🌐 ccomp: clausal complement

Universal Dependencies · 7 key papers · all concepts

Part of: 🗨️ Natural language processing · 🌐 Universal Dependencies

ccomp (clausal complement) is the Universal Dependencies relation linking a predicate to a dependent clause that serves as its core, object-like argument. The embedded clause has its own subject, or at least does not take its subject from the higher clause.

Also: clausal complement, complement clause, ccomp

What it is

In Universal Dependencies (UD), every word of a sentence attaches to one head word, and the attachment carries a typed label. ccomp marks one specific kind of attachment: a whole clause filling the argument slot of a verb or adjective. In "She said that he left", the clause "that he left" is what was said, so it is a clausal complement of "said".

UD treats clauses as one of three basic phrasal units, alongside nominals and modifiers. The relation label tells the reader what type of phrase the dependent is (Marneffe et al. 2021). The scheme pairs phrasal and clausal versions of the same function, such as nsubj/csubj and object/ccomp, and it separates core arguments from other dependents (Nivre et al. 2016). The core/non-core distinction itself is examined across languages in (Zeman 2017). ccomp descends from the earlier universal Stanford typology (Marneffe et al. 2014).

The arc runs from the governing predicate (head) to the predicate of the embedded clause (dependent). Function words hang from the content word they belong to (Marneffe et al. 2021). So the complementizer "that" attaches to "left" as mark, not to "said".

Neighbouring relations

The guidelines allow subtypes written relation:subtype for language-specific distinctions (Nivre et al. 2020). ccomp has no widely shared universal subtypes. Any subtypes that exist are local to particular treebanks.

Examples

• She said [that he left]. Here ccomp(said, left) and mark(left, that).
• I wonder [whether it will rain]. Here ccomp(wonder, rain).
• We are sure [they will come]. Here ccomp(sure, come), with an adjective as head.

Compare "She wants to leave". The understood subject of "leave" is "She", so the clause is xcomp, not ccomp. In English UD, direct quotations after verbs of saying are also commonly analysed as ccomp.

Across languages

In English and French treebanks, the dominant pattern by far is a verb governing a verb.

Smaller groups show other heads and dependents:

• A verb can govern a nominal or adjectival predicate. This happens when the embedded clause is copular, as in "I think he is a doctor".
• English has a noticeable share of adjective heads ("sure", "aware").
• French shows adverb heads, as in "peut-être qu'il viendra".

These differences reflect each language's complementation strategies rather than distinct relations.

Languages differ in how complement clauses look:

• In French and English, the complement is usually finite and introduced by a complementizer (que, that).
• German places the verb at the end of the clause after dass.
• Many languages build complements with nominalized or quotative forms. Annotators must then decide whether the dependent is still a clause or has become a nominal.

The distinction between constructions and the strategies languages use to express them is developed in (Nivre 2025). Borderline clause types are discussed in (Nivre et al. 2026). Practical guidance on such decisions appears in a UD tutorial (Zeman 2021).

In treebanks and research

UD began with release v1.2, covering 37 treebanks for 33 languages, in November 2015 (Nivre et al. 2016). By v2.7 in November 2020 it had reached 183 treebanks for 104 languages (Marneffe et al. 2021). ccomp is part of the inventory of 37 universal relation types in v2 (Nivre et al. 2020). Annotation of clausal relations is not always uniform across corpora, and consistency checks across UD treebanks address such variation (Marneffe et al. 2017).

Parsing

UD data underlay the CoNLL shared tasks of 2017 and 2018, which parsed raw text in many languages (Marneffe et al. 2021). ccomp is therefore a standard target label in multilingual dependency parsing.

Telling ccomp from xcomp is a known source of parser errors. The two relations share surface material, and only control separates them.

The difference matters downstream. Enhanced English UD adds an nsubj:xsubj edge to give a controlled verb its subject (Schuster & Manning 2016). A ccomp clause already contains its own subject, so it needs no such edge.

Meaning and inference

Clausal complements carry claims whose truth the speaker may or may not endorse. The CommitmentBank pulled finite clausal complements from Stanford dependency parses and then curated them by hand. It contains 1,200 discourses with 48 clause-embedding predicates, where the complement sits under negation, a modal, a question or a conditional (Marneffe et al. 2019).

Crowd ratings showed that factivity alone predicts speaker commitment weakly. Predicate, genre, tense and person all contribute (Marneffe et al. 2019). For information extraction, this means a ccomp arc identifies an embedded claim, but it does not show that the speaker asserts that claim.

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-ccomp/