🌐 amod: adjectival modifier
amod (adjectival modifier) is the Universal Dependencies relation that links a nominal head to an adjective or adjectival phrase that modifies its meaning, as in "red car". The adjective is the dependent and the noun or pronoun is the head.
What it is
In Universal Dependencies, a sentence is a tree of typed relations between words. amod is the label for one common case: an adjective that narrows or describes a noun it attaches to. In "a cold morning", the word morning is the head and cold is its amod dependent.
The theory behind UD builds its syntax from three kinds of units. Nominals name entities, clauses name events, and modifiers specify both. The relation label records what kind of phrase the dependent is, so the type of a nominal's modifier can be read off the label alone: amod for an adjectival one, acl for a clausal one, nmod for a nominal one (Marneffe et al. 2021). The label goes back to the cross-linguistic typology of universal Stanford dependencies (Marneffe et al. 2014). It was then carried into the UD v1 inventory of 40 relation types (Nivre et al. 2016) and kept in the v2 inventory of 37 universal types (Nivre et al. 2020).
Neighbouring relations
The boundaries matter more than the core case. In textbook terms, amod differs from several nearby relations:
| Relation | Typical dependent | Contrast with amod |
|---|---|---|
| acl | participle or clause with its own arguments | "the man sitting there" |
| det / nummod | determiner / cardinal number | "this car", "three cars" |
| nmod | nominal | "the roof of the house" |
| compound | noun used as a modifier | "steel bridge" |
A predicative adjective is not amod. In "the car is red", red is the head of the clause and car is its nsubj. Languages may add their own subtypes to universal relations (Nivre et al. 2020). Subtyped forms of the type amod:subtype are rare, however, and most treebanks use plain amod.
Examples
- "She bought a new phone." The relation is amod(phone, new).
- "We fixed the broken window." A bare participle used attributively is still amod: amod(window, broken).
- "Il porte une chemise blanche." In this French example the adjective follows the noun: amod(chemise, blanche).
A pronoun can also be the head, as in "something strange", annotated amod(something, strange). An adjective can modify another adjective-headed nominal, as in "the poor old", where the noun is missing.
Across languages
English places attributive adjectives before the noun. Its amod edges are overwhelmingly noun-to-adjective. Smaller groups have a proper noun as head ("ancient Rome"), a verb-tagged participle as dependent, a pronoun as head, or an adjective as head.
French shows the same dominance of noun heads with adjective dependents. Most of its adjectives follow the noun, and they agree with it in gender and number. A minority of edges have a numeral-tagged dependent, which reflects tagging choices for ordinal-like words.
These cross-language differences are in word order and morphology, not in the relation itself. That is the parallelism the scheme aims for (Marneffe et al. 2021).
Languages with a small or absent adjective class pose a harder question. There, property concepts are often expressed with verbs or relative clauses, and such modifiers fall under acl rather than amod. How UD treats predication that departs from the prototype is itself still being refined (Nivre et al. 2026).
In treebanks and research
UD treebanks store amod in the CoNLL-U format, together with lemma, part-of-speech tag and features (Nivre et al. 2016). Because the label usually pairs a NOUN head with an ADJ dependent, its counts are a natural check on annotation. Inconsistency checks of UD corpora examine exactly such recurring head–dependent patterns (Marneffe et al. 2017).
For dependency parsing, amod is one of the easier relations to predict. The adjective usually sits next to its head and is marked by a distinctive tag. The CoNLL shared tasks of 2017 and 2018 evaluated parsers on raw text in many UD languages, amod included (Marneffe et al. 2021).
Enhanced English UD keeps the basic tree in full and only adds edges or refines labels (Schuster & Manning 2016). As a result, amod edges carry over unchanged. Shared modifiers can also be spread to coordinated nouns, so that in "new cars and trucks" the adjective modifies both nouns.
Outside parsing, amod defines a useful context slot for word meaning. One study of language change parses diachronic corpora with Stanza and compares the adjectival-modifier fillers of a target word across periods using Jensen-Shannon divergence (Tat et al. 2026). For English plane, modifiers such as military rose and horizontal declined. This traces the spread of the aircraft sense, and each such detected shift can be checked against the specific collocates that drove it (Tat et al. 2026).
What amod 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 → ADJ | 37,846 |
| English: PROPN → ADJ | 2,548 |
| English: NOUN → VERB | 2,005 |
| English: PRON → ADJ | 507 |
| English: ADJ → ADJ | 298 |
| French: NOUN → ADJ | 37,356 |
| French: PROPN → ADJ | 1,304 |
| French: NOUN → NUM | 403 |
| French: PRON → ADJ | 374 |
| French: NUM → ADJ | 99 |
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-amod/