The second is actually a decent example of where the semantic web would probably fail because of the imprecision, but Google has succeeded to some degree. “When was grass invented” has no good answer because it’s a stupid question. You could interpret this as “when was grass discovered” but I suspect this is not the question being asked. Google has interpreted this as “when was the grass lawn invented”, which is probably what the user was actually asking. In both cases, the results are actually great (my first link was to the Wikipedia Lawn article).
I don't think it's a stupid question, just oddly phrased (more on that later). I think there are two stable interpretations. "When did grass evolve" or "when were grass lawns invented".
Later:
Why isn't it an obviously stupid question? Because I think the accompanying "who (or what) invented grass" is validly answered as "by evolution". I feel the act of invention requires no intentionality and is simply the output of learning processes where generated artifacts have material and dynamic properties embodying deep knowledge of physical laws and help achieve some goal relative to an environment. Evolution learns in the sense that the mathematics of natural selection mirror that of bayesian filters.
