Basically, the idea is that the favourite metaphors of a given culture will be reflected both in its language (notably by compounds, but also in semantic shifts) and in its arts. Thus, to quote one of his examples, in Tewa "roof" is literally "wooden coil-basket", although modern Tewa roofs do not look much like that, while the roofs of Mesa Grande kivas were built to resemble coil baskets. He takes both to exemplify a metaphor BUILDINGS ARE CONTAINERS, which he takes to be supported not only by this case but by a number of other features, such as the use of pottery design motifs on walls and the polysemy of a word meaning "lake", "ceremonial bowl", and "kiva".
I'm not sure how often this is likely to work in practice. For it to work, your metaphors have to be reflected in the kind of material culture that archeologists can dig up – buildings, pottery, baskets if you're lucky. It would seem to require, minimally, a strong tradition of more or less representational art. I would be hard-pressed to think of such cases in, say, North Africa, unless you go further back than we can reconstruct the languages. But where those preconditions are fulfilled, it does strike me as an interesting approach to try, because it targets the kinds of meaning that the speakers themselves would have considered important.