Using an Implicit Method for Coreference Resolution and Ellipsis Handling in Automatic Student Answer Assessment
Rajendra Banjade, Vasile Rus, Nobal Bikram Niraula · The Florida AI Research Society · 2015
The automatic student answer assessment problem is challenging because it requires natural language understanding. This problem is even more challenging in conversational Intelligent Tutoring Systems (ITS) because in such conversations the speakers develop common ground as the dialogue proceeds, which means contextual information from previous utterances in the dialogue is heavily relied upon to understand a speaker’s utterances. Different linguistic phenomena should be addressed in order to improve the performance of automatic answer assessment systems in conversational ITS. Two such important phenomena are: references to entities mentioned earlier in the dialogue and ellipsis (i.e., answers with contextually implied parts) . In this paper, we present an implicit approach to resolving coreferences and handling elliptical responses in the context of automatic student answer evaluation in dialogue based intelligent tutoring systems.