Automatic allocation of training data for rapid prototyping of speech understanding based on multiple model combination

Kazunori Komatani, Masaki Katsumaru, Mikio Nakano, Kotaro Funakoshi, Tetsuya Ogata, Hiroshi G. Okuno

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

2 Citations (Scopus)

Abstract

The optimal choice of speech understanding method depends on the amount of training data available in rapid prototyping. A statistical method is ultimately chosen, but it is not clear at which point in the increase in training data a statistical method become effective. Our framework combines multiple automatic speech recognition (ASR) and language understanding (LU) modules to provide a set of speech understanding results and selects the best result among them. The issue is how to allocate training data to statistical modules and the selection module in order to avoid overfitting in training and obtain better performance. This paper presents an automatic training data allocation method that is based on the change in the coefficients of the logistic regression functions used in the selection module. Experimental evaluation showed that our allocation method outperformed baseline methods that use a single ASR module and a single LU module at every point while training data increase.

Original languageEnglish
Title of host publicationColing 2010 - 23rd International Conference on Computational Linguistics, Proceedings of the Conference
Pages579-587
Number of pages9
Volume2
Publication statusPublished - 2010
Externally publishedYes
Event23rd International Conference on Computational Linguistics, Coling 2010 - Beijing
Duration: 2010 Aug 232010 Aug 27

Other

Other23rd International Conference on Computational Linguistics, Coling 2010
CityBeijing
Period10/8/2310/8/27

Fingerprint

Rapid prototyping
Speech recognition
Statistical methods
Logistics
statistical method
language
Module
Rapid Prototyping
logistics
regression
evaluation
performance

ASJC Scopus subject areas

  • Language and Linguistics
  • Computational Theory and Mathematics
  • Linguistics and Language

Cite this

Komatani, K., Katsumaru, M., Nakano, M., Funakoshi, K., Ogata, T., & Okuno, H. G. (2010). Automatic allocation of training data for rapid prototyping of speech understanding based on multiple model combination. In Coling 2010 - 23rd International Conference on Computational Linguistics, Proceedings of the Conference (Vol. 2, pp. 579-587)

Automatic allocation of training data for rapid prototyping of speech understanding based on multiple model combination. / Komatani, Kazunori; Katsumaru, Masaki; Nakano, Mikio; Funakoshi, Kotaro; Ogata, Tetsuya; Okuno, Hiroshi G.

Coling 2010 - 23rd International Conference on Computational Linguistics, Proceedings of the Conference. Vol. 2 2010. p. 579-587.

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

Komatani, K, Katsumaru, M, Nakano, M, Funakoshi, K, Ogata, T & Okuno, HG 2010, Automatic allocation of training data for rapid prototyping of speech understanding based on multiple model combination. in Coling 2010 - 23rd International Conference on Computational Linguistics, Proceedings of the Conference. vol. 2, pp. 579-587, 23rd International Conference on Computational Linguistics, Coling 2010, Beijing, 10/8/23.
Komatani K, Katsumaru M, Nakano M, Funakoshi K, Ogata T, Okuno HG. Automatic allocation of training data for rapid prototyping of speech understanding based on multiple model combination. In Coling 2010 - 23rd International Conference on Computational Linguistics, Proceedings of the Conference. Vol. 2. 2010. p. 579-587
Komatani, Kazunori ; Katsumaru, Masaki ; Nakano, Mikio ; Funakoshi, Kotaro ; Ogata, Tetsuya ; Okuno, Hiroshi G. / Automatic allocation of training data for rapid prototyping of speech understanding based on multiple model combination. Coling 2010 - 23rd International Conference on Computational Linguistics, Proceedings of the Conference. Vol. 2 2010. pp. 579-587
@inproceedings{cbe1c8f37da84ddfb437a32cff932e20,
title = "Automatic allocation of training data for rapid prototyping of speech understanding based on multiple model combination",
abstract = "The optimal choice of speech understanding method depends on the amount of training data available in rapid prototyping. A statistical method is ultimately chosen, but it is not clear at which point in the increase in training data a statistical method become effective. Our framework combines multiple automatic speech recognition (ASR) and language understanding (LU) modules to provide a set of speech understanding results and selects the best result among them. The issue is how to allocate training data to statistical modules and the selection module in order to avoid overfitting in training and obtain better performance. This paper presents an automatic training data allocation method that is based on the change in the coefficients of the logistic regression functions used in the selection module. Experimental evaluation showed that our allocation method outperformed baseline methods that use a single ASR module and a single LU module at every point while training data increase.",
author = "Kazunori Komatani and Masaki Katsumaru and Mikio Nakano and Kotaro Funakoshi and Tetsuya Ogata and Okuno, {Hiroshi G.}",
year = "2010",
language = "English",
volume = "2",
pages = "579--587",
booktitle = "Coling 2010 - 23rd International Conference on Computational Linguistics, Proceedings of the Conference",

}

TY - GEN

T1 - Automatic allocation of training data for rapid prototyping of speech understanding based on multiple model combination

AU - Komatani, Kazunori

AU - Katsumaru, Masaki

AU - Nakano, Mikio

AU - Funakoshi, Kotaro

AU - Ogata, Tetsuya

AU - Okuno, Hiroshi G.

PY - 2010

Y1 - 2010

N2 - The optimal choice of speech understanding method depends on the amount of training data available in rapid prototyping. A statistical method is ultimately chosen, but it is not clear at which point in the increase in training data a statistical method become effective. Our framework combines multiple automatic speech recognition (ASR) and language understanding (LU) modules to provide a set of speech understanding results and selects the best result among them. The issue is how to allocate training data to statistical modules and the selection module in order to avoid overfitting in training and obtain better performance. This paper presents an automatic training data allocation method that is based on the change in the coefficients of the logistic regression functions used in the selection module. Experimental evaluation showed that our allocation method outperformed baseline methods that use a single ASR module and a single LU module at every point while training data increase.

AB - The optimal choice of speech understanding method depends on the amount of training data available in rapid prototyping. A statistical method is ultimately chosen, but it is not clear at which point in the increase in training data a statistical method become effective. Our framework combines multiple automatic speech recognition (ASR) and language understanding (LU) modules to provide a set of speech understanding results and selects the best result among them. The issue is how to allocate training data to statistical modules and the selection module in order to avoid overfitting in training and obtain better performance. This paper presents an automatic training data allocation method that is based on the change in the coefficients of the logistic regression functions used in the selection module. Experimental evaluation showed that our allocation method outperformed baseline methods that use a single ASR module and a single LU module at every point while training data increase.

UR - http://www.scopus.com/inward/record.url?scp=80053389668&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=80053389668&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:80053389668

VL - 2

SP - 579

EP - 587

BT - Coling 2010 - 23rd International Conference on Computational Linguistics, Proceedings of the Conference

ER -