TomazErjavec
commented
3 years ago Dear Maria, I have been neglecting you most shamefully! I actually did some work, but didn't push it before today to your repo, neither did I open new issues. Will do now, at least some, so you won't be too bored! :)
Very good news that you found somebody to do the linguistic annotation. These are the rules:
- the annotated files look exaclty the same as the unannotated files, except that their filenames have ".ana" at the end, that their segments (so,
<seg>elements) are linguistically annotated, and that the root file has some extra bits (we will do this last once the annotation of the texts is done) - the annotation should be according to the Universal Dependencies (for Spanish), so, tokenisation, sentence segmentation, lemmas, morphological features and dependency relations
- NER should categorise according to the standard classes, so PER, ORG, LOC, MISC; NER annotations should not be nested
- only the text content of the
<seg>elements should be annotated - inside segs the only tags are
<pb/>and<note>; page break is empty anyway, and the notes should be ignored in the annotation, but it is important that they are present in the final annotated TEI output - how the XML encoded annotations looks like, you can see in the example corpora, e.g. https://github.com/clarin-eric/ParlaMint/blob/a1110008eae5bc837d111bf46aa405671948fd13/ParlaMint-PL/ParlaMint-PL_2015-11-12-senat-01-1.ana.xml#L1735-L1797.
And this is mostly it I think; and whoever is doing it can consult the ParlaMint samples, and I can of course also help with advice if necessary.
calzada
Dear Tomaz, How are you? I hope you have managed to get some rest this wekk-end. So that I do not overload you with work, (and taking into account I do not know mucho so far about UD and NER) I have managed to find some help fo both stages. My friends have just started to look into this as we speak. Any suggestions or important recommendations we should know.
Best for now. mc