STATISTICAL RELATIONAL LEARNING AND SCRIPT INDUCTION FOR TEXTUAL INFERENCE
Abstract
Deep, logic-based approaches and statistical, weighted approaches to understanding natural-language text are often viewed as alternatives. However, they are complementary in their strengths. Logic-based approaches can draw inferences from complex, nested sentences. Statistical approaches can judge semantic similarity, and can learn highly useful regularities from large amounts of data including inference rules encoding probabilistic common-sense knowledge. In this project, we have advanced statistical methods for learning common-sense knowledge and for identifying entity relations in text, and we have integrated logical and statistical methods to induce and effectively utilize probabilistic knowledge for appropriate, accurate inferences when comprehending documents. We have developed four different algorithmic components for aiding relation extraction and textual inference Distributed Markov Logic Semantics, Learning Bayesian Logic Programs for Textual Inference; Stacking for Relational Extraction and Statistical Script Induction.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 2017
- Accession Number
- AD1044908
Entities
People
- Raymond Mooney
Organizations
- University of Texas at Austin