Clues in the Text: Stylistic Signals of Success in Early Detective Fiction

This thesis investigates whether measurable stylistic features are associated with literary “success” in late nineteenth- and early twentieth-century detective fiction, and whether those features can support prediction within a controlled corpus. The study analyzes 281 English-language detective fiction works (1880–1935) from Project Gutenberg. Success is operationalized as contemporary public-domain engagement using a fixed snapshot of download counts (20 December 2025): texts in the top decile are labeled “successful,” and the remainder form the comparison group. Situated in digital humanities and computational stylistics, the research adopts a compact, interpretable feature set capturing broad dimensions of narrative language: lexical concreteness, affective arousal, narrative stance (first- and third-person pronoun density), and syntactic packaging (subordination density). A lexical domain analysis further examines how concreteness is employed in semantic fields related to perception, inference, and procedural detection, contextualized through genre expectations often discussed via the Holmesian tradition. Empirical evaluation combines regression modelling of download intensity with an exposure control for time since Project Gutenberg release and supervised classification (logistic regression, random forest, and a feed-forward neural network) with conservative cross-validation and held-out testing. Results indicate that stylistic properties tied to emotional activation, lexical abstraction, and syntactic complexity are meaningfully associated with the download-based proxy of success, and that simple, interpretable models can achieve above-chance predictive performance, though conclusions are tempered by class imbalance and the small number of successful texts.