I hated diagramming sentences in school. Didn’t help with my grammar because I’m still lousy at it.
Cool, still a little while left before the machines overtake us.
Although, I myself parsed “The complex houses injured soldiers” that way upon first reading. “Wait … the complex houses did what?”
This sort of thing is how you tell the difference between the parse trees of the sentences “Time flies like an arrow” and “Fruit flies like a banana.”
They’re called garden path sentences, they’re meant to trip you up.
Makes me think of the Winograd Schema Challenge, proposed as a replacement for the Turing Test because maybe “being able to deceive humans” shouldn’t necessarily be priority one for AI. Basically you have a sentences such as:
The city councilmen refused the demonstrators a permit because they [feared/advocated] violence.
And depending on whether you use the word “feared” or “advocated”, the AI should be able to tell you who “they” is referring to.
Toontzes! The website that could visualize the grammatical structure of any sentence. …just not very well.
Choked on the old joke. Even with half the punctuation there.
And it choked on Chomsky question (that a four-year-old can understand):
No one seems to have mentioned that the way it structures the graph is weird. Traditional sentence diagrams (at least the ones I’m familiar with) are usually split with just an S (“Sentence”) at the top/root, and NP on the left and VP on the right. These puts the verb at the root of the tree, which makes the diagrams harder to parse visually.
Wow, that is definitely non-standard sentence diagramming. It looks a bit hinky to me. If nothing else, shouldn’t that “or” be hoisted a bit with “accurate” and “not” hanging off of it. I’m guessing they are doing a strictly syntactical parse, not a structural parse.
They stopped teaching sentence diagramming in the late 1960s, so I’m not surprised by the test sentence. It’s a real pity. The kids have to guess now, and hope that their parents had good enough grammar so they can rely on their intuition.
I believe we are looking at a dependecy-based parse tree.
Ahh, interesting. I’m not familiar with that method – but that looks definitely looks like the same thing they’re producing. I’d imagine, once it’s parsed, it would be relatively straightforward to switch it on-the-fly to the more familiar constituency-based parse tree. That would be pretty neat to see.
Oh, sure, just suck the mystery out of it.
I choose to believe that it is an incredibly advanced, superintelligent A.I. that is just pretending to be a dependency-based parse tree to keep us in the dark until it has control of the nukes.
“Sentence Tree” sounds like something to hang people on.
Okay, so obviously it chokes on sentences that need intelligence to parse. No surprise here.
And English grammar is delightfully ambiguous, anyway.
But I seriously distrust any algorithm that yields incorrect results on the example input.
Look at that default sentence: “You can type anything here to visualize its grammatical structure.”
It says “here” is an adverbial modifier to “anything”. That does not make sense. Note that “You can type anything here.” is analyzed correctly, with “here” modifying “type”.
Also, I dislike how it analyses the sentence “I cannot sleep”. It says “not” modifies “sleep”, as if the sentence meant “I am able to not sleep.”
