Fine-tune language models on 566 surviving early modern plays, watch them classify genre, company, playhouse and print — and make them explain every verdict, word by word. Built for scholars: no code, no installation, a browser is enough.
This page is the project's permanent home. The interactive tools below run live on the research group's GPU server; if they are briefly offline, this page — and the project — remain here.
Pick any field of the DEEP catalogue — genre, company, theater, format, decade — choose 2–6 classes, and fine-tune BERT on the corpus. Results stay browsable: per-play votes, confusion matrices, LIME & SHAP explanations.
Open the trainer → HANDS-ON TUTORIALNine small machines to poke: fill-in-the-blank with BERT, the tokenizer's scissors, word-space arithmetic, nearest-play search, live judging of your own passages — with explanations — and a generator to compare.
Enter the playground →Play texts are diplomatic transcriptions of the earliest editions from EEBO-TCP; every label comes from DEEP: Database of Early English Playbooks. Genre by the Annals of English Drama:
This site accompanies the workshop Training and Fine-Tuning Large Language Models for Early Modern Drama Research. Models run live on the research group's GPU server; every result on this site is computed for your input at the moment you ask.
Please cite: Heejin Kim, Interpretable Deep Learning Stylometry of Early Modern English Drama, Kyungpook National University — https://stylometry.digihumeng.org. Code and data accompany a forthcoming article and will be archived openly (with a DOI) upon publication.