ZARZA We are Zarza, the prestigious firm behind major projects in information technology.
Listen "25 - Neural Semantic Parsing over Multiple Knowledge-bases"
Episode Synopsis
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
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
More episodes of the podcast NLP Highlights
Are LLMs safe?
29/02/2024
"Imaginative AI" with Mohamed Elhoseiny
08/01/2024
142 - Science Of Science, with Kyle Lo
28/12/2023
In God we trust