Discovering archetypes of French detective novels using NLP (arxiv.org)

Researchers applied quantitative NLP to trace the evolution of the detective archetype in French crime fiction, training a supervised model with character-level embeddings on a 150-year corpus from M. Lecoq (1866) to Commissaire Adamsberg (2017). The model reliably learned a coherent “detective” embedding across the period, showing that computational features can capture the archetype’s persistent role signals despite stylistic and historical change. The work includes code, data and demos, and uses character-centric representations rather than whole-text stylometry to isolate role-specific patterns. Significantly for AI/ML and digital humanities, the study demonstrates that supervised embedding methods can detect and track literary archetypes diachronically, revealing narrative shifts: the detective moves from a subsidiary presence to the central “reasoning machine” of classical mysteries, and after WWII—with the influx of hardboiled influences—becomes morally ambiguous and entangled in social violence. Implications include new tools for computational narratology, character modeling, and genre analysis, as well as prospects for interpretable archetype detection, transfer learning across periods, and richer metadata generation for literary corpora.