While semantic parsing receives a long-standing interest from the community, developing robust semantic parsing algorithms remains a challenging problem. In this talk, we will consider the following challenges in semantic parsing: 1) representing the semantics of multiple natural languages in a single semantic analysis; 2) developing parsing systems for broad-coverage semantics; 3) designing unifying parsing paradigms to support distinct meaning representation frameworks; and 4) training systems with limited amounts of labeled data. We approach semantic parsing as sequence-to-graph transduction problems, and introduce novel algorithms/components into transductive settings that extend beyond what a typical neural machine translation system would do on this problem. Our approach achieves the state-of-the-art performance on a number of tasks, including cross-lingual open information extraction, cross-lingual decompositional semantic parsing, and broad-coverage semantic parsing for Abstract Meaning Representation (AMR), Semantic Dependencies (SDP) and Universal Conceptual Cognitive Annotation (UCCA).

Lunch will be served.