In linguistics, a distinctive feature is the most basic unit of phonological structure that may be analyzed in phonological theory.

Distinctive features are grouped into categories according to the natural classes of segments they describe: major class features, laryngeal features, manner features, and place features. These feature categories in turn are further specified on the basis of the phonetic properties of the segments in question.^[1] For phonemes to be in a particular natural class, they have to share the same distinctive features such as articulation and/or sound similar to each other. We can find distinctive features between two words by finding the minimal pair between them. The minimal pair are when two words sound the same, but they are different in definition because the pair has different phonemes from each other.^[2]

Since the inception of the phonological analysis of distinctive features in the 1950s, features traditionally have been specified by binary values to signify whether a segment is described by the feature; a positive value, [+], denotes the presence of a feature, while a negative value, [−], indicates its absence. In addition, a phoneme may be unmarked with respect to a feature. It is also possible for certain phonemes to have different features across languages. For example, [l] could be classified as a continuant or not in a given language depending on how it patterns with other consonants.^[3] After the first distinctive feature theory was created by Jakobson in 1941, it was assumed that the distinctive features are binary and this theory about distinctive features being binary was formally adopted in “Sound Pattern of English” by Chomsky and Halle in 1968. Jakobson saw the binary approach as the best way to make the phoneme inventory shorter and the phonological oppositions are naturally binary.^[4]

In recent developments^[when?] to the theory of distinctive features, phonologists have proposed the existence of single-valued features. These features, called univalent or privative features, can only describe the classes of segments that are said to possess those features, and not the classes that are without them.^[5]

Major class features: The features that represent the major classes of sounds.

Laryngeal features: The features that specify the glottal states of sounds.

Manner features: The features that specify the manner of articulation.

However, laryngoscopic studies suggest^{[citation needed]} the features are

The concept of a distinctive feature matrix to distinguish similar elements is identified with phonology, but there have been at least two efforts to use a distinctive feature matrix in related fields. Close to phonology, and clearly acknowledging its debt to phonology, distinctive features have been used to describe and differentiate handshapes in fingerspelling in American Sign Language.^[9] Distinctive features have also been used to distinguish proverbs from other types of language such as slogans, clichés, and aphorisms.^[10]

Analogous feature systems are also used throughout Natural Language Processing (NLP). For example, part-of-speech tagging divides words into categories. These include "major" categories such as Noun vs. Verb, but also other dimensions such as person and number, plurality, tense, and others. Some mnemonics for part-of-speech tags conjoin multiple features, such as "NN" for singular noun, vs. "NNS" for plural noun, vs. "NNS$" for plural possessive noun (see Brown Corpus). Others provide more explicit separation of features, even formalizing them via markup such as the Text Encoding Initiative's feature structures. Modern statistical NLP uses vectors of very many features, although many of those features are not formally "distinctive" in the sense described here.