Automatic thematic indexing has transformed the way large literary corpora are accessed, particularly in the case of Voltaire's complete works. A recent study by Miguel Arana-Catania, Gillian Pink, and Glenn Roe, submitted on July 10, 2026, explores the use of machine learning to automate this traditionally manual process.
The research focuses on two significant sub-corpora of Voltaire’s works: the Essai sur les mœurs et l'esprit des nations and the Questions sur l'Encyclopédie. The authors frame the task as a multi-label classification problem, where a model must assign index entries similar to those used by professional indexers for specific text sections.
Machine Learning Techniques for Thematic Indexing
The study compares various approaches, including encoder-based models with classification heads and generative large language models (LLMs) that are fine-tuned using Low-Rank Adaptation (LoRA). These models range in size from approximately 3 billion to 120 billion parameters.
The best-performing model, part of the Mistral family and configured in a 4-bit quantized format, achieved F1 scores of up to 0.67. The authors argue that these scores represent lower bounds due to the subjective nature of professional indexing.
Implications for Literary and Historical Access
Significantly, the study assesses cross-corpus generalization and includes a qualitative analysis of the model's handling of literary and rhetorical features. These elements often resist automated treatment, highlighting the challenges faced in providing structured thematic access to extensive literary and historical texts.
The findings have substantial implications for digital libraries and information retrieval systems. By enhancing thematic indexing, the research aims to improve scholarly access to large-scale literary editions, making them more navigable for researchers and readers alike.
Future Directions in Thematic Indexing
As machine learning continues to evolve, the potential for further advancements in thematic indexing is vast. The ability to automate and refine this process could lead to more efficient access to a multitude of literary works, beyond just those of Voltaire.
This research marks a significant step towards integrating artificial intelligence into the humanities, paving the way for future innovations in literary scholarship.
🤖 This article was rewritten by Feed and Figures' editorial AI from a report originally published by arXiv NLP. Facts and quotes are preserved from the original; the rewrite focuses on clarity and structure. For the unedited original, see the source link below.
