Big Bidirectional Insertion Representations for Documents

TL;DR

This paper introduces Big BIRD, a scalable insertion-based model for document translation that uses sentence alignment to improve long document translation quality, achieving significant BLEU score gains.

Contribution

The paper presents Big BIRD, a novel insertion-based model with sentence alignment for improved long document translation performance.

Findings

Achieved +4.3 BLEU on WMT'19 English-German document translation

Scaled insertion models to handle long documents effectively

Introduced sentence-positional embeddings for source-target alignment

Abstract

The Insertion Transformer is well suited for long form text generation due to its parallel generation capabilities, requiring $O(\log_2 n)$ generation steps to generate $n$ tokens. However, modeling long sequences is difficult, as there is more ambiguity captured in the attention mechanism. This work proposes the Big Bidirectional Insertion Representations for Documents (Big BIRD), an insertion-based model for document-level translation tasks. We scale up the insertion-based models to long form documents. Our key contribution is introducing sentence alignment via sentence-positional embeddings between the source and target document. We show an improvement of +4.3 BLEU on the WMT'19 English$\rightarrow$German document-level translation task compared with the Insertion Transformer baseline.