We (Katherine DeLong and colleagues in the Kutas Cognitive Electrophysiology lab) have used the event-related brain potential (ERP) method to demonstrate that the brain not only builds up representations of sentences as individual words are read or heard, but sometimes (perhaps all the time) even predicts (=neutrally pre-activates) specific words before receiving them, to varying degrees. This is an important demonstration because many language researchers do not believe in prediction. To test for prediction, we used our knowledge of the N400 brainwave component in combination with a good experimental design. The N400 is especially large to semantic anomalies, but is present for meaningful words to the extent that it is expected and/or fits with the ongoing semantic context. Operationally N400 amplitude is an inverse function of the eliciting item’s cloze probability -- the percentage of individuals providing that particular item in an offline sentence continuation task. The design took advantage of the fact that in English “a” precedes words beginning with consonant-initial sound whereas “an” precedes words beginning with a vowel initial sound (e.g., a response, an answer). A difference in N400 amplitude to “a” vs. “an” in sentences such as – "She was waiting for a/an … " when participants do not yet know what the next word (noun) will be indicates that they must have expectations whether the next word will begin with a vowel or consonant-initial sound – in other words, that they are using prior words/ideas to predict upcoming ones. Wow – caught you! ERPs in the right design are an excellent way to externalize aspects of mental processing. This finding indicates that although language comprehension must occur rapidly, with a nearly infinite number of possible word combinations, the brain nonetheless anticipates continuations to varying degrees.