:::::::::::::: misc.diane :::::::::::::: Time: 60 to 90 minutes:::::::::::::: misc.heeman :::::::::::::: Here are my comments about coding the IU's from the fixed CGUs for verbmobil. The 'suggest meet, ask days' is in the same CGU. Makes it hard to capture the structure in IU's. In fact, this was one of the claims of CGU's: ``CGUs do not have to correspond to a single main dialogue act''. This makes CGU's inappropriate for IU. In your markup, you don't have ``that sounds pretty good'', starting a new CGU. This is more than an acknowledgment. This is showing agreement, and it is definitely adding new information, namely that the speaker is in agreement. THis is a lot different than the ``mm-hm''s that might simply function to acknowledge understanding, which I agree, don't need to start a new CGU. The maptask dialogue was certainty interesting and quite complex from a grounding standpoint. One thing it demonstrates is that the grounding utterance, such as a 'oh okay', might not be grounded itself (i.e. due to overlapping speech). When the speaker does eventually get control, they might ground the utterance. .i.e. 21.1 can be seen as grounding 18.1, especially since 19.1 was not successful. How do we indicate this with the current scheme? We might want to distinguish utterances as the `grounding' utterance for a CGU (and not have it as part of the CGU). This definitely needs work. :::::::::::::: misc.jencc :::::::::::::: Verbmobil: 1. I wasn't sure what to do with CGU #16. Officially B.20.1 was never acknowledged, so it really was never grounded. On the other hand, the participants both ended the conversation, so although it was not grounded verbally, it was grounded by action. Maptask: 1. I'm not sure what to do with dialogue turn 32 where F answer's G's question in the first part of the turn (F.32.1 and F.32.2) and then goes on to the next topic (F.32.3 and F.32.4). G's response in turn 33 explicitly acknowledges the latter part of turn 32, but I think it also implicitly acknowledges the first part of that turn. :::::::::::::: misc.traum :::::::::::::: 1) Verbmobil I did IU analysis *much* later than CGU analysis. IU analysis took approx. 10 minutes, after setting up the files and listening to the dialogue. - combination of overall goal suggestion and day question in same cgu forces this to be put at a lower level, wheras 1.2, if it were by itself would be placed under iu.1 2) maptask (about 30 minutes - a little over an hour real time, but with interruptions because of machine problems, etc.) -7.2, unsure whether to put this with 7.1 (as elaboration on the ack with evaluation), or as it's own (unacknowledged cgu), or, just leave out - chose the latter. 13.2 - should this perhaps be with 13.1, as one utterance? In any case, in some ways, cgu6 should be split, because the local resident info is not grounded and picked up with 16.1 cgu10 is a tough one it's all one unit, since they keep repairing location, the overlap makes it hard for them to converge that they have grounded this info. 29.1 is cut off, can't tell if it belongs to cgu12 or is initiating something new. cgu13 - in some ways there's argument for cutting off at 32.2 with a cgu about identifying the babbling brook (actually pulling this grounded part out of the larger question). Alternatively, see the intervention at 30.3 as a repair, and all within same unit, not fully grounded until 33.1 :::::::::::::: misc.venditti :::::::::::::: vermobil CGU labs: approx. 30min IU labs: approx. 15min data prep: approx. 1hr re-reading manual: approx. 30min ### CGU comments/notes ### Many of speaker B's utterances seemed like she was talking to herself and did not contribute to establishing a common ground. These utterances were not what was eventually acknowledged by speaker A. (What was acknowledged was the utterances that B offered which were relevant to making the meeting plan.) So, following (18) and accompanying explanation in the manual, I left these utterances out of my CGU coding. From looking at the fixed CGU labs, I see that the utterances that I dubbed "talking to herself" or "irrelevant for establishing commom ground of meeting plan" are actually *included* in the CGU analysis here. If this is the preferred strategy, then I will need a bit more clarification in the manual on what does and does not contribute to the common ground. For teaching/training purposes, I find it a good idea to have a number of exercise dialogs for practicing CGU (or IU) labeling, then to provide sample analyses (such as the fixed CGU labs here). There were two of these offered in the manual. But what I find most helpful is NOT a list of utt numbers associated with each CGU, but rather a list of utterances in that CGU, like: 2 "on vacation next week" B.2.1 actually next week I am on vacation A.3.1 gosh 3 "ask following monday" A.3.2 I guess we will have to meet the week after that A.3.4 how 'bout Monday B.4.1 uhh Monday the tenth A.5.1 uh-huh I made a representation like this for both 1) my CGU labs, and 2) the fixed labs. This way, it is quite easy to compare different strategies among the two. I find it impossible to read only the utt numbers, then try to go back and figure out which utts in the text these numbers refer to, then try to piece those together to understand why I/they labeled what I/they did. When I compare the two CGU labelings using this format, I see a few differences that it would be good to talk about at some point. 1) As I mentioned, I took example (18) and explanation in the manual to the extreme, and left out all of B's utts that were irrelevant for the current plan. The fixed labs did not do this. If I were to keep these utts in, I think I would have had very similar CGUs to the fixed labs, at least in this respect. 2) In CGU 1, I left the two "corrected" utts (A.1.3 and A.1.6) out of my CGU. I suppose, according to (6) in the manual, my strategy was incorrect here. 3) I have the fixed CGU 3 broken down into two CGUs 3 & 4 : "propose alternative day - Monday" and "clarify which Monday". Either analysis still seems acceptable to me. 4) In my CGU 14 (fixed CGU 13), I have both B.16.1 and B.16.2 as part of the acknowledgement. The fixed analysis leaves out B.16.2. The reason for including both is that I seem to remember the manual stating that an acknowledgement can take more than one utterance. To me, B.16.2 is still part of the acknowledgment, and is not similar to no-content utterances like "umm", etc. 5) The last chunk is tough. I have the fixed CGU 15 broken up into two CGUs (my 16 & 17). Even though the main task of planning the meeting is done, it seems like they start up a new sub-topic regarding the nature of "roger over and out", for which they do start to build some kind of common ground. ### IU comments/notes ### Again, it is difficult to figure out how much embedding is involved. I feel that my structure is pretty flat, but I wasn't sure what sub-IUs I would want. The CGU labeling is too coarse, at least in two cases: CGU1 introduces the dialogue ("let's meet"), and then jumps right into planning the day. This is two IUs. Also, the final CGU 15 not only says "ok, we are done with this goal of arranging a meeting", but it also briefly opens up another sub-topic about the nature of "roger over and out". These are two different highest-level IUs to me. **I would have liked to look at other people's IU analyses for trains/toot, to see if my analyses were on the right track. I got a lot out of comparing my CGU analyses with the others on the web page, and I would have liked to see the IU analyses posted there too.