Muutke küpsiste eelistusi

E-raamat: Statistical Methods for Spoken Dialogue Management

  • Formaat: PDF+DRM
  • Sari: Springer Theses
  • Ilmumisaeg: 08-Jan-2013
  • Kirjastus: Springer London Ltd
  • Keel: eng
  • ISBN-13: 9781447149231
Teised raamatud teemal:
  • Formaat - PDF+DRM
  • Hind: 110,53 €*
  • * hind on lõplik, st. muud allahindlused enam ei rakendu
  • Lisa ostukorvi
  • Lisa soovinimekirja
  • See e-raamat on mõeldud ainult isiklikuks kasutamiseks. E-raamatuid ei saa tagastada.
Statistical Methods for Spoken Dialogue ManagementSuurem pilt
  • Formaat: PDF+DRM
  • Sari: Springer Theses
  • Ilmumisaeg: 08-Jan-2013
  • Kirjastus: Springer London Ltd
  • Keel: eng
  • ISBN-13: 9781447149231
Teised raamatud teemal:

DRM piirangud

  • Kopeerimine (copy/paste):

    ei ole lubatud

  • Printimine:

    ei ole lubatud

  • Kasutamine:

    Digitaalõiguste kaitse (DRM)
    Kirjastus on väljastanud selle e-raamatu krüpteeritud kujul, mis tähendab, et selle lugemiseks peate installeerima spetsiaalse tarkvara. Samuti peate looma endale  Adobe ID Rohkem infot siin. E-raamatut saab lugeda 1 kasutaja ning alla laadida kuni 6'de seadmesse (kõik autoriseeritud sama Adobe ID-ga).

    Vajalik tarkvara
    Mobiilsetes seadmetes (telefon või tahvelarvuti) lugemiseks peate installeerima selle tasuta rakenduse: PocketBook Reader (iOS / Android)

    PC või Mac seadmes lugemiseks peate installima Adobe Digital Editionsi (Seeon tasuta rakendus spetsiaalselt e-raamatute lugemiseks. Seda ei tohi segamini ajada Adober Reader'iga, mis tõenäoliselt on juba teie arvutisse installeeritud )

    Seda e-raamatut ei saa lugeda Amazon Kindle's. 

Speech is the most natural mode of communication and yet attempts to build systems which support robust habitable conversations between a human and a machine have so far had only limited success. A key reason is that current systems treat speech input as equivalent to a keyboard or mouse, and behaviour is controlled by predefined scripts that try to anticipate what the user will say and act accordingly. But speech recognisers make many errors and humans are not predictable; the result is systems which are difficult to design and fragile in use.Statistical methods for spoken dialogue management takes a radically different view. It treats dialogue as the problem of inferring a user's intentions based on what is said. The dialogue is modelled as a probabilistic network and the input speech acts are observations that provide evidence for performing Bayesian inference. The result is a system which is much more robust to speech recognition errors and for which a dialogue strategy can be learned automatically using reinforcement learning. The thesis describes both the architecture, the algorithms needed for fast real-time inference over very large networks, model parameter estimation and policy optimisation. This ground-breaking work will be of interest both to practitioners in spoken dialogue systems and to cognitive scientists interested in models of human behaviour.

Treating dialogue as a problem of inferring a speaker's intentions based on what is said, this book describes the architecture, the algorithms needed for fast real-time inference over very large networks, model parameter estimation and policy optimisation.
1 Introduction
1(6)
1.1 Thesis Outline and Contributions
3(4)
References
4(3)
2 Dialogue System Theory
7(20)
2.1 Components of a Spoken Dialogue System
7(5)
2.1.1 Speech Recognition
7(1)
2.1.2 Spoken Language Understanding
8(2)
2.1.3 Decision Making
10(1)
2.1.4 Response Generation
11(1)
2.1.5 Extensions to the Dialogue Cycle
11(1)
2.2 User Simulation
12(1)
2.3 The Dialogue Manager
13(14)
2.3.1 Hand-Crafted Dialogue Management
13(2)
2.3.2 Partial Observability
15(1)
2.3.3 Policy Learning
16(1)
2.3.4 Partially Observable Markov Decision Processes
17(5)
References
22(5)
3 Maintaining State
27(18)
3.1 Bayesian Networks
28(1)
3.2 Bayesian Networks for Dialogue State
29(4)
3.2.1 Goal Dependencies
30(1)
3.2.2 The History Nodes
31(1)
3.2.3 Sub-Components for the User Act Nodes
31(1)
3.2.4 Remaining Nodes
32(1)
3.3 TOWNINFO States: An Example
33(1)
3.4 Factor Graphs
34(1)
3.5 Belief Propagation
35(5)
3.5.1 The Approximation
36(1)
3.5.2 The Aim
37(1)
3.5.3 The Calculation
38(1)
3.5.4 The Belief Propagation Algorithm
39(1)
3.6 Comparison to Previous Work
40(1)
3.7 The Loss of the Markov Property
41(1)
3.8 Limiting the Time-Slices
41(2)
3.9 Conclusion
43(2)
References
43(2)
4 Maintaining State: Optimisations
45(12)
4.1 Expectation Propagation
45(1)
4.2 k-Best Belief Propagation
46(5)
4.2.1 The Idea
46(1)
4.2.2 The New Update Equation
47(1)
4.2.3 Reducing Complexity
48(2)
4.2.4 Choosing the k-Best
50(1)
4.2.5 Related Work
50(1)
4.3 Grouped Belief Propagation
51(1)
4.4 Mostly Constant Factors
51(2)
4.5 Experimental Comparison of Inference Algorithms
53(2)
4.6 Conclusion
55(2)
References
55(2)
5 Policy Design
57(14)
5.1 Policy Learning Theory
57(1)
5.2 Summary Actions
58(1)
5.3 TOWNINFO Summary Acts: An Example
59(2)
5.4 Function Approximations for Dialogue Management
61(1)
5.5 TOWNINFO Function Approximations: An Example
62(2)
5.5.1 Grid-Based Features with No Parameter Tying
62(1)
5.5.2 Grid-Based Features with Parameter Tying
63(1)
5.6 Natural Actor Critic
64(3)
5.6.1 Sampling Methods
66(1)
5.7 Simulation
67(1)
5.8 TOWNINFO Learning: An Example
68(1)
5.9 Conclusion
69(2)
References
69(2)
6 Evaluation
71(12)
6.1 TownInfo Systems
71(5)
6.1.1 Hand-Crafted State Transitions
72(2)
6.1.2 Partially Observable State Transitions
74(2)
6.2 Simulated Comparison
76(1)
6.3 User Trial
76(2)
6.4 Evaluating the Effects of Semantic Errors
78(3)
6.5 Conclusion
81(2)
References
81(2)
7 Parameter Learning
83(20)
7.1 An Extended TOWNINFO System
84(3)
7.1.1 Extended TOWNINFO Dialogue State
84(1)
7.1.2 Extended TOWNINFO Summary Acts
84(2)
7.1.3 Extended TOWNINFO Summary Features
86(1)
7.2 Specialised User Act Factors
87(2)
7.3 Learning Dirichlet Distributions
89(6)
7.3.1 The Approximation and Target Functions
90(3)
7.3.2 Matching the Target Function
93(1)
7.3.3 The Algorithm
94(1)
7.4 Tied Dirichlet Distributions
95(3)
7.5 Parameter Learning for Re-Scoring Semantics
98(2)
7.5.1 Simulated Evaluation on TOWNINFO
98(1)
7.5.2 User Data Evaluation on TOWNINFO
99(1)
7.6 Parameter Learning for Improving the Dialogue Manager
100(1)
7.7 Conclusion
101(2)
References
102(1)
8 Conclusion
103(2)
Appendix A Dialogue Acts Formats 105(2)
Appendix B Proof of Grouped Loopy Belief Propagation 107(4)
Appendix C Experimental Model for Testing Belief Updating Optimisations 111(2)
Appendix D The Simulated Confidence Scorer 113(2)
Appendix E Matching the Dirichlet Distribution 115(4)
Appendix F Confidence Score Quality 119(14)
Author Biography 133(2)
Index 135
Blaise Thomson is a Research Fellow at St John's College in the University of Cambridge. He obtained a Bachelors degree in Pure Mathematics, Computer Science, Statistics and Actuarial Science at the University of Cape Town, South Africa, before completing an MPhil at the University of Cambridge in 2006 and a PhD in Statistical Dialogue Modelling in 2010.  He has published around 35 peer-reviewed journal and conference papers, focusing largely on the topics of dialogue management, automatic speech recognition, speech synthesis, natural language understanding and collaborative filtering. In 2008 he was awarded the IEEE Student Spoken Language Processing award for his paper at the International Conference on Acoustics, Speech, and Signal Processing (ICASSP) and in 2010 he co-authored best papers at both the IEEE Spoken Language Technologies workshop and Interspeech. He was co-chair of the 2009 ACL Student Research Workshop and co-presented a tutorial on POMDP dialogue management at Interspeech 2009.

In his spare time, he enjoys playing guitar and dancing and represented England at the 2010, 2011 and 2012 world formation Latin championships.
Lisainfo e-raamatute kohta Lisainfo e-raamatute kohta
Püsilink: https://www.kriso.ee/db/97814471492312e.html
Märksõnad: