MoRTy: Unsupervised Learning of Task-specialized Word Embeddings by Autoencoding

Nils Rethmeier, Barbara Plank

In: Proceedings of the 4th Workshop on Representation Learning for NLP (RepL4NLP-2019). Annual Meeting of the Association for Computational Linguistics (ACL-2019) The 57th Annual Meeting of the Association for Computational Linguistics (ACL) located at ACL July 28-August 2 Florence, Italy RepL4NLP Pages 49-54 Association for Computational Linguistics 7/2019.


Word embeddings have undoubtedly revolutionized NLP. However, pretrained embeddings do not always work for a specific task (or set of tasks), particularly in limited resource setups. We introduce a simple yet effective, self-supervised post-processing method that constructs task-specialized word representations by picking from a menu of reconstructing transformations to yield improved end-task performance (MORTY). The method is complementary to recent state-of-the-art approaches to inductive transfer via fine-tuning, and forgoes costly model architectures and annotation. We evaluate MORTY on a broad range of setups, including different word embedding methods, corpus sizes and end-task semantics. Finally, we provide a surprisingly simple recipe to obtain specialized embeddings that better fit end-tasks.


Weitere Links

W19-4307.pdf (pdf, 296 KB )

German Research Center for Artificial Intelligence
Deutsches Forschungszentrum für Künstliche Intelligenz