Proceedings

Modeling the Learning of Tier Projection with Blocking in Non-Local Phonological Processes

---

Abstract

Phonological theories often address long-distance dependencies by projecting segments onto tiers, but relatively little work has explored how such tiers can be learned from data. The D2L model (Belth, 2024) offers a learning-based account of tier induction by incrementally removing all the projected adjacent segments that cannot account for the alternation from the tier. However, this model fails to learn the correct tier in cases involving inert blockers, which it incorrectly removes from the tier. The present paper improves on D2L by distinguishing transparent segments from inert blockers during tier updating, ensuring that blockers remain on the tier but are excluded from the rule condition. Blockers are distinguished from transparent segments by checking whether the non-triggering adjacent segments consistently predict the default target form, and the default form is established and updated via a trial-and-error process, updated together with the tier and rule condition. By adding these procedures to the D2L model, the current model successfully learned vowel harmony involving blocking, and thus provides a more comprehensive learning procedure of phonological tiers.

Keywords: phonological learning, long-distance dependencies, blocking, tiers

How to Cite:

Yin, Y., (2026) “Modeling the Learning of Tier Projection with Blocking in Non-Local Phonological Processes”, Proceedings of the Annual Meetings on Phonology 2(1). doi: https://doi.org/10.7275/amphonology.3701