SciText2Eq: Assessing LLMs for Explainable Equation Generation for Scientific Creativity

The paper introduces SciText2Eq, a framework for evaluating large language models (LLMs) in generating mathematical equations from scientific texts. It presents a dataset of AI research papers with paired passages and ground-truth equations, along with an evaluation protocol that combines automatic metrics, LLM-based rubrics, and human assessments. Results show that while LLMs demonstrate moderate performance in lexical and syntactic similarity, they struggle with semantic accuracy and alignment with human judgments, underscoring the need for improved models and evaluation methods in scientific equation generation.