Relative Truth and the First Person*
Friederike Moltmann
IHPST, Paris
fmoltmann@univ-paris1.fr
January 2007
In recent work on context-dependency, it has been argued that certain types of sentences give
rise to a notion of relative truth. In particular sentences containing predicates of personal taste
and moral or aesthetic evaluation as well as epistemic modals are held to express a
proposition (relative to a context of use) which is true or false not only relative to a world of
evaluation, but other parameters as well, such as standards of taste or knowledge or an agent.
Thus, a sentence like chocolate tastes good would express a proposition p that is true or false
not only at a world of evaluation, but relative to the additional parameter as well, such as a
parameter of taste or an agent.
I will argue that the sentences that apparently give rive rise to relative truth should be
understood by relating them in a certain way to the first person. More precisely, such
sentences express what I will call first-person-based genericity, a form of generalization that
is based on or directed toward an essential first-person application of the predicate. This
account, I will argue, avoids a range of problems that standard relative truth theories of the
sentences in question face and explains a number of further features that such sentences
display.
First-person-oriented genericity is expressed explicitly in English by sentences that
contain the generic pronoun one as in (1a) or its empty counterpart, socalled arbitrary PRO (as
generative syntacticians call it), as in (1b):
(1) a. It is nice when one is walking in the park.
b. It is nice PROarb to walk in the park.
2
(1a) and (1b) have a natural reading on which they express a generalization on the basis of the
speaker’s own, perhaps onetime, experience or action. That is, (1a, b) are naturally used as an
expression of the speaker’s own evaluation of his walking in the park and at the same time
express a generalization: for anyone x, x’s walking in the park is nice for x.
The key observation is that the same intuitions of relative truth that are displayed by
sentences with evaluative predicates or epistemic modals arise with sentences such as (1a)
and (1b). In previous work I have developed an account of generic one on which generic one
expresses (contextually restriction) quantification over entities qua being identified with by
the relevant agent. In this paper, I will argue that this analysis can and should be carried over
to other sentences displaying the intuitions of relative truth, in particular predicates of
personal taste as well as epistemic modals. However, not all types of sentences for which a
relativist treatment has been proposed should be analysed in terms of first-person-based
genericity. I will introduce a distinction between two types of truth-relative expressions, one
of which does not give rise to all the relative truth intuitions and is to be treated in just the
same way as de se interpreted pronouns. While the latter in fact does not involve relative
truth, the former I will argue involves truth relativism in some fundamental, though firstperson-related way.
I will start with presenting the core intuitions that have motivated the relativist account,
and point out further crucial features of the relevant linguistic expressions. I then introduce
the notion of first-person-based genericity on the basis of my previous analysis of generic one
showing that generic-one-sentence exhibit the intuitions of relative truth in the same way. I
then argue that the same semantic analysis can and should be carried over to other sentences
displaying intuitions of relative truth. I finally come to a crucial distinction between two sorts
of first-person connections.
1. The intuitions of relative truth
There are two fundamentally different types of expressions to distinguish that have motivated
a relativist account. The first type consists of predicates of evaluation and epistemic modals;
the second type consists of ‘self-locating’ expressions such as nearby, local, or right and left.
Let us call them type 1 and type 2 expressions. I will come to the characteristic features of the
second type, which will also include de se interpreted pronouns, at the end. I first will focus
3
on predicates of evaluation and epistemic modals, more in particular on predicates of personal
taste.
1.1. Faultless disagreement
One main motivation for a relativist account has been the possibility of faultless disagreement
(Koelbel 2002, 2003). Faultless disagreement consists in a situation in which two agents
disagree about the truth of a proposition, with neither apparently being at fault. Thus, below,
neither A nor B may be at fault, yet they disagree:
(2) A: White chocolate tastes good.
B: No, I disagree, white chocolate does not taste good.
In fact A and B subsequently may enter a dispute, let’s say whether they should start
producing white chocolate. Faultless disagreement can arise not only in a situation of
conversation, but also manifests itself in intuitions about two agents being involved in
different conversations (MacFarlane, ms 2) or about two agents’ beliefs (Koelbel 2003). What
is crucial about faultless disagreement is that both agents seem to be right in their claims or
beliefs, but yet they disagree. Faultless disagreement does not arise with sentences that
express different propositions when uttered by different speakers. Moreover, faultless
disagreement does not arise when the ‘judge’ is made explicit, as in (3):
(3) A: White chocolate tastes good to me.
B: White chocolate does not taste good to me.
Another important observation, generally disregarded in the literature, is that faultless
disagreement does not arise with purely subjective attitudes, such as find or think (on a
‘subjective’ reading):
(4) a. A finds / thinks that white chocolate tastes good.
b. B finds / thinks that white chocolate does not taste good.
It only arises with ‘truth-directed attitudes’, such as believe:
4
(5) a. A believes that white chocolate tastes good.
b. B believes that white chocolate does not taste good.
Only in (5) not in (4) do A and B intuitively disgree.1
The same contrast holds for speech acts: expressive speech acts do not give rise to faultless
disagreement, only truth-directed ones, such as assertions, do:
(6) a. ? John expressed his feeling that the wine tastes good. Mary disagreed with him.
b. John claimed that white chocolate tastes good. Mary disagreed with him.
The difference between the two kinds of propositional attitudes and illocutionary act
types obviously has to do with the fact that attitudes like ‘find’ and (non-truth-directed)
‘think’ as well as expressive speech acts do not aim at truth, but involve inherently subjective
contents. By contrast, attitudes and speech acts whose aim is truth are incompatible with such
purely subjective contents. I will later argue that the two kinds of attitude or speech act verbs,
even if they take the same clausal complements, take in fact different kinds of contents as
arguments. That is, the ‘purely subjective contents’ which are the arguments of attitudes like
‘find’ are of a different sort from the contents that are required by truth-directed speech act
and attitude verbs, the latter being in fact first-person-based generic propositions.
It is also important to notice that truth-directed attitudes and speech acts allow for
evaluative contents in which the ‘judge’ is made explicit:
(7) I believe / claim that white chocolate tastes good to me.
This tastes good to me is not a subjective, but an objective content, an objective content about
a subjective evaluation.
Related to the possibility of faultless disagreement is the observation that two agents may
agree about the content of an evaluative sentence even if it is clear that the relevant
parameters of evaluation of the two agents are different:
(8) John and Mary agreed that the wine tastes good (but for very different reasons).
5
This intuition is suitably called faulty agreement Like faultless disagreement, it arises in just
the same way in relation to different agents’ beliefs. That is, (8) is true even if John and Mary
did not talk to each other, but just believe, for different reasons, that the wine tastes good..
To summarise, for either agreement or disagreement the parameters of assessment do not
matter, but only the truthconditionally incomplete content alone. Or so the data have been
interpreted by relativist theories.
1.2. Retraction
Another intuition for the truth of a sentence being relative is the possibility of retracting a
proposition, once truthfully asserted or believed, at some later point in time. Retraction, as a
matter of fact, can be viewed simply as faultless disagreement involving the same agent at
different times. The possibility of retraction has been most often discussed with epistemic
modals. Thus, the content of an utterance of John may be in France, correctly asserted at
some point in time, may later be withdrawn in view of evidence that excludes John’s being in
France. Retraction is also possible, though, with sentences involving predicates of personal
taste. Thus, John may withdraw the content of his ‘faultless’ claim or belief that Brouilly
tastes good, at some later point in his life, having changed or refined his taste of wine. But
John was of course not at fault in his belief that Brouilly tastes good at the earlier time and
thus was right on both occasions.
1.3. Sharing
By ‘sharing’ I mean the possibility of sharing of propositional contents by agents involved in
different contexts of evaluation. Sharing is a phenomenon actually not considered in the
recent literature on relative truth. However, it has been discussed at length in regard to moral
predicates by Schiffer (1990). If a sentences S involves relative truth, then sharing of
propositional contents consists in the intuition that agents, even if they are clearly involved in
different contexts of evaluation, share the same propositional content when they have a
propositional attitude that S. There are various linguistic manifestations of sharing.
On of them is the possibility of sentences with a conjunctive or plural subject and a single
clausal complement of an attitude verb:
(9) John and Mary / These two people both believe that the wine tastes good.
6
(9) will requires a single interpretation of the clausal complement, which can thus not involve
two different taste parameters or agents.
Another linguistic manifestation is the validity of the inference in (10), assuming that A’s
and B’s criteria for evaluating wine are known to be quite different (A, but not B, lets say,
being a connoisseur):
(10) A believes the wine tastes good.
B believes the wine tastes good.
A and B believe the same thing.
Such an inference holds with any propositional attitude or speech act verb. Evaluative
predicates, in licensing the inference, thus differ from other context-dependent expressions
such as demonstratives, with which the inference is not valid:
(11) John believes that Mary is there (in New York).
Bill believes that Mary is there (in Boston).
John and Bill believe the same thing.
The validity of the inference in (10) does not hinge on some ‘looseness’ of uses of the
expression the same thing. With a conclusion containing a free relative as in (12) the same
sort of inference is validated, assuming again that John’s and Mary’s taste parameters are
rather different:
(12) John believes what Mary believes, namely that the wine tastes good.
Even more importantly, the same sort of inference is valid with a conclusion containing the
corresponding nominalization of the verb, such as the same belief or the same claim:
(13) John believes that the wine tastes good.
Mary believes that the wine tastes good
John and Mary have the same belief / share a belief.
(14) John claimed that the wine tastes good.
Mary claimed that the wine tastes good.
7
John and Mary made the same claim.
These inferences are valid even if John’s and Mary’s taste parameters’ are different (and even
are known to be different).
Such ‘linguistic’ criteria for sharing are not the only criteria for establishing shared
meaning when different contexts of assessment are involved. There are also a range of other
intuitions that indicate that the meaning, that is, ‘conceptual role’ of evaluative predicates
when used by different agents is the same even if the agents’ criteria for applying the
predicates are rather different. This point has been made rather thoroughly, in the context of a
discussion of moral predicates, by Schiffer (1990). Schiffer discusses various intuitions that
show that moral predicates have the same meaning for two agents even if the agents differ in
their moral criteria and principles for applying the predicates. For example, even if two agents
have different moral principles, their concepts of good and wrong are the same given the roles
of those concepts in determining how the agents want the world to be, given the kinds of
emotional responses they trigger, given the relation of those concepts to related ones (ought,
just), given that the concepts involve the same process of moral training (of punishment and
reward), and given the predicates’ common ‘point’, in getting people to behave in a certain
way. Needless to say similar criteria for the individuation of conceptual roles can be found for
other evaluative predicates as well.
Epistemic modals are another often discussed class of expressions displaying the intuitions
of relative truth (Egan/Hawthorne/Weatherson 2005, Egan, to appear, MacFarlane, ms 1).
They clearly display faultless disagreement: one person may believe or claim that John may
be in Paris, while another person with more knowledge disagrees. Epistemic modals also
allow for retraction: a person may at a given point have believed or claimed that John may be
in Paris, but withdraw the content of that belief or claim later in view of more evidence.
Epistemic modals also lead to sharing of content, as shown by the validity of the following
inference:
(15) A believes that it may rain (because she heard the weather forecast).
B believes that it may rain (because he noticed the cloud formation).
A and B believe the same thing (but for different reasons).
The inference is also valid with the conclusion in (16a) or, as faulty agreement, with the
conclusion in (16b):
8
(16) a. A and B have the same belief.
b. A and B agree that it may rain.
There are other context-dependent expressions which give rise to sharing, but not faultless
disagreement (or faulty agreement). These are certain temporal and spatial expressions such
as right, left, local or neighboring.2
Thus, from (17a) (which itself involves sharing already), one can infer both (17b) and
(17c):
(17) a. John and Mary both believe that the tree is to the left.
b. John and Mary believe the same thing.
c. John and Mary share the same belief.
However, given (17), it would not be the case that John and Mary agree, if what counts left
for Mary is right for John, an observation pointed out by John Hawthorne (oral
communication). Rather, they clearly disagree. That is, the inference from (17a) to (18) is
invalid:
(18) John and Mary agree.
Also retraction does not seem to hold for such expression. Thus, if John a minute ago believed
that the tree is to the left, he will not withdraw his belief once he turns around, maintaining
now that the tree is to the right and rejecting his former belief. Rather his former belief
content is individuated partly by John’s former spatial position, and thus the belief remains
true.
The crucial point that the three intuitions, faultless disagreement, retraction and sharing,
appear to establish is that the propositional content of sentences with predicates of personal
taste is the same even when the context-dependent criteria of evaluation involved (such as
standards of taste etc.) are clearly distinct.
2. Problems for standard relativist theories
9
The relativist theory accounts for the intuitions of relative truth by postulating additional
parameters as part of the context relative to which the truthvalue of a proposition is to be
evaluated. Truth then is relative in the following sense:
(19) Relative Truth
The truth of a sentence S is relative iff for an utterance context u, the proposition that S
expresses in u is true or false only relative to a context of evaluation c that contains not
only a world (and time) but other elements as well.
There are different views as to what the context of assessment may contain. On one view, it
may contain whatever parameters seem necessary for the evaluation of the expression in
question, such as parameters of taste or epistemic standards (MacFarlane 2005a, b). On
another view, it will just contain, besides a world (and perhaps time) of evaluation, the
relevant agent, or equivalently, a centred world (a pair consisting of a world and an agent)
(Egan/Hawthorne/Weatherson 2005, Lasersohn 2005). It appears that a relativist account,
especially on the first version, is problematic in several respects.
2.1. Explaining disagreement
The most important problem for the relativist account, it seems to me, is that it does not really
explain faultless disagreement. The problem is: why on the relativist account can there be
disagreement when the truth of an utterance is to be evaluated relative to different contexts of
assessment? How can the relativist really account for the difference between white chocolate
tastes good to me on the one hand and white chocolate tastes good on the other hand? Here
the answer would depend on the way the context of evaluation is supposed to be determined.
Suppose that the context of evaluation depends in fact on the speaker’s own choice, his
taking a particular perspective for evaluating the proposition, as Lasersohn (2005) has
proposed. If the context of assessment depends on an agent’s choosing a perspective of
evaluation, then one would think this choice of a perspective to be part of an overall
contextually completed intended meaning of the utterance. But then this contextually
completed meaning would not be relevantly different from the meaning of chocolate tastes
good to me (or chocolate tastes good to d, where d is the relevant judge). But we have seen
that the latter does not give rise to faultless disagreement, whereas the former does.3,4
10
The common version of a relativist account, in particular that of MacFarlane (2005a, b),
takes the context of assessment to be uniquely determined by whoever assesses the truthvalue
of the proposition. This obviously would not predict an identity of the intended meaning of
white chocolate tastes good and white chocolate tastes good to d (for d being the relevant
agent). The problem with this version of the relativist account, however, is that it is unable to
distinguish the content of a truth-directed propositional attitude or speech act from a merely
subjective content (such as that of the attitude expressed by find). That is, why is there the
previously noted difference between John finds that chocolate tastes good and John claims
that chocolate tastes good? It seems that the relativist account has difficulties in particular in
accounting for the use of evaluative sentences the context of truth-directed speechacts or
attitudes. If a speaker utters chocolate tastes good then, knowing the truth-relative semantics
of the sentence, the speaker should know that the content of his truth-directed attitude or act
should relate only to his own context and thus, from his point of view, should just be about
just his own context. From his point of view, no considerations need to be made that the
content of his utterance also target (be evaluable as true at) the context of the addressee. Of
course, the speaker may know that the addressee will truth-evaluate the utterance at his
context. But why should he be bothered about that and why should he engage in a
disagreement? The addressee, in turn, given his knowledge of the relativist semantics of the
sentence uttered, should of course know that too. It thus remains a mystery why the situation
should give rise to disagreement. The situation appears entirely undistinguishable from the
one where the speaker expresses or upholds his own subjective opinion without targeting the
addressee’s parameters of evaluation in any way, that is, the situation made explicit by
attitude reports like John finds that chocolate tastes good.
Another problem for the account concerns the parameter themselves that are used for the
relativization of truth. Those parameters, be they agents or evaluative or epistemic standards,
are individuated externally and should thus be equally accessible and identifiable for all the
interlocutors, given the circumstances of the utterance situation or propositional attitude.
Given the standard view about assertion and belief as aiming for truth, why shouldn’t a truthconditionally complete content including the relevant parameters be identified in a given
context by each one of the interlocutors and understood as the content to be communicated?
That is, why should agreement or disagreement not pertain to such a truth-conditionally
completed content as it would serve the aim of assertion or belief, rather than the
truthconditionally incomplete content expressed by a sentence relative to a context of
utterance?5, 6
11
The relativist account is also difficult to maintain for the content of a sentence like
chocolate tastes good when it acts as the object of belief. There are two options of how to
construe the object of belief in this case. First, the object of belief could be viewed as a pair
consisting of a proposition and context of assessment (Lasersohn 2005). But again this could
not give justice to the intuition that two people one of whom believes that white chocolate
tastes good and the other of whom believes that white chocolate does not taste good disagree,
rather than have different, but compatible overall beliefs. Alternatively, the content of belief
could be taken to be a truth-relative proposition with the judge being identified with whoever
is the believer (Egan/Hawthorne/Weatherson 2005). The problem then is that no justice would
be given to the distinction between the purely subjective content of attitudes like think or find
and the content of belief that aims at truth. Intuitively, if John thinks or finds that white
chocolate tastes good and Joe thinks or finds that it does not taste good, then they have
different thoughts or rather ‘tastes’, but they do not disagree. However, when John believes
that white chocolate tastes good and Joe believes that it does not taste good, then the intuition
is that they disagree. Taking the contents of the two sorts of attitudes, the subjective one and
the truth-directed one, to be exactly the same fails to account for this crucial difference.7
To summarize, the relativist account takes care of the intuition that the same propositional
content is involved in the relevant examples, thus explaining the properties of sharing and
retraction. But it does not provide a real explanation of why there is disagreement (and thus in
fact also not really explain retraction either, that is, disagreement across times).
2.2. The meaning-intention problem
The first version of the relativist account posits parameters, such as parameters of taste, as
part of the context of assessment. The problem with such entities is the same as arises in a
number of cases in which entities of a certain sort are posited as implicit arguments, namely
what Schiffer (1987) has discussed as the meaning-intention problem. A speaker generally
would not be able to identify entities such as parameters of taste as part of a context relative to
which he intends his utterance to be made. A speaker may be entirely justified in uttering a
sentence like this tastes good without in any way being able to identify a ‘taste parameter’ as
part of his intentions. This problem of cognitive accessibility is a natural generalization of
Schiffer’s meaning-intention problem.
2.3. The first-person orientation
12
Relativist accounts of the second sort allow for readings involving error through
misidentification that in fact do not exist. Suppose that Joe reads a description of someone
that he fails to realize is a description of himself and that misdescribes him as someone that
likes white chocolate when Joe in fact does not. In this case, given a relativist account, of
either sort, Joe believes that white chocolate tastes good would be true (though Joe’s belief
would not). An agent d could not believe p to be true relative to a taste parameter that d does
not believe to be his own, that is, d could not believe p to be true relative to d without
recognizing that d is himself. In other words, an agent must identify himself as the judge
when standing in an attitudinal relation to an evaluative propositional content. The problem
for the relativist account consists in that it assigns to the additional parameter of assessment a
status of a value of a pronoun interpreted de re, when in fact the parameter should have the
status of the semantic value of a pronoun interpreted de se.
2.4. Predictions about attitude contexts
It is not quite clear what the predictions of the relativist account are for attitude reports.
McFarlane (2005b) assumes that the additional parameter of assessment of the embedded
proposition always becomes the parameter of the assessment of the entire sentence. This is
adequate, however, only for nonattitudinal embedding predicates, such as modal and temporal
ones, as in (20):
(20) It could be the case that chocolate tastes good.
In (20), the taste parameter is clearly that of the speaker. This is different with attitude reports:
(21) a. John thinks that chocolate tastes good.
b. John thinks that it might rain.
Clearly the standard of taste in (21a) and the epistemic state in (21b) is that of John, not the
speaker’s. In general, with embedding attitude verbs, the ‘additional parameter’ will be the
one of the described agent. This is not a problem for a relativist account as such, though. As
Egan/Hawthorne/Weatherson (2005) make clear, for a truth-relative proposition p, an attitude
report d believes that S can be made to be understood as: d believes S to be true relative to d.
13
2.5. The quasi-first-person orientation
There is another important feature of evaluative predicates and epistemic modals and that is
that their first-person-orientation in independent contexts need not be strict, but may relate to
another agent with whom the speaker only identifies. For example, a mother may persuade a
child to eat by uttering (22), without thereby expressing her own taste judgment:
(22) Applesauce tastes good.
The same point can be made with because-clauses and questions:
(23) a. John took another spoon because it tasted so good.
b. Does this taste good?
Here the speaker may just identify with (or projects himself onto) John (in 23a) or the
addressee (23b), not being interested in his own taste judgments.8
Also with epistemic modals may the first-person orientation involve an identification with
another agent. Thus Egan/Hawthorne/Weatherson (2005) note that (24) could be uttered by a
speaker who knows better but identifies himself with a person trying to find a way out of a
maze:
(24) The exit may be this way.
Let me call the first-orientation when the speaker in fact identifies with another agent quasifirst-person-orientation. This quasi-first-person-orientation with evaluative predicates and
epistemic modals is something not available with first-person pronouns except in certain
contexts, such as imagination, as in (25) (Williams 1979):
(25) I imagine that I am Napoleon.
In (25) the content of imagination does not involve an identification of the speaker’s actual
person with Napoleon, but rather speaker simply projects himself onto Napoleon. In fact, I
14
will later propose to incorporate what is particular to contexts of imagination into the
semantic analysis of evaluative predicates and epistemic modals themselves.
2.6. Explaining the kind of context dependency involved
A final problem for the standard relativist account is that it does not explain why the
phenomena in question give rise to a relativization of truth rather than context-dependency in
the traditional sense. The approach makes it look like an accidental fact that evaluative
predicates and epistemic modals give rise to an enrichment of the context of assessment,
rather than the context of use, unlike, for example, spatial indexicals like there.
This objection does not obtain for all relativist accounts, though, namely not for those that
only admit an inclusion of the relevant agent in the context of assessment
(Egan/Hawthorne/Weatherson 2005, Lasersohn 2005).
The relativist account also seems unable to distinguish type 1 expressions and type 2
expressions. If the relativist were successful, as it aims to be, in accounting not only for
sharing, but also for faultless disagreement and retraction, it would be inapplicable to type 2
expressions. Standard relativist theories provide no means for distinguishing the two types of
context-dependent expressions.
3. First-person-based genericity with generic one
I now turn to generic one whose analysis is to shed a significant light on the range of issues
relating to relative truth. Generic one leads to the expression of what I will call first-personbased genericity, which in turn I will take to be the true source of intuitions of relative truth.
3.1. Some crucial data concerning generic one
Let me first present some basic linguistic data concerning generic one.9 Generic one is a
pronoun that always leads to generic sentences and it bears a particular connection to the first
person: it leads to genericity that is either based on a first-person attribution or directed
toward a first-person attribution. Generic one alternates with arbitrary PRO, which is its
empty counterpart and thus occurs in those contexts in which an empty pronominal element,
rather than an overt noun phrase is required (Moltmann 2006). Arbitrary PRO and generic one
15
can covary in contexts like (26a) (that is, they take the same semantic values under the
relevant assignments), and arbitrary PRO may act as the antecedent for one, the possessive
one’s or the reflexive oneself, as in (26b):
(26) a. PROarb to live a great life is to realize one’s true potential.
b. The tailor knows what PROarb to wear at one’s own wedding.
Generic one always occurs in generic sentences, and as such it can occur in apparently two
distinct ways: as genericity-inducing, as in the first occurrence in (27), and as a bound
variable, as in the second occurrence in (27):
(27) One sometimes thinks one’s life is too short.
In both occurrences generic one is best taken to be an expression that introduces a variable
subsequently to be bound by a generic quantifier Gn, which is formally associated with a
syntactic element in sentence-initial position, as in (28) (Moltmann 2006):
(28) Gen x
x sometimes thinks that x’s life is too short.
The generic quantifier Gn in (28) is the kind of quantifier used in the linguistic literature on
genericity (see for example Krifka et al. 1995, Cohen 2002). It is a quantifier that allows for
exceptions and has modal force (ranging not just over actual individuals). A plausible way of
understanding it, quite suited for present purposes, is to take it to be a combination of a
universal quantifier ranging over possible worlds, restricted by some accessibility relation R
(relating the actual worlds to the ‘normal’ worlds) and a universal quantifier ranging over
individuals, possibly restricted by a contextually given property C:
(29) w x (wRwo & C(x)  P(x))
The contextual restriction C is needed to account for why one in (10), for example, may range
only over the students in a particular class:
(30) One has to hand in the essay tomorrow.
16
Of course, the logical form in (29) does not yet capture any first-person-orientation of
generic one. Before introducing the crucial modification, let us see how the first-personorientation manifests itself.
The first-person-based genericity of generic one and its empty counterpart manifests itself
in what at first sight appears to be a general availability of an inference to the first person, as
in (31):
(31) One can see the picture from the entrance.
I can see the picture from the entrance.
However, at ‘second sight’, it turns out that this inference is not in fact generally valid. (31)
can also be uttered by someone for some reason unable to see the picture himself. Thus,
generic one may display in fact a quasi-first-person-orientation.
The same point is made by (32), which does not display any conflict between the
grammatical first person and generic one because one here obviously involves identification
with people different from the speaker:
(32) One can see me from the entrance.
The first-person-orientation of generic one and arbitrary PRO concerns not only the
speaker, but also, in embedded contexts, whoever may be the described agent of the reported
attitude or speech act:
(33) a. John said that one can see the picture from the entrance.
b. John said that it is nice PROarb to walk in the park.
In attitude contexts, the first-person-orientation of generic one is particularly transparent,
for example when a generic one-sentence is embedded under an epistemic predicate:
(34) John found out that one can see the picture from the entrance.
For (34) to be true it is sufficient that John has had the experience of seeing the picture from
the entrance. Generic one-sentences differ thus from universally quantified and other generic
17
sentences, such as those in (35):
(35) John found out that people can see the picture from the entrance.
In (35), John has to have made sure in other ways that people other than himself can see the
picture from the entrance.
The first-person-orientation of generic one manifests itself also in the ability of those
pronouns to serve in an immediate description of a first-person experience:
(36) I find that one can easily forget one’s own past experiences.
The embedded sentences in (36) naturally serve as direct descriptions of a first-personal
psychological state, though the generalizing force is there as well. (36) thus differ markedly
from (37), where the attitude described takes as its immediate source third-person
observations, or else has a derived content, obtained only inferentially from a first-person
experience:
(37) I find that people can easily forget their / his past experiences.
Generic one (and its empty counterpart) is the most suited expression for generalizing
irreducibly subjective experiences as types of experiences.
Another manifestation of the first-person-orientation of generic one are certain restrictions
on which predicates generic one can accept. The predicates has a nose and lives in a big city
for example are hardly acceptable with generic one, though they are fine in other generic
sentences:
(38) a. ?? One has a nose.
b. The typical person has a nose.
(39) a. ?? One lives in a big city.
b. People live in a big city.
Roughly the restriction on predicates acceptable with generic one is that the predicate must
describe possible experiences or actions, that is, it should be a predicate with an irreducible
first-person application.10
18
3.2. The semantic analysis of generic one
The general idea is that sentences with generic one as a whole express a generalization based
on a (generally) irreducible first-person application of a predicate, that is, they express firstperson-based genericity. First-person-based genericity involves the ability of abstracting from
the particularities of one’s own person and situation, judging oneself to be normal in relevant
respects, and then generalizing to anyone meeting the same conditions. This way of
generalizing self-attributions of properties is a form of abstraction, requiring distinguishing
between relevant and irrelevant features of a given person and his situation. First-person
based genericity can also be viewed as a form of simulation in the sense of Gordon (1986,
1995a, 1995b), more precisely as what one may call generic simulation. In the case of generic
simulation, the relevant intentional agent generalizes simply his own situation, abstracting
from the features of his situation that are particular to himself. He does not need to project
himself onto a particular other person and make adjustments to adopt the other person’s point
of view (as in ordinary cases of simulation).
The notion of simulation also helps to understand the quasi-first-person-orientation of
generic one: first-person-based genericity does not require the agent to actually self-ascribe
the predicate. He may just identify with someone who could.
Thus, first-person-oriented pronouns involve self-reference that is detached from the
relevant agent’s actual person: it may involve self-ascribing a property while identifying
oneself with someone else and will always involve self-ascribing a property while identifying
with each one of a collection of individuals.
The formal semantic analysis of generic one that I have proposed makes use of a primitive
notion of identification ‘I’, a relation between an agent and another individual with which the
agent ‘identifies’. The basic idea is that generic one does not just range over individuals, but
individuals qua being identified with by the relevant agent (the speaker or agent of the
reported attitude or speech act). In other words, generic one now ranges over individuals
‘insofar as the agent identifies with them’. I roughly make use of the notion of a qua object in
the sense of Fine (1982), an object that is obtained from an ordinary object x and a property P
by a function qua (qua(x, P)). The crucial modification of (29) then consists in replacing ‘x’
by the two-place functional term ‘qua(x, λy[I y z])’, where the variable ‘z’ is to stand for the
relevant agent. When generic one is not genericity-inducing, it is represented by a simple
variable.
19
While, intuitively, the ‘ordinary’ variable concerns the truth conditions of a one-sentence,
the gloss λy[I y z] gives the ‘mode of presentation’ that is to govern the applicability of
predicates, providing the epistemic basis (or, as we will see, the practical purpose) for
applying the predicate.
What is the status of the variable ‘z’ in the overall logical form of the sentence? Here I
make use of Lewis’ (1979) account of attitudes de se as self-ascriptions of properties, that is,
the view that attitudes de se do not take propositions, but properties as their content. Thus, on
this view, (40a), which has an obligatory de se reading, would be represented as in (40b),
rather than as in (40c), which corresponds to an interpretation de re:
(40) a. John expects [PRO to win].
b. expect(John, x[win(x)])
c. expect(John, ^ he will win)
Given this, the variable ‘z’ can similarly taken to be bound by a lambda operator defining the
meaning of a generic-one-sentence as a whole as a property. This is illustrated in the logical
form of (41a) in (41b):
(41) a. One can see the picture from the entrance.
b. z[Gn x can see the picture from the entrance(qua(x, y[I y z]))]
That is, (41a) expresses the property of being a z such that for any contextually relevant x, x
qua being identified with by y can see the picture from the entrance.
3.3. A second use of generic one
The gloss of the qua objects generic one quantifies over not only provides epistemic grounds
for the application of the predicate, it may also serve a practical purpose, allowing for an
immediate self-application of an independently established generalization This generalization
when using generic one, crucially, is presented with the intention to be at least potentially
applied in a first-person way by the speaker or, more likely, the addressee, or both. This is the
case in particular in deontic sentences:
(42) a. One is not allowed to enter the room.
20
b. One should not lie.
c. One should be respectful toward the elder.
(43) a. The tailor knows what PROarb to wear.
b. John knows how PROarb to behave oneself.
Whereas with Strategy 1, the speakers’ own experience leads to the generalization expressed
by the one-sentence, in these cases the speaker presents an internalized, but already
established generalization, a law, general requirement, or general recommendation. The
generalizations expressed in (42a-c) crucially play a role in the speakers’ reasoning for his
actions, or better are meant to play a role in the addressee’s reasoning. For example, if the
addressee accepts (42a), then this is likely to prevent him from entering the room. That is,
deontic one-sentences are generally uttered with the intention that they play a future role as
premises in the addressee’s practical reasoning, reasoning in which self-reference is essential.
The reason why generic one-sentences are so suited for governing an agent’s practical
reasoning is that they allow for an immediate first-person application by anyone who accepts
them.
As before, ‘first-person application’ need not be understood as application to one’s own
actual person: generic one-sentences can be used in such a way that the speaker identifies with
a group of human beings not including himself:
(44) One has to hand in the essay today. (teacher to students)
The first-person connection of sentences with generic one or arbitrary PRO using Strategy
2 can also be seen from the suitedness of such sentences for the expression of practical
knowledge as in (45a, b), where generic one or arbitrary PRO could not be appropriately
replaced by another generic NP, as in (45b):
(45) a. I know what one can do.
b. I know what PROarb to do.
c. I know what people can do.
The second use of generic one is not always independent of the first use. For example,
sentences expressing physical possibility could be licensed by either use.
21
4. Relative truth and first-person-oriented pronouns
The main problem for the standard relativist theories is that they disregard the fundamental
first-person character of ‘relative truth’ as well as the genericity that is involved in the
sentences in question. At least with predicates of personal taste, the intuitions giving rise to
relative truth can be traced to the first-person-based genericity that is involved in the
semantics of such predicates in just the same way as in the semantics of generic one. That is,
the properties [1] – [3] will be explained as going along with first-person-based genericity.
This account does entirely away with such entities as parameters of taste or epistemic states as
part of a context of assessment, as they are posited on one version of the relativist approach.
Moreover, rather than making use of a context of assessment enriched by an agent (as on the
other version of the relativist approach), the account will construe the first-person-orientation
involved in terms of the nature of the propositional content itself.
4.1. Relative truth intuitions with sentences involving first-person-oriented pronouns
As a point of departure for the parallelism between generic one-sentences on the one hand and
sentences with evaluative predicates and epistemic modals on the other hand, let us observe
that the semantic relation of (46a) to (46b) is about the same as the relation of (47a) to (47b):
(46) a. I can see the picture from the entrance.
b. One can see the picture from the entrance.
(47) a. Chocolate tastes good to me.
b. Chocolate tastes good.
Just as (46b) is a generalization of (46a), (47b) is a generalization of (47a). That is, both
sentences (46b) and (47b) express a first-person-based generalization.
Unlike predicates of personal taste, however, generic-one-sentences do not display any
difference in content with truth-directed and purely subjective attitudes and speech acts:
(48) a. John claims / believes that one can see the picture from entrance.
b. John thinks / finds that one can see the picture from the entrance.
22
The content of the embedded sentence in (48b) is as generic as that in (48a). This supports the
view that predicates of personal taste are ambiguous in their meaning in a sentence,
depending on the kind of attitude or speech act involved, whereas generic one-sentences are
unambiguously generic.
The reason why (48b) lacks a subjective content reading is precisely that the
quantificational content of generic one-sentences is incompatible with such a content. The
attitude verb find requires a clausal complement with an evaluative predicate whose evaluator
is to be identified with the first person, that is, the agent of the attitude.
The three intuitions of relative truth exhibited by predicates of personal taste and epistemic
modals are all shared by sentences with first-person-oriented pronouns.
First, faultless disagreement is possible with generic one-sentences in just the same way
as with predicates of personal taste. One person might be right in asserting (49a), whereas
another person, used to a greater level of comfort, may be right in his way in asserting (49b):
(49) a. One can sleep on this sofa.
b. One cannot sleep on this sofa.
Yet the two clearly disagree.
Also faulty agreement is possible. Thus, (50) is possible even if John and Mary’s grounds
for their generalization are quite different:
(50) John and Mary agreed that one can sleep on this sofa.
Also retraction is possible with generic one-sentences. Someone who previously was
correct in asserting (46b) may retract his assertion later when the place of the picture has
changed.
Finally, two people, with quite different experiences as their epistemic source, may share
the content of a generic one sentence. Thus, an inference of the following sort is always valid:
(51) A discovered that one can see the picture from the entrance.
B discovered that one can see the picture from the entrance.
A and B discovered the same thing (namely that one can see the picture from the
entrance).
23
Free relative clauses and conjunction further support the criterion of ‘sharing’ with generic
one-sentences. Thus, (52a, b) are equally possible as conclusions of (51), as is (52c), with a
nominalization:
(52) a. A discovered what B found out, namely that one can see the picture from the entrance.
b. A and B discovered that one can see the picture from the entrance.
c. A and B made the same discovery.
These parallels between generic one-sentences and sentences with predicates of personal
taste strongly suggest [1] that the three intuitions of relative truth are to be explained in terms
of first-person-based genericity, and [2] that first-person-based genericity is also involved in
the semantics of sentences with predicates of personal taste in truth-directed contexts. Let me
turn to the latter first.
4.2. Explaining relative truth intuitions with first-person-based genericity
On the present analysis, generic-one-sentences (on one reading) express a generalization
based on the agent identifying with the individuals the generalization concerns. As such
generic one-sentences express a first-person-oriented content, which was construed as a
property.
How does this account explain the three intuitions of relative truth? Sharing of contents of
generic one-sentences by different agents is possible for just the same reasons as sharing of
contents of sentences involving de se pronouns is possible on Lewis’ (1979) account of
attitudes de se: it is simply sharing of a property, a property that is to be self-ascribed in the
attitudinal state involved. Lewis himself in fact took the possibility of sharing of the contents
of self-locating beliefs (which I will discuss more to at the end of this paper) to be evidence
for his account of attitudes de se. Thus expressions like the same thing will stand for
properties, the contents of the attitudes or speech acts in question.
This view is harder to maintain for the same belief, the same claim or the same discovery,
with which sharing was possible in the same way. A belief is not a property, nor is it in fact a
proposition. A belief can be surprising, can be abandoned and acquired, can be widespread
and can last, all properties a property or a proposition cannot have. At the same time a belief
generally has truth conditiions and can be about particular objects. Entities like beliefs, claims
24
and discoveries are then better viewed as particulars, individuated both by a content (which
may be a property), as well as an attitudinal, illocutionary, or epistemic mode. They are
contents qua being entertained in the relevant mode. To be more correct, an entity that is ‘the
belief that S’ is best viewed as a type or ‘kind’ of object, a universal, whose instances are
entities of the sort d’s belief that S, for some agent d. (cf. Moltmann 2003a). Entities like d’s
belief that S, d’s claim that S, or d’s discovery that S are individuated by a content, a mode as
well as an agent, and can be called attitudinal objects (Moltmann 2003a). Kinds of attitudinal
objects, unlike attitudinal objects themselves, are sharable by different agents. The view that it
is kinds of attitudinal objects that shared, not properties, also corresponds to the view that
expressions such as the same thing are nominalizing quantifiers, ranging over entities the
corresponding nominalization of the verb would refer to, that is, over entities like the belief
that S or entities like John’s belief that S (Moltmann 2003b). I will make use of the notion of
an attitudinal object on two occasions again later.
One of the main advantages of the analysis of generic one-sentences that I have given is
that it can adequately account for faultless disagreement as it arises with generic onesentences, but also in general. The problem with the standard theories of relative truth was
that sentences giving rise to faultless disagreement are not treated as fundamentally different
(in their meaning-intention) from sentences expressing merely subjective opinion. The present
analysis of generic one-sentences takes them to express both a self-ascription of a property
and to draw a generalization on the basis of that. It is clear that the self-ascription part itself
cannot be the source of the disagreement (lets say because the interlocutor refuses to selfapply the content in question). For the account would then not be significantly different from
standard relativist theories, leading to the same problem of distinguishing subjective opinion
from disagreement. Rather the source of the disagreement is the genericity expressed by
generic one-sentences. If a generalization is claimed or believed even if it has a first-personal
source, it will be in conflict with the negation of that claim, even if it comes from a different
first-personal source. Disagreement is faultless because of the first-personal source of the
generalization expressed.
It is also clear why generic one-sentences allow for the possibility of retraction: retraction
will go along with a resetting of the source of the generalization (such as a different
experience), with two different generalizations then being in conflict with each other.
One final and important issue is: is the truth of sentences expressing first-person-based
genericity relative? Certainly it is, but relative truth here should be understood in a different
sense than standard relativist theories would have it. The truth value of a sentence expressing
25
first-person-based genericity is not simply to be relativized to a person, the person that when
self-ascribing the content of the sentence will identify with the individuals quantified over.
This would not capture the fact that the person should have a ‘de se’, not a de re status.
Instead the truth value of a sentence expressing first-person genericity should be relativized to
the entire attitudinal object, an object such as John’s claim that one can see the picture from
the entrance. This attitudinal object itself is linked to the first person experience that is the
source of the generalization. The relativism then is distinct from that of the standard relativist
accounts: it is relativism as regards the epistemic source (irreducible first-person experience)
that is the basis of a generalization, not relativism relative to an external context.
4.3. Generalizing first-person-based genericity
We can now see how the analysis of generic one-sentences can be carried over to sentences
with predicates of personal taste and epistemic modals. There are in fact also linguistic links
between generic one-sentences and sentences with predicates of personal taste, which
independently require carrying the analysis of the one over to the other.
4.3.1. First-person-based genericity and predicates of personal taste
Like generic-one-sentences expressing first-person-based genericity, the application of a
predicate of personal taste is generally grounded in a first-person ascription. That is,
generally, when a speaker utters, for example, white chocolate tastes good, it implies that
white chocolate tastes good to him. However, while predicates of personal taste involve a
self-attribution of the predicate, this is not all there is to predicates of taste. We have seen that
faultless disagreement arises only with truth-directed attitudes and speech acts, not with
purely subjective attitudes as expressed by the verbs find and (certain uses of) think. This
would mean that in truth-directed contexts, sentences with predicates of taste have a generic
interpretation, the kind of generic interpretation that is obligatory for generic-one-sentences in
any context. Truth-directed propositional attitudes and speech acts thus will along with a
generic interpretation of predicates of personal taste, whereas propositional attitudes and
speech acts of individual judgment will go along with a subjective interpretation. Note that as
with generic-one-sentences, a single first-person experience suffices to make the
generalization expressed by a sentence with a predicate of personal taste in a truth-directed
context:
26
(53) John just found out that chocolate tastes good.
Even though the source for the epistemic attitude described in (53) is most likely a single
first-person experience, the content of the attitude is a generalization.
Furthermore, we have already seen that sentences with evaluative predicates allow for
quasi-first-person-based genericity.
4.3.2. Semantic connections between evaluative predicates and generic one
Given that predicates of personal taste have either a subjective or a generic interpretation and
given that the generic interpretation is based on the subjective interpretation, we should first
address the question: how should the subjective interpretation be represented formally?
Should the dependence on an agent be represented by a bound variable or by a shiftable
contextual parameter. While some authors such as Lasersohn (2005) and Stephenson (2006)
treat them as shiftable contextual parameters, there are two important reasons to treat them as
bound variables. One of them is the possibility of the variable being bound by quantified NP,
as in (54):
(54) Everyone finds that chocolate tastes good.
The variable could either be part of an implicit for-phrase or occupy an argument position of
the predicate of personal taste. For present purposes, let us simply augment the relevant
argument position of the predicate. Then the meaning of chocolate tastes good in the context
of (55a) would be as in (55b), whereas in the context of (56a) as in (56b):
(55) a. I find that chocolate tastes good.
b. λx[tastes good(c, x)]
(56) a. I claim that chocolate tastes good.
b. λx[Gn y tastes good(c, qua(y, λz[I z x]))]
Sentences with predicates of personal taste in truth-directed contexts (and without an overt
subject being specified) always express first-person-based genericity, the generalization being
made from a first-personal subjective experience to anyone like the agent in question.
27
The second reason for positing a ‘judge’-variable is the possibility of a link between
generic one and predicates of taste. In fact this possibility is what gives significant plausibility
for the analysis in (56b) in the first place. The relevant examples include those this paper
started out with, that is:
(1) a. It is nice when one is walking in the park.
b. It is nice PROarb to walk in the park.
In (1a, b), the understood judge is the same as, or rather covaries with, the referent of generic
one and arbitrary PRO. This means that both generic one or arbitrary PRO and what is
understood as the ‘implicit judge’ introduce variables that would have to be bound by a single
generic quantifier, as in the analysis of (1a) in (57):12
(57) λx[Gn z nice(^walk in the park(qua(z, λy[I y x]), qua(z, λy[I y x]))]
Also in the contexts below is the implicit judge of predicates of personal taste understood as
covarying with arbitrary PRO or else generic one:
(58) a. It is sometimes more pleasant when one walks home than when one drives.
b. One should walk home because it is so pleasant.
c. When one drinks this with milk, it is delicious.
d. When one is young, rollerblading is lots of fun.
The possibility of a single operator binding both the ‘judge’ variable and the variable
introduced by generic one or arbitrary PRO means that the genericity involved in predicates
of personal taste cannot be a matter of the lexical meaning of those predicates (or of one of
their lexical meanings). Instead it must be tied to the presence of a generic quantifier in the
semantic structure of the sentence which is able to bind other elements as well.
5. Other type 1 expressions
Until now I have focused on intuitions of relative truth with first-person-oriented pronouns
and predicates of personal taste, making use of the notion of first-person-oriented genericity.
28
In what way and to what extent then can the account be carried over to other expressions or
constructions that have been argued involve a notion of relative truth?
First of all, other evaluative predicates in general exhibit the intuitions of relative truth,
that is, moral predicates, aesthetic predicates, and relative adjectives such as tall or rich.13 It is
less obvious in which sense those predicates express genericity based on a first-person
attribution though. Any ‘subjective’ application of an aesthetic or moral predicate obviously
involves considerations of universal applicability. But at the same time, specially with
aesthetic predicates, first-person aspects of meaning (triggering emotions) are obviously
essential as well. With moral predicates, both emotional aspects of meaning and practical
orientation (guidance of actions) require an essential first-person ascription. It may be then
that the meaning of predicates of aesthetic or moral evaluation is just more complex,
involving first-person attributions of possibly different sorts as only part of their meaning.
This may also be said for relative adjective in an evaluative use as discussed by Richard
(2004).
In such a still to be clarified sense then first-person-oriented genericity is involved in the
semantics of evaluative predicates in general. Note that moral predicates also display the
interaction with first-person-oriented pronouns:
(59) a. It is good PROarb to treat others with respect.
b. PROarb to take the exam is obligatory.
Recall that generic one had two uses, one on which the first-person-connection relates to
the epistemic grounds for applying the predicate, the other on which the first-person
connection gives the practical purpose for applying the predicate. These two kinds of firstperson connections, the epistemic and the practical one, are also at play with evaluative
predicates. With predicates of moral evaluation the practical connection will clearly be
involved, whereas with predicates of taste, the epistemic connection will generally be the
relevant one. At the same these generalizations are not strict: predicates of moral evaluation
involve not only hypothetical practical reasoning, but also emotions of various sort (I am
outraged to have to done X, that you have done X). Similarly, predicates of emotional
evaluation will not just involve an inference from a first-person emotional state, but also will
govern future practical reasoning. Even for predicates of personal taste, this may be the case,
as Blackburn (1998, Chap. 1) emphasizes. Given this, we can say that evaluative predicates
29
involve the two directions: inference from the first person (generalizing first-person mental
states or acts states) and inference to the first person (anticipating potential practical
reasoning).
The notion of first-person-based genericity is applicable naturally to epistemic modals as
well. First, we observe that epistemic modals allow the knower to be made explicit, as in (60):
(60) For all I know, John may be in Paris.
Without a specification of the knower, as in (61a), we seem to have a use equivalent to a use
specifying the knower as ‘one’, as in (61b), the only use of epistemic modals that give rise to
faultless disagreement:
(61) a. John may be in Paris.
b. For all one knows, John may be in Paris.
Thus, the analysis of epistemic modals that suggests itself would be one according to which
John may be in Paris could be paraphrased as ‘It is compatible with my epistemic state, as far
as it is generalizable to anyone of relevance, that John is in Paris’.
With epistemic modals, again the question arises how to represent the dependence on an
agent. We can note also here that with a quantificational subject, binding of the first-personoriented content is possible:
(62) Everyone thinks that John may be in Paris.
It is therefore reasonable to assume that epistemic modals involve an implicit argument
position which can be represented by a variable that can be bound by a generic quantifier.
Again using qua objects of the sort ‘y qua being someone the relevant agent identifies with’,
the content of an epistemic modal sentence can be represented as in (63), where M represents
the epistemic modal in question:
(63) λx[Gn y (M(qua(y, λz[I z x]), p))]
Faultless disagreement would now be explained as before: it is disagreement about what from
the point of view of the two agents is the common epistemic state. Again the crucial notion is
30
first-person-based genericity, which in this case means genericity based on first-personal
access to one’s own epistemic state insofar as it is generalizable.14
6. Explaining relative truth intuitions with type 1 and type 2 expressions
The explanation of faultless disagreement with sentences containing predicates of personal
taste will now be the same as for generic one-sentences.
Also the analysis avoids the particular problems that standard relativist theories face. First
of all, the present analysis won’t face the meaning-intention problem: no additional
parameters such as standards of taste are postulated, but simply a first-person attribution of
evaluative and epistemic properties.
The analysis moreover makes the right prediction as to which agent serves as the ‘judge’,
the person on the basis of whose judgment the generalization is made. In independent
contexts, the analysis predicts that the judge is the speaker or whoever the speaker may
identify with. In attitude contexts, the analysis predicts that the judge is the described agent or
whoever the described agent may identify with. Finally, in independent and modal contexts,
the judge will be the speaker or whomever the speaker may identify with.15
The de se problem can be accounted if the notion of an attitudinal object is used. For a
given attitudinal object, the additional parameter of evaluation, the judge, is self-identified by
the relevant agent self-ascribing the property expressed by a first-person-based generic
sentence. This gives the judge (that agent) not a de re, but a de se status, within the attitudinal
object.
How does the notion of truth apply to contents of first-person-based generic sentences
construed as properties? In cases of faultless disagreement, as I have argued in regard to
generic one-sentences, the disagreement arises because of the generality of conflicting claims
that are being made, and the faultlessnees because the epistemic source is first-personal. It is
also important to note the conditions of acceptance of such sentences: accepting a firstperson-based generic sentence can only mean self-applying the property in question (which is
of course also the source of faulty agreement). But this accounts only for the grounds, the
epistemic basis, for the claim being made, not for its truth. For truth, the account seems to
coincide exactly with relativist accounts that posit an agent as part of a context of assessment:
However, here the relativization serves a different purpose: it ensures that the sentence is
epistemically grounded or practically oriented. It is not truth in one context, as opposed to
31
another, but truth being grounded as opposed to not being grounded in a first-person
ascription.
7. Type 2 expressions
Type 2 expressions such as neighboring, right; left differ from type 1 expressions in that they
exhibit only sharing, not faultless disagreement or retraction. Not only for that reason would it
be inappropriate to take them to involve first-person-based genericity. Those expressions in
fact pattern just like singular de se interpreted pronouns.
De se interpreted pronouns obviously exhibit sharing, as seen from the validity of the
inference in (64a), with the two occurrences of he being interpreted de se, as well as the same
inference with (64b) as conclusion:
(64) a. John thinks that he is the winner.
Bill thinks he is the winner.
John and Bill think the same thing.
b. John and Bill share the same thought.
De se interpreted pronouns do not allow for faultless disagreement, or faulty agreement:
Thus, the premises in (64a) could not possibly describe a situation of John and Bill agreeing.
Given these parallelisms between de se interpreted pronouns and type 2 expressions, it is
reasonable to assume that the context-dependency involved is of the same, distinct from the
truth-relativity of evaluative and epistemic sentences. To evaluate as true or false a claim by
John that he is the winner, the evaluator will not self-ascribe the property of being a winner,
but rather apply it to John. Similarly, when evaluating a claim by John that the tree is to the
left, the evaluator will not self-apply the property of an x such that the tree is to the left of x,
but rather apply it to John.
The same difference between type 1 and type 2 expressions manifests itself in conditions
on the acceptance of assertions: Joe’s assertion that he himself is a hero aims at making the
addressee accept that Joe is a hero, not that he, the addressee, is a hero. That is, accepting the
assertion of a sentence with a pronoun X interpreted de se means accepting a content in which
the person making the assertion counts as the semantic value of X.16 By contrast, accepting
Joe’s assertion that it is nice to walk in the park or his assertion that one can see the picture
32
from the entrance can only mean self-applying the property in question.17
Thus, a relativist account of truth for sentences with type 2 expressions would be
inadequate. How then is a truthconditionally complete content to be obtained from a property
as expressed by a sentence with a de se interpreted pronoun? Here recourse can be made to
the corresponding attitudinal object, such as for example in (64a) John’s thought that he is the
winner. As I have argued in Moltmann (2003a), there are reasons to take attitudinal objects,
entities that inherently aim at truth, to be the primary bearers of truth values, with kinds of
attitudinal objects, sentences (relative to a context), and propositions acting as truth bearers
only in a derivative way. If attitudinal objects are the truth bearers, then sentences with de se
interpreted pronouns do not require a notion of relative truth. Attitudinal objects are
furthermore suited as the objects of agreement or disagreement.
Attitudinal objects should of course play exactly the same role for sentences with type 2
expressions. Thus while the museum is to the left expresses a mere property, John’s claim that
the museum is to the left is truthconditionally complete, and it is this object that is the bearer
of the truth value and acts as the object of agreement and disagreement.
The same option, however, is not available for the contents of first-person-based-generic
sentences. Of course, there are same sorts of attitudinal objects, such as John’s claim that one
can see the picture from the entrance or John’s claim that the wine tastes good. However,
when someone else than John evaluates such a claim as true or false, he will still have to selfascribe the property expressed by that one can see the picture from the entrance or that the
wine tastes good. This means with first-person-based generic sentences, there is a genuine
relativity of truth, though of an epistemic, not a contextual sort. First-person-based generic
sentences require for their evaluation as true or false a relativization to an intentional act (an
attitudinal object) on the part of the evaluator himself.
This account of type 1 and type 2 expressions can also shed light on the particular kind of
context-dependency the two types of expressions may involve . Relative truth and the
property of sharing by itself are closely tied to the first person. Relative truth arises because of
first-person-based genericity, that is, because a generalization is made on the basis of an
esssential first-person application of a predicate. Unlike on relativist accounts, this account
predicts that the only relativity of truth there is is relativity with respect to the first person.
This explains why relative truth does not arise with other indexical expressions such as there
or here.18 Sharing by itself arises because the contents in questions are properties and thus
contents of kinds of self-locating attitudes or speech acts, which are the objects shared.
33
8. Conclusion
First-person-oriented genericity is a form of generalization that is associated most explicitly in
English with generic one, but, as I have argued, is more generally associated with the most
important expressions giving rise to intuitions of relative truth. It involves a form of detached
generalizing self-reference and appears to be at the heart of various phenomena that have been
argued to require a notion of relative truth. First-person-based genericity can explain why
certain kinds of context-dependent expressions display intuitions of relative truth and only
those expressions, namely precisely the expressions whose application conditions involve an
essential first-person attribution of a property. It is thus no accident that some contextdependent expressions do not give rise to a relative notion of truth and that conversely
evaluative predicates and epistemic modals do not allow for a contextualist or implicitargument account.
Notes
* I would like to thank Peter Lasersohn, John MacFarlane, Robert Stalnaker for very useful
comments as well as audiences at SPR05 in San Sebastian, at NYU, and at the Ecole Normale
Superieure in Paris for discussion.
1
The use of believe in (5a) suggests that the agent does not himself have the experience. To
get the nonsubjective first-person-based reading, know would be the right verb:
(1) John knows that white chocolate tastes good.
But then faultless disagreement is actually hard to get. It is an interesting question in fact why
that is so, as it does not seem to fit well with the relativist account of faultless disagreement.
The difference between know and believe seems to be a pragmatic one: given that the
sentence with know implies the agents’ own experience, there should be a point in using the
corresponding sentence with believe instead, which will therefore be forced into a weakened
meaning.
The point is more clearly made by the contrast between verbs for expressive and for truthdirected speechacts.
2
It appears that also certain presupposition triggers allow for sharing, for example expressions
like another time, a second time, or again:
(1) John believes that he won in 2006.
Mary believes that she won in 2005.
John and Mary hope to win another / a second time / again.
(2) a. John and Mary hope for the same thing.
34
b. John and Mary have the same hope
Such presuppositional expressions are generally held to be anaphoric to some element in a
preceding textual or mental representation. But obviously, the identity of the antecedent
element does not bear on the identity of the content. This is a problem for standard anaphoric
accounts of such presupposition triggers. The present paper will not try to offer an account
either.
3
Lasersohn (2005) allows the context of evaluation to depend on the speaker taking one
perspective or another, so that the context of evaluation, to an extent, will be a matter of the
speaker’s choice. At the same time, though, Lasersohn emphasizes that this should not be
taken to mean that the context is dependent on the speaker’s intentions. Thus, Lasersohn’s
account may not be a clear case of this version of the relativist account.
4
Stevenson (2006) proposes that when Chocolate tastes good is asserted (as opposed to just
thought), the intended judge is in fact not the speaker, but rather an imagined ‘common
judge’. The common judge is what is needed when a sentence like Chocolate tastes good is to
be made part of the common ground. The problem with making use of the notion of the
common ground to explain faultless disagreement is that this could not be carried over to
faultless disagreement arising even with beliefs.
5
MacFarlane (2003) takes the intuitions of relative truth to require a modification of the
notion of assertion, truth not being the aim of assertion anymore.
6
The examples were chosen with sentential operators because negation in the same sentence
is not expected to take wide scope over generic one.
7
Lasersohn (2005), actually, proposes a third option according to which the belief relation is a
three-place relation between agents, contexts of evaluation, and propositions. This is needed
because for Lasersohn the context of evaluation depends on whose perspective the agent
takes, not on the agent directly.
8
See Castaneda (1966, 1967), Evans (1982), and Perry (1979) on pronouns interpreted de se.
9
For the examples of because-clauses see Egan/Hawthorne/Weatherson (2005); for the
example with questions see Lasersohn (2005). Of course for those authors the examples
support somewhat different accounts.
10
For a linguistically much more detailed discussion of what follows see Moltmann (2006,
ms).
11
There are contexts, however, in which generic one imposes no restriction on the predicate
whatsoever, namely when occurring as a bound variable, as in (1) and in conditionals as in
(2):
(1) a. Sometimes one forgets that one has a nose.
b. One can doubt that one has a soul.
(2) a. If one lives in a big city, one lives in a city.
b. If one has a nose, one can breathe.
35
See Moltmann (2006) for discussion.
12
This covariation is not obligatory in all cases, for example not in:
(1) It is incomprehensible for PROarb [PROarb to have someone like this].
The natural reading of (1) is one on which the speaker generalizes his own evaluation of
incomprehensibility of the kind of behaviour the addressee has engaged in. What we have
here is in fact a first-person based genericity of evaluation of a second-person-based
genericity of acting; in addition to engaging in first-person-based genericity, the speaker also
identifies with the addressee and generalizes (a behaviour) from that point of view. There are
thus two generic quantifiers involved in the semantic structure of the sentence.
13
See Koelbel (2002) for discussion and Richard (2004) in particular for relative adjectives
such as rich.
14
There is one further kind of expression that in the literature has been taken to involve
relative truth, namely future contingents. Here first-person genericity hardly gives further
insights. Rather the relativity here appears tied to relativity of future itself.
More controversially, the verb to know has been argued to lead to relative truth
(MacFarlane 2005b).
15
An account in the same direction is suggested in Koelbel (2003), who proposes that
chocolate tastes good be relativised to a ‘perspective’ quite simply because the conditions on
the possession of the concept tastes good involve a first-person application, which means the
content of chocolate tastes good could not be grasped otherwise than by a self-application.
However, again, unless a perspective is something other than as a centred world, it is not
obvious how the content of chocolate tastes good could be conceived differently on a
standard relativist account.
16
Stalnaker (1981) in fact takes the behaviour of sentences with de se interpreted pronouns to
be grounds for rejecting Lewis’ account of such sentences as expressing properties. Stalnaker
instead takes them to express propositions like any other sentences.
17
Egan (to appear) discusses the same distinction with a sentence like my pants are on fire
and sentences with epistemic modals (see Section 2.4.). He argues that my pants are on fire
cannot be asserted when expressing a self-locating proposition and gives a pragmatic account
of why this is so. It is clear, however, that the distinction is not a pragmatic one but resides
strictly in the formal distinction between type1 and type 2 expressions (expressions like
pronouns interpreted de se as opposed to expressions giving rise to relative truth, see also
Sections 6, 7.).
18
For this point see also Stephenson (2006).
References
Anscombe, E.(1975): ‘The First Person’. In S. Guttenplan (ed.): Mind and Language. The
36
Wolfscombe Lectures. Oxford UP, Oxford.
Blackburn, S. (1998): Ruling Passions. Clarendon Press, Oxford.
Castaneda, H.-N. (1966): ‘He*: A Study in the Logic of Self-Consciousness’. Ratio 8,
130-157.
-------------------- (1967): ‘The Logic of Self-Attributions to Others’. American
Philosophical Quarterly 4, 85-100.
Chierchia, G. (1990): ‘Anaphora and Attitudes De Se’. In R. Bartsch et al. (eds.):
Language in Action. Foris Publications, Dordrecht,
Cohen, A. (2002): ‘Genericity’. Linguistische Berichte 10.
Davies, M. / T. Stone (eds) (1995): Mental Simulation. Evaluations and Applications.
Oxford, Blackwell.
Egan, A. (to appear): ‘Epistemic Modals, Relativism, and Assertion’. Philosophical Studies.
Egan, A. / J. Hawthorne / B. Weatherson (2005): ‘Epistemic Modals in Context’. In G. Preyer
/ P. Peter (eds): Contextualism in Philosophy. Oxford UP, Oxford.
Evans, G. (1982): ‘Self-Identification’. In J. McDowell ed.): The Varieties of Reference.
Oxford UP, Oxford, 205-33.
Fine, K. (1982): ‘Acts, events, and things’. In W. Leinfellner et al. (eds.): Sprache und
Ontologie. Proceedings of the Eighth Wittgensetin Symposion. Hoelder-Pichler-Tempsky,
Vienna, 97-105.
Goldman, A. I. (1989): 'Interpretation Psychologized'. Mind and Language 4, 161-85.
------------------ (1995): ‘In Defence of Simulation Theory’. In Davies and Stone (eds.).
Gordon, R. M. (1986): 'Folkpsychology as Simulation'. Mind and Language 1, 158-71.
------------------ (1995a): ‘The Simulation Theory: Objections and Misconceptions’. In M.
Davies / T. Stone (eds.).
------------------ (1995b): ‘Simulation without Introspection or Inference from me to you’.
In M. Davies / T. Stone (eds.).
Koelbel, Max (2002): Truth without Objectivity. Routledge, London.
---------------- (2003): ‘Faultless Disagreement’. Aristotelian Society 104, pp. 53-73.
Krifka, M, et al. (1995) ‘Genericity. An Introduction’. In G. Carlson / F. Pelletier (eds.):
The Generic Book. Chicago UP, Chicago.
Lasersohn, Peter (2005): ‘Context Dependence, Disagreement, and Predicates of Personal
Taste’. Linguistics and Philosophy 28,6, 643-686.
Lewis, David (1979): ‘Attitudes De Dicto and De Se’. The Philosophical Review 88, 513543.
37
MacFarlane, J. (2003): ‘Future Contingents and Relative Truth’. Philosophical Quarterly
53, pp. 321-36.
------------------ (2005a): ‘Making Sense of Relative Truth’. Aristotelian Society 105, 321-39.
------------------ (2005b): ‘The Assessment Sensitivity of Knowledge Attributions’. In T.
Szabo Gendler / J. Hawthorne (eds.): Oxford Studies in Epistemology. Oxford UP.
----------------- (ms 1): ‘Epistemic Modalities and Relative Truth’.
----------------- (ms 2): ‘Relativism and Disagreement’.
Moltmann, F. (2003a): 'Propositional Attitudes without Propositions'. Synthese 135, 2003,
70-118.
---------------- (2003b): ‘Nominalizing Quantifiers’. Journal of Philosophical Logic 32,
445-481.
--------------- (2006): ‘Generic One, Arbitrary PRO, and the First Person’. Natural
Language Semantics.
---------------- (ms): ‘Generic One: A First Person-Oriented Generic Pronoun’. IHPST, Paris.
Perry, J. (1979): ‘The Problem of the Essential Indexical’. Nous 13, 3-21.
Richard, M. (2004): ‘Contextualism and Relativism’. Philosophical Studies 119, 215-242.
Schiffer, S. (1987): Remnants of Meaning. MIT Press.
------------- (1990): ‘Meaning and Value’. Journal of Philosophy 1990, 602-614.
Stalnaker, R. (1981): ‘Indexical Belief’. Synthese 49.
Stephenson, T. (2006): ‘Assessor-Sensitivity: Epistemic Modals and Predicates of Personal
Taste’. In J. Gajewski et al. (eds.): New Work on Modality. MIT Working Papers in
Linguistics, v. 51.
Williams, B. (1973): ‘Imagination and the Self’. In B. Williams: Problems of the Self.
Cambridge UP, Cambridge, 26-45.