A hybrid approach toward Natural Language Understanding

• Event: Seminar
• Lecturer: Daisuke Bekki
• Date: 09 December 2020
• Duration: 2 hours
• Venue: Gothenburg

In this talk, I will introduce and discuss our recent development of the natural language understanding (NLU) system under a “hybrid approach” between formal linguistics and machine learning, one of the alternative approaches to fully-neural NLU. I will mainly focus on the end-to-end RTE system “ccg2lambda”, which we developed in 2015, that integrates CCG parsers, formal semantics, higher-order theorem provers, and machine learning. I will also discuss some extensions of ccg2lambda to the tasks such as visual-textual entailment and semantic textual similarity.