18 May 2026 20:06:04 UTC (41 KB)
— Originally published at arxiv.org
More from arXiv cs.CL
See more →arXiv cs.CL·Leyao Wang, Yanan He, Peng Chen, Asaf Yehudai, Yixin Liu, Rex Ying, Michal Shmueli-Scheuer, Arman Cohan
17h ago FeaturedOriginal
Time to REFLECT: Can We Trust LLM Judges for Evidence-based Research Agents?
AI Summary
The reliability of LLM judges for evaluating deep research agents is critically assessed using the REFLECT benchmark.