🏷️ PRON: pronoun
PRON is the universal part-of-speech tag for pronouns in Universal Dependencies. It covers words that stand in for a noun or a noun phrase and get their reference from context, such as I, she, who, nothing or themselves.
What it is
In Universal Dependencies, every syntactic word receives one tag from a closed list of 17 universal parts of speech (Nivre et al. 2020). PRON is one of them. The guidelines describe a pronoun by what it does. It takes the place of a nominal, and its meaning comes from the discourse or the situation, not from a fixed concept. This textbook view goes back to traditional grammar.
The class is meant to include:
- personal pronouns (I, you, they)
- reflexive and reciprocal pronouns (herself, each other)
- interrogative and relative pronouns (who, what, which)
- indefinite, negative and total pronouns (someone, nobody, everything)
- demonstratives used on their own (that in "I like that")
Borderline cases
The tag sits next to several other tags, and the boundaries are where most errors happen.
- PRON vs DET. A word that modifies a noun is a determiner ("this book"). The same word standing alone is a pronoun ("this is fine").
- Possessives. English treebanks tag my or their as PRON with the feature Poss=Yes. Many other languages treat comparable possessives as DET.
- Adverbs. Pro-forms for places or times, such as there, where or then, are ADV.
- Nouns. Nominalized forms like the others usually stay pronouns, but a noun such as people does not become one.
- Expletive it and there are still PRON. Their special status is marked in the syntax, not in the part of speech.
Features
Pronouns usually carry more morphological features (FEATS) than any other class:
- PronType gives the subtype, with values such as Prs, Rel, Int, Dem, Ind, Neg, Tot and Rcp.
- Person, Number, Gender and Case describe the form itself.
- Reflex=Yes marks reflexives, and Poss=Yes marks possessives.
UD v2 added features that matter mostly for pronouns. Polite marks formal address, and Clusivity marks inclusive versus exclusive "we" (Nivre et al. 2020).
Relations
As heads of nominal phrases, pronouns attach mostly as subjects, objects and obliques. In English treebanks, nsubj is far ahead of every other relation, followed by obj and nmod, where the nmod cases are mainly possessives. In French, nsubj also leads, but expl and iobj rank much higher. This reflects reflexive "se" and impersonal "il" (expl) and the dative clitics lui and leur (iobj). Pronouns also make up a larger share of all tokens in English than in French.
Examples
- English: "She/PRON gave it/PRON to me/PRON." Here she is nsubj, it is obj and me is obl.
- English: "Who/PRON called?" The pronoun carries PronType=Int.
- French: "Il/PRON pleut." The subject il is expl, because nothing refers.
- German: "Ich sehe Sie/PRON." The formal pronoun Sie is marked Polite=Form.
- Spanish: "dámelo" ("give it to me") is split into da + me/PRON + lo/PRON.
Across languages
Languages differ in how often pronouns appear and in what form. In pro-drop languages such as Spanish or Italian, subjects are often left out, so fewer pronoun tokens occur. Japanese words for "I" and "you" behave much like nouns, and treebanks must decide where they belong.
Because UD takes the word as the basic unit, clitic and affixal pronouns are split off as separate tokens even when the spelling joins them (Marneffe et al. 2021). The original token is preserved so the text can be recovered (Nivre et al. 2016). This applies, for example, to Arabic object suffixes and Romance clitics. Case systems in languages such as Russian or German give pronouns rich Case values. Languages with inclusive and exclusive "we", such as Tagalog with tayo and kami, use Clusivity.
In treebanks and research
PRON was already one of the twelve coarse categories proposed by Petrov, Das and McDonald (Petrov et al. 2012). Their set also included NOUN, DET, ADP and others. They mapped 25 treebanks onto it, and the coarse tags supported cross-lingual part-of-speech tagging and delexicalized parser transfer (Petrov et al. 2012).
UD v1 extended that inventory to 17 tags. It took its morphological features from Interset (Nivre et al. 2016). Interset is a feature interlingua that converts tagsets through per-tagset decoders and encoders (Zeman 2008).
The PRON/DET boundary in particular varies between treebanks of a single language. This has led to work on annotation consistency (Marneffe et al. 2017). One proposal compares UPOS trigram distributions between two treebanks A and B (Aggarwal & Zeman 2020):
θpos(A,B) = KLcpos3(A,B) + KLcpos3(B,A)
KLcpos3 is an asymmetric Kullback–Leibler divergence over UPOS-tag trigrams. Adding both directions gives a symmetric score θpos. A pair of treebanks counts as consistent when θpos stays below a threshold that depends on genre. The threshold is 0.5 for treebanks of the same genre (Aggarwal & Zeman 2020).
Pronouns also matter outside tagging. They are central to coreference resolution and dependency parsing, since a correct PRON tag and its features help a system link a pronoun to the entity it refers to.
The relations PRON words take most often
The five most frequent in each language's treebanks, by tokens.
The numbers
| Pair | Tokens |
|---|---|
| English: nsubj | 72,645 |
| English: obj | 17,084 |
| English: nmod | 16,314 |
| English: obl | 5,346 |
| English: expl | 3,382 |
| French: nsubj | 41,558 |
| French: obj | 13,990 |
| French: expl | 10,146 |
| French: obl | 9,001 |
| French: iobj | 6,703 |
Key papers
Related concepts
- Universal Dependencies — Universal Dependencies is a framework and an open collection of treebanks that annotates the grammar of many languages with one shared…
- Part-of-speech tagging — Part-of-speech tagging is the task of assigning each token in a text a grammatical category, such as noun, verb, adjective or determiner…
- Morphology — Morphology is the branch of linguistics that studies the internal structure of words. It covers how words are built from meaningful units…
More papers
Topics: 🔤 Part-of-speech tagging · 🌐 Universal Dependencies
Shelves: 🗣️ Language and Linguistics
This page in Markdown → https://md.research.elamur.ai/free/en/concepts/upos-pron/