This paper presents an approach to multilingual alignment for medieval languages, focusing on the prior step of"phrase" segmentation. It outlines the challenges posed by historical data and describes different strategies forsegmenting texts in multiple languages. It releases a gold-standard segmentation corpus based on various literaryand historical works from the late Middle Ages in Europe. This corpus consists of texts in seven medieval languages (French, Castilian, Catalan, Portuguese, Latin, Italian, English). Several architectures are tested with both in-domain and out-of-domain evaluation sets.