Statistical phrase-based translation

TLDR

The study introduces a new phrase-based translation model and decoding algorithm designed to evaluate and compare existing phrase-based translation approaches. The authors conduct extensive experiments within this framework to assess and explain why phrase-based models outperform word-based models. Empirical results across all language pairs show that simple heuristic phrase learning from word alignments combined with lexical weighting yields the best performance, while longer phrases, high-accuracy alignments, or syntactically motivated phrases provide little benefit or even degrade results.

Abstract

We propose a new phrase-based translation model and decoding algorithm that enables us to evaluate and compare several, previously proposed phrase-based translation models. Within our framework, we carry out a large number of experiments to understand better and explain why phrase-based models out-perform word-based models. Our empirical results, which hold for all examined language pairs, suggest that the highest levels of performance can be obtained through relatively simple means: heuristic learning of phrase translations from word-based alignments and lexical weighting of phrase translations. Surprisingly, learning phrases longer than three words and learning phrases from high-accuracy word-level alignment models does not have a strong impact on performance. Learning only syntactically motivated phrases degrades the performance of our systems.