Notes

Argumentative Discourse Notes

I removed this section, because I think it's easier (for now) to steer clear of modelling agent's preferences/beliefs as much as we can. We can still classify dialogue acts as proposing/disagreeing etc.:

Next we describe the roles played and attitudes taken by participants toward the Issue related concepts.

Some Examples

John saying "hello" to Matt:

• Signal is the acoustic signal, with Medium of Air (or is this the Carrier?)
• Transmit has that Signal as its object; that Medium as base; John as the agent; Matt as the recipient
• Embody is Speak, has agent John, object is the Message (see below) and result is the Signal
• Sense is Hear, has agent Matt, object is the Signal and result is the Message ...

• Message is the word string "hello". It may have several other annotations (PoS? tag, parse tree, semantic LF etc). Its base is a Language, SpokenEnglish.
• Convey has agent John, object the Message, base the Signal.
• Express is GenerateSpeech, with agent John, object the Information and result the =Message.
• Interpret is InterpretSpeech, with agent Matt, object the Message and result the Information.

• Information is the propositional content of 'hello', which I guess isn't very much (a 'greeting' in fact).
• Communicate is of type Greet, with agent John, recipient Matt and object the Information.

John pointing at Matt (as PART of an overall communicative act):

• Signal is the visual signal, with Medium of Light (or is this the Carrier?)
• Transmit has that Signal as its object; that Medium as base; John as the agent; everyone(?) as the recipient
• Embody is Gesture, has agent John, object is the Message (see below) and result is the Signal
• Sense is See, has agent everyone(?), object is the Signal and result is the Message ...

• Message is the gesture "point(Matt)". It may have several other annotations (arm, hand etc). Its base is a Language, Gesture.
• Convey has agent John, object the Message, base the Signal.
• Express is GenerateGesture, with agent John, object the Information and result the =Message.
• Interpret is InterpretGesture, with agent everyone(?), object the Message and result the Information.

• Information contains the propositional content of 'point(Matt)', which I guess is the proposition that Matt is being referred to (perhaps by a corresponding linguistic expression). It may contain other propositional information from speech, writing etc.
• Communicate has agent John, recipient Matt and object the Information. We don't know its type until we know more about the Information.

Ballim and Karakapilidis Model

Knott and Sanders Coherence Relations

DAMSL (Core and Allen) Model

• CommunicativeStatus?
  • Uninterpretable
  • Abandoned
  • SelfTalk?
• Information
  • TaskInformation?
  • TaskManagementInformation?
  • CommunicationManagementInformation?
• ForwardLookingFunction?
  • Statement
    • Assert
    • Reassert
    • Other-statement
  • Influencing-addressee-future-action
    • Open-option
    • Action-directive
  • Info-request
  • Committing-speaker-future-action
  • Offer
    • Commit
  • Conventional Opening Closing
  • Explicit-performative
  • Exclamation
  • Other-forward-function
• BackwardLookingFunction?
  • Agreement
    • Accept
    • AcceptPart?
    • Maybe
    • RejectPart?
    • Reject
    • Hold
  • Understanding
    • SignalUnderstanding?
      • Acknowledgement
      • RepeatRephrase?
      • Completion
    • SignalNonunderstanding?
    • CorrectMisspeaking?
  • Answer
  • InformationRelation?

A First Shot at a Content/QUD-Based Model

(see Larsson thesis (IBis, GoDiS? ), Ginzburg)

Model both IUN and QUD as Questions. IUN is the long-term topic (the Issue related to the current thread = minor topic) used to relate argumentative structure to. QUD is the short-term question used for answer interpretation, ellipsis resolution.

For QUD, we have these categories:

• y/n questions (proposition with an empty lambda-abstracted set - truth value of proposition undetermined)
• wh- questions (proposition with a non-empty lambda-abstracted set)
• alternative question (sets of y/n questions)

Here, 'content' means Information I think (propositional content of the move). A=assert(P) means A is a move (Communicate event) whose content (Information) is the proposition P. Both preconditions and update effects are assumed to be inherited by subtypes from supertypes.

?P is the polar question 'whether(P)', the question of whether the proposition P holds. ?X.P is the wh-question 'which X Ps', the question of for which X P holds (where P contains X). & is supposed to be logical conjunction. PTT means 'Poesio-Traum Theory' i.e. the kind of theory which explicitly models whether or not an utterance has been grounded by other participants - I don't think we'll want to do this at this stage, but quite possibly later.

Our Suggested PA Categories

(not likely that all will be technically feasible - just a list of what appear to be useful from a single CMU meeting video)

• WHOLE BODY
  • movement
    • stationary
    • walking (to/from X)
    • rolling chair (to/from X)
  • location
    • near X
    • next to X
    • in front of/behind X
• TORSO
  • body pose
    • leaning forwards (towards X)
    • leaning backwars (away from X)
    • turning (towards/away from X)
    • seated/standing
    • sitting down/standing up
    • facing towards/away from X
    • bending down
    • shifting weight
• HEAD/FACE
  • head
    • nodding (at X)
    • shaking (at X)
    • tilting
    • facing towards/away from X
    • turning to face X
    • leaning on 1/2 hands
  • face
    • smiling (at X)
    • frowning (at X)
    • surprised
  • eyes
    • glancing at X
    • gazing at/fixated on X
  • mouth
    • speaking (to X)
    • yawning
• ARMS/HANDS
  • pointing
    • deictic finger pointing at X
    • otherwise indicating X
  • hands (no contact)
    • waving (at X)
    • clapping
    • hands to/on/behind head
    • thumbs up/down
    • beckoning X
    • tapping/drumming (on X)
    • (moving) hands together/apart
    • counting on fingers
    • indicating levels
  • hands (contact)
    • touching X
    • holding X
    • holding up/showing X
    • picking up/putting down X
    • pushing/pulling X
    • moving/turning X
    • waving X
    • writing on X
    • drawing on X
    • typing on X
    • passing X to Y
    • playing with X
    • folding down a laptop screen! (often to see what's behind)
  • arms
    • fidgeting
    • scratching
    • stretching
    • shrugging
    • holding up hand (turn request / 'wait!' / emphasis)
    • (moving) arms in
    • (moving) arms out

X is intended to be any person/object, including electronic documents and items of instrumentation (e.g. headsets/mics/laptops)

Ideally, information would be presented as probability distributions over features/categories (or scored n-best lists) rather than just best guesses

-- JohnNiekrasz - 23 Nov 2004