Date: 2022-01-21 / 4:00 ~ 5:00 PM
Meaning Representation (AMR) is a graphical meaning representation language designed to represent propositional information about argument structure. However, it fails to adequately represent a number of important semantic features, including number (singular and plural), definiteness, quantifiers, and intensional contexts, which would license inappropriate inferences. Several proposals have been made to improve the representational adequacy of AMR by enriching its graph structure, but are rarely implemented on existing AMR corpora due to the labor costs associated with manual annotation. The goal of this project includes three parts: (1) to improve the representational adequacy of AMR graph structure, (2) to implement the existing proposals through developing an automated annotation tool, and (3) to evaluate different forms of meaning representations including enriched AMR and logical representation in achieving higher parsing accuracy.