Now we have a different problem -- in that the new examples aren't idiomatic. There's no reason we should expect it to identify the phrases 'L' train line and

Now we have a different problem -- in that the new examples aren't idiomatic. There's no reason we should expect it to identify the phrases 'L' train line and 'L' line with equal accuracy, because the former is something basically nobody says -- while the latter is something everyone (even someone unfamiliar with that particular city) understands. But hey -- I'm not defending the API's accuracy ;)