Parsing Heterogeneous Corpora with a Rich Dependency Grammar

Philologist: I need to parse Old French texts of different types (verse, prose, dialects etc.). Do I have to train separate parser models? Computational Linguist: You won’t lose much if you train the parser on all the data you have. P: I can’t do the training myself. What can I expect from existing parser models? C: If the training corpus contained 12th century verse texts, you are best prepared for most flavours of Old French, including prose ––– except for the very oldest texts. P: And if I want to parse very old texts? C: Then the time lapse between your text and the training data should be as small as possible. P: A golden rule to go home with? C: Don’t train on prose if you want to parse verse. P: AOI.