The question can be raised whether all propositions fit properly into one of these pigeon-holes, whether the dichotomies are exhaustive (is there a third option?), and whether the distinctions involved are absolutely binary opposites with no overlap or ambiguity between them.
One type of question which can be raised is about the nature of analytic propositions. The tautological nature of analytic propositions seems clear, especially when examples are taken from mathematics or abstract physics - the types of examples Kant gives in his Kritik der Reinen Vernunft. Yet more everyday examples from natural language can seem less clear. The question can be posed, whether Kant and others have selectively chosen their linguistic examples.
Philosophers, mathematicians, and logicians tend toward idealized versions of language when they look to investigate the structure of propositions. Gareth Evans writes:
Frege was the first to formulate a systematic theory of meaning for a fragment of natural language; systematic in that it sought to provide an explanation of how the significance of complex expressions, particularly sentences, depends upon the significance of their parts. Unsurprisingly, given Frege’s larger purposes in investigating the foundations of mathematics, the fragment which concerned him was free of many of the characteristic features of natural language; in particular, indexical expressions like ‘I’, ‘now’, ‘here’, etc. However, Frege did offer suggestions as to how his apparatus could be brought to bear upon such devices.
Language interested Frege, initially and primarily, to the extent that it bore upon his investigations into logic and mathematics. Frege’s ruminations upon language tended to restrict itself to a small and overly tidy subset of linguistic phenomena.
Let us consider a more typical proposition, and the sentence which represents it: Lee Harvey Oswald is the man who shot JFK. While this proposition does not contain any of the words which are normally considered indexicals, its analysis will nonetheless be conducted with a view to the time in which such analysis in executed.
In the year 1900, neither of the men who are referents of terms in the proposition had been born - although there may have been, at that time, a man whose initials were JFK, and there might conceivably even have been a man named Lee Harvey Oswald. Yet in conducting an analysis, we know that whoever generated the proposition meant this JFK and not that JFK, and meant this Oswald, and not that Oswald. A whole host of problems arise with such natural language sentences, sentences which are quite messier than uncluttered sentences about “bodies having extension” or “7 + 5 = 12” and so forth. Questions arise about ostensiveness, about meaning, about proper nouns, about referents which do not exist, and about synonymy.
Although some philosophers have been hastily dismissive toward Frege’s symbol of assertion in his system, writing that assertion is a merely psychological phenomenon, one might yet acknowledge that Frege, however misguided his assertion operator might have been, was responded perceptively to the complexities which natural language offers.
Our proposition about JFK and Oswald will seem different if we analyze it in 1962, or in 1964. By a much later time - let’s analyze it in the year 2013 - it will be very much like what Kant calls an analytic sentence. Considered from the standpoint of the knowing subject, many knowing subjects know about Oswald only that he shot JFK, and nothing else. For such individuals, our proposition will seem almost like the quasi-mystical a = a of the post-Kantian German idealists.
Simply put: for many natural language sentences, the question of whether they are analytic or synthetic seems to be a temporal question and a psychological question. Its analysis may be different if such analysis is carried out before or after specific events. Its analysis may depend on the epistemological state of the analyzer - whether he knows certain facts.
The murkiness resulting from the analysis, or attempted analysis, of natural language sentences has caused some to doubt the framework of analytic / synthetic and a priori / a posteriori. Willard Van Orman Quine writes:
I am not satisfied that a clear general distinction has yet been drawn between analytic and synthetic. I am even more in the dark on the Kantian distinction between analytic and apriori. This much, nevertheless, I can say: If the statements of the usual higher mathematical logic are analytic, then so are such platonistic statements as ‘There are classes’, ‘There are numbers’.
In Quine’s book Word and Object, his linguistic example begins with a word which has a fully established place in language; the task of the linguist is to discover that word’s use and meaning. An alternative example might begin with the coining of a word: astronauts return from a distant planet, bringing a sample of an unknown substance. They hand it to a scientist in a laboratory, whose task is to analyze the substance. As the scientist takes notes on the sample, he has need to name it, for the mere purpose of writing his observations.
The scientist begins by naming the sample: he calls it the ‘ABCXYZ sample’ and uses this term consistently in his laboratory journals. We might ask the question: which sentences using the name ‘ABCXYZ’ are analytic, and which are synthetic?
At first, a small handful of simple sentences, mainly observations recording sense-data, will be analytic: “the ABCXYZ sample is solid at room temperature” or “the ABCXYZ sample has a density of so-and-so many grams per cubic centimeter.” As the scientist conducts his research, synthetic sentences will emerge as new and not-directly-perceptible information emerges: “the ABCXYZ substance is composed mainly of iron and silicon” or “the melting point of the ABCXYZ sample is so-and-so many degrees Fahrenheit.”
As data about ABCXYZ becomes more clear, it is printed in reference books, and eventually in textbooks. It is taught to undergraduate students and later even to high school students. Eventually, a generation of students arises, a generation which has never seen an ABCXYZ sample, or even a photograph of a sample, and doesn’t know the heroic tale of astronauts who brought it back from a distant planet. For this later generation, ABCXYZ has always been found in all reference books and textbooks. For this later generation, the proposition that ABCXYZ is composed mainly of iron and silicon may be analytic; indeed, it may be the only thing that this generation knows about ABCXYZ.
Quine discusses a similar example:
Suppose a scientist introduces a new term, for a certain substance or force. He introduces it by an act either of legislative definition or of legislative postulation. Progressing, he evolves hypotheses regarding further traits of the named substance or force. Suppose now that some such eventual hypothesis, well attested, identifies this substance or force with one named by a complex term built up of other portions of his scientific vocabulary. We all know that this new identity will figure in the ensuing developments quite on a par with the identity which first came of the act of legislative definition, if any, or on par with the law which first came of the act of legislative postulation. Revisions, in the course of further progress, can touch any of these affirmations equally.
If we accept Quine’s objections, we have at least two options: we might take a moderately Quinian approach, and say that any decision about a proposition’s analyticity is relative to time and circumstance. Along this route we might say that a proposition is a one point in time, for one analyzer, analytic, while at some other point in time, for some other analyzer, it is synthetic.
A radically Quinian - Quine himself apparently preferred the spelling ‘Quinian’ to ‘Quinean’ - position might abolish the distinction between analytic and synthetic altogether, declaring it illegitimate.
