SMILE - SEmantIc roLe Extraction
The research question this proposal addresses is:
- Is it possible to create a Natural Language Understanding (NLU) system that can extract the components of a SPO’s Impact Model, including stakeholders, programs, services, outcomes and indicators, from unstructured text, in an explainable, tractable, governable, and verifiable way?
Our research objectives are to:
- Develop an architecture that can extract Impact Model components from unstructured texts where the components are not explicitly stated but must be inferred from the text.
- Extend existing social services ontologies (Fox et al., 2021a; Fox et al., 2021b; Gajderowicz et al., 2022b; Rosu et al., 2022; Fox et al., 2022), implemented as a knowledge graph, and are capable of storing learned information about Impact Models and responding to queries about learned knowledge.
- Develop metrics for measuring the architecture’s correctness, tractability, explainability, and governance.