We have, then, the three classes--(1) propositions which assert "relations of ideas"; (2) propositions which assert "matters of fact" for which we have the evidence of direct observation or personal memory; (3) propositions which assert "matters of fact_"_ for which we have _not_ this evidence. And as regards propositions of the first two classes, Hume does not seem to doubt our capacity for knowledge. He does not doubt that we can know _some_ (though, of course, not _all)_ propositions about "relations of ideas" to be true; he never doubts, for instance, that we can know that twice two are four. And he generally assumes also that each of us can know the truth of _all_ propositions which merely assert some matter of fact which we ourselves are, in the strictest sense, directly observing, or which we have so observed and now remember. He does, indeed, in one place, suggest a doubt whether our memory is _ever_ to be implicitly trusted, but he generally assumes that it _always_ can. It is with regard to propositions of the third class that he is chiefly anxious to determine which of them (if any) we can know to be true and which not. In what cases can any man know any matter of fact which he himself has not directly observed? It is Hume's views on this question which form, I think, the main interest of his philosophy. He proposes, first of all, by way of answer to it, a rule, which may, I think, be expressed as follows: No man, he says, can ever know any matter of fact, which he has not himself observed, unless he can know that it is connected by "the relation of cause and effect," with some fact which he _has_ observed. And no man can ever know that any two facts are connected by this relation, except by the help of his own past _experience._ In other words, if I am to know any fact, A, which I have not myself observed, my past experience must give me some foundation for the belief that A is causally connected with some fact, B, which I have observed. And the only kind of past experience which can give me any foundation for such a belief is, Hume seems to say, as follows: I must, he says, have found _facts like_ A "constantly conjoined" in the past with _facts like_ B. This is what he _says;_ but we must not, I think, press his words too strictly. I may, for instance, know that A is _probably_ a fact, even where the conjunction of facts like it with facts like B has not been quite constant. Or instead of observing facts like A conjoined with facts like B, I may have observed a whole series of conjunctions--for instance, between A and C, C and D, D and E, and E and B; and such a series, however long, will do quite as well to establish a causal connection between A and B, as if I had directly observed conjunctions between A and B themselves. Such modifications as this, Hume would, I think, certainly allow. But, allowing for them, his principle is, I think, quite clear. I can, he