Strong Associations Can Be Weak: Some Thoughts on Cross-lingual Word Webs for Translation

This paper discusses the implications of human word association norms on the modelling of word associations from large corpora and the relevance of different types of associations in the process of translation, with a focus on adjectives. It is observed that the proportion of paradigmatic responses found in English norms tends to be higher, whereas a clear preference for syntagmatic associations is exhibited in Chinese norms. Further comparison with corpus-based extracted associations, using various functions in the Sketch Engine, shows that collocational associations might be more effectively extracted, but there is also considerable individual variation for different words. It is suggested that although free associations elicited in isolated context serve to reveal a wide range of potential lexical relations, their usefulness and relevance in real language applications should consider the actual task and its information demand. A purposebased approach to construct cross-lingual word webs for computer-aided translation is thus proposed.