Ivan P. Yamshchikov
Max Planck Institute for Mathematics in the Sciences
Title: Generating Natural Language: capturing and representing style and semantics
Date: Friday, October 2, 2020
Place and Time: Zoom, 3:35-4:25 pm
In this talk, we discuss several bottlenecks of current natural language generation methods that use end-to-end statistical learning. We start with the general formulation of stylized language generation, introduce the empiric definition of literary style, and make a deep dive into the decomposition of stylistic and semantic information in the context of style transfer. The talk is an overview of several results on poetry generation, style transfer, and semantic representations.
Tikhonov A, Yamshchikov IP. Guess who? Multilingual approach for the automated generation of author-stylized poetry. In 2018 IEEE Spoken Language Technology Workshop (SLT) 2018 (pp. 787-794). IEEE.
Yamshchikov IP, Shibaev V, Nagaev A, Jost J, Tikhonov A. Decomposing Textual Information For Style Transfer. EMNLP-IJCNLP 2019.
Tikhonov A, Shibaev V, Nagaev A, Nugmanova A, Yamshchikov IP. Style transfer for texts: Retrain, report errors, compare with rewrites. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) 2019 Nov (pp. 3927-3936).