UD for Kullvi
Kullvi UD Treebank (KDTB) | Linguistic annotation overview
Kullvi is an Indo-Aryan language spoken primarily in the Kullu region of Himachal Pradesh, India. The Kullvi UD Treebank (KDTB) provides token-level part-of-speech and dependency annotations following the Universal Dependencies (UD) framework. Its language-specific XPOS tags follow the Bureau of Indian Standards (BIS) POS tagset.
Dataset inspected: kfx_kdtb-ud-test.conllu (test split); 429 sentences and 4,601 integer-ID syntactic tokens. The observations below describe this supplied file, not necessarily every split of the complete treebank.
Tokenization and Word Segmentation
-
Words are segmented into individual CoNLL-U token rows; word boundaries in the sentence text are generally whitespace-delimited.
-
Punctuation is annotated as a separate token when present; e.g., “तुसरा घोर कोखे सा ?” has “?” tagged PUNCT with relation punct.
-
Grammatical and discourse elements are separately represented when annotated: “बे” is tagged PART/discourse in “हाऊँ चोली घोरा बे”.
-
Postpositions are represented as ADP and commonly attached with the case dependency relation.
Morphology
Tags
The supplied treebank uses two POS layers: UPOS (Universal POS tags) and XPOS (BIS-based language-specific tags).
| UPOS | Tokens | Representative XPOS |
|---|---|---|
| NOUN | 954 | NN |
| VERB | 712 | VM |
| PRON | 495 | PRP, WQ |
| ADP | 484 | PSP |
| AUX | 406 | VAUX |
| ADV | 391 | RB |
| PUNCT | 338 | SYM |
| ADJ | 291 | JJ |
| PART | 237 | NEG, RP |
| NUM | 90 | QC |
| CCONJ | 79 | CC |
| PROPN | 77 | NNP |
| SCONJ | 39 | CC |
| DET | 8 | DT |
Examples from annotated sentences: “तुसरा” → PRON / PRP; “घोर” → NOUN / NN; “कोखे” → ADV / RB; “सा” → VERB / VM (sentence 1); “रा” → AUX / VAUX (sentence 5). The category of a surface form is determined by the individual annotation and context.
Auxiliary Verbs (AUX)
-
The dataset distinguishes verbal heads (VERB) from auxiliaries (AUX). In “ओज़ बड़ा भारी गाश लागी रा”, “लागी” is VERB/root and “रा” is AUX/aux.
-
“सी” is annotated AUX/aux in “आसे रोहा सी हिमाचल प्रदेश रे कुल्लू ज़िला न”.
-
Copular constructions also occur: “सा” is AUX/cop in “हिमाचल री राजधानी शिमला सा , जो घूमने री बाँकी जगह सा।”
Modal Auxiliaries and Phase Verbs
The test file contains AUX annotations, but its FEATS fields do not specify a complete modal/phase-verb inventory. A complete language-wide modal or phase verb list should therefore be established by manual linguistic review rather than copied from Mandyali.
Compound and Light-Verb Constructions
The relation compound appears in the file, including “चोली” (VERB/compound) in “हाऊँ चोली घोरा बे”. However, compound also labels proper-name sequences (e.g., “हिमाचल प्रदेश”), so not every compound arc denotes a light verb. The distinction between lexical verbs, light verbs and auxiliaries needs construction-by-construction review.
Features
The morphological FEATS column (column 6 of CoNLL-U) is “_” for all 4,601 syntactic tokens in the supplied test file. Therefore, Gender, Number, Person, Case, Tense, Aspect, Mood and other UD morphological features cannot be reported as annotated features of this dataset on the available evidence. These may be added in a later annotation pass.
Syntax
The test annotations use standard UD relations to encode clausal heads, nominal arguments, modifiers, complements, coordination and discourse structure. Kullvi is typically described as having an SOV tendency, but no corpus-wide word-order analysis was performed for this document.
-
Core arguments: nsubj marks nominal subjects (e.g., “पाहुने” in “मेरे पाहुने लोड़ी आए”); obj marks direct objects (e.g., “गोशटे” in “होरी बे ज्ञान आपू बे गोशटे ।”).
-
Adpositional phrases: case marks adpositions and obl marks many oblique dependents.
-
Clausal embedding: “कि” is SCONJ/mark and “चले” is VERB/ccomp in “हाऊँ बिल्कुल भी नहीं चाहदीं थी कि तेईबे मेरे गूँगे होने रा पता चले।”
-
Nonverbal predicates: cop is present, as in “हिमाचल री राजधानी शिमला सा , जो घूमने री बाँकी जगह सा।”
-
Discourse: particles such as “बे” may receive PART/discourse.
Dependency relations observed in the supplied split (28 types): acl, advcl, advmod, amod, appos, aux, case, cc, ccomp, compound, conj, cop, csubj, dep, det, discourse, iobj, mark, nmod, nsubj, nummod, obj, obl, parataxis, punct, root, vocative, xcomp.
Treebanks
There is only one Kullvi UD treebank:
- Kullvi-KDTB