Add some features and bump oracle on turk_2k data

stanfordnlp / sempre-plot

Semantic Parser with Execution

Other

28 stars 12 forks source link

Closed robinjia closed 7 years ago

robinjia commented 7 years ago

Main changes:

Add floating cross-unigram features between utterance tokens and paths & value types
Generate true/false/0.0 automatically, even when joining on anypath (so we don't have to enable star-star joins)
Strip trailing "s" and punctuation ("s" thing is a hack to handle things like "change this to squares")

Also see corresponding pull request in plot-data

robinjia commented 7 years ago

Turns out just deferring to the current lemmatizer works fine, so I undid some changes in JsonFn and VegaFn