As a programmer, I could recognize that MIDI editing and scoring have not so many direct relations. That is not obvious looking at graphic programs, since many MIDI editors have some score system (f.e. Sonar) and most score editors can produce MIDI files.
But behind the graphics, it is in direction of speech recognition vs book typesetting. Using Text To Speech engine it is possible to "convert" a book into more or less reasonable audio, since not only words but the whole punctuation is there. The same for Score editors to MIDI conversion. For triplets example, it is possible to calculate precise timing when they should be played, add some "human feel", accents, slurs, etc. In other words, more or less formal procedure to interpret scores into music till some degree exists (the primary purpose for scores!).
Reversed process is far more blurred. If in book you see words and punctuation only, in scores there are way more "extra" information. And apart from notes, nothing exists in the MIDI file to guess how that should be scored. To understand the problem, have a look at text based representations for MIDI and for scores. For MIDI, you can use Event List in Sonar. For scores:
http://lilypond.org/ For triplets: MIDI has NO triplets! Note duration is specified in 1/960 of a Beat. For scoring, any human played MIDI should be hard quantized. And if you have tried such function, you know how it works for complex material.
Also there are hard limitations. For books, you can put commas based on syntax of recognized sentence. You can try to predict exclamations. But how, for example, can you recognize a citation as such? The same for music. As soon as you get some unregular material (imagine a triplet in a sequence played with human feel, "abused" (from mathematic perspective) by swing), the complexity of guessed formal procedure to produce something reasonable explode.
I am no way against improvements in Sonar scoring system. But as MIDI based program, it is unable to create final scores from the information it has (and that is current approach there). Build-in score editor or better cooperation with external score editors are feasible and will be welcome, the same for visualization improvements. But not artificial intelligence for real-time conversion between MIDI and ready to print complex scores.