Recent findings suggest that communicative context affects the timing and magnitude of emotion effects in word processing. In particular, social attributions seem to be one important source of plasticity for the processing of affectively charged language. Here, we investigate the timing and magnitude of ERP responses toward positive, neutral, and negative trait adjectives during the anticipation of putative socio-evaluative feedback from different senders (human and computer) varying in predictability. In the first experiment, during word presentation participants could not anticipate whether a human or a randomly acting computer sender was about to give feedback. Here, a main effect of emotion was observed only on the late positive potential (LPP), showing larger amplitudes for positive compared to neutral adjectives. In the second study the same stimuli and set-up were used, but a block-wise presentation was realized, resulting in fixed and fully predictable sender identity. Feedback was supposedly given by an expert (psychotherapist), a layperson (unknown human), and again by a randomly acting computer. Main effects of emotion started with an increased P1 for negative adjectives, followed by effects at the N1 and early posterior negativity (EPN), showing both largest amplitudes for positive words, as well as for the LPP, where positive and negative words elicited larger amplitudes than neutral words. An interaction revealed that emotional LPP modulations occurred only for a human sender. Finally, regardless of content, anticipating human feedback led to larger P1 and P3 components, being highest for the putative expert. These findings demonstrate the malleability of emotional language processing by social contexts. When clear predictions can be made, our brains rapidly differentiate between emotional and neutral information, as well as between different senders. Attributed human presence affects emotional language processing already during feedback anticipation, in line with a selective gating of attentional resources via anticipatory social significance attributions. By contrast, emotion effects occur much later, when crucial social context information is still missing. These findings demonstrate the context-dependence of emotion effects in word processing and are particularly relevant since virtual communication with unknown senders, whose identity is inferred rather than perceived, has become reality for millions of people.