# CaT-BENCH: Benchmarking Language Model Understanding of Causal and Temporal Dependencies in Plans

arXiv: https://arxiv.org/abs/2406.15823 · HTML: https://mlsanity.com/paper/2406.15823/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- Builds on 34 of its 53 references that are in the map.
- First posted on arXiv 22 Jun 2024.

## Builds on (most cited first, 30 of 34)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [ALFWorld: Aligning Text and Embodied Environments for Interactive Learning](https://mlsanity.com/paper/2010.03768/) (2020)
- [e-SNLI: Natural Language Inference with Natural Language Explanations](https://mlsanity.com/paper/1812.01193/) (2018)
- [PlanBench: An Extensible Benchmark for Evaluating Large Language Models on Planning and Reasoning about Change](https://mlsanity.com/paper/2206.10498/) (2022)
- [Explain Yourself! Leveraging Language Models for Commonsense Reasoning](https://mlsanity.com/paper/1906.02361/) (2019)
- [Measuring and Improving Consistency in Pretrained Language Models](https://mlsanity.com/paper/2102.01017/) (2021)
- [NATURAL PLAN: Benchmarking LLMs on Natural Language Planning](https://mlsanity.com/paper/2406.04520/) (2024)
- [The Generative AI Paradox: "What It Can Create, It May Not Understand"](https://mlsanity.com/paper/2311.00059/) (2023)
- [NILE : Natural Language Inference with Faithful Natural Language Explanations](https://mlsanity.com/paper/2005.12116/) (2020)
- [Reasoning about Goals, Steps, and Temporal Ordering with WikiHow](https://mlsanity.com/paper/2009.07690/) (2020)
- [Tracking State Changes in Procedural Text: A Challenge Dataset and Models for Process Paragraph Comprehension](https://mlsanity.com/paper/1805.06975/) (2018)
- [Simulating Action Dynamics with Neural Process Networks](https://mlsanity.com/paper/1711.05313/) (2017)
- [Tracking the World State with Recurrent Entity Networks](https://mlsanity.com/paper/1612.03969/) (2016)
- [Task Planning in Robotics: an Empirical Comparison of PDDL-based and ASP-based Systems](https://mlsanity.com/paper/1804.08229/) (2018)
- [A Dataset for Tracking Entities in Open Domain Procedural Text](https://mlsanity.com/paper/2011.08092/) (2020)
- [Everything Happens for a Reason: Discovering the Purpose of Actions in Procedural Text](https://mlsanity.com/paper/1909.04745/) (2019)
- [Goal-Oriented Script Construction](https://mlsanity.com/paper/2107.13189/) (2021)
- [Curious Case of Language Generation Evaluation Metrics: A Cautionary Tale](https://mlsanity.com/paper/2010.13588/) (2020)
- [TellMeWhy: A Dataset for Answering Why-Questions in Narratives](https://mlsanity.com/paper/2106.06132/) (2021)
- [Analogous Process Structure Induction for Sub-event Sequence Prediction](https://mlsanity.com/paper/2010.08525/) (2020)
- [Causal Reasoning of Entities and Events in Procedural Texts](https://mlsanity.com/paper/2301.10896/) (2023)
- [Author's Sentiment Prediction](https://mlsanity.com/paper/2011.06128/) (2020)
- [Understanding Multimodal Procedural Knowledge by Sequencing Multimodal Instructional Manuals](https://mlsanity.com/paper/2110.08486/) (2021)
- [A Recipe for Creating Multimodal Aligned Datasets for Sequential Tasks](https://mlsanity.com/paper/2005.09606/) (2020)
- [Multi-modal Cooking Workflow Construction for Food Recipes](https://mlsanity.com/paper/2008.09151/) (2020)
- [PROC2PDDL: Open-Domain Planning Representations from Texts](https://mlsanity.com/paper/2403.00092/) (2024)
- [Sequence to Sequence Learning for Event Prediction](https://mlsanity.com/paper/1709.06033/) (2017)
- [OpenPI2.0: An Improved Dataset for Entity Tracking in Texts](https://mlsanity.com/paper/2305.14603/) (2023)
- [Improved Instruction Ordering in Recipe-Grounded Conversation](https://mlsanity.com/paper/2305.17280/) (2023)

## Built on by (most cited first, 11 of 11)

- [Large Language Models for Planning: A Comprehensive and Systematic Survey](https://mlsanity.com/paper/2505.19683/) (2025)
- [Benchmarking Agentic Workflow Generation](https://mlsanity.com/paper/2410.07869/) (2024)
- [ProMQA-Assembly: Multimodal Procedural QA Dataset on Assembly](https://mlsanity.com/paper/2509.02949/) (2025)
- [Transparent and Coherent Procedural Mistake Detection](https://mlsanity.com/paper/2412.11927/) (2024)
- [How LLMs Comprehend Temporal Meaning in Narratives: A Case Study in Cognitive Evaluation of LLMs](https://mlsanity.com/paper/2507.14307/) (2025)
- [Chat2Workflow: A Benchmark for Generating Executable Visual Workflows with Natural Language](https://mlsanity.com/paper/2604.19667/) (2026)
- [ISO-Bench: Benchmarking Multimodal Causal Reasoning in Visual-Language Models through Procedural Plans](https://mlsanity.com/paper/2507.23135/) (2025)
- [Instructional Text Across Disciplines: A Survey of Representations, Downstream Tasks, and Open Challenges Toward Capable AI Agents](https://mlsanity.com/paper/2410.18529/) (2024)
- [Retell, Reward, Repeat: Reinforcement Learning for Narrative Theory-Informed Story Retelling](https://mlsanity.com/paper/2601.17226/) (2026)
- [Relationally Guided Use Case Modeling with LLMs](https://mlsanity.com/paper/2609.18291/) (2026)
- [Sensitivity or Criterion? Measuring and Moving When Language Models Act](https://mlsanity.com/paper/2610.15415/) (2026)

## Abstract (arXiv)

Understanding the abilities of LLMs to reason about natural language plans, such as instructional text and recipes, is critical to reliably using them in decision-making systems. A fundamental aspect of plans is the temporal order in which their steps needs to be executed, which reflects the underlying causal dependencies between them. We introduce CaT-Bench, a benchmark of Step Order Prediction questions, which test whether a step must necessarily occur before or after another in cooking recipe plans. We use this to evaluate how well frontier LLMs understand causal and temporal dependencies. We find that SOTA LLMs are underwhelming (best zero-shot is only 0.59 in F1), and are biased towards predicting dependence more often, perhaps relying on temporal order of steps as a heuristic. While prompting for explanations and using few-shot examples improve performance, the best F1 result is only 0.73. Further, human evaluation of explanations along with answer correctness show that, on average, humans do not agree with model reasoning. Surprisingly, we also find that explaining after answering leads to better performance than normal chain-of-thought prompting, and LLM answers are not consistent across questions about the same step pairs. Overall, results show that LLMs' ability to detect dependence between steps has significant room for improvement.