25 - Neural Semantic Parsing over Multiple Knowledge-bases

ACL 2017 short paper, by Jonathan Herzig and Jonathan Berant. This
is a nice, obvious-in-hindsight paper that applies a
frustratingly-easy-domain-adaptation-like approach to semantic
parsing, similar to the multi-task semantic dependency parsing
approach we talked to Noah Smith about recently. Because there is
limited training data available for complex logical constructs
(like argmax, or comparatives), but the mapping from language onto
these constructions is typically constant across domains, domain
adaptation can give a nice, though somewhat small, boost in
performance. NB: I felt like I struggled a bit with describing this
clearly. Not my best episode. Hopefully it's still useful.
https://www.semanticscholar.org/paper/Neural-Semantic-Parsing-over-Multiple-Knowledge-ba-Herzig-Berant/6611cf821f589111adfc0a6fbb426fa726f4a9af