SlideShare a Scribd company logo
1 of 26
Dialog System Review
Tran Trung Kien
Saltlux Development Center - VDC
Communicating Knowledge
Vietnam Development Center
 What is Dialogue System?
 Definition
 3 generations of Dialog System
 Evaluation
 Spoken Dialogue System
 Architecture
 Components
 Some approaches
 A Neural Conversation Model
 Deep Reinforcement Learning for Dialogue Generation
 Common Frameworks and Data sets
 Discussion
Contents
Communicating Knowledge
Vietnam Development Center
 Definition:
 DS is a computer program developed to converse with human, with a coherent structure.
 DS can use text, speech, graphics, haptics, gestures and other modes for
communication on both the input and output.
 Nowadays, speech is most commonly used for the input and output => Spoken Dialogue
System.
 3 Generations of DS
 G1: Symbolic Rule/Template Based QA
 Focus on grammatical rule & ontological design by human experts (early AI approach)
 Easy interpretation, debugging, and system update
 Popular before late 90’s
 Still in use in commercial systems and by bots startups
 Limitations:
heavily reliance on experts
hard to scale over domains
data used only to help design rules, not for learning
What is Dialogue System - 1/3
Communicating Knowledge
Vietnam Development Center
 G2: Data Driven, Learning
 Data used not to design rules for NLU and action, but to learn statistical parameters in
dialogue systems
Reduce cost of hand-crafting complex dialogue manager
Robustness against speech recognize errors in noisy environment
 MDP(Markov Decision Process)/POMDP (Partially Observed MDP) & RL for dialogue
policy
 Discriminative (CRF) & generative (HMM) methods for NLU
 Popular in academic research until 2014 (before deep learning arrived at the dialogue
world); in parallel with G1 (BBN, AT&T, CMU, SRI, CU ...)
 Limitations:
Not easy to interpret, debug, and update systems
Still hard to scale over domains
Models & representations not powerful enough; no end-2-end, hard to scale up
Remained academic until deep learning arrived
What is Dialogue System – 2/3
Communicating Knowledge
Vietnam Development Center
 G3: Data-Driven Deep Learning:
 Like G2, data used to learn everything in dialogue systems
Reduce cost of hand-crafting complex dialogue manager
Robustness against speech recognize errors in noisy environment & against NLU
errors
MDP/POMDP & reinforcement learning for dialogue policy (same)
 Neural models & representations are much more powerful
 End-to-End learning becomes feasible
 Attracted huge research efforts since 2015 (after deep learning’s success in
vision/speech and in deep RL shown success in Atari games)
 Limitations:
Still not easy to interpret, debug, and update systems
Lack interface btw cont. neural learning and symbolic NL structure to human users
Lack active research in scaling over domains via deep transfer learning & RL
 No clear commercial success reported yet
 Evaluation:?
 Still argueing, no evaluation method is set as standard.
 BLEU is usually used.
 Some researchers define their own evaluation metrics to measure quality.
What is Dialogue System 3/3
Communicating Knowledge
Vietnam Development Center
 Use speech as input and output
 Architecture:
Spoken Dialog System
Communicating Knowledge
Vietnam Development Center
 Automatic Speech Recognition (ASR):
 Convert from voice signal to Words and Manage uncertainty.
 Challenges:
 Environment noises
Speech production: low fluency, false starts, filled pauses, repeats, corrections,
accent, age, gender, differences between human-human and human-machine
speech
Technological familiarity of user
 Spoken Language Understanding (SLU)”
 Spoken Language Understanding is the task of extracting meaning from utterances
 Convert from words to concepts
Dialog acts (the overall intent of an utterance)
Domain specific concepts
Syntactic/Semantic parser
 Very difficult under noisy conditions
 Challenges:
 Recognizer error, background noise resulting in indels (insertions / substitutions /
deletions), word boundary detection problems
 Language production phenomena: low fluency, false starts, corrections, repairs are difficult
to parse
 Meaning must often be assembled from multiple speaker turns
 There are many, many possible ways to say the same thing.
Spoken Dialog System - Components
Communicating Knowledge
Vietnam Development Center
 Dialogue Management:
 Map concepts to action.
 Manage dialog history, states and general flow of the conversation
 Language Generation:
 Generate response for the input.
 Text To Speech Synthesis:
 Convert the generated response to speech and present to user.
Spoken Dialog System - Components
Communicating Knowledge
Vietnam Development Center
 Previous approaches are often restricted to specific domains (e.g., booking an airline
ticket) and require hand-crafted rules.
 Proposed a model based on their “Sequence to sequence learning with neural networks”
(NIPS, 2014).
 Can be trained end-to-end and thus requires much fewer hand-crafted rules.
 Allows researchers to work on tasks for which domain knowledge may not be readily
available, or for tasks which are simply too hard to design rules manually.
 The model:
A Neural Conversation Model – Oriol Vinyals, Quoc V.Le – Google
Using the seq2seq framework for modeling conversations
Communicating Knowledge
Vietnam Development Center
 Data sets:
 IT Helpdesk Troubleshooting:
Typical interaction word length: 400
Turn talking is clearly signaled
30M tokens (3M used as validation)
 OpenSubtitles (Tiedemann, 2009):
Noisy data set
Movie conversation in XML format
After preprocessed:
– Train set: 62M sentences, 923M tokens
– Validation set: 26M sentences, 295M tokens
A Neural Conversation Model – Oriol Vinyals, Quoc V.Le – Google
Communicating Knowledge
Vietnam Development Center
 Experiments:
 IT Helpdesk:
Trained single layer LSTM with 1024 memory cells using stochastic gradient descent
with gradient clipping.
Vocabulary: 20K words
Conversation 1: VPN issues
A Neural Conversation Model – Oriol Vinyals, Quoc V.Le – Google
Communicating Knowledge
Vietnam Development Center
 Experiments:
 OpenSubtitles:
Train 2-layered LSTM, 4096 memory cells for each layer.
Vocabulary: 100k most frequently words.
A Neural Conversation Model – Oriol Vinyals, Quoc V.Le – Google
Communicating Knowledge
Vietnam Development Center
Communicating Knowledge
Vietnam Development Center
Communicating Knowledge
Vietnam Development Center
 Conclusion:
 A simple language model based on the seq2seq framework can be used to train a
conversational engine .
 It can generate simple and basic conversations, and extract knowledge from a noisy but
open-domain dataset.
 Purely data driven without any rules, but can generate quite proper answers.
 A big limitation: lack of a coherent personality.
A Neural Conversation Model – Oriol Vinyals, Quoc V.Le – Google
Communicating Knowledge
Vietnam Development Center
 Authors: J. Li, W. Monroe, A. Ritter, M. Galley, J. Gao, D. Jurafsky
 Despite the success of SEQ2SEQ models in dialogue generation,
two problems emerge:
 How to keep the conversation longer?
Seq2seq models tend to generate generic responses like “I don’t know” regardless
the input. => Responses like this will close the conversion.
The cause is seq2seq use MLE objective function. But the frequency if those generic
responses is very high in training set.
 System becomes stuck in infinite loop of repetitive responses. This is due to MLE-base
seq2seq models’ inability to account for repetition.
Deep RL for Dialogue Generation
Communicating Knowledge
Vietnam Development Center
 => we need a conversation framework that has the ability to:
 (1) integrate developer-defined rewards that better mimic the true goal of chatbot
development.
 (2) model the long- term influence of a generated response in an ongoing dialogue.
 Proposed a neural RL generation method:
 can optimize long-term rewards designed by system developers.
 uses the encoder- decoder architecture as its backbone
 simulates conversation between two virtual agents to explore the space of possible
actions while learning to maximize expected reward.
 We define simple heuristic approximations to rewards that characterize good
conversations: good conversations are forward-looking or interactive (a turn suggests a
following turn), informative, and coherent.
 Use policy gradient method instead of MLE objective function.
 Authors goal is to integrate Seq2seq and RL to get advantages of
both.
Deep RL for Dialogue Generation
Communicating Knowledge
Vietnam Development Center
 Reward: r
 Ease of answering: generated answer should be easy to respond.
S: set of 8 manually collected dull response (“I don’t know”, …)
NS: size of S, s: a sequence in S, Ns: # of token in s.
Pseq2seq: the likelihood calculated by Seq2seq model.
 Information flow: agent should contribute new information to keep dialogue moving =>
penalizing semantic similarity between 2 consecutive turns of agent:
hpi, hpi+1 resulted from encoder for pi, pi+1
Deep RL for Dialogue Generation
Communicating Knowledge
Vietnam Development Center
 Reward: r
 Semantic Coherence: to avoid high reward but not grammatical and coherent
Pseq2seq(a|pi, qi): probability of generating a given the previous utterances [pi, qi]
2nd part: backward probability of generating the previous dialogue utterance
qi based on response a
 Final reward r:
lamda1 + lamda2 + lamda3 = 1, lamda1 = lamda2 = 0.25, lamda3 = 0.5
Deep RL for Dialogue Generation
Communicating Knowledge
Vietnam Development Center
 Simulation:
Deep RL for Dialogue Generation
Communicating Knowledge
Vietnam Development Center
 Experiment results:
 Sub set of 10M messages from OpenSubtitles dataset and extract 0.8M message with
lowest likelihood of generating dull response to ensure the initial input is easy to respond
Deep RL for Dialogue Generation
Communicating Knowledge
Vietnam Development Center
 Experiment results:
Communicating Knowledge
Vietnam Development Center
Communicating Knowledge
Vietnam Development Center
 TensorFlow:
 Open source software library for numerical computation using data flow graphs
 IrisTK:
 Java-based framework for developing spoken dialogue systems.
 Url: http://www.iristk.net/
 OpenDial:
 Java-based, domain-independent toolkit for developing spoken dialogue systems
 Url: http://www.opendial-toolkit.net/
 CSLU Toolkit:
 a comprehensive suite of tools to enable exploration, learning, and research into speech and human-computer
interaction.
 http://www.cslu.ogi.edu/toolkit/
 NADIA: (developed by MARKUS M. BERG)
 set of Java-based components that deals with the creation of spoken dialogue systems.
 Detail information (Phd Thesis, paper: http://mmberg.net/nadia/
 Reference source code (include data model): https://github.com/mmberg
 Datasets:
 https://github.com/karthikncode/nlp-datasets
 Unbutu Dialogue Corpus
Frameworks and Datasets for SDS
Communicating Knowledge
Vietnam Development Center
 Three generations of SDS – Li Deng, Chief Scientist of AI, MS AI
 The Unreasonable Effectiveness of Recurrent Neural Networks
 A neural conversation model – Oriol Vinyals, Quoc V.Le – Google - 2015
 Deep reinforcement learning for Dialogue Generation – Jiwei Li, Will Monroe,
Dan Jurafsky (Standford Univ.), Alan Ritter (Ohio State Univ.), Michel Galley,
Jianfeng Gao (MS Research) - 2016
 Neural responding machine for short-text conversation – Lifeng Shang,
Zhengdong Lu, Hang Li – Huawei Tech., 2015
 Deep RL: An overview – Yuxi Li - 2017
 Dialogue System – Wikipedia: https://en.wikipedia.org/wiki/Dialog_system
 Speech Recognition: https://en.wikipedia.org/wiki/Speech_recognition
 Neural Network Dialog System Papers:
https://github.com/snakeztc/NeuralDialogPapers
 Datasets for Natural Language Processing:
https://github.com/karthikncode/nlp-datasets
References
Communicating Knowledge
Vietnam Development Center
THANK YOU!

More Related Content

What's hot

End-to-End Joint Learning of Natural Language Understanding and Dialogue Manager
End-to-End Joint Learning of Natural Language Understanding and Dialogue ManagerEnd-to-End Joint Learning of Natural Language Understanding and Dialogue Manager
End-to-End Joint Learning of Natural Language Understanding and Dialogue ManagerYun-Nung (Vivian) Chen
 
Deep Learning for Dialogue Modeling - NTHU
Deep Learning for Dialogue Modeling - NTHUDeep Learning for Dialogue Modeling - NTHU
Deep Learning for Dialogue Modeling - NTHUYun-Nung (Vivian) Chen
 
How the Context Matters Language and Interaction in Dialogues
How the Context Matters Language and Interaction in DialoguesHow the Context Matters Language and Interaction in Dialogues
How the Context Matters Language and Interaction in DialoguesYun-Nung (Vivian) Chen
 
Deep Learning, an interactive introduction for NLP-ers
Deep Learning, an interactive introduction for NLP-ersDeep Learning, an interactive introduction for NLP-ers
Deep Learning, an interactive introduction for NLP-ersRoelof Pieters
 
Practical Deep Learning for NLP
Practical Deep Learning for NLP Practical Deep Learning for NLP
Practical Deep Learning for NLP Textkernel
 
Anthiil Inside workshop on NLP
Anthiil Inside workshop on NLPAnthiil Inside workshop on NLP
Anthiil Inside workshop on NLPSatyam Saxena
 
Engineering Intelligent NLP Applications Using Deep Learning – Part 2
Engineering Intelligent NLP Applications Using Deep Learning – Part 2 Engineering Intelligent NLP Applications Using Deep Learning – Part 2
Engineering Intelligent NLP Applications Using Deep Learning – Part 2 Saurabh Kaushik
 
Leveraging Behavioral Patterns of Mobile Applications for Personalized Spoken...
Leveraging Behavioral Patterns of Mobile Applications for Personalized Spoken...Leveraging Behavioral Patterns of Mobile Applications for Personalized Spoken...
Leveraging Behavioral Patterns of Mobile Applications for Personalized Spoken...Yun-Nung (Vivian) Chen
 
Language Empowering Intelligent Assistants (CHT)
Language Empowering Intelligent Assistants (CHT)Language Empowering Intelligent Assistants (CHT)
Language Empowering Intelligent Assistants (CHT)Yun-Nung (Vivian) Chen
 
Recent Advances in NLP
  Recent Advances in NLP  Recent Advances in NLP
Recent Advances in NLPAnuj Gupta
 
An Intelligent Assistant for High-Level Task Understanding
An Intelligent Assistant for High-Level Task UnderstandingAn Intelligent Assistant for High-Level Task Understanding
An Intelligent Assistant for High-Level Task UnderstandingYun-Nung (Vivian) Chen
 
Deep Learning for NLP Applications
Deep Learning for NLP ApplicationsDeep Learning for NLP Applications
Deep Learning for NLP ApplicationsSamiur Rahman
 
Talk from NVidia Developer Connect
Talk from NVidia Developer ConnectTalk from NVidia Developer Connect
Talk from NVidia Developer ConnectAnuj Gupta
 
Deep Learning for NLP (without Magic) - Richard Socher and Christopher Manning
Deep Learning for NLP (without Magic) - Richard Socher and Christopher ManningDeep Learning for NLP (without Magic) - Richard Socher and Christopher Manning
Deep Learning for NLP (without Magic) - Richard Socher and Christopher ManningBigDataCloud
 
Matrix Factorization with Knowledge Graph Propagation for Unsupervised Spoken...
Matrix Factorization with Knowledge Graph Propagation for Unsupervised Spoken...Matrix Factorization with Knowledge Graph Propagation for Unsupervised Spoken...
Matrix Factorization with Knowledge Graph Propagation for Unsupervised Spoken...Yun-Nung (Vivian) Chen
 
Visual-Semantic Embeddings: some thoughts on Language
Visual-Semantic Embeddings: some thoughts on LanguageVisual-Semantic Embeddings: some thoughts on Language
Visual-Semantic Embeddings: some thoughts on LanguageRoelof Pieters
 
Deep Learning Architectures for NLP (Hungarian NLP Meetup 2016-09-07)
Deep Learning Architectures for NLP (Hungarian NLP Meetup 2016-09-07)Deep Learning Architectures for NLP (Hungarian NLP Meetup 2016-09-07)
Deep Learning Architectures for NLP (Hungarian NLP Meetup 2016-09-07)Márton Miháltz
 
Detecting Actionable Items in Meetings by Convolutional Deep Structured Seman...
Detecting Actionable Items in Meetings by Convolutional Deep Structured Seman...Detecting Actionable Items in Meetings by Convolutional Deep Structured Seman...
Detecting Actionable Items in Meetings by Convolutional Deep Structured Seman...Yun-Nung (Vivian) Chen
 

What's hot (20)

End-to-End Joint Learning of Natural Language Understanding and Dialogue Manager
End-to-End Joint Learning of Natural Language Understanding and Dialogue ManagerEnd-to-End Joint Learning of Natural Language Understanding and Dialogue Manager
End-to-End Joint Learning of Natural Language Understanding and Dialogue Manager
 
Deep Learning for Dialogue Modeling - NTHU
Deep Learning for Dialogue Modeling - NTHUDeep Learning for Dialogue Modeling - NTHU
Deep Learning for Dialogue Modeling - NTHU
 
How the Context Matters Language and Interaction in Dialogues
How the Context Matters Language and Interaction in DialoguesHow the Context Matters Language and Interaction in Dialogues
How the Context Matters Language and Interaction in Dialogues
 
Deep Learning, an interactive introduction for NLP-ers
Deep Learning, an interactive introduction for NLP-ersDeep Learning, an interactive introduction for NLP-ers
Deep Learning, an interactive introduction for NLP-ers
 
Practical Deep Learning for NLP
Practical Deep Learning for NLP Practical Deep Learning for NLP
Practical Deep Learning for NLP
 
Anthiil Inside workshop on NLP
Anthiil Inside workshop on NLPAnthiil Inside workshop on NLP
Anthiil Inside workshop on NLP
 
NLP Bootcamp
NLP BootcampNLP Bootcamp
NLP Bootcamp
 
Engineering Intelligent NLP Applications Using Deep Learning – Part 2
Engineering Intelligent NLP Applications Using Deep Learning – Part 2 Engineering Intelligent NLP Applications Using Deep Learning – Part 2
Engineering Intelligent NLP Applications Using Deep Learning – Part 2
 
Leveraging Behavioral Patterns of Mobile Applications for Personalized Spoken...
Leveraging Behavioral Patterns of Mobile Applications for Personalized Spoken...Leveraging Behavioral Patterns of Mobile Applications for Personalized Spoken...
Leveraging Behavioral Patterns of Mobile Applications for Personalized Spoken...
 
Language Empowering Intelligent Assistants (CHT)
Language Empowering Intelligent Assistants (CHT)Language Empowering Intelligent Assistants (CHT)
Language Empowering Intelligent Assistants (CHT)
 
Recent Advances in NLP
  Recent Advances in NLP  Recent Advances in NLP
Recent Advances in NLP
 
Deeplearning NLP
Deeplearning NLPDeeplearning NLP
Deeplearning NLP
 
An Intelligent Assistant for High-Level Task Understanding
An Intelligent Assistant for High-Level Task UnderstandingAn Intelligent Assistant for High-Level Task Understanding
An Intelligent Assistant for High-Level Task Understanding
 
Deep Learning for NLP Applications
Deep Learning for NLP ApplicationsDeep Learning for NLP Applications
Deep Learning for NLP Applications
 
Talk from NVidia Developer Connect
Talk from NVidia Developer ConnectTalk from NVidia Developer Connect
Talk from NVidia Developer Connect
 
Deep Learning for NLP (without Magic) - Richard Socher and Christopher Manning
Deep Learning for NLP (without Magic) - Richard Socher and Christopher ManningDeep Learning for NLP (without Magic) - Richard Socher and Christopher Manning
Deep Learning for NLP (without Magic) - Richard Socher and Christopher Manning
 
Matrix Factorization with Knowledge Graph Propagation for Unsupervised Spoken...
Matrix Factorization with Knowledge Graph Propagation for Unsupervised Spoken...Matrix Factorization with Knowledge Graph Propagation for Unsupervised Spoken...
Matrix Factorization with Knowledge Graph Propagation for Unsupervised Spoken...
 
Visual-Semantic Embeddings: some thoughts on Language
Visual-Semantic Embeddings: some thoughts on LanguageVisual-Semantic Embeddings: some thoughts on Language
Visual-Semantic Embeddings: some thoughts on Language
 
Deep Learning Architectures for NLP (Hungarian NLP Meetup 2016-09-07)
Deep Learning Architectures for NLP (Hungarian NLP Meetup 2016-09-07)Deep Learning Architectures for NLP (Hungarian NLP Meetup 2016-09-07)
Deep Learning Architectures for NLP (Hungarian NLP Meetup 2016-09-07)
 
Detecting Actionable Items in Meetings by Convolutional Deep Structured Seman...
Detecting Actionable Items in Meetings by Convolutional Deep Structured Seman...Detecting Actionable Items in Meetings by Convolutional Deep Structured Seman...
Detecting Actionable Items in Meetings by Convolutional Deep Structured Seman...
 

Similar to Deep Dialog System Review

EXTENDING OUTPUT ATTENTIONS IN RECURRENTNEURAL NETWORKS FOR DIALOG GENERATION
EXTENDING OUTPUT ATTENTIONS IN RECURRENTNEURAL NETWORKS FOR DIALOG GENERATIONEXTENDING OUTPUT ATTENTIONS IN RECURRENTNEURAL NETWORKS FOR DIALOG GENERATION
EXTENDING OUTPUT ATTENTIONS IN RECURRENTNEURAL NETWORKS FOR DIALOG GENERATIONgerogepatton
 
IRJET - Mobile Chatbot for Information Search
 IRJET - Mobile Chatbot for Information Search IRJET - Mobile Chatbot for Information Search
IRJET - Mobile Chatbot for Information SearchIRJET Journal
 
Understanding Human Conversations with AI
Understanding Human Conversations with AI Understanding Human Conversations with AI
Understanding Human Conversations with AI Rajath D M
 
Tomáš Mikolov - Distributed Representations for NLP
Tomáš Mikolov - Distributed Representations for NLPTomáš Mikolov - Distributed Representations for NLP
Tomáš Mikolov - Distributed Representations for NLPMachine Learning Prague
 
A black-box-approach-for-response-quality-evaluation-of-conversational-agent-...
A black-box-approach-for-response-quality-evaluation-of-conversational-agent-...A black-box-approach-for-response-quality-evaluation-of-conversational-agent-...
A black-box-approach-for-response-quality-evaluation-of-conversational-agent-...Cemal Ardil
 
Improving Dialogue Management Through Data Optimization
Improving Dialogue Management Through Data OptimizationImproving Dialogue Management Through Data Optimization
Improving Dialogue Management Through Data Optimizationkevig
 
IMPROVING DIALOGUE MANAGEMENT THROUGH DATA OPTIMIZATION
IMPROVING DIALOGUE MANAGEMENT THROUGH DATA OPTIMIZATIONIMPROVING DIALOGUE MANAGEMENT THROUGH DATA OPTIMIZATION
IMPROVING DIALOGUE MANAGEMENT THROUGH DATA OPTIMIZATIONkevig
 
Bridging the gap between AI and UI - DSI Vienna - full version
Bridging the gap between AI and UI - DSI Vienna - full versionBridging the gap between AI and UI - DSI Vienna - full version
Bridging the gap between AI and UI - DSI Vienna - full versionLiad Magen
 
A Research Paper on HUMAN MACHINE CONVERSATION USING CHATBOT
A Research Paper on HUMAN MACHINE CONVERSATION USING CHATBOTA Research Paper on HUMAN MACHINE CONVERSATION USING CHATBOT
A Research Paper on HUMAN MACHINE CONVERSATION USING CHATBOTIRJET Journal
 
Natural Language Processing .pdf
Natural Language Processing .pdfNatural Language Processing .pdf
Natural Language Processing .pdfAnime196637
 
A comprehensive guide to prompt engineering.pdf
A comprehensive guide to prompt engineering.pdfA comprehensive guide to prompt engineering.pdf
A comprehensive guide to prompt engineering.pdfStephenAmell4
 
IRJET- Semantic Question Matching
IRJET- Semantic Question MatchingIRJET- Semantic Question Matching
IRJET- Semantic Question MatchingIRJET Journal
 
#1 Berlin Students in AI, Machine Learning & NLP presentation
#1 Berlin Students in AI, Machine Learning & NLP presentation#1 Berlin Students in AI, Machine Learning & NLP presentation
#1 Berlin Students in AI, Machine Learning & NLP presentationparlamind
 
Stefan Geissler kairntech - SDC Nice Apr 2019
Stefan Geissler kairntech - SDC Nice Apr 2019 Stefan Geissler kairntech - SDC Nice Apr 2019
Stefan Geissler kairntech - SDC Nice Apr 2019 Stefan Geißler
 
IRJET- Querying Database using Natural Language Interface
IRJET-  	  Querying Database using Natural Language InterfaceIRJET-  	  Querying Database using Natural Language Interface
IRJET- Querying Database using Natural Language InterfaceIRJET Journal
 

Similar to Deep Dialog System Review (20)

EXTENDING OUTPUT ATTENTIONS IN RECURRENTNEURAL NETWORKS FOR DIALOG GENERATION
EXTENDING OUTPUT ATTENTIONS IN RECURRENTNEURAL NETWORKS FOR DIALOG GENERATIONEXTENDING OUTPUT ATTENTIONS IN RECURRENTNEURAL NETWORKS FOR DIALOG GENERATION
EXTENDING OUTPUT ATTENTIONS IN RECURRENTNEURAL NETWORKS FOR DIALOG GENERATION
 
ijeter35852020.pdf
ijeter35852020.pdfijeter35852020.pdf
ijeter35852020.pdf
 
IRJET - Mobile Chatbot for Information Search
 IRJET - Mobile Chatbot for Information Search IRJET - Mobile Chatbot for Information Search
IRJET - Mobile Chatbot for Information Search
 
Understanding Human Conversations with AI
Understanding Human Conversations with AI Understanding Human Conversations with AI
Understanding Human Conversations with AI
 
ms_3.pdf
ms_3.pdfms_3.pdf
ms_3.pdf
 
Gobinath.T Resume - Copy
Gobinath.T Resume - CopyGobinath.T Resume - Copy
Gobinath.T Resume - Copy
 
Tomáš Mikolov - Distributed Representations for NLP
Tomáš Mikolov - Distributed Representations for NLPTomáš Mikolov - Distributed Representations for NLP
Tomáš Mikolov - Distributed Representations for NLP
 
A black-box-approach-for-response-quality-evaluation-of-conversational-agent-...
A black-box-approach-for-response-quality-evaluation-of-conversational-agent-...A black-box-approach-for-response-quality-evaluation-of-conversational-agent-...
A black-box-approach-for-response-quality-evaluation-of-conversational-agent-...
 
Improving Dialogue Management Through Data Optimization
Improving Dialogue Management Through Data OptimizationImproving Dialogue Management Through Data Optimization
Improving Dialogue Management Through Data Optimization
 
IMPROVING DIALOGUE MANAGEMENT THROUGH DATA OPTIMIZATION
IMPROVING DIALOGUE MANAGEMENT THROUGH DATA OPTIMIZATIONIMPROVING DIALOGUE MANAGEMENT THROUGH DATA OPTIMIZATION
IMPROVING DIALOGUE MANAGEMENT THROUGH DATA OPTIMIZATION
 
Bridging the gap between AI and UI - DSI Vienna - full version
Bridging the gap between AI and UI - DSI Vienna - full versionBridging the gap between AI and UI - DSI Vienna - full version
Bridging the gap between AI and UI - DSI Vienna - full version
 
A Research Paper on HUMAN MACHINE CONVERSATION USING CHATBOT
A Research Paper on HUMAN MACHINE CONVERSATION USING CHATBOTA Research Paper on HUMAN MACHINE CONVERSATION USING CHATBOT
A Research Paper on HUMAN MACHINE CONVERSATION USING CHATBOT
 
Natural Language Processing .pdf
Natural Language Processing .pdfNatural Language Processing .pdf
Natural Language Processing .pdf
 
A comprehensive guide to prompt engineering.pdf
A comprehensive guide to prompt engineering.pdfA comprehensive guide to prompt engineering.pdf
A comprehensive guide to prompt engineering.pdf
 
IRJET- Semantic Question Matching
IRJET- Semantic Question MatchingIRJET- Semantic Question Matching
IRJET- Semantic Question Matching
 
#1 Berlin Students in AI, Machine Learning & NLP presentation
#1 Berlin Students in AI, Machine Learning & NLP presentation#1 Berlin Students in AI, Machine Learning & NLP presentation
#1 Berlin Students in AI, Machine Learning & NLP presentation
 
Stefan Geissler kairntech - SDC Nice Apr 2019
Stefan Geissler kairntech - SDC Nice Apr 2019 Stefan Geissler kairntech - SDC Nice Apr 2019
Stefan Geissler kairntech - SDC Nice Apr 2019
 
IRJET- Querying Database using Natural Language Interface
IRJET-  	  Querying Database using Natural Language InterfaceIRJET-  	  Querying Database using Natural Language Interface
IRJET- Querying Database using Natural Language Interface
 
CHI2021
CHI2021CHI2021
CHI2021
 
Question answering
Question answeringQuestion answering
Question answering
 

More from Nguyen Quang

Deep Reinforcement Learning
Deep Reinforcement LearningDeep Reinforcement Learning
Deep Reinforcement LearningNguyen Quang
 
Sequence to Sequence Learning with Neural Networks
Sequence to Sequence Learning with Neural NetworksSequence to Sequence Learning with Neural Networks
Sequence to Sequence Learning with Neural NetworksNguyen Quang
 
Introduction to cassandra
Introduction to cassandraIntroduction to cassandra
Introduction to cassandraNguyen Quang
 
Web browser architecture
Web browser architectureWeb browser architecture
Web browser architectureNguyen Quang
 
X Query for beginner
X Query for beginnerX Query for beginner
X Query for beginnerNguyen Quang
 
Redistributable introtoscrum
Redistributable introtoscrumRedistributable introtoscrum
Redistributable introtoscrumNguyen Quang
 
Text categorization
Text categorizationText categorization
Text categorizationNguyen Quang
 
A holistic lexicon based approach to opinion mining
A holistic lexicon based approach to opinion miningA holistic lexicon based approach to opinion mining
A holistic lexicon based approach to opinion miningNguyen Quang
 

More from Nguyen Quang (13)

Apache Zookeeper
Apache ZookeeperApache Zookeeper
Apache Zookeeper
 
Apache Storm
Apache StormApache Storm
Apache Storm
 
Deep Reinforcement Learning
Deep Reinforcement LearningDeep Reinforcement Learning
Deep Reinforcement Learning
 
Sequence to Sequence Learning with Neural Networks
Sequence to Sequence Learning with Neural NetworksSequence to Sequence Learning with Neural Networks
Sequence to Sequence Learning with Neural Networks
 
Introduction to cassandra
Introduction to cassandraIntroduction to cassandra
Introduction to cassandra
 
Web browser architecture
Web browser architectureWeb browser architecture
Web browser architecture
 
Eclipse orion
Eclipse orionEclipse orion
Eclipse orion
 
X Query for beginner
X Query for beginnerX Query for beginner
X Query for beginner
 
Html 5
Html 5Html 5
Html 5
 
Redistributable introtoscrum
Redistributable introtoscrumRedistributable introtoscrum
Redistributable introtoscrum
 
Text categorization
Text categorizationText categorization
Text categorization
 
A holistic lexicon based approach to opinion mining
A holistic lexicon based approach to opinion miningA holistic lexicon based approach to opinion mining
A holistic lexicon based approach to opinion mining
 
Overview of NoSQL
Overview of NoSQLOverview of NoSQL
Overview of NoSQL
 

Recently uploaded

Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...
Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...
Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...stazi3110
 
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASEBATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASEOrtus Solutions, Corp
 
Folding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a seriesFolding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a seriesPhilip Schwarz
 
GOING AOT WITH GRAALVM – DEVOXX GREECE.pdf
GOING AOT WITH GRAALVM – DEVOXX GREECE.pdfGOING AOT WITH GRAALVM – DEVOXX GREECE.pdf
GOING AOT WITH GRAALVM – DEVOXX GREECE.pdfAlina Yurenko
 
What is Advanced Excel and what are some best practices for designing and cre...
What is Advanced Excel and what are some best practices for designing and cre...What is Advanced Excel and what are some best practices for designing and cre...
What is Advanced Excel and what are some best practices for designing and cre...Technogeeks
 
Balasore Best It Company|| Top 10 IT Company || Balasore Software company Odisha
Balasore Best It Company|| Top 10 IT Company || Balasore Software company OdishaBalasore Best It Company|| Top 10 IT Company || Balasore Software company Odisha
Balasore Best It Company|| Top 10 IT Company || Balasore Software company Odishasmiwainfosol
 
Cloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEECloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEEVICTOR MAESTRE RAMIREZ
 
Russian Call Girls in Karol Bagh Aasnvi ➡️ 8264348440 💋📞 Independent Escort S...
Russian Call Girls in Karol Bagh Aasnvi ➡️ 8264348440 💋📞 Independent Escort S...Russian Call Girls in Karol Bagh Aasnvi ➡️ 8264348440 💋📞 Independent Escort S...
Russian Call Girls in Karol Bagh Aasnvi ➡️ 8264348440 💋📞 Independent Escort S...soniya singh
 
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed DataAlluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed DataAlluxio, Inc.
 
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...Matt Ray
 
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)jennyeacort
 
Cloud Management Software Platforms: OpenStack
Cloud Management Software Platforms: OpenStackCloud Management Software Platforms: OpenStack
Cloud Management Software Platforms: OpenStackVICTOR MAESTRE RAMIREZ
 
Buds n Tech IT Solutions: Top-Notch Web Services in Noida
Buds n Tech IT Solutions: Top-Notch Web Services in NoidaBuds n Tech IT Solutions: Top-Notch Web Services in Noida
Buds n Tech IT Solutions: Top-Notch Web Services in Noidabntitsolutionsrishis
 
Der Spagat zwischen BIAS und FAIRNESS (2024)
Der Spagat zwischen BIAS und FAIRNESS (2024)Der Spagat zwischen BIAS und FAIRNESS (2024)
Der Spagat zwischen BIAS und FAIRNESS (2024)OPEN KNOWLEDGE GmbH
 
CRM Contender Series: HubSpot vs. Salesforce
CRM Contender Series: HubSpot vs. SalesforceCRM Contender Series: HubSpot vs. Salesforce
CRM Contender Series: HubSpot vs. SalesforceBrainSell Technologies
 
What is Fashion PLM and Why Do You Need It
What is Fashion PLM and Why Do You Need ItWhat is Fashion PLM and Why Do You Need It
What is Fashion PLM and Why Do You Need ItWave PLM
 
Xen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdfXen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdfStefano Stabellini
 
ODSC - Batch to Stream workshop - integration of Apache Spark, Cassandra, Pos...
ODSC - Batch to Stream workshop - integration of Apache Spark, Cassandra, Pos...ODSC - Batch to Stream workshop - integration of Apache Spark, Cassandra, Pos...
ODSC - Batch to Stream workshop - integration of Apache Spark, Cassandra, Pos...Christina Lin
 

Recently uploaded (20)

Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...
Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...
Building a General PDE Solving Framework with Symbolic-Numeric Scientific Mac...
 
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASEBATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
BATTLEFIELD ORM: TIPS, TACTICS AND STRATEGIES FOR CONQUERING YOUR DATABASE
 
Folding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a seriesFolding Cheat Sheet #4 - fourth in a series
Folding Cheat Sheet #4 - fourth in a series
 
GOING AOT WITH GRAALVM – DEVOXX GREECE.pdf
GOING AOT WITH GRAALVM – DEVOXX GREECE.pdfGOING AOT WITH GRAALVM – DEVOXX GREECE.pdf
GOING AOT WITH GRAALVM – DEVOXX GREECE.pdf
 
What is Advanced Excel and what are some best practices for designing and cre...
What is Advanced Excel and what are some best practices for designing and cre...What is Advanced Excel and what are some best practices for designing and cre...
What is Advanced Excel and what are some best practices for designing and cre...
 
Balasore Best It Company|| Top 10 IT Company || Balasore Software company Odisha
Balasore Best It Company|| Top 10 IT Company || Balasore Software company OdishaBalasore Best It Company|| Top 10 IT Company || Balasore Software company Odisha
Balasore Best It Company|| Top 10 IT Company || Balasore Software company Odisha
 
2.pdf Ejercicios de programación competitiva
2.pdf Ejercicios de programación competitiva2.pdf Ejercicios de programación competitiva
2.pdf Ejercicios de programación competitiva
 
Cloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEECloud Data Center Network Construction - IEEE
Cloud Data Center Network Construction - IEEE
 
Russian Call Girls in Karol Bagh Aasnvi ➡️ 8264348440 💋📞 Independent Escort S...
Russian Call Girls in Karol Bagh Aasnvi ➡️ 8264348440 💋📞 Independent Escort S...Russian Call Girls in Karol Bagh Aasnvi ➡️ 8264348440 💋📞 Independent Escort S...
Russian Call Girls in Karol Bagh Aasnvi ➡️ 8264348440 💋📞 Independent Escort S...
 
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed DataAlluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
Alluxio Monthly Webinar | Cloud-Native Model Training on Distributed Data
 
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...
Open Source Summit NA 2024: Open Source Cloud Costs - OpenCost's Impact on En...
 
Hot Sexy call girls in Patel Nagar🔝 9953056974 🔝 escort Service
Hot Sexy call girls in Patel Nagar🔝 9953056974 🔝 escort ServiceHot Sexy call girls in Patel Nagar🔝 9953056974 🔝 escort Service
Hot Sexy call girls in Patel Nagar🔝 9953056974 🔝 escort Service
 
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)
Call Us🔝>༒+91-9711147426⇛Call In girls karol bagh (Delhi)
 
Cloud Management Software Platforms: OpenStack
Cloud Management Software Platforms: OpenStackCloud Management Software Platforms: OpenStack
Cloud Management Software Platforms: OpenStack
 
Buds n Tech IT Solutions: Top-Notch Web Services in Noida
Buds n Tech IT Solutions: Top-Notch Web Services in NoidaBuds n Tech IT Solutions: Top-Notch Web Services in Noida
Buds n Tech IT Solutions: Top-Notch Web Services in Noida
 
Der Spagat zwischen BIAS und FAIRNESS (2024)
Der Spagat zwischen BIAS und FAIRNESS (2024)Der Spagat zwischen BIAS und FAIRNESS (2024)
Der Spagat zwischen BIAS und FAIRNESS (2024)
 
CRM Contender Series: HubSpot vs. Salesforce
CRM Contender Series: HubSpot vs. SalesforceCRM Contender Series: HubSpot vs. Salesforce
CRM Contender Series: HubSpot vs. Salesforce
 
What is Fashion PLM and Why Do You Need It
What is Fashion PLM and Why Do You Need ItWhat is Fashion PLM and Why Do You Need It
What is Fashion PLM and Why Do You Need It
 
Xen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdfXen Safety Embedded OSS Summit April 2024 v4.pdf
Xen Safety Embedded OSS Summit April 2024 v4.pdf
 
ODSC - Batch to Stream workshop - integration of Apache Spark, Cassandra, Pos...
ODSC - Batch to Stream workshop - integration of Apache Spark, Cassandra, Pos...ODSC - Batch to Stream workshop - integration of Apache Spark, Cassandra, Pos...
ODSC - Batch to Stream workshop - integration of Apache Spark, Cassandra, Pos...
 

Deep Dialog System Review

  • 1. Dialog System Review Tran Trung Kien Saltlux Development Center - VDC
  • 2. Communicating Knowledge Vietnam Development Center  What is Dialogue System?  Definition  3 generations of Dialog System  Evaluation  Spoken Dialogue System  Architecture  Components  Some approaches  A Neural Conversation Model  Deep Reinforcement Learning for Dialogue Generation  Common Frameworks and Data sets  Discussion Contents
  • 3. Communicating Knowledge Vietnam Development Center  Definition:  DS is a computer program developed to converse with human, with a coherent structure.  DS can use text, speech, graphics, haptics, gestures and other modes for communication on both the input and output.  Nowadays, speech is most commonly used for the input and output => Spoken Dialogue System.  3 Generations of DS  G1: Symbolic Rule/Template Based QA  Focus on grammatical rule & ontological design by human experts (early AI approach)  Easy interpretation, debugging, and system update  Popular before late 90’s  Still in use in commercial systems and by bots startups  Limitations: heavily reliance on experts hard to scale over domains data used only to help design rules, not for learning What is Dialogue System - 1/3
  • 4. Communicating Knowledge Vietnam Development Center  G2: Data Driven, Learning  Data used not to design rules for NLU and action, but to learn statistical parameters in dialogue systems Reduce cost of hand-crafting complex dialogue manager Robustness against speech recognize errors in noisy environment  MDP(Markov Decision Process)/POMDP (Partially Observed MDP) & RL for dialogue policy  Discriminative (CRF) & generative (HMM) methods for NLU  Popular in academic research until 2014 (before deep learning arrived at the dialogue world); in parallel with G1 (BBN, AT&T, CMU, SRI, CU ...)  Limitations: Not easy to interpret, debug, and update systems Still hard to scale over domains Models & representations not powerful enough; no end-2-end, hard to scale up Remained academic until deep learning arrived What is Dialogue System – 2/3
  • 5. Communicating Knowledge Vietnam Development Center  G3: Data-Driven Deep Learning:  Like G2, data used to learn everything in dialogue systems Reduce cost of hand-crafting complex dialogue manager Robustness against speech recognize errors in noisy environment & against NLU errors MDP/POMDP & reinforcement learning for dialogue policy (same)  Neural models & representations are much more powerful  End-to-End learning becomes feasible  Attracted huge research efforts since 2015 (after deep learning’s success in vision/speech and in deep RL shown success in Atari games)  Limitations: Still not easy to interpret, debug, and update systems Lack interface btw cont. neural learning and symbolic NL structure to human users Lack active research in scaling over domains via deep transfer learning & RL  No clear commercial success reported yet  Evaluation:?  Still argueing, no evaluation method is set as standard.  BLEU is usually used.  Some researchers define their own evaluation metrics to measure quality. What is Dialogue System 3/3
  • 6. Communicating Knowledge Vietnam Development Center  Use speech as input and output  Architecture: Spoken Dialog System
  • 7. Communicating Knowledge Vietnam Development Center  Automatic Speech Recognition (ASR):  Convert from voice signal to Words and Manage uncertainty.  Challenges:  Environment noises Speech production: low fluency, false starts, filled pauses, repeats, corrections, accent, age, gender, differences between human-human and human-machine speech Technological familiarity of user  Spoken Language Understanding (SLU)”  Spoken Language Understanding is the task of extracting meaning from utterances  Convert from words to concepts Dialog acts (the overall intent of an utterance) Domain specific concepts Syntactic/Semantic parser  Very difficult under noisy conditions  Challenges:  Recognizer error, background noise resulting in indels (insertions / substitutions / deletions), word boundary detection problems  Language production phenomena: low fluency, false starts, corrections, repairs are difficult to parse  Meaning must often be assembled from multiple speaker turns  There are many, many possible ways to say the same thing. Spoken Dialog System - Components
  • 8. Communicating Knowledge Vietnam Development Center  Dialogue Management:  Map concepts to action.  Manage dialog history, states and general flow of the conversation  Language Generation:  Generate response for the input.  Text To Speech Synthesis:  Convert the generated response to speech and present to user. Spoken Dialog System - Components
  • 9. Communicating Knowledge Vietnam Development Center  Previous approaches are often restricted to specific domains (e.g., booking an airline ticket) and require hand-crafted rules.  Proposed a model based on their “Sequence to sequence learning with neural networks” (NIPS, 2014).  Can be trained end-to-end and thus requires much fewer hand-crafted rules.  Allows researchers to work on tasks for which domain knowledge may not be readily available, or for tasks which are simply too hard to design rules manually.  The model: A Neural Conversation Model – Oriol Vinyals, Quoc V.Le – Google Using the seq2seq framework for modeling conversations
  • 10. Communicating Knowledge Vietnam Development Center  Data sets:  IT Helpdesk Troubleshooting: Typical interaction word length: 400 Turn talking is clearly signaled 30M tokens (3M used as validation)  OpenSubtitles (Tiedemann, 2009): Noisy data set Movie conversation in XML format After preprocessed: – Train set: 62M sentences, 923M tokens – Validation set: 26M sentences, 295M tokens A Neural Conversation Model – Oriol Vinyals, Quoc V.Le – Google
  • 11. Communicating Knowledge Vietnam Development Center  Experiments:  IT Helpdesk: Trained single layer LSTM with 1024 memory cells using stochastic gradient descent with gradient clipping. Vocabulary: 20K words Conversation 1: VPN issues A Neural Conversation Model – Oriol Vinyals, Quoc V.Le – Google
  • 12. Communicating Knowledge Vietnam Development Center  Experiments:  OpenSubtitles: Train 2-layered LSTM, 4096 memory cells for each layer. Vocabulary: 100k most frequently words. A Neural Conversation Model – Oriol Vinyals, Quoc V.Le – Google
  • 15. Communicating Knowledge Vietnam Development Center  Conclusion:  A simple language model based on the seq2seq framework can be used to train a conversational engine .  It can generate simple and basic conversations, and extract knowledge from a noisy but open-domain dataset.  Purely data driven without any rules, but can generate quite proper answers.  A big limitation: lack of a coherent personality. A Neural Conversation Model – Oriol Vinyals, Quoc V.Le – Google
  • 16. Communicating Knowledge Vietnam Development Center  Authors: J. Li, W. Monroe, A. Ritter, M. Galley, J. Gao, D. Jurafsky  Despite the success of SEQ2SEQ models in dialogue generation, two problems emerge:  How to keep the conversation longer? Seq2seq models tend to generate generic responses like “I don’t know” regardless the input. => Responses like this will close the conversion. The cause is seq2seq use MLE objective function. But the frequency if those generic responses is very high in training set.  System becomes stuck in infinite loop of repetitive responses. This is due to MLE-base seq2seq models’ inability to account for repetition. Deep RL for Dialogue Generation
  • 17. Communicating Knowledge Vietnam Development Center  => we need a conversation framework that has the ability to:  (1) integrate developer-defined rewards that better mimic the true goal of chatbot development.  (2) model the long- term influence of a generated response in an ongoing dialogue.  Proposed a neural RL generation method:  can optimize long-term rewards designed by system developers.  uses the encoder- decoder architecture as its backbone  simulates conversation between two virtual agents to explore the space of possible actions while learning to maximize expected reward.  We define simple heuristic approximations to rewards that characterize good conversations: good conversations are forward-looking or interactive (a turn suggests a following turn), informative, and coherent.  Use policy gradient method instead of MLE objective function.  Authors goal is to integrate Seq2seq and RL to get advantages of both. Deep RL for Dialogue Generation
  • 18. Communicating Knowledge Vietnam Development Center  Reward: r  Ease of answering: generated answer should be easy to respond. S: set of 8 manually collected dull response (“I don’t know”, …) NS: size of S, s: a sequence in S, Ns: # of token in s. Pseq2seq: the likelihood calculated by Seq2seq model.  Information flow: agent should contribute new information to keep dialogue moving => penalizing semantic similarity between 2 consecutive turns of agent: hpi, hpi+1 resulted from encoder for pi, pi+1 Deep RL for Dialogue Generation
  • 19. Communicating Knowledge Vietnam Development Center  Reward: r  Semantic Coherence: to avoid high reward but not grammatical and coherent Pseq2seq(a|pi, qi): probability of generating a given the previous utterances [pi, qi] 2nd part: backward probability of generating the previous dialogue utterance qi based on response a  Final reward r: lamda1 + lamda2 + lamda3 = 1, lamda1 = lamda2 = 0.25, lamda3 = 0.5 Deep RL for Dialogue Generation
  • 20. Communicating Knowledge Vietnam Development Center  Simulation: Deep RL for Dialogue Generation
  • 21. Communicating Knowledge Vietnam Development Center  Experiment results:  Sub set of 10M messages from OpenSubtitles dataset and extract 0.8M message with lowest likelihood of generating dull response to ensure the initial input is easy to respond Deep RL for Dialogue Generation
  • 22. Communicating Knowledge Vietnam Development Center  Experiment results:
  • 24. Communicating Knowledge Vietnam Development Center  TensorFlow:  Open source software library for numerical computation using data flow graphs  IrisTK:  Java-based framework for developing spoken dialogue systems.  Url: http://www.iristk.net/  OpenDial:  Java-based, domain-independent toolkit for developing spoken dialogue systems  Url: http://www.opendial-toolkit.net/  CSLU Toolkit:  a comprehensive suite of tools to enable exploration, learning, and research into speech and human-computer interaction.  http://www.cslu.ogi.edu/toolkit/  NADIA: (developed by MARKUS M. BERG)  set of Java-based components that deals with the creation of spoken dialogue systems.  Detail information (Phd Thesis, paper: http://mmberg.net/nadia/  Reference source code (include data model): https://github.com/mmberg  Datasets:  https://github.com/karthikncode/nlp-datasets  Unbutu Dialogue Corpus Frameworks and Datasets for SDS
  • 25. Communicating Knowledge Vietnam Development Center  Three generations of SDS – Li Deng, Chief Scientist of AI, MS AI  The Unreasonable Effectiveness of Recurrent Neural Networks  A neural conversation model – Oriol Vinyals, Quoc V.Le – Google - 2015  Deep reinforcement learning for Dialogue Generation – Jiwei Li, Will Monroe, Dan Jurafsky (Standford Univ.), Alan Ritter (Ohio State Univ.), Michel Galley, Jianfeng Gao (MS Research) - 2016  Neural responding machine for short-text conversation – Lifeng Shang, Zhengdong Lu, Hang Li – Huawei Tech., 2015  Deep RL: An overview – Yuxi Li - 2017  Dialogue System – Wikipedia: https://en.wikipedia.org/wiki/Dialog_system  Speech Recognition: https://en.wikipedia.org/wiki/Speech_recognition  Neural Network Dialog System Papers: https://github.com/snakeztc/NeuralDialogPapers  Datasets for Natural Language Processing: https://github.com/karthikncode/nlp-datasets References