Among the challenges facing the promotion of Kanak languages (New Caledonia), the accessibility of up-to-date, freely available, and expandable lexicographical resources is a recurring demand from both speaker communities and non-speakers (ALK, 2021). In this context, our project aims to develop a multilingual database based on lexicographical data from some twenty existing dictionaries (bilingual Kanak-French, or even multilingual several Kanak languages-French), whose formats (typescript or compuscript), types (thematical or alphabetical), layouts (double-column or full-page), and the granularity of the information. This paper aims to present a methodological and technological innovation that makes it possible to extract OCRised content from digitized pages of these dictionaries in order to feed the multilingual relational lexical database. The challenge is to assign a label to each identified lexicographical entry that characterizes the type of information (headword; translation; example; translation of the example; metalinguistic information, etc.). In this purpose, we developed an application designed to train a model using a sample of labelled pages, in order to automate the extraction and identification of information types for the entire dictionary. More specifically, a generic JSON-driven pipeline transforms labelling work (manual annotation) into a complete and reproducible chain of hierarchical lexicographic data (datasets → training → inference → structured entries export). This JSON pipeline is then duplicated in the application's declarative parameters for other dictionaries, enabling automation and considerable time savings, while minimizing recognition errors and interference. In addition, this paper also addresses the challenges of the protocol next steps, namely the cross-correlation of lexicographic data between different dictionaries/languages, in order to achieve the expected end result: a freely accessible online tool enabling multilingual lexical queries. Finally, this tool allows to incorporate spelling reforms resulting from the gradual standardisation of Kanak languages. It also has the advantage of giving new life to lexicographic works that are often out of print, difficult to consult, or contain obsolete lexical information. Their digital transition allows to update them, including by making it possible for speakers themselves to participate online to lexical data updating. In the end, by converting printed-only dictionaries to digital database, this project participates to continuing the work of linguists who have been committed to promoting these languages for more than 50 years.