I’m not sure if you’re being facetious, but in fact it’d be easy, if you could solve the problem of computer reading in general (which is a big if). Which is to say, understanding the meaning of any given word in a sentence depends on the context of all the surrounding words already and our evaluation of the most likely meaning based on our culture: if I write “Mary had a little lamb” it’s not clear whether she owned a lamb or ate a lamb or had some sort of sexual congress with a lamb.

Thus each word is more of a probability field (like an electron) than a point, in some virtual, n-dimensional meaning space, and the task is to write a program that can collapse all those fields.

But if I had a program that could figure that out, it wouldn’t be any harder for me to understand jumbled or even misspelled words; computers would be very good at looking up all permutations of a word, they love brute-force algorithms. Really, for an n letter word, there are only ((n-1)! – 1) jumbles of that word ((n-1) because the first letter must be in the first spot, and -1 because one of the permutations is the correct spelling), which is a tiny number to a computer that can process billions of instructions per second.

And if I hit on two possible interpretations for a word (words that were anagrams of each other) that’d just be an increase in the volume of the probability field for that word.

So the mind isn’t just reading whole words at one bite, it’s whole sentences. At the very least, it’s giving words ‘provisional’ meaning which are only confirmed by reference to the whole sentence. Where’s Steve Pinker when you need him?