Med-CoReasoner: Reducing Language Disparities in Medical Reasoning via Language-Informed Co-Reasoning

A new language-informed co-reasoning framework called Med-CoReasoner aims to reduce language disparities in medical reasoning by combining English and local-language clinical knowledge, according to a preprint posted to arXiv [1]. The framework, detailed in a paper revised in May 2026, elicits parallel reasoning in English and a local language, abstracts them into structured concepts, and integrates local clinical knowledge into an English logical scaffold through concept-level alignment and retrieval [1][2]. This approach is designed to combine the structural robustness of English reasoning with the practice-grounded expertise encoded in local languages [2]. The work addresses a persistent multilingual gap where reasoning-enhanced large language models perform substantially weaker in local languages, limiting equitable global medical deployment [2]. To evaluate the system beyond multiple-choice settings, the researchers constructed MultiMed-X, a benchmark covering seven languages with expert-annotated long-form question answering and natural language inference tasks, comprising 350 instances per language [1][2]. Experiments across three benchmarks showed that Med-CoReasoner improves multilingual reasoning performance by an average of 5%, with particularly substantial gains in low-resource languages [1][2]. Model distillation and expert evaluation analysis further confirmed that the framework produces clinically sound and culturally grounded reasoning traces [2]. The development of tools like Med-CoReasoner comes amid broader concerns about healthcare equity and access. In the United States, healthcare is largely provided by private sector facilities and paid for through a mix of public programs, private insurance, and out-of-pocket payments [3]. The U.S. is the only developed country without a system of universal healthcare, and a significant proportion of its population lacks health insurance [3]. Disparities in access to care exist based on factors such as income, race, and geographical location [3]. The COVID-19 pandemic, which caused severe social and economic disruption globally and led to over 7 million confirmed deaths as of May 2026, raised issues of racial and geographic discrimination and health equity [4]. Discrimination in healthcare settings also affects other marginalized groups. Transgender people regularly encounter healthcare discrimination, workplace discrimination, and violence, and the stress created by transphobia causes negative mental health outcomes [5]. The submission history shows the paper was first posted on 13 January 2026 and revised twice, with the latest version appearing on 26 May 2026 [1]. The work is hosted on arXiv under the Computation and Language category [1].