🌐 advcl: adverbial clause modifier

Universal Dependencies · 6 key papers · all concepts

Part of: 🗨️ Natural language processing · 🌐 Universal Dependencies

advcl (adverbial clause modifier) is the Universal Dependencies relation that attaches a subordinate clause to a predicate it modifies. The clause works like an adverbial and expresses circumstances such as time, cause, condition, purpose, concession or manner.

Also: adverbial clause modifier, advcl:relcl, advcl:cmp

What it is

In Universal Dependencies (UD), a sentence is a tree of typed relations between words. Content words act as heads, and function words such as auxiliaries and prepositions depend on them (Marneffe et al. 2021).

The label advcl links two predicates:

• The head is the predicate of the clause being modified. This is usually the main verb, but it can be an adjective or noun when that word carries the predication.
• The dependent is the predicate of the adverbial clause.
• The subordinator itself (because, when, if) hangs from the dependent as mark.

A copula never serves as head, because the predicate word heads its clause. That is why pairs like VERB → ADJ also occur.

The relation descends from the Universal Stanford Dependencies, which UD adopted as its starting typology (Marneffe et al. 2014). In UD the label tells you what kind of dependent is attached. Clauses denote events and modifiers specify them, so advcl says "a clause in a modifying role" (Marneffe et al. 2021).

Two design axes separate advcl from its neighbours (Nivre et al. 2016):

• Phrase versus clause. advcl is the clausal counterpart of advmod.
• Core argument versus other dependent. Adverbial clauses are non-core, which separates them from complement clauses (Zeman 2017).

Languages may add subtypes but not new basic categories (Nivre et al. 2020). Some treebanks use advcl:relcl for relative clauses that modify a whole clause, and advcl:cmp for comparative clauses.

The enhanced representation for English adds the marker to the relation label (nmod:after is one example), and the same mechanism extends to clausal modifiers (Schuster & Manning 2016).

Examples

In the sentences below, the dependent is in brackets and the arrow runs from head to dependent.

1. She left early [because she was tired]. advcl(left → tired). The word tired heads its clause because the copula was depends on it. because is mark(tired).
2. [If it rains], we will stay home. advcl(stay → rains), with mark(rains → If).
3. He read the letter, [smiling]. advcl(read → smiling). This is a non-finite clause with no marker.

Compare "the letter [that he read]", which is acl:relcl on letter. Compare also "I know [that he left]", which is ccomp on know.

Across languages

UD aims for parallel annotation across typologically very different languages (Marneffe et al. 2021). Languages differ mainly in how they mark the adverbial clause:

• English and French mostly use a subordinating conjunction before a finite clause (because, quand, si) or a non-finite form. French uses the gérondif for this, as in "en marchant".
• Turkic languages, Japanese and others rely on converbs and case-marked verbal nouns. These are also annotated as advcl, and the morphology carries what a conjunction would carry elsewhere.
• Clauses without an ordinary verbal predicate are harder to fit into the scheme. Nonprototypical predication has received separate treatment (Nivre et al. 2026).
• More broadly, these choices can be seen as language-specific strategies for one shared construction (Nivre 2025).

In the current English and French treebanks, advcl covers a little over one percent of tokens. Verb-to-verb links dominate in both languages, and adjective-headed links come second. French also shows a noticeable share of nouns heading adverbial clauses, typically nominal predicates in copular sentences.

In treebanks and research

UD grew from a few treebanks into a collection covering more than a hundred languages (Marneffe et al. 2021). It supplied the data for the CoNLL 2017 and 2018 shared tasks on parsing raw text (Nivre et al. 2020).

advcl is a known source of difficulty in dependency parsing, for three reasons:

• Attachment height is ambiguous. In a sentence with several clauses, it is often unclear which predicate the adverbial clause modifies.
• The boundary with neighbouring labels is fuzzy. advcl can be confused with xcomp, with conj, and with acl when the clause is non-finite.
• Annotation varies across treebanks. Annotation-consistency studies find such boundary cases among the places where UD corpora diverge (Marneffe et al. 2017).

For downstream tasks such as semantic role labeling and relation extraction, the enhanced graphs are especially useful. Marker-specific labels and propagated subjects make the link between a subordinate verb and its arguments explicit (Schuster & Manning 2016). Practical introductions to the guidelines and tools cover these conventions for newcomers (Zeman 2021).

See also treebanks, syntax, English and French.

Key papers

The works this page cites, the most cited first; citations as counted by OpenAlex plus the papers of this library that cite the work, read 2026-10-24.

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-advcl/