Experiments in morphological evolution

Richard Sproat
University of Illinois at Urbana/Champaign


Morphology, inflectional morphology in particular, can often show surprising complexity in natural languages. In highly inflected languages, one often finds the situation where words fall into different inflectional classes, showing different markings for the same function. Syncretism – the same morphological marker serving several different functions – abounds, and sometimes this syncretism is systematic, motivating "rules of referral" that tie the expression of a particular set of morphosyntactic features to the expression of another. In a similar fashion, words may show seemingly arbitrary stem alternations, with particular stem variants being associated with particular slots in a paradigm: often these stem variants have no apparent phonological motivation.

How does such complexity come about?

I will attempt to provide partial answers by discussing some experiments in morphological evolution in multi-agent systems; this work follows very much in the tracks of previous work such as that of Kirby or Nettle or Wang and colleagues. One of the conclusions that seems to follow from this research is that some phenomena, such as "rules of referral", may be less interesting than their prominence in the linguistics literature might suggest. That is, the conditions that would motivate a linguist to posit a rule of referral can arise due to weak biases in the system, without any particular reference to global notions of the form "render slot X in the same way as slot Y".