The meaning of a natural language utterance is largely determined from its\nsyntax and words. Additionally, there is evidence that humans process an\nutterance by separating knowledge about the lexicon from syntax knowledge.\nTheories from semantics and neuroscience claim that complete word meanings are\nnot encoded in the representation of syntax. In this paper, we propose neural\nunits that can enforce this constraint over an LSTM encoder and decoder. We\ndemonstrate that our model achieves competitive performance across a variety of\ndomains including semantic parsing, syntactic parsing, and English to Mandarin\nChinese translation. In these cases, our model outperforms the standard LSTM\nencoder and decoder architecture on many or all of our metrics. To demonstrate\nthat our model achieves the desired separation between the lexicon and syntax,\nwe analyze its weights and explore its behavior when different neural modules\nare damaged. When damaged, we find that the model displays the knowledge\ndistortions that aphasics are evidenced to have.\n