Elamur Research⌕
☰
home / concepts / acl: clausal modifier of noun

🌐 acl: clausal modifier of noun

Universal Dependencies · 6 key papers · all concepts

Part of: 🗨️ Natural language processing · 🌐 Universal Dependencies

acl (clausal modifier of noun) is the Universal Dependencies relation that attaches a clause to the nominal it modifies. Examples are relative clauses, participial and infinitival clauses, and content clauses after nouns such as "fact".

Also: clausal modifier of noun, adnominal clause, acl:relcl (relative clause subtype)

What it is

In Universal Dependencies (UD), every dependent of a nominal is labelled by the kind of phrase it is. A nominal dependent is nmod, an adjectival one is amod, and a clausal one is acl. The label therefore records the phrase type, so no separate constituent annotation is needed (Marneffe et al. 2021). This follows from UD's three basic units. Nominals denote entities, clauses denote events, and modifiers specify either of them (Marneffe et al. 2021).

The relation joins two words:

Function words such as relativizers, subordinators and infinitive markers attach inside the clause. They are not the head of the acl link, because UD treats content words as heads (Marneffe et al. 2021).

Neighbouring relations

UD consistently separates phrasal dependents from clausal ones, as in the pairs nsubj/csubj and advmod/advcl (Nivre et al. 2016). acl is the clausal partner of nmod and amod. It differs from these neighbours as follows:

Languages may refine universal relations with subtypes, but they may not add new basic categories (Nivre et al. 2020). The most widespread subtype is acl:relcl for relative clauses. Some treebanks also use acl:cleft for the clause of a cleft sentence.

Examples

  1. The book that she wrote sold well. Here acl:relcl joins book (head) to wrote (dependent), and "that" depends on wrote.
  2. A man carrying a ladder knocked. Here acl joins man to carrying, a participial clause with no relative word.
  3. Her decision to resign surprised us. Here acl joins decision to resign, an infinitival clause, and "to" attaches as a marker.

A content clause such as "the claim that prices rose" also takes plain acl (claim → rose). The clause states what the claim is, rather than restricting which claim is meant.

Across languages

Because the label is universal, the same construction can be compared across typologically different languages (Marneffe et al. 2021). Languages differ in which clause types carry most of the load.

In the English and French UD treebanks, the dominant pattern is a noun head with a verb dependent. A pronoun head with a verb dependent comes second in both languages. English shows more noun-to-adjective cases than proper-noun heads, while French has the reverse order. French uses the relation for a larger share of its tokens than English does. Plausible causes are its productive participial modifiers and infinitival constructions with "à" or "de", as in "une maison à vendre" ('a house for sale').

Languages with prenominal clauses, such as Japanese and Korean, often keep plain acl. Their noun-modifying clauses do not mark a clear contrast between relative and content clauses, so acl:relcl would impose a distinction the grammar does not draw. The underlying question of which strategies realize which constructions across languages is a current topic in UD theory (Nivre 2025). The same holds for clauses with unusual predicates (Nivre et al. 2026). Work on argument structure bears on how a relativized noun's role is recovered (Zeman 2017).

In treebanks and research

The acl relation goes back to the cross-linguistic version of Stanford dependencies (Marneffe et al. 2014). It was part of the first UD release, whose inventory had 40 dependency types (Nivre et al. 2016). Under the v2 guidelines it is one of 37 universal relations (Nivre et al. 2020).

Enhanced representation

A basic tree cannot show that the head noun of a relative clause is also an argument inside that clause. The enhanced representation is a graph that adds such implicit relations (Marneffe et al. 2021). For English it was specified in detail and released with converters in Stanford CoreNLP (Schuster & Manning 2016). That work served applications that need direct links between content words, such as relation extraction (Schuster & Manning 2016).

Parsing

In dependency parsing, acl is evaluated like any other label. It is one of the relations tested in the 2017 and 2018 CoNLL shared tasks on parsing raw text (Marneffe et al. 2021). Errors tend to come from three sources:

Uneven annotation of this kind is the subject of studies of treebank consistency in UD (Marneffe et al. 2017). Tutorials on UD principles and tools cover how such clauses should be analysed (Zeman 2021).

How often the relation acl occurs: its share of all tokens

Counted by code over the Universal Dependencies 2.18 treebanks of English and French (every token once, subtypes with their base relation).

How often the relation acl occurs: its share of all tokensEnglish treebanks: 1.43% of tokens (16,121 tokens)English treebanks1.43% of tokens (16,121 tokens)French treebanks: 2.14% of tokens (23,298 tokens)French treebanks2.14% of tokens (23,298 tokens)
The numbers
TreebanksShare
English treebanks1.43% of tokens (16,121 tokens)
French treebanks2.14% of tokens (23,298 tokens)

What acl joins: the most frequent head tag → dependent tag

The five most frequent in each language's treebanks, by tokens.

What acl joins: the most frequent head tag → dependent tagEnglish: NOUN → VERB: 11,890English: NOUN → VERB11,890English: PRON → VERB: 1,431English: PRON → VERB1,431English: NOUN → ADJ: 538English: NOUN → ADJ538English: PROPN → VERB: 526English: PROPN → VERB526English: NOUN → NOUN: 349English: NOUN → NOUN349French: NOUN → VERB: 16,944French: NOUN → VERB16,944French: PRON → VERB: 2,927French: PRON → VERB2,927French: PROPN → VERB: 1,405French: PROPN → VERB1,405French: NOUN → ADJ: 453French: NOUN → ADJ453French: NOUN → NOUN: 335French: NOUN → NOUN335
The numbers
PairTokens
English: NOUN → VERB11,890
English: PRON → VERB1,431
English: NOUN → ADJ538
English: PROPN → VERB526
English: NOUN → NOUN349
French: NOUN → VERB16,944
French: PRON → VERB2,927
French: PROPN → VERB1,405
French: NOUN → ADJ453
French: NOUN → NOUN335

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.

More papers

Topics: 🌐 Universal Dependencies · 🌳 Dependency parsing

Shelves: 🗣️ Language and Linguistics

This page in Markdown → https://md.research.elamur.ai/free/en/concepts/ud-acl/