Antonyms in Context : A Corpus-Based Semantic Analysis of Swedish Descriptive Adjectives

Abstract: How are antonym relations acquired? What types of lexical information can be extracted from corpora and how? How can this information be encoded in a lexicon? The work in this book was developed within the framework of WordNet. A further elaborated lexical model is suggested, as well as methods for implementing it. The first part of the book is a study of adjectives co-occurring with adjectives. English direct antonyms, e.g. short-long, have previously been shown to co-occur in the same sentence significantly more often than expected according to the null hypothesis that the words of a corpus are randomly distributed. Using further elaborated methods that account for variation in sentence length, the studies in this book show that this is true also for Swedish direct antonyms, e.g. kort-long 'short'-'long'. However, there are various reasons why words co-occur; most word pairs from the same semantic scale co-occur more often than expected according to the null hypothesis. It is further shown that so called indirect antonyms, such as småväxt-reslig 'short of stature'-'stately' co-occur sententially less often than the direct antonyms of the same scales. It is also shown that the vast majority of the co-occurring antonyms appear in parallel context. These characteristics may facilitate the acquisition of antonym relations, and can also be used to find semantically related words in corpora. The second part of the book focus on the co-occurrence of adjectives and nouns, i.e. the semantic range of adjectives, which can be used to distinguish the various meanings of an adjective. The semantic ranges of full 'full' and tom 'empty' are described and compared in a case study. It is found that the semantic ranges of the words overlap only in the so-called container sense while only full is used in the so-called rod sense. The polysemy of the words and their diverging semantic ranges may explain why the two words do not co-occur significantly more often than expected, as other direct antonyms do. A case study of stor 'large' and 28 synonyms of stor describes the semantic ranges of the 29 words. The words were organised according to the most frequently modified semantic category of each word. This grouping was used as a basis to distinguish groups of synonyms, so called synsets, and to code the words in the Swedish WordNet. Four meanings of stor were distinguished: concrete dimension, importance, countable quantity, and uncountable quantity. The most frequent meaning found in corpora was not the core meaning, concrete dimension, which was outnumbered by both countable quantity and uncountable quantity. A study of the early use of stor in children shows that they in fact only use stor in the concrete-dimension sense, validating that this is the core meaning of stor. An explanation of the semantic shift of stor from the central meaning to importance, countable quantity, and uncountable quantity is suggested within a cognitive semantic framework. Various computer programs facilitating research such as presented in this book is also described.