SEQ^3: Differentiable Sequence-to-Sequence-to-Sequence Autoencoder for Unsupervised Abstractive Sentence Compression

TL;DR

The paper introduces SEQ^3, an unsupervised sequence-to-sequence autoencoder that compresses sentences by learning discrete latent representations, enabling effective sentence compression without parallel data.

Contribution

It proposes a novel differentiable autoencoder with discrete latent variables for unsupervised sentence compression, combining length constraints and pretrained language models.

Findings

Achieves promising results on benchmark datasets

Operates without parallel text-summary pairs

Uses continuous relaxations for gradient-based training

Abstract

Neural sequence-to-sequence models are currently the dominant approach in several natural language processing tasks, but require large parallel corpora. We present a sequence-to-sequence-to-sequence autoencoder (SEQ^3), consisting of two chained encoder-decoder pairs, with words used as a sequence of discrete latent variables. We apply the proposed model to unsupervised abstractive sentence compression, where the first and last sequences are the input and reconstructed sentences, respectively, while the middle sequence is the compressed sentence. Constraining the length of the latent word sequences forces the model to distill important information from the input. A pretrained language model, acting as a prior over the latent sequences, encourages the compressed sentences to be human-readable. Continuous relaxations enable us to sample from categorical distributions, allowing gradient-based optimization, unlike alternatives that rely on reinforcement learning. The proposed model does not require parallel text-summary pairs, achieving promising results in unsupervised sentence compression on benchmark datasets.