Researchers have shown growing interest in how pragmatic reasoning shapes the social impressions listeners form of speakers. This study investigates the interaction between linguistic choice, contextual constraints, pragmatic inference, and social meaning, introducing the Social Evaluation Model (SEM) as a formal account of how listeners derive social judgments from language use. SEM captures how evaluations of a speaker's competence and likability emerge from inferences about the speaker's knowledge state and communicative motives within a given situation. We test the model's predictions using experimental data on numerical (im)precision and demonstrate how the framework extends to other domains in which pragmatic reasoning influences social inference. Beyond its theoretical contribution to pragmatics and sociolinguistics, this work also points to an important avenue: examining whether large language models exhibit more human-like social reasoning when explicitly guided through structured pragmatic inference steps, as characterized by SEM.