Logic and Probabilistic Methods for Dialog
0 likes84 views
The document discusses advancements in dialog systems and virtual assistants, highlighting the importance of combining symbolic methods with machine learning for improved performance. It outlines the evolution of AI as a conversational user interface and reviews the current landscape of natural language processing and understanding. Additionally, it emphasizes research on dialogue coherence, turn-taking, and knowledge representation to enhance artificial agents' conversational abilities.
![© 2014 Nuance Communications, Inc. All rights reserved. 23
Deepak:
Growing POT for everyone…
(loads that I don’t understand)
1st Dialog State Tracking Challenge
Relational Belief Tracking
[Ramachandran & Ratnaparkhi’15]
Belief Tracking with Tree Surgery](https://image.slidesharecdn.com/probsymworkshopintroduction-210815001211/85/Logic-and-Probabilistic-Methods-for-Dialog-23-320.jpg)
Logic and Probabilistic Methods for Dialog
- 1. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 1 Logic and Probabilistic Methods for Dialog Valeria de Paiva & Charlie Ortiz ESSLLI Barcelona, 10-14 August 2015
- 2. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 2 Vision Overview of work on dialog to organize our own
- 3. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 3 + Ron Kaplan “Beyond the GUI: It’s Time for a Conversational User Interface”, Wired 2013
- 4. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 4
- 5. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 5 Source: http://www.bloomberg.com/news/articles/2015-02-03/i-ll-be-back-the-return-of-artificial-intelligence Setting the Scene: AI as an ecosystem
- 6. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 6 Functional Areas Big players Prediction Knowledge Retrieval Sentiment Personalization Learning Summarization VA Hundreds of startups & many capabilities
- 7. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 7 Personalized & proactive. Consistent assistance on every channel. Intelligence is an expectation Knows how to talk.
- 8. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 8 Siri Oct 2011 Google Now Jul 2012 Nina Aug 2012 USAA EVA CVS Cortana Aug 2014 Dom Oct 2014 USAA Coach May 2015 Voice based Virtual Assistants paving the way for general acceptance of AI INGE Lily May 2013 Tangerine Dragon Oct 2012
- 9. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 9 CoreASR performance • Continuous server ASR word error rate (WER) reduction ~18% / year: combination of algorithms, data, and computing • Deep learning (DNNs) is driving recent performance improvements in ASR and meaning extraction
- 10. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 10 DeepLanguageUnderstanding Symbolic methods complement machine learning in a common architecture Amount of domain-specific data Accuracy Combined Symbolic-only Data only
- 11. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 11 Structured Data VA Building Tools & Runtime (NLU + Dialog + Knowledge) SDK/API How Virtual Assistants are built today…
- 12. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 12 NLP + Dialog + Knowledge?
- 13. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 13 Dialogue, Dialog – (Raquel Fernández, Oxford Bibliographies, 2013) – Research on dialogue is concerned with the study of how communication takes place through language in conversation. – Investigates what makes a dialogue coherent; – how dialogue participants coordinate to take turns in speaking and to achieve mutual understanding; – and how knowledge about how dialogue works can be used to design artificial agents that are able to converse with humans. – By now well-established field of research at the interface of linguistics, psycholinguistics, sociolinguistics, and artificial intelligence. – Ginzburg, Jonathan, and Raquel Fernández. 2010. Computational models of dialogue. In Handbook of computational linguistics and natural language processing. Edited by Alexander Clark, Chris Fox, and Shalom Lappin
- 14. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 14 Timetable
- 15. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 15 Let the fun begin!
- 16. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 16 Recap:
- 17. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 17 So far: synergy here: 3 courses, workshop, plus Invited Talk wide area, very different approaches, lots left uncovered, e.g. ellipsis, (more) discourse, formal models of game- boards? consensus on annotations/ segmentation?
- 18. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 18 Taking home more than Elephant theory of dialog
- 19. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 19 Matrix of genres/emphasis
- 20. © 2014 Nuance Communications, Inc. All rights reserved. 20 Deepak: I’m interested in building systems that can handle simple dialogs but do so in a robust manner.
- 21. © 2014 Nuance Communications, Inc. All rights reserved. 21
- 22. © 2014 Nuance Communications, Inc. All rights reserved. 22 Deepak again…
- 23. © 2014 Nuance Communications, Inc. All rights reserved. 23 Deepak: Growing POT for everyone… (loads that I don’t understand) 1st Dialog State Tracking Challenge Relational Belief Tracking [Ramachandran & Ratnaparkhi’15] Belief Tracking with Tree Surgery
- 24. © 2002-2015 Nuance Communications, Inc. All Rights Reserved. Page 24 Further Plans? Shall we do another workshop? Shall we do a survey publication?
- 25. © 2014 Nuance Communications, Inc. All rights reserved. 25 Thank you!