The reason: All thesauri I've seen *suck*. Visuwords is actually pretty cool, but even it is lacking some. The more minds working on a single thesaurus, the more words will appear in cross-reference, because it's simply a matter of the relevant words coming to mind -- especially considering that it greatly broadens the dimensions in which relevancy becomes relevant..which brings me to my second point:

The policy on relevant words should not be too strict. The benefits of having the right word at the right time far outweigh the benefits of any puristic notions of parallelism.

Furthermore, we can easily have the best of two words by introducing orders of separation into any given entry. That is, you have first-order relations, then second-order, then maybe third order, and of course you can click on any synonym (or pseudo-synonym) which brings up another three orders of relation. This is essentially the idea behind visuwords, only in visuwords it's even more convenient because it's all visualized. We may provide an analogous visual interface to the thesaurus..the way they did it is actually pretty damned good.

There needs to be a mechanism of collaboration, editing, undoing others changes, etc. A good solution might be like Wikipedia where people can a) freely make changes, b) roll back changes, and c) discuss changes in the discussion page .. or possibly only allow a change after a vote. If not that, possibly only allow registered users to make changes.

Changes would include
 - adding a synonym
 - removing a synonym
 - moving a "synonym" to a higher or lower order
Definitions, parts of speech and semantic relationships should be provided by WordNet and therefore shouldn't have to be edited by wiki users--except to tell to it which sense of a word is being referred to. In other words, when a user adds a word, it should bring up the options for part of speech and/or sense that are provided by WordNet (when there is more than one possibility), and let the user select which one.

Multi-word expressions should be allowed. In some cases they might not be available on WordNet. Perhaps then, the grammatical and semantical information should be editable..but if we're going to provide the feature, we might as well allow it for any word, you know, just in case WordNet is wrong sometimes...maybe with a vote.

Visuwords has 24 elements in its legend. Incorporate them all, and make it look good as Visuwords, or die.

There is a rather sophisticated interface for Python to WordNet, called the Natural Language Toolkit: http://nltk.org

users should be able to enter words that are analogous in meaning but occupy different parts of speech, but should be able to filter out those words in search.