This presentation covers dialogue systems: their definition, basic structure (covering all modules: natural language understanding, dialogue manager, natural language generation), evaluation and the way they can be used. We also provide details about future directions and discusses current personal assistants: SIRI, S-Voice, Cortana, Maluuba etc.
Formation of low mass protostars and their circumstellar disks
Dialogue systems and personal assistants
1. 1
Introduction to Dialogue SystemsIntroduction to Dialogue Systems
Personal Assistants are becoming a realityPersonal Assistants are becoming a reality
Dr Natalia Konstantinova
University of Wolverhampton
11 April 2014
2. 2
OutlineOutline
• What is a dialogue system?
• System structure and classification;
• Evaluation;
• Examples of existing systems;
• Future directions;
• IQA;
3. 3
DefinitionDefinition
• Artificial intelligence – idea to teach
machines to think and act as humans.
• NLP – give machines the ability to read,
understand and use natural language.
• Dialogue systems – part of artificial
intelligence challenge.
6. 6
What is a dialogue system?What is a dialogue system?
Ideas?
•
•
•
•
7. 7
DefinitionDefinition
• Editors of the Journal of Dialogue Systems :
“A dialogue system is a computational device or
agent that
• (a) engages in interaction with other human
and/or computer participant(s);
• (b) uses human language in some form such as
speech, text, or sign;
• and (c) typically engages in such interaction
across multiple turns or sentences.”
8. 8
Other termsOther terms
• Conversational agents (Jurafsky and
H.Martin, 2006), (Lester, Branting, and
Mott, 2004)
• “Chatterbot” or “chatbot”, first coined
by Mauldin (1994):
• simple dialogue systems, primarily based on
simple analysis of keywords in the input and
usage of different templates
9. 9
Where are they used?Where are they used?
Usually embedded in such applications as:
• customer service,
• help desks,
• website navigation,
• guided selling,
• technical support
(Lester, Branting, and Mott, 2004)
10. 10
““Body” for a dialogue systemBody” for a dialogue system
• Embodied conversational agents
(Cassell et al., 2000):
• has a “body”, where both verbal and
nonverbal devices advance and regulate the
dialogue between the user and the computer.
• Financial advisers, sales agents at online
shops
12. 12
System structureSystem structure
• They generally consist of 5 main
components (Jurafsky and H.Martin,
2006):
1. speech recognition;
2. natural language understanding (NLU);
3. dialogue management;
4. natural language generation (NLG);
5. speech synthesis.
13. 13
System structureSystem structure
• Some modules are optional:
• e.g. speech recognition and speech synthesis
• Dialogue systems involving speech are
more complicated:
• need to deal with errors in speech
recognition
• Speech recognition can be dialogue-state
dependant
14. 14
NLUNLU
• Aim of NLU module:
• produce a semantic representation
appropriate for a dialogue task.
15. 15
Dialogue managerDialogue manager
• One of the most important parts of DS
(Dale, Moisi, and Somers, 2000):
• interpret the speech acts;
• carry out problem-solving actions;
• formulate response;
• in general maintain the system's idea of the
state of the discourse (e.g. dialogue move
tree)
16. 16
Dialogue managerDialogue manager
• Interlink of NLU and NLG
• Responsible for the content generation
• (taking decisions about what to say and how)
17. 18
NLGNLG
• Chooses syntactic structures and words to
express the intended meaning, which was
formulated by a dialogue manager.
• How?:
• Templates to generate “prompts” (generated
outputs)
• Advanced natural language generators
21. 22
Frame/form based DMFrame/form based DM
• Simple and the most widely used
• Asks questions to fill in the slots in the
frame
• Perform a database query
• E.g. booking a holiday
22. 23
Information-state DMInformation-state DM
• More complicated
• Incorporates several ways to achieve a result.
• Components:
• the information state (the “discourse context” or
“mental model”);
• dialogue act interpreter (or “interpretation engine”);
• dialogue act generator (or “generation engine”);
• set of update rules (to update information state);
• control structure to select needed update rule.
23. 24
Plan-based DMPlan-based DM
• The most sophisticated one
• It interprets conversation as creation of a
plan and then interprets a plan “in reverse”
• Is often referred as BDI (belief, desire and
intentions) model.
24. 25
Other classificationsOther classifications
• system-initiative (or single initiative systems)
mixed initiative systems
• spoken dialogue systems text dialogue
systems
• multi-modal dialogue systems unimodal
dialogue systems
• domain restricted dialogue systems Open
domain dialogue systems
26. 27
EvaluationEvaluation
• How to make an objective evaluation?
• Task-based evaluation (Dale, Moisi, and
Somers, 2000):
• task completion success;
• efficiency cost;
• quality costs.
27. 28
EvaluationEvaluation
• Asking people to complete a question list
and rank the quality of the system giving
grades:
• E.g. evaluate naturalness
• Maybe not very objective
30. Competitors of SIRICompetitors of SIRI
• Cortana by Microsoft;
• Voice Mate by LG;
• S-Voice by Samsung;
• Google Now;
• E.g. Android versions: Maluuba; Robin; Iris;
Vlingo; Skyvi;
• More similar apps;
31
31. Further directionsFurther directions
• Currently DM in all commercial systems
is rule- based;
• What can be used?
• Reinforcement learning (hierarchical RL);
• Online learning;
• Dialogue manager based on partially observable
Markov decision process (POMDP);
• Quality-adaptive DM;
32
32. 33
ReferencesReferences
• Cassell, Justine, Joe Sullivan, Scott Prevost, and Elizabeth F. Churchill, editors. 2000. Embodied
Conversational Agents. Cambridge, MA: MIT Press.
• Dale, Robert, Hermann Moisi, and Harold Somers, editors. 2000. Handbook of Natural
Language Processing. Marcel Dekker, Inc.
• Jurafsky, Daniel and James H.Martin. 2006. Speech and language processing an introduction to
natural language processing, computational linguistics, and speech recognition. Prentice-Hall,
Inc.
• Lester, James, Karl Branting, and Bradford Mott. 2004. Conversational agents. In Munindar P
Singh, editor, The Practical Handbook of Internet Computing. Chapman & Hall.
• Mauldin, Michael L. 1994. Chatterbots, Tinymuds, and the Turing test: Entering the Loebner
prize competition. In Proceedings of the Eleventh National Conference on Artificial Intelligence.
AAAI Press.
• Mitkov, Ruslan, editor. 2003. Handbook of Computational linguistics. Oxford University Press,
USA.
• Sacks, H., E. A. Schegloff, and G. Jefferson. 1974. A simplest systematics for the organization of
turn-taking for conversation. Language, 50(4):696-735.
• Varges, S., F. Weng, and H. Pon-Barry. 2007. Interactive question answering and constraint
relaxation in spoken dialogue systems. Natural Language Engineering, 15(1):9-30.
• Webb, Nick and Bonnie Webber. 2009. Special issue on interactive question answering:
Introduction. Natural Language Engineering, 15(1):1-8, January.