Ref trnka aac

• User Interaction with Word Prediction: The Effects of Prediction Quality
• hypotheses:
  • increase communication rate for those writing at AAC speeds
  • advanced word prediction improves communication rate more than would (theoretical) keystroke saving imply (how come?)
• many references to similar testing (AAC)
• references Newell et al. [1992] who mentioned (for children):
  • higher quantity and improved quality of written texts
• references Venatagiri [1993] who found:
  • word prediction doesn't increase communication rate
• their experiment:
  • not impiared people, paid $30 :-)
  • rewriting senteces, on-screen keyboard, enforced delay 1.5 s (impairment simulation)
  • used Wivik keyboard, that now uses WordQ tool
  • five suggestions are common AAC standard (p. 14)
  • word prediction:
    • none
    • basic -- recently used words + alphabeticaly sorted dictionary (prefix filtered)
    • advanced -- trigram LM with with backoff smoothing (p. 15)
• results:
  • measures
    • input rate -- keystroke per second
    • commuincation rate -- words per minute
    • prediction utilization -- ratio between actual keystroke savings and theoretical
  • derived formula expressing trade-off between keystroke savings and input rate (p. 30)
  • increase in communication rate:
    • 45.4 % for advanced prediction against basic prediction
    • 58.6 % for advanced prediction against no prediction
    • those are numbers from conclusion, figure 7 says a bit different (46.6 % and 61.2 % (figure itself), 45.8 % and 61.4 % (comments to figure)),