In almost all cases he mentions, the result has been a net simplification of the language--e.g. fewer cases.
Aside from enforced vocabulary mixing (e.g. beef/cow), can you think of instances when a language became more complex?
If not, then we have a problem.
- Perhaps we have had a bias for observing simplifications.
- Maybe linguists have been using the wrong tools to project into the past. Since these are dead languages, they have to make some assumptions.
- Maybe languages tend to become simpler when they get recorded, either natively or when interacting with literate cultures. Pidgins do tend to have simpler structure--but the changes don't seem that dramatic.
- If not, then we have an "expanding universe" kind of problem. Extrapolating back points to greater and greater complexity. I'm not sure how to extrapolate with any precision--it depends partly on what sorts of other cultures your test language happened to run into. The best I can do is say: "It was more complicated."
The language families are different enough that you can't point back to a single ur-language--you wind up with several of them. Sort of like Babel.
Are there good examples of complexification?