Pragmatics at Home

Here's an interesting case of implicatures I noticed the other day when discussing what movie to watch with my wife. (>> means "pragmatically implies" and *>> means "does not pragmatically imply")

Case One:
W: Do you want X?
H: Only if you want it.
>> I don't want it.
*>> I do want it.

Case Two:
W: Do you want X?
H: Not if you don't.
>> I do want it.
*>> I don't want it.

What is strange about this case is that, presumably, the husband H's responses in both cases are logically equivalent to each either; assuming I'm parsing them right, they both say "I want X only if you want X" or, equivalently, "I do not want X if you do not want X." (***See bottom of page for an explanation, if this isn't clear.)

But the response in Case One (at least sometimes) implies something different than the response in Case Two.  (I say 'at least sometimes', because, as with many implicatures, it may depend somewhat on the sonic properties of one's utterance too -- i.e., the way one pronounces the words.) But this seems to imply that the implicature is "detachable" in Grice's sense (see the bottom of p.57 and ff., here).

However, according to classical Gricean pragmatics, conversational implicatures are non-detachable; hence, if these were conversational implicatures, they would both imply the same things (which they don't). So it seems that they must be conventional implicatures. (See here on that distinction.) That's sort of weird though, because conventional implicatures are usually associated with syncategorematic expressions that do not contribute any additional truth-conditional meaning to the sentence (for instance, "however," "but," "even though," "nevertheless," etc.).

Also, these implicatures seem to be more like conversational implicatures than conventional ones, since they do seem to sort of follow from something like Grice's Maxim of Manner (or, better, Levinson's M-Principle); i.e., saying something in an equivalent but roundabout way implies a non-standard meaning. For depending on whether one uses the double negation form or not you get a different implicature. However, it doesn't *quite* fit this rule I think, because it doesn't seem like either of the response from Case One or Case Two is more "roundabout" than the other; in other words, the responses in both cases seem to be symmetric as far as the "oddness" of their phrasing goes.

Anyway, kind of an interesting case. FYI, in the actual situation, I *did* want it, but I didn't want it if she didn't. : - )

***
To see the equivalence, note that all of the following are equivalent:

• I want it only if you want it.
• I only want it if you want it.

(These are clearly equivalent. For consider the following:
x goes to the store only if x is hungry.
x only goes to the store if x is hungry.)

• If you do not want it, I do not want it.
• I do not want it if you don't want it.

Modernizing (and Medievalizing) Analyticity

A project I've been hoping to do eventually is to re-work the concept-containment notion of analyticity. Something has always seemed to me intuitive about it, at least in the "Bachelors are unmarried" sorts of cases. (From the 19th century onward, for various reasons, the analytic truths seem to have become equated with the logical truths; this seems wrong to me.)

I'd also like to see whether the notion of analyticity comes up at all in medieval philosophy, and whether any of the tools of medieval philosophy could be of use for this project (or, what would be just as interesting, whether they would find the concept-containment notion of analyticity hopelessly confused).

There are several problems for the concept-containment approach, but a big one is this:

1. Forms of Analytic Judgments: Kant (sometimes) defined analyticity in terms of conceptual containment, roughly as: A judgment of the form 'A are B' is analytic iff the concept B is contained in the concept A.

The worry is that there are many judgments that do not seem to have this form, but where they seem to also be true in virtue of 'meaning' or 'concepts' (or something close). Some examples, taken from SEP:

(11) If Bob is married to Sue, then Sue is married to Bob.
(12) Anyone who's an ancestor of an ancestor of Bob is an ancestor of Bob.
(13) If x is bigger than y, and y is bigger than z, then x is bigger than z.
(14) If something is red, then it's colored.

Other examples, from Jerrold Katz:

(15) Mary walks with those with whom she herself walks.
(16) Mary walks with those with whom she herself strolls.
(17) Poor people have less money than rich people.
(18) Rich people have more money than poor people.

It's not totally clear how all of these examples can be analytic (assuming they all are) if we take analytic truth to mean that the predicate-concept is contained in the subject-concept.

I don't have a worked-out answer to this yet, but I suspect that some of the work in cognitive linguistics might be helpful. In fact, Katz' own work might be helpful here, though from what I understand Katz is a Platonist (rather than a "conceptualist") about meaning, and so it would have to be properly adapted.

In addition to these strategies, where medieval philosophy might be useful here is in seeing how we might, in fact, be able to reformulate all "basic" sentences and then parse them out so that they technically obey the constraint of "subject-copula-predicate" form.

As Terence Parsons points out, medieval scholastic Latin is unique in that it is a natural language and yet there is no distinction between a sentence's ordinary surface grammar and its logical form.

Now, if one had a close enough association between words and mental concepts, one might then be able to get a nice conceptualist-type semantics going. And it seems certain medieval thinkers did have precisely this sort of close association between mental concepts and meaning, viz., in the theory of "subordination" (I'm thinking of Buridan and Ockham right now).

This suggests that we could translate basic ordinary-language sentences into medieval subject-copula-predicate sentences, and the concept-containment idea could become more useful again.

I'm not sure it will be as simple as that or that this will solve everything, but I suspect it will make things easier.

There are some other issues for a conceptual containment theory of analyticity too:

2. Conceptual Containment: There are really at least two problems here: (a) a theory of concepts, and (b) a notion of containment. Can we give a plausible and clear theory of both? (And, what would be even better: Can we give a theory of both that is mathematizable and subject to rigor and computation once we are given a case?) And how neutral can we be here with respect to different theories of concepts and containment?

How can medieval philosophy help here? Medieval philosophy certainly contains much discussion of concepts, and that should certainly be useful.

As for the notion of containment, I can't help but think of Scotus' theory of "repugnance" and "non-repugnance" by which he assesses the modal status of basic propositions -- the kinds of propositions concept-containment seems to be after (see page 162, here). What's interesting is that Scotus seems to define repugnance as a relation holding between terms. That seems to make it a semantic relation rather than a conceptual relation for Scotus; moreover, the relation's holding is said to be grounded in "notae," which are in some sense objective features of the external world.

So this isn't exactly concept-containment analyticity; but still, in terms of its formal/logical properties, non-repugnance/repugnance behaves similarly to concept-containment/exclusion. (This seems to be an instance of a more "externalist," metaphysical picture of containment and exclusion relations; I get the sense that this sort of externalism is the norm in medieval philosophy, especially pre-Nominalism, though even after that as well.) Moreover, like repugnance-relations, concept-containment relations are supposed to ground the modal status of propositions, and moreover, they both seem to deal with the same sorts of propositions. So I can't help but think Scotus will be helpful here, even if he probably wouldn't have a view of analyticity like Kant's.

3. Rigor: How would a rigorous conceptual semantics go? Can we make it as formal, precise and mathematical as the non-conceptualist semantics have been? Some work has already been done on this sort of thing, but there's more to be said. Again, I wonder if the medieval logic might help us here, since Parsons and others have shown it to be entirely rigorous and fit for mathematical treatment. Medieval logic at its height was generally far more sophisticated than anything after it -- certainly more sophisticated than anything Kant did on logic -- and so I can only imagine it will make this project easier.

4. Externalism: Although the conceptual-containment approach seem plausible in certain cases, so does semantic externalism. Gillian Russell, a professor here at UNC, has done some brilliant work on updating analyticity to take these post-Kripkean insights into account. However, she tends toward the more externalist side of things, and I'd like to see whether we can salvage more of the connections between meaning, analyticity and concepts than she does, while still giving a reasonable account of the externalist insights (something I worry cognitive semantics a la Peter Gardenfors hasn't quite done -- see 4.1 here, for instance).

One thing I worry about in trying to find analyticity in medieval philosophy is that medieval philosophy seems to tend much more toward externalism about semantic content (though this is only a hunch I get -- I can't point to anything specific). But maybe I'm wrong and there is room for analyticity in medieval philosophy; and even if not, the project may still be worthwhile, since maybe we will find new reasons to either abandon or revise our conception of analyticity in interesting ways we've never thought of.

So I plan to do some research on medieval logic and semantics in the next few weeks, and maybe this will help my concept-containment project. Moreover, I think it is an interesting historical project in its own right to see whether something like analyticity can be found (or reformulated) in terms of medieval semantic categories.

Perceptual Representation: Pictures and Sounds, Seeing and Hearing

Perceptions are representations. What representations are can be cashed out in several ways.

First, they are about something.

Second, they represent the world as being a certain way.

Moreover, they have satisfication notions associated with them (and corresponding satisfication conditions, i.e., conditions that must obtain for them to be satisfied). The satisfaction notion of desires is 'being fulfilled' or not; with belief it is 'being true' or not; with volitions it is 'being done' or not; with commands it is 'being obeyed' or not; and with perceptions it is 'being accurate' or not.

Representations include words, sentences, beliefs, pictures, questions, paintings, videos, signs, diagrams, maps, hand gestures, commands, recordings, desires, and, in the case at hand, perceptions.

If a given representation has a satisfaction notion associated with it, it will also follow that it has satisfaction conditions. Giving the satisfaction conditions for a type of representation is (at least part of) giving a semantics for that type of representation.

Much work has been done on giving semantics for various sorts of linguistic representations. Whole classes of sentences have been given rigorous semantics. For instance, famously, modal language has been given a whole semantic theory, rigorously formulated in terms of mathematical models involving possible worlds.

Arguably, something similar can be done for pictures.

If a rigorous semantics can be given for pictures, this may hold some promise for giving a rigorous semantics for visual perceptions. Naively, we think of vision as giving us an "image". Think for instance of Ernst Mach's drawing of his own visual field:

Mach's picture of the visual field

With appropriate modifications for the perspective and particularities of our visual field, the naive application of pictorial semantics to visual-perceptual semantics would be to straightforwardly give the same semantics for our visual perceptions as we would for a drawing of our visual field.

This is obviously simplifying hugely and leaves a host of questions unanswered. But it allows us to get a preliminary grip on things.

What I'm interested in lately is whether, suitably modified, the semantics of auditory perception will be similar to the semantics of visual perception.

You can think of visual perception as giving you a picture, where the picture includes such properties as {OBJECT, SHAPE, MOTION, COLOR, DISTANCE, SPACE, TEXTURE, CAUSE, AGENCY}. All of these properties are represented in visual perception (and, with most of them, in pictures too, at least generally).

(Note: One of the simplifications we have to make in assuming the semantics of visual perception is just like the semantics of pictures is that, as with the semantics of pictures, visual perceptions do not have a temporal-duration aspect: A good assumption for pictures maybe, which can be thought of as "instantaneous" in some way; on the other hand, a bad assumption for visual perception, which as perceptual psychology demonstrates represents motion, and an even worse assumption for auditory perception.)

So we can think of vision as giving a picture. Can we think of hearing as giving us a picture too? If we can, then if the move from pictorial semantics to visual-perceptual semantics is relatively straightforward, then so will the move be from pictorial semantics to auditory-perceptual semantics (and maybe we can even find a better representational correlate than pictures for sound; hint: recordings maybe?).

Well, clearly, hearing doesn't immediately represent all of the same properties as vision. And neither does vision represent all of the same properties as hearing.

However, that doesn't need to stop us from thinking of hearing as presenting an auditory "picture," at least if we understand "picture" as picking out a general representational structure that is in some way common to normal, physical pictures and our mental, visual images. After all, as we said before, we can think of the visual image as a picture representing such properties as {OBJECT, SHAPE, LOCATION, MOTION, CHANGE, COLOR, DISTANCE, SPACE, TEXTURE, CAUSE, AGENCY}. But then maybe we can also think of hearing as giving us an auditory representation, structurally similar to the visual image, but instead representing such features as {OBJECT, LOCATION, MOTION, CHANGE, DISTANCE, SPACE, CAUSE, VOLUME, PITCH, TIMBRE}. We might even be able to include SHAPE and AGENCY in there in some cases (almost certainly in the case of bats), though I'd have to double-check the scientific literature.

In other words, the perceptual "structure" could remain the same between visual perception and auditory perception, and indeed, some of the qualities that are represented in vision may be represented in hearing, and vice versa. For instance, just from what I've listed, the intersection of properties represented in hearing and vision will include such features as {OBJECT, LOCATION, MOTION, CHANGE, DISTANCE, SPACE, CAUSE). On the other hand, maybe it is unique to vision to include {COLOR, TEXTURE}, and maybe it is unique to hearing to include {VOLUME, PITCH, TIMBRE}.

But again, despite the differences in features that are represented in hearing and vision, the overall representational structure might be, generically, the same at a certain level of abstraction.

What would be interesting would be to work out a precise semantics for visual perceptions and a precise semantics for auditory perceptions and see where, in the details, the two actually differ. Obviously hearing and vision will have their accuracy conditions assigned by different systems of depiction. But it's exciting to think that the two might not be so far apart, that they may, structurally speaking, having a lot in common, and that light might be shed on both by looking at conventional, physical representations (such as pictures, recordings, etc.).

Essence and Counterpossibles

In my last post I made the point that the predicate position in real definitions is hyperintensional. So, even if two predicates have the same intension, i.e. necessarily apply to all the same things, they might not be able to be substituted for each other in the real definition while preserving truth value. This means that, among the necessary properties of a thing, we have to distinguish those which are essential to the thing from those which are not.

Maybe one way to do this is by using counterfactuals with impossible antecedents, also known as counterpossibles. 

The general idea is this. Counterpossibles, according to a certain semantics, are also hyperintensional. You can insert intensionally equivalent antecedents into the same counterfactual, but only some of these counterfactuals will be true while others will be false. So maybe we can use a particular counterpossible schema (as we will see, that in (1*)) to discriminate between properties that are essential and those that are non-essential. 

More specifically, we can take two intensionally equivalent properties F and G, insert F into the antecedent, insert G into the antecedent, and the counterfactual may have a different truth value depending on which of F or G is substituted. Thus, the essential vs. non-essential distinction will be able to be defined in terms of the truth or falsity of instances of a certain counterfactual schema. To make the point more vivid, the counterfactual schema will be like a box we can insert intensionally equivalent properties such as F or G into. If we put in G for instance and the box outputs TRUE then G is essential; if it outputs FALSE then G is not essential. Since counterpossibles are hyperintensional, the 'box' won't always give the same output for properties with the same intension.

So: Consider two intensionally equivalent properties F and G. This means that, necessarily, if anything has F in a world then it also has G in the same world, and vice versa. If F and G are intensionally equivalent then the properties λz[□Fz] and λz[□Gz] are also intensionally equivalent, and thus the formulas λz[□Fz](x) and λz[□Gz](x) are intensionally equivalent. Now, consider a counterfactual of the form:

• (C) If φ had been the case then A.

Since we are going with an interpretation allowing for non-trivially true counterpossibles, we can substitute in for φ either of two necessarily false propositions, P or Q, but it won't automatically follow that (C) will come out true under both substitutions.

So suppose it is true that a is necessarily F and necessarily G.

Let P be '¬λz[□Fz](a)', let Q be '¬λz[□Gz](a)' and let A be '¬λz[∃yy=z](a)'.

Again, it doesn't follow automatically from the semantics of counterpossibles that substituting P in for φ will give you the same truth value as substituting Q for φ, despite the fact that these two formulas are intensionally equivalent. What we can do then to distinguish whether F is essential or G is essential (or neither) is substitute in P for φ and Q for φ in (C). If (C) comes out true, then the property is essential; if it comes out false, then it is not.

With this hypothesis in mind, here is a very rough first stab. For any x:

• (1) For any P, if P is a de re necessary property of x, then P is an essential property of x if and only if had x lacked P then x would not exist.

More formally, let |x| be λz[z=x]: 

• (1*) For any P: If λz[□Pz](x) then: □|x|Px iff (¬λz[Pz](x) □→¬λz[∃yy=z](x)])

Again, keep in mind that the counterfactuals here are to have a semantics where they can be read as non-trivially true counterpossibles, and thus not according to the standard Lewis-Stalnaker semantics. Also, the 'essentialist box', □F, comes from Kit Fine's 'Logic of Essence'. Roughly, □FA means that in virtue of the essence of the F's, A holds. So if |x| is the property of being identical to x, then □|x|Px turns out to mean that in virtue of the essence of x P holds of x. Fine parses out the logic this way for its semantical and logical elegance.

Consider one example given by Kit Fine: Suppose we have the property λz[z∈{z}]. Intuitively, this is the property satisfied by something whenever it is in its singleton. Assuming that, necessarily, I am a member of my singleton, then this is a de re necessary property of me, i.e. λz[□z∈{z}](a) However, it seems to not be an essential property of me.

So, is it true that the following holds?

• (A) Had Alfredo lacked λz[z∈{z}], Alfredo would not exist.
• (A*) ¬λz[z∈{z}](a) □→ ¬λz[∃yy=z](a)

It seems that (A) does not hold. For it's irrelevant to my nature whether there are any abstract objects, such as sets, at all. After all, it seems that if nominalism were true, I would still exist. This lends some weight toward thinking that, if I had lacked λz[z∈{z}], I would still exist. But then by the criterion in (1), since the counterpossible does not hold for the property λz[z∈{z}], it must follow that λz[z∈{z}] is not essential to me. 

Given a de re necessary property of x, P, it might also be that the truth of the appropriate counterpossible is just a necessary condition for a property's being essential. In other words:

• (NEC) If P is essential to x, then were x to lack P x would not exist. 
• (NEC*) If □|x|Px, then (¬λz[Pz](x) □→¬λz[∃yy=z](x)]).

Or it might be a sufficient condition as well, i.e.:

• (SUFF) Given that if x were to lack P x would not exist, then P is essential to x.
• (SUFF*) If (¬λz[Pz](x) □→¬λz[∃yy=z](x)]), then □|x|Px.

Keeping in mind of course the sense of the term 'essence' in mind, and the relevant semantics for counterpossibles, (NEC) seems definitely true, and probably uncontentious. I suppose the interesting question is whether (SUFF) is true. I think the examples lend some support to the idea, such as the case of the singleton set given above.

In the case of (SUFF) it is particularly important that we use the right semantics for counterpossibles. If (SUFF) is true then this is very useful when talking to those who don't recognize the sense of essence at stake here: If they already know how to evaluate the counterpossible in the antecedent according to a 'non-trivial' semantics, then we simply say, "Plug in the property for the antecedent; if the counterfactual holds non-trivially, the property is essential. Now you know what I mean." This might also be a nice way to interpret people who give multiple definitions of 'essential' and 'accidental' properties, such as Aristotle. Aristotle gives a modal definition of essential properties which can sound like it might contradict other definitions of his; but (1) is 'modal' too, and it might be a way to interpret Aristotle that makes him consistent.

Probably potential counter-examples to the hypothesis come to mind. I know that I already see some issues. But it might be useful to see how far this hypothesis can go. And maybe if the hypothesis doesn't hold in general (I bet it doesn't) it might at least help us pick out an important class of the essential properties. After all, it seems in part that the reason we recognize λz[z∈{z}] as non-essential to me is because in some (non-trivial) sense had I lacked it λz[z∈{z}] I would still exist. Had nominalism been true, I'd have still been real (save if you're a Platonist/Pythagorean of a certain sort, in which case maybe you'd have good, non-trivial reason to deny that the counterfactual holds).

It might seem overly complicated to do this quasi-formally as I have, but one thing I'd like to do is look more at the formal semantics of essence such as Fine's for instance (hence the essentialist 'box' operator from Fine's papers), and see how this relates to the formal semantics of counterpossibles (whatever that happens to be). Given the inter-dependence of the two notions, maybe the correct semantics for counterpossibles will help us find the correct semantics for essence, and vice versa. Maybe the notion of essence will help us give a more principled similarity metric for counterpossibles in certain contexts. Also, maybe the notions of essence and counterpossible will relate closely to other notions, such as grounding, dependence and explanation. It might be an interesting project to see how far formal methods can help us here in finding relations between these concepts.