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

Keyster - Fast, Accurate, Image-Aligned Text Entry

  See Keyster examples

  From Assistive Technology to Assisting Technology

CRCL is pleased to announce the beta release of Keyster, a system for high-speed, high-accuracy, image-aligned text and tag entry.

    Keyster was developed to help Digital Divide Data (DDD), an award-winning organization that provides jobs and training to the disadvantaged in Cambodia and Laos. DDD specializes in single and double-key entry of pages that are complex, poorly scanned, or use exotic fonts - if DDD gets the job, OCR didn't work.
    Keyster relies on assistive technology, both to help disabled or less-well educated operators, and to ward off repetitive stress injuries that plague this kind of work.
    But we also built Keyster to explore ways of assisting technology - to let humans step in right at the point where software fails - and by doing so, to create skilled jobs in the broad divide between low-wage and high-tech.

What Keyster Does   see examples
Keyster is a 'heads-up' text transcription tool, used for copying newspapers, handwritten notes, computer printouts, and the like from scanned images, 

    Keyster's primary strength is high-accuracy applications; particularly when multiple independent entry, and close visual (as well as textual) matching to the original are required.  Typed text and image can be precisely aligned and font-matched, and paired XML tags and user-defined entities can be inserted by button or hotkey.
    A variety of adaptive features make Keyster exceptionally sensitive to the needs of DDD employees, many of whom are disabled, and none of whose first language is English. These include:
 -  real-time predictive word completion and multi-language spell checking,
 -  automatic extension of the user interface to accommodate arbitrary XML tags and entities; buttons and other shortcuts are automatically derived from the customer's stylesheet or DTD.
 -  intuitive visual access to Unicode and user-defined character entities.
    Keyster development was initially prompted by CRCL's own need for extreme accuracy in copying difficult multilingual dictionaries (like this example, which includes Khmer, Thai, Indic transliteration, and phonetic transcription):

    Below, the same sample as it is being entered in Keyster.  Assistive tools handle a variety of tasks, from entering balanced XML brackets to finding and entering Unicode character entity codes. Operators can do more complex and demanding work without decreased accuracy, or increased stress: