Rapid progress of deep learning technologies for natural language processing opened up new opportunities for its application. The goal of this project is to create methodology of how the recent Transformer deep learning architecture can be employed for automatic text rewriting/editing. Model will be trained with pairs of flawed – correct text samples and in the end will be able to rewrite the given “illiterate” text into the correct one. It will be the first one monolingual Transformer for Lithuanian language and will use a novel training data by messing text in a various level. Results of this project will be twofold: 1) an automatic text editing, rewriting tool and 2) publicly available weights for other researchers to fine-tune their Lithuanian text tasks.
KTU Research and Innovation Fund
Period of project implementation: 2021-04-01 - 2021-12-31
Project partners: Vytautas Magnus University