Based on Mandarin question markers' embeddability profiles in nonquotational subordinating contexts, this paper argues in favor of Dayal's (2025) three-way split in projections that contribute to the illocutionary force, attributing the non-co-occurrence of question markers to their semantics. Specifically, we propose that (i) LowIForceP houses the most embeddable meiyou, OpWH, and A-not-A; (ii) PerspP houses question markers restrictedly embeddable like bu and ma; and (iii) HighIForceP houses the unembeddable ba. Looking deeper into the embeddability patterns of bu and ma, we argue that our PerspP should be interpreted `a la Dayal (2025) in the sense that it introduces a not-at-issue requirement for the embedded question to be potentially active for the matrix subject. We explicate that the embedding of bu and ma is restricted by two constraints: (i) the embedded clause size must satisfy the syntactic selectional requirement of the embedding context; and (ii) the overall interpretation must be compatible with a not-at-issue requirement lexically introduced. Our proposal calls for a theory of selection that manifests a conspiracy of syntax and semantics.
Liu et al. (2026) studied this question.