--- title: "SCONJ: subordinating conjunction — definition and key papers" description: "A subordinating conjunction (SCONJ) is, in Universal Dependencies, the universal part-of-speech (UPOS) tag for function words that…" canonical: "https://research.elamur.ai/free/en/concepts/upos-sconj/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/concepts/upos-sconj/) [home](https://md.research.elamur.ai/free/en/) / [concepts](https://md.research.elamur.ai/free/en/concepts/) / SCONJ: subordinating conjunction # 🏷️ SCONJ: subordinating conjunction Universal Dependencies · 7 key papers · [all concepts](https://md.research.elamur.ai/free/en/concepts/) Part of: [🗨️ Natural language processing](https://md.research.elamur.ai/free/en/concepts/natural-language-processing/) · [🌐 Universal Dependencies](https://md.research.elamur.ai/free/en/concepts/universal-dependencies/) A subordinating conjunction (SCONJ) is, in Universal Dependencies, the universal part-of-speech (UPOS) tag for function words that introduce a clause and mark it as dependent on another clause, as a complement or an adverbial modifier. Examples are English that, if and because. Also: subordinating conjunction, subordinator, complementizer (for clause-complement markers), UPOS SCONJ ## What it is A subordinating conjunction is a small, closed-class word that turns a whole clause into a part of a larger sentence. In *She left because it rained*, the word *because* makes *it rained* an adverbial clause modifying *left*. In *I know that he lied*, the word *that* makes *he lied* the clausal complement (`ccomp`) of *know*. The coordinating conjunction does something different: it joins two units of equal rank (*and*, *or*). In [Universal Dependencies](https://md.research.elamur.ai/free/en/concepts/universal-dependencies/) it is tagged CCONJ (coordinating conjunction). UD defines SCONJ as a conjunction that links constructions by making one of them a constituent of the other. Typically it marks the incorporated clause as an argument or an adverbial. Under UD's content-head design, a function word depends on the content word it serves rather than heading it ([Marneffe et al. 2021](https://md.research.elamur.ai/free/en/papers/marneffe-2021-universal-dependencies/)). A SCONJ token therefore hangs from the predicate of the subordinate clause, usually through the relation `mark` (marker of a subordinate clause). ## Examples - English: *I wonder whether/if she knows* (complement clause). *Although it was late, we stayed* (concessive adverbial). *He spoke as if he were ill*. - French: *Je crois que tu as raison* (complement). *Il est parti parce qu'il pleuvait*. Here the multiword *parce que* is linked internally by `fixed` (fixed multiword expression). - German: *Ich bleibe, weil es regnet*. The clause it introduces is verb-final. - Spanish: *Dijo que vendría*. *Si llueve, no salgo*. Several neighbouring tags compete with SCONJ for the same words: | Word or case | Usual UD choice | Reason | | --- | --- | --- | | English *before/after* + noun | ADP (adposition) | it heads a nominal, so `case` (case marker) | | *before/after* + clause | SCONJ | it introduces a clause, so `mark` | | relative *that* (English) | PRON (pronoun) | it fills a role inside the relative clause | | *when/where* (English) | ADV (adverb) | wh-adverbs with their own function, attached as `advmod`; German *wenn* and *als* are SCONJ | | infinitival *to* (English) | PART (particle) | English UD convention treats it as a particle, though it still attaches as `mark`; other languages differ | SCONJ words usually carry no features. The Czech treebanks follow a language-specific convention here, not a general UD rule. Forms such as *aby* and *kdyby* fuse a conjunction with a conditional auxiliary, so they bear Mood=Cnd plus Person and Number agreement. ## Across languages Subordination is coded very differently across languages, and the frequency of SCONJ tokens varies with it. Head-initial European languages use free subordinators regularly. Japanese, Korean and Turkish rely mostly on verbal suffixes, nominalizations and converbs (non-finite verb forms used for adverbial subordination). As a result, many of their clauses contain no SCONJ word at all. UD takes the word as its basic unit and splits clitics into separate words ([Marneffe et al. 2021](https://md.research.elamur.ai/free/en/papers/marneffe-2021-universal-dependencies/)). Accordingly, Hebrew treebanks treat the prefixed subordinator *še-* "that" as its own SCONJ token. In both English and French data, `mark` is the usual relation for SCONJ. `fixed` appears where a subordinator spans several words. That is common in French, with *parce que*, *bien que* and *alors que*. ## In treebanks and research The twelve-tag universal tagset of Petrov, Das and McDonald had no SCONJ category. Both kinds of conjunction fell under a single CONJ tag ([Petrov et al. 2012](https://md.research.elamur.ai/free/en/papers/petrov-2012-universal-part-speech-tagset/)). The UD v1 guidelines, published in October 2014, extended that set to 17 tags with SCONJ as a tag of its own. The v1 collection was described in a later overview ([Nivre et al. 2016](https://md.research.elamur.ai/free/en/papers/nivre-2016-universal-dependencies-v1-multilingual-treebank/)). The second version of the guidelines (UD v2) then renamed the remaining CONJ to CCONJ, which made the two-way distinction explicit ([Nivre et al. 2020](https://md.research.elamur.ai/free/en/papers/nivre-2020-universal-dependencies-v2-evergrowing-multilingual/)). Mapping legacy tagsets onto this distinction can rely on feature-based conversion. Interset is one such intermediate representation: a shared inventory of morphosyntactic features and their values. UD's morphological features were based on it ([Zeman 2008](https://md.research.elamur.ai/free/en/papers/zeman-2008-reusable-tagset-conversion-tagset-drivers/); [Nivre et al. 2016](https://md.research.elamur.ai/free/en/papers/nivre-2016-universal-dependencies-v1-multilingual-treebank/)). Part-of-speech annotation is not always consistent across the UD corpora ([Marneffe et al. 2017](https://md.research.elamur.ai/free/en/papers/marneffe-2017-assessing-annotation-consistency-universal-dependencies/)). Differences arise even between two treebanks of one language annotated under the same guidelines ([Aggarwal & Zeman 2020](https://md.research.elamur.ai/free/en/papers/aggarwal-2020-estimating-pos-annotation-consistency-different/)). Words on the SCONJ/ADP/ADV boundary are one plausible place for such differences. One proposed check compares the UPOS trigram distributions of two treebanks ([Aggarwal & Zeman 2020](https://md.research.elamur.ai/free/en/papers/aggarwal-2020-estimating-pos-annotation-consistency-different/)): ``` θpos(A,B) = KLcpos3(A,B) + KLcpos3(B,A) ``` Here A and B are two treebanks. KLcpos3, introduced by Rosa and Žabokrtský, is an asymmetric Kullback–Leibler divergence between their UPOS-trigram frequency distributions. Trigrams missing from one treebank are given a count of one. Summing both directions gives θpos, a symmetric measure proposed by Aggarwal and Zeman ([Aggarwal & Zeman 2020](https://md.research.elamur.ai/free/en/papers/aggarwal-2020-estimating-pos-annotation-consistency-different/)). Divergence also rises with smaller data and with differences in genre, so the authors derived genre-dependent thresholds. Whatever the genre mix, θpos at or below 0.5 counts as consistent annotation, and 4.0 or above as inconsistent ([Aggarwal & Zeman 2020](https://md.research.elamur.ai/free/en/papers/aggarwal-2020-estimating-pos-annotation-consistency-different/)). Tagging *before* as SCONJ in one treebank and as ADP in another could contribute to such trigram differences. For [part-of-speech tagging](https://md.research.elamur.ai/free/en/concepts/part-of-speech-tagging/) and [dependency parsing](https://md.research.elamur.ai/free/en/concepts/dependency-parsing/), a correctly tagged subordinator is useful evidence about where a clause begins and what kind of clause it may be. It can help a parser choose between relations such as `ccomp` and `advcl` (adverbial clause modifier). Still, words like *that* and *if* introduce both kinds of clause, and the choice also depends on the head predicate. How often the tag SCONJ 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 | 1.41% of tokens (15,907 tokens) | | French treebanks | 1.89% of tokens (20,562 tokens) | The relations SCONJ words take most often The five most frequent in each language's treebanks, by tokens. **The numbers** | Pair | Tokens | | --- | --- | | English: mark | 15,249 | | English: fixed | 190 | | English: root | 83 | | English: obj | 64 | | English: cc | 63 | | French: mark | 17,189 | | French: fixed | 2,622 | | French: case | 416 | | French: advmod | 94 | | French: dep | 60 | ## 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/) — Joakim Nivre, Marie-Catherine de Marneffe, Filip Ginter et al. · 2016 · 1,741 citations - [A Universal Part-of-Speech Tagset](https://md.research.elamur.ai/free/en/papers/petrov-2012-universal-part-speech-tagset/) — Slav Petrov, Dipanjan Das, Ryan McDonald · 2012 · 1,180 citations - [Universal Dependencies](https://md.research.elamur.ai/free/en/papers/marneffe-2021-universal-dependencies/) — Marie-Catherine de Marneffe, Christopher D. Manning, Joakim Nivre et al. · 2021 · 480 citations - [Universal Dependencies v2: An Evergrowing Multilingual Treebank Collection](https://md.research.elamur.ai/free/en/papers/nivre-2020-universal-dependencies-v2-evergrowing-multilingual/) — Joakim Nivre, Marie-Catherine de Marneffe, Filip Ginter et al. · 2020 · 325 citations - [Reusable Tagset Conversion Using Tagset Drivers](https://md.research.elamur.ai/free/en/papers/zeman-2008-reusable-tagset-conversion-tagset-drivers/) — Daniel Zeman · 2008 · 181 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/) — Marie-Catherine de Marneffe, Matias Grioni, Jenna Kanerva et al. · 2017 · 30 citations - [Estimating POS Annotation Consistency of Different Treebanks in a Language](https://md.research.elamur.ai/free/en/papers/aggarwal-2020-estimating-pos-annotation-consistency-different/) — Akshay Aggarwal, Daniel Zeman · 2020 · 7 citations ## Related concepts - [Universal Dependencies](https://md.research.elamur.ai/free/en/concepts/universal-dependencies/) — Universal Dependencies is a framework and an open collection of treebanks that annotates the grammar of many languages with one shared… - [Part-of-speech tagging](https://md.research.elamur.ai/free/en/concepts/part-of-speech-tagging/) — Part-of-speech tagging is the task of assigning each token in a text a grammatical category, such as noun, verb, adjective or determiner… - [Morphology](https://md.research.elamur.ai/free/en/concepts/morphology/) — Morphology is the branch of linguistics that studies the internal structure of words. It covers how words are built from meaningful units… ## More papers Topics: [🔤 Part-of-speech tagging](https://md.research.elamur.ai/free/en/topics/part-of-speech-tagging/) · [🌐 Universal Dependencies](https://md.research.elamur.ai/free/en/topics/universal-dependencies/) Shelves: [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) - [← previous concept Scaling laws](https://md.research.elamur.ai/free/en/concepts/scaling-laws/) - [next concept → Semantic parsing](https://md.research.elamur.ai/free/en/concepts/semantic-parsing/) - [↑ back to concepts](https://md.research.elamur.ai/free/en/concepts/)