When a sender interprets data to persuade a receiver's belief about a state, the receiver faces two sources of content simultaneously: the strategic content - what the sender's choice of interpretation reveals about her private information, and the semantic content - what her interpretation literally says about the state. This paper develops a framework that accommodates consideration of both sources and characterizes a class of inferential behaviors. We axiomatize the receiver's inferential rule and show it admits a representation as a convex combination of strategic and semantic content, parameterized by a weighting function $\lambda$. The framework unifies cheap talk $(\lambda = 1)$ and narrative persuasion $(\lambda = 0)$ as special cases of a common primitive, and yields tractable comparative statics on how the inferential rule shapes equilibrium. When $\lambda$ partially weights both sources, the sender's language choice generates novel tradeoffs: semantic content anchors beliefs, which constrains feasible communication, and even negligible semantic sensitivity makes the cheap talk benchmark a knife-edge rather than a robust prediction. A receiver applying purely strategic inference in a cheap talk setting may be susceptible to narrative persuasion once messages with semantic content are available.