--- title: "root: root β€” definition and key papers β€” Elamur Research" description: "root is the Universal Dependencies relation that attaches the head of a sentence, usually its main predicate, to an artificial root node outside the…" canonical: "https://research.elamur.ai/free/en/concepts/ud-root/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) Β· [this page in HTML](https://research.elamur.ai/free/en/concepts/ud-root/) [home](https://md.research.elamur.ai/free/en/) / [concepts](https://md.research.elamur.ai/free/en/concepts/) / root: root # 🌐 root: root Universal Dependencies Β· 6 key papers Β· [all concepts](https://md.research.elamur.ai/free/en/concepts/) Part of: [πŸ—¨οΈ Natural language processing](https://md.research.elamur.ai/free/en/concepts/natural-language-processing/) Β· [🌐 Universal Dependencies](https://md.research.elamur.ai/free/en/concepts/universal-dependencies/) root is the Universal Dependencies relation that attaches the head of a sentence, usually its main predicate, to an artificial root node outside the sentence. Every basic dependency tree has exactly one root edge, and every other word depends, directly or indirectly, on that root word. Also: ROOT, sentence root ## What it is In [Universal Dependencies](https://md.research.elamur.ai/free/en/concepts/universal-dependencies/) (UD), the syntax of a sentence is encoded as a tree of typed relations between words ([Marneffe et al. 2021](https://md.research.elamur.ai/free/en/papers/marneffe-2021-universal-dependencies/)). The tree needs a starting point. That starting point is a virtual root node, conventionally numbered 0, which is not a real word. The root relation joins this node, as head, to the single word that heads the whole clause, as dependent. In the CoNLL-U files that store UD data, that root word carries head index 0 and the label `root`. UD picks content words as heads, and function words such as auxiliaries, copulas and adpositions depend on them ([Marneffe et al. 2021](https://md.research.elamur.ai/free/en/papers/marneffe-2021-universal-dependencies/)). So the root word is usually the main lexical verb. When a sentence has a copula, the root word is the nonverbal predicate instead, such as an adjective or a noun. The `root` relation is one of the universal relation types of UD v2 ([Nivre et al. 2020](https://md.research.elamur.ai/free/en/papers/nivre-2020-universal-dependencies-v2-evergrowing-multilingual/)). It was already part of the first release of the guidelines ([Nivre et al. 2016](https://md.research.elamur.ai/free/en/papers/nivre-2016-universal-dependencies-v1-multilingual-treebank/)), which inherited it from the Stanford tradition ([Marneffe et al. 2014](https://md.research.elamur.ai/free/en/papers/marneffe-2014-universal-stanford-dependencies-cross-linguistic/)). ### Neighbouring relations Several relations are sometimes confused with `root`: - `parataxis` joins further clauses that sit side by side with the root clause. - `conj` joins coordinated predicates, with the first conjunct taking the `root` relation. - `ccomp` and `advcl` attach subordinate clauses below the root word. - `dep` is the fallback label for links that cannot be classified more precisely. The head of a sentence still takes `root` even when its other links are unclear. UD v2 lets languages add subtypes to universal relations ([Nivre et al. 2020](https://md.research.elamur.ai/free/en/papers/nivre-2020-universal-dependencies-v2-evergrowing-multilingual/)). However, no universal subtype of `root` is defined, and the label normally appears bare. ## Examples In these examples, the word in capitals is the dependent of the root relation: 1. She READS old books. Here `root` β†’ *reads*. The words *she* and *books* attach to *reads* as `nsubj` and `obj`. 2. The soup is HOT. Here `root` β†’ *hot*. The copula *is* attaches to *hot* as `cop`. 3. FIRE! Here `root` β†’ *fire*. This is a one-word utterance with no verb, so its root word heads the sentence without being a predicate. Example 2 shows the content-word principle at work. A verb-headed analysis would make *is* the root word, but UD instead keeps *hot* as the head of the clause. Recent work examines how UD should treat clauses whose predication is unusual or absent (Nivre et al. 2026). A related study places such choices within a broader account of constructions and strategies (Nivre 2025). ## Across languages In UD, every sentence in every language has exactly one root. What varies is the part of speech of the root word, which is where [part-of-speech tagging](https://md.research.elamur.ai/free/en/concepts/part-of-speech-tagging/) and syntax meet. UD makes the nonverbal predicate the head even when a copula is present ([Marneffe et al. 2021](https://md.research.elamur.ai/free/en/papers/marneffe-2021-universal-dependencies/)). As a result, nominal and adjectival predicates are root words in English or French just as in languages without a present-tense copula, such as Russian. The copula-less clause simply lacks a `cop` dependent. Pro-drop languages often have root verbs with no overt subject at all. Headlines, titles and list items push many nouns and proper nouns into root position. Because each sentence has one root, the share of `root` tokens in a treebank is simply the inverse of the mean sentence length. Comparing that share between, say, [English](https://md.research.elamur.ai/free/en/topics/english/) and [French](https://md.research.elamur.ai/free/en/topics/french/) corpora therefore compares sentence length. Sentence length depends heavily on the genres a corpus contains, such as web text versus news, rather than on the language alone. Work on core arguments looks at how UD separates them from obliques across languages (Zeman 2017). UD's tooling and principles for keeping such choices uniform are summarised in tutorial form (Zeman 2021). ## In treebanks and research The `root` relation anchors the basic tree, which every one of the UD [treebanks](https://md.research.elamur.ai/free/en/topics/treebanks/) must provide ([Nivre et al. 2020](https://md.research.elamur.ai/free/en/papers/nivre-2020-universal-dependencies-v2-evergrowing-multilingual/)). The enhanced English representation leaves that basic tree untouched while adding edges and making relation labels more specific. Its stronger enhanced++ variant may also remove and restructure basic edges ([Schuster & Manning 2016](https://md.research.elamur.ai/free/en/papers/schuster-2016-enhanced-english-universal-dependencies-improved/)). Because the root word is fixed, validation tools can check for a single root per sentence. Consistency studies compare annotation choices, including heads and relation labels, across corpora ([Marneffe et al. 2017](https://md.research.elamur.ai/free/en/papers/marneffe-2017-assessing-annotation-consistency-universal-dependencies/)). In [dependency parsing](https://md.research.elamur.ai/free/en/concepts/dependency-parsing/), the artificial root node (node 0) is standard machinery. Transition-based parsers typically begin with the root node on the stack. Graph-based parsers search for a spanning tree rooted at node 0. Limiting that node to a single dependent is an added restriction that not all graph-based parsers enforce. Root attachment is a useful check on clause-head and sentence segmentation errors. If the root word is wrong, the error can propagate to many attachments. It then lowers the labelled attachment score (LAS) and the unlabelled attachment score (UAS). LAS is the share of words given both the correct head and the correct label, and UAS is the share given the correct head regardless of label. UD served as the basis for the CoNLL 2017 and CoNLL 2018 shared tasks, in which parsers worked from raw text in many languages ([Marneffe et al. 2021](https://md.research.elamur.ai/free/en/papers/marneffe-2021-universal-dependencies/)). In such a setting, sentence segmentation errors create or merge roots, so the relation is closely tied to how sentences are cut. How often the relation root 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). **The numbers** | Treebanks | Share | | --- | --- | | English treebanks | 8.79% of tokens (98,986 tokens) | | French treebanks | 5.42% of tokens (59,091 tokens) | What root joins: the most frequent head tag β†’ dependent tag The five most frequent in each language's treebanks, by tokens. **The numbers** | Pair | Tokens | | --- | --- | | English: ROOT β†’ VERB | 60,308 | | English: ROOT β†’ NOUN | 17,063 | | English: ROOT β†’ ADJ | 6,879 | | English: ROOT β†’ PROPN | 3,982 | | English: ROOT β†’ PRON | 3,117 | | French: ROOT β†’ VERB | 47,471 | | French: ROOT β†’ NOUN | 5,736 | | French: ROOT β†’ ADJ | 3,322 | | French: ROOT β†’ PRON | 758 | | French: ROOT β†’ ADV | 524 | ## 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. - [Universal Dependencies v1: A Multilingual Treebank Collection](https://md.research.elamur.ai/free/en/papers/nivre-2016-universal-dependencies-v1-multilingual-treebank/) β€” Joakim Nivre, Marie-Catherine de Marneffe, Filip Ginter et al. Β· 2016 Β· 1,741 citations - [Universal Stanford dependencies: A cross-linguistic typology](https://md.research.elamur.ai/free/en/papers/marneffe-2014-universal-stanford-dependencies-cross-linguistic/) β€” Marie-Catherine de Marneffe, Timothy Dozat, Natalia Silveira et al. Β· 2014 Β· 627 citations - [Universal Dependencies](https://md.research.elamur.ai/free/en/papers/marneffe-2021-universal-dependencies/) β€” Marie-Catherine de Marneffe, Christopher D. Manning, Joakim Nivre et al. Β· 2021 Β· 480 citations - [Enhanced English Universal Dependencies: An Improved Representation for Natural Language Understanding Tasks](https://md.research.elamur.ai/free/en/papers/schuster-2016-enhanced-english-universal-dependencies-improved/) β€” Sebastian Schuster, Christopher D. Manning Β· 2016 Β· 336 citations - [Universal Dependencies v2: An Evergrowing Multilingual Treebank Collection](https://md.research.elamur.ai/free/en/papers/nivre-2020-universal-dependencies-v2-evergrowing-multilingual/) β€” Joakim Nivre, Marie-Catherine de Marneffe, Filip Ginter et al. Β· 2020 Β· 325 citations - [Assessing the Annotation Consistency of the Universal Dependencies Corpora](https://md.research.elamur.ai/free/en/papers/marneffe-2017-assessing-annotation-consistency-universal-dependencies/) β€” Marie-Catherine de Marneffe, Matias Grioni, Jenna Kanerva et al. Β· 2017 Β· 30 citations ## Related concepts - [Universal Dependencies](https://md.research.elamur.ai/free/en/concepts/universal-dependencies/) β€” Universal Dependencies is a framework and an open collection of treebanks that annotates the grammar of many languages with one shared… - [Dependency parsing](https://md.research.elamur.ai/free/en/concepts/dependency-parsing/) β€” 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](https://md.research.elamur.ai/free/en/topics/universal-dependencies/) Β· [🌳 Dependency parsing](https://md.research.elamur.ai/free/en/topics/dependency-parsing/) Shelves: [πŸ—£οΈ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) - [← previous concept Retrieval-augmented generation](https://md.research.elamur.ai/free/en/concepts/retrieval-augmented-generation/) - [next concept β†’ Scaling laws](https://md.research.elamur.ai/free/en/concepts/scaling-laws/) - [↑ back to concepts](https://md.research.elamur.ai/free/en/concepts/)