Abstract

This work investigates Differential Object Marking (DOM) in Brazilian Portuguese (BP), specifically a-marked objects, or prepositional accusatives (PP-ACCs), across four variables: semantic requirements, constituent order, verb semantics, and textual genre.An optimized parsing model was trained to recognize instances of PP-ACCs and automatically annotate historical documents for these objects for the Tycho Brahe and Colonia corpora. Contrary to expectations based on the low frequency of these objects and prior diachronic studies on European Portuguese (EP), our results reveal that PP-ACCs remain present in BP from the 18th century onward. Our findings confirm previous patterns for EP and present textual genre (specifically, narrative texts and theater plays) as a possible relevant variable, but warrants further investigation. Constituent order was proved to be less significant than previously suggested. This work also reveals methodological challenges in using computational models and NLP tools for research in historical Portuguese.