UD for Gujarati 
Gujarati is an Indo-Aryan language originating from the western Indian state of Gujarat. The language is widely spoken by over 56 million speakers and is one of the 22 languages with official status in India. Yet, the Gujarati Computational Linguistics community is still in its infancy. Earlier literature classifies Gujarati in the “Scraping-Bys” category (category 1) in their taxonomy indicating a scant availability of labeled datasets.
Tokenization and Word Segmentation
- In general, words are delimited by whitespace characters. Description of exceptions follows.
- Punctuation marks are tokenized as separate tokens (words), even when typographically attached to a neighboring word.
- Postpositions and case markers that are written fused to their host are split off as separate syntactic words using multiword tokens (MWT), e.g. સભ્યતાથી → સભ્યતા (NOUN) + થી (ADP), મારાથી → મારા (PRON) + થી (ADP), and દેશને → દેશ + ને.
- Compounds written as one orthographic word may also be split into multiple words, e.g. વૈવિધ્યપૂર્ણ → વૈવિધ્ય + પૂર્ણ.
- Postpositions written separately (e.g. સાથે, પ્રત્યે, વિષે, હેઠળ) are separate tokens from the start.
Morphology
Gujarati morphology is agglutinative and has a rich system of inflectional and derivational morphology. The language has a complex system of verb conjugation, noun declension, and postpositions.
Tags
- Gujarati uses all 17 universal POS categories, including particles (PART).
- PART is used for emphatic and discourse particles such as જ, પણ (“also”), ભલે (concessive), and the negators ન and નહિ.
- CCONJ covers coordinators such as અને, અથવા, પણ (“but”), and છતાંપણ.
- SCONJ covers subordinators and complementizers such as કે, કારણકે, and the correlative તો in conditionals.
- Gujarati has the following auxiliary verbs (AUX):
- છે che and થવું thavuṁ are present and past equivalents of “to be”.
- ન na is the negative auxiliary.
- રહેવું rahevuṁ “to remain” expresses progressive aspect.
- આવવું āvavuṁ “to come” expresses perfect aspect.
- ગયું gayuṁ “to go” expresses perfect aspect.
- દેવું devuṁ “to give” expresses perfect aspect.
- હતું hatuṁ expresses past tense.
- જોઈતું joītuṁ “to want” expresses modality (wish or necessity).
- પડવું paḍavuṁ “to fall” expresses necessity.
- હોવું hovuṁ “to be” expresses necessity; its conditional forms (e.g. હોત) are also analysed with this lemma.
- શકવું śakavuṁ “can” expresses possibility, ability.
- શું śuṁ is used in interrogative clauses.
- Vector verbs in compound verbs (e.g. નાખવું in કચડી નાખું) that are not on the auxiliary list are tagged VERB and attached with
compound:svc.
Features
The treebank annotates the following morphological features, all of which are enabled for Gujarati in the UD validator.
| Feature | Values | Used on |
|---|---|---|
| Case | Nom, Acc, Dat, Erg, Gen, Ins, Loc, Voc | NOUN, PROPN, PRON |
| Gender | Masc, Fem, Neut | NOUN, PRON, DET, VERB, AUX |
| Number | Sing, Plur | NOUN, PRON, DET, VERB |
| Person | 1, 2, 3 | PRON, VERB |
| PronType | Prs, Dem, Rel, Int, Ind, Tot | PRON, DET |
| Poss | Yes | possessive PRON |
| Tense | Past, Pres, Fut | VERB |
| Mood | Ind, Imp, Sub | VERB, AUX |
| Aspect | Perf, Hab | VERB |
| VerbForm | Fin, Inf, Part, Conv | VERB |
| ExtPos | ADP | fixed multiword expressions |
| Typo | Yes | heads of goeswith relations |
Notes on feature use:
- Conditional (counterfactual) forms such as કરત, થાત, and હોત are annotated as
Mood=Sub. - Conjunctive participles (ઈને forms) are
VerbForm=Convand carry no Mood. - Infinitives, including ablative infinitives such as જોવાથી, are
VerbForm=Inf. - Postpositions (ADP) carry no features; case is marked on the host noun or pronoun.
- AUX tokens carry no VerbForm; PROPN tokens carry only Case and Gender is not annotated.
- Perfective transitive verbs take an ergative subject (
Case=Erg) and agree in Gender and Number with the object.
Syntax
Standard dependency relations are used, except for clf which is not used in Gujarati.
- Gujarati is a head-final, or left-branching language.
- Adjectives precede nouns, direct objects come before verbs, and there are postpositions.
- The word order of Gujarati is SOV, and there are three genders and two numbers.
- Subjects are frequently dropped (pro-drop) and no empty node is added for them.
Constructions
- Postpositional phrases: the postposition is a
casedependent of its noun; the noun isoblof a verb ornmodof a noun. - Conjunctive participles (ઈને forms) are
advclof the main verb. - Ablative infinitives (e.g. જોવાથી, હોવાથી) express cause and are
advcl. - Conditionals (e.g. તું આવ્યો હોત, તો …): the protasis is
advclof the apodosis, and તો is amarkon the protasis. - Correlative and relative clauses (જે…તે, જ્યાં…ત્યાં, જ્યારે…ત્યારે): relative clauses are
acl:relcl(oradvclfor adverbial correlatives), and the correlative word is adet,advmod, or pronoun in the main clause. - Light verb and compound verb constructions (e.g. પ્રયત્ન કરવું, સહન થવું): the nominal is
objorcompoundof the verb; vector verbs areauxorcompound:svc. - Direct speech introduced by a verb of saying is attached with
parataxis. - Vocatives (e.g. ભાઈ) are
vocative; discourse particles such as કેમ arediscourse. - Genitives (મારો, તેની, વારસાનો) are
nmod:possand agree in Gender and Number with the possessed noun.
Treebanks
There are 2 Gujarati UD treebanks: