Outline Language Technology II Tasks of dialogue management - PowerPoint PPT Presentation

Outline Language Technology II • Tasks of dialogue management Dialogue Management • Dialogue-flow control • Finite State-Based DM • Frame-Based DM • ISU-Based DM Ivana Kruijff-Korbayová • Grounding and Verification korbay@coli.uni-sb.de • Inititative and Cooperation www.coli.uni-sb.de/~korbay/ � Teaching • Current challenges 6/29/06 Ivana Kruijff-Korbayová 1 6/29/06 Ivana Kruijff-Korbayová 2 Language Technology II: Dialogue Management Language Technology II: Dialogue Management Tasks of Dialogue Management • Dialogue flow control • Dialogue modeling Dialogue Flow Control – Dialogue context – Dialogue moves when to say something, • Error handling when to stop • Initiative and cooperation � turn taking • Adaptivity • … 6/29/06 Ivana Kruijff-Korbayová 3 6/29/06 Ivana Kruijff-Korbayová 4 Language Technology II: Dialogue Management Language Technology II: Dialogue Management

Turn Taking Rules Turn Taking • Conversational analysis • Dialogue participants take turns (like in a game): • When does turn-taking occur: A, B, A, B – Transition-relevance places (TRPs) ---points where the • Dialogue turn = a continuous “contribution” to dialog/utterance structure allows speaker shift to occur (typically at utterance boundaries, but also smaller units, e.g., phrases) the dialogue from one speaker – TRP signals include syntax (phrase boundaries), intonation, gaze, • Though it is generally not obvious when a turn in gesture; Also cultural conventions apply • Who speaks next natural dialog is finished, turn-taking appears – At each TRP (current speaker A): fluid in normal conversation: • If A selected B as next speaker, B should speak – Minimal pauses between speakers (few hundred ms) • If A did not select the next speaker, then anyone may take a turn • If no-one else takes a turn, then A may (continue) – Less than 5% speech overlap – To get a turn if not selected, a speaker must “jump in” at a TRP • How does it work? • When do we get pauses or lapses? When do we get overlaps? 6/29/06 Ivana Kruijff-Korbayová 5 6/29/06 Ivana Kruijff-Korbayová 6 Language Technology II: Dialogue Management Language Technology II: Dialogue Management Turn Taking in Human-Computer Dialogue • Rigid: strict separation of system/user turns – How to determine the end of user’s turn? (Is s/he finished?) – How long to wait for user’s turn? (Is the user still engaged? Did Dialogue Modeling s/he hear?) – Avoid user’s speaking too early by explicit turn-taking signals • Flexible, with barge-in: – User barge-in: system stops speaking when it detects input Where we are • Open-mic: system listening all-the-time – Problems: talk vs. noise; system’s own talk is also “noise” & • Push-to-talk: user pushes a button to open the mic (take a turn) Problem: What has actually been conveyed to the user? What is the What to say next resulting common ground between the system and the user? E.g., list with several options, complex info --> reference resolution – System barge-in: When appropriate at all? When is a TRP? 6/29/06 Ivana Kruijff-Korbayová 7 6/29/06 Ivana Kruijff-Korbayová 8 Language Technology II: Dialogue Management Language Technology II: Dialogue Management

Global Dialogue Structure Local Dialogue Structure Opening • Adjacency pairs or dialogue games: Opening – Turns produced by different speakers + task info yes – Ordered: First^Second (initiation - response) Novic? + control options – Typed: particular First requires a particular Second Task no • Greet-greet, ask-answer, request-grant, offer-accept, abandon Task compliment-downplay, etc. � preferences, expectations yes restart More? • Insertion sequences: APs can be embedded no yes – E.g., “sub-dialogue”, misapprehension-correction, More? clarification Closing no Closing 6/29/06 Ivana Kruijff-Korbayová 9 6/29/06 Ivana Kruijff-Korbayová 10 Language Technology II: Dialogue Management Language Technology II: Dialogue Management Local Structure: Insertions Methods of DM • “Sub-dialogue”: • Script-based: Finite automata A: Where are you going? B: Why do you want to know? – Sequence of pre-defined steps (dialogue script) Task complexity A: I thought I’d come with you. • Frame-based (also: form-filling) B: I’m going to the supermarket. • Clarification: – Set of slots to be filled (task template) and A: I’d like three sausages. corresponding prompts B: Which ones? Merquez or Lyoner? A: Merquez. • Information-State Update B: Here you go. • Misapprehension-Correction: – Declarative rules for updating dialogue context A: When is the next train from SB to Hamburg? B: The next train to Homburg is at 1 p.m. A: Hamburg, not Homburg. B: Ah, Hamburg? A: Yes. B: The next connection to Hamburg Hauptbahnhof is at 3 p.m. 6/29/06 Ivana Kruijff-Korbayová 11 6/29/06 Ivana Kruijff-Korbayová 12 Language Technology II: Dialogue Management Language Technology II: Dialogue Management

DM Based on Finite Automata • Automaton describes all possible dialogues • Set of states and transitions Script-Based DM – State determines system utterance (Finite Automata) – User utterance determines transition to next state (deterministic) • No recursion! (= no nested subdialogues) • Fixed dialogue script • System-driven interaction 6/29/06 Ivana Kruijff-Korbayová 13 6/29/06 Ivana Kruijff-Korbayová 14 Language Technology II: Dialogue Management Language Technology II: Dialogue Management Finite Automaton FSM-Based Models (Finite State Machine) U: Elevator? • <States, Init-State, Alphabet, Transition-fction> S: Hello. Which floor would you like to go to? Floor_n U: Third floor. • Variants: machines having S: OK, I am taking you to the third floor. – actions associated with states (Moore machine) … floor n – actions associated with transitions (Mealy machine) – multiple start states init Floor_1 Welcome Ask_floor – transitions conditioned on no input symbol (a null) – more than one transition for a given symbol and state (nondeterministic finite state machine) floor 1 – states designated as accepting states (recognizer) � States: … unknown – etc. Init-State: … Alphabet: … See, e.g., NIST http://www.nist.gov/dads/HTML/finiteStateMachine.html Transition function: … Not_und 6/29/06 Ivana Kruijff-Korbayová 15 6/29/06 Ivana Kruijff-Korbayová 16 Language Technology II: Dialogue Management Language Technology II: Dialogue Management

FSM-Based Models FSM-Based Models bad # bad # States: … U: Elevator? Person>Floor # Init-State: … S: Hello. Where would you like to go to? � Alphabet: … U: Prof. Barry. Transition function: … S: Prof. Barry is on the fourth floor. I am taking you to the fourth floor. person repeat acct# repeat pin init unknown correct pin Go_floor welcome Ask_floor good # bad # bad # � floor # what init get_pin# Lookup get_acct# unknown Extension: variable for floor number service good # Not_und correct pin balance [McTear 2002] 6/29/06 Ivana Kruijff-Korbayová 17 6/29/06 Ivana Kruijff-Korbayová 18 Language Technology II: Dialogue Management Language Technology II: Dialogue Management FSM-Based Models FSM-Based DM: Sum Up unknown unknown States: … Init-State: … • Advantages Alphabet: … – Fixed prompts can be pre-recorded Transition function: … – Speech recognition and input interpretation can be tuned for each state repeat depart repeat dest • Disadvantages unknown city – Very rigid dialogue flow city unknown – Inhibiting user initiative unknown – Only suitable for simple tasks – In principle can make more flexible, but it quickly gets very what init …. dest_city Lookup depart_city complex info However: modular solutions are possible (--> DiaManT) city city flight 6/29/06 Ivana Kruijff-Korbayová 19 6/29/06 Ivana Kruijff-Korbayová 20 Language Technology II: Dialogue Management Language Technology II: Dialogue Management

Frame-Based Models • Frame (form): what info should be supplied by user departure_city ? Frame-Based DM departure_date ? destination_city ? (Form Filling) return_date ? … • Dialogue states: which slots are filled • General routines for what system should do next (given which slots are filled) 6/29/06 Ivana Kruijff-Korbayová 21 6/29/06 Ivana Kruijff-Korbayová 22 Language Technology II: Dialogue Management Language Technology II: Dialogue Management Frame-Based Models Frame-Based Models S: What can I do for you? departure_city ? • Deciding what to do next U: I want to fly to Paris departure_date ? – Next unfilled slot destination_city Paris return_date ? – Slot-combination weighting ... – Ontology-based coherence • Database lookup S: Where will you fly from? departure_city Berlin – Delayed (typically; after certain slots filled) U: From Berlin on August 1 st . departure_date 1/8/05 destination_city Paris – Immediate (can be “expensive” = take time, return_date ? but enables more helpful system behavior) “Overanswering” ... 6/29/06 Ivana Kruijff-Korbayová 23 6/29/06 Ivana Kruijff-Korbayová 24 Language Technology II: Dialogue Management Language Technology II: Dialogue Management