🌐 nmod: nominal modifier
nmod (nominal modifier) is the Universal Dependencies relation that attaches a noun phrase to another nominal, which it modifies or completes. Examples are the possessor in «Mary's book» and the noun introduced by «of» in «the roof of the house».
What it is
In Universal Dependencies (UD), every word of a sentence is attached to one head by a typed relation. The head is always the content word, and function words such as prepositions hang from it (Marneffe et al. 2021).
The relation nmod covers one specific pairing. The head is a nominal: a noun, proper noun, pronoun or numeral. The dependent is also a nominal, one that narrows down or completes the head. If a preposition or case marker introduces the dependent, that marker attaches to the dependent noun as case. It does not mediate the link between the two nouns.
UD describes syntax with three kinds of phrasal units: nominals, clauses and modifiers. The relation label tells you which kind of unit the dependent is, so phrase types can be read off the labels without a separate layer (Marneffe et al. 2021).
Neighbouring relations
nmod is defined by its contrasts with nearby labels:
- obl: a nominal that depends on a verb, adjective or adverb rather than on a noun. In v1 such dependents were also labelled nmod; version 2 gave them their own label.
- amod: an adjective modifying a noun.
- acl: a clause modifying a noun.
- nummod and det: numbers and determiners attached to a noun.
- compound and flat: noun–noun combinations treated as word-like units or names.
The basic UD inventory has 37 relation types. Languages may refine them with subtypes but may not invent new basic categories (Nivre et al. 2020).
Subtypes
Common subtypes of nmod include:
- nmod:poss for possessors.
- nmod:tmod for bare temporal nouns.
- nmod:npmod for other bare noun-phrase modifiers. Some recent releases merge these bare-noun subtypes under different names.
Enhanced representations add another kind of subtype. They copy the preposition into the label, as in nmod:on. Multiword prepositions are fused into one label, giving a direct nmod:in_front_of edge between house and hill (Schuster & Manning 2016).
Examples
| Sentence | Relation | Note |
|---|---|---|
| the roof of the house | nmod(roof, house) | case(house, of) |
| Mary's book | nmod:poss(book, Mary) | 's is a separate word |
| the meeting on Monday | nmod(meeting, Monday) | the head is a noun |
| She left on Monday | obl(left, Monday) | the head is a verb, so not nmod |
The possessive 's in English is split off as its own syntactic word even though spelling joins it to the noun. This follows UD's lexical integrity principle (Marneffe et al. 2021).
Across languages
How a language marks noun-on-noun modification determines how nmod looks in its treebanks:
- Languages with morphological case, such as Czech, Finnish or Russian, often express the dependent noun with a genitive ending and no adposition.
- English and French rely mostly on prepositions (of, de). In English, possessive pronouns are also analysed as nominal dependents.
- French possessives (mon, sa) are determiners, so they do not appear under nmod.
This difference is visible in the English and French UD treebanks. Noun-to-noun links dominate in both languages, and more strongly in French. In English, the next most frequent pairing is a pronoun dependent on a noun, followed by a proper noun dependent. In French, proper nouns and numerals come next, and pronouns barely figure. In both languages, nmod accounts for a similar modest share of all tokens, slightly higher in French.
The choice between nmod and the argument relations raises its own cross-linguistic questions. Zeman examined this in his work on what counts as a core argument (Zeman 2017). Nivre has framed it in terms of constructions and the strategies languages use to express them (Nivre 2025). Language-specific needs are met through subtypes and additional morphology features rather than new labels (Nivre et al. 2020).
In treebanks and research
nmod descends from the cross-linguistic typology of universal Stanford dependencies proposed by de Marneffe, Dozat, Silveira, Haverinen, Ginter and Nivre (Marneffe et al. 2014).
The first UD guidelines appeared in October 2014, with 40 dependency types (Nivre et al. 2016). Release v1.2 (November 2015) covered 37 treebanks in 33 languages (Nivre et al. 2016). By v2.7 (November 2020) the collection had grown to 183 treebanks for 104 languages (Marneffe et al. 2021).
The need for a shared label set was stark before UD. For parallel Swedish, Danish and English sentences, no two annotation schemes agreed on more than 40% of dependencies (Nivre et al. 2016).
Parsing
In dependency parsing, nmod versus obl is a classic attachment ambiguity. A prepositional phrase after a noun may belong to the noun or to the verb, and only the head decides the label. The CoNLL shared tasks of 2017 and 2018 trained parsers on UD for many languages directly from raw text (Marneffe et al. 2021).
Annotation consistency across treebanks, including for frequent relations like this one, has been measured and corrected in follow-up work (Marneffe et al. 2017). Teaching material on UD principles and tools covers the label alongside its neighbours (Zeman 2021).
Applications
For information extraction and natural language understanding, preposition-specific enhanced labels give direct edges between content words. This makes relations like nmod:of immediately usable. Converters producing these graphs ship with Stanford CoreNLP (Schuster & Manning 2016).
What nmod joins: the most frequent head tag → dependent tag
The five most frequent in each language's treebanks, by tokens.
The numbers
| Pair | Tokens |
|---|---|
| English: NOUN → NOUN | 24,155 |
| English: NOUN → PRON | 15,144 |
| English: NOUN → PROPN | 13,422 |
| English: PROPN → PROPN | 2,771 |
| English: NUM → NOUN | 965 |
| French: NOUN → NOUN | 38,833 |
| French: NOUN → PROPN | 12,858 |
| French: NOUN → NUM | 3,148 |
| French: PROPN → PROPN | 1,730 |
| French: NUM → NOUN | 1,479 |
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…
- Dependency parsing — Dependency parsing is the task of recovering a sentence's syntactic structure as directed, labelled links between its words. Each word is…
More papers
Topics: 🌐 Universal Dependencies · 🌳 Dependency parsing
Shelves: 🗣️ Language and Linguistics
This page in Markdown → https://md.research.elamur.ai/free/en/concepts/ud-nmod/