home edit page issue tracker

UD for Pāli

Pāli is an Indo-Aryan language, widely studied as the scriptural and liturgical language of Theravāda Buddhism.

Orthography

• While Pāli can traditionally be written with a variety of scripts, for UD we use the IAST Romanization standard.
• If your corpus or text is in another script or alternate Romanization, you can use a tool like aksharamukha to automatically convert it to IAST.
• As different scripts use different, and sometimes no, punctuation, we further standardize our orthography by removing all punctuation from our texts.
• IAST is case insensitive. Here, we use lowercase, with the first letter of each sentence capitalized by convention.

Sentence Segmentation

While we strip out punctuation in our # text rows, we do recommend using any punctuation in your source to help segment into sentences. In the Pāli Text Society and Mahāsaṅgīti editions, you should generally use both semicolons (;) and periods (.) to split sentences. However, be aware that Pāli verses may sometimes use semicolons to mark pāda “line” boundaries which are not semantic. Always segment sentences by syntax.

Tokenization and Word Segmentation

• Words are generally delimited by spaces.
• Compounds are split and sandhi undone as it is the custom in Sanskritic linguistics.
• Compounds receive multiword token ranges (e.g. 10-12) and should get the appropriate compound relation subtype. Only proper Pāli compounds get MWTs. We treat attached particles (like “-ti” and “-pi”) as sandhi and not as a MWT. Note that we don’t use the Compound=Yes feature and rely on the MWT and compound deprel to mark compound words.
• Sandhi should be marked with a SpaceAfter=No in the earlier token’s MISC and do not get a range id.
• Multitoken words (e.g. words with spaces inside them) are generally not used. Parts of a multitoken name (e.g. “Saccaka Nigaṇṭhāputta”) should get the flat:name dependency relation.

Morphology

Part of Speech Tags

Participles are tagged as VERBs.

Sometimes in Pāli, a word derived from a participle has taken on a new meaning beyond its root verb. Nouns like buddha (“the Buddha”, from bujjhati, “knows”), carita (“behavior”, from carati, “walks”), khādanīya (“non-staple food”, from khādati, “chews”), etc or adjectives like anāgata (“future”, from āgacchati, “comes”), etc should be tagged as NOUN or ADJ as appropriate and not as VERB despite the etymological connection. Participles that are used in the same sense as their underlying verb, are just VERBs, even if they are syntactically close to being used as adjectives or nouns.

When a form of bhavati or atthi is used in a copula, it should get the AUX UPOS. When they are used in the intransitive sense of “exists,” they are VERBs.

Because Pāli is originally an oral language, we strip out symbols and punctuation, meaning we do not use the SYM or PUNCT tags.

Determiners

If a pronoun such as imaṃ is used as a determiner (i.e. it gets deprel det pointing to a substantive) then it should get the UPOS DET instead of PRON. If it’s used as a pronoun in the sentence, then it should get the PRON UPOS and the appropriate deprel (such as nsubj).

Particles

The following particles get the PART UPOS:

• The quote-ending particle ti
  • LEMMA: iti
  • DEPREL: mark to the head of the quoted clause
  • The preceding token should usually get SpaceAfter=No (as appropriate)
• The negation particle na / n’
  • The lemma is na
  • Mark it as Polarity=Neg in FEATS
  • Give it the advmod relation to the negated head.
• api / pi
  • The lemma is api
  • It should normally get the discourse dependency relation when used rhetorically, but can get advmod if used conjunctively.
• eva and kho
  • Normally get the discourse dependency, except in fixed phrases
• kira
  • When used as “apparently” it should also get the discourse relation.
  • Note that “kira” is not a particle but a VERB if it is the imperative 2nd sg or the reflexive aorist 1st sg of kirati.

Lemmatization

We generally follow the lemmatization found in the Pāli Text Society’s dictionaries: The Pali-English Dictionary and A Dictionary of Pāli.

Verb Lemmas

The LEMMA for all verb forms is the 3rd-person singular present indicative of the underlying verb. For example, the past participle gata “gone” gets the LEMMA gacchati.

For passive and causative voiced finite verbs, we give the passive and causative lemma and do not collapse the lemmas down to the simplex. For example, for the passive verb nirujjhanti, the lemma should be nirujjhati and for the causative verb desemi, the causative lemma deseti should be given. Similarly, for a double causative verb like yojāpetvā, the lemma should be yojāpeti. Note that passive participles get the active verb as their lemma, for example karaṇīya gets the lemma karoti.

Forms of the verb hoti (honti, ahosi, etc) always get the lemma bhavati. Forms of the verb atthi (amhi, atthu, santi, etc), get the lemma atthi.

Noun Lemmas

Lemmas are always lowercase, even for proper nouns.

For personal pronouns, collapse the cases down to the nominative. For example, amhākaṃ gets mayaṃ and te gets tvaṃ. Keep first and second, singular and plural separate.

For demonstrative pronouns, collapse the cases, gender, and number to the nominative, singular, neuter form:

• Use tad for all ta- forms: sā, sa, taṃ, tena, etc.
• Use etad for all eta- forms: eso, etassa, esā, etc.
• Use yad for all ya- forms: yo, yā, yaṃ, yamhi, etc.
• Use idaṃ for its various forms: ayaṃ, imaṃ, imāni, imehi, anena, etc.

Similarly, use kiñci for forms of the indefinite pronoun: koci, kāci, kenaci, kesuci, etc.

Substantives and adjectives should get the stem form as their lemma. For example, itthiyo has the lemma itthi, bhikkhave gets the lemma bhikkhu, rasso the lemma rassa, rāja the lemma rājan, bhagavato bhagavant, etc.

Features

Some forms in Pāli are ambiguous. For example, a feminine noun of the -ā class with a -āya suffix might be Singular Instrumental, Ablative, Genitive, Dative, or Locative! When Case can be inferred from context, feel free to mark only the semantically correct case. If multiple parses are reasonable in a given context, list the plausible values in alphabetical order, separated by commas (e.g. if you have an -āya noun, and all but instrumental are reasonable parses, mark it Case=Abl,Dat,Gen,Loc|Gender=Fem|Number=Sing, though such cases should be exceedingly rare).

Nouns

Nouns in Pāli have Case, Gender, and Number.

We use Case=Dat for the “possessive dative” construction: Tassa tayo puttā atthi, “For him there are three sons.” = “He has three sons.” or Na me ācariyo atthi, “There is no teacher for me.” = “I have no teacher.”

Verbs

Finite Verbs

Finite verbs should have the VerbForm=Fin along with Mood, Tense, Person, Number, and Voice.

UD Pāli supports the custom, double causative (Voice=CauCau) voice. See Voice for details.

Absolutives

Absolutives (sometimes called “gerunds”) formed in Pāli with -tvā / -ya get the VerbForm=Conv feature and are attached to the main verb via advcl.

Infinitives

Infinitives in Pāli end in -tuṃ. They are noted with VerbForm=Inf and are typically xcomp.

Participles

Participles in Pāli are VERBs with the following FEATS:

• VerbForm=Part
• Voice=Act or Voice=Pass as appropriate
• Full nominal agreement features (Case, Gender, Number)

Injunctives

In Pāli, the injunctive (prohibitive) mood is formed with “mā” + the verb in aorist form (with or without augment). Here, we annotate the verb as simply aorist (Mood=Ind|Tense=Past) and we mark the injunctive by adding to the mā particle Mood=Imp|Polarity=Neg. We do not use the Sanskrit jussive (Mood=Jus).

Syntax

Since all the participles have VERB as their UPOS, they cannot be used as amod. They are either acl if they modify a noun or advcl if they modify a verb or a clause.

We have two subtypes of the obl relation:

• obl:agent
• obl:goal

We do not (at the moment) use enhanced dependency graphs, so DEPS should always be _.

We have a number of unique deprels for our compound subtypes. See the compound doc pages for details on those.

Pāli conjunctions vā (“or”) and ca (“and”) are cc (coordinating conjunctions) pointing to the previous entry in the list if that previous entry is in the same sentence. If the sentence as a whole is being coordinated with another sentence, the conjunction should just point to the root.

When a name is followed by their job (title, etc), the name is the main noun and the occupation (etc) is an appos. For example, in the genitive clause “vipassissa bodhisattassa” bodhisattassa is an appos pointing back to vipassissa which acts as the main noun. Note that this is different from Pāli compounds, where the last token of the compound is its head.

Treebanks

There is one Pāli UD treebank:

• Pali-PaliCanon

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

.