Re:Re:Instant Text Revision 303 Now Available
Re: Re:Instant Text Revision 303 Now Available -- Jim Kerslake
Jim, Choosing between alternative rule-sets is an option for the future. Of course, one constraint to keep in mind is that we have to compute the set of candidates in the time that separates two keystrokes. In practice we make sure that our computations stay below 1/100 of a second on a slow machine. This most recent improvement was something we had wanted to do for three years and it is only in the past months that we found a technique allowing us to do it fast enough. Could frequencies add something? Perhaps yes and no. On one hand, they can be useful with very large general glossaries, say with 50,000 entries or more. But the best use of frequencies is in glossary creation itself. If you do smaller glossaries covering a narrower subject, then the chances that you are going to have more strictly left-to-right candidates than advisory lines is small. Also you can always add an additional letter to prune the current list. In our own experimentation, we have found frequencies to be of marginal value for the advisory choice itself. Perhaps after a few months of experience with the new system, we'll have a better appreciation of their potential usefulness. Note also that the problem for phrases is entirely different: you have complete control of the order in which phrases of a given two-letter group appear in the phrase advisory and you can modify it in the Move dialog. In the future, what we do will be also influenced by the level of horsepower (megahertz) that can be expected on average machines. Currently, unlike Dragon which assumes 128MB and 400Mhz, we are assuming much less, but the situation may change and we may take advantage of the extra power available on average machines. Finally, while the current improvement is significant, I remain convinced that what allow faster-than-speech entry (hence much faster than voice) are continuations. This is an area where you are going to see further improvements in version 4 next year. Thanks for your comments Jean
|