Negation height and NPI licensing in Jordanian Arabic: Experimental support for an agree-based account

This paper presents the first experimental investigation of licensing of Negative Polarity Items (NPIs) in Jordanian Arabic (JA). NPIs are expressions that are restricted to non-affirmative environments such as negation and other nonveridical contexts. JA is an ideal testing ground because it employs two structurally distinct negation strategies: a high clausal negator mā and a bipartite low negation ma–V–š . We report two offline experiments, a 7-point acceptability judgment task and a forced-choice interpretation task, crossing NPI position (preverbal vs. postverbal) with negation type (high vs. bipartite). The results reveal a robust interaction: preverbal NPIs are licensed only under high mā , while postverbal NPIs are licensed only under bipartite ma–V–š . Since both negators create equally nonveridical environments, the asymmetry cannot be reduced to semantic weakness. Instead, it follows from structural licensing: mā occupies a high Neg head above TP, ma–V–š realizes a single low Neg head under T, and NPIs are licensed accordingly, with adverbial ʕumr (‘ever’) requiring a high licensor and walā ḥada (‘anyone’) belonging to a low concord system. Licensing therefore reduces to Agree between [iNEG] and [uNEG] features under c-command. These findings provide experimental confirmation for feature-based syntactic analyses of polarity (Zeijlstra, 2004; Bjorkman & Zeijlstra, 2019), while restricting scope-based alternatives to explicitly structure-sensitive formulations (Barker, 2018). More broadly, the JA data align with typological evidence from German, Japanese, and Czech, showing that polarity sensitivity is determined not by semantic nonveridicality alone but by fine-grained structural dependencies at the syntax–semantics interface. • First experimental test of NPI licensing in Jordanian Arabic. • Negation height—not just meaning—drives where NPIs are allowed. • Preverbal NPIs license with high mā ; postverbal NPIs with ma –V– š. • Findings support a structure-based (Agree) account of polarity. • Two tasks (ratings, interpretations) show the same robust interaction.