Dependency Parsing seems to be present in most of the NLP toolkits out there.

What is not clear and what I have trouble finding in Google is what are actual practical, or even research, applications of parsing.

  • 1
    The syntax of a (natural) language has two dimensions: constituency and dependency. What dependency parsing gives your are the relations between words, which are important for semantics (the meaning of the sentence). – Atamiri Mar 16 '19 at 7:00

Parsing is mostly an important building block used for improving the performance of downstream tasks, not as an application per se.

So, for example, if we are training NER, it helps to hint to the model what the parser guesses the noun phrases are. Similar for dialogue systems, translation and so on. See also: POS language model.

In the old days (before 2015 or so) parsing was typically an early step in a pipeline. Nowadays with end-to-end models it could be used to create the representation, either by marking up the text before vectorising, or as an additional vector. But anecdotally I can say that in the early days of deep learning mania raw text vector input had a lot of appeal, now we are in a phase of adding back a few of the things like parse trees and parts of speech.

Intuitively, the point is: a good parsing model has been trained on very large datasets, so your model can effectively incorporate all that knowledge of the world, for an aspect of language that generalises reasonably well across a language.

  • So how do you feel about DG vs PSG effect on parsers? – amI Mar 16 '19 at 6:09
  • (As an engineer...) Whatever is actually available, practical and works better for the given task. The way deep learning is, you can just use both. The way reality is, for most languages you are lucky if you even have one. – Adam Bittlingmayer Mar 16 '19 at 12:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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