I am wondering what it takes to parse a sentence with incomplete knowledge. That is, take a sentence like this:

If I use timeout I have to call again my function at the end of the execution of the callback of timeout (like I'm doing right now). If something goes wrong, my callback will never be called again.

I'm not sure I spotted the verbs all correctly. But I can find them because I know what those words mean from experience. I'm wondering though if there are hints in the structure of these sentences (that is, hints in the structure of any sentence, these just being one example) that tell you when something is a verb. For example, maybe verbs typically come after I <verb> or something <verb>. Wondering if there are any patterns here, so if you weren't familiar with the language, given some bare minimum knowledge, you could identify the verbs. If so, wondering what the bare minimum knowledge would be, and what the patterns typically are (or where I should look for further search).

  • 1
    Is this question specifically about NLP, or more general morphological and syntactic tests for POS? – curiousdannii Jul 31 at 13:52

Yes, with probabilistic parsing and part-of-speech tagging.

See:

  • The OP said '...if you weren't familiar with the language...', so I think unsupervised POS-tagging would be a helpful keyword, considering that there may not be POS-tagged corpora available. – WavesWashSands Jul 31 at 10:06
  • Yes, some familiarity with the language is needed, but you do not need a complete list of verbs (or a complete dictionary that is part-of-speech annotated). A semi-supervised approach is perfectly feasible. – jknappen Jul 31 at 10:27
  • But he said 'given some bare minimum knowledge', which could include a labelled dataset. A (supervised) tagger is not familiar with the language, it just knows what it needs to do its task. Anyway, the POS tagging article covers both, if he is curious he should read about them all, for one thing the unsupervised type are more standard for now so there are just more resources on it that would be accessible to beginners. – A. M. Bittlingmayer Jul 31 at 10:29
  • I kinda interpreted 'minimal' to mean literally the least amount of knowledge needed for POS tagging (which would exclude such a dataset, since it's possible to POS tag without it). But yeah, I agree, the OP can read about all of them! – WavesWashSands Jul 31 at 11:30
  • Lots of reading! Thank you. By minimal knowledge, I was imagining like a kindergartner or someone just learning to speak. – Lance Pollard Jul 31 at 14:24

In a word, if you don't know, you have to guess, then try to verify your guess. You will probably use some heuristics (that is, rules of thumb) to guide your guesswork. At least, as a linguist, that's what I do, and I think it is a plausible model for what any hearer does in figuring out what is being said. Your first guesses will lead to further guesses about the structure of an expression you've heard.

A technical term for this process is top-down parsing, which means that deciphering the tree structure of an expression, you proceed from the top of a structure tree, with the root, and work your way down to lower branches, and eventually to leaves (which will be morphemes, or perhaps phonemes).

Although, as I said, I think this is a plausible idea of how language understanding works, using it as a conscious strategy in grammatical analysis, is not very intuitive, and is something that has to be taught to beginning syntax students, who are more naturally led to bottom up parsing, as an alternative model.

In bottom up parsing, you start with phonemes and morphemes, look those up in a mental dictionary of what you know of the language, and work your way up the tree, winding up with a complete analysis of structure only at the end.

For instance, for figuring out where English verbs are, in bottom up parsing, perhaps you could look for /d/s and /t/s at the ends of words as a clue, with the idea that they might be past tense suffixes, which would identify words as finite verbs.

In top down parsing, however, if the communicative context were that someone is giving you instructions, you might guess that the first words of sentences are verbs, since English imperatives start with verbs.

  • 'Bottom up' uses syntactic context and 'top down' uses semantic context. You can't really parse anything without using syntactic clues, but (for example) "I can fish all day" is ambiguous without also using 'top down' context. – amI Nov 10 at 5:18
  • @amI, yes, at least in the examples I gave. Syntactic clues also work in top-down parsing, as for example when we guess that an expression is an S when it is coordinated with an S. – Greg Lee Nov 10 at 18:31

A word is a verb if it is used as a verb.

I will gribicaluate you.

In this sentence, gribicaluate is not a valid word but it is used as a verb, so it is a verb.

A more realistic example is

Let me google it.

Here google is used a verb.

...

And how we can figure it out, the above answer answers this.

We're looking for long answers that provide some explanation and context. Don't just give a one-line answer; explain why your answer is right, ideally with citations. Answers that don't include explanations may be removed.

  • 'The above answer nothing' about how we determine usage. – amI Nov 10 at 5:06

Your Answer

 

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.