🌐 aux: auxiliary
aux (auxiliary) is the Universal Dependencies relation that attaches a function word to the predicate it modifies. The function word marks tense, mood, aspect, voice, polarity or evidentiality, for example English has, will, can and was.
What it is
In Universal Dependencies (UD), aux is one of the typed relations that link the words of a sentence into a tree. The scheme makes content words the heads and lets function words depend on them. That is why an auxiliary hangs below the main verb instead of governing it (Marneffe et al. 2021).
The head of aux is the predicate of the clause. Usually this is a lexical verb. In a copular clause it can be an adjective or a noun, because the nonverbal predicate is itself the head. The dependent is a word tagged AUX, one of the closed set of universal part-of-speech tags (Nivre et al. 2016; Nivre et al. 2020).
The design has a purpose. Placing function words under content words makes predicate–argument structure easier to read off the tree. It also makes sentences in different languages look more alike (Marneffe et al. 2021). The approach goes back to the universal Stanford dependencies, from which UD grew (Marneffe et al. 2014; Nivre et al. 2016).
Neighbouring relations
aux has to be told apart from several close relations:
- cop: a copula (English be in "she is tired") attaches as cop, not aux. If another auxiliary appears in the same clause ("she has been tired"), has is aux and been is cop, and both attach to the adjective.
- xcomp: verbs that take a verbal complement ("want to leave") are lexical heads. The second verb is their xcomp; it is not an aux relation.
- mark: subordinating particles and infinitival to are mark.
- advmod: English not is a modifier (advmod), even though it sits inside the verb group.
Subtypes
Languages may refine universal relations with subtypes but may not invent new basic ones. aux:pass is cited as a standard example of such a subtype (Nivre et al. 2020). As a matter of common annotation practice, some treebanks also use aux:tense for tense auxiliaries and aux:caus for causative auxiliaries. French uses all three.
Examples
| Sentence | aux edge(s) |
|---|---|
| She has left. | aux(left, has) |
| The letter was written yesterday. | aux:pass(written, was) |
| You might have been right. | aux(right, might), aux(right, have); cop(right, been) |
In the third sentence, all the function words attach directly to the adjective. They do not form a chain, so the result is a flat fan of dependents under the predicate.
Across languages
How much aux is used depends on how a language builds its verb phrases. Three cases are common:
- Languages that use many analytic forms, such as English modals and perfects, or the French passé composé with avoir/être, produce many aux edges.
- Agglutinative languages often express the same meanings as suffixes. These leave no separate word to attach.
- In some languages the negative element is itself an auxiliary verb, as with Finnish ei, and is annotated aux. Its English counterpart is an advmod.
Treating the word rather than the morpheme as the unit of annotation is what produces these differences in morphology versus syntax (Marneffe et al. 2021).
In the current English and French treebanks, the great majority of aux edges have a verbal head. In English, adjectival and nominal heads follow at a distance, reflecting copular clauses with modals or perfects. An auxiliary heading another auxiliary is rare and mostly arises in ellipsis. French auxiliaries are proportionally less frequent than English ones, and almost all of them attach to verbs.
Questions of where auxiliary-like strategies stop and lexical predication begins are still being worked out (Nivre 2025; Nivre et al. 2026).
In treebanks and research
The content-head choice was argued for at length as a coherent typological stance, not a convenience (Marneffe et al. 2021). It is taught as one of the core UD principles (Zeman 2021). Its separation of function words from core arguments fits the scheme's treatment of subjects and objects (Zeman 2017).
For dependency parsing, aux is usually among the easier relations. It is short-range and strongly signalled by the AUX tag. Errors still occur where tagging is uncertain, for instance on English have and get, or French être as auxiliary versus copula. Studies of annotation consistency across UD corpora examine such label confusions (Marneffe et al. 2017).
Multilingual parsers trained on treebanks for the CoNLL shared tasks benefited from aux being defined the same way across languages (Nivre et al. 2020; Marneffe et al. 2021). In applications, the enhanced representations keep basic aux edges intact. Downstream tools such as relation extraction and semantic role labeling can then skip auxiliaries and connect the main predicate directly with its arguments (Schuster & Manning 2016).
What aux joins: the most frequent head tag → dependent tag
The five most frequent in each language's treebanks, by tokens.
The numbers
| Pair | Tokens |
|---|---|
| English: VERB → AUX | 39,785 |
| English: ADJ → AUX | 1,244 |
| English: NOUN → AUX | 800 |
| English: AUX → AUX | 213 |
| English: ADV → AUX | 156 |
| French: VERB → AUX | 24,223 |
| French: NOUN → AUX | 195 |
| French: ADJ → AUX | 183 |
| French: PRON → AUX | 31 |
| French: AUX → AUX | 16 |
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-aux/