Computers and Language Topic 6: Dialogue Systems Introduction Human dialogue Basic facts Rules The Computer and Natural Language Speech acts Discourse structure (Ling 445/515) Early dialogue systems Turing test Topic 6: Dialogue Systems ELIZA Other chatterbots Modern dialogue systems Markus Dickinson TRIPS system Dialogue manager Dept. of Linguistics, Indiana Frame-based systems Natural Language Generation Autumn 2007 Pronoun resolution Human-computer interaction 1 / 64
What are dialogue systems good for? Computers and Language Topic 6: Dialogue Systems Introduction ◮ We can book airline tickets over the phone without Human dialogue Basic facts dealing with error-prone humans. Rules Speech acts ◮ We can interact with our computer while keeping our Discourse structure Early dialogue hands free. systems Turing test ◮ We can talk to a system that won’t have the same ELIZA Other chatterbots predujices a human might (... depending on the Modern dialogue programmer) systems TRIPS system ◮ When robots someday take over the world, we’ll be able Dialogue manager Frame-based systems to understand their demands ;) Natural Language Generation Pronoun resolution Human-computer interaction Before we look at dialogue systems, though, we need to understand something about human dialogue 2 / 64
Human dialogue Computers and Language Topic 6: Dialogue Systems Introduction Human dialogue Basic facts Rules How does dialogue, or discourse (= conversation), work? Speech acts Discourse structure Early dialogue We will look at a few aspects of human-human interactions: systems Turing test ◮ Basic facts about dialogues ELIZA Other chatterbots ◮ Rules of conversation: Gricean maxims Modern dialogue systems ◮ What utterances do: Speech acts TRIPS system Dialogue manager ◮ Organization of conversation: Discourse structure Frame-based systems Natural Language Generation Pronoun resolution Human-computer interaction 3 / 64
Basic facts about dialogues Computers and Language Utterances Topic 6: Dialogue Systems People do not necessarily speak in sentences Introduction Human dialogue ◮ Utterances : basic unit of conversation Basic facts Rules ◮ may span over several turns Speech acts Discourse structure ◮ may have several utterances within one turn Early dialogue systems AGENT: Yeah yeah the um let me see here Turing test ELIZA we’ve got you on American flight Other chatterbots nine thirty eight Modern dialogue systems CUSTOMER: Yep. TRIPS system Dialogue manager AGENT: leaving on the twentieth of June Frame-based systems Natural Language Generation out of Orange County John Wayne Pronoun resolution Human-computer interaction Airport at seven thirty p.m. CUSTOMER: Seven thirty. AGENT: and into uh San Fransisco at eight fifty seven. 4 / 64
Basic facts about dialogues Computers and Language Taking turns Topic 6: Dialogue Systems Introduction How humans communicate is relatively systematic Human dialogue Basic facts Rules ◮ Turn-taking : must know when it’s the right time to Speech acts Discourse structure contribute your turn Early dialogue systems ◮ Comes naturally to humans: overlaps and long pauses Turing test are actually somewhat rare ELIZA Other chatterbots ◮ Adjacency pairs : two-part conversational structures Modern dialogue systems where turn-taking is usually quite clear TRIPS system Dialogue manager ◮ question/answer: question is looking for an answer from Frame-based systems Natural Language Generation dialogue partner Pronoun resolution Human-computer interaction ◮ greeting/greeting ◮ request/grant ◮ etc. 5 / 64
Basic facts about dialogues Computers and Language Common ground Topic 6: Dialogue Systems Introduction Human dialogue Common ground : the set of things which both speakers Basic facts believe to be true of the conversation. Rules Speech acts Discourse structure ◮ Part of classroom common ground is that we are at Early dialogue Indiana in a linguistics class. Not part of the common systems Turing test ground is what I had for lunch. ELIZA Other chatterbots ◮ People assess what others know before making a new Modern dialogue systems contribution TRIPS system Dialogue manager ◮ Starting a conversation about Greg would be confusing Frame-based systems Natural Language Generation because we have no Greg in our common ground Pronoun resolution ◮ Starting a conversation by saying I was just thinking Human-computer interaction about an old friend of mine from Illinois named Greg is less confusing 6 / 64
Grounding Computers and Language Topic 6: Dialogue Systems Introduction In order to establish common ground, speakers do various Human dialogue Basic facts things: Rules Speech acts Discourse structure ◮ Backchannels = short utterance which indicates the Early dialogue utterance was heard and that the speaker should systems Turing test continue. ELIZA Other chatterbots A: That’s a nice shirt. Modern dialogue systems B: Mm-hmm. TRIPS system Dialogue manager ◮ Acknowledgment of utterance: continued attention, Frame-based systems Natural Language Generation completing speaker’s utterance, and so on Pronoun resolution Human-computer interaction ◮ Asking clarifying questions 7 / 64
Rules of Conversation: Gricean Maxims Computers and Language Topic 6: Dialogue Systems Introduction Human dialogue People often speak “indirectly,” but it’s very clear what they Basic facts Rules mean. Speech acts Discourse structure Early dialogue ⇒ They obey what Grice (1975) referred to as the four systems Turing test maxims (rules) or conversation ELIZA Other chatterbots ◮ All based on the cooperative principle = speakers are Modern dialogue systems both trying to contribute to the purposes of the TRIPS system Dialogue manager conversation. Frame-based systems Natural Language Generation ◮ We use these to infer what a speaker is really saying Pronoun resolution Human-computer interaction 8 / 64
Gricean maxim 1: Quantity Computers and Language Topic 6: Dialogue Systems Quantity: Be exactly as informative as is required. Introduction Human dialogue ◮ Make your contribution as informative as is required (for Basic facts Rules the current purposes of the exchange) Speech acts Discourse structure A: How many pencils do you have? Early dialogue systems B: Two. Turing test ELIZA → means exactly two (or would have said more) Other chatterbots Modern dialogue ◮ Do not make your contribution more informative than is systems TRIPS system required. Dialogue manager Frame-based systems Violation: Natural Language Generation Pronoun resolution (passing by somebody on the way to class) Human-computer interaction A: How’s it going? B It’s complicated. Yesterday I was sick ... 9 / 64
Gricean maxim 2: Quality Computers and Language Topic 6: Dialogue Systems Introduction Human dialogue Basic facts Rules Quality: Try to make your contribution one that is true. Speech acts Discourse structure ◮ Do not say what you believe to be false Early dialogue systems ◮ Do not say that for which you lack adequate evidence Turing test ELIZA Other chatterbots Violation: Modern dialogue systems A: Do you know how to drive a stick-shift? TRIPS system B: Yes, I do. I’ve seen my dad do it many times. Dialogue manager Frame-based systems Natural Language Generation Pronoun resolution Human-computer interaction 10 / 64
Gricean maxim 3: Relevance Computers and Language Topic 6: Dialogue Systems Introduction Human dialogue Basic facts Rules Relevance: Be relevant. Speech acts Discourse structure Early dialogue A: Is Gail dating anyone these days? systems B: Well, she goes to Cleveland every weekend. Turing test ELIZA Other chatterbots We make an inference that Gail is dating somebody in Modern dialogue systems Cleveland TRIPS system Dialogue manager ◮ Otherwise, B’s statement doesn’t make much sense. Frame-based systems Natural Language Generation Pronoun resolution Human-computer interaction 11 / 64
Gricean maxim 4: Manner Computers and Language Topic 6: Dialogue Systems Introduction Manner: Be perspicuous (easy to understand). Human dialogue Basic facts ◮ Avoid obscurity of expression Rules Speech acts Discourse structure ◮ In other words, avoid jargon when it’s not a part of the Early dialogue common ground, e.g., words like prolixity systems Turing test ◮ Avoid ambiguity ELIZA Other chatterbots ◮ Be brief (avoid unnecessary prolixity) Modern dialogue systems TRIPS system ◮ Similar to maxim of quantity, but think here of Dialogue manager long-winded conversations that provide the same Frame-based systems Natural Language Generation amount of information as a shorter one Pronoun resolution Human-computer interaction ◮ Be orderly 12 / 64
Recommend
More recommend
