🌐 det: determiner
det (determiner) is the Universal Dependencies relation that attaches a determiner, such as an article, demonstrative or quantifier, to the nominal it introduces. The head is the noun, proper noun or other nominal, and the determiner is its dependent.
What it is
In Universal Dependencies (UD), every word of a sentence is linked to one head by a typed grammatical relation. Content words act as heads and function words depend on them (Marneffe et al. 2021). The label det belongs to this design. It connects a nominal head to a word whose job is to fix the reference of the noun phrase: whether it is definite or indefinite, near or far, how many or which ones. Typical determiners are articles (the, a), demonstratives (this, those), quantifiers (every, some, no) and interrogatives (which).
UD v2 has 37 universal relation types, and languages may refine them with subtypes but may not add new basic types (Nivre et al. 2020). The det label goes back to the earlier universal Stanford dependencies, one of the schemes that UD merged (Marneffe et al. 2014; Nivre et al. 2016).
Neighbouring relations
Several relations sit close to det, and the guidelines keep them apart as follows.
- amod is used for adjectival modifiers, which describe the referent rather than identify it.
- nummod is used for cardinal numbers (three dogs), in languages where numbers are not treated as determiners.
- nmod:poss is used for possessors that are full nominals (John's book). Possessive determiners take det or det:poss, depending on the treebank.
- case is used for adpositions and case clitics attached to the nominal.
- clf is used for numeral classifiers in languages that have them.
Common subtypes
The following subtypes are textbook practice in many treebanks:
- det:poss marks possessive determiners.
- det:predet marks predeterminers such as all in all the books.
- det:numgov and det:nummod are used in Slavic treebanks, where they record whether a quantifier governs the case of its noun or agrees with it.
The enhanced++ representation for English adds det:qmod. In this representation, the quantifying part of a partitive or light-noun phrase (a bunch of people) becomes a flat unit that depends on the content noun (Schuster & Manning 2016).
Examples
Each relation below is written as relation(head, dependent).
The dog chased a cat.
det(dog, The) det(cat, a)
All these books are mine.
det:predet(books, All) det(books, these)
Quelle maison as-tu vue ? ('Which house did you see?')
det(maison, Quelle)
In the first sentence, the and a each hang from the noun they introduce, not from the verb. In the second, all precedes another determiner, so it receives the predeterminer subtype. The French example shows an interrogative determiner, which takes plain det.
A determiner that stands alone, as in I'll take this, is not det at all. In that case the word heads its own nominal and fills the argument position itself.
Across languages
Languages differ sharply in how much work det does. Languages with obligatory articles show the label constantly. Languages without articles, such as Russian, Czech, Japanese or Latin, use it mainly for demonstratives, quantifiers and possessives.
French is a clear contrast to English. French marks nearly every noun with an article, including partitive du and de la, so det makes up a larger share of French tokens than of English ones. In both languages, the noun-to-determiner pair is by far the most common use. Proper nouns are the next most common heads, as in la France or the Thames. Pronouns, numerals and adjectives head determiners much less often. Arabic and Hebrew definite prefixes are usually treated as features of the noun or segmented off. Languages where a noun can take several determiners at once rely on the predet subtype.
Two classification questions vary from treebank to treebank. The first is which possessive words count as determiners rather than nominals. The second is whether quantifiers like many count as DET or ADJ. Such divergences are among the kinds of inconsistency that cross-corpus checks of UD set out to find (Marneffe et al. 2017).
In treebanks and research
UD treebanks annotate each word on two layers. The morphological layer covers the lemma, the universal tag and features, and the syntactic layer covers the dependency (Nivre et al. 2016). As a result, det interacts with part-of-speech tagging. Most det dependents are tagged DET, but tag and relation are independent decisions, and mismatches between them are a standard target of validation (Zeman 2021).
For dependency parsing, det is one of the easiest relations to predict. It is short, local, frequent and usually signalled by a closed class of words. Parsers in the CoNLL shared tasks of 2017 and 2018, which parsed raw text in many languages on UD data, typically scored near the top on it (Marneffe et al. 2021; Nivre et al. 2020). Because det is nearly always correct, overall attachment scores can hide errors on harder relations. For this reason, researchers often report content-word scores that leave function-word relations like det aside.
In cross-lingual work, det is a useful probe of transfer. Articles are present in some languages and absent in others, and the same meaning may surface as a determiner, an affix or nothing. Typologically minded work on UD treats these as different strategies for encoding the same construction (Nivre 2025). For corpus studies, see treebanks.
What det 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 → DET | 78,156 |
| English: PROPN → DET | 4,531 |
| English: ADJ → DET | 1,317 |
| English: PRON → DET | 610 |
| English: NUM → DET | 363 |
| French: NOUN → DET | 123,687 |
| French: PROPN → DET | 8,408 |
| French: PRON → DET | 2,527 |
| French: NUM → DET | 1,343 |
| French: ADJ → DET | 1,245 |
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-det/