Behavioral profiles: A corpus-based perspective on synonymy and antonymy*

1 Introduction 1.1 Two empirical perspectives in the study of synonymy and antonymy The domain of linguistics that has arguably been studied most from a corpus-linguistic perspective is lexical, or even lexicographical, semantics. Already the early work of pioneers such as Firth and Sinclair has paved the way for the study of lexical items, their distribution, and what their distribution reveals about their semantics and pragmatics / discourse function(s). A particularly fruitful area has been the study of (near) synonyms; 1 probably every corpus linguist has come across the specific example of strong and powerful – the fact that one would say strong tea but not powerful tea – as well as the general approach of studying synonyms on the basis of their distributional characteristics. However, while synonymy is probably the most frequently corpus-linguistically studied lexical relation, there is by now also quite some corpus-based work on its counterpart relation, antonymy, and we will briefly discuss examples of both below. The study of the semantics of lexical items quite obviously presupposes two central concepts. First, one requires a notion of what it means to know a word, and for reasons that we will elaborate on more below, we follow here an approach by Miller and Charles (1991), who proposed the notion of a contextual representation, which is: • " knowledge of how that word is used " (p. 4; cf. also Miller 1999); • " some abstraction or generalisation derived from the contexts that have been encountered " (p. 5); • " a mental representation of the contexts in which the word occurs, a representation that includes all of the syntactic, semantic, pragmatic, and stylistic information required to use the word appropriately. " (p. 26). Second, one requires some kind of scale or, more likely, multidimensional space of semantic similarity along/within which words or, more precisely, concepts or