A law of meaning

This article rejects the canonical ideal of a one-to-one correspondence between meaning and marker and proposes a set-theoretical and optimality-based law for the relationship between meaning and its markers which allows for distinguishing true markers (such as not, no, never for negation) from otherwise associated items (such as negative polarity items as but, any):a meaning is expressed by the set of non-randomly recurrent markers that together are the best collocation of that meaning. We implement the law in an algorithm using Dunning’s log-likelihood and illustrate it by extracting markers for ‘know’, negation, first person subject and complementizers from translations of the New Testament in a variety sample of 83 languages with manual evaluation of all extracted markers. Markers are extracted from unannotated texts (with lexemes being just a special case of marker-set coalition phenomena) considering just one meaning at a time (without any need for accounting for specific coexpression types).