Live blog post from LeWeb 2011: Deb Roy, Co-Founder & CEO, Bluefin Labs

There are four prin­ci­ples to Deb Roy’s con­cept of pre­dict­ing learning:

1. Record every­thing and store this data
2. Develop machine learn­ing algo­rithms to under­stand the data
3. Visu­al­ize those pat­terns to cre­ate understanding
4. Lever­age those under­stand­ings to bet­ter cer­tain areas of society

Lan­guage is the defin­ing fea­ture of the human species. The basic data to under­stand how a child learns is nat­u­rally in the home. Deb Roy recorded video of his baby son through­out the house as he grew up and started to talk. The video pro­duced move­ment data that could be traced and dat­a­mined, par­tic­u­lar when the son tran­si­tioned from say­ing “gaga” to say­ing “water.”

There is a nat­ural rela­tion­ship between the num­ber of times a word is heard and learn­ing that word. How­ever, there is a social con­text to learn­ing. The activ­ity one does when learn­ing also affects learn­ing of words related to the activ­ity. The 3-D move­ment land­scape pat­tern asso­ci­ated with learn­ing words can be referred to as a visual word­scape. The unique­ness of the word­scape is a pre­dic­tor of how early the word will be learned. The more unique the word­scape com­pared to other words,  the ear­lier the word will be learned. This has impli­ca­tions for those hav­ing trou­ble learn­ing lan­guage; you need to look at the non-linguistic aspects of learning.

The same prin­ci­ples of lin­guis­tic analy­sis can be applied to a whole new are: TV. His com­pany, Bluefin Labs, con­nects what shows are run­ning to what peo­ple are say­ing about the shows. Bluefin is index­ing about 215 US net­works, 200,000 shows and 2 mil­lion ads per month and link­ing these to con­ver­sa­tions gen­er­ated by 20 mil­lion peo­ple. With text min­ing and seman­tic analy­sis, you can under­stand what peo­ple are think­ing about collectively.

Here is Deb Roy’s really inter­est­ing talk at TED: