How do a neural network model knows if a title is a letter title?

We are glad to share our model to predict whether a phrase can be a letter title in Ming Dynasty. For example, this model can tell you 上丞相康思公書 can be a letter title. 劉先生墓誌銘 probabily can't be a letter title.

In collaboration with the Center for Chinese Studies 漢學研究中心 and the efforts of our research assistant Katherine Enright, the China Biographical Database project created 438,000 records training data for letter titles. We fine-tuned a transformer model by these training data. The author of the codes are Queenie Luo queenieluo[at]g.harvard.edu.