From Newman’s Grammar of Assent:
Logical inference…will be found partly to succeed and partly to fail; succeeding so far as words can in fact be found for representing the countless varieties and subtleties of human thought, failing on account of the fallacy of the original assumption, that whatever can be thought can be adequately expressed in words…The concrete matter of propositions is a constant source of trouble to syllogistic reasoning…In inferential exercises it is the very triumph of that clearness and hardness of head…to have starved each term down till it has become the ghost of itself…a relation, a generalization, or other abstraction, a notion neatly turned out of the laboratory of the mind, and sufficiently tame and subdued, because existing only in a definition. (From Ian Ker’s anthology, The Genius of Newman, 97-99)
Logical inference can only work upon abstractions made from concrete knowledge, which it fashions by “starving each term down till it has become the ghost of itself.” (This implies a generally Aristotelian-empiricist account of knowledge, according to which “we name as we know” (as O’Callaghan puts it, Thomist Realism, 26)). The precision of the abstraction is inference’s strength and weakness; it allows thought to master the entailments of and connections among terms, though it proves, at the limit, to have trapped only a ghost in its machine, and not things themselves.