🏷️ INTJ: interjection

Universal Dependencies · 7 key papers · all concepts

Part of: 🗨️ Natural language processing · 🌐 Universal Dependencies

INTJ (interjection) is the Universal Dependencies part-of-speech tag for words used mainly as exclamations or as parts of an exclamation. They express a feeling, a reaction, a greeting or a response, and usually stand outside the grammatical structure of the clause.

Also: interjection, exclamation, UPOS INTJ

What it is

In Universal Dependencies, every word receives one tag from a closed list of seventeen universal parts of speech (Nivre et al. 2020). INTJ is the tag for words whose job is to react rather than to describe. Examples are "ouch", "wow", "hello", "yes" and the hesitation sound "um". Such words are not arguments of a predicate and do not modify nouns. They bear on the utterance as a whole, which places them closer to pragmatics than to syntax.

The guideline idea, in plain terms, is that a word counts as INTJ by how it is used. Several kinds of word fall under it:

• exclamations of feeling
• greetings and farewells
• answer words such as "yes" and "no"
• fillers in spontaneous speech
• onomatopoeic words, when they are used as exclamations

The twelve-tag universal set of Petrov, Das and McDonald had no interjection class. Its categories were noun, verb, adjective, adverb, pronoun, determiner, adposition, numeral, conjunction, particle, punctuation and a residual X (Petrov et al. 2012). When UD v1 extended that Google set to seventeen tags, INTJ became a category of its own, along with PROPN, AUX and others (Nivre et al. 2016).

Borderline cases

The difficult decisions lie where INTJ meets its neighbouring tags:

• Words such as "well", "so" or "like" are adverbs or conjunctions in their ordinary use. They get INTJ only when they work as discourse fillers.
• Interjection-like particles that are part of the grammar go to PART. An example is the English infinitive marker "to".
• Emoticons and similar signs go to SYM.
• Words with clear nominal or verbal grammar normally keep their lexical tag even in an exclamation. Examples are "Thanks!" and "Shame!" in many treebanks.
• "Please" and "OK" are handled differently from one treebank to the next.

Features

INTJ words usually carry no FEATS at all, because they rarely inflect. UD v2 added a Polite feature (Nivre et al. 2020). It is available where a language marks politeness on response or greeting words, but most treebanks leave interjections unmarked.

Examples

Below are four short sentences, each with the interjection and the relation it takes:

When an interjection forms a whole utterance by itself, it becomes the root of its sentence. When a speaker starts a word, stops and starts again, the abandoned fragment can be linked to the repair as a reparandum. Multi-part exclamations such as "ha ha ha" are chained with flat.

Across languages

Interjections exist in all known languages, but their inventory and sound shape vary a great deal. Some examples:

• French: "ah", "bah", "euh", "oui", "hein".
• German: "ach", "na", "ja".
• Spanish: "ay", "ojalá".
• Japanese: "ええ" and "あのう", which are common as responses and fillers.

Some languages turn originally nominal or verbal forms into fixed exclamations. In those cases annotators must decide whether the form is still a noun or verb or has already become an interjection.

The theory behind UD treats the word as the basic unit and looks for parallel analyses across languages (Marneffe et al. 2021). As a result, the same discourse relation is used for "oh" and for its French counterpart.

The English and French treebanks show the same general picture. In both, the attachment of an interjection as a discourse dependent clearly dominates, and its use as a sentence root comes a distant second. Coordination and reparandum follow, with reparandum somewhat more prominent in French. INTJ is noticeably more frequent in English, which reflects the share of spoken and web material in its treebanks.

In treebanks and research

Interjections are rare in edited prose and frequent in transcribed speech, chat and social media. They therefore matter most for corpus linguistics of spoken language, for dialogue systems and for speech research.

Older treebanks often lacked a separate interjection tag or encoded it differently. Converting them into UD relied on mapping layers such as Interset. Interset decodes each source tag into universal features and re-encodes it into a target tagset, so each new tagset needs only one driver (Zeman 2008).

Because INTJ depends on usage, it is a classic source of disagreement between treebanks of the same language. Consistency studies compare annotation across UD corpora (Marneffe et al. 2017). One such study measures the divergence of UPOS-trigram distributions with a symmetric score (Aggarwal & Zeman 2020):

θpos(A, B) = KLcpos3(A, B) + KLcpos3(B, A)

Here A and B are two treebanks, and KLcpos3 is the Kullback–Leibler divergence between their distributions of three-tag sequences. Under the study's thresholds, a sum of 0.5 or less signals consistent annotation, and a value of 4.0 or more signals inconsistent annotation (Aggarwal & Zeman 2020). A treebank that tags fillers as INTJ in one genre and as ADV in another shows up in such comparisons.

For part-of-speech tagging and dependency parsing, INTJ is easy in clean text and hard in noisy text. The principles and tools of UD are surveyed in tutorial form by Zeman (Zeman 2021).

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.

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-intj/