Source-side Preordering for Translation using Logistic Regression and Depth-first Branch-and-Bound Search

Abstract

We present a simple preordering approach for machine translation based on a featurerich logistic regression model to predict whether two children of the same node in the source-side parse tree should be swapped or not. Given the pair-wise children regression scores we conduct an efficient depth-first branch-and-bound search through the space of possible children permutations, avoiding using a cascade of classifiers or limiting the list of possible ordering outcomes. We report experiments in translating English to Japanese and Korean, demonstrating superior performance as (a) the number of crossing links drops by more than 10% absolute with respect to other state-of-the-art preordering approaches, (b) BLEU scores improve on 2.2 points over the baseline with lexicalised reordering model, and (c) decoding can be carried out 80 times faster.

Recursive Neural Network Based Preordering for English-to-Japanese Machine Translation

“…Preordering reorders source sentences before translation, while post-ordering reorders sentences translated without considering the word order after translation. In particular, preordering effectively improves the translation quality because it solves long-distance reordering and computational complexity issues (Jehl et al, 2014;Nakagawa, 2015).…”

A POS-Based Preordering Approach for English-to-Arabic Statistical Machine Translation

“…For the task of translation from Chinese to Japanese, Sudoh et al [8] used a learning-to-rank model based on a pairwise classification method to predict the target Japanese word order. In the same spirit, Jehl et al [9] proposed a featurebased reordering model for English-to-Japanese and English-to-Korean translation. Their model predicts whether a pair of sibling nodes on the source-side of the parse tree needs to be swapped.…”

Examining the Relationship between Preordering and Word Order Freedom in Machine Translation

“…As we are interested in the bilingual case and, specifically, in preordering, we content ourselves with using the same syntactic representation, i.e. dependency trees, that many preordering models use (e.g., Jehl et al (2014), Lerner and Petrov (2013)). …”

“…More recently, Jehl et al (2014) learn to order sibling nodes in the source-side dependency parse tree. The space of possible permutations is explored via depth-first branch-and-bound search (Balas and Toth, 1983).…”

“…Based on their earlier work, which used logistic regression and graph search for preordering (Jehl et al, 2014), de Gispert et al (2015 introduce a neural preordering model. In this model, a feedforward neural network is trained to estimate the swap probabilities of nodes in the source-side dependency tree.…”