Grounding in dialogue systems Staffan Larsson Inst. för lingvistik, GU [email protected] OFTI 2002, Göteborg.

Grounding in dialogue systems

Staffan LarssonInst. för lingvistik, GU

[email protected] 2002, Göteborg

Overview

• Background• Interactive Communication

Management (ICM)• Action levels and metaissues• Feedback properties• Update strategies• ICM and grounding for a dialogue

system

Background

• Research on dialogue and dialogue systems– TRINDI (1997-2000)– SDS (1997-1998)– SIRIDUS (2000-2002)– D’Homme (2001)

• Implementation– TrindiKit: toolkit for building dialogue systems,

information state approach– GoDiS: dialogue system; issue-based dialogue

management; implemented using TrindiKit

GoDiS in SIRIDUS• explore and implement issue-based dialogue

management– adapt Ginzburg’s KOS to dialogue system (GoDiS) and

implement– extend theory to handle more flexible dialogue (incl.

grounding, accommodation, action-oriented dialogue, negotiation, conditional responses)

– implement extensions

• separating general and domain-dependent phenomena helps reconfigurability– general theory of dialogue– extended into subtheories for different dialogue genres– domain knowledge clearly separated– minimize effort for adapting to new genres and domains

TrindiKit

GoDiS

GoDiS-I GoDiS-A

TravelAgency

Auto-route

Xeroxmanual

VCRmanager

basic IBDM

homedevice

manager

ISapproach

action-oriented

IBDM

T.A. domain

knowledge

inquiry-oriented

IBDM

Basic issue-based dialogue management

• dialogue is, basically, all about raising and addressing issues– incl. short answers– issue reraising and accommodation

• starting point: KOS framework [Ginzburg]– Dialogue Gameboard (DGB)– related DGB update protocols

• dialogue moves: ask, answer, (greet, quit)• other features

– dialogue plans– handling multiple simultaneous issues– information sharing between plans

• initial genre: enquiry-oriented dialogue (database search)• sample domain: travel agency

Interactive Communication Management [Allwood]

• feedback– purpose: regulate grounding (adding to common ground)

[Clark]– feedback moves reflect grounding status of utterances

• turntaking ICM– purpose: regulate turntaking– turntaking moves reflects turntaking structure of dialogue

• sequencing– purpose:

• coordination of common ground other than grounding• indicating ”internal” mental moves affecting common ground

– sequencing moves reflects dialogue structure (part of common ground)

Action levels in dialogue [Allwood, Clark]

• contact• perception• understanding• acceptance

Grounding and action levels

• ”To ground a thing … is to establish it as part of common ground well enough for current purposes.” [Clark]

• grounding applies to all action levels– not just understanding

• U is grounded on level L iff– the grounding issue on level L is positively resolved

• grounding assumptions correspond to information state updates in system

Feedback polarity [Allwood et al ’91]

• polarity: positive, negative– indicating e.g. understanding (+) or lack thereof (-)

• eliciting/non-eliciting (evocative/non-evoc.)– whether utterance introduces obligation to respond

• Examples– ”What do you mean?”

• negative, eliciting– ”Do you mean that the destination is Paris?”

• ??negative??, eliciting– ”To Paris.”

• positive, non-eliciting– ”Pardon?”

• negative, eliciting

Form and content of ICM dialogue moves

• Form:– declarative: ”I didn’t hear what you said.”; ”The

destination city is Paris.”– interrogative: ”What did you say?”; ”Do you want to go

to Paris?”– imperative: ”Please repeat your latest utterance!”– elliptical

• interrogative: ”Paris?”, ”To Paris or from Paris?”• declarative: ”To Paris.”

• Content:– object-level: ”To Paris?”, ”Do you want to go to Paris?”– metalevel: ”Did you mean you want to go to Paris?”– none (except polarity): ”Pardon?”, ”OK”

ICM in GoDiS

• Grounding moves– all four action levels– simplified polarities– coarse-grained semantics– no detailed account of form; template-based generation

• Sequencing moves– reraising issues– loading dialogue plans– question accommodation

• Turntaking moves– no account of turntaking moves; strict turntaking

enforced

Feedback polarities in GoDiS

• how far can we get with meta-issues? – we don’t model obligations– all feedback introduces or answers meta-issues – meta-issues may or may not be responded to; system

must be able to deal with both

• 3 ”polarities”, mutually exclusive– positive: pos

• implicitly introduces question such as ”was p a correct interpretation of U?”

– negative: neg• answers question such as ”did B understand U?”

– eliciting->interrogative: int• explicitly raises question, e.g. ”What does U mean?”

Some ICM dialogue moves

• feedback– icm:Level{*Polarity}{:Content}– icm:und*neg – ”I don’t understand”– icm:und*pos:P – ”To Paris.”– icm:und*int:Q – ”Did you mean to Paris or from

Paris?”– icm:acc*neg:Q – ”Sorry, I can’t answer Q”– icm:acc*pos – ”Okay”

• sequencing– icm:Type{:Content}– icm:reraise:Q – ”Returning to the issue Q”– icm:loadplan – ”Let’s see…”

System feedback for user utterances

• contact– negative (”I didn’t hear anything from you.”, ”hello”)

• perception– negative: fb-phrase (”Pardon?”, ”I didn’t hear what you said”)– positive: repetition (”I heard ’to paris’”)

• understanding– negative: fb-phrase (”I don’t quite understand”)– positive: reformulation (”To Paris.”)– interrogative: reformulation (”To Paris, is that correct?”, ”To

Paris?”)• acceptance/integration

– negative: fb-phrase with reformulation (”Sorry, I cannot answer Q”, ”Sorry, Paris is not a valid destination city.”)

– positive: fb-word (”Okay.”)

User feedback for system utterances

• contact: -• perception

– negative: fb-phrase (”Pardon?”, ”I didn’t hear what you said”)

• understanding: -• acceptance/integration

– negative (for questions): fb-phrase (”I don’t know”, ”Never mind”)

– positive: fb-word (”okay.”)

Grounding update strategies

• strategic questions:– When should U assumed to be grounded on level L?

• as soon as it has been uttered (of course, the hearer cannot assume grounding until grounding wh-issues have some answer, e.g. ”what did A say?” )

• if B does not give negative feedback• when B gives positive feedback• when B has given eliciting feedback which has been

confirmed by A

– What to do if the grounding assumption turns out to be mistaken

Grounding update strategies cont’d

• optimism on level L:– assume U is grounded on level L as soon as

U has been uttered

• cautious optimism:– make sure the optimistic assumption can

easily be retracted

• pessimism:– don’t assume U grounded until there has

been some positive feedback (or at least no negative feedback)

choice of strategies in system

• system utterances– optimistically assumed to be grounded on all levels– negative feedback on perception or acceptance levels ->

backtrack to saved state• user utterances

– if problem on any level, give corresponding negative fb• if several alternatives available, give interrogative fb

– if OK on all levels, update strategy and feedback determined by recognition score

– S > 0.9• optimistic update, icm:acc*pos

– 0.9 >= S > 0.8• optimistic update, icm:acc*pos, icm:und*pos:Content

– 0.8 >= S >0.5• pessimistic update, ask(?Content)• if ?Content recieves answer ”yes”, add assume Content grounded

S> Welcome to the travel agency! U> price information pleaseS> OK, you want to know about price, is that correct?icm:acc*pos icm:und*int:?issue(price)U> yesS> Lets see. How do you want to travel? U> by flightS> OK, by flight. What city do you want to go to?icm:acc*pos icm:und*pos:means_of_transport(fly) …U> do I need a visa ?…S> Lets see. What country are you from? icm:loadplanU> swedenS> Okay. You are a citizen of Sweden. Yes, you need a Visa. S> Returning to the issue of price… icm:reraise:?x.price(x)…S> What month do you want to leave?U> Don’t know icm:acc*neg…

Recommended reading!

• Staffan Larsson (2002): Issue-based Dialogue Management. PhD Thesis.

• Download the system– www.ling.gu.se/~sl/Thesis

• SIRIDUS project– www.ling.gu.se/projekt/siridus

rejections

• rejection of question Q– inability to answer Q

• ”Sorry, I can’t answer that question”– unwillingness to answer Q

• ”I don’t want to discuss that”

• rejection of proposition as issue– unwillingness to discuss whether ?P

• ”I don’t want to discuss that”– other reasons?

• rejection of proposition– ”Sorry, I don’t agree.”, ”You’re wrong!”, ”That’s

impossible!”– can be expected to lead to argumentation

problematic cases

S: ”Where do you want to go?”U1: ”Nowhere”U2: ”I don’t know”U3: (silence) OR ”I want first class!”

• do these count as rejections?– U1: negative answer? presupposition failiure?

rejection?– U2: rejection?

• but not as definite as ”No comment!”

– U3: rejection? • in any case, irrelevant followup