Revisiting Iterative Back-Translation from the Perspective of Compositional Generalization

TL;DR

This paper investigates how iterative back-translation can enhance compositional generalization in neural seq2seq models, demonstrating significant improvements on benchmarks and proposing curriculum strategies to further boost performance.

Contribution

The study empirically shows iterative back-translation improves compositional generalization and introduces curriculum iterative back-translation for better pseudo-parallel data quality.

Findings

Iterative back-translation significantly improves compositional generalization.

It corrects errors in pseudo-parallel data over iterations.

Curriculum iterative back-translation further enhances performance.

Abstract

Human intelligence exhibits compositional generalization (i.e., the capacity to understand and produce unseen combinations of seen components), but current neural seq2seq models lack such ability. In this paper, we revisit iterative back-translation, a simple yet effective semi-supervised method, to investigate whether and how it can improve compositional generalization. In this work: (1) We first empirically show that iterative back-translation substantially improves the performance on compositional generalization benchmarks (CFQ and SCAN). (2) To understand why iterative back-translation is useful, we carefully examine the performance gains and find that iterative back-translation can increasingly correct errors in pseudo-parallel data. (3) To further encourage this mechanism, we propose curriculum iterative back-translation, which better improves the quality of pseudo-parallel data, thus further improving the performance.