The ability to ground conversational referents is a key requirement
for human dialogue. This process, known as reference
resolution, has received much attention from both psycholinguists
seeking to understand how humans process language
and computer scientists seeking to improve the performance
of language-capable agents. However, the majority of previous
research has focused on what we term closed-world reference
resolution, in which the set of possible referents is assumed
to be known a priori. In this paper we present a domainindependent
model of open-world reference resolution which
appropriately handles uncertain knowledge, and the results of
an empirical human-subject experiment conducted to verify
the model’s predictions.
