What about this...? A simple tool might be useful as a first line of defense against silly filings. First, the tool could monitor trademark filings. Second, it could compare against common words, perhaps from a dictionary. Third, it could score filings using a 'frivolity' score. Fourth, perhaps it could expose its a moderation queue via a mailing list or web interface. Fifth, maybe it could be gamified?
Maybe you should use Google Search result count. If a phrase doesn't appear frequently online, its ok to trademark. If there are 10 million results, it should be flagged.
For such a tool to be most useful, I'd recommend gathering data from several places (such as DuckDuckGo, Wiktionary, Wikipedia). Also, it would be better not to be tied to fixed thresholds (such as 10 million results); rather, it would likely be better to construct a feature vector and try out various machine learning approaches.
