In linguistics, a hyponym is a word or phrase whose semantic range is included within that of another word. For example, scarlet, vermilion, carmine, and crimson are all hyponyms of red (their hypernym), which is, in turn, a hyponym of colour.
According to Fromkin and Rodman, hyponyms are a set of related words whose meaning are specific instances of a more general word. For example, red, white, blue, etc., are hyponyms of colour. Hyponymy is thus the relationship between a general term such as polygon and specific instances of it, such as a triangle.
Computer science often terms this relationship an is-a relationship. For example, Red is a colour that can be used to describe the hyponymic relationship between red and colour. The term Hypernym denotes a word, usually somewhat vague and broad in meaning, that other more specific words fall under or are fairly encompassed by.
For example, vehicle denotes all the things that are separately denoted by the words train, chariot, dogsled, aeroplane, and automobile and is, therefore, a hypernym of each of those words. Conversely, the words train, chariot etc. are hyponyms of vehicle.
Hypernymy is the semantic relation in which one word is the hypernym of another. Hypernymy, the relation words stand-in when their extensions stand in the relation of class to subclass, should not be confused with holonymy which is the relation words stand-in when the things that they denote stand in the relation of the whole to part. A similar warning applies to hyponymy and meronymy.
Automatically Finding Hypernyms
One of the first suggestions on how to find hypernym/hyponym pairs in a text came from Marti Hearst, who suggested looking at the output of a parser and taking all of the terms linked by constructions such as X and other Y; X could be considered a possible hyponym of Y. This method was extended by Snow et al, who developed an automated method of finding possible constructions that could signal such a pair.
Their process works by taking hypernym/hyponym pairs from WordNet and finding many noun-noun pairs from a parsed corpus. They train a classifier to select those pairs of words that have a high probability of being hypernym pairs given the constructions which link the terms in the corpus.