Markov matcher for dabbrev

Question

While writing plain text (not code), I'd like a completion based on more context, an exmple would speak more so here it is:

Given the following buffer (a preloaded big text corpus would make more sense in reality):

foo bar baz
foo bar baz
foo bar boo
poo bar bot

and then typing:

foo bar

(followed by a space and M-/) I'd like dabbrev (or something else) to propose, in the following order:

• baz, because it appears twice after foo bar, so it's the more probable one.
• boo, because it appears after foo bar
• bot, because it appears after bar

Currently dabbrev would propose bot first, as the nearest possible word following bar, if I understand correctly.

In other words, I'd like completion based on Markov chains. Does it already exists? Would it be hard to implement?