Mapping natural language sentences to semantic graphs
As statistical natural language processing systems have progressed to provide deeper representations, there has been renewed interest in graph-based representations of semantic structures, and in algorithms to produce them. In this talk, we present three different approaches that take as input a sentence, and produce a semantic graph representation of the sentence as the output. We make an analogy of semantic parsing to the task of phrase-based machine translation and present a synchronous-graph-grammar based approach that uses a hyperedge replacement grammar (HRG) to analyze the semantic graph side. We further describe a neural sequence-to-sequence model which treats the target semantic graph representation as the output sequence. As a third approach, we propose a cache transition system that generalizes standard transition-based dependency parsing techniques to generate a graph rather than a tree, where a cache with fixed size m is included to analyze different classes of semantic graphs. We evaluate these approaches on a semantic graph representation called Abstract Meaning Representation (AMR) and empirical results show that these approaches achieve competitive results in comparison with other systems.
This talk will be live-streamed.