🌐 csubj: clausal subject
csubj (clausal subject) is the Universal Dependencies relation linking a predicate to a subordinate clause that serves as its subject. The arrow runs from the governing predicate to the head word of the subject clause, usually its verb.
What it is
In Universal Dependencies every word attaches to one head through a typed relation. Subjects come in two kinds. When the subject is a noun phrase, the relation is nsubj. When the subject is a whole clause, with its own predicate, the relation is csubj. This split between phrasal and clausal dependents has been part of the scheme since version 1, alongside dobj/ccomp and advmod/advcl (Nivre et al. 2016). It goes back to the universal Stanford dependencies (Marneffe et al. 2014).
UD takes content words as heads, and copulas and auxiliaries depend on them (Marneffe et al. 2021). So the head of csubj is the main predicate of the clause. That may be a verb, or the adjective or noun of a copular predicate. The dependent is the head of the subject clause, normally its main verb. Complementizers such as that attach to that verb as mark.
Neighbouring relations
- nsubj: a nominal subject. In English, free relatives headed by what are treated as nominals, so they take nsubj rather than csubj.
- ccomp and xcomp: clauses in object or complement position, not subject position.
- advcl: an adverbial clause, which modifies the predicate rather than filling an argument slot.
- expl: an expletive placeholder such as it, which often co-occurs with an extraposed csubj.
Zeman treats subjects, including clausal ones, as core arguments distinct from obliques and modifiers (Zeman 2017).
Subtypes
UD lets languages add subtypes but not new basic relations (Nivre et al. 2020). The most common subtype is csubj:pass, used for a clausal subject of a passive verb. It replaced the separate v1 label csubjpass. The guidelines also provide csubj:outer, which is textbook knowledge from the guidelines rather than from the papers cited here. It marks a clausal subject of a predicate that already contains its own inner subject.
Examples
| Sentence | Relation |
|---|---|
| That he lied surprised everyone. | csubj(surprised, lied) |
| It is clear that she won. | csubj(clear, won); expl(clear, It) |
| To leave now would be rude. | csubj(rude, leave) |
| That taxes rise was announced. | csubj:pass(announced, rise) |
In the second and third sentences the head is an adjective, because the copula is not the head. The second also shows extraposition: the clause follows the predicate and an expletive fills the subject slot.
Across languages
How often csubj appears depends on the language. Languages differ in how freely they allow clauses in subject position, whether they prefer extraposition with an expletive, and whether they nominalize clauses instead. Nominalized clauses would be annotated as nominals with nsubj.
The current treebanks of English and French show the relation is rare in both, with English using it somewhat more often. The dominant head types differ:
- In English, adjective heads come first, reflecting the frequent "It is ADJ that/to …" pattern. Verb heads come second and noun heads third.
- In French, verb heads lead and adjective heads follow, with nouns again third.
In both languages the dependent is overwhelmingly a verb.
Nivre's work on constructions and strategies places relations like csubj within a typological account of how languages encode clauses and their arguments (Nivre 2025). Recent work on nonprototypical predication examines clauses whose predicate is not a verb (Nivre et al. 2026). This bears directly on which word heads a csubj.
In treebanks and research
csubj is one of the 37 universal relation types of UD v2 (Nivre et al. 2020). It is annotated in the basic tree of every treebank that has clausal subjects.
In the enhanced representation, a controlled clause can receive an explicit subject edge. Examples are the infinitival subject in "To leave now would be rude", or a subordinate verb whose subject comes from the main clause. Schuster and Manning introduced labels such as nsubj:xsubj for this in enhanced English UD (Schuster & Manning 2016). Their converters derive these graphs from basic trees.
Because the label is rare, it is hard for dependency parsing. Parsers often confuse an extraposed csubj with ccomp, or an infinitival subject with xcomp or advcl. Low-frequency labels like this one also show up in studies of annotation consistency. De Marneffe and colleagues assessed how uniformly UD corpora apply their relations, which matters for rare clausal labels (Marneffe et al. 2017).
Multilingual parsing campaigns, such as the CoNLL shared tasks of 2017 and 2018, evaluated parsers on all UD relations, csubj included (Marneffe et al. 2021). Tutorials on UD principles use csubj to illustrate the clausal/nominal split in syntax annotation (Zeman 2021). For downstream tasks such as semantic role labeling, csubj marks an event-denoting argument rather than an entity.
What csubj joins: the most frequent head tag → dependent tag
The five most frequent in each language's treebanks, by tokens.
The numbers
| Pair | Tokens |
|---|---|
| English: ADJ → VERB | 500 |
| English: VERB → VERB | 381 |
| English: NOUN → VERB | 212 |
| English: VERB → ADJ | 24 |
| English: PRON → VERB | 23 |
| French: VERB → VERB | 346 |
| French: ADJ → VERB | 262 |
| French: NOUN → VERB | 128 |
| French: ADV → VERB | 26 |
| French: ADJ → ADJ | 10 |
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-csubj/