Explaining Character-Aware Neural Networks for Word-Level Prediction: Do They Discover Linguistic Rules?

TL;DR

This paper investigates whether character-aware neural networks learn linguistically meaningful patterns, revealing that these models implicitly discover understandable morphological rules across different languages.

Contribution

It extends the contextual decomposition technique to CNNs and LSTMs, enabling qualitative analysis of learned character-level patterns in NLP models.

Findings

Models discover linguistically meaningful patterns

Comparison of CNNs and LSTMs in morphological tagging

Models implicitly learn morphological rules

Abstract

Character-level features are currently used in different neural network-based natural language processing algorithms. However, little is known about the character-level patterns those models learn. Moreover, models are often compared only quantitatively while a qualitative analysis is missing. In this paper, we investigate which character-level patterns neural networks learn and if those patterns coincide with manually-defined word segmentations and annotations. To that end, we extend the contextual decomposition technique (Murdoch et al. 2018) to convolutional neural networks which allows us to compare convolutional neural networks and bidirectional long short-term memory networks. We evaluate and compare these models for the task of morphological tagging on three morphologically different languages and show that these models implicitly discover understandable linguistic rules. Our implementation can be found at https://github.com/FredericGodin/ContextualDecomposition-NLP .