Participants had their eye-movements recorded as they read vignettes containing implied promises and threats. We observed a reading time penalty when participants read the word 'threat' when it anaphorically referred to an implied promise. There was no such penalty when the word "promise" was used to refer to an implied threat. On a later measure of processing we again found a reading time penalty when the word 'threat' was used to refer to a promise, but also when the word 'promise' was used to refer to a threat. These results suggest that anaphoric processing of such expressions is driven initially by sensitivity to the semantic scope differences of 'threats' versus "promises". A threat can be understood as a type of promise, but a promise cannot be understood as a type of threat. However, this effect was short lived; readers were ultimately sensitive to mismatched meaning, regardless of speech act performed.