Elamur Research⌕
☰
home / concepts / iobj: indirect object

🌐 iobj: indirect object

Universal Dependencies · 6 key papers · all concepts

Part of: 🗨️ Natural language processing · 🌐 Universal Dependencies

The indirect object (iobj) is the Universal Dependencies (UD) relation linking a predicate to a nominal core argument that is neither its subject nor its direct object. Typically this is the recipient, addressee or beneficiary in a ditransitive clause, such as *me* in *she gave me a book*.

Also: dative object, IO

What it is

In Universal Dependencies, every word attaches to exactly one head through a typed relation, and the main predicate attaches to an artificial root node. Content words serve as heads, and function words depend on them (Marneffe et al. 2021). iobj belongs to the core arguments. These are the small set of dependents that a predicate selects and that a language codes as central rather than oblique. Like nsubj and obj, it is kept apart from obl, which covers both oblique arguments and adjuncts (Nivre et al. 2016; Zeman 2017). Which coding counts as core is decided language by language. A bare nominal or a dative case can be core in some languages, while adpositions and other oblique cases mark an obl dependent (Zeman 2017; Marneffe et al. 2021).

The head is usually a verb, and the dependent is a nominal: a noun, a proper noun or a pronoun. UD makes the content word the head and lets the copula depend on it (Marneffe et al. 2021). An adjectival predicate can therefore host the relation, as in French Il lui est fidèle "He is faithful to him/her", iobj(fidèle, lui).

The relation is defined against its neighbours:

The decision rests on how an argument is coded, not on its semantic role. The recipient role links the relation to semantic role labeling, but the label itself is syntactic. Subtypes are permitted in UD v2 only as language-specific refinements of the universal set (Nivre et al. 2020). Refinements of iobj itself are rare and treebank-specific, and none has universal status.

Examples

Compare She gave a book to me. Here me carries a preposition, so it is obl (Marneffe et al. 2021).

Across languages

Languages code the indirect object differently. Some use a dative case, some use word order, and some use clitic pronouns. Others use only prepositions, and in those languages the relation barely occurs. UD's typological framing compares these strategies within a single inventory of relations (Marneffe et al. 2021; Nivre 2025).

In English, the relation corresponds to the first object of the double-object construction (gave me a book). The prepositional variant goes to obl (Marneffe et al. 2021).

In French, the indirect object is realised mainly by dative clitics: lui, leur, and me and te when dative. Full nominals with à go to obl. Some French treebanks give them the language-specific subtype obl:arg, which refines obl rather than iobj (Nivre et al. 2020).

For Portuguese, the boundary between indirect objects and adverbial adjuncts has been contested. The borderline between iobj and adverbial obliques has been re-examined directly (Souza & Freitas 2022). Some predicates are not prototypical verbs, as in the French adjectival case above. There it becomes a question which word heads the clause and therefore hosts the indirect object (Nivre et al. 2026).

In treebanks and research

Origins

iobj was inherited from the universal Stanford dependencies (Marneffe et al. 2014). It belonged to the first UD inventory, where the direct object was still called dobj (Nivre et al. 2016). UD v2 revised the guidelines and widened the use of subtypes (Nivre et al. 2020). Among the changes, dobj became obj, and the old nmod was split. Nominals attached to predicates became obl, while nmod was kept for dependents of nouns. In v1, to me in gave a book to me was therefore nmod; since v2 it is obl (Nivre et al. 2020). This split is what now separates prepositional recipients from the indirect object.

Parsing and consistency

In dependency parsing, the relation must be told apart from obj and obl on the basis of coding that differs from language to language. Cross-treebank checks have assessed how consistently the UD corpora are annotated (Marneffe et al. 2017).

Enhanced UD is an optional graph layer built on top of the basic tree (Marneffe et al. 2021). It has been specified in detail for English only, keeping the basic tree while adding edges and refining labels (Schuster & Manning 2016). For indirect objects this has two effects. A dependent shared by coordinated verbs, as in She bought and gave me a book, is propagated to each conjunct. A prepositional recipient carries its preposition in the label, which was nmod:to in that English specification (Schuster & Manning 2016).

Indirect objects in neural models

Outside treebanks, "indirect object" names a benchmark task, indirect object identification (IOI), used in mechanistic interpretability research. In IOI, a model completes sentences such as After John and Mary visited the shop, John handed a drink to …. A circuit of attention heads that carries out IOI has been reverse-engineered in GPT-2 small (Wang et al. 2022). The target there is a prepositional recipient, which UD itself would label obl.

How often the relation iobj 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 iobj occurs: its share of all tokensEnglish treebanks: 0.34% of tokens (3,781 tokens)English treebanks0.34% of tokens (3,781 tokens)French treebanks: 0.62% of tokens (6,742 tokens)French treebanks0.62% of tokens (6,742 tokens)
The numbers
TreebanksShare
English treebanks0.34% of tokens (3,781 tokens)
French treebanks0.62% of tokens (6,742 tokens)

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

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

What iobj joins: the most frequent head tag → dependent tagEnglish: VERB → PRON: 3,131English: VERB → PRON3,131English: VERB → NOUN: 346English: VERB → NOUN346English: VERB → PROPN: 289English: VERB → PROPN289English: VERB → DET: 4English: VERB → DET4English: NOUN → PRON: 3English: NOUN → PRON3French: VERB → PRON: 6,483French: VERB → PRON6,483French: ADJ → PRON: 90French: ADJ → PRON90French: NOUN → PRON: 87French: NOUN → PRON87French: ADV → PRON: 37French: ADV → PRON37French: VERB → ADV: 22French: VERB → ADV22
The numbers
PairTokens
English: VERB → PRON3,131
English: VERB → NOUN346
English: VERB → PROPN289
English: VERB → DET4
English: NOUN → PRON3
French: VERB → PRON6,483
French: ADJ → PRON90
French: NOUN → PRON87
French: ADV → PRON37
French: VERB → ADV22

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-iobj/