Litmus (Re)Agent: A Benchmark and Agentic System for Predictive Evaluation of Multilingual Models
2026-04-10 • Computation and Language
Computation and LanguageArtificial IntelligenceHuman-Computer InteractionMultiagent Systems
AI summaryⓘ
The authors explore how to guess a model's performance on a language when there is no direct test data available for that language. They created a benchmark with many questions covering various tasks and types of available evidence to test these predictions. They also designed a system called Litmus (Re)Agent, which breaks down questions, gathers clues, and combines them to make predictions. Their system performed best, especially when little direct evidence existed. This work suggests that breaking down reasoning steps helps estimate performance across languages even with limited information.
multilingual evaluationbenchmarkperformance estimationagentic systemDAG (Directed Acyclic Graph)hypothesis decompositionevidence retrievalfeature aggregationtransfer learningincomplete evidence
Authors
Avni Mittal, Shanu Kumar, Sandipan Dandapat, Monojit Choudhury
Abstract
We study predictive multilingual evaluation: estimating how well a model will perform on a task in a target language when direct benchmark results are missing. This problem is common in multilingual deployment, where evaluation coverage is sparse and published evidence is uneven across languages, tasks, and model families. We introduce a controlled benchmark of 1,500 questions spanning six tasks and five evidence scenarios. The benchmark separates accessible evidence from ground truth, enabling evaluation of systems that must infer missing results from incomplete literature evidence. We also present Litmus (Re)Agent, a DAG-orchestrated agentic system that decomposes queries into hypotheses, retrieves evidence, and synthesises predictions through feature-aware aggregation. Across six systems, Litmus (Re)Agent achieves the best overall performance, with the largest gains in transfer-heavy scenarios where direct evidence is weak or absent. These results show that structured agentic reasoning is a promising approach to multilingual performance estimation under incomplete evidence.