🌐 obl: oblique nominal
The oblique nominal (obl) is the Universal Dependencies (UD) relation linking a predicate to a nominal dependent that is not a core argument (subject, object or indirect object). Typical cases are case-marked or prepositional phrases of place, time, manner, instrument or source, and the agent of a passive.
What it is
In Universal Dependencies, every word in a sentence attaches to one head through a typed relation; the main predicate attaches to an artificial root. The obl relation covers nominal phrases that depend on a predicate without being core arguments. Core arguments are the subject and the objects. Obliques are everything else of nominal type: "in the garden", "on Monday", "with a knife", "by the committee". The head is usually a verb, but it can also be an adjective or adverb that acts as a predicate. The dependent is the head noun, pronoun, proper noun or numeral of the phrase.
UD makes content words the heads and attaches function words to them (Marneffe et al. 2021). So in "in the garden" the obl arc goes to "garden", and the preposition "in" hangs from "garden" as case.
The label also records the type of phrase. UD sorts dependents into nominals, clauses and modifiers, and the relation name shows which one is involved (Marneffe et al. 2021). Contrasts such as obl versus advcl versus advmod thus encode the dependent's type. The contrast obl versus nmod encodes the head's type: a predicate or a nominal.
Neighbouring relations
- nsubj, obj and iobj mark core arguments. Where the core–oblique line falls is a cross-linguistic question in its own right (Zeman 2017).
- nmod is used when the same kind of phrase modifies a noun ("the house on the hill") rather than a predicate.
- advcl marks a clause, rather than a nominal, that modifies a predicate.
- advmod marks an adverb.
UD v1 kept separate types for phrasal and clausal dependents and set core arguments apart from other dependents (Nivre et al. 2016). However, v1 used nmod for nominals under both verbs and nouns. The v2 revision introduced obl as a distinct label, and the inventory went from 40 types in v1 (Nivre et al. 2016) to 37 universal types in v2 (Nivre et al. 2020). The count fell despite the new label because several v1 types were removed or merged. Passive marking, for instance, is now expressed through subtypes such as nsubj:pass and aux:pass (Nivre et al. 2020).
Subtypes
Languages may refine universal labels with subtypes but may not add new base categories (Nivre et al. 2020). Common subtypes include:
- obl:agent for the passive agent.
- obl:arg for obliques that a verb's valency requires.
- obl:tmod for bare temporal nominals.
- obl:npmod for other bare nominal adverbials.
Some treebanks, including the English ones, now merge the last two into obl:unmarked.
Examples
Relations are written relation(head, dependent):
| Sentence | obl arc | Note |
|---|---|---|
| The cat sleeps on the sofa. | obl(sleeps, sofa) | "on" attaches to "sofa" as case |
| The letter was written by Anna. | obl:agent(written, Anna) | passive agent |
| We met last Tuesday. | obl:unmarked(met, Tuesday) | no preposition; formerly obl:tmod |
Compare "the cat on the sofa sleeps". Here "sofa" depends on the noun "cat", so the label is nmod, not obl.
Across languages
The relation is defined by function rather than form, so the same label covers several marking strategies:
- English and French use prepositions.
- Finnish, Hungarian and Turkish mark obliques mainly with case suffixes, and postpositions, as in Hungarian, attach as case. The case value is recorded as a morphology feature, for example Case=Ine for the Finnish inessive.
- Japanese uses postpositional case particles such as で and に, attached as case.
Recent work describes UD categories as constructions that languages realize through different strategies (Nivre 2025). It also examines clauses whose predication departs from the prototype (Nivre et al. 2026). Predicate locatives show why such clauses matter for obl. In "The cat is on the sofa", the phrase that looks oblique supplies the root, "sofa", and "is" attaches to it as cop, so no obl arc appears.
Whether a phrase counts as an argument (obl:arg) or an adjunct (plain obl) is decided treebank by treebank. Choices of this kind put cross-treebank consistency at risk, and the consistency of the UD corpora has itself been assessed systematically (Marneffe et al. 2017).
In treebanks and research
UD grew out of universal Stanford dependencies (Marneffe et al. 2014) and other harmonization efforts. These include Google universal POS tags, the Universal Dependency Treebanks, and Interset as used in HamleDT (Nivre et al. 2016). By release v2.7 it held 183 treebanks in 104 languages (Marneffe et al. 2021).
The data underlay the CoNLL 2017 and 2018 shared tasks on parsing raw text (Marneffe et al. 2021). As a result, obl is one of the labels every multilingual parser must predict (see dependency parsing). The obl–nmod split mirrors the classic prepositional-phrase attachment ambiguity, since such a phrase may attach to a verb (obl) or to a noun (nmod).
For downstream use, enhanced representations make obliques more informative. For English, the enhanced graph appends the preposition to the label and turns multiword prepositions into a single labelled edge. The proposal, built on v1, wrote nmod:on, which corresponds to obl:on for predicate dependents in enhanced UD v2. The design targets relation extraction and other natural language understanding tasks (Schuster & Manning 2016). Such graphs bring obl close to semantic role labeling, where location, time and instrument are roles that obliques frequently express.
What obl joins: the most frequent head tag → dependent tag
The five most frequent in each language's treebanks, by tokens.
The numbers
| Pair | Tokens |
|---|---|
| English: VERB → NOUN | 29,537 |
| English: VERB → PROPN | 6,455 |
| English: VERB → PRON | 4,568 |
| English: ADJ → NOUN | 2,785 |
| English: VERB → NUM | 1,170 |
| French: VERB → NOUN | 44,544 |
| French: VERB → PRON | 8,229 |
| French: VERB → PROPN | 6,435 |
| French: VERB → ADV | 4,942 |
| French: VERB → NUM | 3,041 |
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-obl/