search

TraCES-Lexicon / lexicon

Geez lexicon of the TraCES project
1 stars 0 forks source link

Examples for testing #35

Closed sdruskat closed 5 years ago

sdruskat commented 6 years ago

I'm starting to implement automated testing for the API and need a number of examples to test against.

Please provide at least three examples per category, with a complete solution, so that I can test that the API works correctly. "Categories" are, e.g.:

What I need for each example to be testable is:

@PietroLiuzzo Could you please assign to those who might be able to provide the examples? Thanks!

PietroLiuzzo commented 6 years ago

Aren't the ones in the tables fine? I can extract them from the tables which vitagrazia and magda did, but it sounds a bit cumbersome and you will have to wait that I am back from mekelle...

PietroLiuzzo commented 6 years ago

perhaps @AndreasEllwardt, @abausi, @MagdaKrzyz and @SusanneHummel can provide here some examples as describe above, selecting cases which are simple as well as cases which are complex.

sdruskat commented 6 years ago

That'd be great, thanks. I'd like to test against real life queries, not generated ones (I'm doing some of those already).

MagdaKrzyz commented 5 years ago

Dear Stephan,

I am sending you some examples for your queries. In the field "Complete grammatical categorisation" I gave the annotation according to TraCES. Note that some of the features can be determined only in the context (not in isolated words). That is why, espcially for nouns, it is good to work with a text. If sth is unclear, please write me back. Best M Stephan_examples.docx agda

PietroLiuzzo commented 5 years ago

@sdruskat are you happy with this? I actually find a bit confusing that the solutions offered in the examples are the TraCES morphological annotations. This was never a requirement for the parser, also because these annotation are contextual, and a parser working on a word only cannot know all what is in there. The tables provided do not allow to retrive this information as far as I know. Also the prefixes are hard coded in a table, but how would one know from there the state? That is way the demo parser returns searches into the annotations to that end. But Stephan does not have those, or I am missing something? Especially the first part on nouns is IMHO impossible to calculate from the data in the tables, it would need to use the TraCES annotations and this would not be logic, but matching in a corpus only which should return contextualized results.

abausi commented 5 years ago

I agree with this: the parser has NOT to take into account the TraCES annotation; it has exactly to work ignoring all that is there, that is the result of a completely different process of analysis - i.e. the other way around.

Il 02.11.2018 08:50, Pietro Liuzzo ha scritto:

@sdruskat https://github.com/sdruskat are you happy with this? I actually find a bit confusing that the solutions offered in the examples are the TraCES morphological annotations. This was never a requirement for the parser, also because these annotation are contextual, and a parser working on a word only cannot know all what is in there. The tables provided do not allow to retreave this information as far as I know. Also the prefixes are hard coded in a table, but how would one know from there the state? That is way the demo parser returns searches into the annotations to that end. But Stephan does not have those, or I am missing something? Especially the first part on nouns is IMHO impossible to calculate from the data in the tables, it would need to use the TraCES annotations and this would not be logic, but matching in a corpus only which should return contextualized results.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/TraCES-Lexicon/lexicon/issues/35#issuecomment-435298330, or mute the thread https://github.com/notifications/unsubscribe-auth/ATDMaFDsI0EM6-mqNIBypGy2MpfLVEG-ks5uq_kxgaJpZM4XJuFS.

sdruskat commented 5 years ago

Thanks @MagdaKrzyz for the examples. They work well for the representations, lemma, and pattern.

There has been some confusion around the model of the parser vs. TraCES annotations. Just to be clear here, I can only implement what is possible to retrieve from structured information on the word level, not the context, similar to what the demo parser does (including hard-coded information).