This is a very interesting application of WolframAlpha but it appears to be purely luck when "success" was the result. Using things such as "2nd item in a..." or "7th digit in..." work in a lot of cases but lets talk about a few.
"2nd fruit in bear apple goat orange" would result in apple because it is looking for second in a list and neglects context of fruit.
"7th digit in abc123def456ghi789 " would result in d when it should be 7. Again not understanding context and merely looking at logical construction.
> "2nd fruit in bear apple goat orange"[1], "7th digit in abc123def456ghi789"[2]
It barfs on these ones, but not like you predict (it's actually worse). "The 2nd colour in purple, belly, yellow, arm, white and blue"[3] gives back yellow, though, so it's not that stupid.
If you change yellow to silver, it gives you "colour silver," but if you look at the assumptions, it assumes silver is an element. Clearly it's not really getting the meaning of the sentence, and the fact that it's picking "yellow" is mostly coincidental, I think, as is also illustrated by your examples.
Right. And as a little experiment, it was surprising the results I got. It also implies that it is not hard to structure the sentences in a way that makes it harder for WolframAlpha's algorithms to get the right answer. But doing that for 180 million questions? I wonder what the percentage of success using only WolframAlpha would be on the whole data set.
"2nd fruit in bear apple goat orange" would result in apple because it is looking for second in a list and neglects context of fruit.
"7th digit in abc123def456ghi789 " would result in d when it should be 7. Again not understanding context and merely looking at logical construction.