GitXplorerGitXplorer
s

ccg

public
22 stars
1 forks
2 issues

Commits

List of commits on branch master.
Unverified
63ec16edee2616eb423d53710a8600e731b4e17e

Update README.md

ssyllog1sm committed 11 years ago
Unverified
ec3509880805ea12054081347f89b64401705ecb

* Delete test that failed because of absent MU category

committed 11 years ago
Unverified
40b3f713e6ba3c8eb0526c74e5b6170c49b859a6

* Switch to using former develop branch, as it seems to be more recent. 4 tests failing.

committed 11 years ago
Unverified
560a4e3629f67a3275f781a97adf4cffeead7bd6

* Imported updated version from other repo

committed 11 years ago
Unverified
f061791ef92692d64f041689a1ac565a86ed5b7d

* Added hack-up of Treebank code to repo. Could use restyling...

committed 11 years ago
Unverified
61e8541e83d1bd5659714ae5a6e9f06f299d4a03

* Initial commit of old ccg code. May be bad branch, tests are failing.

committed 11 years ago

README

The README file for this repository.
  • Overview

Manipulate Combinatory Categorial Grammar categories and derivations, for natural language processing research.

The library is quite feature rich, but has a pretty messy API, and some bugs.

The "killer feature" is the implementation of the CCG grammar rules and variable binding. After sentence.unify_vars() has been called, all categories will have all slots bound to "global" variables, which are unified to other variable bindings, and may have words attached.

Aside from ugliness, there are two main sources of remaining problems:

  1. Coordination is very difficult to get right with respect to unification, as we need a set of words, and we don't necessarily unify when we coordinate (think "red bus and green train". We do not unify "bus" and "train"!).

  2. When a word is missing from the "markedup" file, we do a terrible job of guessing its annotation.