🏷️ DET: determiner
DET (determiner) is the Universal Dependencies part-of-speech tag for words that modify a noun or noun phrase and fix its reference in context: articles, demonstratives, quantifiers and, in many languages, possessive and interrogative words that accompany a noun.
What it is
In Universal Dependencies, every word gets one tag from a closed inventory of seventeen universal parts of speech (Nivre et al. 2020). DET is the tag for words that sit next to a noun and tell the hearer which or how many referents are meant. They do not add descriptive content. The word the in the cat does not describe the cat. It signals that the speaker expects the listener to identify it.
The tag goes back to the twelve-category universal tagset of Petrov, Das and McDonald. That set already had DET beside NOUN, PRON, ADJ and NUM, and it was built by mapping many treebank-specific tagsets onto shared categories (Petrov et al. 2012). UD v1 extended that set to seventeen tags and kept DET unchanged (Nivre et al. 2016).
UD treats determiners as function words. They depend on the content word they modify, which here is the noun (Marneffe et al. 2021). Their typical relation is det. Some treebanks use subtypes such as det:poss or det:predet, since languages may refine universal relations but not invent new basic ones (Nivre et al. 2020).
Borderline cases
| Contrast | Typical decision |
|---|---|
| DET vs PRON | A word inside a noun phrase that modifies the noun is DET (this book). A form standing alone as an argument is often PRON (I like this), but practice varies by treebank. |
| DET vs NUM | Cardinal numbers are NUM with nummod. Quantifiers like many, some and every are DET. |
| DET vs ADJ | Words with descriptive content or full adjectival behaviour (other, such in some treebanks) may be ADJ. |
| Possessives | English my and your are tagged PRON with Poss=Yes in the main treebanks. Many languages tag possessives as DET. |
Features
Determiners usually carry PronType, which says which kind they are: Art (article), Dem (demonstrative), Ind, Tot, Neg, Int or Rel. Articles also carry Definite (Def or Ind). Determiners in inflecting languages agree with their noun in Number, Gender and Case. Possessives carry Poss, Person and sometimes the possessor's number. This feature inventory descends from Interset, a universal set of morphosyntactic features designed as an interlingua between tagsets (Zeman 2008).
Examples
- English the dog, a dog, every dog, which dog?: each determiner is DET and attaches to dog by
det. - English all the dogs: all is a predeterminer, attached as
det:predetin some treebanks. - French le chat, la maison, ces livres, mon ami: articles, demonstratives and possessives are all DET and agree with the noun.
- German dem Mann: the article marks dative case, singular, masculine.
- Italian del pane: the partitive is a contraction of preposition plus article. It is split into ADP and DET in the annotation.
Across languages
Languages differ in how much they grammaticalize reference. English and the Romance languages have obligatory articles. Many Slavic languages have none, so their DET words are mostly demonstratives, quantifiers and possessives. Some languages mark definiteness with a suffix rather than a separate word, as Swedish huset ('the house') does. In such languages definiteness shows up as a feature on the NOUN and not as a DET token. UD splits contracted forms and clitics into syntactic words while keeping the original tokens recoverable (Nivre et al. 2016). French du ('of the') and similar fused forms therefore yield a DET token of their own.
In treebanks and research
In current UD releases, determiners make up a noticeably larger share of tokens in French than in English. This reflects the obligatory article of French, which appears even before plural and mass nouns.
In both languages the relation det overwhelmingly dominates. The secondary relations show how the languages differ. In English, a DET word that stands alone as a phrase most often takes nmod, then obj and nsubj, as in both or each used on their own, with some discourse uses as well. In French, the next most common relations are fixed, for determiners frozen inside multiword expressions, then advmod and reparandum, the latter for self-corrections in spoken data.
Because determiners are frequent and short, the DET versus PRON decision has a measurable effect on cross-treebank agreement. Work on annotation consistency in UD examines exactly such tag disagreements (Marneffe et al. 2017). One proposal compares UPOS trigram distributions across treebanks of the same language with a symmetric divergence score, so that it does not rely on a trained tagger (Aggarwal & Zeman 2020).
For part-of-speech tagging and dependency parsing, the coarse DET class is part of why parsers can be transferred across languages. A delexicalized English parser using the universal tags transferred directly to other languages better than the unsupervised methods of its time (Petrov et al. 2012). Introductory accounts of UD's principles and tooling cover these tag choices for annotators (Zeman 2021).
The relations DET words take most often
The five most frequent in each language's treebanks, by tokens.
The numbers
| Pair | Tokens |
|---|---|
| English: det | 85,373 |
| English: nmod | 848 |
| English: obj | 612 |
| English: nsubj | 556 |
| English: discourse | 369 |
| French: det | 138,049 |
| French: fixed | 532 |
| French: advmod | 353 |
| French: reparandum | 329 |
| French: obj | 74 |
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-det/