INTRODUCTION

In our understanding of how language works, the word is a crucial element. The lexicon is usually seen as collection of individual words, which form the building blocks of larger utterances. One important characteristic of words is that they seem to participate in different kinds of “lexical partnerships” (Singleton, 2000, p. 47). Because some words are often found in the company of certain other words, lexical partnerships of various kinds have long been recognized in language research.

As the ubiquity of such language phenomena is increasingly noticed, the term “formulaicity” seems to be emerging as the most popular “umbrella” term used to refer to the quality of language that makes it impossible to reduce it simply to individual words and syntactic rules, and to refer to the many different kinds of connections between words in a language (e.g. Meunier, 2012).

To provide the broadest possible definition of formulaicity, it is useful to take as the starting point the traditional “slot and filler” approach to language. In this view, for each “slot” along the syntagmatic axis, one word is chosen from a number of options available along the paradigmatic axis, depending on the intended meaning. The choices for each slot, however, are restricted to certain classes of words by syntagmatic restraints which result from the syntactic rules of the language. In its broadest sense, formulaicity can be taken to mean all the phenomena which affect how words are combined together which fall outside the “slot and filler” model. Boers and Lindstromberg offer a good practical definition of formulaicity: “the use of word strings that have become conventionalized in a given language as attested by native-speaker judgment and/or corpus data” (2012, p. 83). Many authors use definitions of formulaic expressions which are based on the assumed holistic processing, such as: “frequent multiword combinations that are stored and retrieved holistically from the mental lexicon at the moment of speech” (Nekrasova, 2009, p. 647). However, such definitions are better avoided, since proving that multiword combinations are stored/retrieved holistically is in itself problematic, as will be explained in more detail below.