Abstract
Division of words into parts of speech (POS) is one of the most traditional fields of grammar and rarely the subject of dispute. Although the characteristics used for defining POS vary from language to language, three criteria are relatively common – semantic, syntactic and morphological.
In this paper I would like to introduce a new criterion and a new approach to POS division, based on context. Since we know that the immediate context is more influenced by the key word than the distant one, we can assume that this context reflects the original word (the repertory of contexts of a word can be referred to as its fingerprint).
Adopting an approach widely used in semantics which assumes that the lexical meaning of a word is given by its contexts, we constitute classes of words (POS) that have similar contextual fingerprints. This corpus-driven approach can lead to a somewhat different set of POS with a different classification, based solely on formal and textual corpus data.