Curriculum Learning for Natural Language Understanding

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6095–6104 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 6095 Curriculum Learning for Natural Language Understanding Benfeng Xu1∗, Licheng Zhang1∗ , Zhendong Mao1†, Quan Wang2 , Hongtao Xie1 and Yongdong Zhang1 1School of Information Science and Technology, University of Science and Technology of China, Hefei, China 2Beijing Research Institute, University of Science and Technology of China, Beijing, China {benfeng,zlczlc}@mail.ustc.edu.cn, quanwang1012@gmail.com {zdmao,htxie,zhyd73}@ustc.edu.cn Abstract With the great success of pre-trained lan- guage models, the pretrain-finetune paradigm now becomes the undoubtedly dominant solu- tion for natural language understanding (NLU) tasks. At the fine-tune stage, target task data is usually introduced in a completely random order and treated equally. However, examples in NLU tasks can vary greatly in difficulty, and similar to human learning procedure, language models can benefit from an easy-to-difficult curriculum. Based on this idea, we propose our Curriculum Learning approach. By re- viewing the trainset in a crossed way, we are able to distinguish easy examples from diffi- cult ones, and arrange a curriculum for lan- guage models. Without any manual model ar- chitecture design or use of external data, our Curriculum Learning approach obtains signifi- cant and universal performance improvements on a wide range of NLU tasks. 1 Introduction Natural Language Understanding (NLU), which re- quires machines to understand and reason with hu- man language, is a crucial yet challenging problem. Recently, language model (LM) pre-training has achieved remarkable success in NLU. Pre-trained LMs learn universal language representations from large-scale unlabeled data, and can be simply fine- tuned with a few adjustments to adapt to various NLU tasks, showing consistent and significant im- provements in these tasks (Radford et al., 2018; Devlin et al., 2018). While lots of attention has been devoted to de- signing better pre-training strategies (Yang et al., 2019; Liu et al., 2019; Raffel et al., 2019), it is also valuable to explore how to more effectively solve downstream NLU tasks in the fine-tuning stage. ∗Equal contribution. †Corresponding author. Easy cases: easy, comfortable positive most purely enjoyable positive most plain, unimaginative negative badly edited negative Hard cases: why didn’t Hollywood think of this sooner positive I simply can’t recommend it enough positive supposedly funny movie negative occasionally interesting negative Table 1: Examples from SST-2 sentiment classification task. Difficulty levels are determined by our review method (detailed later). Most current approaches perform fine-tuning in a straightforward manner, i.e., all training examples are treated equally and presented in a completely random order during training. However, even in the same NLU task, the training examples could vary significantly in their difficulty levels, with some easily solvable by simple lexical clues while others requiring sophisticated reasoning. Table 1 shows some examples from the SST-2 sentiment classifi- cation task (Socher et al., 2013), which identifies sentiment polarities (positive or negative) of movie reviews. The easy cases can be solved directly by identifying sentiment words such as “comfortable” and “unimaginative”, while the hard ones further require reasoning with …

Content truncated to 500 words