kornai/pymachine

Eilenberg machines for computational semantics

pymachine's Issues

create gold machine representations for a sample set of sentences

We should get started on this, maybe for words in the Mr. Hug story.

def. parser bug: round brackets get multiple types of treatment, why?

case in point:
forget: lack(remember), before(know)
@Eszti @makrai FYI

Support multiple instances of machines for contexts

So that links created based on some input will only affect the relevan context(s), not the global Lexicon

kornai: Would like two-level context:: level zero - stable knowledge level - document/paragraph-wide knowledge. Maybe separate issue (not sure)

must handle pointers to other headwords in 4lang

This is a serious bug, e.g. society is now parsed as "society/2285" instead of as "group, MEMBER, HAS rule"

need to handle machines with identical printnames

• In 4lang we do several things wrong: we do not handle pointers to other headwords (see #1), but this comes up a lot because we always use the last occurence of a printname (an old issue comment in machine suggests this has been fixed but in fact it hasn't, try "society",), which should at least be changed to using the first occurence.
• In Longman we simply use the first definition of each word and it'll require more discussion before we can do anything smarter
• In the long run we shouldn't have too many of these multiple entries (since we want to derive polysemy), but we'll always need to do something about e.g. yard ('udvar') and yard ('0.914m')