Concept Mention Grounding Task

From GM-RKB
Jump to navigation Jump to search

A concept mention grounding task is a mention grounding task that can map a concept mention to a grounded concept mention (with a canonical referencer).



References

1992

  • (Webber et al., 1992) ⇒ Bonnie L. Webber, Norman I. Badler, F. Breckenridge Baldwin, Welton Becket, Barbara Di Eugenio, Christopher W. Geib, Moon Ryul Jung, Libby Levison, Michael B. Moore, and Michael White. (1992). “Doing What You're Told: Following Task Instructions in Changing, but Hospitable Environments." Technical Report, University of Pennsylvania
    • QUOTE: Because multi-clause instruction steps may evoke more than one situational context [56], part of an agent's cognitive task in understanding an instruction step is to determine that situation in which he is meant to find a referent for each of its referring expressions. This is the process of grounding referring expressions. … When given this instruction (or similar ones such as "Open the box and hand me the yellow block", which one of the authors (Webber) has publically tested on several individuals), agents appear to develop an expectation that after they perform the action, they will be in a context in which it makes sense to try to ground the expression and determine its referent.

      The second instruction (". ..and wash out the coffee urn") is different: if the agent sees a coffee urn in the current context, prior to acting, he will happily ground the referring expression "the coffee urn" against that object. If he doesn't see a coffee urn prior to acting, he develops the same expectation as in the first example, that when he gets into the kitchen, he will be able to ground the expression then. The fact that agents will look around when they get to the kitchen if a coffee urn isn't immediately visible, opening cabinets until they find one, shows the strength of this expectation and the behavior it leads to.

      This decision as to the context to use in grounding a referring expression is based on distinguishing the information (and assumptions) used to resolve a referring expression from that used to ground it. Reference resolution precedes reference grounding (cf. Figure 2) and involves using information from the interpretation of the current utterance (i.e., the explicit description), information from the previous discourse (i.e., the existence of a salient discourse entity with that description in the agent's discourse model) [26, 54, 551, and hypotheses about the intended relationship between actions.