Learning Disentangled Representations for Natural Language Definitions
This work addresses the need for better interpretability and control in neural models for NLP, though it is incremental as it builds on existing disentanglement methods by applying them to a specific type of textual data.
The paper tackled the problem of learning disentangled representations in NLP by leveraging semantic structures in definitional sentences to train a Variational Autoencoder, resulting in improved performance over unsupervised baselines on disentanglement benchmarks and enhanced results in definition modeling.
Disentangling the encodings of neural models is a fundamental aspect for improving interpretability, semantic control and downstream task performance in Natural Language Processing. Currently, most disentanglement methods are unsupervised or rely on synthetic datasets with known generative factors. We argue that recurrent syntactic and semantic regularities in textual data can be used to provide the models with both structural biases and generative factors. We leverage the semantic structures present in a representative and semantically dense category of sentence types, definitional sentences, for training a Variational Autoencoder to learn disentangled representations. Our experimental results show that the proposed model outperforms unsupervised baselines on several qualitative and quantitative benchmarks for disentanglement, and it also improves the results in the downstream task of definition modeling.