Since Wierzbicka (1982) inquired in the title of her paper why one can have a drink , but not have an eat , there have been several efforts coming from different schools of thought to semantically justify restricted lexical co-occurrence between the components of collocations. Even though these analyses can be considered well-founded, I believe that the distinctions made are so complex and so overspecifi ed that they are in no way useful either for the language learner or for a natural language processing system. Regardless of their usefulness, semantic explanations tend to blur the difference between free co-occurrence and restricted co-occurrence, and, consequently, they deny the existence of collocate lexical units (LUs), i.e. LUs whose selection is lexically controlled by other lexical elements. Nevertheless, for the extreme cases, it is diffi cult to deny that there are two different types (free and restricted) of lexical combinations. For example, there is no doubt that in the lexical entry for table we do not need to indicate that this noun can be combined with large or black because these adjectives denote properties

that can be predicated of any physical object such as the one designated by the noun table . In contrast, it is necessary to specify that the noun walk is combined with the verb take in English, while, in French, promenade ‘walk’ goes with faire ‘do/make’, because from the perspective of production or encoding, we cannot deduce from the meaning of these nouns which verbs they are combined with. Therefore, the aim of this chapter is to argue in favour of there being something intrinsically different between a free combination such as large table and a restricted combination such as to take a walk . Although not all collocations are as prototypical as to take a walk , from a lexicographical perspective [→ Section 2], we also have to take the intermediate cases into account and be able to describe them.