公司 標誌 Question Answering System Introduction to Q-A System 資訊四 B91902009 張弘霖 資訊四 B91902066 王惟正.

Slides:



Advertisements
Similar presentations
DOMAIN DEPENDENT QUERY REFORMULATION FOR WEB SEARCH Date : 2013/06/17 Author : Van Dang, Giridhar Kumaran, Adam Troy Source : CIKM’12 Advisor : Dr. Jia-Ling.
Advertisements

Natural Language and Speech Processing Creation of computational models of the understanding and the generation of natural language. Different fields coming.
SMS-Based Web Search for Low-end Mobile Devices Jay Chen New York University Lakshmi Subramanian New York University Eric Brewer University of California.
6/16/20151 Recent Results in Automatic Web Resource Discovery Soumen Chakrabartiv Presentation by Cui Tao.
What can humans do when faced with ASR errors? Dan Bohus Dialogs on Dialogs Group, October 2003.
The Informative Role of WordNet in Open-Domain Question Answering Marius Paşca and Sanda M. Harabagiu (NAACL 2001) Presented by Shauna Eggers CS 620 February.
Feature vs. Model Based Vocal Tract Length Normalization for a Speech Recognition-based Interactive Toy Jacky CHAU Department of Computer Science and Engineering.
ASR Evaluation Julia Hirschberg CS Outline Intrinsic Methods –Transcription Accuracy Word Error Rate Automatic methods, toolkits Limitations –Concept.
1 Information Retrieval and Extraction 資訊檢索與擷取 Chia-Hui Chang, Assistant Professor Dept. of Computer Science & Information Engineering National Central.
Text Mining: Finding Nuggets in Mountains of Textual Data Jochen Dijrre, Peter Gerstl, Roland Seiffert Presented by Huimin Ye.
Text Mining: Finding Nuggets in Mountains of Textual Data Jochen Dijrre, Peter Gerstl, Roland Seiffert Presented by Drew DeHaas.
Overview of Search Engines
SMS-Based web Search for Low- end Mobile Devices Jay Chen New York University Lakshmi Subramanian New York University
Search is not only about the Web An Overview on Printed Documents Search and Patent Search Walid Magdy Centre for Next Generation Localisation School of.
AQUAINT Kickoff Meeting – December 2001 Integrating Robust Semantics, Event Detection, Information Fusion, and Summarization for Multimedia Question Answering.
1 7-Speech Recognition (Cont’d) HMM Calculating Approaches Neural Components Three Basic HMM Problems Viterbi Algorithm State Duration Modeling Training.
A Technical Seminar on Question Answering SHRI RAMDEOBABA COLLEGE OF ENGINEERING & MANAGEMENT Presented By: Rohini Kamdi Guided By: Dr. A.J.Agrawal.
Aardvark Anatomy of a Large-Scale Social Search Engine.
“How much context do you need?” An experiment about context size in Interactive Cross-language Question Answering B. Navarro, L. Moreno-Monteagudo, E.
Evaluation Experiments and Experience from the Perspective of Interactive Information Retrieval Ross Wilkinson Mingfang Wu ICT Centre CSIRO, Australia.
Author: James Allen, Nathanael Chambers, etc. By: Rex, Linger, Xiaoyi Nov. 23, 2009.
Experiments on Building Language Resources for Multi-Modal Dialogue Systems Goals identification of a methodology for adapting linguistic resources for.
Question Answering.  Goal  Automatically answer questions submitted by humans in a natural language form  Approaches  Rely on techniques from diverse.
A Probabilistic Graphical Model for Joint Answer Ranking in Question Answering Jeongwoo Ko, Luo Si, Eric Nyberg (SIGIR ’ 07) Speaker: Cho, Chin Wei Advisor:
A Language Independent Method for Question Classification COLING 2004.
1 Dept of Information and Communication Technology Creating Objects in Flexible Authorization Framework ¹ Dep. of Information and Communication Technology,
21/11/2002 The Integration of Lexical Knowledge and External Resources for QA Hui YANG, Tat-Seng Chua Pris, School of Computing.
1 Just-in-Time Interactive Question Answering Language Computer Corporation Sanda Harabagiu, PI John Lehmann John Williams Paul Aarseth.
LML Speech Recognition Speech Recognition Introduction I E.M. Bakker.
Comparing syntactic semantic patterns and passages in Interactive Cross Language Information Access (iCLEF at the University of Alicante) Borja Navarro,
Improving out of vocabulary name resolution The Hanks David Palmer and Mari Ostendorf Computer Speech and Language 19 (2005) Presented by Aasish Pappu,
Search Engine Architecture
Collocations and Information Management Applications Gregor Erbach Saarland University Saarbrücken.
Searching the web Enormous amount of information –In 1994, 100 thousand pages indexed –In 1997, 100 million pages indexed –In June, 2000, 500 million pages.
AQUAINT Kickoff Meeting Advanced Techniques for Answer Extraction and Formulation Language Computer Corporation Dallas, Texas.
Next Generation Search Engines Ehsun Daroodi 1 Feb, 2003.
DIALOG SYSTEMS FOR AUTOMOTIVE ENVIRONMENTS Presenter: Joseph Picone Inst. for Signal and Info. Processing Dept. Electrical and Computer Eng. Mississippi.
AQUAINT June 2002 Workshop June 2002 Just-in-Time Interactive Question Answering Sanda Harabagiu: PI Language Computer Corporation.
Carnegie Mellon Novelty and Redundancy Detection in Adaptive Filtering Yi Zhang, Jamie Callan, Thomas Minka Carnegie Mellon University {yiz, callan,
A Word Clustering Approach for Language Model-based Sentence Retrieval in Question Answering Systems Saeedeh Momtazi, Dietrich Klakow University of Saarland,Germany.
Automatic Question Answering  Introduction  Factoid Based Question Answering.
A Classification-based Approach to Question Answering in Discussion Boards Liangjie Hong, Brian D. Davison Lehigh University (SIGIR ’ 09) Speaker: Cho,
Intelligent Key Prediction by N-grams and Error-correction Rules Kanokwut Thanadkran, Virach Sornlertlamvanich and Tanapong Potipiti Information Research.
Answer Mining by Combining Extraction Techniques with Abductive Reasoning Sanda Harabagiu, Dan Moldovan, Christine Clark, Mitchell Bowden, Jown Williams.
Copyright © 2013 by Educational Testing Service. All rights reserved. Evaluating Unsupervised Language Model Adaption Methods for Speaking Assessment ShaSha.
Finding document topics for improving topic segmentation Source: ACL2007 Authors: Olivier Ferret (18 route du Panorama, BP6) Reporter:Yong-Xiang Chen.
Unsupervised Relation Detection using Automatic Alignment of Query Patterns extracted from Knowledge Graphs and Query Click Logs Panupong PasupatDilek.
Concepts and Realization of a Diagram Editor Generator Based on Hypergraph Transformation Author: Mark Minas Presenter: Song Gu.
1 ICASSP Paper Survey Presenter: Chen Yi-Ting. 2 Improved Spoken Document Retrieval With Dynamic Key Term Lexicon and Probabilistic Latent Semantic Analysis.
Tracking Hands with Distance Transforms Dave Bargeron Noah Snavely.
NTNU Speech Lab 1 Topic Themes for Multi-Document Summarization Sanda Harabagiu and Finley Lacatusu Language Computer Corporation Presented by Yi-Ting.
1 7-Speech Recognition Speech Recognition Concepts Speech Recognition Approaches Recognition Theories Bayse Rule Simple Language Model P(A|W) Network Types.
A Simple English-to-Punjabi Translation System By : Shailendra Singh.
AQUAINT Mid-Year PI Meeting – June 2002 Integrating Robust Semantics, Event Detection, Information Fusion, and Summarization for Multimedia Question Answering.
Christoph Prinz / Automatic Speech Recognition Research Progress Hits the Road.
Tight Coupling between ASR and MT in Speech-to-Speech Translation Arthur Chan Prepared for Advanced Machine Translation Seminar.
Linguistic knowledge for Speech recognition
Kenneth Baclawski et. al. PSB /11/7 Sa-Im Shin
Information Retrieval and Web Search
Search Engine Architecture
Word AdHoc Network: Using Google Core Distance to extract the most relevant information Presenter : Wei-Hao Huang   Authors : Ping-I Chen, Shi-Jen.
Personalized, Interactive Question Answering on the Web
Information Retrieval and Web Search
Issues in Spoken Dialogue Systems
Generating Natural Answers by Incorporating Copying and Retrieving Mechanisms in Sequence-to-Sequence Learning Shizhu He, Cao liu, Kang Liu and Jun Zhao.
Introduction into Knowledge and information
CSE 635 Multimedia Information Retrieval
Search Engine Architecture
Information Retrieval and Web Design
customer / university logo
Presentation transcript:

公司 標誌 Question Answering System Introduction to Q-A System 資訊四 B 張弘霖 資訊四 B 王惟正

Reference  “Open-Domain Voice-Activated Question Answering” - COLING’02  "Study on spoken interactive open domain question answering" - SSPR’03  “Language models and dialogue strategy for a voice QA system “ - ICA’04

Open –Domain Voice- Activated Question Answering  Date: 2006/5/22  Author: Sanda Harabagiu, Dan Moldvan, Joe Picone.2002  Abbreviation: VAQA

Outline 1. Instruction and Motivation 3. Experiment result and conclusion 2. Four major components VAQA

Instruction and Motivation  Open-Domain Question Answering (ODQA) is popular, especially on Internet domain because of its rich source.  Text-based: yahoo, google  Why voice-activated QA is need?  Mobile device, keyboard bottleneck  Voice input is fast and convenient VAQA

Instruction and Motivation  Basic component for VAQA  Automatic Speech Recognition(ASR)  Q&A system  Simple path: ASR  Q&A (not good, latter)  Our solution: ASR  Q&A VAQA

Instruction and Motivation VAQA Global view for Voice- Activated Question Answering System On-line Documents

Instruction and Motivation  Filtering: ill-formed questions from the word lattice.  Alternation of keywords.  Interactive Q&A module.  Enhanced language model. VAQA

Instruction and Motivation  Simple path (ASR  Q&A)  TREC8 and 9  ISSP with 30% WER  76% ↓ 7%  Iterative refinement! Interaction between ASR and Q&A make better performance than individual components. VAQA

Four components  Alternation (Harabagiu et al. 2001)  Three keyword variants.  Morphological invent, invention, inventor  Semantic murderer and killer  Lexical paraphrase Like better and prefer VAQA

Four major components  Filtering Goal: significantly reduce the large number of outputs produced by the word lattice search module. VAQA

Four major components 1.Syntactic filter: “The was President Cleveland wife” “When President Cleveland life” 2.Semantic filer: “It was President Cleveland lawyer” for “Who is President Cleveland wife”; “Who is President Cleveland’s life” VAQA

Four major components 3.Pragmatic filter: ”How far is Yaroslavl from Moscow?” Even if the city name is not recognized, one of question pattern set will identify the first concept after the question stem is “location.” VAQA

Four major components VAQA Architecture of filtering component in VAQA

Four major components VAQA Global view for Voice- Activated Question Answering System On-line Documents

Four major components  Enhanced Language Model  Language Model  A mechanism to estimate the probability of some word w in q word sequence W given the surrounding words. [prob.”I am” > prob.”I aim” ]  Linguistic, domain, pragmatic knowledge.  N-gram model for most ASR, local dependencies between words. VAQA

Four major components  Enhanced Language Model  N-gram is insufficient for the recognition of spoken question words.  “How far is Yaroslavl from Moscow”  “Affair is yes level from Moscow”  Probability for Affair and Moscow VAQA

Four major components  Enhanced Language Model  Semantic transformation of questions (Harabagiu et al. 2000)  Graphs in witch the edges are binary dependencies and question stems are replaced by semantic classes e.g. PERSON, DISTANCE. VAQA How far is Yaroslavl from Moscow

Four major components  Semantic Transformation of Questions  “Affair is yes level from Moscow”  “from Moscow” + ST + Stem  correction of “affair” to “how far”. VAQA

Four major components  Enhanced Language Model  Recognized Question  Semantic transformation S  The Question Q  A set of binary dependency  The base NPs recognized by the parser  Question semantic template information VAQA

Four major components  Question Template  Semantic Class v.s. Expected Answer Type VAQA

Four major component  Recognized Question  Semantic transformation S  The question Q VAQA

Four major components VAQA Global view for Voice- Activated Question Answering System On-line Documents

Four major components  Interactive Question Answering  Due to the errors from ASR.  Clarification from user by asking question.  Steps:  Finding the conflicts in the question recognized.  Deciding what the question is about.  Re-solve the question with the feedback.  Rank the keywords for answering. VAQA

Four major components  Interactive Question Answering  Example:  “Where” identifies expected answer type as LOCATION.  “leader” is a member the PERSON subhierarchy.  “leader” is the focus of the question. VAQA

Four Major components  Interactive Question Answering  “No” indicates the system did not comprehend the topic of the question.  “musical” and “summer” are new keywords. VAQA

Four major components  Interactive Question Answering  “Where”  at least one location  “musical”  “summer”, dropped because the number of paragraphs is too small. VAQA

Four major components VAQA Global view for Voice- Activated Question Answering System On-line Documents

Experiment result and conclusion  Experiment result  RAR (reciprocal value of the rank)  RAR = 1 / rank  MRAR = 1/n ΣRAR VAQA

Experiment result and conclusion  Experiment result  Word Error Rate (WER) for ASR VAQA

Experiment Result and conclusion  Conclusion  Performance of VAQA depends mostly on ELM and correction in IQA module.  To train the ELM, filtering component is essential. why?  Experiment result reveals VAQA here both improves:  Accuracy of spoken Q&A  Better WER of ASR VAQA

Global view for Voice- Activated Question Answering System On-line Documents

ENDEND End