I want an online language translation service that uses genetic search result queries to improve performance


I have used several online translation sites from Google language tools to Babelfish. For western languages the results seem to be very good. For asian languages like Chinese or Japanese, the translation often results in a haiku like jumble of stream of consciousness words. Although not very effective, its still better than nothing.


What I wonder is why the language results are simply 1 result. Wouldn’t it be better to get multiple results like in a search and let the user pick the best one that they feel matches the translation that they seek. The translation service can log which choices were chosen and therefore work to improve the algorithm or at least the validity of the translation result for future users or use. Currently, using translation is like hitting the “I feel lucky” button on Google Search.


By helping train the engine on what constitutes a good translation, the output and algorithm can improve in an evolutionary way like genetic selection.


If you check out www.pandora.com which uses genetic selection of music styles, you will see that you as a user giving feedback helps to train the database and selection engine to make better more valid music choices for your listening pleasure. Sure, some music choices are duds but the more I use the service the more it seems to understand the type of music I like and has helped me to discover loads of new music that I would otherwise never have encountered.


Language translation should be similar to this brute learning approach and should be hooked onto a dynamic database that changes with user choices and context.


Just my two fen but it seems like a good idea…(I’m not on any search projects so perhaps this is already underway)


Frank Yu