Phonetic and phonological enhancement strategies in Tarifit robot-directed speech

This study examines phonological and phonetic adaptations in Robot-directed speech (Robot-DS) in Tarifit, an Amazigh language of Morocco. Thirty native speakers (younger and older adults) produced CCəC verbs in two contexts: baseline reading and interaction with a robot. Analyses focused on three features: (1) vowelless word realizations (schwa deletion), (2) schwa epenthesis in onset clusters, and (3) schwa duration. Results reveal categorical and gradient hyperarticulation in Robot-DS: vowelless forms are never produced; epenthetic schwas occurred more frequently, reducing consonant clusters; and prosodic template schwas were significantly lengthened. Age modulated durational, but not categorical, patterns: older adults produced greater lengthening in initial and confirm productions, suggesting pre-emptive hyperarticulation, while younger adults reserved maximal durational enhancement for error-repair contexts. These findings indicate that Tarifit speakers model robots as low-competence interlocutors. The results inform typological accounts of schwa variation and provide practical implications for Amazigh ASR and voice interface design.