Word learning can help individuals master the basic knowledge of science and culture, promoting social interaction. A large number of researchers carried out studies on word learning from different angles. But most previous studies of word learning are less to inspect in context and ignored the impact of the emotional valence. This makes the learning process too unrealistic and the results are lack of a certain ecological validity. So the present study determined to simulate natural learning context to investigate the regulation role of angry, happy and neutral emotional prosody in word learning in high limited context combining with event-related potential technique with high temporal resolution. During the encoding phase, the prosodic contexts were presented via loud speakers firstly and then the to-be-learned words appeared. Participants should infer and learn the meaning of the pseudo-word according to the prosodic context. The correspondent real words were subsequently given to the participants for checking their guessing. To help the consolidation process, the pseudo-words were presented again after the checking. During the testing phase, two kinds of tasks were employed. In experiment 1, participants were asked to judge the concreteness of real words which were primed by consistent or inconsistent pseudo words. While in experiment 2, the task was to explicitly judge the meaning consistency between the real words and pseudo-words. Results of experiment 1 showed that there were significant effects of concreteness of words and emotional prosody, ie., the concrete words had processing advantage over the abstract words and angry prosody showed longer reaction time than happy and neutral prosody. More importantly, there were significant interactions between emotional prosody and words’ concreteness, and the further simple analysis revealed that emotional prosody effects only showed in abstract words not in concrete words, indicating that abstract words compared to concrete words are more susceptible to the impact of emotional prosody. Behavior results of experiment 2 were similar to the experiment 1, and the processing for abstract words learned in angry contexts was impact compared to the words learned in neutral and happy contexts. ERP results experiment 2 showed that:(1) During the testing phase, angry prosody elicited larger N300 than neutral prosody, indicating that in the early lexical semantic processing component, angry prosody entailed more cognitive load; concrete words evoked smaller N400, LPC than abstract words, indicating that concrete word showed semantic processing advantages at late lexical semantic processing component. (2) During the encoding phase, the angry and happy elicited larger N400 and smaller LPC than neutral prosody, which reflected that there more cognitive afford for retrieving the meaning of the target words, and the reanalysis and integration process of them was inhibited since the cognitive resources were taken up due to emotional arousal. Furthermore, concrete words evoked larger N400 than abstract words, indicating that concrete word showed semantic processing advantages at late lexical semantic processing component. Taken together, conclusions as follow could be made through our experiments. (1) Contextual emotional prosody has different modulating effects on word learning. Angry prosody has a significant negative effect on word learning. (2) The influence of emotional prosody was moderated by the concreteness of the learned; emotional prosody has a bigger effect on abstract words learning. (3) The modulating role of emotional prosody in word learned could be originated from the earlier semantic mapping and later semantic integration processes during the encoding phase.