home edit page issue tracker

UD for Nepali

Tokenization and Word Segmentation

• Nepali words are generally delimited by whitespace. Each orthographic word written in Devanagari is normally treated as one token.
• Punctuation marks such as ।, ,, ?, !, quotation marks, ellipses and dashes are tokenized separately and annotated as PUNCT.
  • Example: जंगलमा जाने , चराउने ।
• Quotation marks and paired punctuation are treated as separate tokens on each side of the quoted span.
  • Example: " मलाई एकदम थकाई ला छ , जिउ दुखि रहेछ । "
• Reduplicated or hyphenated forms written as one orthographic unit are kept as one token.
  • Example: ह्वार-ह्वार in अनि ह्वार-ह्वार बलेछ रे ।
• Case and postpositional morphology may be written in two ways in the current data:
  • When case/postpositional material is fused with the nominal stem, it is kept inside the same token and represented with morphological features such as Case=Loc, Case=Gen, Case=Abl, Case=Ins, or Case=Erg.
    • Examples: जंगलमा, कर्तव्यको, धनबाट, विचारले, मान्छेले.
  • When postposition-like elements are written as separate orthographic tokens, they are tokenized separately and annotated as ADP, normally with the relation case.
    • Examples: ले, लाई/लाइ, सम्म.
• Discourse and focus particles such as रे, त, नै, पनि, केरे, and चाहिँ appear as independent tokens. They are normally tagged as PART and attached with discourse.
• Foreign or Sanskrit quotation material written in Devanagari is segmented word by word. Such tokens are tagged as X and receive Foreign=Yes when appropriate.
  • Example: न्याययतेन द्रव्यम् कर्तव्यम् पारलौकिकम् ।
• There are no multiword-token ranges in the current treebank data.

Morphology

Tags

The Nepali UD treebank uses the following universal POS tags:

• ADJ, ADP, ADV, AUX, CCONJ, DET, INTJ, NOUN, NUM, PART, PRON, PROPN, PUNCT, VERB, X

The tags SCONJ and SYM do not occur in the current data.

Tag usage

• ADJ is used for adjectives that modify nouns or occur as nonverbal predicates.
  • Examples: महान्, पूर्वीय, गहन, प्राकृतिक, न्यायपूर्ण.
• ADP is used for separate postposition-like tokens.
  • Examples: ले, लाई, लाइ, सम्म.
• ADV is used for lexical adverbs expressing manner, degree, time, or discourse connection.
  • Examples: समेत, सधैं, एकदम, त्यसपछि, अर्थात, यसर्थ.
• AUX is used for auxiliaries and copular forms when they have a grammatical function.
  • Examples: छ, छन्, हो, रहेछ, भएका.
• CCONJ is used for coordinating conjunctions.
  • Examples: र, वा, अनि.
• DET is used for determiners modifying a following noun.
  • Examples: त्यो गोठ, एउटा मान्छे, यस कुरा.
• INTJ is used for interjections, especially in the spoken narrative data.
  • Examples: ए, ल.
• NOUN is used for common nouns, including nouns with fused case/postpositional morphology.
  • Examples: कर्तव्य, अधिकारभन्दा, ग्रन्थमा, कुरालाई, जंगलमा, मान्छेले.
• NUM is used for cardinal numerals.
  • Example: एक.
• PART is used for discourse, focus, emphasis, evidential, and reportative particles.
  • Examples: रे, त, नै, पनि, केरे, चाहिँ.
• PRON is used for independent pronouns and pronominal forms.
  • Examples: मलाई, हाम्रो, जुन, त्यसलाई, के, आफ्ना.
• PROPN is used for proper names.
  • Examples: खड्केरिरहा, विवेकान्नंद, जर्ज, इलियट.
• PUNCT is used for punctuation, including Nepali danda ।, commas, quotation marks, ellipses and dashes.
• VERB is used for lexical verbs, participles, converbs, infinitives, and verbal predicates.
  • Examples: गर्छन्, भनिएको, गर्नुपर्छ, फर्केर, आएको, घस्दिएछ.
• X is used for foreign or Sanskrit quotation material that is not analyzed internally.
  • Examples: न्याययतेन, द्रव्यम्, कर्तव्यम्, पारलौकिकम्.

AUX vs. VERB

The distinction between AUX and VERB is based on grammatical function.

• AUX is used when a form of हुनु or another auxiliary-like form contributes tense, aspect, mood, evidentiality, or copular meaning rather than introducing the main lexical predicate.
  • Examples: छ, छन्, हो, रहेछ, भएका.
• VERB is used for the lexical predicate that carries the main event or state meaning.
  • Examples: गर्छन्, भनिएको, आएको, गर्नुपर्छ, घस्दिएछ.
• In verbal complexes, the lexical verb is the head and the auxiliary attaches with aux.
  • Example: in भनिएको छ, भनिएको is VERB and छ is AUX.
• In nonverbal clauses, copular forms such as छ and हो are tagged AUX and attached with cop to the nominal or adjectival predicate.
  • Example: in कर्तव्य अधिकारभन्दा महान् छ ।, महान् is the predicate and छ is cop.

DET vs. PRON

The distinction between DET and PRON is syntactic.

• A form is tagged DET when it modifies a noun.
  • Examples: त्यो गोठ, एउटा मान्छे, यस कुरा.
• The same or similar form is tagged PRON when it functions independently as a nominal argument.
  • Examples: त्यसलाई, जुन, मलाई, के.

Deverbal forms

Participial and converbal forms are normally tagged as VERB when they preserve verbal syntax.

• Participial modifiers are tagged VERB and attached with acl .
  • Examples: भनिएको, गरिएको, गर्ने.
• Converbs are tagged VERB and attached with advcl.
  • Examples: फर्केर, आएर, खर्चेर.
• Deverbal lexicalized nominal or adjectival forms may be tagged according to their syntactic distribution when they behave like ordinary nouns or adjectives.

Features

The current Nepali treebank uses the following morphological features:

• Aspect, Case, Evident, Foreign, Gender, Mood, NumType, Number, Person, Polarity, PronType, Reflex, Tense, VerbForm, Voice.

Only features attested in the current data are listed here.

Nominal and pronominal features

• Case is one of the central features in the Nepali data. It is used both for unmarked nominative forms and for forms with fused case/postpositional morphology.

Value	Meaning	Examples
Nom	nominative	कर्तव्य, खुशी
Acc	accusative/direct object, often morphologically unmarked	काम, घ्यू, कर्तव्य, कर्म, सोच (could be marked with लाई in certain situations)
Dat	dative	मानिसलाई, कसैलाई
Gen	genitive	कर्तव्यको, हाम्रो, उसको
Loc	locative	ग्रन्थमा, जंगलमा, गोठमा
Abl	ablative/source/comparative	अधिकारभन्दा, धनबाट
Ins	instrumental	विचारले, तरिकाले
Erg	ergative/instrumental subject marking	मान्छेले, चिन्तकले

• Number is used for singular and plural nominal and verbal forms.
  • Number=Sing: कर्म, पैसा, कथा.
  • Number=Plur: विचारहरू, उपनिषद्हरू.
• Gender occurs only where it is morphologically or lexically relevant.
  • Attested values include Masc, Fem, and Neut.
• PronType is used for pronouns and determiners.
  • PronType=Dem: demonstratives such as यो, त्यो, यस.
  • PronType=Prs: personal pronouns and possessive pronouns such as हाम्रो, उसको.
  • PronType=Int: interrogative forms such as के.
  • PronType=Rel: relative forms such as जुन.
• Reflex=Yes is used for reflexive forms such as आआफ्नो.
• NumType=Card is used for cardinal numerals such as एक.

Verbal features

• VerbForm is used to distinguish finite, participial, converbal and infinitival forms.

Value	Meaning	Examples
Fin	finite verb	गर्छन्, हुन्छ, छ
Part	participle	गरेको, गरिएको, भएको
Conv	converb	फर्केर, आएर, खर्चेर
Inf	infinitive	गर्न, चराउन

• Tense is used for present, past and future forms.
  • Tense=Pres: छ, गर्छन्, हुन्छ.
  • Tense=Past: भएको, गरिएको, रहेछ.
  • Tense=Fut: future forms attested in the narrative data.
• Aspect is used for imperfective, perfective and progressive meanings.
  • Attested values: Imp, Perf, Prog.
• Mood is used for indicative, imperative and necessity forms.
  • Mood=Ind: ordinary statements.
  • Mood=Imp: imperatives such as हेर्नुहोस्.
  • Mood=Nec: necessity forms such as गर्नुपर्छ.
• Person and Number are used mainly on finite verbs and auxiliaries.
  • Example: गर्छन् has third person plural agreement.
• Polarity=Neg is used for negative forms such as हुदैन and नहुने.
• Voice=Pass is used where passive morphology is present.
• Evident=Nfh occurs in the narrative data, especially with reportative or non-firsthand forms.

Other features

• Foreign=Yes is used for foreign or Sanskrit material in quotations.
• Nepali has honorific and politeness distinctions in the language, especially in pronouns and verb agreement. These distinctions are visible linguistically, but no separate Polite feature is used in the current data.

Syntax

Basic clause structure

Nepali is a head-final language. The normal constituent order is SOV, and the finite verbal predicate or verbal complex normally occurs at the end of the clause.

• Example: यस्तो पस्चात्य चिन्तकले पनि कर्तव्यको महानतालाई उदघाटित गरेका छन् ।

Core arguments

Subjects (nsubj)

Canonical subjects are annotated as nsubj. They may be unmarked nominative NPs or case-marked NPs.

• Example: कर्तव्य अधिकारभन्दा महान् छ ।
  • कर्तव्य is nsubj.
• Nepali also shows ergative/instrumental subject marking with -ले, especially with perfective or agentive predicates. Such subjects are still annotated as nsubj.
  • Examples: मान्छेले, चिन्तकले, विचारले.

Direct objects (obj)

Direct objects are annotated as obj when they are the core patient/theme argument of a lexical verb.

• Example: in त्यो घ्यू घस्दिएछ, घ्यू is obj.
• Bare objects and case-marked object-like nominals are treated according to their syntactic role in the clause.

Indirect objects (iobj)

Dative or recipient-like arguments are annotated as iobj when they are core arguments of the predicate.

• Example: कसैलाई पनि धारा पानीको भुत लाग्दैन ।
  • कसैलाई is iobj.
• Dative-marked or लाई-marked arguments that behave like recipients, experiencers, or affected participants are often annotated as iobj.

Obliques (obl)

Non-core arguments are annotated as obl. These include locative, ablative, instrumental, temporal, source, manner and other adverbial nominals.

• Examples: जंगलमा, गोठमा, धनबाट, अधिकारभन्दा, रूपमा, तरिकाले.
• The current data uses plain obl rather than language-specific oblique subtypes such as obl:loc or obl:tmod.

Copula and nonverbal clauses

In nonverbal clauses with nominal or adjectival predicates, the lexical predicate is the syntactic head and the copular auxiliary is attached with cop.

• Example: कर्तव्य अधिकारभन्दा महान् छ ।
  • महान् is the root predicate.
  • छ is attached as cop.
  • कर्तव्य is nsubj.
  • अधिकारभन्दा is obl.
• Example: कर्तव्यप्रतीको निष्ठा वा भक्ति उपासनाको सर्वोच्च रूप हो ...
  • रूप is the nominal predicate.
  • हो is attached as cop.

In verbal predicates, auxiliary forms such as छ and छन् attach to the lexical verb with aux.

• Example: in भनिएको छ, भनिएको is the verbal head and छ is aux.

Relational constructions

Participial and relative clause modifiers

Participial clauses modifying a noun are attached with acl.

• Examples: भनिएको, गरिएको, गर्ने.

Converbs and clause chaining

Converb forms are annotated as adverbial clauses with advcl.

• Examples: फर्केर, आएर, लागेर, खर्चेर.

Coordination and parataxis

Coordinating conjunctions such as र, वा, and अनि are annotated as cc, and the non-initial conjunct is attached with conj.

Loosely sequenced main clauses, especially in narrative passages, may be connected with parataxis when they are not clear cases of coordination or subordination.

Discourse and special relations

Discourse particles such as त, नै, रे, केरे, चाहिँ, and पनि are attached with discourse.

• Example: in भन्छ रे, रे is attached with discourse.

The following subtyped relations occur in the current Nepali data:

• acl:relcl – relative clauses.
• compound:redup – reduplicated compounds, e.g. repeated or echo-like expressions.
• flat:name – multi-token proper names.
• nmod:poss – possessive nominal modifiers.

Other important relations used in the data include advcl, advmod, amod, appos, aux, case, cc, ccomp, compound, conj, cop, dep, det, discourse, dislocated, iobj, nmod, nsubj, nummod, obj, obl, parataxis, punct, reparandum, root, and xcomp.

Treebanks

There is 1 Nepali UD treebank:

• Nepali-BK

© 2014–2024 Universal Dependencies contributors. Site powered by Annodoc and brat

.