Naturalistic Dialog Understanding
Abstract
The contractor will configure a semantic text analysis system whose capabilities focus on language phenomena encountered in dialog. The system will incrementally build up a semantic analysis, working from left to right. Residual ambiguities that remain after the processing of a given word will be retained, waiting to be informed by subsequent input. The system will cover such advanced capabilities as lexical and referential disambiguation, the treatment of fragments and ellipsis, the analysis of noun-noun compounds, and the detection of superfluous utterances that need not be fully analyzed or stored to agent memory. The output of semantic analysis will be text meaning representations (TMRs) grounded in a formal ontology. This work will offer an essential bridge between raw text and the types of interpreted knowledge structures that have long been waited for by the automatic reasoning community. The system will be evaluated by comparing its automatically generated TMRs against manually crafted gold-standard TMRs, and by analyzing what the system can do in comparison with select existing systems. Although all functionalities will be domain independent, dialogs in the domain of first aid will be used for system demonstrations.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 03, 2016
- Source ID
- N000141612118
Entities
People
- Marjorie Joan McShane
Organizations
- Office of Naval Research
- Rensselaer Polytechnic Institute
- United States Navy