Phonotactic constraints and learnability: analyzing Dagaare vowel harmony with tier-based strictly local (TSL) grammar

Abstract This paper examines vowel harmony in Dagaare using the Tier-Based Strictly Local (TSL) framework, focusing on tongue root, rounding, backness, and height harmonies. While vowel harmony in Niger-Congo languages, particularly Dagaare, has been explored from phonetic, phonological and typological perspectives, computational insights remain limited. The study applies the TSL framework to model the phonotactic constraints governing harmonic patterns, projecting only harmony-relevant features onto a tier to capture non-local dependencies while ignoring irrelevant segments. This approach allows for precise modeling of agreement among non-adjacent vowels and demonstrates that all harmony types in Dagaare can be represented within a single TSL grammar, avoiding the need for separate tiers for each feature. The findings indicate that the Dagaare system is robustly learnable from surface data under TSL constraints, offering a computationally tractable path for both human and machine learners. A key limitation is also identified: TSL fails to account for harmony exceptions in morphologically complex words, such as compounds, due to its lack of morphological domain sensitivity. The study contributes to the typological understanding of Dagaare, illustrates the utility of TSL for modeling complex harmony systems, and recommends extensions such as domain-sensitive tier projection to better handle morphologically complex contexts.