Fast BTG-Forest-Based Hierarchical Sub-sentential Alignment

TL;DR

This paper introduces a fast, hierarchical sub-sentential alignment method based on BTG-forest, which improves phrase table size and translation quality, especially for distant language pairs, while maintaining efficiency.

Contribution

It presents a novel BTG-forest-based alignment approach with a fast unsupervised initialization, achieving comparable speed to fast_align but with smaller phrase tables and better performance on distant languages.

Findings

Achieves similar runtime to fast_align

Produces smaller phrase tables

Outperforms in English-Japanese translation

Abstract

In this paper, we propose a novel BTG-forest-based alignment method. Based on a fast unsupervised initialization of parameters using variational IBM models, we synchronously parse parallel sentences top-down and align hierarchically under the constraint of BTG. Our two-step method can achieve the same run-time and comparable translation performance as fast_align while it yields smaller phrase tables. Final SMT results show that our method even outperforms in the experiment of distantly related languages, e.g., English-Japanese.