Paper

Towards a Learning-Based Account of Underlying Forms: A Case Study in Turkish

---

Abstract

A traditional concept in phonological theory is that of the underlying form. However, the history of phonology has witnessed a debate about how abstract underlying representations ought to be allowed to be, and a number of arguments have been given that phonology should abandon such representations altogether. In this paper, we consider a learning-based approach to the question. We propose a model that, by default, constructs concrete representations of morphemes. When and only when such concrete representations make it challenging to generalize in the face of the sparse statistical profile of language, our proposed model constructs abstract underlying forms that allow for effective generalization. As a case study, we consider the highly agglutinative language, Turkish. We demonstrate that the underlying forms that our model constructs account for the complexities of Turkish phonology resulting from its multifaceted vowel harmony. Moreover, these underlying forms enable the highly-accurate prediction of novel surface forms, demonstrating the importance of some underlying forms to generalization.

Keywords: underlying representations, morphophonological learning, alternations, Turkish

How to Cite:

Belth, C., (2023) “Towards a Learning-Based Account of Underlying Forms: A Case Study in Turkish”, Society for Computation in Linguistics 6(1), 332-342. doi: https://doi.org/10.7275/bc8q-vj22