Written and maintained by CASRAI Editorial Board
Published
On this page
Morphological features are the grammatical properties of a word form, such as its number, case, tense or person, recorded as attribute-value pairs. In the Universal Dependencies (UD) annotation scheme they are written as Feature=Value pairs in the FEATS column of a treebank, for example Number=Plur for a plural noun. This page explains the format, the most common universal features, how to read an annotation and how to read the same data in Python.
What are morphological features? (definition)
A morphological feature describes one grammatical property that a word form carries through its shape. Researchers in linguistics and natural language processing use them to describe how a language marks distinctions such as singular versus plural or past versus present, without relying on the word’s meaning alone. The features sit beside, not instead of, the word’s lemma (its dictionary form) and its part-of-speech tag; the part-of-speech tagging guide covers the tag layer.
Universal Dependencies states that its universal feature list contains features “acknowledged as ‘universal’ in the universal part of the UD guidelines”, and that treebanks “may use additional features and values if they are properly documented”. A feature name therefore is not guaranteed to be universal, and a language may use values another language never does.
Lexical versus inflectional features
UD sorts features loosely into lexical and inflectional, but calls these labels “approximate categories only”. Its own example: gender is a lexical feature of nouns but an inflectional feature of adjectives or verbs. Treat the split as a reading aid, not a classification you can apply mechanically.
The Feature=Value format (how the FEATS column is written)
The UD file format defines the FEATS field as a list of morphological features. There is no formula in the arithmetic sense; instead there are serialization rules, and every symbol in them has a fixed meaning.
| Symbol or rule | Meaning in the FEATS field |
|---|---|
= |
Separates the attribute (feature name) from its value |
| (vertical bar) |
Separates one feature from the next in the list |
_ (underscore) |
Stands for an empty list: the token has no morphological features |
| Comma | Separates several values of one feature, sorted alphabetically |
| Order | Features are sorted alphabetically by attribute name |
The same file format has related columns: LEMMA holds the “lemma or stem of word form”, and UPOS holds the tag from the universal part-of-speech tag set. FEATS is the third piece of the token’s description.
Universal morphological features and their values
The UD universal feature list names these features: PronType, Gender, VerbForm, NumType, Animacy, Mood, Poss, NounClass, Tense, Reflex, Number, Aspect, Case, Voice, Abbr, Definite, Evident, Typo, Deixis, Polarity, Foreign, DeixisRef, Person, ExtPos, Degree, Polite and Clusivity. The data table below lists values for three of them (Number, Tense and VerbForm) with the definition and English example that each UD feature page gives. A blank in the example column means the page gives no English example for that value.
Less familiar values that show how much languages differ
The UD pages also define values that English rarely needs, which is why a feature inventory is language-specific. Under Number, UD lists Dual (two persons, animals or things), Tri (three, found in some Austronesian pronouns), Pauc (a few), Grpa (greater paucal: more than several but not many), Grpl (greater plural: many, potentially all, instances), Inv (a non-default number for a particular noun), Count (a special plural form used after numerals, exemplified with Bulgarian and Russian) and Coll (collective, mass or singulare tantum, exemplified with Czech). Under Tense, Imp marks the imperfect, a special case of the past tense (exemplified with Bulgarian), and Pqp marks the pluperfect, an action that happened before another action in the past. Under VerbForm, Conv is the converb (exemplified with Czech), Gdv the gerundive (Latin and Ancient Greek, not the same as a gerund), Sup the supine (some Slavic languages and Swedish) and Vnoun a verbal noun other than an infinitive, which inflects like a noun (Czech example).
The practical consequence for a researcher is that the same feature name can have a different set of attested values in each language. Before comparing two corpora, list the values that actually occur in each, then compare only values whose UD definitions match.
Worked example: reading a FEATS string
The three rows below are an illustration of the format that this page constructed by hand. They are not taken from a real treebank, and a real treebank may annotate the same words with more features.
| Form | Lemma | UPOS | FEATS |
|---|---|---|---|
| cars | car | NOUN | Number=Plur |
| went | go | VERB | Tense=Past|VerbForm=Fin |
| seeing | see | VERB | VerbForm=Ger |
Reading the second row step by step: the FEATS string is Tense=Past|VerbForm=Fin. Splitting on the vertical bar gives two features, so the token carries 2 features. Splitting each on = gives the pairs Tense with value Past and VerbForm with value Fin. The attribute names are in alphabetical order (T before V), which matches the UD rule. UD’s definition of Past, “actions that happened before a reference point”, tells you how to interpret the first pair. The third row shows a single feature, so there is no bar. A token with no features would show _ instead.
Interpretation: how to read a result
| What you see | How to read it |
|---|---|
Number=Plur |
The form denotes several persons, animals or things (UD’s example: cars) |
Number=Ptan |
Plurale tantum: appears only in the plural although it denotes one thing (UD’s examples: scissors, pants) |
Tense=Pres on “he goes” |
Actions in progress, or states valid, at a reference point; UD’s English examples are “he goes home” and “he was going home” |
VerbForm=Fin |
A finite verb: UD defines it as a verb with a non-empty Mood feature |
_ |
No features listed for this token |
| A feature name not on the universal list | A treebank-specific extension; check that treebank’s own documentation |
Note that UD’s own English examples under Past include “he has gone home” and under Pres include “he was going home”. The label describes the relation to a reference point, not just the surface form, so do not infer a feature value from the ending of a word alone.
Assumptions and common mistakes
- Treating every feature as universal. UD allows extra features and values when documented, so a feature seen in one treebank may be absent in another.
- Assuming one tag set. Other tagsets exist. UD’s VerbForm page warns that some tagsets merge verb forms and moods, so values from different schemes are not interchangeable.
- Reading Pqp or Fut for English. The Tense page notes that English forms the pluperfect analytically, so English forms are not tagged Pqp, and its Future example is Spanish.
- Treating the lexical/inflectional label as exact. UD calls it approximate.
- Sorting by hand. FEATS has a defined order (alphabetical by attribute), so differently ordered strings may be flagged as invalid or compare unequal as plain text.
- Ignoring the biology sense. See the short note on disambiguation below.
How to report morphological features in your work
Name the annotation scheme (Universal Dependencies), the treebank or model that produced the features, and the version you used. State whether the features were hand-annotated or predicted by a model, since the two have different reliability. If you use features outside the universal list, say so and point to their documentation. Cite the UD guidelines page for the features you use, for example the UD feature index at universaldependencies.org/u/feat, and the format specification when you describe the FEATS column.
How to read morphological features in Python (spaCy)
spaCy stores morphological analyses in the UD FEATS string format: by default a vertical bar between features, an equals sign between field and value, and a comma between multiple values. A MorphAnalysis object returns the FEATS string with str(), a dictionary with to_dict(), the list of values for one feature with get(field) (an empty list if the feature is unset), the feature count with len(), and supports in to test a feature and value pair.
import spacynlp = spacy.load("en_core_web_sm") # any pipeline that sets morphological featuresdoc = nlp("The cars went home.")for token in doc: m = token.morph # a MorphAnalysis print(token.text, str(m), m.to_dict(), m.get("Number"), len(m))
The values printed depend on the pipeline and its version, because a trained model predicts them. Use Morphology.feats_to_dict and Morphology.dict_to_feats to convert between a FEATS string and a dictionary. There is no Excel or base-R equivalent worth recommending for a treebank; for CoNLL-U files, split the FEATS column on the bar and the equals sign as in the table above.
Not to be confused with morphological features in biology
The phrase also appears in biology, where it refers to observable form and structure rather than grammatical properties; for one such use see the colony morphology guide. The two senses share only the word “morphology”, the study of form. Everything on this page concerns the linguistic sense.
Sources
Data: Selected Universal Dependencies morphological feature values with definitions and English examples
Compiled from the Universal Dependencies feature pages for Number, Tense and VerbForm: each row gives the feature, its value code, UD's definition in condensed form and the English example the page gives, if any.
| Feature=Value | Value meaning | UD definition (condensed) | English example on the UD page |
|---|---|---|---|
| Number=Sing | Singular | One person, animal or thing | car |
| Number=Plur | Plural | Several persons, animals or things | cars |
| Number=Ptan | Plurale tantum | Appears only in the plural although it denotes one thing | scissors, pants |
| Tense=Past | Past tense | Actions that happened before a reference point | he went home; he has gone home |
| Tense=Pres | Present tense | Actions in progress, or states valid, in a reference point | he goes home; he was going home |
| Tense=Fut | Future tense | Actions that will happen after a reference point | none (Spanish example only) |
| VerbForm=Fin | Finite verb | A verb with a non-empty Mood feature | I do, he does |
| VerbForm=Part | Participle | Non-finite form combining verb and adjective traits | he could have been prepared if he had foreseen it |
| VerbForm=Ger | Gerund | Use discouraged by UD as the term is confusing; kept available | I look forward to seeing you |
Sources
- UD: Number (accessed 2026-11-17)
- UD: Tense (accessed 2026-11-17)
- UD: VerbForm (accessed 2026-11-17)
Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.







