An infrastructure for Historical Dutch Corpus Development

Abstract

We describe an infrastructure for linguistic annotation of historical Dutch texts, consisting of tagging and lemmatisation guidelines, gold standard data, trained tagging models, the GaLAHaD platform for automatic linguistic annotation and evaluation and the LAnCeLoT manual annotation tool. Users can upload unannotated materials to the GaLAHaD platform, where trained PoS taggers and lemmatizers (including modern deep learning models) annotate the data. Results can be evaluated against newly expanded gold standard corpora (13th to 19th centuries) with various metrics, then exported for further analysis. The LAnCeLoT tool supports manual annotation correction at both type and token level, thus supporting the development of in-domain gold standard material. The infrastructure was developed at the Instituut voor de Nederlandse Taal (INT, Dutch Language Institute) and first released in 2025.