From Popper to Lakatos: A case for cumulative computational modeling
Ardi Roelofs · 2005
An important problem with several modeling enterprises in psycholin-guistics is that they are not cumulative, unlike successful experimental research. For example, in the field of language production, quite a few models focus on a few findings only instead of trying to account simulta-neously for a wide range of data. Even worse, some investigators treat their models like their toothbrushes by using them only for their own data. There is no guarantee that these micromodels can be integrated into a single comprehensive macromodel, because micromodels are often mu-tually incompatible. Moreover, experimental tests of models developed by others are often conducted in the world of a misinterpreted Popper, where testing models is like skeet shooting. 1 The aim is to shoot down 1 Lakatos (1970) distinguished three Poppers: Popper0, Popper1, and Popper2. “Popper0 is the dogmatic falsificationist who never published a word: he was invented—and ‘criticized’—first by Ayer and then by many others. Popper1 is the naive falsificationist, Popper2 the sophisticated falsificationist. The real Popper developed from dogmatic to a naive version of methodological falsificationism in the twenties; he arrived at the ‘acceptance rules ’ of sophisticated falsificationism in the fifties. Thus the real Popper consists of Popper1 together with some elements of Popper2 ” (p. 181). Skeet shooting is often defended by referring to the mythical