Fodor and Pylyshyn on Connectionism
Levels of description
The (in)famous argument
- Connectionism is either a theory at the implementation level, or it is a theory at the level of representation and algorithm.
- If it is the former, then it is compatible with LOT.
- If it is the latter, then it is also compatible with LOT.
- So connectionism is compatible with LOT.
Argument for P2
- LOT can be implemented by neural networks or in other ways.
Argument for P3
- Connectionism must be compatible with LOT at this level in order to explain systematicity, productivity, inferential coherence.
- Systematicity - if you can think A loves B, you can also think B loves A.
- Productivity - you can have an indefinite number of different thoughts
- Inferential coherence - being able to carry out coherent reasoning, e.g. if every cat can meow, and Garfield is cat, then Garfield can meow.
- Any counterexample to systematicity?