🏷️ X: other
X (other) is the universal part-of-speech tag of Universal Dependencies for words that cannot be given any real part of speech, such as unanalysed foreign material, fragments of split words and other residue. It is meant as a rare last resort.
What it is
In part-of-speech tagging, every token needs a label, even tokens that fit no grammatical class. The tag X fills that gap. In Universal Dependencies (UD), the UPOS inventory is a closed list of 17 tags, and X is the one reserved for material with no proper category (Nivre et al. 2020). In the guidelines' terms, X applies when an annotator cannot assign NOUN, VERB or any other real class to a word. A typical case is a stretch of another language that is quoted rather than integrated into the sentence.
The tag predates UD. The universal tagset of Petrov, Das and McDonald had twelve categories, and its X covered everything else, explicitly including abbreviations and foreign words (Petrov et al. 2012). UD v1 extended this Google set to 17 tags, adding PROPN and AUX among others (Nivre et al. 2016). UD v2 narrowed X considerably. It added the morphological features Abbr, Foreign and Typo (Nivre et al. 2020). As a result, an abbreviation keeps its real part of speech and is marked Abbr=Yes, and a misspelled word keeps its class and is marked Typo=Yes.
Borderline cases
- Foreign words used with the host language's syntax get an ordinary tag, such as NOUN or ADJ. Only unanalysed foreign stretches become X, often with Foreign=Yes.
- Symbols such as %, $ and = are SYM, and punctuation is PUNCT, not X.
- Interjections and fillers are INTJ when the language treats them as words.
- Proper names, even foreign ones, are normally PROPN.
- The non-first parts of a word wrongly split by a space are typically X attached by goeswith.
Because UD treats the syntactic word as the basic unit (Marneffe et al. 2021), wrongly split or broken words are where X most often appears.
Examples
- English: She whispered "je ne sais quoi". The French words can be tagged X with Foreign=Yes and chained by flat (or flat:foreign) to the first word.
- English: I'll do it with out help. Here out is X, attached to with by goeswith, because together they spell without.
- French: Il a dit « no comment ». The English phrase is X, held together by flat.
- Spoken French: je vou- je voulais. The fragment vou- is X and depends on the repaired word via reparandum.
Across languages
X behaves differently in different languages because what counts as "unanalysable" depends on the corpus and its genre. In the English UD treebanks, X is rare. Its words most often enter flat relations, followed by discourse and goeswith, which reflects web text with split words and odd tokens. In the French treebanks, X is somewhat more frequent and its usage is dominated more strongly by flat, with appos, nmod, conj and reparandum trailing. That pattern points to quoted foreign titles and names and to disfluencies in speech.
Code-switched treebanks raise a harder question: whether the second language should be analysed or left as X. Historical corpora face a similar choice with damaged or illegible tokens.
In treebanks and research
For dependency parsing, X matters less for its frequency than for its inconsistency. One treebank may tag a foreign quote as X while another analyses it word by word. Earlier harmonization work, including Zeman's Interset, mapped heterogeneous tagsets through a universal feature inventory (Zeman 2008). Residual categories were exactly where such mappings struggled, and the same problem arose in the 12-tag mappings (Petrov et al. 2012).
UD's consistency checks can expose divergent X usage (Marneffe et al. 2017). One measure compares UPOS trigram distributions between two treebanks of the same language (Aggarwal & Zeman 2020):
θpos(A, B) = KLcpos3(A, B) + KLcpos3(B, A)
Here A and B are two treebanks. KLcpos3 is the asymmetric Kullback–Leibler divergence over UPOS trigram frequencies, and θpos is its symmetric sum. A pair with a low θpos is considered consistently annotated, with thresholds that depend on the genres involved (Aggarwal & Zeman 2020). A treebank that overuses X shifts its trigram distribution and can be flagged by this measure.
Tutorials on UD principles likewise advise annotators to reach for X sparingly (Zeman 2021).
The relations X words take most often
The five most frequent in each language's treebanks, by tokens.
The numbers
| Pair | Tokens |
|---|---|
| English: flat | 464 |
| English: discourse | 345 |
| English: goeswith | 268 |
| English: conj | 106 |
| English: root | 82 |
| French: flat | 2,107 |
| French: appos | 428 |
| French: nmod | 292 |
| French: conj | 253 |
| French: reparandum | 208 |
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-x/