...where [\phi]_classical is the conservative semanticist's (e.g. the Gricean's) analysis of sentence \phi, the free choice sentence isn't equivalent to anything of the `implicature-strengthened' form [\phi]_classical + p, except in the trivial case where p = the conclusion of the inference. NB that usually p = ~r, where r is some relevant stronger alternative to the proposition \phi expresses.
Examples of the pattern:
(1) A v B = [A v B]_classical + \neg(A & B)
("A & B" is a relevant alternative to "A v B" on the Horn-scale; obviously "A & B" is stronger than "A v B"_class)
(2) Three F G = [Three F G]_classical + \neg(more than three F G)
("n F G", for n greater than 3, are all relevant alternatives to "three F G" on the Horn-scale; each of these n would entail [three F G]_class, and hence are stronger relevant alternatives)
(3) Warm(a) = [Warm(a)]_classical + \neg[Hot(a)]_classical
(4) Mary likes SUE / Mary likes (only) Sue
= [Mary likes Sue]_classical + \forall x (x \neq Sue) -> neg(Mary likes x)
Possible response on behalf of (neo-)Gricean: maybe you could express the missing p = ~r if you used `only' in the specification of p?
(5?) Might(p v q) = [Might (p v q)]_classical + \neg(Might only p) ??
But that's very odd. In what sense is "Might only p" a relevant alternative to a part or whole of the original sentence? Plus, if you're allowed to use "only"s in the specification of the relevant alternatives, I could run disastrous symmetry arguments against the entire Neo-Gricean enterprise:
(6) A v B = [A v B]_classical + \neg(Only A) => A & B.
(!!!) Perhaps this is what Fox is working to avoid in his 2006 paper.
Tuesday, May 17, 2011
Monday, April 18, 2011
Defending against `contextualism'
I am discussing views on which experience presents (represents?...) certain colors. What about `contextualist' or `situation-dependent' views on which this is not true? Here is one such view from Sean Kelly:
``On my view the phenomenon of perceptual constancy shows us something crucial about the context dependence of perceptual experience. In particular, it shows us that the complete and accurate account of my perceptual experience of the color of an object must contain some reference to the lighting context in which that color is perceived. Without a reference to the context we won't have the resources necessary to explain the change in experience that occurs when the lighting context is varied. If this is right, as all perceptual psychologists agree, that this change in not a change in color (hence the name `color constancy'), then no color concept, not even a demonstrative one, could completely describe the content of a color experience...[the perceptual demonstrative] `that color' is unable to distinguish between that color as presented in the sun and the same color as presented in the shade. Because the relevant difference is not a difference in color, no color term could make such a distinction. Since such a distinction is clearly made in experience---the color looks different in the sun than in the shade---the demonstrative concept is inadequate to account for the experience." (Kelly, ``The non-conceptual content of perceptual experience," PPR May 2001, pg 607)
What exactly to make of this? Let us accept the view that one and the same color looks different in the sun than in the shade. When I carry the sample from sun to shade, my experience changes, but the represented color doesn't change. Conclusion: my experience represents both lighting and color. [Comparison: I am testing to see whether my friend can hold his breath for a minute. I watch him for a full minute while he is underwater. My experience represents that he hasn't moved, but clearly my experience has represented some kind of change: the passage of time---my experience represents both position and time.] What shall we say, then? One thing would be a temporalist-type view of content: my experience maps color-lighting ordered-pairs to truth-values. Another view would be more eternalist: my experience represents a color at a certain fixed lighting. The lighting, being a feature of context, need not be represented by me in any way; it's merely an external feature needed for the `completeness' of the truth-conditions associated with my experience. It seems like the guiding thought for deciding between these view might be: could I have had the exact same experience under different lighting? If yes, then it is fair to say that my experience only represents a function from ordered pairs of color-and-lighting to truth-values; if no, then a more temporalist view is suggested. Either way, it does seem like a simple instruction to `fix the lighting' would be ok.
...Well, I don't know. How is Kelly's argument about demonstratives supposed to work? The idea must be that a demonstrative term like ``that shade" picks out a function from colors to truth-values, while the content of experience must be a function from color-lighting pairs to truth-values. Does it then follow that ``no color concept, not even a demonstrative one, could completely describe the content of a color experience...[since the perceptual demonstrative] `that color' is unable to distinguish between that color as presented in the sun and the same color as presented in the shade"? I suppose so; what the referent of the demonstrative would contribute is a function from lighting-conditions to truth-values; but the content of perception fixes the lighting as well as the shade. Our experiences are `opinionated' w.r.t. the lighting conditions, but the content of the demonstrative ``that shade'' is not.
``On my view the phenomenon of perceptual constancy shows us something crucial about the context dependence of perceptual experience. In particular, it shows us that the complete and accurate account of my perceptual experience of the color of an object must contain some reference to the lighting context in which that color is perceived. Without a reference to the context we won't have the resources necessary to explain the change in experience that occurs when the lighting context is varied. If this is right, as all perceptual psychologists agree, that this change in not a change in color (hence the name `color constancy'), then no color concept, not even a demonstrative one, could completely describe the content of a color experience...[the perceptual demonstrative] `that color' is unable to distinguish between that color as presented in the sun and the same color as presented in the shade. Because the relevant difference is not a difference in color, no color term could make such a distinction. Since such a distinction is clearly made in experience---the color looks different in the sun than in the shade---the demonstrative concept is inadequate to account for the experience." (Kelly, ``The non-conceptual content of perceptual experience," PPR May 2001, pg 607)
What exactly to make of this? Let us accept the view that one and the same color looks different in the sun than in the shade. When I carry the sample from sun to shade, my experience changes, but the represented color doesn't change. Conclusion: my experience represents both lighting and color. [Comparison: I am testing to see whether my friend can hold his breath for a minute. I watch him for a full minute while he is underwater. My experience represents that he hasn't moved, but clearly my experience has represented some kind of change: the passage of time---my experience represents both position and time.] What shall we say, then? One thing would be a temporalist-type view of content: my experience maps color-lighting ordered-pairs to truth-values. Another view would be more eternalist: my experience represents a color at a certain fixed lighting. The lighting, being a feature of context, need not be represented by me in any way; it's merely an external feature needed for the `completeness' of the truth-conditions associated with my experience. It seems like the guiding thought for deciding between these view might be: could I have had the exact same experience under different lighting? If yes, then it is fair to say that my experience only represents a function from ordered pairs of color-and-lighting to truth-values; if no, then a more temporalist view is suggested. Either way, it does seem like a simple instruction to `fix the lighting' would be ok.
...Well, I don't know. How is Kelly's argument about demonstratives supposed to work? The idea must be that a demonstrative term like ``that shade" picks out a function from colors to truth-values, while the content of experience must be a function from color-lighting pairs to truth-values. Does it then follow that ``no color concept, not even a demonstrative one, could completely describe the content of a color experience...[since the perceptual demonstrative] `that color' is unable to distinguish between that color as presented in the sun and the same color as presented in the shade"? I suppose so; what the referent of the demonstrative would contribute is a function from lighting-conditions to truth-values; but the content of perception fixes the lighting as well as the shade. Our experiences are `opinionated' w.r.t. the lighting conditions, but the content of the demonstrative ``that shade'' is not.
Sunday, April 10, 2011
Sennet on Binding and ``Free Enrichment"
I discovered this article, ``The Binding Argument and Pragmatic Enrichment, or, why philosophers care even more than weathermen about `raining'" on Philpapers while searching for``Free Enrichment." I must regrettably say that I am still somewhat in the dark as to what ``Free Enrichment" means. But let us meditate on the article all the same.
Consider the simple fact that
(1) It's raining.
...uttered by John in Oxford at time t, is true iff it is raining *in Oxford* at t. Is there a hidden variable in (1)? I thought that the hidden variable at issue would be a variable for *times*. This would be sensical since we analyze
(2) It's always raining
as
(2') ALWAYS: it's raining
as
(2'') At : it rains at t.
i.e., we need a `t' in (1) so that we can bind it with the quantifier ``Always." [I hope this is right anyway...check email exchange with MacF...] But I was mistaken! We were actually considering (1) qua argument that there is a hidden variable l for *location* in (1), because intuitively (1), uttered in Oxford, is only true if it's raining in Oxford. Ok.
It gets weirder, though. It seems (from the expanded title and remarks at the end of the paper) that Sennet is considering the view that these considerations about (1) show that ``rains" is context-sensitive. WHOAH! That seems like weird view to motivate on (1). After all (1) has more than one word. *Which* constituent of the sentence that expresses p should we declare context-sensitive on the basis that p is true, uttered at l, iff p is true at l? This seems dangerously underdetermined. And bizarre.
Let's start again, putting context-sensitivity first, a la Cappelen and Lepore. Consider the dude who wishes to hold that ``tall" is context-sensitive, and that it can sometimes mean ``tall for a basketball player." Now I understand that there are two ways to make the claim. One is that there is an unarticulated constituent in the sentence
(3) Billy is tall
...on some occasion of use in which (3) intuitively means that he's tall for a basketball player. So the LF of (3) is really
(3') Billy is tall-for-x
where x is a variable for a comparison class. Now, when (3') is uttered unembedded, context saturates the value for x, giving us x_c, the comparison-class-of-the-context. So far, so good.
To one who protests that putting extra junk in (3) to get (3') is undermotivated, we urge that there are often hidden variables in sentences. Consider any garden-variety use of PRO:
(4) Mary wanted PRO to leave.
...Sennet will call an argument that (3') is the LF of (3), on analogy with (4), a view on which there are hidden semantic constituents. It is an example of *semantic enrichment,* not *pragmatic enrichment.* Pragmatic enrichment is a no-no, according to certain doctrines that take the semantics-pragmatics divide seriously. So we can see this argument that there is a hidden variable in LF for (1) and (3) as a CONSERVATIVE move against people who think there is pragmatic enrichment. The question seems to hinge on considerations like this: when you interpret (3) as meaning that Billy is tall for a basketball player, does your analysis go by way of settling some locations and values of some hidden variables, as it does when you disambiguate different readings of e.g. ``Every boy loves his mother"?
...It turns out that the consideration of (2) I advanced is called THE BINDING ARGUMENT, and it is controversial! The binding argument takes a sentence s and considers embedded versions of it, where the embedded s seems to be quantified into/over by the embedding quantifier. For example:
(1) It's raining.
(5) Everywhere Bianca goes, it's raining.
(2) It's always raining.
The binding argument seeks out embeddings of s that seem (semantically) ``dependent" on the embedding environment, i.e. as in (2) and (5). It then concludes that there is a hidden constituent---two actually!---available for binding in (1).
Sennet locates two assumptions on which the Binding Argument depends. The first he calls Semantic Innocence (SI). This says, roughly, that what looks like an embedded occurrence of s really is one--that embedded and unembedded occurrences have the same LF. The second is called the Binding Assumption (BA). I'm not sure what to make of it, except that it says that where binding occurs, it occurs(?): that detected semantic ``dependence" (i.e., semantic interaction) can be traced to binding.
We then consider rejections of the binding argument. I must say it is not at all clear what the broader dialectic is here: if we reject the binding argument, are we back to ``pragmatic enrichment" a la Perry for (1)? Surely not ?...
Anyway, attacks on the binding argument fall into several cases. We may deny SI and we may deny BA. Recanati, Sennet suggests, denies SI. This is just to say he holds that there might need to be invisible variables in embedded occurrences of s, but that doesn't mean that there are invisible variables in *unembedded occurrences* of s. Sennet is dismissive of this approach because a lot of dependent occurrences are dependent occurrences of individual words, not whole sentences. He invites us to consider (8):
(8) Every sports league supports the application of a tall player.
...there is no well-formed subsentential sentence in (8). But we still have the intuition that `tall' may well be context-sensitive in (8).
I feel that this point is a little inconclusive because it's completely unclear whether we're talking about the variable-sensitivity of sentence-truth to certain parameters, or the *context-sensitivity* of certain *words*!
The second way is to deny BA. How does one do this? Well, perhaps the kind of evidence we've considered isn't sufficient to show the presence of (hidden) variables. Two sorts of considerations: Partee's and Cappelen/Lepore's. Partee points out that hidden variables should be capable of felicitously being made explicit, and it's not clear this can be done:
(12?!) A man from every country hates a foreigner *to it.*
...this seems like a bad argument to me because, as has been pointed out by Ninan et al., ``John wants PRO to dance" is actually not semantically equivalent to ``John wants John to dance."
Another anti-binding consideration suggests that maybe the inference [to richer structure in the unembedded case, on considerations stemming from the embedded, bound-feeling cases] comes from the claim that ``variables have semantic values of their own." I guess this is just another way of denying that we have identical LFs in both cases: it's just that it isn't so weird to say that the variable's `gone missing' in one case and not the other if it's simply the case of a denotation-bearing lexical item that's failed to appear. Sennet quotes Stanley criticizing this move: while we can e.g. use ostension to make salient the referent of an explicit pronoun, can we do this in the case of a variable? It doesn't seem so. [I don't really understand this argument. You need ostension and salience for REFERENTIAL pronouns. You do not need them--can't have 'em, can you?--for BOUND pronouns.]
Another view would hold that pragmatic processes result in the insertion of a boundable (and bound) pronoun in the bound cases; this also allows us to keep variables out of unembedded cases. As Sennet puts it, `the dialogue gets murky here.' Is this semantic or pragmatic enrichment, or both? He also points out that if we could always posit hidden boundable (and bound) pronouns wherever we needed to to make a sentence grammatical, we should be able to hear
(15) Every woman likes house.
...as grammatical, by positing a covert ``her" (bound) between ``likes" and ``house." But we can't really do that. So it doesn't seem like that explains what's going on in ``it's always raining" or ``Wherever Bianca goes, it's raining."
One important comment to make here concerns Lewisian unselective quantifiers: while ``every'' is selective, plausibly, other quantifiers like ``it is rarely the case that..." or ``whenever..." are not. Indeed, we can see this with sentences like
(16) It is rarely the case that, wherever Bianca goes, it's raining.
The (possible) moral is that we *can* sometimes posit (type-unselective?) variables--variables for *cases*, perhaps--when a quantifier demands it, despite there being very little to be said for positing it in the unembedded case.
Anyway, the biggest objection to the binding argument is OVERGENERATION. This comes in two flavors: syntactic and semantic. For a purely syntactic one, consider the Cappelen and Lepore examples with `2+2=4':
(17) Wherever Bianca goes, 2+2=4.
(18) Whenever I step outside, 2+2=4.
etc...
(17) and (18) are not semantically dependent, in the sense that their truth-conditions presumably do not differ from the unembedded `2+2=4': no one is forced to hold that the truth-conditions of mathematical statements varies NON-VACUOUSLY with times and places. But perhaps to stuff the unembedded sentence with these hidden variables is unpalatable, even if they are truth-conditionally inert.
Sennet writes that the semantic version of the complaint of overgeneration is one about predicted, but unattested, context-sensitivity. We consider:
(19) John is anorexic, but whenever his father cooks mushrooms, he eats.
The idea is that (19) doesn't mean `...he eats THEM.'
I'm pretty bewildered by this. Why would it? That would have to be a hidden variable for a direct object--not a time, but a thing--which couldn't be quantified over by `whenever', anyway. The appropriate version of the sentence would be:
(20) John is anorexic, but *whatever* his father cooks [t], he eats [t].
...now that IS a genuinely bound reading! But...it has nothing to do with (19). This is absolutely puzzling. Perhaps the only point is that we should beware of what Sennet has loosely termed ``semantic dependence" or ``dependent readings." An intuition of dependence may be merely pragmatic, related to salience: the *suggestion*, but not the content, of (19) includes the proposition that John eats his father's mushrooms whenever his father cooks mushrooms.
We have reached the end of the article. I have to say I'm still not clear on the connection between two things: (i) context-dependence, and (ii) hidden variables. If some words, like `tall', are context-sensitive, perhaps there is a hidden variable in them saturated by context. Perhaps these variables can be bound. Perhaps, then, we can test for bound readings. But we don't have quantifiers that bind contrast classes, so...uh...maybe not. Worse, presumably people are claiming that `tall' works relatively like an INDEXICAL---MacFarlane's nonindexical contextualism being a rather exotic view (and going unmentioned by Sennet.) But if that's true, the context-sensitive items SHOULDN'T be bindable--any more than indexicals are! It's a dogma of Kaplanian semantics that there are no natural language binders or shifters of indexicals: these would be `monsters.'
To conclude: what is free enrichment? Well, minimally, it's the positing of a hidden variable at LF. Boring view: so there's really an unarticulated constituent at LF. Weird crazy view: there isn't one there...AT FIRST...but pragmatic processes `insert' it!! Uh...that's pretty weird, since presumably facts in the meat of the language-module of your brain don't depend on how I interpret your sentence by applying Gricean maxims. I take it that the real puzzle of (1)---though maybe I've just got the debate all wrong---is whether we want to take `metaphysical' considerations to put stuff in LF. I've suggested elsewhere that the way to consider that question is to consider a language where there are no temporal operators like `always:'. More broadly: whether we can freely insert things in LF so that sentences come out to have the truth-conditions we think they should, in some sense, `really' have...when are intuitions about what truth-conditions they REALLY have are obviously contaminated by the binders, etc. we have in our *own* language!
Consider the simple fact that
(1) It's raining.
...uttered by John in Oxford at time t, is true iff it is raining *in Oxford* at t. Is there a hidden variable in (1)? I thought that the hidden variable at issue would be a variable for *times*. This would be sensical since we analyze
(2) It's always raining
as
(2') ALWAYS: it's raining
as
(2'') At : it rains at t.
i.e., we need a `t' in (1) so that we can bind it with the quantifier ``Always." [I hope this is right anyway...check email exchange with MacF...] But I was mistaken! We were actually considering (1) qua argument that there is a hidden variable l for *location* in (1), because intuitively (1), uttered in Oxford, is only true if it's raining in Oxford. Ok.
It gets weirder, though. It seems (from the expanded title and remarks at the end of the paper) that Sennet is considering the view that these considerations about (1) show that ``rains" is context-sensitive. WHOAH! That seems like weird view to motivate on (1). After all (1) has more than one word. *Which* constituent of the sentence that expresses p should we declare context-sensitive on the basis that p is true, uttered at l, iff p is true at l? This seems dangerously underdetermined. And bizarre.
Let's start again, putting context-sensitivity first, a la Cappelen and Lepore. Consider the dude who wishes to hold that ``tall" is context-sensitive, and that it can sometimes mean ``tall for a basketball player." Now I understand that there are two ways to make the claim. One is that there is an unarticulated constituent in the sentence
(3) Billy is tall
...on some occasion of use in which (3) intuitively means that he's tall for a basketball player. So the LF of (3) is really
(3') Billy is tall-for-x
where x is a variable for a comparison class. Now, when (3') is uttered unembedded, context saturates the value for x, giving us x_c, the comparison-class-of-the-context. So far, so good.
To one who protests that putting extra junk in (3) to get (3') is undermotivated, we urge that there are often hidden variables in sentences. Consider any garden-variety use of PRO:
(4) Mary wanted PRO to leave.
...Sennet will call an argument that (3') is the LF of (3), on analogy with (4), a view on which there are hidden semantic constituents. It is an example of *semantic enrichment,* not *pragmatic enrichment.* Pragmatic enrichment is a no-no, according to certain doctrines that take the semantics-pragmatics divide seriously. So we can see this argument that there is a hidden variable in LF for (1) and (3) as a CONSERVATIVE move against people who think there is pragmatic enrichment. The question seems to hinge on considerations like this: when you interpret (3) as meaning that Billy is tall for a basketball player, does your analysis go by way of settling some locations and values of some hidden variables, as it does when you disambiguate different readings of e.g. ``Every boy loves his mother"?
...It turns out that the consideration of (2) I advanced is called THE BINDING ARGUMENT, and it is controversial! The binding argument takes a sentence s and considers embedded versions of it, where the embedded s seems to be quantified into/over by the embedding quantifier. For example:
(1) It's raining.
(5) Everywhere Bianca goes, it's raining.
(2) It's always raining.
The binding argument seeks out embeddings of s that seem (semantically) ``dependent" on the embedding environment, i.e. as in (2) and (5). It then concludes that there is a hidden constituent---two actually!---available for binding in (1).
Sennet locates two assumptions on which the Binding Argument depends. The first he calls Semantic Innocence (SI). This says, roughly, that what looks like an embedded occurrence of s really is one--that embedded and unembedded occurrences have the same LF. The second is called the Binding Assumption (BA). I'm not sure what to make of it, except that it says that where binding occurs, it occurs(?): that detected semantic ``dependence" (i.e., semantic interaction) can be traced to binding.
We then consider rejections of the binding argument. I must say it is not at all clear what the broader dialectic is here: if we reject the binding argument, are we back to ``pragmatic enrichment" a la Perry for (1)? Surely not ?...
Anyway, attacks on the binding argument fall into several cases. We may deny SI and we may deny BA. Recanati, Sennet suggests, denies SI. This is just to say he holds that there might need to be invisible variables in embedded occurrences of s, but that doesn't mean that there are invisible variables in *unembedded occurrences* of s. Sennet is dismissive of this approach because a lot of dependent occurrences are dependent occurrences of individual words, not whole sentences. He invites us to consider (8):
(8) Every sports league supports the application of a tall player.
...there is no well-formed subsentential sentence in (8). But we still have the intuition that `tall' may well be context-sensitive in (8).
I feel that this point is a little inconclusive because it's completely unclear whether we're talking about the variable-sensitivity of sentence-truth to certain parameters, or the *context-sensitivity* of certain *words*!
The second way is to deny BA. How does one do this? Well, perhaps the kind of evidence we've considered isn't sufficient to show the presence of (hidden) variables. Two sorts of considerations: Partee's and Cappelen/Lepore's. Partee points out that hidden variables should be capable of felicitously being made explicit, and it's not clear this can be done:
(12?!) A man from every country hates a foreigner *to it.*
...this seems like a bad argument to me because, as has been pointed out by Ninan et al., ``John wants PRO to dance" is actually not semantically equivalent to ``John wants John to dance."
Another anti-binding consideration suggests that maybe the inference [to richer structure in the unembedded case, on considerations stemming from the embedded, bound-feeling cases] comes from the claim that ``variables have semantic values of their own." I guess this is just another way of denying that we have identical LFs in both cases: it's just that it isn't so weird to say that the variable's `gone missing' in one case and not the other if it's simply the case of a denotation-bearing lexical item that's failed to appear. Sennet quotes Stanley criticizing this move: while we can e.g. use ostension to make salient the referent of an explicit pronoun, can we do this in the case of a variable? It doesn't seem so. [I don't really understand this argument. You need ostension and salience for REFERENTIAL pronouns. You do not need them--can't have 'em, can you?--for BOUND pronouns.]
Another view would hold that pragmatic processes result in the insertion of a boundable (and bound) pronoun in the bound cases; this also allows us to keep variables out of unembedded cases. As Sennet puts it, `the dialogue gets murky here.' Is this semantic or pragmatic enrichment, or both? He also points out that if we could always posit hidden boundable (and bound) pronouns wherever we needed to to make a sentence grammatical, we should be able to hear
(15) Every woman likes house.
...as grammatical, by positing a covert ``her" (bound) between ``likes" and ``house." But we can't really do that. So it doesn't seem like that explains what's going on in ``it's always raining" or ``Wherever Bianca goes, it's raining."
One important comment to make here concerns Lewisian unselective quantifiers: while ``every'' is selective, plausibly, other quantifiers like ``it is rarely the case that..." or ``whenever..." are not. Indeed, we can see this with sentences like
(16) It is rarely the case that, wherever Bianca goes, it's raining.
The (possible) moral is that we *can* sometimes posit (type-unselective?) variables--variables for *cases*, perhaps--when a quantifier demands it, despite there being very little to be said for positing it in the unembedded case.
Anyway, the biggest objection to the binding argument is OVERGENERATION. This comes in two flavors: syntactic and semantic. For a purely syntactic one, consider the Cappelen and Lepore examples with `2+2=4':
(17) Wherever Bianca goes, 2+2=4.
(18) Whenever I step outside, 2+2=4.
etc...
(17) and (18) are not semantically dependent, in the sense that their truth-conditions presumably do not differ from the unembedded `2+2=4': no one is forced to hold that the truth-conditions of mathematical statements varies NON-VACUOUSLY with times and places. But perhaps to stuff the unembedded sentence with these hidden variables is unpalatable, even if they are truth-conditionally inert.
Sennet writes that the semantic version of the complaint of overgeneration is one about predicted, but unattested, context-sensitivity. We consider:
(19) John is anorexic, but whenever his father cooks mushrooms, he eats.
The idea is that (19) doesn't mean `...he eats THEM.'
I'm pretty bewildered by this. Why would it? That would have to be a hidden variable for a direct object--not a time, but a thing--which couldn't be quantified over by `whenever', anyway. The appropriate version of the sentence would be:
(20) John is anorexic, but *whatever* his father cooks [t], he eats [t].
...now that IS a genuinely bound reading! But...it has nothing to do with (19). This is absolutely puzzling. Perhaps the only point is that we should beware of what Sennet has loosely termed ``semantic dependence" or ``dependent readings." An intuition of dependence may be merely pragmatic, related to salience: the *suggestion*, but not the content, of (19) includes the proposition that John eats his father's mushrooms whenever his father cooks mushrooms.
We have reached the end of the article. I have to say I'm still not clear on the connection between two things: (i) context-dependence, and (ii) hidden variables. If some words, like `tall', are context-sensitive, perhaps there is a hidden variable in them saturated by context. Perhaps these variables can be bound. Perhaps, then, we can test for bound readings. But we don't have quantifiers that bind contrast classes, so...uh...maybe not. Worse, presumably people are claiming that `tall' works relatively like an INDEXICAL---MacFarlane's nonindexical contextualism being a rather exotic view (and going unmentioned by Sennet.) But if that's true, the context-sensitive items SHOULDN'T be bindable--any more than indexicals are! It's a dogma of Kaplanian semantics that there are no natural language binders or shifters of indexicals: these would be `monsters.'
To conclude: what is free enrichment? Well, minimally, it's the positing of a hidden variable at LF. Boring view: so there's really an unarticulated constituent at LF. Weird crazy view: there isn't one there...AT FIRST...but pragmatic processes `insert' it!! Uh...that's pretty weird, since presumably facts in the meat of the language-module of your brain don't depend on how I interpret your sentence by applying Gricean maxims. I take it that the real puzzle of (1)---though maybe I've just got the debate all wrong---is whether we want to take `metaphysical' considerations to put stuff in LF. I've suggested elsewhere that the way to consider that question is to consider a language where there are no temporal operators like `always:'. More broadly: whether we can freely insert things in LF so that sentences come out to have the truth-conditions we think they should, in some sense, `really' have...when are intuitions about what truth-conditions they REALLY have are obviously contaminated by the binders, etc. we have in our *own* language!
Subscribe to:
Posts (Atom)
