davidjurgens / crown Goto Github PK

Hi! I'm running into some problems using CROWN with a popular WordNet library for python.

I've been trying to use CROWN with the nltk WordNet library, but it seems that all of the available versions of the CROWN dataset I found at http://cs.mcgill.ca/~jurgens/crown/ are incompatible with the "newest" official lexnames file (released with WordNet 3.0).

It is my understanding that I cannot generate a new lexnames file using grind without the 45 lexicographer files that are used to create the database files (those files under dict).

Right now I'm trying to generate this file using your library. My guess is that the change needed to include the generated lexnames file in the release only requires a small change to this line:

Is that the case?

I think it would be a good idea to include a new lexnames file in the subsequent CROWN datasets you distribute to better guarantee compatibility with existing libraries.

If you feel the same way, I'd be happy to put in a pull request with the needed change. Anyway, based on what I've read in the NAACL paper, I'm very excited to make use of your library. Thanks!

Feature enhancement to automatically produce this output during the build process to allow for easier integration and transition from WordNet to Crown.

Wiktionary and possibly other resources contain etymology data that may be useful for downstream resources. This information should be extracted integrated into the build process. One option is to produce stand-off files that map synsets to their source language, if known.

Parts of this mapping are already implicitly found during the enrichment process when each EnrichmentProceedure tries to find an existing sense for a given definition and then (correctly) rejects making a new integration when an existing sense is found. However, the sense-mapping data is never explicitly reported, which makes it difficult to enrich these existing synsets with any other information that may be present in Wiktionary or other resources (e.g., domain links, antonym links, or etymological information).

The current GST and NAACL-paper similarity functions are purely surface-form based. It would be good to integrate ADW (https://github.com/pilehvar/ADW) as an option.