N-way composition of weighted finite-state transducers

Cyril Allauzen, Mehryar Mohri

Research output: Contribution to journalArticle

Abstract

Composition of weighted transducers is a fundamental algorithm used in many applications, including for computing complex edit-distances between automata, or string kernels in machine learning, or to combine different components of a speech recognition, speech synthesis, or information extraction system. We present a generalization of the composition of weighted transducers, n-way composition, which is dramatically faster in practice than the standard composition algorithm when combining more than two transducers. The worst-case complexity of our algorithm for composing three transducers T1, T2, and T3 resulting in T, is O(|T|Q min(d(T1)d(T3), d(T2)) + |T|E), where |·|Q denotes the number of states, |·| E the number of transitions, and d(·) the maximum out-degree. As in regular composition, the use of perfect hashing requires a pre-processing step with linear-time expected complexity in the size of the input transducers. In many cases, this approach significantly improves on the complexity of standard composition. Our algorithm also leads to a dramatically faster composition in practice. Furthermore, standard composition can be obtained as a special case of our algorithm. We report the results of several experiments demonstrating this improvement. These theoretical and empirical improvements significantly enhance performance in the applications already mentioned.

Original languageEnglish (US)
Pages (from-to)613-627
Number of pages15
JournalInternational Journal of Foundations of Computer Science
Volume20
Issue number4
DOIs
StatePublished - Aug 2009

Fingerprint

Transducers
Chemical analysis
Speech synthesis
Speech recognition
Learning systems
Processing
Experiments

ASJC Scopus subject areas

  • Computer Science (miscellaneous)

Cite this

N-way composition of weighted finite-state transducers. / Allauzen, Cyril; Mohri, Mehryar.

In: International Journal of Foundations of Computer Science, Vol. 20, No. 4, 08.2009, p. 613-627.

Research output: Contribution to journalArticle

@article{0c10201ec7dc4dfb8e0bef600eb4e9f0,
title = "N-way composition of weighted finite-state transducers",
abstract = "Composition of weighted transducers is a fundamental algorithm used in many applications, including for computing complex edit-distances between automata, or string kernels in machine learning, or to combine different components of a speech recognition, speech synthesis, or information extraction system. We present a generalization of the composition of weighted transducers, n-way composition, which is dramatically faster in practice than the standard composition algorithm when combining more than two transducers. The worst-case complexity of our algorithm for composing three transducers T1, T2, and T3 resulting in T, is O(|T|Q min(d(T1)d(T3), d(T2)) + |T|E), where |·|Q denotes the number of states, |·| E the number of transitions, and d(·) the maximum out-degree. As in regular composition, the use of perfect hashing requires a pre-processing step with linear-time expected complexity in the size of the input transducers. In many cases, this approach significantly improves on the complexity of standard composition. Our algorithm also leads to a dramatically faster composition in practice. Furthermore, standard composition can be obtained as a special case of our algorithm. We report the results of several experiments demonstrating this improvement. These theoretical and empirical improvements significantly enhance performance in the applications already mentioned.",
author = "Cyril Allauzen and Mehryar Mohri",
year = "2009",
month = "8",
doi = "10.1142/S0129054109006772",
language = "English (US)",
volume = "20",
pages = "613--627",
journal = "International Journal of Foundations of Computer Science",
issn = "0129-0541",
publisher = "World Scientific Publishing Co. Pte Ltd",
number = "4",

}

TY - JOUR

T1 - N-way composition of weighted finite-state transducers

AU - Allauzen, Cyril

AU - Mohri, Mehryar

PY - 2009/8

Y1 - 2009/8

N2 - Composition of weighted transducers is a fundamental algorithm used in many applications, including for computing complex edit-distances between automata, or string kernels in machine learning, or to combine different components of a speech recognition, speech synthesis, or information extraction system. We present a generalization of the composition of weighted transducers, n-way composition, which is dramatically faster in practice than the standard composition algorithm when combining more than two transducers. The worst-case complexity of our algorithm for composing three transducers T1, T2, and T3 resulting in T, is O(|T|Q min(d(T1)d(T3), d(T2)) + |T|E), where |·|Q denotes the number of states, |·| E the number of transitions, and d(·) the maximum out-degree. As in regular composition, the use of perfect hashing requires a pre-processing step with linear-time expected complexity in the size of the input transducers. In many cases, this approach significantly improves on the complexity of standard composition. Our algorithm also leads to a dramatically faster composition in practice. Furthermore, standard composition can be obtained as a special case of our algorithm. We report the results of several experiments demonstrating this improvement. These theoretical and empirical improvements significantly enhance performance in the applications already mentioned.

AB - Composition of weighted transducers is a fundamental algorithm used in many applications, including for computing complex edit-distances between automata, or string kernels in machine learning, or to combine different components of a speech recognition, speech synthesis, or information extraction system. We present a generalization of the composition of weighted transducers, n-way composition, which is dramatically faster in practice than the standard composition algorithm when combining more than two transducers. The worst-case complexity of our algorithm for composing three transducers T1, T2, and T3 resulting in T, is O(|T|Q min(d(T1)d(T3), d(T2)) + |T|E), where |·|Q denotes the number of states, |·| E the number of transitions, and d(·) the maximum out-degree. As in regular composition, the use of perfect hashing requires a pre-processing step with linear-time expected complexity in the size of the input transducers. In many cases, this approach significantly improves on the complexity of standard composition. Our algorithm also leads to a dramatically faster composition in practice. Furthermore, standard composition can be obtained as a special case of our algorithm. We report the results of several experiments demonstrating this improvement. These theoretical and empirical improvements significantly enhance performance in the applications already mentioned.

UR - http://www.scopus.com/inward/record.url?scp=68649098986&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=68649098986&partnerID=8YFLogxK

U2 - 10.1142/S0129054109006772

DO - 10.1142/S0129054109006772

M3 - Article

VL - 20

SP - 613

EP - 627

JO - International Journal of Foundations of Computer Science

JF - International Journal of Foundations of Computer Science

SN - 0129-0541

IS - 4

ER -