Modeling distance-based variable sibilant harmony in Moroccan Arabic with a MaxEnt grammar

Abstract The aim of this study is to examine the variable patterns of regressive sibilant harmony in Moroccan Arabic. This process is triggered by the palatal fricative [ʒ], and targets the alveolar fricatives [z] and [s], changing them to [ʒ] and [ʃ] respectively ( Harrell, 1962 ; Heath, 1987 , 2002 ). This paper examines how factors such as the distance between harmonizing segments, voicing of the target segment, and morphological complexity influence the likelihood of harmonization. The findings from an experimental study reveal that the distance between the two harmonizing sibilants significantly influences harmonization, with a shorter distance indicating a higher likelihood of harmonization. Moreover, while voicing of the target sound appears to affect harmonization, a closer examination of the experimental results attributes this effect to the exceptional behavior of some words whose harmonizing sound happens to be [s]. The study uses Maximum Entropy grammar ( Goldwater & Johnson, 2003 ) to learn the variability in applying harmonization. The analysis incorporates Agreement-by-Correspondence constraints ( Rose & Walker, 2000 , 2004) to account for the observed distance effects and lexically-indexed constraints ( Pater, 2000 , 2009 ) to capture the exceptional behavior of certain lexical items that do not follow the general trends.