More or less, as I see it. Part-of-speech tagging does pay; Knowing the function of each word is a first step towards knowing their meaning (especially ambiguity). And indeed, from there on it would seem a matter of laying the elements out in natural patterns until one “makes sense”. Humans match several such possible patterns simultaneously as they read a sentence, but repeated attempts would be just as effective.
I must note that I’ve always found nouns to be a category blind to meaning, so I’m not too crazy about following POS methods to the letter though.

The part-of-speech tagging is where the hundreds of grammar rules mostly come into play, or statistics in some systems . The arrangement phase after that is, certainly in essence, fairly simple, and from there on the sentence will have been translated to the AI’s native inner language. “Understanding”, as I call it.
Nevertheless it still takes quite a bit of ingenuity to pick apart noun-phrases and correctly apply the strewn mess of modifying words to their appropriate elements. And in your Yoda example for instance there is a sub-clause that takes the role of object “[understanding this sentence] [a computer] does.”, which in itself is a sentence. So one would need several levels of POS-tagging upon POS-tagging (not sure if I’m still using the term correctly here). Which also requires the AI to have a more flexible knowledge structure than a sentence structure.

At least, that’s my take on it. The arrangement phase in my AI is still under development, if that wasn’t obvious enough from its jumbled response

You know, I’m going to put that Yoda language on my to-do list for today after all. I recently said “Interesting project you have there.”, so it’s more common than I thought and I’ve been working on relative clauses this week anyway. There’s a clear “object + verb = subject” rule in there. Thanks for the tête à tête, Dave.

Glad I could be of help, Don.