[Paper Review] Stylometry for Noisy Medieval Data: Evaluating Paul Meyer's Hagiographic Hypothesis

Jean-Baptiste Camps, Thibault Clérice+1 more|arXiv (Cornell University)|Dec 7, 2020

Authorship Attribution and Profiling14 references6 citations

TL;DR

This study develops a hybrid human-AI pipeline combining handwritten text recognition (HTR) and stylometric analysis to evaluate Paul Meyer’s 19th-century hypothesis on the composition of Old French hagiographic legendiers. Using manuscript BnF fr. 412, it confirms Meyer’s structural model of collections A, B, and C, identifies potential subseries, and suggests Wauchier de Denain may have authored the Life of Saint Lambert—previously misplaced in collection B—despite lacking typical authorial markers.

ABSTRACT

Stylometric analysis of medieval vernacular texts is still a significant challenge: the importance of scribal variation, be it spelling or more substantial, as well as the variants and errors introduced in the tradition, complicate the task of the would-be stylometrist. Basing the analysis on the study of the copy from a single hand of several texts can partially mitigate these issues (Camps and Cafiero, 2013), but the limited availability of complete diplomatic transcriptions might make this difficult. In this paper, we use a workflow combining handwritten text recognition and stylometric analysis, applied to the case of the hagiographic works contained in MS BnF, fr. 412. We seek to evaluate Paul Meyer's hypothesis about the constitution of groups of hagiographic works, as well as to examine potential authorial groupings in a vastly anonymous corpus.

Motivation & Objective

• To evaluate Paul Meyer’s 1906 hypothesis on the compilation of Old French hagiographic legendiers, particularly the structural grouping of collections A, B, and C.
• To address the challenge of analyzing short, noisy medieval texts with high spelling variation and OCR errors using computational stylometry.
• To identify potential subseries and authorial groupings within the anonymous corpus of hagiographic lives in manuscript BnF fr. 412.
• To test whether stylometric analysis on HTR-processed data can reliably detect authorial or thematic groupings despite data noise and limited training data.
• To explore the feasibility of computational methods in re-evaluating 19th-century philological hypotheses about medieval manuscript transmission.

Proposed method

• Developed a multi-stage data pipeline: image acquisition → line detection → HTR using Transkribus and Kraken → text normalization → linguistic annotation (POS, lemmas).
• Applied stylometric analysis using character 3-grams as the baseline feature set, proven effective for noisy OCR and HTR data in prior studies.
• Compared alternative features (POS 3-grams, lemmas, function lemmas) to assess their robustness against cumulative error from preprocessing steps.
• Used hierarchical clustering and multidimensional scaling (MDS) to visualize textual similarities and identify groupings across the corpus.
• Combined multiple feature types (e.g., character 3-grams and lemmas) to improve stability in low-word-count texts (<3000 words).
• Conducted qualitative validation through close reading to assess stylometric groupings against known authorial markers (e.g., signatures, patrons, Latin substrates).

Experimental results

Research questions

• RQ1 Can stylometric analysis on HTR-processed data from medieval manuscripts reliably detect structural groupings proposed by Paul Meyer in his hypothesis about Old French legendiers?
• RQ2 Do the hagiographic lives in BnF fr. 412 form coherent subseries that align with Meyer’s proposed collections A, B, and C, or reveal new groupings?
• RQ3 Is there evidence of authorial authorship for the Life of Saint Lambert, which does not currently fit within Meyer’s framework or Wauchier de Denain’s known works?
• RQ4 How do different feature types (character n-grams, lemmas, POS 3-grams) perform in detecting groupings when applied to noisy, short medieval texts?
• RQ5 Can computational stylometry detect authorial signatures or stylistic markers in anonymous texts where traditional philological evidence is absent?

Key findings

• The stylometric analysis using character 3-grams on raw HTR data confirmed the structural integrity of Paul Meyer’s hypothesis regarding the composition of collections A, B, and C in the Old French legendiers.
• The analysis revealed a previously unnoted potential subseries, suggesting that the Life of Saint Longin may belong to collection B rather than A, challenging its current classification.
• A strong stylistic link was detected between Li Seint Confessor and the Life of Saint Lambert, despite the latter’s absence from collection C and lack of typical authorial markers.
• The Life of Saint Lambert was not grouped with Wauchier de Denain’s known works in most feature sets (e.g., POS 3-grams, lemmas), but its stylistic proximity to Li Seint Confessor suggests possible authorship by Wauchier, possibly as a translator without self-reference.
• While lemmatization and POS 3-grams offered complementary insights, they did not significantly outperform character 3-grams, likely due to cumulative error from preprocessing steps.
• The study demonstrates that stylometric analysis on HTR-processed data remains effective even for texts below 3000 words when using less sparse features like character 3-grams or feature concatenation, ensuring stable results across variable text lengths.