TY - JOUR AU - AB - the performances are fantastic the films are fantastic the movies are fantastic positive We propose a novel data augmentation for the stories are fantastic labeled sentences called contextual augmen- tation. We assume an invariance that sen- tences are natural even if the words in the performances sentences are replaced with other words with films paradigmatic relations. We stochastically re- movies stories place words with other words that are pre- dicted by a bi-directional language model at the word positions. Words predicted accord- positive ing to a context are numerous but appropri- ate for the augmentation of the original words. Furthermore, we retrofit a language model with a label-conditional architecture, which al- the actors are fantastic lows the model to augment sentences without the actors are fantastic positive breaking the label-compatibility. Through the experiments for six various different text clas- sification tasks, we demonstrate that the pro- Figure 1: Contextual augmentation with a bi- posed method improves classifiers based on directional RNN language model, when a sentence the convolutional or recurrent neural networks. “the actors are fantastic” is augmented by replacing only actors with words predicted based on the context. 1 Introduction Neural network-based models for NLP have been However, TI - Contextual Augmentation: Data Augmentation by Words with Paradigmatic Relations JF - Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers) DO - 10.18653/v1/n18-2072 DA - 2018-01-01 UR - https://www.deepdyve.com/lp/unpaywall/contextual-augmentation-data-augmentation-by-words-with-paradigmatic-juCQYC2iZr DP - DeepDyve ER -