Quantifying aspects of antonym canonicity in English and Swedish: textual and experimental

This paper highlights the potential usefulness of combining corpus methods and experimental methods to gain new theoretical insights into the role of antonymy as an organizing lexicosemantic principle in human thinking and languages’ vocabularies. We are intrigued by what distinguishes so-called canonical antonyms such as good-bad, long-short, thin-thick from other types of contrasts such as cold-scorching, pale-dark and speedy-slow. There are probably various converging reasons for perceptions of ‘goodness of antonymy’, e.g. frequency of co-occurrence, co-occurrence in certain constructions, e.g. whether slow or fast, stylistic co-occurrence preferences and pairwise acquisition (e.g. Muehleisen 1997, Willners 2001, Jones 2002, Murphy 2003). The research reported in this paper forms part of an international collaborative project on antonymy in discourse. This paper proposes (i) a principled method for creating a foundation for cross-linguistic comparisons of adjectival antonym pairings, using corpus methods and (ii) subsequently using the corpus data for the design of experiments on antonym canonicity. Since our investigations concern the nature of antonymy as a semantic relation and the patterns of lexicalization in English and Swedish, it is important to make sure that there is a reasonably robust basis of comparability between the languages under study. The assumption is that there is a scale of canonicity where some pairings are perceived as strongly antonymic while others are less strong, rather than a strict contrast between ‘direct’ and ‘indirect’ antonyms as assumed in some of the literature (e.g. Gross & Miller 1990). For instance, it is likely that speakers of English would regard slow – fast as a good example of a pair of strongly antonymic adjectives, while slow – quick and slow – rapid may be perceived as less good examples of antonyms, but still better than delayed – immediate. One way of making comparative investigations of antonyms is to identify dimensional commonalities and use them as a starting-point. These commonalities are onomasiological (ontological) in nature (Paradis 2005, Koptjevskaja-Tamm (forthcoming)), and frequency of co-occurrence is assumed to play a role for our perceptions about ‘goodness of antonymy’. For this purpose, we used English and Swedish corpus data as our source for selecting antonyms for our experiments of antonym canonicity in English and Swedish. We ran two types of experiments, a judgement experiment investigating goodness of pairings and an elicitation experiment measuring informant responses to individual words. Both experiment types made use of the same test items retrieved from the corpora. This paper reports on the method of selecting test items and some of the results from the elicitation experiments. Previous studies show that antonyms co-occur in sentences significantly more often than chance would predict (Justeson & Katz 1992, Jones 2002) and antonym pairs that we assume are perceived as canonical by native speakers co-occur more often than contextually restricted antonyms (Willners 2001). This textual evidence of sentential co-occurrence of antonyms is used in this study for retrieval of test items for experiments in order to secure a principled basis of test item collection across two languages. Willners (2001:83) and Holtsberg & Willners (2001) developed a computer program called Coco to calculate expected and observed sentential co-occurrences of words in a given set and their levels of probability. Coco was also designed to take sentence length variations into account which the