texttheater / geopar Goto Github PK

It serializes all variables as underscores unless given a TermReader object that knows all these variables.

See test_terms.test_string_roundtrip

Possible fix: no longer use TermReader for storing variable names, keep the name-variable map only during the term reading process, always write terms in a canonical form (with no underscores).

Hi, I'm trying to use your code for chinese semantic parsing. Do you hava any document for generating action sequence? I'm quite confused about the logic.

Some arguments are for dropping/lifting, others are for corefing. Makes no sense to apply one action to the other.

Where should the restrictions be implemented, in parsestacks or in parseitems? Replace stackelement.arg with a more high-level method?

This is currently the main problem on the peripheral_integration branch, leading to an explosion of the search space.

lift should add to conjunctive terms from the left, not right. Need to adapt addresses in StackElement.

Terms dropped into the same argument as a previous one should inherit that term’s secondary stack position, without adding to the secondary stack. Otherwise terms become inaccessible too quickly.