Testing the Potential of Automatically Inferred Affix Colexifications for Linguistic Typology

Cross-linguistic colexification patterns have proven useful for quantitative studies in lexical typology. While most studies focus on full colexification, where senses are co-expressed by the same word form, recent studies have proposed to compute partial colexifications, where senses are not colexificied by entire words, but only by parts of them. Among these, affix colexifications, where one word recurs in the end or the beginning of another word, show interesting properties, potentially reflecting word formation processes giving hints cross-linguistic motivation patterns. In order to test their potential, we conduct a detailed case study. Based on a large sample of cross-linguistic partial colexification patterns, computed from the Database of Cross-Linguistic Colexifications, we first check to which degree partial colexifications reflect true cases of lexical motivation and then carry out a detailed comparison of concept relations underlying frequent partial and full colexification patterns. Our results show that partial affix colexifications that recur across five and more language families tend to reflect true lexical motivation patterns in almost 90% of all cases. Furthermore, we find that majority of affix colexifications and full colexifications reflect contiguity relations. However, the proportion of contiguity relations in partial colexifications exceeds the proportion of contiguity relations in full colexifications (50% vs. 40%), showing that there are differences in the semantics reflected by both colexification types.