We propose LingBench++, a linguistically-informed benchmark and reasoning framework designed to evaluate large language models (LLMs) on complex linguistic tasks inspired by the International Linguistics Olympiad (IOL). Unlike prior benchmarks that focus solely on final answer accuracy, LingBench++ provides structured reasoning traces, stepwise evaluation protocols, and rich typological metadata across over 90 low-resource and cross-cultural languages. We further develop a multi-agent architecture integrating grammatical knowledge retrieval, tool-augmented reasoning, and deliberate hypothesis testing. Through systematic comparisons of baseline and our proposed agentic models, we demonstrate that models equipped with external knowledge sources and iterative reasoning outperform single-pass approaches in both accuracy and interpretability. LingBench++ offers a comprehensive foundation for advancing linguistically grounded, culturally informed, and cognitively plausible reasoning in LLMs.
@article{arxiv.2507.16809,
title = {LingBench++: A Linguistically-Informed Benchmark and Reasoning Framework for Multi-Step and Cross-Cultural Inference with LLMs},
author = {Da-Chen Lian and Ri-Sheng Huang and Pin-Er Chen and Chunki Lim and You-Kuan Lin and Guan-Yu Tseng and Zi-Cheng Yang and Zhen-Yu Lin and Pin-Cheng Chen and Shu-Kai Hsieh},
journal= {arXiv preprint arXiv:2507.16809},
year = {2025}
}
Comments
42p, 17f, 10t. Revisions: Merged paragraphs in Intro to emphasize contributions. Clarified benchmark design (Sec 3.5.1). Added single-agent, OpenAI-guided & 6-round experiments (Sec 5.2). Note: we only ran each experiment once; statistical tests are needed for strong claims. Revised Sec 6. Added acknowledgements, 2 new co-authors, and corrected typos/grammar