That almost completes the description of our version of Eliza. There is one gap in the structure to fill in. This concerns what happens when several possible responses are possible. For instance
i wouldn't pay money for a computer
fits patterns [== money ==] and [== computer ==]. Eliza deals with this in a simple way: it selects one of the matching patterns at random, and generates a response based on that one. Special tests, like whether a sentence is particularly short or long, are just treated as special kinds of patterns that can be matched (though not using the notation above).