Back to LanguageTool Homepage - Privacy - Imprint

[En] Questions: Article usage (a/an)


(Pavan Vedula) #1

Hello all,

So we have been using LT for some time now and we have a few questions. We we have LT installed on a local environment and are testing the rules individually.

While testing the Article usage rules in the grammar.xml, we discovered that the system is able to change the wrong usage of articles (i.e. "a" used incorrectly in place of "an" and vice-e-versa) very accurately, however, for inserting missing articles it has very low success ratio. And there is no provision for unwanted articles.

Is there any specific reason that there are not many rules for insertion of articles? Can we create/add more rules that increase the current success percent of insertion of articles?

Note: We have referred to previous posts on this issue, and do know that some work has been done on this aspect.

Also, we noticed that OpenNLP system does not identify proper nouns in many cases (as compared to Stanford CoreNLP) and has separate tags for uncountable nouns, is there any way that this identification percentage be improved?

Thanks
Pavan


(Mike Unwalla) #2

By default, some grammar rules are not selected. Make sure that all the rules for articles are selected. LT correctly finds the error in "Do not eat a food before your operation."

Lack of time by the volunteers who write the rules.

Yes. Refer to these web pages:
http://wiki.languagetool.org/development-overview
http://wiki.languagetool.org/developing-robust-rules

When you write a rule, I suggest that you include plenty of examples of correct and incorrect usage. It is easy to see only what we want to see, and not think about the counter-examples. The corpora on http://corpus.byu.edu/ will help you to find examples and counter-examples.

After you write the rules and make sure that they don't cause many false warnings, please open an issue on https://github.com/languagetool-org/languagetool/issues and put your rules in that.Then the LT team can add the rules to LT.

I don't know. Possibly, other members of the LT team can answer that question.


(Pavan Vedula) #3

Thanks Mike. That info is very helpful :slight_smile: .


(Daniel Naber) #4

This might be addressed by machine learning, let me know if you're interested in working on it.


(Pavan Vedula) #5

Sure. @dnaber Am currently reading and understanding various NLP concepts and would like to contribute to the OpenNLP system. Will come back with some ideas when I am ready :slight_smile: