🌐 list: list
list is a Universal Dependencies relation for chains of comparable items that are laid out side by side, as in addresses, contact details or itemised records, without forming a grammatical sentence or a coordination. The first item is the head, and each later item attaches to it.
What it is
Universal Dependencies (UD) annotates every word of a sentence with one head and one typed relation. Its basic tree uses content words as heads (Marneffe et al. 2021). Most relations describe ordinary syntax: subjects, objects, modifiers. Real text also contains material that is not built from clauses at all. Examples are a signature block, a business-card line, or a string of names and numbers under a heading. The relation list covers these chains of parallel items.
The label already appeared in the original inventory of 40 dependency types (Nivre et al. 2016). It is kept among the 37 universal relation types of version 2 (Nivre et al. 2020). The design goes back to the cross-linguistic typology of Stanford-style dependencies (Marneffe et al. 2014).
Head and dependent
The head is the first item of the chain. Every following item, however many there are, attaches directly to that first item with list. Items do not attach one to the next. This flat "all items on the first" convention follows standard UD guideline practice. Each item keeps its own internal structure, so a multiword name inside an item still uses flat or compound internally.
Neighbouring relations
| Relation | Used when |
|---|---|
| conj | items are coordinated within a sentence, usually with or without a conjunction such as and |
| parataxis | clauses are juxtaposed in running text |
| appos | the second nominal renames or identifies the first |
| flat | parts of one name or one expression are joined without a clear head |
The key test is whether the items stand in running grammatical discourse. If they do, conj or parataxis applies; list is reserved for non-sentential layouts. No universal subtypes of the form list:subtype are in common use. Languages may define subtypes in principle, but not new basic relations (Nivre et al. 2020).
Examples
In the lines below, the arrow runs from the head to the dependent.
Maria Lopez , Main Street 4 , Springfield
Lopez → Street : list Lopez → Springfield : list
The name is the head. The street address and the town each depend on it with list.
Tel. 0123 4567 , fax 0123 4568
Tel. → fax : list
Results : Oslo 3 , Bergen 1 , Tromsø 2
Oslo → Bergen : list Oslo → Tromsø : list
In the last line, both later entries hang on the first entry, not on each other.
Across languages
How often list is used depends heavily on the genre of the treebank and on each team's guidelines.
English web and email corpora contain many addresses, sign-offs and contact lines. In current English data the relation mostly joins proper nouns to proper nouns. Proper noun to common noun, noun to noun, numeral to numeral and proper noun to numeral pairs follow behind. That profile matches names, places, phone numbers and dates.
French treebanks, by contrast, do not use the label at all in the current release. Comparable material there is either absent from the texts or analysed with other relations. This shows that the label tracks text type and annotation decisions more than grammar.
Whether such a split counts as a different strategy for the same construction is the kind of question raised in work on constructions and strategies in UD (Nivre 2025). It also touches on clauses that lack ordinary predication (Nivre et al. 2026). See the English and French topic pages for the treebanks themselves.
In treebanks and research
In dependency parsing, list is a rare label. Parsers therefore see few training examples of it, and they often confuse it with conj, appos or parataxis. UD data served as the basis for the CoNLL shared tasks of 2017 and 2018, in which parsers processed raw text for many languages (Marneffe et al. 2021). In those tasks such low-frequency labels count in the labelled attachment scores like any other relation.
Uneven use of list across treebanks of related languages is a typical target for consistency checking. Methods for assessing annotation consistency across UD corpora (Marneffe et al. 2017) and tutorials on UD principles and tools (Zeman 2021) give the framework for such audits.
Downstream, list edges matter for information extraction and named entity recognition from semi-structured text. A record of name, address and phone number becomes one star-shaped subtree. The enhanced UD representation adds edges for conjuncts and control (Schuster & Manning 2016). It does not treat list items as conjuncts, so their attachment stays as in the basic tree.
What list joins: the most frequent head tag → dependent tag
The five most frequent in each language's treebanks, by tokens.
The numbers
| Pair | Tokens |
|---|---|
| English: PROPN → PROPN | 411 |
| English: PROPN → NOUN | 245 |
| English: NOUN → NOUN | 229 |
| English: NUM → NUM | 225 |
| English: PROPN → NUM | 223 |
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-list/