The ICDAR 2026 Competition on Multilingual Medieval Handwriting Recognition (CMMHWR26) seeks to evaluate the state of the art in multilingual historical handwritten text recognition. Each of the three increasingly difficult tasks is designed to assess model generalization and robustness under conditions that reflect real-world retrodigitization workflows.

Task 1: Multilingual Recognition

evaluates text recognition across medieval manuscripts written in different languages. For this task, contestants will receive a dataset containing manuscripts written in eight different Romance languages (Castilian, Catalan, French, Gallician, Italian, Latin, Navarrese, and Venitian). Evaluation will be performed on a test set containing material written in the same languages and similar in style to the furnished training set.

Task 2: Intra-language family generalization

extends Task 1 by evaluating recognition of manuscripts written in Occitan, a closely related Romance language that is not present in the training dataset.

Task 3: Cross-language family generalization

evaluates accuracy on historical material linguistically dissimilar to the languages present in the training dataset. The test set will consist of manuscripts with similar handwriting styles to the training set, written in a non-Romance European language undisclosed to the participants in advance.