A huge question in NLP is how the discrete symbolic structures that we think characterize natural language can be embedded in high-dimensional continuous space. This paper proposes a solution that justifies some ad-hoc results from before and could form the basis for better means of embedding in the future.