========================================================================= From: Amy Soller I had a conversation recently with a colleague in which he argued that the principle of least collaborative effort was not useful in tutorial dialog. His argument was based on the idea that a tutor is aiming not to establish a minimal level of understanding, but a maximal level of understanding. I then argued that this principle is perhaps interesting in the sense that a certain level of understanding is necessary (i.e. the level generated with least collaborative effort), but a certain higher level is desired. This higher level might be realized by partners who have achieved the "maximum potential collaborative effort". What are the classs' thoughts on this? --------- From: Antonio Roque "Referring as a Collaborative Process" - in terms of considering how well the principles in section 4 apply to intelligent tutoring system dialogues, as opposed to general conversation: they both seem to hold up well, except that in the principle of mutual responsibility, the student may have a stronger responsibility for understanding than the tutor. This is important because even the best of ITSes will "break" at some point; to some extent the system will always rely on the willingness of the student to work with it. --------- From: Matt Bell Collaborating on Referring Expressions: This paper is good, but (somewhat ironically) I don't get anything outside of its gist due to their referring expressions being unidentifiable to me. I'm not familiar enough with plans and plan generation to be able to fully appreciate this paper, and after reading much of it without gaining in understanding at all, I dispaired and just skimmed through the rest. That said, I find a number of claims made at the beginning of the paper fascinating in that they are somewhat suspicious, to me at least, in that they seem to come out of and lead to a very different conception of languauge than any I am used to comming from a more linguistic background. These are found in the second paragraph on pg. 4: "In order to address the problem that we have set out, we have limited the scope of our work. First, ***we look at referring expressions in isolation***, rather than as part of a larger speach act...Forth, as the input and the output to our system, we use ***representations of surface speech actions, not natural languauge strings***. Finally, although belief revision is an important part of how agents collaborate, we do not explicitly address this." (emph mine) My question: If referring expressions are studied in isolation, and not in language but in some logical formalism, how can we be sure that what we are gathering is at all representative? I note, for example, that they use only two rather clean looking examples taken out of a corpus in order to trace their model. Graham Hirst is good, and after trying and failing to slog through the formalisms Heeman and he have created to handle this I'm convinced that a lot of work has been done here, but find these assumptions troubling and wonder if anyone else did as well. The other observation I make here is that the emphasis seems to be on generation here -- and generation of each individual word in the sentence. This perspective seems quite different from the BDI type models, speech-act models, etc. used earlier. Is this literature motivated by Clark's work turning into a completely different approach to languauge problems? ======================================================= Referring as a Collaborative Process The hypothesis of this paper is that selection of referring expressions is a negotiated activity. One cool aspect of this hypothesis is that it allows for sociolinguistic constraints upon referents -- persons will choose referring expressions that accomplish social as well as task-oriented goals. (If I want to offend or banter with someone, I'll choose a negative slanted term when referring to an object treated in their beliefs, or contrariwise choose more neutral or positive terms for politeness or solidarity). Might this be a way to model politeness for a system? -------- From: Chad Lane Clark & Wiles-Gibbs: The task suits the study very well... the Tangram figures are strange enough to merit verbose descriptions, but look enough like people to offer some obvious grounding for the participants (e.g., praying, ice-skating, etc. - things people do). Is the task "too perfect" for the hypotheses of more words early, less words late? Could there be a task where it worked the other way (less words early, more late)? Doesn't the "minimization of collaborative effort" hypothesis conflict with the principle of mutual responsibility? In other words, an increased desire of CPs to confirm one another's understandings seems to trump the desire to minimize collaboration. Heeman & Hirst: The distinction of no matches vs. more than one match (p.13) is possibly too strong. How would you account for "possible" matches? Just default to a match, and have the speaker refashion? Humans use fuzzy reasoning all the time, it might be a useful computational approach to consider here. When a referring expression fails, it is likely the hearer was still able to garner some useful information from the attempt (e.g., if the #matches < #figures, then some figures were ruled out). Does the Heeman & Hirst model support this idea? -------- From: Stefanie Bruninghaus Heeman/Hirst: While this paper offers a very elaborate solution, I am not sure what problem it addresses. The authors formalize a theory for referring expressions, but I fail to see what that is good for. The authors have not implemented it, and do not suggest that they are. This is probably wise, because the kind of dialogs presented in the paper don't seem to be good candidates for a human-computer dialog. Clark/Wilkes-Gibbs: I am wondering whether one should also consider non-verbal information - pitch, speed, hesitation, and most importantly gestures. In particular for references, many people (most notably Italians!) use their hands to convey additional information. Beyond pointing, hand motions could indicate that the speaker is just looking for the right words and knows what he wants to say, or whether the speaker only has a vague idea. (One can say, for instance, "That woman with the red hair" to mean "I just can't remember her name", or "I vaguely recall there was a female" depending on pitch and hand movements.) Also: The collaborative process in this paper seems to be limited to the director/listener scenario - but aren't there also situations where both participants can contribute? I am thinking of bird watching (yeah, it's spring), when two people try to locate and then identify a bird, by detecting its fieldmarks. There's often a lot of back-and-forth involved, which reminds me somewhat of contract negotiations - with cycles of offer - rejection - counterproposal. And, then there's also a whole group of birders trying to locate and identify that one elusive bird, ie. more than two people involved in the collaboration. The Clark model doesn't seem to cover birding ;-) -------- From: Mihai Rotaru Clark paper: I am little bit confused on how their referring theory can be integrated with contribution theory. Even so the referring theory should be a refinement of the contributions theory it does not seem to be part of it. Or at least the authors did not try to make any link between them.