An Actionable Diagnosis of Multilingual, Multi-Agent Planning Failures

2026-08-04

AI Summary

This work addresses the significant degradation in planning performance, loss of task-critical information, and difficulty in diagnosing failure causes that multilingual multi-agent systems encounter in non-English settings. The study introduces the first actionable taxonomy specifically designed for planning-to-execution failures in multilingual contexts and proposes the TART framework, which explicitly embeds key dimensions of this taxonomy into both the planner and downstream sub-agents. This integration enables consistent improvements across languages, models, and system configurations. Evaluated on the GAIA benchmark covering 11 languages, TART boosts the average accuracy of state-of-the-art systems by 5.6 percentage points, with particularly pronounced gains in low-resource languages.

Abstract

Multilingual multi-agent systems exhibit substantial degradation beyond English, yet prior work rarely identifies how task-critical information is lost when user requests are converted into executable plans. We study the planner in a multi-agent system as the request-to-action interface and derive an actionable taxonomy of planning-grounding failures from failed real-world task executions. LLM-based analysis shows that these failures constitute an increasing share of unsuccessful executions as language-resource availability declines, with the strongest effects in low-resource languages. To test whether the taxonomy supports mitigation, we introduce TART, Taxonomy-Guided Actionable Representation, that makes the taxonomy's key aspects explicit to the planner and downstream sub-agents. Across multiple languages, three LLM backbones, two datasets, and two agentic configurations, TART consistently improves performance. On multilingual GAIA, it raises a state-of-the-art system's accuracy by 5.6 percentage points averaged across eleven languages spanning low- to high-resource settings.

Problem

• multilingual
• multi-agent planning
• planning failures
• low-resource languages
• task execution

Innovation

• multilingual planning
• failure taxonomy
• actionable representation
• multi-agent systems
• low-resource languages