Our thesis is that computing with words needs to account for the uncertainties associated with the meanings of words, and that these uncertainties require using type-2 fuzzy sets. Doing this leads to a proposed architecture for making judgments by means of computing with words - a perceptual computer Per-C. The Per-C includes an encoder, a type-2 rule-based fuzzy logic system, and a decoder. It lets all human-computer interactions be performed using words. In this paper, a quantitative language is established for the Per-C, and many open issues about the perceptual computer are described.