Human language surpasses all other animal communication systems in its complexity and generative power. I use behavioral, neuroimaging, and computational approaches to illuminate the functional architecture of language, with the goal of deciphering the representations and computations that enable us to understand and produce language. I will discuss three discoveries about the language system. First, I will show that the language network is selective for language processing over a wide range of non-linguistic processes. Next, I will challenge current proposals of the neural architecture of language, which argue that syntax (the rules for how words combine into phrases and sentences) is cognitively and neurally dissociable from the lexicon (word meanings). I will show that syntactic processing is not localized to a particular region within the language network, and that every brain region that responds to syntactic processing is at least as sensitive to word meanings, including when probed with a high-spatial/high-temporal-resolution method (ECoG). Finally, I will provide evidence that stimuli that are not syntactically well-formed but allow for meaning composition (operationalized within an information-theoretic framework) elicit as strong a response as intact sentences, suggesting that semantic composition may be the core driver of the response in the language-selective brain regions.