Hi Adam,
aware of this problem, we added some features in the SUD (Surface-Syntactic Universal Dependencies) for coordination: - A feature Shared=Yes (or No) on shared dependents: http://universal.grew.fr/?custom=63495b6f8a644 - A feature @emb on embedded coordinations (either Peter and Bill or Tom): http://universal.grew.fr/?custom=63495b3e5743f
The annotation scheme is explained at https://surfacesyntacticud.github.io/guidelines/u/particular_phenomena/coord.... SUD treebanks are automatically converted into UD, but, as you mentioned, only a part of the information can be recovered in the UD=>SUD conversion.
The Shared feature is presented in our last paper on SUD: Gerdes K., Guillaume B., Kahane S, Perrier G. (2021) Starting a new treebank? Go SUD! https://aclanthology.org/2021.depling-1.4.pdf, Proceedings of 6th international conference on Dependency Linguistics (DepLing), SyntaxFest, ACL, 11 p.
Best Sy
Le 13 oct. 2022 à 09:06, Adam Przepiórkowski via Corpora corpora@list.elra.info a écrit :
Dear All,
I am looking for treebanks (of any kind; dependency, constituency, LFG, HPSG, …) with good – preferably manual – unambiguous annotation of coordinate structures, for any language.
A typical UD treebank does not have a good annotation of coordinations, because vanilla UD does not distinguish between dependents of single conjuncts, as in I [came and [bought a book]], and shared dependents of conjuncts, as in I [[saw and bought] a book]. Enhanced UD can in principle make this distinction, but many EUD treebanks are automatically converted from vanilla UD treebanks, so this information is also often not available or not reliable. On the other hand, many constituency treebanks (including PTB) do not have explicit information about governors of coordinations (in I bought John and Mary interesting books the governor of John and Mary is bought and not, say, books), and – perhaps surprisingly – it is often not easy to guess the governor. So I am looking for treebanks that wear both kinds of information – about shared dependents and about governors – on their sleeves.
Thanks, best, Adam P. _______________________________________________ Corpora mailing list -- corpora@list.elra.info https://list.elra.info/mailman3/postorius/lists/corpora.list.elra.info/ To unsubscribe send an email to corpora-leave@list.elra.info