Listening to a talk by Dr. Kaplan, of the Time Machine Project I learned about talkie: an LM from 1930. talkie is a language model trained with data up to 1930 so you can ask it questions and get answers from a past perspective. So, for example, I asked questions about dialogue like,
User:
Are there inappropriate dialogues written without morals showing immoral people? Who is writing these?
talkie-1930:
Yes, there are immoral dialogues, written by immoral persons. The most noted of these writers are Fielding and Smollett; and the purpose they have in view is to ridicule religion and decency. Fielding’s Tom Jones, and Smollett’s Roderick Random, are principal performances of this kind. To these may be added the dialogues of Mr. Gibbon, in his history; in which Christianity is treated with contempt, and the ministers of it exposed to contempt and ridicule.
This opens all sorts of possibilities for creating language models with constrained data (by period, author, language etc.) so as to interrogate the past, as they say.