🏷️ ADP: adposition
ADP (adposition) is the Universal Dependencies part-of-speech tag for prepositions, postpositions and circumpositions. These are function words that attach to a nominal and mark its grammatical or semantic relation, such as place, time, instrument or possession, to the rest of the clause.
What it is
In Universal Dependencies, ADP is one of the closed classes in the universal part-of-speech inventory. In English the label covers words such as in, of, with and after. In Japanese it covers words that follow the noun, such as ni or de. An adposition usually belongs to a nominal phrase and signals how that phrase relates to its governor, much as a case ending does in a language with rich morphology.
UD treats content words as heads and makes function words their dependents (Marneffe et al. 2021). An adposition therefore does not head its phrase. It attaches to the noun it introduces, and that noun attaches to the verb or noun it modifies. The tag inventory of UD v2 is closed: languages cannot add new basic categories and can only refine existing ones (Nivre et al. 2020).
Borderline cases
- ADP vs ADV: a word like outside used without an object (she went outside) is normally tagged ADV, not ADP.
- ADP vs SCONJ: before or after introducing a full finite clause is usually SCONJ in English. Some languages tag adpositions that introduce clauses as ADP and give them the relation
mark. - ADP vs PART: English verb particles (give up, turn off) are tagged ADP in the English treebanks and attached with
compound:prt. Infinitival to is PART. - Fused forms: French au and German zum are split into an adposition plus a determiner, while the original token is kept as a multiword token. This follows UD's recoverability principle (Nivre et al. 2016).
Features and relations
The typical feature is AdpType (Prep, Post, Circ). Some languages also use Case, for the case an adposition governs, or Abbr. In the basic tree, the usual relation is case. Others are mark before clauses, fixed inside multiword adpositions such as because of, and occasionally obl when the adposition is stranded or elliptical.
Examples
book on the table
NOUN ADP DET NOUN
case→table (table —nmod→ book)
In this example, on is tagged ADP and depends on table via case, and table is the dependent of book via nmod. Some further examples:
- English: according to the report. Here according is ADP, and to attaches to it as
fixed. - French: il a cessé de fumer. Here de is ADP with
mark, because it introduces an infinitive. - Turkish: ev için ("for the house"). Here için is a postposition (
AdpType=Post). - German: um des Friedens willen ("for peace's sake"). This is a circumposition, with both parts tagged ADP.
Across languages
Word order largely decides whether a language uses prepositions (English, French, Arabic) or postpositions (Japanese, Turkish, Hindi). Many languages express similar meanings through case morphology rather than separate words. Because UD takes the syntactic word as its basic unit, case affixes are not tagged ADP, and the same relation can therefore surface as a word in one language and as a feature in another (Marneffe et al. 2021). Clitic adpositions written together with their host, as in Hebrew or Arabic, are segmented into separate words (Marneffe et al. 2021).
Frequency also varies by language. In the current English treebanks, ADP is dominated by case, followed at a distance by compound (verb particles), fixed and mark. In French, ADP accounts for a larger share of all tokens. There case again leads, but mark comes clearly second, reflecting de and à before infinitives. See the English and French topics.
In treebanks and research
ADP originates in the twelve-category universal tagset of Petrov, Das and McDonald (Petrov et al. 2012). They mapped many incompatible treebank tagsets onto one coarse set so that part-of-speech tagging results could be compared and parsers transferred across languages. In that coarse mapping, a tag like the Penn Treebank's IN, which covers both prepositions and subordinators, falls into ADP.
UD v1 extended the set to seventeen tags. It added categories such as PROPN and AUX and drew its morphological features from Interset (Nivre et al. 2016). Interset is a feature interlingua that converts tagsets through decoder and encoder drivers rather than pairwise converters (Zeman 2008). UD v2 renamed CONJ to CCONJ, which separated coordinators from the subordinators that sit next to ADP in the inventory (Nivre et al. 2020). The principles and tooling behind these choices are presented in tutorial form by Zeman (Zeman 2021).
The ADP/SCONJ/ADV boundary is a common source of disagreement in treebanks. Consistency studies of the UD corpora examine such annotation divergences (Marneffe et al. 2017). One symmetric measure of this kind, built on UPOS trigram distributions, flags inconsistent treebank pairs within a single language without depending on any tagger (Aggarwal & Zeman 2020). For dependency parsing, ADP's role as a dependent of the content word keeps argument structure parallel across languages, whatever their surface morphology (Marneffe et al. 2021).
The relations ADP words take most often
The five most frequent in each language's treebanks, by tokens.
The numbers
| Pair | Tokens |
|---|---|
| English: case | 89,873 |
| English: compound | 3,771 |
| English: fixed | 1,425 |
| English: mark | 1,212 |
| English: obl | 606 |
| French: case | 127,402 |
| French: mark | 10,644 |
| French: advmod | 1,655 |
| French: fixed | 660 |
| French: reparandum | 258 |
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…
- 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 — 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 · 🌐 Universal Dependencies
Shelves: 🗣️ Language and Linguistics
This page in Markdown → https://md.research.elamur.ai/free/en/concepts/upos-adp/