Scalable Reordering Models for SMT based on Multiclass SVM
In state-of-the-art phrase-based statistical machine translation systems, modelling phrase reorderings is an important need to enhance naturalness of the translated outputs, particularly when the grammatical structures of the language pairs differ significantly. Posing phrase movements as a classifi...
Main Authors: | , |
---|---|
Format: | Article |
Language: | English |
Published: |
Sciendo
2015-04-01
|
Series: | Prague Bulletin of Mathematical Linguistics |
Online Access: | https://doi.org/10.1515/pralin-2015-0004 |
id |
doaj-a3f1687bb4274cfbaf89d27de5eceb6b |
---|---|
record_format |
Article |
spelling |
doaj-a3f1687bb4274cfbaf89d27de5eceb6b2021-09-05T13:59:53ZengSciendoPrague Bulletin of Mathematical Linguistics 1804-04622015-04-011031658410.1515/pralin-2015-0004pralin-2015-0004Scalable Reordering Models for SMT based on Multiclass SVMAlrajeh Abdullah0Niranjan Mahesan1School of Electronics and Computer Science, University of Southampton/Computer Research Institute, King Abdulaziz City for Science and Technology (KACST)Computer Research Institute, King Abdulaziz City for Science and Technology (KACST)In state-of-the-art phrase-based statistical machine translation systems, modelling phrase reorderings is an important need to enhance naturalness of the translated outputs, particularly when the grammatical structures of the language pairs differ significantly. Posing phrase movements as a classification problem, we exploit recent developments in solving large-scale multiclass support vector machines. Using dual coordinate descent methods for learning, we provide a mechanism to shrink the amount of training data required for each iteration. Hence, we produce significant computational saving while preserving the accuracy of the models. Our approach is a couple of times faster than maximum entropy approach and more memory-efficient (50% reduction). Experiments were carried out on an Arabic-English corpus with more than a quarter of a billion words. We achieve BLEU score improvements on top of a strong baseline system with sparse reordering features.https://doi.org/10.1515/pralin-2015-0004 |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Alrajeh Abdullah Niranjan Mahesan |
spellingShingle |
Alrajeh Abdullah Niranjan Mahesan Scalable Reordering Models for SMT based on Multiclass SVM Prague Bulletin of Mathematical Linguistics |
author_facet |
Alrajeh Abdullah Niranjan Mahesan |
author_sort |
Alrajeh Abdullah |
title |
Scalable Reordering Models for SMT based on Multiclass SVM |
title_short |
Scalable Reordering Models for SMT based on Multiclass SVM |
title_full |
Scalable Reordering Models for SMT based on Multiclass SVM |
title_fullStr |
Scalable Reordering Models for SMT based on Multiclass SVM |
title_full_unstemmed |
Scalable Reordering Models for SMT based on Multiclass SVM |
title_sort |
scalable reordering models for smt based on multiclass svm |
publisher |
Sciendo |
series |
Prague Bulletin of Mathematical Linguistics |
issn |
1804-0462 |
publishDate |
2015-04-01 |
description |
In state-of-the-art phrase-based statistical machine translation systems, modelling phrase reorderings is an important need to enhance naturalness of the translated outputs, particularly when the grammatical structures of the language pairs differ significantly. Posing phrase movements as a classification problem, we exploit recent developments in solving large-scale multiclass support vector machines. Using dual coordinate descent methods for learning, we provide a mechanism to shrink the amount of training data required for each iteration. Hence, we produce significant computational saving while preserving the accuracy of the models. Our approach is a couple of times faster than maximum entropy approach and more memory-efficient (50% reduction). Experiments were carried out on an Arabic-English corpus with more than a quarter of a billion words. We achieve BLEU score improvements on top of a strong baseline system with sparse reordering features. |
url |
https://doi.org/10.1515/pralin-2015-0004 |
work_keys_str_mv |
AT alrajehabdullah scalablereorderingmodelsforsmtbasedonmulticlasssvm AT niranjanmahesan scalablereorderingmodelsforsmtbasedonmulticlasssvm |
_version_ |
1717812839873249280 |