Ambiguity is an inherent feature of all natural languages and presents a particular challenge in Natural Language Processing (NLP), especially when working with historical corpora. This study addresses the identification and resolution of specific ambiguities in MedText, a historical corpus of medieval Latin last wills and testaments from the city of Zadar (13th–15th centuries). Within the corpus of approximately 500,000 tokens, around 5,800 ambiguities were identified. The analysis distinguishes three types of ambiguity: morphological, syntactic, and lexical. Special attention is given to morphosyntactic ambiguities, including overlaps between adjectives and nouns, verbs and pronouns, as well as prepositions and conjunctions. To reduce these ambiguities, the NooJ development environment was used to create context-sensitive syntactic grammars. These grammars disambiguate forms by analyzing their syntactic context, thereby improving the accuracy of morphological tagging and syntactic parsing. Although a substantial portion of ambiguous cases was resolved through syntactic grammars, some instances still require manual correction. Nevertheless, this work represents an advancement in the disambiguation of morphosyntactic structures in non-standardized medieval Latin, contributing to the field of historical language processing.