Ref Mubaid notes

• name: Application of word prediction and disambiguation to improve text entry for people with physical disabilities (assistive technology)

• in the begining they repeat quite often the same information (those from abstract about approaches)

• their project

  • main goal: minimize keystrokes
  • side goal: reduce cognitive load (limit no. of suggestions)
  • they use machine learning for syntactic information, not only n-gram statistics
  • combine two sources of information: a) corpora texts b) user texts
  • methods they use are SVM (support vector machine) and Lsquares (in the article explained only roughly, I did not understand it much, referencing other works)
  • emphasis on real-time processing
  • results
    • they tested on medical texts where some names are ambiguous, they're same for genes and proteins – they predicted
    • they express results in terms of accuracy, precision and recall (??), in the case of proteins/genes they are over 99%
    • they refer to tags-and-words (POS + ngrams??) methods that have hit rate of 37% (intended word appeared in the suggestions list) and key stroke savings of 53%