0

The only web-based automatic parser I know is CoreNLP version 4.5.5, where you can put in an English text and get a constituency tree (when you select 'parts-of-speech' as Annotations and click 'Submit').

It works fine with simple texts, but when you put in a complicated text, you might get an incorrect tree under any grammatical frame. For example, if you put this noun phrase even all the preposterous salary from Lloyds that Bill gets, you get this tree:

enter image description here Here, the relative clause that Bill gets combines with Lloyds to form an NP, which is incorrect under any grammatical framework.

So I'm looking for a better automatic web-based parser that's error-free, hopefully. Alternatively, I wonder if I can fix whatever error I find in this parser by tweaking it. Any help is appreciated.

Improve this question

1 Answer 1

Reset to default
1

Ignoring the fact that there are ambiguous sentences, but parsers usually give only one parse (that may reflect the intention of the speaker or not), there are no error-free parsers out there. All recent parsers I am aware of use some kind of statistical machine learning and training. They have an error rate by design. You can try to employ several different parsers and accept a parse only when a majority of the parsers agree, but even this procedure will leave some sentences with wrong parses, and you have some sentences with no accepted parse at all.

Improve this answer
4
  • Thanks. With the advent of LLM, I wonder why there hasn't been any automatic parer developed that analyzes enough context to parse any sentence without ambiguity.
    – JK2
    Commented Sep 3 at 9:28
  • 1
    Because LLMs only regurgitate their training data based on probabilities. There is no new knowledge gained from using them.
    – Oliver Mason
    Commented Sep 5 at 11:40
  • Also, language in general is not precise enough to allow for unambiguous parsing of arbitrary sentences. A lot of possibilities can be rejected with world knowledge, but a syntactic parser typically does not have that available.
    – Oliver Mason
    Commented Sep 10 at 9:43
  • 1
    I have heard that the best dependency parsers have an accuracy rate that reaches 90%, or even a little more. Your parse here is a phrase structure parse, and hence the syntactic analysis is much more complicated than the dependency parse would be. More nodes in the syntactic analysis means more potential for incorrect parsing choices.
    – Tim Osborne
    Commented Sep 10 at 15:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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