I had an idea for something like this a while back but it had a different twist: for any given noun or verb, anyone could upload their attempt at making the most universally recognizable version of that icon. The creator would tag the icon with descriptive words to help categorize it, with maybe one main word that it's trying to convey.
To truly discover which icons are the most clear and recognizable for a given noun, other users would be able to go through the icons one by one and add what they think each icon means by entering in a few key words. With enough input from enough users, it would gradually become clear which icons met their goals and which didn't. These results could be broken down by country and ethnicity, too. This way, we could really be confident that using an icon that's supposed to translate to a word will actually translate as it's supposed to for the maximum number of people.
That sounds like an awesome way to assist in localisation. By analysing the results you could have one set of icons for your app for western countries, another for China etc. It reminds me of the problems planners had in trying to design airport signs for international visiters in the early days of flying and the confusion caused by complex signs asking people to wash their hands. The BBC did a documentary on early UK airports and it featured in it but fortunately I can't find it.
To truly discover which icons are the most clear and recognizable for a given noun, other users would be able to go through the icons one by one and add what they think each icon means by entering in a few key words. With enough input from enough users, it would gradually become clear which icons met their goals and which didn't. These results could be broken down by country and ethnicity, too. This way, we could really be confident that using an icon that's supposed to translate to a word will actually translate as it's supposed to for the maximum number of people.