aiqbench analysis
This research introduces ADAGE, a novel pipeline for evaluating large language models on culturally grounded analogical reasoning without relying on English translations or introducing linguistic artifacts. The study reveals significant performance gaps across Arabic, Amharic, and Japanese benchmarks, demonstrating that current models struggle with abstract reasoning outside their native training distribution compared to English proverb tasks. By releasing these translation-free datasets alongside the evaluation suite, this work provides a critical new standard for assessing true multilingual AI capabilities beyond superficial benchmarking.
ADAGE: A Language-Agnostic Pipeline for Analogical Reasoning Evaluation
arXiv:2607.23058v1 Announce Type: cross Abstract: Multilingual reasoning evaluation overwhelmingly relies on translating English benchmarks, a practice that introduces linguistic artifacts and fails to test culturally-grounded reasoning. We introduce ADAGE (Analogical Difficulty-by-design Assessment for Grounded Evaluation), a language-agnostic pipeline that combines native-speaker curation with LLM-assisted generation to construct challenging, translation-free benchmarks for abstract analogical reasoning. We validate ADAGE by constructing benchmarks for Arabic, Amharic, and Japanese. Evaluating 14 open-weight models, we find a consistent cultural reasoning gap: models that perform well on English proverb reasoning struggle substantially on all three native benchmarks, with accuracy dropping by 12--52 percentage points relative to English. We release the pipeline, all three benchmarks, and the full evaluation suite.
Abstract:Multilingual reasoning evaluation overwhelmingly relies on translating English benchmarks, a practice that introduces linguistic artifacts and fails to test culturally-grounded reasoning. We introduce ADAGE (Analogical Difficulty-by-design Assessment for Grounded Evaluation), a language-agnostic pipeline that combines native-speaker curation with LLM-assisted generation to construct challenging, translation-free benchmarks for abstract analogical reasoning. We validate ADAGE by constructing benchmarks for Arabic, Amharic, and Japanese. Evaluating 14 open-weight models, we find a consistent cultural reasoning gap: models that perform well on English proverb reasoning struggle substantially on all three native benchmarks, with accuracy dropping by 12--52 percentage points relative to English. We release the pipeline, all three benchmarks, and the full evaluation suite.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2607.23058 [cs.CL]
(or arXiv:2607.23058v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2607.23058 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 206 words
