Keyster

C · R · C · L
Center for Research in Computational Linguistics, Bangkok

From Assistive Technology to Assisting Technology

Basic competence at typing can be readily attained at minimal cost; physical disiblity or lack of education do not pose insurmountable barriers. DDD founder Jeremy Hockenstein recognized that data entry could, with some assistance, support a new kind of cottage industry, even in low-technology regions.

But high-speed typing at commodity rates - building 'electronic sweatshops' - would defeat the purpose. CRCL became interested in helping DDD turn simple key-entry into skilled jobs that would not just compete with other low-wage countries.

Keyster's basic premise is to use human talents at pattern recognition and text understanding to do what computer software cannot - make correct decisions under uncertain conditions.

In building Keyster (and its future companions TagBoy and ZoneFever), we deliberately seek out the kind of work that breaks software, including:

- OCR of poor-quality text or handwriting, or text that uses scripts (like those of Southeast Asia) that cannot be OCR'd effectively;

- accurate tagging of page content, both small scale (references to people or businesses), and large (distinguishing section types in textbooks);

- building 'virtual' pages that ignore advertising content, and understand the layout of newspaper or magazine articles;

Then, we develop 'bridge' software and procedures that DDD employees use to help standard or experimental tools work through, or recover from, failure.