Contextual Influences on Near-Synonym Choice

One of the least understood aspects of lexical choice in Natural Language Generation is choosing between near-synonyms. Previous studies of this issue, such as Edmonds and Hirst [4], have focused on semantic differences between near-synonyms, as analysed by lexicographers. Our empirical analysis of near-synonym choice in weather forecasts, however, suggests that other factors are probably more important than semantic differences. These include preferences and idiosyncrasies of individual authors; collocation; variation of lexical usage; and position of a lexeme in a text. Our analysis also suggests that when semantic differences do influence near-synonym choice, they may do so in an author-dependent manner. Thus, at least in our domain, ‘context’ (including author) seems to be more important than semantics when choosing between near-synonyms.