Dynamic Topic Adaptation for Phrase-based MT

Eva Hasler; Phil Blunsom; Philipp Koehn; Barry Haddow

doi:10.3115/v1/E14-1035

Dynamic Topic Adaptation for Phrase-based MT

Eva Hasler, Phil Blunsom, Philipp Koehn, Barry Haddow

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

Abstract

Translating text from diverse sources poses a challenge to current machine translation systems which are rarely adapted to structure beyond corpus level. We explore topic adaptation on a diverse data set and present a new bilingual variant of Latent Dirichlet Allocation to compute topic-adapted, probabilistic phrase translation features. We dynamically infer document-specific translation probabilities for test sets of unknown origin, thereby capturing the effects of document context on phrase translations. We show gains of up to 1.26 BLEU over the baseline and 1.04 over a domain adaptation benchmark. We further provide an analysis of the domain-specific data and show additive gains of our model in combination with other types of topic-adapted features.

Original language	English
Title of host publication	Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2014, April 26-30, 2014, Gothenburg, Sweden
Place of Publication	Gothenburg, Sweden
Publisher	Association for Computational Linguistics (ACL)
Pages	328-337
Number of pages	10
DOIs	https://doi.org/10.3115/v1/E14-1035
Publication status	Published - Apr 2014
Event	14th Conference of the European Chapter of the Association for Computational Linguistics - Chalmers University, Gothenburg, Sweden Duration: 26 Apr 2014 → 30 Apr 2014 http://eacl2014.org/

Conference

Conference	14th Conference of the European Chapter of the Association for Computational Linguistics
Abbreviated title	EACL 2014
Country/Territory	Sweden
City	Gothenburg
Period	26/04/14 → 30/04/14
Internet address	http://eacl2014.org/

Access to Document

10.3115/v1/E14-1035Licence: Creative Commons: Attribution (CC-BY)

Koehn_Hasler et al 2014 Dynamic Topic AdaptationAccepted author manuscript, 287 KB

http://aclweb.org/anthology/E/E14/E14-1035.pdfLicence: Creative Commons: Attribution (CC-BY)

EU-Bridge
Renals, S. (Principal Investigator), King, S. (Co-investigator), Koehn, P. (Co-investigator) & Osborne, M. (Co-investigator)
EU government bodies
1/02/12 → 31/01/15
Project: Research
ACCEPT (Automated Community Content Editing Portal)
Koehn, P. (Principal Investigator)
EU government bodies
1/01/12 → 31/12/14
Project: Research

Barry Haddow
- bhaddowinf.ed.acuk
- School of Informatics - Senior Research Fellow
- Institute of Language, Cognition and Computation
- Language, Interaction, and Robotics
Person: Academic: Research Active (Research Assistant)

Cite this

@inproceedings{605001b36748452ab5786cbfb456f0e1,

title = "Dynamic Topic Adaptation for Phrase-based MT",

abstract = "Translating text from diverse sources poses a challenge to current machine translation systems which are rarely adapted to structure beyond corpus level. We explore topic adaptation on a diverse data set and present a new bilingual variant of Latent Dirichlet Allocation to compute topic-adapted, probabilistic phrase translation features. We dynamically infer document-specific translation probabilities for test sets of unknown origin, thereby capturing the effects of document context on phrase translations. We show gains of up to 1.26 BLEU over the baseline and 1.04 over a domain adaptation benchmark. We further provide an analysis of the domain-specific data and show additive gains of our model in combination with other types of topic-adapted features.",

author = "Eva Hasler and Phil Blunsom and Philipp Koehn and Barry Haddow",

year = "2014",

month = apr,

doi = "10.3115/v1/E14-1035",

language = "English",

pages = "328--337",

booktitle = "Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2014, April 26-30, 2014, Gothenburg, Sweden",

publisher = "Association for Computational Linguistics (ACL)",

note = "14th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2014 ; Conference date: 26-04-2014 Through 30-04-2014",

url = "http://eacl2014.org/",

}

Hasler, E, Blunsom, P, Koehn, P & Haddow, B 2014, Dynamic Topic Adaptation for Phrase-based MT. in Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2014, April 26-30, 2014, Gothenburg, Sweden. Association for Computational Linguistics (ACL), Gothenburg, Sweden, pp. 328-337, 14th Conference of the European Chapter of the Association for Computational Linguistics, Gothenburg, Sweden, 26/04/14. https://doi.org/10.3115/v1/E14-1035

Dynamic Topic Adaptation for Phrase-based MT. / Hasler, Eva; Blunsom, Phil; Koehn, Philipp et al.
Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2014, April 26-30, 2014, Gothenburg, Sweden. Gothenburg, Sweden: Association for Computational Linguistics (ACL), 2014. p. 328-337.

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

TY - GEN

T1 - Dynamic Topic Adaptation for Phrase-based MT

AU - Hasler, Eva

AU - Blunsom, Phil

AU - Koehn, Philipp

AU - Haddow, Barry

PY - 2014/4

Y1 - 2014/4

N2 - Translating text from diverse sources poses a challenge to current machine translation systems which are rarely adapted to structure beyond corpus level. We explore topic adaptation on a diverse data set and present a new bilingual variant of Latent Dirichlet Allocation to compute topic-adapted, probabilistic phrase translation features. We dynamically infer document-specific translation probabilities for test sets of unknown origin, thereby capturing the effects of document context on phrase translations. We show gains of up to 1.26 BLEU over the baseline and 1.04 over a domain adaptation benchmark. We further provide an analysis of the domain-specific data and show additive gains of our model in combination with other types of topic-adapted features.

AB - Translating text from diverse sources poses a challenge to current machine translation systems which are rarely adapted to structure beyond corpus level. We explore topic adaptation on a diverse data set and present a new bilingual variant of Latent Dirichlet Allocation to compute topic-adapted, probabilistic phrase translation features. We dynamically infer document-specific translation probabilities for test sets of unknown origin, thereby capturing the effects of document context on phrase translations. We show gains of up to 1.26 BLEU over the baseline and 1.04 over a domain adaptation benchmark. We further provide an analysis of the domain-specific data and show additive gains of our model in combination with other types of topic-adapted features.

U2 - 10.3115/v1/E14-1035

DO - 10.3115/v1/E14-1035

M3 - Conference contribution

SP - 328

EP - 337

BT - Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2014, April 26-30, 2014, Gothenburg, Sweden

PB - Association for Computational Linguistics (ACL)

CY - Gothenburg, Sweden

T2 - 14th Conference of the European Chapter of the Association for Computational Linguistics

Y2 - 26 April 2014 through 30 April 2014

ER -

Dynamic Topic Adaptation for Phrase-based MT

Abstract

Conference

Access to Document

Fingerprint

Projects

EU-Bridge

ACCEPT (Automated Community Content Editing Portal)

Profiles

Barry Haddow

Cite this