nitinvwaran/carmls-hi

Hindi SNACS (Semantic Network of Adposition and Case Supersenses; Schneider et al., 2018) annotation scheme and guidelines.

CARMLS Hindi

This is a centralized repository of the work of the CARMLS research group on SNACS (Semantic Network of Adposition and Case Supersenses) for Hindi.

Currently, this includes the entire annotated Nanhā Rājkumār (The Little Prince) in Hindi.

Some open guidelines issues are stored (expectedly) as issues on the repo.

Some differences from English STREUSLE annotations

1. Irregular pronouns directly receive the SNACS label. This applies to genitive, dative, and accusative case forms. It's hard to find a theory to extract the genitive post-position from the irregular form.
2. 'Regular' pronouns which can be split into oblique form + post-position, have been re-tokenized into the two tokens, and SNACS label applied to the post-position. This includes case markers [ne, se, mem] and accusative/dative ko and genitive [ka,ke,ki] - basically when these markers are obvious in the morphology.
3. New PART lexcat for tokens with UD tag PART. The exception is negative particles [nahin, na] which get ADV lexcat.
4. Some MWE adpositions can't be validated using the lemma forms, as the tokens [ki, ke] seem to be lemmatized into [ka] but the MWE is always [ki tarah] or [ke/ki upar]. I haven't heard any case myself like [ka upar] or [ka tarah]. So MWE lexlemmas are validated against the raw token, instead of the lemma forms.
5. NONSNACS label has been removed.