DRI Discourse Structure In Dialogue

Discussion Notes on CGU Codings

Verbmobil:

  1. restart-continue: in or out? Manual says borderline. What about issue of content? specific issue: A.1.3, A.1.6
    At the meeting we decided that this kind of speech repair will be pre-marked as being connected and will go together in a CGU
  2. extra info on acks, different style or better evidence - add to prior CGU? case in point a.3.2 , b.16.2
    At the meeting, we decided that only enough evidence to consider something grounded willgo in the prior CGU (modulo decisions about how to mark (different kinds of) acknowledgements, in general)
  3. request repair vs clarification vs repair
    unit 3 - heeman, jencc treat 4.1 as repair request,5.1 as repair, and 6.1 as ack, traum and chn treat 4.1 as repair and 5.1 as ack, diane,owen,venditti,ward treat 4.1 as ack and new initiate (jencc also the latter, to treat the sub-dialogue grounding), if so, does 5.1 need to be an init of answer, e.g., diane, owen?
  4. inclusion of irrelevent material in cgu that is not directly acked - e.g., 6.2,6.3, also 8.2-8.5,14.1-14.2
  5. peter - starting cgus from acceptances, chn treats 13.1 as repair rather than ack/initiate.
  6. oddness at beginning closing of dialogue, due perhap to strange recording situation, leads to disagreements.

TRAINS:

  1. repair around 7.2 - does it go with 7.1, and if so, with 6.1 also? or 9.1? also overlap, 8.1,7.2 unit from 2 people, transcript problems with trains -continuing different tratments of turns 9,10,11.
  2. 19.1, beginning of repair sequence or simple ack with followup question?
  3. multi-utterance answer to question - all or just first as ack (manual says first, but cf. 20.2 repsonses), also 26.1,26.2, answer with elaboration
    The above policy on how much needed to ack as well as eventual decision about marking questions will help decide this
  4. 24.1 - can't leave off answer the same way as acceptance (or can you, cf. Massimo's hypothesis about not needing to ground backward acts. But "and" does seem to ack, or could just be coninuation - maybe prosody helps?
  5. which things don't need to initiate a unit. E.g., various kinds of positive feedback of agreement or yes answer to check question? Lots of peter's replies. Also e.g., 44.1 some contentful answers not seen as initiating, or not seen as implicitly grounded. Several other places, same pattern for Q&A (6.1.4)
  6. turn 43 controversy over cancellations, how to mark them well for agreement - basically traum,heeman,jeanc agree on none of those utterances entering common ground, but marked three diff. ways.
  7. big confusion around turn 47 - 49: overlap? some people identify the same units, but * or don't *. Too tough to come up with consensus.
  8. Peter: look at relation to backward understanding acts, notion of keeping cgu's open - useful at all for speech? use cleaned up speech (speech repairs/disfluencies cleaned up) for CGU coding
  9. Jean: Q&A in same CGU if answer *complex*? -difficult to tell about discourse marker vs. ack
  10. Jencc - 13.2 continues content of 4.2, traum also role of 9.1,11.1

Maptask:

  1. Big Q: did everyone listen carefully to speech - intonation and timing very important here.
    Different strategies: some just listened before, some lots of replaying while listening, some listening for places where transcript is "unintuitive", and listening only to those
  2. traum: overlap 2.1 and 3.1: 3.1 seems to be a repair given lack of immediate response to 1, thus whole thing to 5 seems one cgu; similar phenomenon in 20.1 coming at same time as 19.1. (different subcodings, but all see endings at 5.1, then agree about 5.2)
  3. how to do 7, with ack then accceptance - include both or only one?
  4. 8.1 in next unit? complex answer and explanation, so seems yes), repair which is not complete rstart in 8.2, also 13.2 is negative response after ack of 13.1)
  5. diane9 is strange unit
  6. 11.1 seems like reqrepair not ack, hence joined with next unit
  7. 14.1 as implicit ack of 13.1, also, with 16.1 followup of 10.2, and by implication ack of 13.1, since asking about not-mentioned part.
  8. 16.1 unit should include 14.1, because not complete restart - need q part (perhaps clear from intonation).
  9. 20.1 as repair when ack comes too late.
  10. complete disagreement about where to include 20.2! Starting with owen14: 8 coders and 8 different codings for this unit ending with venditti 14
  11. transcription error in 22.4, really just disfluency
  12. multiple ack? 28.1, included in previous unit or initiating unit?
  13. very different codings for 30.1-32.2

TOOT:

  1. starting new cgu on different topic without ack, e.g., 1.1,1.2 from 1.3
  2. 6.2, extra info, misunderstood - combine?
  3. question about splitting 11 at 11.3 or 11.0 or not at all.
  4. julia has nothing from 14.1 to 21.1 - missing part of file?
  5. repair also counting for ack? 16.2 - don't think it should)
  6. general confusion from 16.1 to 20.1
  7. initiate unit for yes answer 22.1
  8. missing bits in transcript in turn 45 - venditti, owen tried to add uu tokens
  9. new unit at 45.7?
  10. turn 51 ack, 51.1 or 51.2 or both - why split there?

Last updated on June 4th 1998