Learning Joint Semantic Parsers from Disjoint Data
This addresses a challenge in natural language processing for researchers and practitioners by enabling joint learning from disjoint datasets, but it appears incremental as it builds on existing baselines.
The paper tackles the problem of learning semantic parsers from multiple datasets with different semantic formalisms and no overlapping data, by treating unobserved formalisms as latent variables. It shows improvements in frame-semantic parsing and semantic dependency parsing through joint modeling, though no concrete numbers are provided in the abstract.
We present a new approach to learning semantic parsers from multiple datasets, even when the target semantic formalisms are drastically different, and the underlying corpora do not overlap. We handle such "disjoint" data by treating annotations for unobserved formalisms as latent structured variables. Building on state-of-the-art baselines, we show improvements both in frame-semantic parsing and semantic dependency parsing by modeling them jointly.