[en] Chunker improvement

[MikeUnwalla]

Is it possible change the behaviour of the chunker?

With a plural noun in a singular noun phrase, the chunker incorrectly gives the last word the chunk E-NP-plural. Some examples:

The aircraft maintenance manager is in the hangar.
Does your box converter operate correctly?
The happy sheep is grazing in the field.
The shiny new aircraft was on the runway.
I’d like a fish pie and a kilo of sprouts.

(Only a very small proportion of plural noun phrases end with a singular noun. Example, ‘sergeants major’: https://www.merriam-webster.com/dictionary/sergeants%20major).

[danielnaber]

I have made a change locally. The change wasn't difficult, but at least 12 tests break now. These seem to be cases where the chunker is wrong and has always been, but now it causes an error. One example:

Adults average in length; the largest recorded specimen weighed 2.65 kg.

"Adults average" is considered a noun phrase, and now it's considered a singular noun phrase, breaking ADJECTIVE_IN_ATTRIBUTE. I won't have time to fix these cases. @MikeUnwalla, if you're interested in working on those cases, I could commit my changes to a branch. Of course, more issues might show up after those 12 tests are fixed...

[MikeUnwalla]

Hi Daniel,

Yes, I will work on these cases.

I will probably need help with setting up a branch.

Another type of chunker error is where there are different readings for upper-case text and lower-case text:

Sudden operation of the gyro can cause unwanted movement of the horizontal stabilizer.
SUDDEN OPERATION OF THE GYRO CAN CAUSE UNWANTED MOVEMENT OF THE HORIZONTAL STABILIZER.

[danielnaber]

The work-in-progress branch (which tests that don't work) is now available here: https://github.com/languagetool-org/languagetool/tree/issue-7044-chunker

[MikeUnwalla]

@languagetool-org/developers, if @danielnaber is out of the office today, can one of you please look at #7050 ? Many thanks. I think that I have possibly not used GitHub correctly. I am not sure what to do.

[danielnaber]

PR has been merged now.

[MikeUnwalla]

@danielnaber, thank you.

I know about more chunker errors. Do you want them? If yes, I will make a new issue.

[danielnaber]

I know about more chunker errors. Do you want them? If yes, I will make a new issue.

We cannot really fix them, but maybe we can add antipatterns, so please send them.