Cite as: Brogaard, B. 2013. “It’s not what it seems.

A semantic account of
‘seems’ and seemings”. In Herman Cappelen and Stewart Shapiro, eds.
Contextualism and Relativism, special issue of Inquiry. In Press.

It’s not what it seems. A semantic account of ‘seems’ and seemings

Berit Brogaard
01/31/2013

Contents
1. ‘Seem’ and ‘Intuitive’
2. Epistemic Versus Perceptual Uses of “Seem’
3. ‘Seem’ as a Raising Verb
4. Why the Semantics of ‘Seem’ does not Lend Support to Adverbialism
5. ‘Seem’ as a Contextually Flexible Expression
6. Semantic and Logical Properties of ‘Seem’
7. Perceptual Seemings Versus Other Kinds of Mental States
8. Preliminaries: Assumptions and Definitions
9. The Perspectival Nature of Perceptual Seemings
10. Perceptual Seemings Require the Possession of Concepts
11. Conclusion

I start out by reviewing the semantics of ‘seem’. As ‘seem’ is a subject-raising

verb, ‘it seems’ can be treated as a sentential operator. I look at the semantic
and logical properties of ‘it seems’. I argue that ‘It seems’ is a hyperintensional
and contextually flexible operator. The operator distributes over conjunction but
not over disjunction, conditionals or semantic entailments. I further argue that ‘It
seems’ does not commute with negation and does not agglomerate with
conjunction. I then show that the mental states expressed by perceptual uses of
‘seem’ have nonconceptual, yet perspectival contents. In the final part of the

  1  
paper I argue that while the content of the mental states expressed by perceptual
uses of ‘seem’ are nonconceptual, having a mental state of this type requires
possessing conceptual abilities corresponding to what the mental state
represents.

1. ‘Seem’ and ‘Intuitive’

‘Seem’-reports are utterances of sentences that contain the perceptual verb ‘to
seem’. Consider:

(1)
(a) You seem to have lost weight
(b) Her skin seemed very pale
(c) She seemed more amused than shocked
(d) It seems that Hurricane Sandy is not the scariest of them all
(e) This hurricane seems worse than Hurricane Katrina
(f) The Dewey school seems to have the best educational philosophy
(g) She seemed like a fairy that, after playing its tricks for a while on
the cottage floor, would flit away with a mocking smile
(h) I always thought she seemed like a lazy pillow princess
(i) This election seems a lot like the election in 2000

One way of dividing up ‘seem’-reports is in terms of the mental state they purport
to refer to in the conversational context. ‘Seem’-reports can give off the sense
that they report perceptual seemings, memory seemings, introspective seemings
and intellectual seemings. The following are some illustrative examples:

Perceptual
(2)
(a) My chair seems to be on fire
(b) John seems to be in a bad mood
(c) Lisa seemed to really enjoy the party

  2  
 Memory-Based
(3)
(a) It seems to me that you wore a red shirt last Monday
(b) It seems that Barack Obama won the last election
(c) It seems that ‘ranarian’ means frog-like.

Introspective
(4)
(a) My headache seems to be getting worse
(b) My visual image doesn't seem to have any boundaries
(3) My thoughts seem to be in English

Intellectual
(5)
(a) The theory of superstrings seems right
(b) It would seem that the shortest distance between two points in a
Euclidean is a straight line
(c) It doesn’t seem to me that all unmarried men are bachelors

The hardest use of ‘seem’ to get a good grip on is the intellectual use. I take the
following to be marks of intellectual uses:

The intellectual Use of ‘Seem’

(i) You can determine the truth-value of ‘it seems that p’ from the
armchair on the basis of available evidence (a priori knowledge
as well as prior experience)
(ii) ‘Seem’ does not express a mental state that is exclusively
perception- introspection- or memory-based

  3  
As Herman Cappelen (2012) points out, ‘intuitive’ can be used in numerous
ways. We can say of computer programs, dances and people that they are
intuitive. It can also mean ‘pre-theoretic’, as in ‘Intuitively, the speed of light is not
a constant‘. However, I do believe the intellectual use of ‘seem’ can express
intuition in a philosophically interesting sense. ‘Seem’, in the intellectual use,
expresses intuition when it satisfies three further constraints:

Intuitively
(iii) Either p is necessary or ∼p is necessary
(iv) ‘It seems that p’ expresses a mental state that confers prima
facie justification on the belief that p
(v) ‘It seems that p’ expresses a mental state that forms immediately
upon considering p and not as a result of extensive reasoning

Only necessarily true or false propositions can be the operant clauses of

‘intuitively’. This may explain why it would be odd to say on the basis of a hunch:
‘Intuitively, it’s raining’. Empirical propositions are not the sorts of things that can
be intuitively true or false. Some illustrative examples:

(6)
(a) Intuitively, a man with 50,000 hairs on his head isn’t bald
(b) Intuitively the subject [in the Gettier case] doesn’t have knowledge
(c) Intuitively, proper names are rigid (Kripke)
(d) Intuitively, the whole world is not fundamental (Sider)
(e) Intuitively, zombies are conceivable (Chalmers)

It first glance it appears that there are some obvious counterexamples to (iii).
Consider:

(7) Intuitively, these line fragments [in the Muller-Lyer illusion] have the
same length

  4  
The operant close does not express a necessary proposition. Upon further
scrutiny, however, it becomes clear that (7) does not express intuition.
Even if I do happen to believe on the basis of how things seem that the line
fragments in the Müller-Lyer illusion have the same length, intuition does not
confer prima facie justification on that belief. If my belief is justified, its
justification derives from a perceptual state.

(v) rules out cases in which ‘seem’ expresses an intellectual use based on
reflection. When ‘seem’ expresses intuition, it is incompatible with careful
reflection on available evidence. To borrow an example from Cappelen: After
carefully looking at and reflecting on the crime scene a detective may say: “It
seems he was killed in the bedroom and then moved to the bathroom”.
(Cappelen 2012: 46). It would be odd, however, to think the detective somehow
based his judgment on intuition. In this case, the operant clauses express an
empirical proposition. However, we can modify the example to involve a
necessary proposition. Suppose you believe that you have just proven a new
theorem. If you conclude with the remark ‘So, it seems that T25 is true after all’,
your seeming state does not form immediately but is the result of extensive
reasoning. In this case, it would be equally odd to say ‘intuitively, T25 is true’.

What sort of mental state does ‘Intuitively, p’ express in the limited sense? The
most straightforward answer is that it expresses a complex mental state the
content of which consists at least partially in information retrieved from semantic
memory. To rationally intuit p is in some cases to retrieve a term from semantic
memory and apply it to an envisaged scenario.

When p is a priori, ‘It seems that p’ and ‘Intuitively, p’ can be awkward-sounding.

Even if it’s appropriate for you to say that 2 + 2 = 4 in some context, claiming ‘It
seems to me that 2 + 2 = 4’ is odd. We can explain the infelicity on Gricean
grounds. Since you could say something more succinct, viz. that 2 + 2 = 4, your

  5  
claim is violating Grice’s maxim of quantity. So, your claim implicates, mistakenly,
that you are not fully certain that 2 + 2 = 4.

Cappelen holds that ‘seem’ often is used as a hedging verb. I think he is right
about that. If you are explaining your philosophical position to a colleague who is
opinionated about the opposite view, you may politely throw in a few epistemic
‘seems’ even when you don’t find it strictly necessary. We can explain that along
Gricean lines as well. By adding ‘seems’ your interlocutor will infer that the bare
statement is too strong. As he takes you to be conversationally cooperative, your
contribution implicates that your belief is probabilistic. Your credence on the
evidence is somewhat lower than 1. This sort of hedging has a social function:
You avoid upsetting your colleague. We act similarly when submitting papers to
avoid offending referees who happen to have views strictly in opposition to our
own.

2. Epistemic Versus Perceptual Uses of ‘Seem’

Roderick Chisholm familiarly drew a distinction among three uses of “appear
words”—perceptual verbs such as ‘seem’, ‘appear’ and ‘look’—that cuts across
the aforementioned uses of ‘seems’ (1957: chap. 4). He distinguished among
epistemic, comparative and non-comparative uses of “appear words”. The
language of the report will reveal whether a use is (explicitly) comparative or non-
comparative. For example, ‘John seemed pale’ is non-comparative, whereas
‘John looked like a ghost’ is comparative. Whether a use is epistemic cannot be
read off of the language of the report. For example, if I utter the sentence ‘John
seems to be worn out’, my use of ‘seem’ could be either epistemic or non-
epistemic. If my utterance is based on John having a greyish complexion and
dark circles under his eyes, then my use is non-epistemic. Call this the
‘perceptual use’. If, on the other hand, my utterance is based on a CNN reporter
reporting that John didn’t land any film roles this year, then my use is epistemic.
Whether a use of ‘seem’ is perceptual or epistemic depends on which mental
state is expressed by the report. Here, ‘express’ is a technical word to be defined

  6  
below.

A perceptual use of ‘seem’ expresses a perceptual seeming, a seeming based

on experience. For example, ‘John seems to be stinking rich’, if used
perceptually and if spoken sincerely, expresses a mental state that is based on
the speaker’s experiences of John. ‘Experience’ here should be taken to include
bodily experience, emotional experience, introspective experience and memory
experience. So ‘perceptual’ should not be understood in the narrow sense in
which it is often used in the philosophy of perception literature.

An epistemic use of ‘seem’ expresses a cognitive state concerning what is

subjectively probable conditional on (total, total inner, total relevant, total relevant
presented so far,...) evidence.1 If we hear on the radio that there will be flooding
in our area, I might say ‘It seems we ought to evacuate.’ My states expresses a
cognitive state about what is subjectively probable, for example given my total
inner evidence.

We can say that an epistemic use of ‘seem’ is evidence bearing for the speaker if
the report correctly describes something that is subjectively probable for the
speaker. It is in this sense that ‘seems’ functions as an (generic) evidential. I
shall take it as a definitive mark of epistemic uses of ‘seem’ that when the use is
evidence bearing, the cognitive state ceases to exist in the presence of a
defeater, if the agent is rational. More on that below.

Comparative ‘seems’ reports can be either epistemic or perceptual. For example,

your past visual acquaintance with Mary may lead you to conclude that Mary
shares certain features in common with nice girls. In this case you could correctly
utter the sentence ‘Mary seems like a nice girl’. Here ‘seems’ presumably is
perceptual. If, on the other hand, you hear on the radio that Hurricane Sandy will
cause more damage than Hurricane Katrina, you may form a probabilistic belief
in the proposition that Hurricane Sandy will cause more damage than Hurricane
Katrina. You can now correctly utter the sentence ‘Hurricane Sandy seems worse
                                                                                                               
1
I owe this proposal to Hannes Leitgeb.

  7  
than Hurricane Katrina’. Here ‘seems’ presumably is epistemic.

As comparative ‘seems’ reports have a distinctly comparative structure, it is likely

that they are structurally related to more familiar comparative sentences.
Consider:

(8)

(a) Kevin is taller than every girl is.

(b) Krista dances like Hannah

(c) Chris eats like a bird

(d) Matt is as rich as his mother

There is an extensive literature on the semantics of comparative sentences. The

one I find most plausible is the degree account offered by Irene Heim (2006). On
this view, comparative sentences contain semantically vacuous ‘wh’-items in the
sentence structure. 8(a) can be read as: ‘Kevin is taller than every girl is wh’. To
a first approximation, ‘every girl is wh’ can be understood as ‘every girl x: x is this
tall’. This item scopes out of the comparative clause, and the ‘wh’-item raises to a
wide-scope position. Hence, 7(a) has the following underlying structure:

[wh1[every girl is t1]]2 [Kevin is taller than t2]

The truth-condition for 7(a) can be articulated as follows: for every girl x, there is
a height y such that x is y tall and Kevin's height is greater than y.

Likewise, 7(b) can be cashed out as: ‘Krista dances like Hannah does wh’, where
‘Krista does wh’ is to be understood as: ‘Krista dances this way’. This item
scopes out of the comparative clause and takes wide scope. So, 7(b) is of the
form: [wh1[Krista dances t1]]2 [Hannah dances t2]. We can assign the following
truth-condition to 7(b): for some way w such that w is a way that Krista dances,

  8  
Hannah dances that way too.

If we suppose that putatively comparative ‘seems’-reports are truly comparative,

which their grammar suggests, then it makes sense to apply Heim’s analysis of
comparative sentences to them too. The natural way to do so is as follows. Take
‘X seems worse than Y’. This structure contains the implicit wh-clause ‘wh1[Y
seems t1 bad]’, which takes wide scope. So we get: [wh1[Y seems t1 bad]]2 [X
seems worse than t2]. For example, ‘Hurricane Sandy seems worse than
Hurricane Katrina’ contains the implicit clause wh-clause: wh1[Hurricane Katrina
seems t1]. This item takes wide scope, so the sentence is of the form:
[wh1[Hurricane Katrina seems t1]]2 [Hurricane sandy seems worse t2]. We can
assign the following truth-condition to this structure: there is an x such that x is
how bad Katrina seems, and this hurricane seems worse than x.

As this analysis of comparative ‘seem’-reports makes unreduced appeal to the

notion of ‘seem x’, it is not meant to provide a complete answer to the question of
how to assign truth-conditions to the underlying linguistic forms. Because ‘seem
x’ can be perceptual or epistemic, the truth-conditions for comparative ‘seems’-
reports are parasitic on the truth-conditions for non-comparative perceptual and
epistemic ‘seems’-reports.

The question here arises how to categorize memory-derived, introspective and

intellectual uses of ‘seem’. There is good reason to think that they could fall
under either the perceptual or the epistemic uses, depending on the
conversational context. Consider a dialogue between a counselor and her client.

Counselor: When you recall that episode from your childhood, what are
you seeing?

Client: I am seeing a little girl walk down the stairs. She seems sad.

Despite the introspective or memory-derived nature of the report, ‘she seems

sad’ here appears to be used perceptually. If, on the other hand, you tell me
‘Your colleagues seem to think you are a really strict teacher’, then the use of

  9  
‘seem’ probably is both memory-derived and epistemic.

Even intellectual uses of ‘seem’ can be used perceptually. Suppose a rational

agent utters the following Moore sentences:

(A) Your view is correct but it doesn’t seem correct to me

(B) Newton’s theory of gravity is incorrect, but it seems right to me

Unlike analogous Moore paradoxes with belief and knowledge both instances are
perfectly fine. But the propositions expressed by the initial conjuncts are
defeaters of the operant clause of the second conjunct. So if the uses were
epistemic, we should expect (A) and (B) to be awkward-sounding. The fact that
they are not awkward-sounding suggests that the uses here are perceptual,
despite being intellectual. Compare:

(C) It seems like we ought to evacuate but we ought not to evacuate

(D) The Dewey school seems to have the best educational philosophy but
it really doesn’t

If I utter (C) immediately after hearing on NPR that there will be flooding in my
area, then my utterance would be awkward-sounding. Likewise, (D) would be
odd if all the evidence I have supports the proposition that the Dewey school has
the best educational philosophy. This is because my uses of ‘seem’ in these
contexts are epistemic and all my evidence supports the embedded clause in the
first conjunct. There are contexts that would render these sentences felicitous.
For example, I might have evidence that supports the embedded clause in the
first conjunct but overruling evidence that supports the second conjunct.

3. ‘Seem’ as a Raising Verb

Verbs such as ‘to seem’, ‘to appear’, ‘to feel’, ‘to prove’, ‘to turn out’ function as

  10  
subject raising verbs. I will use ‘raising verbs’ as shorthand for ‘subject raising
verbs’. Raising verbs, like linking verbs (e.g., ‘to grow’ as in ‘to grow stronger’),
join the sentence subject with an adjectival or infinitive complement, as in:

(9)

(a) Lisa seemed angry

(b) John turned out to be a crook

(c) Publishing in the top journals proved to be difficult

(d) Paul’s students were expected to turn in their papers on time

Some raising verbs also function as transitive verbs, as in ‘John looked (shy,
shyly) at Mary’, ‘Tom (eagerly) expected the car crash’ and ‘Alice
(enthusiastically) tasted the soup’. When they function as transitive verbs, they
describe acts or actions of the referent of the semantic subject. When they
function as intransitive raising verbs, they describe a passive experiential or
epistemic state of an implicitly or explicitly mentioned perceiver. For example,
‘Lisa seemed angry to Paul’ describes a passive experiential or epistemic state of
Paul, and ‘The tomato seems rotten’ describes a passive experiential or
epistemic state of the speaker.

Raising verbs are followed by adjectives or infinitive clauses rather than adverbs.
The ‘to be’ of the infinitive clauses takes an adjectival complement, not an
adverbial one, as in apparent in ‘John was found to be missing’ and ‘Susan
turned out to be guilty’. Hence, while the complements of raising verbs can be
modified by adverbs, as in ‘extremely beautiful’, they cannot themselves be
adverbs or ‘to be’ plus adverbial clauses.

On the face of it, sentences with raising verbs have the same surface grammar
as sentences containing intensional verbs, such as ‘John wants to be happy’.
However, this is not so. One of the big advances of transformational grammar
was that it offered a way to distinguish between the different underlying forms of

  11  
sentences like ‘John wants to be happy’ and ‘John seems to be happy’. The
‘want’- and the ‘seem’-sentences may be taken to have the underlying forms:

John wants [John to be happy]2

Seems[John to be happy]

The surface forms are produced by applying the transformation rules Equi-NP-
Deletion and Subject-to-Subject-Raising to the sentences, respectively. Equi-NP-
Deletion allows identical phrases, for example noun-phrases or ‘for’-phrases, to
be deleted, as in ‘It’s good for her for her to stay here’ or ‘It’s good for her for her
to stay here’ (Partee 1975). In Subject-to-Subject-Raising a subject that belongs
semantically to a subordinate clause becomes realized in the surface-grammar
as a constituent of a higher clause. In the case of ‘John seems happy’ the subject
‘John’ is the surface-grammatical subject of the raising verb ‘seems’ but it is the
semantic subject of ‘to be happy’. ‘John seems happy’ has the underlying
derivational structure:

[e seems [John to be happy]]

In the transformation of deep grammar into surface grammar ‘John’ becomes

raised to become the subject of ‘seems’.

The subjects of raising verbs like ‘seem’, ‘proven’ and ‘assumed’ thus have no
semantic relation to the raising verbs. Rather, they are associated with the
infinitive predicate or the verb of the embedded clause. For example, in ‘the
apple seems red’ the subject ‘the apple’ is associated with ‘to be red’ and in
‘John seems to prefer red wine’ the subject ‘John’ is associated with the verb
‘prefers’. The formal way to put this is that raising verbs do not assign a theta-
role to their subjects.

One test that a verb does not assign a theta-role to a subject is that we can
                                                                                                               
2
On a more recent analysis, the subject of the infinitival clause is treated as some sort of hidden
pronoun which is anaphoric to John, not a second occurrence of John. Thanks to Peter
Lasersohn here.

  12  
express the same meaning by raising different elements in the derivational
structure. 3 For example, ‘John seems to be the first author of the article’ and
‘The first author seems to be John’ have the same meaning. This shows that
‘seem’ does not assign a theta-role to ‘John’ or ‘The main author of the article’.
Rather, these semantic roles are assigned in the subordinate clause ‘John to be
the main author of the article’.

Raising is specified in the lexical entry of raising verbs. For example, ‘seem’
states in its lexical entry, among other things, that it doesn’t assign a theta-role to
its subject.

There are different theories of why raising happens. In Relational Grammar

subject-to-subject-raising is driven by the rule that in English all clauses must
ultimately have a subject, which can be either the expletive subject ‘it’ or the
raised subject (Postal 1974). In Government and Binding Theory subject-raising
it is an instance of DP movement (Chomsky 1981).

4. Why the Semantics of ‘Seem’ does not Lend Support to Adverbialism

One might object to an account of ‘seem’ as a raising verb on the grounds that
perceptual ‘seem’-reports involve manner adverbials, that is, adverbials that
occur in final position and describe the manner of the activity picked out by the
verb. Consider:

(9)

(a) John spilled the beans clumsily

(b) John dances clumsily

                                                                                                               
3
The argument is somewhat problematic. Many speakers do not have the intuition that John
appears to have passed the exam is equivalent to The exam appears to have been passed by
John, for example (since the conclusions one would draw based on John’s appearance might be
very different from the conclusions one would draw based on the exam’s appearance). However,
the argument is less problematic when the seeming involves identity. Thanks to Peter Lasersohn
here.

  13  
9(a) means ‘John spilled the beans in a clumsy manner’, and 9(b) means ‘John
dances in a clumsy manner’. Adverbs that describe the manner of the activity
picked out by the verb are also known as ‘manner adverbials’. 9(a)-(b) can be
assigned the following truth-conditions using Davidsonian event semantics:

∃e[spill(e, John, beans) & clumsily(e)]

∃e[dance(e, John) & clumsily(e)]

If perceptual ‘seem’-reports involve manner adverbials, the truth-condition for ‘X

seems x’ involves ‘seems x-ly,’ where ‘x-ly’ is a manner adverbial. Consider:

The tomato seems red to me

On the adverbial reading, the truth-condition comes out as follows:

∃e[seem(e, tomato, me) & redly(e)]

In other words, there is a seeming event with the tomato and the speaker as
participants (agent and patient, respectively), and the event takes place in a redly
manner. This would be consistent with an adverbial theory of perception,
according to which perceiving is an object acting upon a perceiver in a certain
manner. Roderick Chisholm is a legendary defender of the adverbial theory.
Wylie Breckenridge is one of its contemporary defenders.

However, the semantics of ‘seem’ does not support this view. Only verbs that
assign structural case are true accusative (or agentive) verbs. For example, ‘kick’
as it occurs in ‘John kicked her’ assigns structural case to ‘her’ and hence is a
true accusative. Verbs that assign inherent case are not true accusatives. For
example, ‘believe’ as it occurs in ‘John believed in her’ assigns inherent case to
‘her’ and hence is not a true accusative. The same goes for ‘good’ in ‘It’s good for

  14  
her [to be happy]’, where ‘good’ presumably is used with a dative.4

Intransitive verbs do not allow manner adverbials and do not select affected
objects. Consider:

(11)

(a) John dried the dishes enthusiastically

(b) *John resembled her carefully

(c) *John aggravated me revoltingly

(d) *Mary desired a raise slowly

(e) Mary walked to school quickly

Sentences 11(b)-(c) are clearly infelicitous. ‘John resembled her closely’ and
‘Mary desired a raise immensely’, of course, are felicitous. But ‘closely’ and
‘immensely’ are not manner adverbials. They do not specify a way of resembling
or desiring but a quantity.

Unlike true agentive verbs such as ‘kick’ and ‘eat’, ‘seem’ does not assign
structural case. ‘Methinks’ is a modern construction of an old English dative form:
‘It seems me’ [‘thyncan’ = ‘seem’]. ‘Me’ in ‘seems to me’ is thus a morphologically
marked dative case rather than a true accusative. So, ‘seem’ does not allow an
adverbial interpretation of the adjectival phrase in ‘seem [adjectival phrase]’.5

5. ‘Seem’ as a Contextually Flexible Expression

                                                                                                               
4
I am using the phrase ‘dative’ loosely here. It is pretty controversial whether modern English has
a dative case.
5
Seem does permit manner adverbials in different types of position. Take a sentence like John
suddenly seemed a lot less attractive. Here, suddenly is a manner adverbial, and belongs to
seem, not to less attractive. (That is, the sentence is paraphrasable as It suddenly seemed that
John was a lot less attractive, rather than as It seemed that John was suddenly a lot less
attractive.) Thanks to Peter Lasersohn here.

  15  
‘Seem’-reports express particular kinds of mental states. ‘Express’, as I shall use
the word here, is a term of art. Let’s say that a report attributing a seeming to S
expresses mental state m iff if the report were true, then S would be in m. For
example, my utterance of the sentence ‘the neighbors seem to be arguing’
expresses a seeming with the content [the neighbors are arguing]. Even if I am
insincere when making my utterance, my utterance still expresses a seeming
with the content [the neighbors are arguing] in this sense of ‘expresses’. I will say
more about the content of seemings in subsequent sections.

Which type of mental state is expressed by ‘seem’ depends on what the report
communicates about the grounds of the mental state in the conversational
context. If I utter the sentence ‘My current thoughts seem to be in English’, my
report likely communicates that the mental state expressed is based on
introspection. So my utterance likely expresses an introspective seeming.

Whether my use of ‘seem’ is epistemic or perceptual depends on what is

communicated about the would-be behavior of the expressed cognitive state in
the presence of a defeater. If I say ‘The lines in the Müller-Lyer illusion seem to
have different lengths, even though I know that they have the exact same length’,
then my report communicates that the mental state expressed by ‘seem’ would
not go away in the presence of a defeater. So, the use of ‘seem’ is perceptual. If,
on the other hand, I say ‘I am not totally sure but it seems to me that John is
going to get tenure’, then my report communicates that the mental state
expressed by ‘seem’ would go away in the presence of a defeater. So the use of
‘seem’ is epistemic.

Whether ‘seem’-reports are true or false turns on whether the mental state is
correctly ascribed to the subject in question. If it seems to me that your shirt is
green, then ‘It seems to me that your shirt is green’ is true. In some cases, the
truth-value is not determinable. For example, knowing what I know about
dichromats, I might say ‘If I were a dichromat, this tomato would seem gray to

  16  
me.’ Though some dichromats do describe red things as gray, it is not clear that
we can make any conclusions about how things would seem to me if I were a
dichromat.

The question here arises whether ‘seem’-reports have a contextual semantics.

Bare uses of ‘seem’, of course, are context-sensitive. For example, ‘It seems that
John has arrived’ can express the proposition that it seems to me that John has
arrived when I utter it, and it can express the proposition that it seems to you that
John has arrived when you utter it.

It is tempting to think that ‘seem’ is a flexible contextual expression, like ‘local’

and ‘nearby’. On the standard treatment of these words, they are associated with
a hidden variable at the level of logical form. Context can supply just about any
value for the variable. ‘John went to a local bar’ need not be interpreted as ‘John
went to a bar that is local to the speaker’, but may be interpreted as ‘John went to
a bar that is local to John’, ‘John went to a bar that is local to the hearer’, ‘John
went to a bar that is local to his grandmother’, and so on, depending on the
conversational context. In this case, the speaker can more or less freely fix the
value of the hidden indexical variable associated with ‘local’.

‘It seems’, however, is not flexible in this liberal way. Suppose you are on the
phone with your aunt in California. She tells you that it’s raining. After you get off
the phone, your girlfriend asks you: ‘How is the weather there?” You reply with ‘It
seems that it’s raining’. Even if the aunt did say that it seemed to her that it’s
raining, the hidden variable associated with ‘seems’ in your utterance can only
take you as a value (or you and your girlfriend), not the aunt. If you had
responded with ‘It seems that it’s raining. But it doesn’t seem to me that it’s
raining,’ your response would have been infelicitous. So, ‘seems’ is not flexible in
the same way as ‘nearby’ and ‘local’.

However, despite not being as flexible as ‘nearby and ‘local’, ‘seem’ does have
some similarities to these expressions. When I turn to you and say ‘It seems that

  17  
we need to evacuate’, the hidden variable could be occupied by the speaker or
the speaker and her audience (or group). Linguistic context can also sometimes
fix the value of the variable. Consider, for example, John  peered  into  the  room.  It  
seemed  to  be  empty.6

Furthermore, when epistemic uses of ‘seem’ are not evidence bearing for the
speaker, the variable may take a third party as a value. Suppose after stating the
usual antidote and Fink problems, I say ‘It seems, prima facie, like one cannot
analyze dispositions in terms of conditionals’. I then give a new more elaborate
theory in which the proposal is in fact to analyze dispositions in terms of
conditionals.7 Here, the report is not evidence bearing for the speaker because it
makes implicit reference to the evidence of a person who is not acquainted with
the new theory. So, ‘seem’ here is not semantically equivalent to ‘seem to me’. In
this case, ‘seem’ is equivalent to ‘seem to you’ or ‘seem to someone not familiar
with my theory’.

Here is another example.8 In the 2012 election Republicans lost in the national
race because they bet on the angry white male vote and disregarded the overall
force of minority-trend voting in key states. So it’s true to say: ‘It seems like
Republicans got it all wrong on the demographics’. The embedded claim is not
absolutely certain as a claim but it's probable (given known data after the fact).
Here ‘seem’ likely is not be equivalent to ‘seem to me’ but is probably equivalent
to ‘seem to those who are rational and have the available evidence.’ So ‘seem’ is
a special kind of flexible contextual expression.

Given that values other than the speaker can occupy the variable associated with
‘seem’, the question arises whether ‘seem’-reports have relativistic contents
(Kölbel 2002, 2003, 2007, MacFarlane 2005a, MacFarlane 2005b, Egan,
Hawthorne, Weatherson 2004 and López de Sa 2007). One piece of evidence
traditionally used to argue for relativistic contents in the case of taste predicates,

                                                                                                               
6
Thanks to Peter Lasersohn here.
7
I owe this example to Hannes Leitgeb
8
I owe this example to Alan White

  18  
knowledge and epistemic modals is the possibility of meaningful disagreement
among interlocutors about discourse containing the relevant term. For example,
the fact that you and I can disagree about whether Brian knows where his car is
parked, despite having the same knowledge about Brian’s evidence has
sometimes been taken to indicate that the truth-value of knowledge claims is
relative to a perspective, or a judge. The thought is that if the speakers commit
no factual errors, then the difference in their assignment of truth-values must
originate in the fact that the truth-values are relative to perspectives.

Another piece of information that has been used to argue for a relativistic
framework for certain word groups is the felicity of retraction claims for relative
terms. For example, John might say ‘I know where my car is parked.’ When told
(perhaps wrongly) that car theft is prevalent in the area, he might retract his
previous claim: ‘I guess I don’t know where my car is parked after all. I was
wrong in thinking that I did’. In the case of knowledge, the retraction claim sounds
felicitous. This has been taken to indicate that the assignment of truth-value even
to one’s own earlier utterances is relative to one’s perspective at the time.

There has been a lot of criticism of these tests for assessing whether a certain
word class is relativistic or not (see e.g., Cappelen and Hawthorne 2009). Even if
we grant the effectiveness of the tests, however, ‘seem’ does not appear to be
relativistic verb.

Suppose we both hear on the radio that there will be a hurricane in our area. ‘It
seems that our home will be flooded’, I say. You reply that it doesn’t seem that
way. If we are equally rational, one of us has evidence not available to the other,
for example, evidence that the radio station is notoriously unreliable. In this case,
then, we disagree about facts about the situation.

Retraction data do not support a relativistic semantics for ‘seems’-reports either.

Consider the following dialogue:

John: It seems that we need to evacuate.

  19  
 Radio host: The earlier announcement about flooding was a hoax.

John: It seems that we don’t need to evacuate after all.

Mary: But earlier you said that it seemed that we ought to evacuate.

John: It seemed that way to me then. I didn’t know that the announcement
was a hoax

Here John is not willing to retract this earlier claim that it seemed to him that they
ought to evacuate. What he admits is that he didn’t have all the information that
was needed for the complement clause turning out true. It would seem, then, that
‘seem’-reports do not have a relativistic semantics.

6. Semantic and Logical Properties of ‘Seem’

As ‘seem’ is a raising verb, it can be treated as a kind of sentential operator when

the sentence occurs in its un-raised form. For example, ‘it seems’ as it occurs in
‘it seems that 2 + 2 = 4’ can be treated as a sentential operator on the operant
clause ‘that 2 + 2 = 4’. Like ‘According to the Sherlock Holmes stories’ and ‘I
believe’, ‘seem’ is hyperintensional. An operator is hyperintensional just in case
replacing a referring expression in the complement clause with a logically
equivalent expression can affect the truth-value of the sentence in which the
operator occurs. Consider, for example:

(12)

(a) It seems to John that 2 + 2 = 5

(b) It seems to Lois Lane that Superman is not Clark Kent

(c) It seems to the person viewing the waterfall illusion that the waterfall is
moving and not moving

  20  
 (d) It seems to Mary that water is xyz

Substituting ‘4’ for ‘2 + 2’, ‘Superman’ for ‘Clark Kent’, ‘2 + 2 = 5’ for ‘the moving
waterfall is moving and not moving’ and ‘H20’ for ‘water’ could affect the truth-
value of the sentences in (12). ‘It seems’ thus satisfies the criterion for
hyperintensionality.

Logically, ‘it seems’ behaves in some respects like well-known quantifiers and
sentential operators. Sentential operators such as the knowledge operator and
the universal quantifier distribute over conjunction but don’t distribute over
disjunction. ‘It seems’ is similar in this respect. We can infer '(it seems that p) & (it
seems that q)' from ‘it seems that (p & q)’ but we cannot infer '(it seems that p) or
(it seems that q)' from ‘It seems that (p or q)'. It seems to me that it's raining or it
isn't. But it doesn't follow that it seems to me that it is raining, or it seems to me
that it isn't.

Like the knowledge operator, ‘it seems’ does not commute with negation. Your
shirt may not seem like anything to me, in which case it’s not the case that it
seems to me that your shirt is blue. So from ‘It’s not the case that it seems to that
your shirt is blue’ it doesn’t follow that it seems to me that your shirt isn’t blue.

There is reason to think that 'it seems' does not agglomerate with conjunction
(closure fails). Let p be: Ticket 1 will win. And let q be: Ticket 2 will win. There is
0.8 probability that 1 will win and 0.8 probability that 2 will win. It might seem to
me that ticket 1 will win, and it might seem to me that ticket 2 will win. But it may
not seem to me that (both ticket 1 & ticket 2 will win). The chance of both tickets
winning is 0.64. The relatively low probability may not be high enough for it to
seem to me that both will win. Alternatively, it may not seem that way to me
because I know that only one ticket can win in this particular lottery.

Here is another example due to Mark Lance. It really does seem to Mark that the
Axiom of Choice is true, and it seems to him that Well Ordering is false. But it

  21  
doesn't seem to him that AC&∼WO because he realizes that AC and WO are
equivalent.

It’s a bit more difficult to determine whether ‘it seems’ distributes over indicative
conditionals. It seems to X that if your argument is correct, then it’s raining. Does
it follow that if it seems to X that your argument is correct, then it seems to X that
it’s raining? In other words, is the following inference valid?

It seems to X that (if your argument is correct, then it’s raining)

So, if it seems to X that your argument is correct, then it seems to X that
it’s raining

I am inclined to think that it doesn’t. Of course, we have to be careful not to

equivocate on ‘seems’. We cannot read some of the occurrences of ‘seems’ as
perceptual and others as epistemic. But even on an unequivocal reading, the
premise could be true, while the conclusion is false. The premise might be true in
virtue of X’s belief that your argument is incorrect. As she knows that a
conditional with a false antecedent is true, it seems to her that the conditional is
true. Even if X believes that your argument is incorrect, it can still seem to her
that it’s correct. So the antecedent of the conclusion could be true. But it need
not seem to X that it’s raining (or that it’s not). So, the antecedent of the
conclusion could be false

Presumably ‘seems’ does not distribute over strict conditionals or semantic

entailments either. Consider:

It seems to me that necessarily (if water exists, then there are H20
molecules)

So, necessarily (if it seems to me that water exists, then it seems to me

that there are H20 molecules)

  22  
Assume that the premise is true on the grounds that in the actual world I have a
probabilistic belief in the proposition that necessarily, water is H20. In a world in
which the clear, potable liquid that fills oceans, rivers and lakes consists of XYZ
and not H20, I probably won’t have a probabilistic belief in the proposition that
necessarily, water is H20 or the proposition that there are H20 molecules.
However, it may seem to me that water exists.

The failure of distribution of ‘seem’ over generalized conditionals seems involved

in the phenomenon of paradox. A paradox is an argument for which each of the
premises seems true, and which seems valid, but in which the conclusion not
only fails to seem true but positively seems false. Consider an instance of the
sorites paradox.

A man with 50,000 hairs on his head is not bald

If a man with n hairs on his head is not bald, then a man with n-1 hairs on
his head is not bald

So, a man with 0 hairs on his head is not bald

The argument seems valid and the premises seem true but the conclusion
seems false. Since the argument is valid, this is a good example of a case in
which ‘seems’ fails to distribute over semantic entailment.

The case also adds further support to the view that ‘it seems’ does not
agglomerate over conjunction.9 The premises seem true to me, the argument
seems valid but the conclusion seems false. It does not follow from this that it
seems to me that (the argument is valid, the premises are true and the
conclusion false). I know that a valid argument with true premises cannot have a
false conclusion.

                                                                                                               
9
I owe this observation to Mark Lance.

  23  
The preface paradox is another paradox that can illustrate the failure of ‘it seems’
to agglomerate over conjunction. Individually the sentences in a preface seem
true but taken together they do not seem true.

7. Perceptual Seemings Versus Other Kinds of Mental States

I turn now to the mental states expressed by ‘seem’-reports. Chisholm talked

about uses of “appear words”. He didn’t say what makes a seeming epistemic as
opposed to perceptual. However, as hinted at above, I suggest that a definitive
mark of epistemic seemings is that they go away in the presence of a rebutting
defeater if the agent is rational. According to John Pollock’s (1984) original
distinction, an undercutting defeater is a reason for supposing that one’s ground
for believing p is not sufficiently indicative of the truth of the belief that p. If the
dean of the college informs you that they have just implemented rain machines
on campus, then that might undercut your reasons for thinking that it’s raining. A
rebutting defeater, on the other hand, is a reason for holding the negation of p or
for holding some proposition, q, incompatible with p. For example, you believe
that it’s raining but you hear on CNN that it’s not raining. So you update your
belief. In this case, the information from CNN is a rebutting defeater of your
original belief.

Suppose again that you hear on the radio that a public health care reform has
been accepted. You say ‘it would seem that I won’t need my private health
insurance anymore’. A minute later the radio host comes back on and announces
that the previous statement was a hoax. This is a rebutting defeater of your
probabilistic belief that you won’t need your private health insurance. So, if you
are rational it will no longer seem to you that you won’t need your private health
insurance.

  24  
Epistemic seemings can be captured fairly well in terms of probabilistic belief.
Perceptual seemings, however, do not reduce to other more commonly
discussed mental states, such as belief or perceptual experience. Perceptual
seemings, I will argue, are derived from perceptual experience together with
background beliefs. If you wish, you can think of them as high-level perceptual
experiences, which are more conceptual than perceptual in nature. Though
perceptual seemings are derived from perceptual experience, they are not
beliefs. To see this, consider the Müller-Lyer Illusion:

I believe that the horizontal lines in the Müller-Lyer Illusion have the same length
but they perceptually seem to have different lengths. Another case in which
beliefs and seemings come apart is the McGurk Effect. The McGurk effect is an
illusion that occurs when one sound is played while a person is shown uttering a
different sound. For example, the speaker is mouthing the word /ba-ba/ while the
audio is dubbed as /ga-ga/. What you end up hearing is /da-da/. The illusion
occurs because the brain is attempting to bind visual information with conflicting
auditory information and has to make a guess as to what the true sound is. In this
case it perceptually seems that the speaker is mouthing the word /da-da/, even
though you know that isn’t the case.

There is also good reason to think that perceptual seemings and perceptual
experiences are distinct types of mental states. Cases of change blindness and

  25  
intentional blindness give good evidence that we must distinguish between
perceptual experiences and perceptual seemings. In change blindness cases,
you consciously see a change (e.g., a change in clothing) but it doesn’t seem to
you that a change has taken place. You would be unable to report it. Inattentional
blindness cases are similar. When you are busy paying attention to a certain
task, you don’t pay much attention to what goes on elsewhere in your visual field.
So things can occur in the off-zones without it seeming that way to you.

One could deny that we can have perceptual experiences of things we don’t pay
attention to. If consciousness requires attention, as Jesse Prinz (forthcoming)
has recently argued, then we do not perceptually experience changes or items
we don’t pay attention to.10 The cases of change blindness and inattentional
blindness then do not lend support to the view that perceptual experiences and
perceptual seemings are distinct types of mental states.

However, there are other reasons to think perceptual seemings and perceptual
experience are distinct types of mental states. As I argued in a previous section
of this paper, the sentential operator ‘it seems’ does not commute with negation.
The main reason for this presumably is that it can seem to us that something isn’t
the case, even in the perceptual case. For example, if I had a visual experience
of your shirt as red, it would seem to me that your shirt isn’t blue. But it’s doubtful
that I can have a visual experience of your shirt lacking in blueness.

Just as we don’t perceive the lack of certain properties, we don’t perceptually

experience certain high-level properties. Suppose it perceptually seems to me
that there are more than five people in the seminar room in Lucas Hall. My
perceptual experience presumably doesn’t represent this sort of “high-level”
quantificational assumption (Brogaard 2012b). At least that’s not consistent with
how the brain processes information. As different areas of the brain are activated
bottom-up, what’s represented changes.
                                                                                                               
10
For the view that conscious experience does not require attention, see e.g., Ned Block (XXXX).

  26  
Another, related, reason to think perceptual seemings and perceptual experience
are distinct types of mental states is that perceptual seemings appear to require
that we have certain beliefs. My experience that that is red, for example, can
transmute into a seeming that your car is red only if I believe that that is your car,
which suggests that perceptual seemings rest on interpretation of perceptual
experience.

8. Preliminaries: Assumptions and Definitions

Let’s turn now to the argument for the position that the contents of the mental
states expressed by perceptual uses of ‘seem’ are both perspectival and non-
conceptual. The argument assumes a weak version of representationalism,
which we may articulate as follow:

Weak Representationalism
The contents of visual seemings supervene on phenomenology

The assumption seems fairly innocent for mental states that have a strong visual
phenomenology. To account for the contents of subpersonal and dispositional
mental states we would need to invoke behavioristic measures, as these states
do not have a phenomenology. Behavioristic measures, however, are notoriously
unreliable ways to determine the content of a mental state. One reason for this is
that people can engage in pretense behavior and succeed in fooling us. For
example, Mary may act afraid even when she is not. Attributing fear to her on the
basis of her behavior would be mistaken. The behavioristic approach becomes
more plausible if we look at the neurological bases of behavior before attributing
mental states to people on the basis of their behavior. In the case of Mary, we
could measure brain activity in areas of the brain that have been associated with
fear processing, such as the ventromedial prefrontal cortex and the amygdala,
and conclude that she is not really experiencing fear. Here I shall set aside the
issue of whether there can be subpersonal and dispositional seemings.

  27  
Before turning the argument for the view that the content of perceptual seemings
is nonconceptual yet perspectival, we will also need some definitions. I will define
‘object-dependent content’ as follows.

Object-Dependent Content
To a first approximation: Object-dependent content is content that supervenes
on the individual essence of the objects denoted.

Following Saul Kripke (1980), we can take the individual essence of people to be
the zygote from which they came and the individual essence of artifacts to be the
material from which they came. Given this definition, descriptive content, ∃x(Px &
∀y(Py à x=y) & Dx), is not object-dependent. Take the content of ‘The President
is a democrat’. As indefinitely many individuals can be the President, the content
of the description ‘The President’ is not object-dependent. One example of
object-dependent content is content that has objects as constituent parts. As
Russell put it, “Mont Blanc and all its snow fields” may be a constituent of a
proposition about Mont Blanc. Another example of object-dependent content is
content that consists of relational properties with an object as a relatum, for
example, Jonathan Cohen’s color properties (e.g., red-for-S) (Cohen 2004,
2009). A third example of object-dependent content is sui generis abstract
content that is determined by its relations to external objects (Bealer 1998).

I will also need the notion of concept possession. The notion can be
characterized as follows:

Concept Possession
To a first approximation: to possess (“grasp”) a concept is to have a
reliable ability to tell that items falling under the concept are of the same
kind.

  28  
In this sense, I possess the concept of tomato but not the concept of Eucalyptus
tree. We can imagine testing subjects on whether they possess a certain concept
by showing them pairs of items, asking them questions such as “Are those
tomatoes?” If they respond reliably to cases in which the objects are of the same
kind, we can say that they possess the concept in question. In certain persistent
skeptical scenarios, it is not possible to possess concepts. For example, if all
Martini-looking and -tasting drinks, except for those you are drinking, are made of
flavored water rather than gin, you do not possess the concept of Martini. We can
safely assume, however, that the actual world does not contain persistent
skeptical scenarios.

The characterization of ‘concept possession’ just provided should not be taken to

imply that people who don’t speak English do not possess concepts. One could
use other symbolic devices or languages to test whether people possess the
concept, including written language, baby sign language, American Sign
Language and braille. The characterization does have the consequence that very
young children and most non-human animals don’t possess concepts. But that is
how it should be.

The notion of concept possession, as cashed out, may not seem to extend to
individual concepts. If, however, we allow that worldly stages can be the
referents of proper names, then the definition on offer can be extended to this
word class. When extended in this way, we can say that Lois Lane possesses
the concept of Superman, despite the fact that she doesn’t know that Superman
is Clark Kent.

Some define ‘concept possession’ as being capable of having thoughts or beliefs

involving the concept (Crane 1992, Heck 2000, Speaks 2005). I shall not here
discuss the relative virtues of these two definitions of concepts at any great
length. Suffice it to say that I find this latter characterization either too strong or
too weak. If ‘having thoughts or beliefs involving the concept’ is to be understood

  29  
as having thoughts or beliefs with a conceptual content, then it is too strong. So
understood, it presupposes a controversial view about belief content. If it is to be
understood as having thoughts or beliefs whose constituents one grasps by
possessing concepts, then it is circular and hence uninformative.

Following the literature we can draw a distinction between state and concept
conceptualism as follows (Heck 2000, Speaks 2005):

State Conceptualism
The content of a subject S’s perceptual experience R is limited by the
concepts S possesses, i.e., S possesses concepts corresponding to what
R represents (e.g., you can have a perceptual experience representing
something red only if you possess the concept of red)

Content Conceptualism
All perceptual states have conceptual contents (e.g., Fregean contents as
opposed to Russellian contents)

The difference between state and content conceptualism comes down to a

difference between whether your mental states can represent independently of
your possessing certain concepts or your mental states involve concepts as
constituents of their contents.

9. The Perspectival Nature of Perceptual Seemings

Let’s turn now to the argument for the view that the content of perceptual
seemings is perspectival. Envisage that you are looking at a beautiful garden
from a window in a tall tower. As you are looking out from your perspective in the
tower, the following image is what fills your visual field.

  30  
The first part of the argument that seemings have perspectival content is an
argument for the position that the contents of seemings are object-independent.

Perceptual Content is Not Object-Dependent

Scenario: Garden (see image)
1. My physical and functional twin and I can have perceptual seemings
with the same phenomenology.
2. If seemings have object-dependent contents, they have different
contents (e.g., “in front of me/you”).
3. Weak representationalism: The contents of seemings supervene on
phenomenology.
Conclusion: The contents of seemings are not object-dependent.

  31  
The first argument follows from absurdity in assuming dualism about the mind
upfront. Suppose for reduction that my physical and functional twin and I can
have perceptual seemings with a different phenomenology. As we are physically
and functionally identical, it follows that phenomenology does not supervene on
our physical makeup. As my twin and I are distinct individuals with distinct
individual essences or haecceities (this-nesses), the differences in
phenomenology could supervene on our individual essences or our haecceities.
If the differences supervene on our individual essences—the zygotes from which
we derived, then phenomenology could supervene on physical facts. However, it
is highly implausible that the differences in the phenomenology of our current
experiences supervene on genetic differences that once were but no longer exist.
So the differences must supervene on our haecceities. Haecceities are not
physical. So if the differences supervene on our haecceities, then
phenomenology does not supervene on physical facts. So dualism is true.

While dualism could be true for all I have said and will say here, the position is
too controversial to take for granted. As premise 1 is consistent with both
physicalism and dualism, it seems hard to challenge. Premise 2 follows from the
fact that perceptual seemings are inherently perspectival. When I look at the
garden, it seems to me that some things are in front of me or to the right of me,
whereas other things are not. But if the content of the complement clause in the
seeming is object-dependent, then the experiences of my physical and functional
twin and I have different contents. Yet our assumption that weak
representationalism is true yields that if we have the same phenomenology, we
have the same content. So premise 2 is true. It follows that the contents of the
visual seemings are not object-dependent.

Before moving onto the next step of my argument, we need the notion of a
centered world. A centered world is a world with additional, marked parameters.
Centered worlds used in semantics often are worlds with a marked time and a
marked individual. But any world that has an additionally, marked parameter is a

  32  
centered world. So a world with a marked time is a centered world. If English
contains tense operators, as I argue in my Transient Truths (2012a), then the
content of the operant clauses has a truth-value only relative to a centered world.

Centered worlds serve as circumstances of evaluation (indices). They can also

be rendered as n-tuples of parameters, as is more familiar from the work of
Richard Montague (1973) and David Kaplan (1989). Centered content is content
that is assessable for truth only relative to a centered world. It’s an interesting
question whether centered contents are propositions (functionally defined). I deal
with that question in my book Transient Truths.

Given these definitions and the previous argument, we can now show that the
contents of visual seemings are centered. The argument is quite similar to the
former.

Argument for Centered Content

Scenario: Garden from my perspective (some trees are further away than
others)
1. If the content of my perceptual seeming that that tree is further away
than that tree of the garden is non-centered (e.g., consisting of
classical properties: functions from worlds to extensions), then it is
object-dependent.
2. The content of perceptual seemings is not object-dependent.
Conclusion: The content of the experience is centered.

The argument for the first premise runs as follows. If the content of my perceptual
seeming is not centered, then it has a truth-value relative to a possible world. No
further parameters are needed to assess its truth. If this were the case, we would
be able to assess the truth of the complement of a perceptual seeming like “It
seems that that tree is further away from me than that tree,” but only by taking
the content of the complement to depend on objects, like me and you. So,

  33  
premise 1 is true: If the content is non-centered, then it is object-dependent. But I
argued above that the contents of perceptual seemings are not object-
dependent. It follows that the contents of perceptual seemings are centered.

There are two types of centered content. One type involves de se modes of
presentation that pick out the individual (or the time) in the center of the world;
the other does not. A de se mode of presentation (involving, e.g., /me/ and /now/)
is identical to the intension of a description. For example, /me/ may be identical
to the intension of the description “The person who has a perceptual seeming on
Russell Blvd, Monday November 5, 2012.”

Two Types of Centered Perceptual Content

1. With a de se mode of presentation (conceptual): Content that involves a
de se mode of presentation (e.g., that tree is further away from /me/ than
that tree)
2. Without de se mode of presentation (nonconceptual): Content that does
not involve a de se mode of presentation but is (a) a set of centered
worlds or (b) a structured entity consisting of centered properties (i.e.,
functions from centered worlds to extensions)

To decide between the two kinds of content I shall first argue that memory
seemings do not involve de se modes of presentation. The argument then can be
put as follows:

Against Modes, Memory Seemings

1. Mary, a war veteran, can have visual memory seemings with the same
phenomenology one and two days/hours after the event
2. If the seemings involve de se modes of presentation, they have different
contents
3. Weak representationalism: The content of visual seemings supervenes on
phenomenology

  34  
 Conclusion: The contents of the visual memory seemings do not involve de
se modes of presentation

Premise 1 presumably could be partially confirmed on empirical grounds. If we

make the time intervals for Mary’s flashbacks short enough, it is likely that they
involve the same phenomenology. The argument for premise 2 is a bit more
complicated. It may be proposed that the content of Mary’s flashbacks could be
cashed out as follows:

§ It was the case /yesterday/ [that smoke is all around /me yesterday/]
§ It was the case /the day before yesterday/ [that smoke is all around /me
the day before yesterday/]

But this clearly won’t work. If the seemings have different de se modes of
presentation, then they have different contents. Suppose instead that we take it
to be part of Mary’s visual memory seeming that it was the case [that the smoke
is all around /me/. This won’t do either. For /me/ is identical to the intension of a
description, for example, “The person who is recalling the memory on Russell
Blvd 11/5/12.” But it is false that it was the case that the smoke is all around the
person who is recalling the memory on Russell Blvd 11/5/12. Given weak
representationalism, then, it follows that the contents of the memory seemings do
not involve de se modes of presentation.

It is now straightforward to show that not all perceptual seemings involve modes
of presentations. Most of our perceptual seemings involve the seeming of things
that move. If you look at a room full of people, it’s hard to find a single one who is
sitting completely still. Someone is playing with her pen, someone is taking
notes, someone is reaching into his bag and someone is nodding in agreement.
But having perceptual seemings of things that move requires having memory
seemings and hence involves seemings that do not contain modes of
presentation. We can articulate the argument as follows:

  35  
 Against Modes, Perceptual Seemings
1. We have perceptual seemings of motion
2. If we have perceptual seemings of motion, we have perceptual
seemings of the past
3. Having a perceptual seeming of the past requires having a memory
seeming that is part of the perceptual seeming
4. The content of memory seemings do not involve modes of presentation
Conclusion: The contents of perceptual seemings do not (always) involve
modes of presentation

This concludes my argument for the hypothesis that some perceptual seemings
do not involve de se modes of presentation. So the contents of perceptual
seemings are nonconceptual despite being perspectival.

10. Perceptual Seemings Require the Possession of Concepts

Though the content of perceptual seemings does not involve senses or modes of
presentation, there is reason to think that you cannot have a perceptual seeming
without possessing concepts corresponding to what the seeming represents. My
argument for this view requires the familiar distinction between subjective and
objective reasons, to use Mark Schroeder’s (1998) terminology. A subjective
reason is a reason you have for believing something; an objective reason is a
reason there is for believing something. Even if I am sitting in an office with no
windows, there is a reason for me to believe that it is raining if indeed it is raining.
But if I have no information about the weather, I do not have a reason to believe
that it is raining. Given this distinction, the first part of the argument for the
position that perceptual seemings can represent something only if the agent
possesses corresponding concepts can be articulated as follows.

Subjective Reasons Require the Possession of Concepts

  36  
 1. R is a subjective reason for S’s empirical belief that p only if R is
inferentially related to p
2. R is a subjective reason for S’s empirical belief that P only if S is able to
recognize R as her reason for p
3. R can be inferentially related to p and recognized as a reason for p only
if S possesses concepts corresponding to what R represents.
Conclusion: Subjective reasons require possession of the relevant
concepts.

The first premise follows from the notions of being inferentially related and
subjective reason. I take two propositions p and q to be inferentially related for an
agent S just in case S, if asked, could derive one proposition from the other
proposition, together with her background beliefs. If I believe that it’s raining and I
also believe that the streets are wet, then my beliefs are inferentially related
given my background belief that if the streets are wet, then it’s raining. It seems
that the ability to come to hold a belief by implicitly or explicitly deriving it from a
reason is a minimal requirement on having a reason for one’s belief.

The second premise follows even more straightforwardly from the notion of
subjective reason. If S is unable to recognize that R is a reason for her belief that
P, then R is not a reason she has for her belief but merely a reason there is for
her belief.

The third premise follows from considerations of what it takes to be able to derive
one proposition from another and being able to recognize a proposition as a
reason. Both cognitive events are relatively cognitively difficult acts that involve
being able to classify items into types. Given the definition of concept
possession, it follows that these cognitive events require possessing concepts
corresponding to what the reason represents.

  37  
We are now in a position to provide the final step of the argument for the position
that concept possession is required for perceptual seemings to represent.

Perceptual Seemings Require the Possession of Concepts

1. The content of S’s perceptual seeming can be a subjective reason R
for S’s empirical belief that p
2. Subjective reasons require possession of concepts corresponding to
what R represents
Conclusion: The contents of perceptual seemings require possession of
concepts corresponding to what R represent.

The first premise is in need of justification. Subjective reasons are sometimes

assumed to be the contents of beliefs. If we assume this up front, then premise 1,
of course, is false. However, the main reason for denying that the contents of
perceptual seemings can be reasons appears to be that it can give rise to
bootstrapping. Here is an example of this phenomenon, inspired by Susanna
Siegel’s (2011) argument against taking perceptual experiences to provide
immediate justification of belief.

Many first users of microscopes claimed to see embryos in sperm cells that they
examined using a microscope. They held this belief prior to the invention of
microscopes. So it seemed to them that the sperm cells contained embryos
because of their beliefs that the cells contained embryos. While they had no
reason for their initial belief, once the belief is based on their seeming, they
suddenly had a reason for their belief. But it would be odd if a belief that had no
initial reason generated a reason for that very belief.

My reply to this sort of argument is that the seemings in this case are epistemic,
not perceptual. If someone convinces a first microscope user M that sperm cells
do not contain embryos prior to the invention of the microscope, then M will not

  38  
believe that the cells contain embryos. So if M is rational, it will not seem to her
as if the cells contain embryos when she finally looks in the microscope. So, the
seemings in question are epistemic. But epistemic seemings, in general, do not
provide us with a reason for our empirical beliefs. So the main argument for
denying that perceptual seemings can serve as subjective reasons is unsound.
Together with my previous argument, it follows that perceptual seemings
represent only if the agent possesses the relevant concepts.

11. Conclusion
I have argued that ‘seem’ functions semantically as a subject-raising verb. As
‘seem’ is a raising verb, ‘It seems’ can be treated as a sentential operator. I have
argued that ‘It seems’ is a hyperintensional and contextually flexible operator.
The operator distributes over conjunction but not over disjunction, conditionals or
semantic entailments. I have further argued that the operator does not commute
with negation and does not agglomerate with conjunction.

In the second part of the paper I offered reasons for thinking that the mental
states expressed by perceptual uses of ‘seem’ have contents that are
nonconceptual despite being perspectival: Their contents have a truth-value only
relative to a centered world, i.e., a world in which an individual and a time are
marked. But the contents do not involve a de se mode of presentation picking out
the individual in the center. Finally, I argued that the mental states expressed by
perceptual uses of ‘seem’ represent only if the agent possesses concepts
corresponding to what is represented.

Unlike perceptual experiences, perceptual seemings can represent only once we

have the requisite conceptual tools in our possession. The idea that we can
sometimes store perceptual experience for later interpretation when we have
acquired the right conceptual tools is familiar. Madeleine Ransom, a student at
University of British Columbia, told me the following story from her childhood. In
her hometown they occasionally had fairs with pony rides. As she was standing

  39  
in line waiting for her five-minute turn on a pony, she suddenly realized that one
of the ponies appeared to be giving birth. Furiously she jumped the line and
grabbed the organizer’s sleeve. “Mister! Look. Look,” she shouted with her tiny
voice, “one of the ponies is about to give birth. It looks like the baby is stuck; its
leg is hanging out.” The organizer looked at her in disbelief. “Go back to your
mom,” he said calmly. Only years later did it all click for Madeleine.11

References
Bealer, G. (1998). “Propositions”, Mind 107, 1-32.
Block, N. (2007). “Consciousness, accessibility, and the mesh between
psychology and neuroscience”. Behavioral and Brain Sciences; 30(5–6): 481–
499
Brogaard, B. (2012a). Transient Truth: An Essay in the Metaphysics of
Propositions, Oxford University Press.
Brogaard, B. (2012b). “Do We Perceive Natural Kind Properties”, Philosophical
Studies.
Cappelen, H. (2012). Philosophy Without Intuition, Oxford: Oxford University
Press.
Cappelen, H. and Hawthorne, J. (2009). Relativism and Monadic Truth, Oxford:
Oxford University Press.
Chisholm, R. M. (1957). Perceiving: A Philosophical Study, Ithaca: Cornell
University Press.
Chomsky, N. (1986). Knowledge of language. New York: Praeger.
Cohen, J. (2004). “Color Properties and Color Ascriptions: A Relationalist
Manifesto”, The Philosophical Review 113: 451-506.
Cohen, J. (2009). The Red and the Real: An Essay on Color Ontology, Oxford:
Oxford University Press.
Crane, T. (1992). “The Nonconceptual Content of Experience”, in T. Crane, ed.
                                                                                                               
11
For helpful comments on a draft of this paper I am grateful to Murat Aydede, David Chalmers,
Kenny Eswaran, Jonathan Jenkins Ichikawa, Dan Kervick, Peter Lasersohn, Mark Lance, Janet
Levin, Hannes Leitgeb, Chris Mole, Lewis Powell, Brian Rabern, Madelaine Ransom, Ori
Simchen, Ralph Wedgwood, Alan White and audiences at Duke University, University of British
Columbia, University of Missouri, St. Louis and University of Southern California.

  40  
The Contents of Experience, Cambridge: Cambridge University Press, pp. 136-
157.

Egan, A., Hawthorne, J. and Weatherson, B. (2004). “Epistemic Modals in

Context”, in Contextualism in Philosophy, ed. G. Preyer and G. Peter, Oxford:
Clarendon Press: 131-68.

Heck, R. (2000). “Nonconceptual Content and the ‘Space of Reasons’ ”,

Philosophical Review 109: 483-524.

Heim, I. (2006). “Remarks on comparative clauses as generalized quantifiers”,

Ms, MIT.

Kaplan, D. (1989). “Demonstratives”, J. Almog, J. Perry and H. Wettstein, eds.,

Themes from Kaplan, New York: Oxford University Press.

Kripke, S. (1980). Naming and Necessity, Cambridge, MA: Harvard University

Press.
Kölbel, M.( 2002). Truth Without Objectivity, London: Routledge.

Kölbel, M. (2003). Faultless Disagreement, Proceedings of the Aristotelian

Society 104: 53-73.

Max Kölbel, M. (2007). “How to Spell Out Genuine Relativism and How to Defend
Indexical
Relativism”, International Journal of Philosophical Studies 15: 281-288.

López de Sa, D. (2007). “The Many Relativisms and the Question of

Disagreement”, International Journal of Philosophical Studies 15: 269 – 279.

  41  
MacFarlane, J. (2005a). “The Assessment Sensitivity of Knowledge Attributions”,
in Oxford Studies in Epistemology 1, ed. T. Szabo-Gendler and J. Hawthorne,
Oxford: Oxford University Press: 197-233.
MacFarlane, J. (2005b). “Making Sense of Relative Truth”, Proceedings of the
Aristotelian Society 105: 321-39.
Montague, R. (1973). “The Proper Treatment of Quantification in English”, in
Hintikka, K. et al., ed., Approaches to Natural Language, pp. 221-242
Partee, B. (1975). “Deletion and Variable Binding”, in E. Klima, ed., Formal
Semantics of Natural Language, Cambridge University Press, Cambridge.
Pollock, John. (1984). “Reliability and Justified Belief” Canadian Journal of
Philosophy 14: 103-114.
Postal, P. (1974). On Raising, Cambridge, Mass. MIT Press.
Prinz, J. (forthcoming). Consciousness Equals Attention.
Schroeder, M. (1998). “Having Reasons”, Philosophical Studies 139: 57-71.
Siegel, S. (2011). “Cognitive Penetrability and Perceptual Justification”, Nous.
Speaks, J. (2005) “Is There a Problem About Nonconceptual Content?,"
Philosophical Review 114: 359-398.

  42