On Sun, Nov 30, 2008 at 10:50:48AM +0000, Paul Smith wrote: > >> Ideally, of course, autocompletion should work already with the first > >> keystroke in a document and produce ready-to-submit papers. > >> > >> Of course, that still leaves the author at the mercy of the (mostly) > >> human reviewers... > > > > Ideally, of course, we would like to have less ironic and more > > constructive LyX developers... > > The idea of using machine learning algorithms to the problem of > ranking text autocomplete suggestions is not, in fact, new; it has > been applied to determine the suggestions of urls for Firefox/Mozilla: > > http://www.mozilla.org/projects/ml/autocomplete/
The "bugzilla example" they cite as bad case of conventional autocompletion can probably be handled well by "conventional autocompletion plus wildcards". To get to the sixth completion there one would e.g. type b, u, g, *, 9 I am not sure about the general merits of machine learning in such an environtment, though, especially if the input comes from several users. We all know about the possible outcomes when asking Aunt Google for LaTeX ;-) Andre'