The 3rd Workshop on Mathematical Natural Language Processing

MathNLP 2025: The 3rd Workshop on Mathematical Natural Language Processing

at EMNLP 2025

Suzhou, China

Submit your paper on OpenReview!

A One-day Workshop on Mathematical Language Processing

The articulation of mathematical arguments is a fundamental part of scientific reasoning and communication. Across many disciplines, expressing relations and interdependencies between quantities is at the centre of scientific argumentation. Nevertheless, despite its importance, the application of contemporary NLP models for inference over mathematical text remains under-explored or subject to important limitations. MathNLP represents a forum for discussing new ideas to advance research on Mathematical Natural Language Processing, welcoming novel contributions on model architectures, evaluation methods and downstream applications.

Recent advances in Natural Language Processing (NLP) enabled by Deep Learning-based architectures bring the opportunity to support the interpretation of textual content at scale. The application of these methods can facilitate scientific discovery, reducing the gap between current research and the available large-scale scientific knowledge. Previous work has shown the potential of designing neural architectures for different mathematical natural language inference tasks, such as premise selection in natural language, expression derivation, and mathematical information retrieval.

However, there are still technical gaps that need to be addressed such as the availability of datasets and evaluation tasks, techniques for the joint interpretation of different modalities present in mathematical text (equational and natural language), the understanding of unique aspects of mathematical discourse and multi-hop models for mathematical inference.

We propose this workshop as a continuation of our previous editions, with a new emphasis on the integration of Large Language Models (LLMs) and symbolic approaches with the goal of addressing these challenges and connect different experts in this field.

Call for Papers

We welcome contributions of previously unpublished papers, which could be either long (8 pages) or short (4 pages). All submissions will be peer-reviewed by multiple reviewers. The authors' identities must be concealed to enable a double-blind peer review.

MathNLP welcomes both archival and non-archival submissions. Only archival submissions will be included in the proceedings.

We are particularly interested in (but not limited to) works related to the following topics:

Neural/Neuro-symbolic architectures to support mathematical natural language inference;
Large Language Models for Mathematics;
Equational embeddings;
Autoformalisation and translation from natural language to formal languages (and vice-versa);
Linguistic analysis of mathematical discourse and argumentation relations in the context of mathematical text;
Probing mathematical understanding of state-of-the-art models;
Adaptation of NLP tasks for mathematical discourse;
NLP applied to mathematics education;
Premise selection over mathematical text;
Understanding and typing of variables in mathematical text;
Retrieval of equations/formulas/expressions based on textual queries;
Retrieval of textual context based on equational queries.

Submissions can be made via OpenReview. Please, use the official ACL template.

Please note: new profiles created on OpenReview without an institutional email will go through a moderation process that can take up to two weeks. New profiles created with an institutional email will be activated automatically.