In his “Meaning and Inference” Sellars makes a distinction between material inferences and logical inferences. Logical inferences are classically valid and proceed from ‘p->q,p’ to ‘q’, while material inferences go from ‘p’ to ‘q’ based on their material contents. This distinction is also made in Carnap’s Logical Syntax of Language as L-inferences and P-inferences respectively. The paper goes through a detailed discussion and criticism of Carnap’s work, which more and more seems to be something I wll have to go through. I will mention a couple of points that Selalrs makes that interested me.

First he insisted on the rule part of ‘rule of inference’. He focused on the normative force that rules have and how they are involved in action. Reformulating a rule in such a way that it does not indicate an action is to eliminate the rule in favor of a description of the circumstances in which the rule can be applied. This seems basically right. His example is: X is arrestable =_{def} X broke a law. The left-hand side features a permissable action while the right-hand side features only a description of a state of affairs. He discusses logical necessity as being embodied in rules of inference. Something is logically necessary only if it conforms to a certain inference pattern. This is where he ties together modality and normativity with the phrase that Brandom deployed so well: Modality is a transposed language of norms. Sellars makes a lot of connections between object language and meta-language phenomena. He also shows how counterfactuals within a language can be eliminated if one endorses certain rules of inference in the metalanguage. One of my favorite comments made by Sellars is about the confusion of regularity with following a rule. Both are learned behavior, but there are many differences between the two and ignoring these differences has led to a lot of nonsense about ostensive definition, in roughly his words.

Second, one of the surprising things was how much importance was placed on counterfactual reasoning. The counterfactual inferences that one endorses show what content is assigned to each of the descriptive terms in the inferences. The fact that counterfactual reasoning is not classical and that the subjunctive conditional is not a material implication is seen as a feature and not a bug, pace Quine.

Third, there was a brief argument at the end about how modal and normative vocabulary does not assert psychological facts although it conveys them (I think that is how he put it.). This results in the claim that modal, normative, and psychological vocabulary are not reducible to each other. I did not follow this argument, but it was surprising to see so much of “Between Saying and Doing” show up in some guise in this essay.

Fourth, Sellars seems to launch sustained criticisms of empiricism in his work, particularly of the logical positivist kind. While he praises Carnap a great deal, I think he successfully showed that a functional natural language cannot do without P-inferences and intensional counterfactuals. Carnap thought that extensional L-inferences would do the job just fine. Granted, Carnap was making artificial calculi not natural languages, but he did think that the artificial languages were capable of being adopted by people as a natural language (according to Sellars).