Learning General Purpose Distributed Sentence Representations via Large Scale Multi-task Learning

TL;DR

The paper tackles learning general-purpose fixed-length sentence representations by training a single recurrent encoder across a diverse, large-scale multi-task set that includes multilingual NMT, skip-thought-like objectives, constituency parsing, and natural language inference. By sharing one encoder across tasks and using task-specific decoders, the approach captures a broad range of inductive biases, improving transfer performance on unseen tasks and enabling strong results in low-resource settings. Empirical results show consistent gains over prior methods, competitive word embeddings, and informative signals about how different tasks contribute to encoding sentence characteristics and syntax. The work demonstrates the practicality of multi-task, fixed-length sentence representations for broad NLP tasks and provides a foundation for future explorations in interpretability and controllable text generation.

Abstract

A lot of the recent success in natural language processing (NLP) has been driven by distributed vector representations of words trained on large amounts of text in an unsupervised manner. These representations are typically used as general purpose features for words across a range of NLP problems. However, extending this success to learning representations of sequences of words, such as sentences, remains an open problem. Recent work has explored unsupervised as well as supervised learning techniques with different training objectives to learn general purpose fixed-length sentence representations. In this work, we present a simple, effective multi-task learning framework for sentence representations that combines the inductive biases of diverse training objectives in a single model. We train this model on several data sources with multiple training objectives on over 100 million sentences. Extensive experiments demonstrate that sharing a single recurrent sentence encoder across weakly related tasks leads to consistent improvements over previous methods. We present substantial improvements in the context of transfer learning and low-resource settings using our learned general-purpose representations.

Figures (1)

• Figure 1: T-SNE visualizations of our sentence representations on 3 different datasets. SUBJ (left), TREC (middle), DBpedia (right). Dataset details are presented in the Appendix.