As Katz says (p. 137), there is a need to trim the space of possible senses--that is, if we’re talking about a real or ideal psychological account of hearers’ language understanding (as he must here be assuming).  Almost every English sentence is ambiguous, and most sentences are many-ways ambiguous due to the kind of paronymy Katz illustrates with his “gold” example on the following page.  All a hearer hears is a sentence uttered in a context.  How does the hearer pick out a single contextually appropriate sense from among the uttered sentence’s many possible senses, given contextual features?

    A deep, excellent and terrible question.  Its answer is simple: No one has the faintest idea.  But, belying that answer, Katz suggests that selection restrictions play a significant role in paring down the space of possible senses.  Selection restrictions are needed for a second, more specific reason also:  “…some sentences have no sense, even though their individual words are meaningful,” and selection restrictions are needed to predict those cases in particular.

    They do that by ruling out some otherwise licit combinations of lexical senses as failing to amount to a collective derived sense, as being (corporately) “meaningless….”  Even such subsentential combinations as “metal truth,” “yellow thought” and “yellow virtue” are “senseless” (p. 138).  These must be ruled out by any adequate semantic theory, through selection restrictions: “metal” and “yellow” each require (Physical) and (Object), while “truth,” “thought” and “virtue” rule out (Physical) if not (Object).  That ruling out is often expressed using a minus sign within the selection restriction brackets: <–(Physical)>.

    I have never met or heard of a linguist who didn’t agree with Katz about all that.  Previously, Chomsky had given the famous example, “Colorless green ideas sleep furiously,” allegedly meaningless/senseless despite being made of perfectly good words.  Linguists call the phenomenon “sortal incorrectness.”  It is cognate with (though not quite the same thing as) what Ryle called a category mistake.  The idea they share is that sometimes it is not just false that X is F; sometimes X is not even the right kind of thing to be called “F.”  Virtue is not the right kind or category of thing to be called “yellow,” or “metal” for that matter.  Neither is the number 7.53.  Mist can be called “gold” in the sense of the color, but not in that of the metal.  Such pseudo-predications are not “not just” false; they are not false at all, but meaningless and without sense.  To utter one is to evince conceptual confusion, not just—not even—erroneous belief.

    As I may have hinted here and there, I think this whole idea is misguided.  (The late W.V. Quine firmly agreed with me.)  Well, perhaps not the whole idea, since it does sound right to say that virtue is not the right kind of thing to be called “yellow,” and that only physical objects can rightly be called yellow.  But “kind” there should not be italicized.  I maintain that there is no fixed difference, but only one of degree, between a “kind” or “category” and an ordinary group or collection of things.  A right-wing loony is not the right kind of person to be U.S. Attorney General, and a manual typewriter is not the best kind of thing to use for writing handouts.  I also reject the alleged distinction between “conceptual confusion” and erroneous belief; that too is a matter of smooth degree.

    I concede a modal difference as well:  My jacket and my tennis balls are not yellow, but they could have been.  Virtue, truth and thought could not have been yellow, or metal, because they are not physical objects to begin with.

    But why do those admitted facts amount to the dramatic charge of meaninglessness?  On the contrary: If you didn’t know what the “sortally” incorrect predications meant, you couldn’t know that they can’t be true—from which it follows that they mean something, not nothing.  I quote myself (from Logical Form in Natural Language (Bradford/MIT, 1984, pp. 100-01), not distinguishing between sortal incorrectness and category mistakes):

     Ryle’s original doctrine of “sortal incorrectness” or “category mistake” (1938, 1949) was that offending instances of (56a), (57a), and (58a), such as

 (59)  Well-played golf is red,

 (60)  Quadruplicity drinks procrastination,

and

 (61)  pi is tuned to G#

are meaningless.  If they are meaningless, then plainly their “denials” are too; “It is false that three spadlaps sat on a bazzafrazz” is no more meaningful than “Three spadlaps sat on a bazzafrazz.” …

     Of course, the doctrine of the literal meaninglessness of category mistakes is highly suspect (I doubt that even Ryle himself would have assimilated category mistakes strictly to gibberish such as “Three spadlaps sat on a bazzafrazz” or “Umph the but g kreesplat blunk”).  For one thing, none of (59)-(61) contains either any non-word or any illicit surface-grammatical concatenation.  If any of them is marked “ungrammatical” by any theorist’s syntax, it is only because the theorist has stipulated by means of a “selectional restriction” or a “meaning postulate” that category mistakes like these are not to be counted as well-formed.  I think the doctrine is also impugned by the same sort of argument that worked plausibly against verificationism:  We have to know what (59)-(61) say in order to tell that they cannot be true.  For example, we have to know that (60) ascribes to quadruplicity (an abstract numerical attribute of all manner of things) the property of drinking (hence ingesting) procrastination (a habit indulged in by certain people).  Otherwise we would not be able to apply our prior knowledge that such things cannot be (that the items in question are not the kinds of things that can drink each other), in order to judge that (60) cannot be true.  If we know what (60) says, then there is something that (60) says.  If (60) says something, then (60) is not meaningless, even if there is something drastically wrong with (60).
     If they are not literally meaningless, what is wrong with (59)-(61)?  To say that they violate selectional restrictions or the like is no answer, since this is just to repeat that they are category mistakes.
 

    On p. 143 Katz backs off “meaningless,” substituting “semantically anomalous; “meaningless” “has a stronger connotation than is desired” (footnote 17).  Some linguists and philosophers. notably the logician John Martin,<1>  have assimilated sortal incorrectness to failure of “semantic presupposition”: When S1 presupposes S2 and S2 is false, S1 (though perfectly meaningful) lacks truth-value entirely rather than being false, and so does S1’s negation:  If the diaper service was never robbed at all, then neither “It was Grannie who robbed the diaper service” nor “It was not Grannie who robbed the diaper service” is true; their common presupposition having failed, each sentence just goes to zip.<2>   Martin says the same thing about sentences like (59)-(61).  What’s wrong with them is not that they are literally meaningless/senseless, but merely that they are truth-valueless owing to presupposition failure.

    Indeed, (59) does in some sense presuppose that well-played golf is colored and that it is a physical object.  But listen to Lycan:

     I am not sure what solid motivation one might find for accepting Martin’s hypothesis.  If someone simply announces that what is wrong with (59)-(61) is that they lack truth-value, and lets it go at that, he has raised more questions than the has answered: Why do these sentences lack truth-value instead of being false, given that they are not meaningless?  Why should truth-valueless (though meaningful) sentences sound any funnier than sentences that are just so absurdly false that no one could possibly believe them?  What exactly distinguishes a “sortally incorrect” sentence from an ordinary falsehood, and how might we represent the distinguishing property in our semantics without simply marking certain predicates as “restricted against” or “contraindicated for” other expressions?
     Besides these awkward questions, there are the standard difficulties for Ryle’s notion of a “category mistake” that do not turn specifically on a strong and literal interpretation of the pejorative “meaningless.”  First, what is the difference between a “category” or kind (as in “X is not the kind of thing that could ß”) and a mere group?  Is a “category” simply larger or more inclusive than an ordinary collection of things? But size and inclusiveness come in degrees; they do not lend themselves to the absolute distinction one draws when one marks some sentences as merely false but others as neuter.  Indeed, intuitions of sortal incorrectness themselves seem to come in degrees- “category mistakes” shade off rather smoothly into ordinary falsehoods (“pi weighs twenty pounds” - “Honesty weighs twenty pounds” - “Ali’s honesty weighs twenty pounds” - “Ali’s left hook weighs twenty pounds” - “The singularity in field F weighs twenty pounds” - “The electron in the chamber weighs twenty pounds” - “Fort Knox weighs twenty pounds” - “Jumbo the elephant weighs twenty pounds”).  These facts strongly suggest that the deficiency of (59)-(61), whatever exactly it may be, is itself something that comes in degrees (such as outrageousness or obviousness of falsity), rather than something absolute such as lack of truth-value.
     Martin has a more powerful motivation in mind for maintaining the truth-valuelessness of category mistakes (pp. 250, 274).  He contends that negations like (56b), (57b), and (58b) “imply” or necessitate the alleged presupposita just as (56a), (57a), and (58a) do.  If this is correct, then the truth-valuelessness of (59)-(61) is not a dubious explanatory claim that enlightens less than it mystifies, but an apodeictic consequence of facts of logical necessitation.
     My objection to this argument is that Martin’s claims of necessitation seem plainly false.  I suppose the negations of (59), (60), and (61) respectively suggest (62), (63), and (64) via the Maxim of Strength:

 (62)  Well-played golf is colored.

 (63)  Quadruplicity is animate and procrastination is a liquid.

 (64)  pi is capable of producing a musical tone.

But these suggestions are easily cancellable:

 (65)  Well-played golf is not red, or any other color.

 (66)  It’s false that quadruplicity drinks procrastination; quadruplicity isn’t even animate, and procrastination is a habit people have, not a liquid.

 (67)  pi is not tuned to G# or any other pitch.  It’s not even capable of producing a musical tone--it’s a number, for God’s sake.

Any strangeness I can hear in (65)-(67) is due entirely to their forehead-smacking obviousness.  Certainly they are not contradictions; indeed, I would call them excessively plain truths.

    But never mind semantic presupposition.  There is a deeper question to be put to Katz and other linguists about sortal incorrectness:  Why think it is a linguistic phenomenon at all?

    I conceded that “sortal” incorrectness reflects modal facts.  But, contra the Positivists, not all modal truths are linguistic truths.  I don’t see that there’s anything at all linguistically wrong with the foregoing examples; they just say things we know can’t be true, or refer to things we know can’t exist.  Linguistically, (60) is perfectly in order.  It means that quadruplicity drinks procrastination, which we know can’t happen because quadruplicity is an abstract numerical property and procrastination is not a liquid.

    If I am right, then “sortal” incorrectness is no business of the linguist’s in the first place, and there should be no selection restrictions in linguistic theory.  Not that you would ever get any linguist to believe a word of that.

    At least selection restrictions get rid of some of the gabillions of unwanted senses, Katz will say, and so they have the virtue of making a start on the hideous Disambiguation Problem with which we began.  But the start is a minuscule one at best, hardly worth the ink.  Because there is so much ambiguity even without counting the sortally incorrect readings that the Problem is still hideous, intractable and in need of a drastic and revolutionary solution.
 
 

1 “A Many-Valued Semantics for Category Mistakes,” Synthese 31 (1975).

2 For the record, I myself think this doctrine of semantic presupposition is itself perniciously false; see the puissant Chapter 4 of Logical Form in Natural Language.  (Reading the whole book wouldn’t do you any harm.)