discourse and dialogue

12 październik 2006 bruckenkurs – text structure and dialogue 1/71

discourse and dialogue


What is a unit of communication?

M: hi. d4 to d6.J: uh–huh.(week passes)

J: a3 to a7.M: hmmm.(2 weeks pass)

M: Queen beats the laufer at e1. Check....

Theories of discourse meaning depend in part on a specification of the basic units of a dicouse and the relations that can hold among them. Discourse processing requires an ability to determine to which portions of a discourse an individual utterance relates. Thus the role of discourse structure in discourse processing derives both from its role in delimiting units of discourse meaning and...


What is a discourse?

“Assume that you have collected an arbitrary set of well-formed

and independently interpretable utterances, for instance,

by randomly selecting one sentence from each of the previous

chapters of this book. Do you have a discourse? Almost certainly

not. The reason is that these utterances, when juxtaposed,

will not exhibit coherence. Consider, for example, the difference

between passages (18.71) and (18.72).”(Jurafsky and Martin:695)

“Consider, for example, the difference between passages

(18.71) and (18.72). Assume that you have collected an arbitrary

set of well-formed and independently interpretable utterances, for

instance, by randomly selecting one sentence from each of the

previous chapters of this book. Almost certainly not. Do you have

a discourse? The reason is that these utterances, when

juxtaposed, will not exhibit coherence.”

vs….

“Assume that you have collected an arbitrary set of well-formed

and independently interpretable utterances, for instance,

by randomly selecting one sentence from each of the previous

chapters of this book. Do you have a discourse? Almost certainly

not ~. The reason is that these utterances, when juxtaposed,

will not exhibit coherence. Consider, for example, the difference

between passages (18.71) and (18.72).”(Jurafsky and Martin:695)



The pool for members only. Please use the toilet, not the pool.The pool for members only.Please use the toilet, not the pool.



sentences are (typically) not processed in isolation

discourse, unlike an arbitratry collection of utterances, forms an intentionally meaningful whole (discourses are „about” something)

discourse has structure

segmentation and orderingcoherencecohesion


Discourse is internally linked; it „hangs together”

patterns of lexical connectivity cohesion

linguistic text-forming devices:

lexical repetition, synonymy/antonymy, ellipsis/pro-forms, enumeration, parallelism, co-reference (anaphora)

– Time flies.– You can’t; they fly too quickly.(Halliday and Hasan 1982)

– Time flies.– You can’t ~; they fly too quickly.(Halliday and Hasan 1982)


Anaphora: pronominal

My neighbor has a monster Harley 1200. They are huge but gas-efficient bikes.

One should mind their own business.

Anaphora: nominal (definite NP)

Al bought a car the other day. […] He took it out of the garage last night with the help of George Cottrell, and the thing gave forth such immense clouds of smoke that one man came running up and asked me where the fire was.

[…] I wanted a Trumpeter Swan who could play like Louis Armstrong, and I simply created him and named him Louis. The cutting of the webs between his toes is also fantastical, just as the bird itself is; […].


Anaphora: surface-count and demonstrative

Sarah could leave but she was also given an option to stay; she chose the latter.

Have just driven to town, carrying our cook1 and our cook’s dog2. Gave the one1

$300 in currency and placed the other2 in the infirmary, with eczema.


Temporal anaphora

If I must declare today that I am not a Communist, tomorrow I shall have to testify that I am not a Unitarian. And the day after, that I never belonged to a dahlia club.

Spatial anaphora

The awful hot spell broke last night and today is clear and beautiful, […] Across the street, the entire janitorial family has blossomed out in pink carnations, […]


Strained anaphora (bridging)

John became a guitarist because he thought that it was a beautiful instrument.

The house was beautiful. The door was painted white and the windows

had blue shutters.


Abstract entity anaphora

Each Fall, penguins migrate to Fiji.

That’s where they wait out the winter.

That’s when it’s cold even for them.

That’s why I’m going there next month.

It happens just before the eggs hutch.(Webber 1988)

Send an engine to Elmira.

That’s six hours.(Byron 2002)


Ellipsis

The well water had chemicals in it and nothing in the house worked as it should [work].

[I] Have been uncommunicative lately, and [I have been] lagging in life’s race.

I’m afraid my poem isn’t as nicely written as “Paradise Lost,” but anyway, it’s shorter [than “Paradise Lost”] .

Ultimately, even after Garcia was gone, Ruelas was able to cope and move on with his career. And indeed, he has [coped and moved on with his career].


To form the intended „whole” discourse segments can beconnected in a limited number of ways coherence

there exist linguistic devices that make structure explicit

identity (sameness): that is, that is to say, in other words, ...

opposition (contrast): but, yet, however, nevertheless, whereas, in contrast...

addition (continuation): and, too, also, furthermore, moreover, in addition,...

cause and effect: therefore, so, consequently, thus, it follows that, ...

concession (willingness to consider the other side): admittedly, true, I grant,...

exemplification (shift from general/abstract to specific/concrete idea): for example, for instance, after all, an illustration of, indeed, in fact, specifically,...

discourse comprehension consists of recognizing the structure


Coherence vs. Cohesioncoherence: structural, functional relations between sentences

cohesion: non-structural, text-forming relations that “tie” parts of discourse together

John went to his favourite music store to buy a piano.

He was excited that he could finally buy a piano.





He had frequented the store for many years.



It was a store John had frequented for many years.








He arrived just as the store was closing for the day.




It was closing just as John arrived.




When Teddy Kennedy paid a courtesy call on Ronald Reagan recently, he made only one Cabinet suggestion. Western surveillance satellites confirmed huge Soviet troop concentrations virtually encircling Poland.(Hobbs 1982)

E: Forks have windows.P: Yes they do. Augmented pretension. Four plus four equals sixteen. It is a larger element, it’s photographic and phototrophic, but it is a higher number, higher course-work. It grows through evaporation or nocturnalism, it is sleepy, you rediscover it and I suppose forks could have windows through evaporation.




Discourse modeling: intentional approach

discourse participants have certain goals (agendas) to achieve

utterances : actions that realize the intentions

speaker’s plan wrt. communicating intentions ties the discourse

together

discourse understanding : recognizing speaker’s intentions


Intentional Approach (Grosz and Sidner 86)

three dimensions of discourse

linguistic structure : the utterances

intentional structure : hierarchy of intentions (communicative goals)

attentional structure : model of objects, properties and relationsthat are salient at each point in discourse

(dynamically changing)



Linguistic structure

discourse segments + relations that hold between them



Linguistic structure

discourse segments + relations that hold between them

(para-) linguistic expressions reflect discourse structure

cue phrases, aspect, tense, intonation, gesture

discourse structure constraints discourse interpretationanaphora resolution



Attentional structure

participants’ focus of attention (what is „attended to”)

modeled by focus spaces: objects and relations in focus

changes: insertion and deletion rules



Intentional structure

Discourse Purpose (DP)

purpose/intention held by discourse initiator e.g. make hearer:

intend to perform a task,

believe a fact,

believe that one fact supports another fact,

identify an object,

identify a property of an object

assumption: one per discourse




Discourse Segment Purpose (DSP)

how given segment contributes to DP




Hierarchy of intentions

dominanceDSP1 dominates DSP2 if satisfying DSP2 is intended to provide part of satisfaction of DSP1

precedence

DSP1 precedes DSP2 if DSP1 must be satisfied before DSP2


Discourse modeling: functional approach

relations between discourse units

relations may be made explicit by linguistic cues

model: domain-independent rhetorical structure

compositionally built discourse tree


Rhetorical Structure Theory (Mann and Thompson 87)

nucleus(N) vs. satellite(S) segments core vs. peripheral part of the message

„nuclearity principle”

relations defined in terms of:constraints on the nucleus

constraints on the satellite

constraints on the comination of N and S

effect achieved on the text receiver

„classical RST”: 24 relations, (Mann, 2005): 30 relations



example relations

Elaboration: set/member, class/instance/whole/part…Contrast: multinuclearCondition: S presents precondition for NPurpose: S presents goal of action in NSequence: multinuclearResult: N results from something presented in S



Evidence: S provides evidence for what N claimsconstraints on N: Reader might not believe N to a degree satisfactory to Writer

on S: R believes S or will find it credible

on N and S: R's comprehending S increases R's belief of N

effect of W: R's belief of N is increased



Evidence: S provides evidence for what N claimsconstraints on N: Reader might not believe N to a degree satisfactory to Writer

on S: R believes S or will find it credible

on N and S: R's comprehending S increases R's belief of N

effect of W: R's belief of N is increased

[ George Bush supports Big Business. ]N

[ He is sure to veto House Bill 1711. ]S



(volitional) Cause: S presents a cause that motivates Nconstraints on N: N is a volitional action or else a situation that could have

arisen from a volitional action

on N and S: S could have caused the agent of the volitional action in N to perform that action; without the presentation of S, R might not regard the action

as motivated or know the particular motivation;N is more central to W's purposes than S.

effect of W: R recognizes S as a cause for the volitional action in N



(volitional) Cause: S presents a cause that motivates Nconstraints on N: N is a volitional action or else a situation that could have

arisen from a volitional action

on N and S: S could have caused the agent of the volitional action in N to perform that action; without the presentation of S, R might not regard the action

as motivated or know the particular motivation;N is more central to W's purposes than S.

effect of W: R recognizes S as a cause for the volitional action in N

[ George Bush supports Big Business. ]S

[ He is sure to veto House Bill 1711. ]N


Problems with RST


Problems with RST (cf. Moore and Pollack 92)

how many Rhetorical Relations are there?

how can we use RST in dialogue as well as monologue?

how to incorporate speaker’s intentions into RST?

RST does not allow for multiple relations holding between parts of a discourse

RST does not model overall structure of the discourse


Computation of discourse coherence

grammar-basedanalogous to sentence grammar: encode RRs as rules, parse (Polanyi)

inference-basedproof-system: encode RRs as axioms, prove coherence, e.g. by abduction (Hobbs et al.)

plan-basedencode RRs as plan operators, instantiate plan given disourse goal (Litman&Allen)

shallow rules:schemata/templates, lexical clues (Marcu)


Automatic identification of rhetorical structure (Marcu 99 and later work)

parser trained on a discourse treebank– 90 hand-annotated rhetorical structure trees

– Elementary Discourse Units (EDU) linked by Rhetorical Relations (RR)

– parser learns to identify N and S and their RR

– mainly shallow features: lexical, structural, Wordnet-based similarity

discourse segmenter (to identify EDUs)– trained to segment on hand-labeled corpus (C4.5)

– mainly shallow features: 5-word POS window, presence of discourse markers, punctuation, presence/absence of particular syntactic items

– 96-8% accuracy


Automatic identification of rhetorical structure(Marcu 99 and later work)

evaluation of Marcu’s parser

hierarchical structure easier to identify than rhetorical structure

recall precision

EDU identification: 75% 97%

hierarchical structure (related EDUs): 71% 84%

nucleus/satellite labels: 58% 69%

rhetorical relation: 38% 45%


Dialog


Dialog

linguistic properties (cohesive devices)

structure manifested in the dialog partys’ contributions

speech-related phenomena:pauses and fillers („uh”, „um”, „..., like, you know,...”)

prosody, articulation

disfluencies

overlapping speech

spontaneous vs. „practical” dialogs

topic drifts vs. goal-orientedness


Dialog

dialog is made up of turnsspeaker A says sth, then speaker B, then A...

how do speakers know when it’s time to contribute a turn?

there are points in dialog/utterance structure that allow for a speaker shift

Transition-Relevance Points (TRP)

e.g. intonational phrase boundaries


Dialog

dialog is made up of turnsspeaker A says sth, then speaker B, then A...

turn taking rules determine who is expected to speak next

at each TRP of each turn:

if current speaker has selected A as next speaker, then A must speak next

if current speaker does not select next speaker, any other speaker may take next turn

if no one else takes next turn, the current speaker may take next turn


Dialog

some turns specifically select who the next speaker will be

adjacency pairs

regularly occuring, conventionalized sequences

conventions introduce obligations to respond (and preferred responses)

greeting : greeting question : answer

complement : downplayer accusation : denial

offer : acceptance request : grant

set up next speaker expectations (‘significant silence’ dispreferred)


Dialog

entering a conversation we (typically) have a certain intention

paradigmatic use of language: making statements...

...BUT there are also other things we can do with words

e.g. make requests, ask questions, give orders, make promises, give thanks, offer apologies

aspects of the speaker's intention:the act of saying something, what one does in saying it (requesting or promising)how one is trying to affect the audience


Dialog: speech acts

certain actions we take in communication are designed to get our interlocutor(s) to do things on the basis of understanding of what we mean

doing things with words: Austin, 1962, later Searle, Davis speech acts

utterances are multi-dimentional acts that affect the context in which theyare spoken


Dialog: speech acts

dimensions

locutionary act: uttering something with a certain „meaning”

illocutionary act: act performed by means of uttering the words utterance’s „conventional force”

perlocutionary act: what is brought about as a result (intentionally or not)

how hearer is affected: convincing the hearer,

persuading, surprising, making sad, laugh, etc.


Dialog: speech acts

examples of illocutionary acts

assertive: get H to form or attend to a belief; e.g. „claim” „conclude”

directive: get H to do sth; e.g. „order”, „request”, „beg”

commissive: S commits to doing sth; e.g. „promise”, „plan”, „vow”, „bet”

expressive: S expresses a psychological state, feeling twrd. H „thank”, „apologize”, „hate”, „love”

declarations: S changes the state of the world; e.g. „resign”, „fire”, „name”, „baptize”, „pronounce husband and wife”


Dialog: joint activity

when entering a conversation, we pressupose that there exists certain shared knowledge common ground

introduced by Stalnaker (1978) based on older family of notions: common

knowledge (Lewis, 1969), mutual knowledge or belief (Schiffler, 1972)




stock of knowledge taken for granted, i.e. assumed to be known both by the Speaker and the Hearersum of their mutual, common or joint knowledge, beliefs, and

suppositions

sources of the assumptions:evidence about social, cultural comunities people belong to, academicbackgrounds, etc. (communal common ground)

direct personal experiences (personal common ground)




What does it mean „You and I (mutually) know that p”?





I know that p You know that p






I know that you know that p You know that I know that p






I know that you know that p You know that I know that p

I know that you know that I know that p You know that I know that you know that p

...ad infinitum...



communication relies on collaboration

Gricean Cooperative Principle + principles of rational behaviour

cooperatively interpret and contribute



communication relies on collaboration

Gricean Cooperative Principle + principles of rational behaviour

cooperatively interpret and contribute

crucial: establishing shared knowledge (adding to common ground)

grounding


Dialog: grounding

levels of interpretation of performed communicative act:channel: S executes, H attendssignal: S presents, H identifiesproposition: S signals that p, H recognizes that pintention: S proposes p, H considers p


Dialog: grounding


the Hearer must ground or acknowledge Speaker’s utterance OR signal, at the level that satisfies the Speaker, that there was a problemin reaching common ground


Dialog: grounding


grounding feedback possible at all levels:continued attentionrelevant next contributionacknowledgementdemonstration (e.g. paraphrase, completion)display (verbatim)


Dialog: grounding


problems ...possible at all levels:lack of perceptionlack of understandingambiguitymisunderstanding

clarification and repair strategies


Dialog: grounding


S: I can upgrade you to an SUV at that rate.H gazes appreciatively at S (continued attention)

H: Do you have a RAV4 available? (relevant next contribution)

H: ok / mhmmm / Great! (acknowledgement/backchannel)

H: An SUV. (demonstration/paraphrase)

H: You can upgrade me to an SUV at the same rate? (display/repetition)

H: I beg your pardon? (request for repair)


goal-oriented conversational systems

challenges:

need to understand

interpretation context-dependent

intention recognition

anaphora resolution

people don’t talk in sentences...

user’s self-revisions

dialog systems


goal-oriented conversational systems

how:

interactions in a limited domain

prime users to adopt vocabulary the system knows

partition interaction into manageable stages

let the system take the initiative (predictability)

dialog systems


example tasks:

retrieve information information-seeking dialogue

seek to satisfy constraints negotiation dialogue

perform action command-control dialog

collaborate on solving a problem problem-solving dialog

instruct tutorial/instructional dialogue

dialog systems


modality: type of communication channel used to convey or acquire information

natural-language: spoken or textual keyboard-based or both

pointing devices

graphics, drawing

gesture

combination of one of more of above (multi-modal systems)

dialog systems


turn-taking and initiative strategies

system initiativeS: Please give me your arrival city name.

U: Baltimore.

S: Please give me your departure city name….

user initiativeS: How may I help you?

U: I want to go from Boston to Baltimore on November 8.

mixed initiative

S: How may I help you?

U: I want to go to Boston.

S: What day do you want to go to Boston?

dialog systems


ResponseGeneration

Automatic SpeechRecognition

Spoken LanguageUnderstanding

DialogManagement

data,rules,

domain reasoning

Speech

Action

Words spoken

Bill: I need a flight from Washington DC to Denver roundtrip

Meaning

Speech

ORIGIN_CITY: WASHINGTONDESTINATION_CITY: DENVERFLIGHT_TYPE: ROUNDTRIP

getDepartureDate

System: Which date do you want to fly from Washington to Denver?

dialog systems


NLP: grammars, parsers,generation, discourse,

pragmatics

AI: reasoning, communication,planning, learning

human factors: design, performance, usability

speech technology:recognition, synthesis

hello Bill, how may I help you today?

dialog systems


dialog management

control the flow of dialog

when to say something, when to listen (turn-taking), when to stop

update dialog context with current user’s input and output the next

action in the dialog

deal with barge-in, hang-ups

dialog modeling

what is the context

what to say next

goal: achieve an application goal in an efficient way through a series of interaction with the user

dialog systems


`When I use a word,' Humpty Dumpty said, in rather a scornful tone, `it means just what I choose it to mean – neither more nor less.' `The question is,' said Alice, `whether you can make words mean so many different things.' `The question is,' said Humpty Dumpty, `which is to be master – that's all .‘


References

D. Byron. Resolving Pronominal Reference to Abstract Entities. Proceedings of ACL-02, pp.80–87, 2002

B. J. Grosz, K. Sparck-Jones, B. L.. Webber. Readings in Natural Language Processing, Morgan Kaufmann Publishers, 1986

B. J. Grosz and C. L.. Sidner. Attention, intentions, and the structure of discourse. Computational Linguistics 12(3):175–204. 1986

M. Halliday, R. Hasan. Cohesion in English. Harlow: Longman, 1976

J. Hobbs. Towards an Understanding of Coherence in Discourse, in W. Lehnert & M. Ringle (eds.), Strategies for Natural-Language Processing, Hillsdale NJ: Erlbaum, 1982

W. C. Mann and S. A. Thompson. Rhetorical structure theory: A theory of text organization. Technical Report ISI/RS-87-190, USC/ISI, 1987

D. Marcu. A decision-based approach to rhetorical parsing. Proceedings of ACL-99, pp. 365–372, 1999

B. L.. Webber. Discourse deixis: Reference to discourse segments. Proceedings of ACL-88, pp. 113–123, 1988

E.B. White. Letters of E.B. White, ed. D.L. Guth, Harper & Row, New York, 1972 (example sentences with anaphora)

discourse and dialogue

Documents

discourse processing

isolation discourse

theories of discourse

havea discourse

role of discourse structure

delimiting units of

interpretable utterances

theprevious chapters