Statistics of When versus what: How time and valence of contextual descriptions modulate affective ERP components in response to facial expressions