🏷️ NOUN: noun

Universal Dependencies · 7 key papers · all concepts

Part of: 🗨️ Natural language processing · 🌐 Universal Dependencies

The noun (NOUN) tag is the universal part-of-speech (UPOS) tag in Universal Dependencies for common nouns. These are words that name people, animals, things, places, substances or abstract ideas. They typically head nominal phrases that serve as subjects, objects and oblique nominals (for example, the noun inside a prepositional adjunct).

Also: noun, common noun, UPOS NOUN, substantive

What it is

In part-of-speech tagging, every word receives a category label. In Universal Dependencies (UD), NOUN is one item in a closed list of 17 UPOS tags (Nivre et al. 2020). The guidelines describe it as the class of common nouns: words for entities, typically used as the head of a nominal phrase.

UD's theory has three basic units. Nominals refer to entities and clauses to events, while modifiers qualify either of the two (Marneffe et al. 2021). NOUN is the prototypical head of the first unit.

Borderline cases

The tag is defined against its neighbours:

• PROPN (proper noun) covers names such as Paris or Anna. The older 12-tag universal set had no separate class for names (Petrov et al. 2012). UD v1 introduced PROPN (Nivre et al. 2016); see the history below.
• PRON (pronoun) covers words that stand in for nominals without naming a class of entities, such as she or something.
• VERB (verb) is the competitor for forms like the English -ing word. The building collapsed has a NOUN. In Building houses is hard, building is VERB, because it takes an object.
• ADJ (adjective) competes with nominalised adjectives (the rich). These usually stay ADJ unless the noun use is lexicalised, as with German der Abgeordnete 'deputy'.
• NUM (numeral) competes with counting words like dozen or million. Language-specific guidelines settle these cases.

Features

The morphology layer pairs each word with its lemma and a set of standardised features (Nivre et al. 2020). The UD features descend from the Interset inventory. Interset was designed as an interlingua of features such as part of speech, gender, number and case (Zeman 2008; Nivre et al. 2016). Typical features on nouns are:

NounClass was among the features added in v2 (Nivre et al. 2020).

Examples

English: The dog chased two cats across the garden. Here dog (nsubj, nominal subject), cats (obj, object; Number=Plur) and garden (obl, oblique nominal) are NOUN. The is DET (determiner) and across is ADP (adposition).

French: La maison de mon frère est grande 'my brother's house is big'. Maison (Gender=Fem) and frère are NOUN. Frère attaches to maison as nmod (nominal modifier).

Russian: Я читаю книгу 'I am reading a book'. Книгу is NOUN with Case=Acc, Gender=Fem and Animacy=Inan.

English compound: data center. Data is a NOUN that depends on center through the compound relation.

Across languages

Languages differ mainly in how much grammar sits on the noun itself.

• English marks little beyond number. It uses word order and prepositions instead.
• Finnish or Hungarian load case onto the noun.
• Semitic languages add definiteness and construct state.
• Bantu languages add noun class.

UD's basic unit is the word, not the morpheme. Clitics such as English 's are split off as separate words (Marneffe et al. 2021). As a result, a NOUN token corresponds to a syntactic word in every language, although the features it carries differ.

Under UD's content-word-head principle, prepositions and case markers depend on the noun rather than the reverse (Marneffe et al. 2021). This is why a noun inside a prepositional phrase still shows up as obl or nmod.

In treebanks and research

The tag goes back to the universal tagset of Petrov, Das and McDonald, which had twelve coarse categories. The authors mapped 25 treebanks in 22 languages onto it (Petrov et al. 2012). They reported that an English dependency parser lost only about 0.6% in parsing accuracy when it used the 12 coarse tags rather than the 45 Penn Treebank tags. They also showed that a delexicalised parser transferred directly from English outperformed the best unsupervised methods then available (Petrov et al. 2012).

UD v1 adopted this set and extended it to 17 tags, adding among others PROPN and AUX (auxiliary) (Nivre et al. 2016). UD-annotated data, including UPOS tags, served as the basis for the CoNLL 2017 and CoNLL 2018 shared tasks on multilingual dependency parsing (Marneffe et al. 2021).

The annotation consistency of the UD corpora has been examined directly (Marneffe et al. 2017). For several treebanks of the same language, Aggarwal and Zeman compare distributions of UPOS trigrams with a symmetrised divergence (Aggarwal & Zeman 2020):

θpos(A,B) = KLcpos3(A,B) + KLcpos3(B,A)

Here A and B are two treebanks. KLcpos3 is the asymmetric divergence between their UPOS-trigram frequency distributions, a measure introduced by Rosa and Žabokrtský (2015). θpos is the symmetric sum of the two directions (Aggarwal & Zeman 2020).

The authors derived the interpretation thresholds empirically, and the cut-off depends on genre:

• 0.5 for pairs of the same genre;
• 2.0 for pairs of different genres;
• a further 2.0 when adulterant genres are present.

Whatever the genre mix, a value of 0.5 or less counts as consistent and 4.0 or more as inconsistent. Values in between are judged against the genre-specific cut-off (Aggarwal & Zeman 2020). The measure can also help locate where a discrepancy between two treebanks comes from (Aggarwal & Zeman 2020).

Key papers

The works this page cites, the most cited first; citations as counted by OpenAlex plus the papers of this library that cite the work, read 2026-10-24.

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-noun/