Suffixing, prefixing, and the functional order of regularities in meaningful strings

The world’s languages tend to exhibit a suffixing preference, adding inflections to the ends of words, rather than the beginning of them. Previous works has suggested that this apparently universal preference arises out of the constraints imposed by general purpose learning mechanisms in the brain, and specifically, the kinds of information structures that facilitate discrimination learning (St Clair, Monaghan, & Ramscar, 2009). Here I show that learning theory predicts that prefixes and suffixes will tend to promote different kinds of learning: prefixes will facilitate the learning of the probabilities that any following elements in a sequence will follow a label, whereas suffixing will promote the abstraction of common dimensions from a set of preceding elements. The results of the artificial language learning experiment support this analysis: When words are learned with consistent prefixes, participants learned the relationship between the prefixes and the noun labels, and the relationship between the noun labels and the objects associated with them, better than when words were learned with consistent suffixes. When words were learned with consistent suffixes, participants treated similarly suffixed nouns as being more similar than nouns learned with consistent prefixes. It appears that while prefixes tend to make items more predictable and to make veridical discriminations easier, suffixes tended to make items cohere more, increasing the similarities between them.