Computational evidence for Hennen van Merchtenen and De Voortzetting
Three lines of computational evidence toward a shared author between Hennen van Merchtenen's work and De Voortzetting van de Brabantsche Yeesten.
Abstract
De Voortzetting van de Brabantsche Yeesten is a fifteenth-century Middle Dutch rhyme chronicle detailing the history of the Duchy of Brabant. Written as a direct sequel to the earlier Brabantsche Yeesten by Jan van Boendale, its authorship remains a subject of active scholarly debate. A potential candidate proposed by scholars is Hennen van Merchtenen, who authored a similar chronicle on the subject in 1415, titled the Cornicke van Brabant.
De Voortzetting survives in a remarkable autograph manuscript filled with marginal additions, deletions, and corrections, providing rare insight into the author’s writing and editing process Houthuys (2004). Unfortunately no such autograph exists for Merchtenen's Cornicke. Consequently, handwriting comparison cannot be used to settle the identity of the author.
Building on the work of Reynaert (2019), who argued for Merchtenen’s authorship based on (among other evidence) qualitative stylistic and rhyming evidence, this project approaches the problem computationally. Using word frequencies and machine learning representations, we assess and compare the writing styles of these chronicles against a wider corpus of Middle Dutch rhymed texts.
Shared bigrams and trigrams between the known text and De Voortzetting, ranked by their contribution to the two documents' TF‑IDF similarity. Click any term on either side to see how each document actually uses it.
N-gram similarities between documents in the corpus
Shared rhyme words between the documents. Click any word to see the lines it closes in both texts. Investigate the similarities and differences in rhyming style yourself!
Rhyme word similarities between documents in the corpus. Please note that both texts actually have documents they are more similar to in rhyme that we know are from other authors.
Each point is one line of verse, placed by a 2D projection of its learned style embedding. Lines that sit close together should be similar in phrasing and rhythm by the model. Click a point to read the line in context.
Warning: The model fails during training and does not converge, the visualization should be interpreted as a work in progress, and as an indication how such a model could be used.
How it was made
Three independent methods were run over a corpus of 263 Middle Dutch rhymed manuscripts: TF‑IDF bigram/trigram cosine similarity, rhyme-word cosine similarity, and a (not working) contrastive character-level line encoder (based on ByT5) trained across the corpus with author identity as the training signal. For the first two methods, each questioned book's similarity to the known text is checked against a background distribution, all document similarity to every other document in the corpus, so a high score can be judged against how much similarity naturally varies between unrelated documents. The full pipeline and how to regenerate the data behind these tabs is documented in the README.
References
- Reynaert, Joris. "Hennen van Merchtenen en de Voortzetting van de Brabantsche yeesten (ca. 1430-1450): Een bijdrage aan de attributieproblematiek." Nederlandse Letterkunde 24.1 (2019): 1-33. [Access Article]
- Houthuys, Astrid. Middeleeuws kladwerk: de autograaf van de Brabantsche yeesten, boek VI (vijftiende eeuw). Vol. 117. Uitgeverij Verloren, 2009.
- Corpus Middelnederlands Available at the Dutch Language Institute, (January 2021). [Access dataset]