Hopkins and May, EMNLP 2011. Tuning as Ranking

Citation

Mark Hopkins and Jonathan May. 2011. Tuning as Ranking. In Proceedings of EMNLP-2011.

Online Version

Tuning as Ranking

Summary

This paper presents a simple and scalable method for statistical machine translation parameter tuning based on the pairwise approach to ranking. This pairwise ranking optimization (PRO) method has advantages over MERT Och, 2003 as it is not limited to a handful of parameters and can easily handle systems with thousands of features. In addition, unlike recent approaches built upon the MIRA algorithm of Crammer and Singer, 2003 (Watanabe et al., 2007), PRO is easy to implement.

Method

Although, MERT is well-understood, easy to implement, and runs quickly, it can behave erratically and would not scale beyond a handful of features. This is a major bottleneck towards working with richer feature representations and structure.

Hence, the authors propose a simpler approach than MIRA, to tuning that similarly scales to high-dimensional feature spaces. Tuning is treated as a ranking problem (Chen et al., 2009), where the explicit goal is to learn to correctly rank candidate translations. The authors describe a pairwise approach to ranking, in which the ranking problem is reduced to the binary classification task of deciding between candidate translation pairs.

Approach

The goal of tuning is to learn a weight vector ${\textbf {w}}$ such that $H_{w}(p)$ assigns a high score to good translations and a low score to bad translations. $H_{w}(p)$ is given by:

$H_{w}(p)=\sum _{i\in I}h_{w}(i,p(i))$

where, the scores for candidate translations $p(i)$ are represented in the following form, $h_{w}(i,p(i))={\textbf {w}}\cdot {\textbf {x}}(i,j)$

Optimization via Pairwise Ranking

MIRA scales well to high-dimensionality candidate spaces. However, its architecture is complex and different to that of MERT. This method assumes a gold scoring function which can be decomposed in the following way:

$G(p)=\sum _{i\in I}g(i,p(i))$ where, $g(i,j)$ is a local scoring function that scores each candidate translation $e(i,j)$.

In a pairwise ranking approach, the learning task is framed as the classification of candidate pairs into two categories: correctly ordered and incorrectly ordered. For each pair of translation candidates $e(i,j)$ and $e(i,j^{\prime })$, an inequality relation is established:

$g(i,j)>g(i,j^{\prime })$