Expected sequence similarity maximization

Cyril Allauzen, Shankar Kumar, Wolfgang Macherey, Mehryar Mohri, Michael Riley

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

This paper presents efficient algorithms for expected similarity maximization, which coincides with minimum Bayes decoding for a similarity-based loss function. Our algorithms are designed for similarity functions that are sequence kernels in a general class of positive definite symmetric kernels. We discuss both a general algorithm and a more efficient algorithm applicable in a common unambiguous scenario. We also describe the application of our algorithms to machine translation and report the results of experiments with several translation data sets which demonstrate a substantial speed-up. In particular, our results show a speed-up by two orders of magnitude with respect to the original method of Tromble et al. (2008) and by a factor of 3 or more even with respect to an approximate algorithm specifically designed for that task. These results open the path for the exploration of more appropriate or optimal kernels for the specific tasks considered.

Original languageEnglish (US)
Title of host publicationNAACL HLT 2010 - Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Proceedings of the Main Conference
Pages957-965
Number of pages9
StatePublished - 2010
Event2010 Human Language Technologies Conference ofthe North American Chapter of the Association for Computational Linguistics, NAACL HLT 2010 - Los Angeles, CA, United States
Duration: Jun 2 2010Jun 4 2010

Other

Other2010 Human Language Technologies Conference ofthe North American Chapter of the Association for Computational Linguistics, NAACL HLT 2010
CountryUnited States
CityLos Angeles, CA
Period6/2/106/4/10

Fingerprint

scenario
experiment
Kernel
Decoding
Definites
Scenarios
Experiment
Machine Translation

ASJC Scopus subject areas

  • Language and Linguistics
  • Linguistics and Language

Cite this

Allauzen, C., Kumar, S., Macherey, W., Mohri, M., & Riley, M. (2010). Expected sequence similarity maximization. In NAACL HLT 2010 - Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Proceedings of the Main Conference (pp. 957-965)

Expected sequence similarity maximization. / Allauzen, Cyril; Kumar, Shankar; Macherey, Wolfgang; Mohri, Mehryar; Riley, Michael.

NAACL HLT 2010 - Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Proceedings of the Main Conference. 2010. p. 957-965.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Allauzen, C, Kumar, S, Macherey, W, Mohri, M & Riley, M 2010, Expected sequence similarity maximization. in NAACL HLT 2010 - Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Proceedings of the Main Conference. pp. 957-965, 2010 Human Language Technologies Conference ofthe North American Chapter of the Association for Computational Linguistics, NAACL HLT 2010, Los Angeles, CA, United States, 6/2/10.
Allauzen C, Kumar S, Macherey W, Mohri M, Riley M. Expected sequence similarity maximization. In NAACL HLT 2010 - Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Proceedings of the Main Conference. 2010. p. 957-965
Allauzen, Cyril ; Kumar, Shankar ; Macherey, Wolfgang ; Mohri, Mehryar ; Riley, Michael. / Expected sequence similarity maximization. NAACL HLT 2010 - Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Proceedings of the Main Conference. 2010. pp. 957-965
@inproceedings{614ca5f6db564bf689274f217e0edc2b,
title = "Expected sequence similarity maximization",
abstract = "This paper presents efficient algorithms for expected similarity maximization, which coincides with minimum Bayes decoding for a similarity-based loss function. Our algorithms are designed for similarity functions that are sequence kernels in a general class of positive definite symmetric kernels. We discuss both a general algorithm and a more efficient algorithm applicable in a common unambiguous scenario. We also describe the application of our algorithms to machine translation and report the results of experiments with several translation data sets which demonstrate a substantial speed-up. In particular, our results show a speed-up by two orders of magnitude with respect to the original method of Tromble et al. (2008) and by a factor of 3 or more even with respect to an approximate algorithm specifically designed for that task. These results open the path for the exploration of more appropriate or optimal kernels for the specific tasks considered.",
author = "Cyril Allauzen and Shankar Kumar and Wolfgang Macherey and Mehryar Mohri and Michael Riley",
year = "2010",
language = "English (US)",
isbn = "1932432655",
pages = "957--965",
booktitle = "NAACL HLT 2010 - Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Proceedings of the Main Conference",

}

TY - GEN

T1 - Expected sequence similarity maximization

AU - Allauzen, Cyril

AU - Kumar, Shankar

AU - Macherey, Wolfgang

AU - Mohri, Mehryar

AU - Riley, Michael

PY - 2010

Y1 - 2010

N2 - This paper presents efficient algorithms for expected similarity maximization, which coincides with minimum Bayes decoding for a similarity-based loss function. Our algorithms are designed for similarity functions that are sequence kernels in a general class of positive definite symmetric kernels. We discuss both a general algorithm and a more efficient algorithm applicable in a common unambiguous scenario. We also describe the application of our algorithms to machine translation and report the results of experiments with several translation data sets which demonstrate a substantial speed-up. In particular, our results show a speed-up by two orders of magnitude with respect to the original method of Tromble et al. (2008) and by a factor of 3 or more even with respect to an approximate algorithm specifically designed for that task. These results open the path for the exploration of more appropriate or optimal kernels for the specific tasks considered.

AB - This paper presents efficient algorithms for expected similarity maximization, which coincides with minimum Bayes decoding for a similarity-based loss function. Our algorithms are designed for similarity functions that are sequence kernels in a general class of positive definite symmetric kernels. We discuss both a general algorithm and a more efficient algorithm applicable in a common unambiguous scenario. We also describe the application of our algorithms to machine translation and report the results of experiments with several translation data sets which demonstrate a substantial speed-up. In particular, our results show a speed-up by two orders of magnitude with respect to the original method of Tromble et al. (2008) and by a factor of 3 or more even with respect to an approximate algorithm specifically designed for that task. These results open the path for the exploration of more appropriate or optimal kernels for the specific tasks considered.

UR - http://www.scopus.com/inward/record.url?scp=84863391321&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84863391321&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:84863391321

SN - 1932432655

SN - 9781932432657

SP - 957

EP - 965

BT - NAACL HLT 2010 - Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Proceedings of the Main Conference

ER -