Incremental Construction of Robust but Deep Semantic Representations for Use in Responsive Dialogue Systems
Andreas Peldszus, David Schlangen · Publikationen an der Universität Bielefeld (Universität Bielefeld) · 2012
It is widely acknowledged that current dialogue systems are held back by a lack of flexibility, both in their turn-taking model (typically, allowing only a strict back-and-forth between user and system) and in their interpretation capabilities (typically, restricted to slot filling). We have developed a component for NLU that attempts to address both of these challenges, by a) constructing robust but deep meaning representations that support a range of further user intention determination techniques from inference / reasoning-based ones to ones based on more basic structures, and b) constructing these representations incrementally and hence providing semantic information on which system reactions can be based concurrently to the ongoing user utterance. The approach is based on an existing semantic representation formalism, Robust Minimal Recursion Semantics, which we have modified to suit incremental construction. We present the modifications, our implementation, and discuss applications within a dialogue system context, showing that the approach indeed promises to meet the requirements for more flexibility.