Statistical Phrase-based Translation

Philipp Koehn; Franz Josef Och; Daniel Marcu

doi:10.3115/1073445.1073462

Statistical Phrase-based Translation

Philipp Koehn, Franz Josef Och, Daniel Marcu

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

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.

Original language	English
Title of host publication	Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology - Volume 1
Place of Publication	Stroudsburg, PA, USA
Publisher	Association for Computational Linguistics
Pages	48-54
Number of pages	7
DOIs	https://doi.org/10.3115/1073445.1073462
Publication status	Published - 2003
Event	2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Langauge Technology (HLT-NAACL 2003) - Edmonton, Canada Duration: 27 May 2003 → 1 Jun 2003

Publication series

Name	NAACL '03
Publisher	Association for Computational Linguistics

Conference

Conference	2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Langauge Technology (HLT-NAACL 2003)
Country/Territory	Canada
City	Edmonton
Period	27/05/03 → 1/06/03

Access to Document

10.3115/1073445.1073462

Statistical Phrase-Based TranslationFinal published version, 127 KB

http://dx.doi.org/10.3115/1073445.1073462

Cite this

@inproceedings{075fddc1f8c847139a359284a1d8e167,

title = "Statistical Phrase-based Translation",

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.",

author = "Philipp Koehn and Och, {Franz Josef} and Daniel Marcu",

year = "2003",

doi = "10.3115/1073445.1073462",

language = "English",

series = "NAACL '03",

publisher = "Association for Computational Linguistics",

pages = "48--54",

booktitle = "Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology - Volume 1",

note = "2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Langauge Technology (HLT-NAACL 2003) ; Conference date: 27-05-2003 Through 01-06-2003",

}

Koehn, P, Och, FJ & Marcu, D 2003, Statistical Phrase-based Translation. in Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology - Volume 1. NAACL '03, Association for Computational Linguistics, Stroudsburg, PA, USA, pp. 48-54, 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Langauge Technology (HLT-NAACL 2003), Edmonton, Canada, 27/05/03. https://doi.org/10.3115/1073445.1073462

Statistical Phrase-based Translation. / Koehn, Philipp; Och, Franz Josef; Marcu, Daniel.

Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology - Volume 1. Stroudsburg, PA, USA : Association for Computational Linguistics, 2003. p. 48-54 (NAACL '03).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Statistical Phrase-based Translation

AU - Koehn, Philipp

AU - Och, Franz Josef

AU - Marcu, Daniel

PY - 2003

Y1 - 2003

N2 - 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.

AB - 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.

U2 - 10.3115/1073445.1073462

DO - 10.3115/1073445.1073462

M3 - Conference contribution

T3 - NAACL '03

SP - 48

EP - 54

BT - Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology - Volume 1

PB - Association for Computational Linguistics

CY - Stroudsburg, PA, USA

T2 - 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Langauge Technology (HLT-NAACL 2003)

Y2 - 27 May 2003 through 1 June 2003

ER -

Statistical Phrase-based Translation

Abstract

Publication series

Conference

Access to Document

Fingerprint

Cite this