UM-$p$Aligner: Neural Network Based Parallel Sentence Identification Model

UM > Faculty of Science and Technology > DEPARTMENT OF COMPUTER AND INFORMATION SCIENCE

Status	已發表Published
	UM-$p$Aligner: Neural Network Based Parallel Sentence Identification Model
	Leong, C.; Wong, D. F.; Chao, L. S.
	2018-05-01
Source Publication	11th Workshop on Building and Using Comparable Corpora
Pages	53-58
Abstract	This paper describes the UM-pAligner for the parallel sentence identification shared task of BUCC 2018. The proposed UM-pAligner system consists of two main components, alignment candidate identification and classification models. For the identification model, we propose using an orthogonal denoising autoencoder to transform the embedding features of parallel sentences into shared and private latent spaces, with an objective to better capture the translation correspondences of parallel sentences. In classification, a maximum entropy classifier is employed to determine and select the parallel sentences from the candidate list. On Chinese-English track data, the UM-pAligner achieves a retrieval rate up to 83.65% at the identification phase when n-best is set to 80. The classification model obtains an F1-score of 73.47%, 58.54% and 56.00% respectively on sample, training and test data.
Keyword	parallel sentence classification orthogonal denoising autoencoder neural model maximum entropy
Language	英語English
The Source to Article	PB_Publication
PUB ID	39302
Document Type	Conference paper
Collection	DEPARTMENT OF COMPUTER AND INFORMATION SCIENCE
Corresponding Author	Wong, D. F.
Recommended Citation GB/T 7714	Leong, C.,Wong, D. F.,Chao, L. S.. UM-$p$Aligner: Neural Network Based Parallel Sentence Identification Model[C], 2018, 53-58.
APA	Leong, C.., Wong, D. F.., & Chao, L. S. (2018). UM-$p$Aligner: Neural Network Based Parallel Sentence Identification Model. 11th Workshop on Building and Using Comparable Corpora, 53-58.

Files in This Item:
There are no files associated with this item.

If you have any objections to this item, please fill out the form below and the administrator will contact you as soon as possible.
Content:
Email：	*
Affiliation No.
Verification Code:	Refresh

Any comments and suggestions are welcomed.
Title:	*
Content:
Email：	*
Verification Code:	Refresh