Scene Restoring for Narrative Machine Reading Comprehension

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 332–348, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 332 FIND: Human-in-the-Loop Debugging Deep Text Classifiers Piyawat Lertvittayakumjorn, Lucia Specia, Francesca Toni Department of Computing, Imperial College London, UK {pl1515, l.specia, ft}@imperial.ac.uk Abstract Since obtaining a perfect training dataset (i.e., a dataset which is considerably large, unbi- ased, and well-representative of unseen cases) is hardly possible, many real-world text classi- fiers are trained on the available, yet imperfect, datasets. These classifiers are thus likely to have undesirable properties. For instance, they may have biases against some sub-populations or may not work effectively in the wild due to overfitting. In this paper, we propose FIND – a framework which enables humans to debug deep learning text classifiers by disabling irrel- evant hidden features. Experiments show that by using FIND, humans can improve CNN text classifiers which were trained under different types of imperfect datasets (including datasets with biases and datasets with dissimilar train- test distributions). 1 Introduction Deep learning has become the dominant approach to address most Natural Language Processing (NLP) tasks, including text classification. With suf- ficient and high-quality training data, deep learning models can perform incredibly well (Zhang et al., 2015; Wang et al., 2019). However, in real-world cases, such ideal datasets are scarce. Often times, the available datasets are small, full of regular but irrelevant words, and contain unintended biases (Wiegand et al., 2019; Gururangan et al., 2018). These can lead to suboptimal models with unde- sirable properties. For example, the models may have biases against some sub-populations or may not work effectively in the wild as they overfit the imperfect training data. To improve the models, previous work has looked into different techniques beyond standard model fitting. If the weaknesses of the training datasets or the models are anticipated, strategies can be tailored to mitigate such weaknesses. For example, augmenting the training data with gender- swapped input texts helps reduce gender bias in the models (Park et al., 2018; Zhao et al., 2018). Adver- sarial training can prevent the models from exploit- ing irrelevant and/or protected features (Jaiswal et al., 2019; Zhang et al., 2018). With a limited number of training examples, using human ratio- nales or prior knowledge together with training labels can help the models perform better (Zaidan et al., 2007; Bao et al., 2018; Liu and Avci, 2019). Nonetheless, there are side-effects of sub- optimal datasets that cannot be predicted and are only found after training thanks to post-hoc error analysis. To rectify such problems, there have been attempts to enable humans to fix the trained models (i.e., to perform model debugging) (Stumpf et al., 2009; Teso and Kersting, 2019). Since the mod- els are usually too complex to understand, manu- ally modifying the model parameters is not possi- ble. Existing techniques, therefore, allow humans to provide feedback on individual predictions in- stead. Then, additional training examples are cre- ated based on the feedback to retrain the models. However, such local …

Content truncated to 500 words