Abstract
This thesis investigates why some phonological processes are typologically common, while others are typologically uncommon. One such typological asymmetry is that between final devoicing, a common, phonetically natural process, and final voicing, an uncommon, phonetically unnatural process. Two explanations have been proposed to account for this asymmetry: the channel bias account and the substantive bias account. Whereas proponents of the channel bias account hold that phonetic naturalness alone can explain typological asymmetries, proponents of the substantive bias account argue that phonetic naturalness is incorporated into grammatical constraints disfavouring the learning of unnatural processes. In this thesis, the asymmetry between final devoicing and final voicing is investigated using an artificial language learning experiment. Participants were assigned to either of two conditions, DEVOICING or VOICING, and trained on artificial miniature languages displaying these processes. Moreover, the experiment made use of the iterated learning paradigm, in which participants formed diffusion chains and the languages were transmitted from generation to generation in the chains. This design was used with the intention of investigating whether final voicing was learned to a lesser extent than final devoicing, and whether such an effect was amplified through language transmission. Such effects would support a substantive bias view. The results indicate that participants overall did not learn final voicing to a lesser extent than final devoicing. Participants learned the phonotactic restriction in their language, disallowing final voiceless obstruents in VOICING and final voiced obstruents in DEVOICING, equally well in the two conditions. A lower proportion of the items conforming to this phonotactic restriction were alternating in VOICING than in DEVOICING, but this difference was small, and it cannot be straightforwardly interpreted as a substantive bias effect. As for the iterated learning design, there was no indication that a potential substantive bias effect was amplified throughout the generations. In most diffusion chains in both VOICING and DEVOICING, the relevant process was lost, and this experiment thus indicates that larger scale iterated learning experiments must be conducted to determine whether this design can provide new insights to the study of typological asymmetries in phonology.