🏷️ PROPN: proper noun
PROPN (proper noun) is the Universal Dependencies part-of-speech tag for nouns, or noun-like words, that name a specific individual, place, organization or other unique entity rather than a class, such as Paris, Marie or Toyota.
What it is
In Universal Dependencies (UD), every word gets one tag from a closed list of 17 universal part-of-speech (UPOS) categories (Nivre et al. 2020). PROPN is the tag for names.
A common noun like city applies to any member of a class. A proper noun like Lyon refers to one particular entity. Proper nouns refer without describing: what they pick out does not follow from a dictionary sense of the word. In many writing systems they are capitalized. Capitalization is only a hint, though, not the test.
UD treats nominals as the units that denote entities (Marneffe et al. 2021). PROPN marks the subset of nominals whose job is to name. That makes it a useful signal for named entity recognition and coreference resolution.
Origin in the universal tagset
The universal tagset of Petrov, Das and McDonald had twelve coarse categories with no separate tag for names. Proper and common nouns were merged under NOUN (Petrov et al. 2012). UD v1 extended and revised that set to 17 tags, and PROPN, together with AUX, was among the additions (Nivre et al. 2016). The split restores a distinction that many source tagsets already drew, such as NNP in the Penn Treebank.
Borderline cases
- PROPN vs NOUN: common nouns inside a name (bank in Bank of England) are handled differently across treebanks. Some tag every part of the name PROPN; others keep the word's ordinary category.
- PROPN vs ADJ: adjectives derived from names (English French, German deutsch) are ADJ, not PROPN.
- PROPN vs X: foreign-language material that is not used as a name falls under X. UD v2 also offers the Foreign feature for such words (Nivre et al. 2020).
- Months, weekdays and genericized brand names (a hoover): whether these are PROPN or NOUN varies by language and treebank.
Features and relations
Proper nouns carry the nominal features of their language: Number, and where relevant Gender and Case. Abbr=Yes marks abbreviations of any kind, acronyms such as NATO included (Nivre et al. 2020). The feature inventory descends from Interset, the interlingua of morphosyntactic features designed for tagset conversion (Zeman 2008; Nivre et al. 2016).
Syntactically, a proper noun can fill any nominal role. Examples are subject (nsubj), object (obj), oblique (obl) and nominal modifier (nmod).
Multiword names are joined with flat, the relation for multiword expressions without internal head structure. In UD v2 it is often subtyped as flat:name. The first word of the name is the head, and the following words attach to it.
English names also often stack as noun modifiers (Obama administration). The compound relation, used for noun–noun compounding, captures this. In English, the possessive 's is split off as a separate word, so the name itself keeps its plain form (Marneffe et al. 2021).
Examples
- English: Marie/PROPN Curie/PROPN moved to Paris/PROPN. Here Marie is the head of the name, Curie attaches to it by flat (or flat:name), and Paris is obl.
- French: Jean-Paul Sartre est né à Paris. Both name parts are PROPN.
- German: Angela Merkels Rede. Where case is annotated on names, the token Merkels carries Case=Gen.
- Czech: v Praze ('in Prague'). The lemma is Praha, and the form carries Case=Loc.
- Finnish: Helsingissä ('in Helsinki'). The PROPN takes the inessive case ending.
Across languages
Languages differ mainly in how much morphology proper nouns carry, and in whether the script gives cues.
- In Czech or Finnish, names decline like other nouns, so a PROPN token rarely equals its lemma.
- English and French mark names mostly by capitals.
- Chinese, Japanese, Arabic and Hebrew scripts have no capitals. Name detection then rests on context, which can make PROPN one of the harder tags in part-of-speech tagging for these languages.
In treebanks and research
The boundary between PROPN and NOUN depends on annotator decisions. It is therefore a common source of inconsistency within and across treebanks. Systematic checks of UD corpora have looked for annotation inconsistencies in general (Marneffe et al. 2017).
One proposed measure compares UPOS trigram distributions between two treebanks of the same language. It builds on the asymmetric KLcpos3 measure of Rosa and Žabokrtský (Aggarwal & Zeman 2020):
θpos(A,B) = KLcpos3(A,B) + KLcpos3(B,A)
In this formula, A and B are the two treebanks. KLcpos3 is a Kullback–Leibler divergence over trigrams of universal POS tags. A trigram missing from one treebank is given a count of one, so the divergence stays defined.
Summing both directions yields a symmetric score, where a low value suggests consistent annotation. A score of 0.5 or less counts as consistent whatever the genre mix (Aggarwal & Zeman 2020). The score also rises with genre differences and with smaller data. Disagreements over whether names are PROPN or NOUN may therefore contribute to such divergences, but the measure does not isolate them (Aggarwal & Zeman 2020).
For multilingual dependency parsing, the coarse universal tags made direct transfer of delexicalized parsers across languages workable (Petrov et al. 2012). All UD languages share the same tag inventory, PROPN included. This supports training multilingual models and cross-lingual transfer (Nivre et al. 2020).
The relations PROPN words take most often
The five most frequent in each language's treebanks, by tokens.
The numbers
| Pair | Tokens |
|---|---|
| English: nmod | 17,048 |
| English: flat | 8,249 |
| English: nsubj | 7,600 |
| English: obl | 7,307 |
| English: compound | 7,239 |
| French: nmod | 15,000 |
| French: flat | 11,976 |
| French: nsubj | 8,044 |
| French: obl | 6,960 |
| French: conj | 3,908 |
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-propn/