Discussion at SuquamishTop on shared scripting libraries.

There may be some duplication of code it would be nice to reduce.

Things I know of:

  • MT experimentation scripts (Eric)
  • MT rule building (Petter)
  • MT grammar subset selection (Petter)
  • profile2db scripts (Francis)
  • MRS comparison code (Rebecca)

Things I think I heard:

  • MRS to html
  • Dependency to nltk

“Minutes”

Is it possible to make changes to the MRS structure, given enough reason? Laurie says Ann says yes.

Emily: shouldn’t stop us from documenting what exists now.

Is there a final canonical publication describing it? Stephan: yes. but it may not be “final.”

Dan: even if MRS doesn’t exhaust what we want to say about meaning… been talking about adding room to put constraints on coreference between individuals, e.g. “ICONS” like “HCONS”, like “x7 is coreferent to x9”, or “x3 is NOT coreferent to x10”, from binding principals. want to implement that. BUT, I don’t think that will be a showstopper; other implementations can gradually assimilate an extra attribute. Also we only use ‘qeq’ in HCONS currently, but some people may want to use ‘geq’. I support having multiple implementations of this formalism.

Francis: I’ve noticed there’s been some reimplementation of things in Perl or Python; scripts for batches (mixes of Lisp and other stuff); scripts for building and training models; scripts for profile manipulation. I suspect work done in place A could be useful for people in place B. E.g. tools for pulling bits and pieces of information out of TSDB profiles. Or renumbering profiles.

Emily: How about starting with IGT and building a TSDB skeleton?

Is there a list of the tools and what they do, for people new to the community?

Rebecca: Come to the meeting.

Glenn: We talked about worrying that MRS functions depending on Lisp as a problem; what about the tokenizer for the ERG?

Stephan: Hey, Lisp is great, it’s not a problem. But ok, having tools in other languages is nice too.

Rebecca: Releasing a C++ REPP library. With deterministic characterization. Working on integrating to PET. Everyone: document it!

Most useful if we can integrate TNT also.

Dan: Can we have Chart Mapping in the LKB? Stephan: Good idea, go do it.

Stephan: REPP is a good example of evolution. Been OK for a while, but characterization always been a slight question mark. Re-implementations can trigger resolution of old problems.

Dan: Talking about python and other tools… the lex type db is fabulous. Rebecca: I find it useful too.

Francis: Good example of something written in perl currently that could be used by other people too. Dan: Scalability to wikiwoods? Francis: yes. aside from writing a page for every tree. would be great if we could call MRS->HTML or Tree->HTML easily in a lightweight fashion.

Francis: Who still uses Perl? Who uses Python? (around the table, mostly python… UW… but many perl users) Stephan: I use Clozure

Francis: Where next? /logon/scripts? Stephan: DELPH-IN SVN repository, perhaps Francis: First step to collaborating is making the code available. Put it in SVN and also mark it on the Wiki. And announce it on the Developers list. Stephan: Everyone, get your acts together about marking up on the wiki!

Name of tool? What use cases are people trying to fill?

Rebecca: See ToolsTop – there is some stuff there.

Rebecca: I put up C++ training code. And EDM code (on any MRS simple string). Emily: Sign up to be informed when the ToolsTop page changes.

Last update: 2011-06-28 by WoodleyPackard [edit]