Xiaolin Wang Andrew Finch Masao Utiyama Eiichiro Sumita

Slides:



Advertisements
Similar presentations
Rationale for a multilingual corpus for machine translation evaluation Debbie Elliott Anthony Hartley Eric Atwell Corpus Linguistics 2003, Lancaster, England.
Advertisements

A Human-Centered Computing Framework to Enable Personalized News Video Recommendation (Oh Jun-hyuk)
Statistical Machine Translation Part II – Word Alignments and EM Alex Fraser Institute for Natural Language Processing University of Stuttgart
Measuring the Influence of Long Range Dependencies with Neural Network Language Models Le Hai Son, Alexandre Allauzen, Franc¸ois Yvon Univ. Paris-Sud and.
Confidence Measures for Speech Recognition Reza Sadraei.
Perl Practical Extraction and Report Language Senior Projects II Jeff Wilson.
Neural Nets Using Backpropagation Chris Marriott Ryan Shirley CJ Baker Thomas Tannahill.
Distributional Cues to Word Boundaries: Context Is Important Sharon Goldwater Stanford University Tom Griffiths UC Berkeley Mark Johnson Microsoft Research/
VARIABLE PRESELECTION LIST LENGTH ESTIMATION USING NEURAL NETWORKS IN A TELEPHONE SPEECH HYPOTHESIS-VERIFICATION SYSTEM J. Macías-Guarasa, J. Ferreiros,
Towards Natural Clarification Questions in Dialogue Systems Svetlana Stoyanchev, Alex Liu, and Julia Hirschberg AISB 2014 Convention at Goldsmiths, University.
Automating Translation in the Localisation Factory An Investigation of Post-Editing Effort Sharon O’Brien Dublin City University.
Adaptation Techniques in Automatic Speech Recognition Tor André Myrvoll Telektronikk 99(2), Issue on Spoken Language Technology in Telecommunications,
Large Language Models in Machine Translation Conference on Empirical Methods in Natural Language Processing 2007 報告者:郝柏翰 2013/06/04 Thorsten Brants, Ashok.
Advanced Signal Processing 05/06 Reinisch Bernhard Statistical Machine Translation Phrase Based Model.
Active Learning for Statistical Phrase-based Machine Translation Gholamreza Haffari Joint work with: Maxim Roy, Anoop Sarkar Simon Fraser University NAACL.
Better Punctuation Prediction with Dynamic Conditional Random Fields Wei Lu and Hwee Tou Ng National University of Singapore.
Intelligent Database Systems Lab N.Y.U.S.T. I. M. Chinese Word Segmentation and Statistical Machine Translation Presenter : Wu, Jia-Hao Authors : RUIQIANG.
Yun-Nung (Vivian) Chen, Yu Huang, Sheng-Yi Kong, Lin-Shan Lee National Taiwan University, Taiwan.
1 Sentence-extractive automatic speech summarization and evaluation techniques Makoto Hirohata, Yosuke Shinnaka, Koji Iwano, Sadaoki Furui Presented by.
Date: 2012/3/5 Source: Marcus Fontouraet. al(CIKM’11) Advisor: Jia-ling, Koh Speaker: Jiun Jia, Chiou 1 Efficiently encoding term co-occurrences in inverted.
Copyright  2014 Pearson Education, Inc. or its affiliate(s). All rights reserved. Automatic Assessment of the Speech of Young English Learners Jian Cheng,
1 Boostrapping language models for dialogue systems Karl Weilhammer, Matthew N Stuttle, Steve Young Presenter: Hsuan-Sheng Chiu.
Handing Uncertain Observations in Unsupervised Topic-Mixture Language Model Adaptation Ekapol Chuangsuwanich 1, Shinji Watanabe 2, Takaaki Hori 2, Tomoharu.
1 Sentence Extraction-based Presentation Summarization Techniques and Evaluation Metrics Makoto Hirohata, Yousuke Shinnaka, Koji Iwano and Sadaoki Furui.
Chinese Word Segmentation Adaptation for Statistical Machine Translation Hailong Cao, Masao Utiyama and Eiichiro Sumita Language Translation Group NICT&ATR.
UNSUPERVISED CV LANGUAGE MODEL ADAPTATION BASED ON DIRECT LIKELIHOOD MAXIMIZATION SENTENCE SELECTION Takahiro Shinozaki, Yasuo Horiuchi, Shingo Kuroiwa.
National Taiwan University, Taiwan
1 Broadcast News Segmentation using Metadata and Speech-To-Text Information to Improve Speech Recognition Sebastien Coquoz, Swiss Federal Institute of.
Jun-Won Suh Intelligent Electronic Systems Human and Systems Engineering Department of Electrical and Computer Engineering Speaker Verification System.
Voice Activity Detection based on OptimallyWeighted Combination of Multiple Features Yusuke Kida and Tatsuya Kawahara School of Informatics, Kyoto University,
QoS Supported Clustered Query Processing in Large Collaboration of Heterogeneous Sensor Networks Debraj De and Lifeng Sang Ohio State University Workshop.
MINIMUM WORD CLASSIFICATION ERROR TRAINING OF HMMS FOR AUTOMATIC SPEECH RECOGNITION Yueng-Tien, Lo Speech Lab, CSIE National.
Haitham Elmarakeby.  Speech recognition
1 Minimum Error Rate Training in Statistical Machine Translation Franz Josef Och Information Sciences Institute University of Southern California ACL 2003.
Statistical Machine Translation Part II: Word Alignments and EM Alex Fraser Institute for Natural Language Processing University of Stuttgart
St. Petersburg Institute for Informatics and Automation of the Russian Academy of Sciences Recurrent Neural Network-based Language Modeling for an Automatic.
Phone-Level Pronunciation Scoring and Assessment for Interactive Language Learning Speech Communication, 2000 Authors: S. M. Witt, S. J. Young Presenter:
Evaluating Translation Memory Software Francie Gow MA Translation, University of Ottawa Translator, Translation Bureau, Government of Canada
Mobile Speech Translation Systems Design for /19/2013 INST603 Term Project MIM, UMD Makoto Asami.
Pruning Analysis for the Position Specific Posterior Lattices for Spoken Document Search Jorge Silva University of Southern California Ciprian Chelba and.
1 Minimum Bayes-risk Methods in Automatic Speech Recognition Vaibhava Geol And William Byrne IBM ; Johns Hopkins University 2003 by CRC Press LLC 2005/4/26.
Interpretese vs Translationese
Arnar Thor Jensson Koji Iwano Sadaoki Furui Tokyo Institute of Technology Development of a Speech Recognition System For Icelandic Using Machine Translated.
Is Neural Machine Translation the New State of the Art?
Fabien Cromieres Chenhui Chu Toshiaki Nakazawa Sadao Kurohashi
When deep learning meets object detection: Introduction to two technologies: SSD and YOLO Wenchi Ma.
How to teach translation technologies
Online Multiscale Dynamic Topic Models
Wu et. al., arXiv - sept 2016 Presenter: Lütfi Kerem Şenel
Tools for Natural Language Processing Applications
2 Research Department, iFLYTEK Co. LTD.
Chapter 6. Data Collection in a Wizard-of-Oz Experiment in Reinforcement Learning for Adaptive Dialogue Systems by: Rieser & Lemon. Course: Autonomous.
Artificial Intelligence for Speech Recognition
Adversarial Learning for Neural Dialogue Generation
Attention Is All You Need
For Evaluating Dialog Error Conditions Based on Acoustic Information
HOW TO MAKE MACHINE TRANSLATION WORK FOR YOU
Efficient Estimation of Word Representation in Vector Space
Zhengjun Pan and Hamid Bolouri Department of Computer Science
Eiji Aramaki* Sadao Kurohashi* * University of Tokyo
Natural Language Processing
Statistical Machine Translation Papers from COLING 2004
Voice Activation for Wealth Management
Essentials of Oral Defense
Machine Translation(MT)
Two-Sample Inference Procedures with Means
Two-Sample Inference Procedures with Means
Idiap Research Institute University of Edinburgh
Visual Recognition of American Sign Language Using Hidden Markov Models 문현구 문현구.
Auditory Morphing Weyni Clacken
Presentation transcript:

An Efficient and Effective Online Sentence Segmenter for Simultaneous Interpretation Xiaolin Wang Andrew Finch Masao Utiyama Eiichiro Sumita Advanced Translation Research and Development Promotion Center National Institute of Information and Communication Technology, Japan {xiaolin.wang, andrew.finch, mutiyama, eiichiro.sumita}@nict.go.jp

What is Simultaneous Interpretation? Machine translation normally translates text given by users. Spoken language translation translates text from automatic speech recognition systems. Simultaneous interpretation performs spoken language translation in an online manner. Advantage: Much less delay: Users can hear the translation almost as soon as speakers finish a sentence or sub-sentence. Ideal form of bridging the gap of languages.

Online Sentence Segmenter Online Sentence Segmentation is one of many approaches to simultaneous interpretation Sentence segmenters bridge the gap of ASR and MT by segmenting stream of words into sentences

Our Contribution Most existent segmentation methods Require a long context of future words Computationally expensive We propose a new method Efficient: Computationally simple Less latency : fewer future words Effective: End-to-end performance: optimized against BLEU Scalable for industry Provide simultaneous interpretation service for every one

Method: Framework

Method: Framework(II) Confidence of segmenting Segmentation Strategy

Method: Segmenting Confidence Hypothesis I: no boundary after word wi Hypothesis II: break after word wi

Method: Segmenting Confidence(II) The confidence score is defined as the ratio of the two probabilities

Method: Segmenting Confidence (III) Delay is determined by the order of n-gram language models. Delay can be empirically reduced to one word through an approximation. (Please refer to the experiments)

Method: Segmentation Strategy Threshold-based Segmentation Strategy Latency-based Segmentation Strategy

Method: Segmentation Strategy (II) Threshold-Latency-based strategy A hybrid strategy Aim at lowest latency

Experiments: Settings Translation between Japanese - English Simulate ASR output Remove punctuations Concatenate a random number (1 to 10) of sentences into one utterance Performance measurement Efficiency: average latency per source words Effectiveness: BLEU

Experiments: Results Using different sentence segmenters † SRILM toolkit. ‡The method is not online since it operates on a whole sequence of words

Experiments: Results (II) Proposed methods are much better than the trivial method of fixed-length segmentation, comparable to the offline method (hidden N-gram models) The hybrid method consistently outperforms the two basic ones. BLEU of the threshold-based segmenter compared to that of latency-based segmenter is Lower in English-to-Japanese Almost equal in Japanese-to-English Because Japanese has end-of-sentence indicators such as “ます (MA SU)” and “です (DE SU)”

Experiments: Results (II) Using different numbers of future words One future word is sufficient.

An Example

Conclusions We proposed a practical solution to simultaneous interpretation, consists of a confidence score for boundaries a segmentation strategy The method achieves faster speed better translation quality This is a First Step. Next is Neural network (NN) segmentation model Native NN simultaneous interpretation model … Welcome to the real-world demonstration at December 13th, 11:00 – 12:30, Room 1004-1007 Demo Session 1, COLING 2016  Thank You