This document details the model which motivates the thinking here. It is in two versions, bottom-up and top-down. They differ only in their order of presentation of the model's definitions. If you want to cut to the chase then work back to the specifics, the top-down version is the one to consult:  bottom-up  top-down

Effectively everything said here about truth, meaning, consciousness and other matters is a consequence of what I am inventively calling 'the model'. The model imagines a minimal world of agents engaged in a simple activity superficially like speaking a language.

It's augmented incrementally until familiar properties of language emerge. Other people have of course done this kind of thing - what I think distinguishes the approach here is the elements it does and does not include, and its scaled-back explanatory ambitions.

As the model is developed, its elements acquire new properties. To keep things clear, these properties are named following a convention - their names all have a leading '_'. The names make plain the real-world properties they're supposed to model. The idea behind a thought experiment such as this is to separate, in principle, the activities

  1. of imagining a world of people engaged in plausible activities and stipulating definitions to describe them (this is hopefully uncontrovesial in itself), and
  2. of mapping those definitions onto familiar concepts (this is where the controversy comes in, potentially).
Ultimately the goal is to get to a point where we are inclined to say, e.g., 'truth is _truth', i.e., that a particular thing in the real world is true just in case the relations in which it stands to other parts of the world mirror the relations in which its corresponding model element stands to corresponding other parts of the model, when the model element is _true.

The model in the first instance imagines a large set of people, which it terms '_agents', milling about in reasonable proximity. They are prone to uttering strings of sounds, '_sentences', which, as it happens, resemble sentences of (say) English - but we are to imagine these strings having no meaning (initially). The model stipulates that _agents reflexively assign them -utterances of _sentences- each a '_value' between 0 and 1. This is sort-of a gut-reaction, yes-no evaluation, with the valuation being intrinsic to the 'token' _sentences. It is formalized in the model by allowing that every agent has a kind-of table or function which associates a number to every possible token _sentence. Keeping to our clever naming practice, this is called the '_value function'.

The crux of the project is here. Whereas in most conventional accounts of language -I think it's fair to say- the goal is to explain sentence evaluations in terms of other things (meanings-grasped or what have you), in the model these are fundamental. They are the given.

The reason things mostly are the usual way and not as in the model is hardly surprising. There are effectively infinitely many token sentences, yet somehow our valuations are (mostly) predictable. The merest sense of intellectual responsibility would seem to compel us to try to produce a theory which generates predictions about what we understand and agree to, in terms of what is said - to be able to explain the reliability of our expectations about sentence valuations based on their words' meanings. But this, I think (as others have thought), actually misapprehends what needs explaining, and the terms of any possible explanation. Considered as a natural phenomenon -as measuable dispositions towards sentences considered as sound-sequences- there is an important and viable, finitary, scientific project here. Neural net modelling, particularly in the last ten years, has made enormous strides in this respect. What is key for present purposes is that such scientific theorizing has no need of meanings or truth - any semantic or intentional concepts. It is a project of natural science proper. Theorizing of this type carries the whole explanatory burden. We are absolved of the obligation to formulate a theory in semantic terms. Which is a good thing, because any such theory is bound to fail.

Anyway. Grasping that the semanticist's assumed explanatory burden is specious, is a hurdle which must be leapt. Once it is, a much clearer view becomes available.

The way it's imagined, the model also needs to distinguish between token _sentences which an _agent has heard or uttered and those s/he has not (an _agent's _value function maps all token _sentences, heard or not). A token _sentence heard or entertained which is _valued is termed a _belief. With just these elements in place, interesting questions arise as to how best to model real language. One consequential realization is that the _value function should be designed so that whether an _agent _values some given token _sentence depends on what other _sentences s/he has already heard and _values (intuitively, what you're inclined to believe depends in part on what you already believe). This complicates the model in ways that are both interesting and realistic.

_Sentence _valuations are meant to be mostly obvious to _agents, but not of any great intrinsic interest - like, say, basic colour recognition. To give _agents a motive to speak, a further property, _pleasure, has to be added to the model. _Pleasure is meant to be some nice feeling distinct from any familiar feeling, which is experienced when certain _valuable _sentences are encountered. It's intended to correlate to the real-world benefits truth facilitates getting, including avoiding familiar pains. It serves in the model as the only motive for exchanging _sentences – for ‘_conversing’. It is a contrivance of the model which ultimately has to be cashed-out when mapping the model onto reality.

With a bit of finessing, it emerges in the model that _agents would naturally want to cooperate to amass as large a collection of valuable _sentences as possible (a bit like how we naturally cooperate collectively to improve our individual financial circumstances, even as we compete). This gives rise to a property which on inspection looks a whole lot like truth in normal talk - the property of belonging in that collection.

With _truth in place it becomes possible to introduce _words, and to define a concept solely in the model's terms which arguably does all the work which we should expect of a concept of word-meaning. And with _word-_meaning in place, we can get straight-forwardly to a concept of _sentence-_meaning or proposition.

Robert Kermode, 2026
(Notes:
  • This explanation follows the leading-'_' convention for the model's terms, explained in 'The model' -> 'The Idea'.
  • The concept of _proposition is assumed without introduction, here. This concept is specified later, in the discussion of _meaning. A _proposition is in effect a set of token _sentences.)
Possibly what distinguishes the model from other treatments is that it represents a speaker's goal in talking to be, not the acquisition of information, but rather simply the maximization for her or himself of some simple good. The details of the model which make this interesting (and, I think, realistic) are that
  1. Some _propositions are significantly rewarding on first hearing. In the model, they are intrinsically _pleasurable.
  2. It is not trivially easy to come up with _pleasurable _propositions.
  3. It costs effectively nothing to repeat a _sentence to someone else -to share its _pleasure, if you like.
  4. The _values _agents assign to new token _sentences depend in part on what _sentences they've already heard and _value (what you're inclined to _believe depends on what you already _believe).
  5. Where _agents have mostly the same prior _beliefs, their _valuations of new _sentences mostly agree. Where their prior _beliefs differ, the previous point specifies, so may their _valuations.
  6. Lastly, the _values _agents assign to _sentences are in part -but only small part- up to them. They have a limited ability to choose whether to _value new token _sentences. It costs effort, however, to change one's _beliefs. _Agents are at least a bit responsible for what they _believe.
These factors would conspire to create an interesting dynamic, in some ways like what governs an economy of goods. On the one hand, _agents would have a strong incentive to share their _sentences (to talk to each other). That is, the inhabitants of the model would naturally evolve an implicit contract whereby they each volunteer their _sentences, provided others do likewise. _Agents' contracting with one-another in this way would evidently result in their individually accruing greater _pleasure than if they did not. They would, in short, be motivated to cooperate.
But they would also have an incentive sometimes to compete. The idea is that the dependency of new _valuations on prior ones (_beliefs) combined with _agents' dependence on each other, make it advantageous to them to try to bring others into harmony with themselves, in the event of disagreement. To the extent that _agent B's _beliefs overlap with _agent A's and they routinely exchange their _sentences in _conversation, B effectively works as a proxy for A in getting new, _pleasurable _sentences as B goes about the wider community, _conversing. To the applicable extent, A will likely _value what B does and so likely get pleasure from B's new discoveries when they meet. When A's _valuation of a new _sentence differs from B's, it is thus in her/his interest to win B over to her/his _valuation, and vice-versa. Each would, in effect, be striving to get the other to work on her/his behalf, rather than having to forfeit her/his _sentences' _values as a cost of staying in general harmony. Furthermore, if A can bring B round to seeing the _value of her/his new _sentence, then A has gained in B a proselytizer of this _belief, as B travels about, _conversing. This increases the probability of A's getting _pleasure in her/his subsequent _linguistic exchanges with the community.
Repeating the point more succinctly, to an _agent ai, other _agents are instrumental in the first instance as sources of new, _pleasurable _sentences. Another _agent's utility to ai increases monotonically with the extent of the other's belief set's overlap with ai's, up to a limit . ai is thus motivated to maximize this overlap. Noting that there is a personal cost to ai in surrendering their own _valuations, where there is a difference, ai is motivated to cajole the other to see the _value in the mooted _sentence as ai does. In winning another _agent over to their _valuation of a _sentence, ai then gets the secondary benefit provided by the other's dissemination of the _sentence to the wider community .
It will be useful in what follows to have a term for the relation between two _sentences when one influences the other's _value. '_Support' can be defined as follows:
_support: token _sentence s2 _supports token _sentence s1 for _agent ai just in case the _value of s1 for ai is greater with s2 in ai's _belief set B than if not, i.e.,
Vi(s1, B ∪ {s2}) > Vi(s1, B \ {s2})
What would this competition look like? Suppose we inhabit the model, and I utter a _sentence in your presence. Other things being equal, the model implies I _value this _sentence. Suppose that you do not _value it -it is not added to your set of _beliefs. For the reasons just given, I now have an incentive to change your _valuation. How do I do this? Well, my _valuing it will very likely be due to the presence of some other, _supporting _sentence in my _belief set. If this latter _sentence is not in your _belief set, then maybe by uttering it I will get you to _believe it, and as a consequence to tip your _valuation of the first mooted _sentence to positive . However, if, as it happens, you _disvalue also this latter, suggested _supporting _sentence, then clearly this process can recur on it.
It's worth emphasizing that this practice would have a particular, practical role in the model. It would be of use only to the extent that _agents often can ultimately come to agreement (or agree to disagree), terminating the recursion.
If my initial mooted _sentence is just intrinsically _valuable, and not substantially _supported by any other _belief -if, that is, it is an '_observation' _sentence- then one thing I can do to try to bring you into alignment with myself is to ensure that your focus of attention on hearing the token _sentence is the same as mine . If this fails, I may be out of luck .
The natural name for the activity is '_justification'.
To date the linguistic elements of the model have solely been sound sequences, _sentences. We can suppose the model to be augmented to include some finite stock of generally shorter sequences, _words, and stipulate that _sentences consistently resolve into sequences of _words .
Among the stock of _words, certain can be supposed to be of special interest: AND, OR, IF, and NOT. Concerning the first three, it can be supposed that, with strong regularity, in _sentences containing them _valued by _agents, they appear flanked by _phrases which themselves are _sentences. The last is, with strong regularity flanked just on the right by a _sentence. It can be stipulated, further, that among _agents, certain _support relations between _sentences containing these _words and their contained flanking _sentences are found with strong regularity to obtain. Specifically,
  • _sentences of the form <s1, AND, s2> are strongly supportive of both s1 and s2, and vice-versa.
  • _sentences of the form <s1, OR, s2> are strongly supportive of s1 or s2, and vice-versa.
  • _sentences of the form <s2, IF, s1> are strongly supportive of <NOT,s1> or s2, and vice-versa.
  • an _agent's _valuing a _sentence of the form <NOT, s> correlates strongly with their _disvaluing s, and vice-versa.
If the support relations are consistent enough, _agents will do well to expect that deviating from the pattern will lead to '_error' - that is, to a set of _beliefs not all of which are _true.
Support relations involving these words will be called '_logical'.
It's reasonable to suppose that a community of people motivated and constrained as in the model would come to attach some importance to _justificatory exchanges as described just above. Standards in the conduct of such exchanges would naturally emerge. Being a competitive activity ultimately aimed at a shared good, the manner of its conduct, and of its participants' willingness to revise (or not), their _beliefs, would be something worth noticing .
The discussion of _justification has emphasized the advancement of individual purposes. The real point of _justification, though, is that one takes the _sentences one is _justifying to be _true - that one takes them properly to be elements of the _maximal communal set, of _value to all. Entering into the _linguistic contract involves recognizing that one ultimately maximizes one's own _value by maximizing collective _value. The worth of an '_interlocutor' in all this would naturally be a function of their usefulness in providing _truths, and only _truths. An _interlocutor found often to offer _false _sentences would not be good to _converse with . An _interlocutor who, in _justifying a _sentence, routinely offers _sentences which on consideration, don't really _support the initial, mooted _sentence, would not be good to _converse with. An _interlocutor who is found not to be _logical in their _beliefs, would not be good to _converse with.
The obvious name for the set of norms governing such exchanges, and for _agents' practices in revising their _belief sets, would be '_rationality'. It is a counterpart of the norm ascribed to the _sentences themselves, '_truth'.
Robert Kermode, 2026
(Note: this explanation follows the leading-'_' convention for the model's terms, explained in 'The model' -> 'The Idea'.)
The model imagines that _sentences are of interest to the people in it, not for any information they may convey, but rather for a certain contrived good ('_pleasure') associated with some of them (_sentences which express novel '_propositions'). _Agents' sole goal in the model is to maximize their personal quotient of this _pleasure.
Because the good can be shared effectively without cost (just by speaking a _sentence to someone), and because the model's _agents are mostly in agreement, it's easy to see their best strategy individually would be to agree mutually to assist one another. They would naturally form an implicit contract whereby each speaks her valuable _sentences to others on the understanding that they will do likewise.
Crucially, the model stipulates further that _agents' _valuations of newly encountered _sentences depend in part on what they have encountered and _valued in the past, and also that _agents have some lattitude to re-_value _sentences --just a bit. The point of this wiggle room is that if _sentence _valuations are interdependent as stipulated, the possibility will arise of a newly encountered _sentence being in conflict with a big prior collection. The project of _value maximization is best facilitated by allowing a bit of tolerance with respect to individual _sentences, so as not to oblige abandonment of what may after all prove to be a sound collection. _Value is in a way cumulative for _agents, and they do best in the project of getting it if they're allowed a modicum of discretion. If _valued _sentences are understood as beliefs, these stipulations hopefully won't feel too exotic.
_Valuations being thus interdependent leads to a first interesting property of _sentences. We can suppose there to be some one set of them which would maximize overall _value for any given _agent, were she to value all and only its elements . If, as we have said, her goal in trading _sentences is to get as much _value as she can, then she is ultimately looking to _value all and only the elements of this set. At any given time, however, since the aforementioned wiggle room opens the door to _valuing _sentences which aren't really part of the set, she may _value _sentences which in fact are not in it. For any given _sentence, _valued or not, then, there will be the question as to whether it is in the set. Clearly the property of being in that set is of interest, and deserves a name. We will say a _sentence is _true for _agent a just in case it's in _agent a's set, and _false for a if not.
'_truth for a' is of limited interest. For one thing, an _agent has no way of knowing what is in that set, apart from what she _values at any given moment. So the distinction between what she actually _values and what she ought to, to maximize her total _value, would not hold any interest for her. For another, two additional stipulations of the model, that novel _valuable _sentences are hard to come up with but very easy to share, and that people mostly agree in their _valuations, imply that an _agent will be getting most of her _valuable _sentences from exchanges with others. This suggests that the parallel question about collective _value maximization may hold more interest.
Suppose _agent a _values _sentence s and utters it in the presence of b who, as it happens, _disvalues it. As things stand, we have four possibilities.
First, it may be that s is genuinely in a's maximizing set but not in b's - their maximizing sets simply disagree, and s is _true for a and _false for b.
More interestingly, it may be that b is mistaken about s - that in fact his maximal set agrees with a's, but he doesn't realize it. So s is _true for both a and b - a is right and b is mistaken.
Or, vice versa - it may be that s is _false for both - a is mistaken and b is right.
(The last, not so interesting case of course is that a and b's sets diverge on s, but they're both mistaken (s is _false for a and _true for b)).
Given these possibilities, what ought _agents to do when confronted with such differences? In particular, should they treat what is _true for one _agent to be _true for all? If _truth is just _truth-for-a –if it is merely '_idiolectic'– then for every disagreement between two _agents about a token _sentence either there is a genuine divergence between them as to what is _true in their respective _idiolects, or their _idiolects properly agree and one or other of them is simply mistaken about what is _true for him. It becomes apparent that the best strategy for _agents individually to maximize _value is to treat all differences as signalling a mistake for one or other _agent. That is, it emerges that it is quantifiably advantageous to treat _truth as objective– provided enough _agents opt-in to this strategy and assuming, as we have, both that _agents mostly coincide in their appraisals of _value and that appraisals of _value are at least in part malleable.
To see this, suppose that an _agent disagrees with the majority about some widely _valued _sentence. She then forfeits the _value (and attendant possible _pleasure) potentially of all _sentences whose _support depends on it. These may be considerable in number. If she disagrees with the majority about even a small fraction of all _sentences, she loses not just the _pleasure of the _sentences hierarchically supported by them but also the confidence that _theory _sentences offered by others will rest on foundations which she would _value, and hence, in many cases, the possibility of _valuing _sentences on the strength of others' ‘_testimony’. Admittedly, this _agent will get some compensation in the form of hierarchically dependent _sentences she contrives herself on the basis of her _idiolectic _sentences. As even the most prolific of solitary _sentence producers will fall far short of what her community collectively can contrive, however, and since she will be deprived of the initial error-check afforded by communal acceptance, she will on balance be considerably less well-off. The extra _value to be gained by an individual by participating in a community of _agents in which effectively everyone commits to and contributes to the construction of a single shared edifice of _truths will easily offset the cost of sacrificing the _pleasure of any genuinely _idiolectic _sentential affinities.
And so we have in the model a concept of _truth: A _sentence is _true, roughly, just in case it is an element of the set of _sentences which would maximize the combined _value of all _speakers, were they all to _evaluate all of its elements. This being impossible, it stands merely as an ideal. I will note for now just two points.
The first point is just a reminder that the definitions so far are mere stipulations about hypothetical people in hypothetical circumstances. It is fair to ask whether such people are possible, and whether, if they existed, they would behave as described. These are not philosophically loaded questions -should not be, anyway. What is of course loaded is the fitness of the stipulated concepts to model their real counterparts. My claim, evidently, is not just that such people are possible but that we in fact are them, and our usual concepts, the model's.
The model recognizes a superficially truth-like ideolectic property, '_truth-for-an-_agent'. A committed defender of indivualistic rationality might plausibly cleave to this in the hope of resisting the depredations of the current approach. The second point is that this is a mistake on two fronts, I think. First, it makes what is true effectively unknowable, and hence apparently without practical significance. Second, though, I think it fails to see, what the model conveys, that the concept of _truth (for all) has a character substantially different than that of _truth-for-a -that it would play an importantly different role in the lives of _agents- and that this character correctly maps onto our familiar concept of negotiated, public truth.
Robert Kermode, 2026

(Note: this explanation follows the leading-'_' convention for the model's terms, explained in 'The model' -> 'The Idea'.)
So far we have created a simple model of a natural language like English, and seen how its minimal elements would lead to both cooperative and competitive interactions among its speakers as they pursue their individual ends. We have seen, too, that this dynamic would naturally be described in terms of an ideal which on consideration looks a lot like truth. We are now in a position to appreciate how concepts emerge in the model which correlate to our familiar concepts of meaning and proposition.
For perspective, a usual project of semantics is to give an account of the meanings of words and phrases, and to build out from them to an account of the propositions expressed by sentences, and in turn the truth of things said. The present project is the inverse of this. The goal is to show that, being afforded an account of sentence truth formulated in terms of valuations of token sentences (the given), we have everything we need to explicate the semantics of words. We are accustomed to thinking that the references and possibly senses of words do theoretically informative, explanatory work, and likewise the propositions expressed by sentences. This is not the case in the following.
The concept of _truth in the model was arrived at, notably, without assuming _sentences to be composed of _words. _Sentences to date have been just sound patterns of a kind. We can now add _words to the model, understanding them to be just smaller, inert sound sequences. We'll take there to be a large, finite set of them, and stipulate that a _sentence is always decomposable into some sequence of them. The path to their _meanings is a bit tortuous, but I think at each point, locally straight-forward. I will accumulate some promisory notes along the way, which I will return to at the end.
The first point to notice is that we will actually be concerned with _phrases, not specifically _words, a _phrase being understood to be any word-sequence .
The second point is that our primary concept is not of _meaning as such, but rather of sameness of _meaning -specifically, sameness of _meaning of two token _phrases. The _meaning of a token _phrase p will then be identified with the set of all token _phrases having the same _meaning as p .
The first undertaking is to explicate sameness of _meaning of token _phrases at a moment of speech. If attention is narrowed to the interval needed to utter just a given token _phrase, it becomes instructive to investigate the token _sentences (uttered and un-uttered) then and there _true. For simplicity, we will stipulate initially that there are in the model no _phrases which create '_opaque' contexts, and limit our attention to '_atomic' _sentences. These restrictions will have to be revisited. A _moment, m, is specified to be the time, location and focus of attention of an _agent when a _phrase p is uttered, and the _sentential _moment set of p at m will be the set of all token _sentences containing p which are _true at m:
_moment: a _moment m is a composite <t, x, ξ>, where t is a time, x is a position in three-dimensional space corresponding to an _agent's location, and ξ is a position in three-dimensional space corresponding to the _agent's focus of attention .
_sentential _moment set (_s-moment set) of a _phrase at a _moment: the _sentential _moment set of p at m is the set of all and only the (_atomic) _sentences containing p which are _true at m.
A _sentential function of p can be specified as the construct got from a _sentence by replacing (one token of) p from a _true _sentence containing it with a placeholder variable, and a _moment set of p at m can be specified to be a set got by forming a _sentential function from each of the elements of the _sentential _moment set of p at m.
_sentential function: a _sentential function of p is a construct got from a _sentence containing one or more tokens of p by replacing one of the tokens with a placeholder variable.
_moment set of p at m: a _moment set of a _phrase p at a _moment m is a set of _sentential functions got by transforming each element of the _sentential _moment set of p at m into a _sentential function.
With these definitions in place, we get that two _phrases p1 and p2 have the same _meaning at a _moment m iff they have a common _moment set .
sameness of _meaning at a _moment: _phrases p1 and p2 have the same _meaning at a _moment m iff there is a _moment set ms1 of p1 and a _moment set ms2 of p2 and ms1 and ms2 are identical.
_meaning at a _moment: The _meaning of a _phrase p at a _moment m is the set of all _phrases with the same _meaning at the _moment as p.
You and I may know of, and have recently been discussing, two dogs, both named 'Fido'. Let's posit that one is a mutt, the other a pure-bred Dachshund. The thought here is that at the very moment I utter 'Fido' as I say 'Fido is obedient.', a set of true token 'Fido'-containing sentences gets picked-out, all of which are true of (let's say) the mutt, but not all of which will be true of the Dachshund. Normal, informed speakers would not, at that very moment, value 'Fido is a Dachshund' (it would be false). In other words, the narrowness of a moment of speech completely disambiguates which Fido is in question. Sentence valuation disambiguates words, rather than some word-associated property disambiguating sentence valuation.
So: suppose our token _sentence s is
  • "Abe's dog is a mutt".
By the terms of our definition, what would it take to settle that "Fido" means (refers to) the same thing (animal) as "Abe's dog"?
Well, at the _moment, there will be a bunch of other _true token "Abe's dog" _sentences:
  • "Abe's dog is on the couch"
  • "Abe's dog is sleeping"
  • "Abe's dog suffers from arthritis"
  • etc..
Our requirement is that the token _sentences,
  • "Fido is a mutt"
  • "Fido is on the couch"
  • "Fido is sleeping"
  • "Fido suffers from arthritis"
  • etc..
all be _true as well, and that there be no additional _true "Fido" _sentences.
What has been provided-for so far is only synonomy-in-a-_moment -resolution of the question as to whether two token _phrases at some one _moment of speech have the same _meaning. However: any half-way adequate account of meaning has to have something to say about synonomy across different _moments (in my jargon). So, yesterday, I said (truly) "Fido is on the rug." (and maybe, "Fido is not on the couch."); today I say, "Fido is on the couch." (and "Fido is not on the rug."). Clearly it has to be possible that my uses of "Fido" in the two moments have the same meaning -that I was talking about one and the same dog at the two times. So far, nothing has been said about how in the model this can be made sense of. The evident complication here is that the sets of _sentences _true at the _moments may be different, because, after all, things change over time.
In English, it is possible to convey that a sentence reports what was or will be the case at some other point in time and space by formulating it in the historical present tense and explicitly prepending to it time and place indices. So, for example,
s1: Fido was on the rug.
where the _moment of utterance is understood to be after June 12, 2026, might be equivalent to,
s2: June 12, 2026, at 999 Mongrel St, Somecity, Somewhere, Fido is on the rug.
Such specifications can, of course, refer to the present _moment.
Remembering that the model is in the first instance an arbitrary contrivance, it can be stipulated that it should include _words correlating to the words of English used to specify time and place in this way. It can be stipulated, too, that of the very many token _sentences _valued by _agents, only the subset which are of this form should be considered for determination of sameness of _meaning. I'll call such _sentences, their corresponding _sentential functions and the _moment sets of these, 'indexed'. With these stipulations in place, we can now posit -with one qualification to come- that the identity of indexed _moment sets is a necessary and sufficient condition for _phrases to have the same _meaning (punkt - not just at a single _moment).
sameness of _meaning (preliminary): token _phrases p1 and p2 have the same _meaning iff the _indexed _moment sets of p1 and p2 are identical (to be qualified).
_meaning: The _meaning of a token _phrase p is the set of _phrases whose _indexed _moment sets are identical to the _indexed _moment set of p.
Suppose I have two rugs, rug1 and rug2, both of which, in different speech contexts, I may refer to as 'the rug'. Today, Fido is on rug2. Yesterday, Fido was on rug1. Helping ourselves to the _word 'not' in the model for clarity, we might have the _sentential _moment sets,
Sunday (_moment fixes _meaning of 'rug' to be rug1):
{ "Monday, June 22, 2026, Fido is not on the rug." (rug1), //a claim about the future
  "Sunday, June 21, 2026, Fido is on the rug." (rug1),
...
}
Monday (_moment fixes _meaning of 'rug' to be rug2):
{ "Monday, June 22, 2026, Fido is on the rug." (rug2),
  "Sunday, June 21, 2026, Fido is not on the rug." (rug2), //a claim about the past
...
}
The ambiguity of 'rug' entails that the _moment sets will be different. The complicating consequence is that the definition, as-is, will incorrectly imply that the meaning of "Fido" at the two _moments is different.
The potential ambiguity of the _phrases in the _sentential functions of _meaning-determining _moment sets raises a problem. Intuitively, what is needed is to insist that only _sentential functions composed of _phrases which are not ambiguous between the two applicable _moments, be included in the comparison sets. In selecting eligible _sentential functions for the _moment sets, we need to recur in the _meaning-determination process and compare their _phrases' _meanings, and then exclude from consideration any mismatches. There is, however, an apparent hitch with this proposal, in that it requires using sameness of _meaning to explicate 'same _meaning' -we seem to have embarked on an infinite regress. It seems we are being pulled back to the primacy of the semantics of words.
I propose that this apparent problem of a looming regress is not a real problem. Our project, again, is not to stand-up a concept of _meaning fit to do substantive theoretical work, but only to show that the model has the resources fully to capture our normal talk of meanings . For this latter purpose, it is admissable to assume sameness of _meaning of orthographically identical _words by default. In comparing _moment sets, only mismatches need be compared for sameness of _meaning of their constituent _words, and cases of _ambiguity purged. Comparing my Fido _sentences Sunday and Monday, I note the very substantial overlap, and satisfy myself that the few mismatched _sentences owe to _ambiguity. Amending the definition, we get,
sameness of _meaning (revised): token _phrases p1 and p2 have the same _meaning iff the _indexed _moment sets of p1 and p2 are identical once any _sentential functions found to contain _phrases ambiguous between the sets are purged.
For clarity, it's worth repeating that the matter of how speakers come to know or be justified in believing that two words have the same meaning, is not in question here - this is not the model's concern.
The model's treatment of '_propositions' is straight-forward, once _meanings are in place. Any token _sentence can be partitioned into _phrases, potentially in more ways than one. To each partitioning, there corresponds an ordered set of _meanings. Expressing the same _proposition is just a matter of sharing a common ordered set of _meanings:
sameness of _proposition: _sentences s1 and s2 express the same _proposition just in case there are partitionings of s1 and s2 into _phrases such that the ordered sets of _meanings corresponding to the partitionings are identical.
_proposition: An ordered set of _meanings corresponding to a partitioning of a _sentence into _phrases is a _proposition
Is there a potential problem lurking here, in the absence of anything substantial to say about 'intended' parsings? I don't think so. The threatening thought is that there might be two token _sentences, the second of which admits of two parsings. The first of these parsings would align in its _meanings with the first _sentence's, the second would not, and the second -we would want to say- is the intended _parsing. Such a scenario would imply the model lacks the information to resolve token _sentences into _propositions, and so is fundamentally flawed.
What precludes such scenarios is _moments' determinations of _word-meanings. If there is a _parsing expressing a different _proposition, it will do so in virtue of its constituent _words' different _meanings, which preempts the problem.
There's a lot to say about _meanings, thus conceived. Here are just four thoughts, quickly noted.
Sense and Reference
The model does not immediately make the familiar distinction between the sense and reference of a term (though see below). A main value of the distinction in familiar theories is to permit accounting for the cognitive difference between sentences which, in those theories' terms, have the same truth conditions. So, in Frege's familiar example, a person may believe
s3: Hesperus is Hesperus.
but fail to believe the equivalent,
s4: Hesperus is Phosphorus.
... because they don't know that 'Hesperus' and 'Phosphorus' refer to the same thing.
As I have stressed very likely too many times, the model's position is that it is not the job of philosophy to explain cognitive phenomena. Science can tell us what's afoot in this person's brain in a case like this (without assistance from the vocabulary of semantics). And common sense tells us that, yeah, they don't know that 'Hespersus' and 'Phosphorus' are names for the same thing, so you wouldn't expect them to believe s4. And that's everything. The model gives us an account of meaning, justification and rationality which implies this person's beliefs may be fully rational - there's nothing more to explain.
Fictional entities
The model depends on _true token _sentences to fix _phrase _meanings. If sentences making claims about non-existent individuals are false, as a theory might reasonably hold, then fictional individuals' names are deprived of meaning. If sustained in the model, the activity of reading novels would be a bit of a mystery.
The model allows that claims about non-existent individuals may be true. Consider,
s5: Dr. Watson was Holmes's friend.
s6: Dr. Jekyll was Holmes's friend.
There is surely some difference to mark between s5 and s6. Plausibly, it might make the difference between passing and failing a high-school English test. The model's _value function implies the first is _true, the second, _false. Additionally, though, the model has the resources to distinguish referring and non-referring terms. Noticing that the _value function has parameters for focus of attention and for _beliefs, it becomes possible to introduce _observation _sentences:
_observation _sentence: a _sentence s is an _observation _sentence just in case the _value of s is dependent on the focus of attention parameter for _agents (ξ) and largely independent of the _beliefs parameter (B) .
With this in place, we see that a set can be constituted of the space-time points defined by the time and focus of attention parameters of all and only the _moments at which _observation _sentences containing a name for some individual are _true. This set, we can stipulate, always constitutes a continuously differentiable curve in the model.
Because there will be no _true _observation _sentence containing 'Dr. Watson' (having the same _meaning as 'Dr. Watson' in s5), there will be no such curve. 'Dr. Watson', the model may hold, is not a '_referring' term. He does not '_exist'.
Externalism
Again perfunctorily, I note that the model sustains externalism, although without a causal theory of reference. '_True' is defined in the model, we have said, in terms of what would maximize _value for an entire _linguistic community. Individual _agents contract with that community to contribute to its common _value, with corresponding benefits. This contract is the source of the norms of _speech, of _meaning, and ultimately of _rationality. An individual teleported unbeknownst from their community to a different but apparently _linguistically and phenomenally identical community, intuitively would not ispo facto have their contract, so to speak, immediately transferred to the new _linguistic community -they would remain contracted to the community which for all those years afforded them _linguistic _value. The _truth values of their token _sentences, and the _meanings of their token _phrases, accordingly, would remain what they were prior to the teleportation. Confronted with something phenomenally identical but constitutionally different to what in their old world they used to call 'water', and also locally called 'water', their _sentence 'This is water.' would be _false owing to those _semantics-sustaining normative ties.
Default meanings and written text
A final related point under this heading concerns default _meanings. It's plausible that in some cases, one _meaning of a given _phrase will come clearly to dominate in size others associated with the same _phrase. The _phrase type 'Obama' might have associated with its many tokens, many _meanings, but one overwhelmingly dominant. In such cases, the dominant _meaning can be assumed to be the default, permitting economies in context-setting. In this way, a token _sentence like
s8: Obama gave a speech yesterday.
...can be _evaluable even absent any conversational context. The model has resources, moreover, such as the _support relation afforded by the _beliefs parameter to the _value function, to tell a more sophisticated story about how we are able to bootstrap our way into conversations without stage-setting. This point is important to understanding the fundamental topic of the _semantics of written _text, whose tokens evidently have properties different than spoken tokens.
A main take-away here is that the model holds context-bound _meanings to be primary and default _meanings secondary, rather than vice-versa as may be the case in other theories.
As with _meanings just above, here are some quick thoughts, raising more questions than they answer.
_Propositions and _truth
As I have belaboured, token _sentences are the primary bearers of _truth in the model. _Propositions are, in effect, just sets of these. This means that a single _proposition may be _true at one _moment and _false at another. In the first instance, _agents _believe token _sentences - they _believe _propositions only by extension, where all (relevant) tokens constituting the _sentence are _valuable (presumed _true). See the discussion of opacity just below, for further considerations.
Species of _proposition?
There is scope to make distinctions among types of _proposition. For example, there are commonly understood to be at least two constraints on the propositions which figure in logical inferences. First, the phrases appearing in successive premises must have consistent meanings. Ambiguity in phrase meaning from one premise to the next evidently vitiates an inference. This may push us in the direction of insisting that the relata of logical inferences are propositions, not mere sentences. Secondly, however, the phrases appearing in successive premises must -we may think- be the same (orthographically). Patterns of logical inference, in the usual way of thinking, depend on this. This may push us to hold that the relata of logical inferences are sentences, not propositions.
The model provides the machinery to define subsets of _propositions which resolve this tension. An 'orthographic _proposition' could be defined as a set of token _sentences which express the same _proposition, all of which are orthographically identical . Formulating the claims of logic in terms of orthographic _propositions would settle this conflict.
Opaque contexts
Consider,
s9: Angelina believes Bob Dylan has a mellifluous voice.
s10: It's not the case that Angelina believes Robert Zimmerman has a mellifluous voice.
It's possible for both these sentences to be true, even though, we might want to say, in some sense, 'Bob Dylan' and 'Robert Zimmerman' have the same meaning. The problem, familiarly, is introduced by the opacity of the context introduced by 'believes'.
This is a big topic. So far, the model tells us the _meanings of _words in '_transparent' contexts, but does not inform their _meanings in '_opaque' contexts. A few points about this:
  • This is not the huge obstacle for the model which it would be for more familiar theories. The model does not purport to inform how it is that people understand what they do, when they hear a sentence. The account of truth is already securely in place.
  • It's worth noting the model has no trouble treating such _sentences as ambiguous between 'de _dicto' and 'de _re' readings. This is just a question of _sentence _valuation in context .
  • The natural thought about _opaque contexts is that what matters is not whether substituting one _phrase for another preserves the _truth of the embedded _sentence, but whether it preserves _value for the individual in question. Intuitively, we want to relativize _meaning in such contexts to the individuals in question. The problem here for the model is that we are pulled in two directions about those individuals.
    • For the model to come out right, the individuals in question need to be people within it, named by the _language of the model and with spatio-temporal curves associated with their names, as discussed above, and so not _agents .
    • On the other hand, we would like to be able to lean on our concepts of _valuation and _belief to describe those individuals for the purposes of theory. These, however, are properties of _agents specified in the definition of the model -not of individuals named within the model.
In short, the problem encourages us to identify _agents with individuals named within the model, which is inadmissable. To do this is in effect to conflate object-language and meta-language.
The solution to an important part of this problem will involve saying more about _opaque contexts, formalizing the concept of _reference suggested above, and introducing a concept of '_intentional individual' for the _referents of the names in _sentences which introduce such contexts. With all of this in place, the concept of a _moment set can be relativized to _intentional individuals, and individual-_meaning then specified in terms of such relativized _moment sets just as it is in their unqualified versions. We get to the individual-_meanings of Angelina's _phrases, for example, by looking at the relevant substitutions which preserve the _truth of _sentences of the form 'Angelina believes _', 'Angelina hopes _', etc..
It may be noted that none of this says anything about the _opaque contexts introduced by modal terms. Elaborating the understanding which the model recommends of these is another undertaking I defer.
The discussion above has relied on concepts of _sentence _atomicity and the potential _opacity of _sentences' internal contexts. But it has said nothing about how these are to be identified.
_Atomicity
We can specify an _atomic _sentence to be one, none of whose constituent _phrases is itself a _sentence. This, however, raises the question as to what exactly distinguishes a _phrase as a _sentence. The intuitive thought is that it should be a candidate for _truth or _falsity -roughly, that _agents' _value functions would generally assign it something different than 0.5. The problem here of course is that some things which we would want to count as token _sentences should generally have _value = 0.5, namely, _sentences about which _agents are agnostic. The solution, I think, is that for a sound sequence seq to qualify as a _sentence, there must some possible _moment and _value of the _beliefs set parameter B which would generally shift the _value of seq to something other than 0.5 .
_Opacity
As discussed, the account requires that some _phrases in a whole range of non-_atomic _sentences get their own treatment of _meaning, viz., _phrases in _opaque contexts. What, then, marks a _sentential context as _opaque?
The idea of the model is to stipulate as small and simplified a world as possible which gives rise to all of the properties of semantics. For this purpose, it is enough simply to stipulate the existence in its '_vocabulary' of a subset of _phrases whose sentential contexts are _opaque: 'believes', 'hopes', etc., and also 'It is necessary that' and 'It is possible that'.
I posit that this is enough. To try to give a systematic explanation in the vocabulary of _meanings and _propositions of how it is that _agents recognize _opaque contexts as such and treat their _semantics differently, would be to lapse back into the explanatory project we're looking to escape .
Robert Kermode, 2026
(Note: for reasons of simplicity, this explanation mostly drops the leading-'_' convention for the model's terms, used elsewhere).
One particularly interesting consequence of the model is that it represents, without any substantial auxilliary assumptions, sentences we would use to express conscious experience or to report dreams, and permits us to characterise their semantics. Doing this, then descending semantically, helps us to get a grip on the subject matter in question.
The elements of the model were nothing more than a set of speakers or '_agents', a set of '_sentences', a set of functions mapping _sentence utterances to '_values' (one function for each _agent), and a '_pleasure' property associated with some sentences' being _valuable. The model's contribution is to make plain the minimal parameters needed to give rise to convincing concepts of truth and meaning in language. These are,
  1. sentence uttered, being just a sequence of sounds
  2. time of utterance
  3. place of utterance
  4. place of hearer's focus of attention
  5. context of utterance, being the set of token sentences recently heard by the hearer
  6. sentence utterer
  7. full set of encountered token sentences valued by the hearer ('beliefs')
Philosophers traditionally have distinguished beliefs based on sensory experience (the candle in front of me is lit) and those arrived at through rational reflection (17 + 8 = 25). Acknowledging the vagueness of this distinction but otherwise ignoring, for now, the voluminous discussion of the subject in philosophy, the two classes of token sentences corresponding to these two types of belief can be specified in the model, like so:

A token sentence s is an observation sentence for a just in case her valuing of s is (relatively) dependent on her focus of attention and independent of her belief set B. That is, the value function returns the same value for s regardless (almost) of the B parameter but varies with the focus of attention parameter.

A token sentence s is a theory sentence for a just in case a's valuing of s is dependent on B.

In articulating these definitions, what becomes apparent is that there are more distinctions to be made.
The point I want to get to is that the model permits us to isolate a class of sentences relevant to the present subject. These are token sentences which
  1. are positively valued only when utterer = agent valuing
  2. are independent of context
  3. are independent of beliefs
We can stipulate in the model that this class is non-empty - that in fact all agents value some such sentences. Let's label sentences in this class, 'Q-sentences'.
Our thought is to get clear on the semantics of such sentences, and in so doing to shed light on what they purport to be about.
So, first, are such sentences candidates for being true? the model's initially frustrating but ultimately illuminating answer is, in a way, yes, and in a way, no. Recall that the model tells us,
Truth: A token sentence is true just in case it's an element of a set which would maximize the combined aggregate value of all speakers, and which is such that its removal would result in a set of lower combined aggregate value.
By this definition, Q-sentences are indeed true or false as the case may be. Adding valued such sentences to the set does increase total value, even if they are valued only by one person.
However: they are in a sense degenerate, and so not fully-fledged peers of true sentences of other classes. As the elaboration of the concepts relevant to truth in the model made plain, its interest is its role as a norm of conversation. The purpose of conversation is, for each of us, to maximize our individual quotient of value, our effectiveness in doing which vastly increases with increasing conversational participants. Conversation is regulated by a contract which binds participants to strive to maximize collective sentential value, other things being equal; truth is what does this. Adding a new token sentence to your belief set is expected to expand your facility in aquiring others.
Q-sentences, being directly valued solely by their utterer and independent of the belief parameter to the value function, 'B', are disengaged from this activity. My learning that your utterance of 'I dreamt last night that there was a blue armadillo on the dresser.' is true increases my net sentential value only by the default amount I accord your sentences on account of trust - the sentence does precisely nothing else for me. Were you to value the denial of this sentence - were the denial true in the degenerate sense we are allowing such sentences to be- it would make no difference at all to anyone. Indeed, we could posit a more restrictive definition of truth which excluded such sentences altogether, one which would do all the work expected of a concept of truth, though at a cost of simplicity.
The model introduced the contrived concept of '_pleasure', to serve as a motive for people to converse. It was meant to function as a stand-in for all of the mundane benefits which the learning of truths affords. The apposite question is whether Q-sentences ever afford _pleasure —that is, whether they ever can provide the kind of mundane benefits, the getting of which is the genesis of truth and meaning. Let's divide this into two questions:
  • Do Q-sentences ever provide benefit to their hearers?
  • Do Q-sentences ever provide benefit to those who speak them?
The answer to the first question is straight-forwardly, 'No', which is the crux of the matter. Q-sentences' being removed from the benefits provided by truth maximization is what distinguishes them . What about the second question? I think the answer here, too, is 'No', but it is less clear. The point of the model, in any case, is that any benefits such sentences might afford would be disengaged from anything to do with the semantic properties of language .
Q-sentences are true, then, but only in this qualified, degenerate sense. What now about the meanings of their relevant words and phrases? Recall that the model manages solely with a concept of sameness of meaning, and that this is effectively a matter of truth-preservation in context of utterance. Two token words have the same meaning just in case one can be swapped for the other in all (atomic) sentences true at the moment without changing the sentences' truth values.
With this in mind, consider two relevantly true token sentences,

s1, uttered at 1:00PM: The pain sensation in my toe is now throbbing.

s2, uttered at 2:00PM: The pain sensation in my toe is now dull and constant.

Is it possible that the two tokens of 'the pain sensation in my toe' mean -refer to- the same thing? I think this is an interesting question for any theory of sensation talk, but we can set aside any general concerns here about object identity as they apply to sensations. Let us allow that by the terms of the model, they do. The interesting point is that insofar as sameness of meaning is wholly a matter of preservation of truth, and the truth being preserved is degenerate in the way just discussed, the sense in which these tokens have the same meaning is similarly degenerate. The meanings of expressions used in Q-sentences inherit the degeneracy of their containing sentences' truth.
The significance of all this lies in the contrast with the sentences of science. The hallmark of science is objective verifiability. A scientific sentence is true or false, as the case may be, in precisely the way Q-sentences are not: believing a scientific sentence, and so having it included in the B parameter to one's value function, implies there will be other token sentences whose value will be different than had the sentence not been included. The propositions of natural science are characterised by the reliability of their connections to theory and experience and so, in the terms of the model, by their support connections to other sentences. All of this being the case, we should expect Q-sentences to be inescapably opaque to science.
Q-sentences, I submit, correctly model qualia reports such as interest theoreticians seeking an understanding of consciousness. Q-sentences have enough to them for us possibly to take interest in them —to treat them as more than mere jabberwocky-nonsense— but not enough to make them worthy of public discussion; not enough for science to be able to get a grip on them. The problem of consciousness arises because we fail to grasp the differences between the two classes of sentence. We treat Q-sentences as though they are full-fledged peers of the sentences of science, and so expect them to be worthy participants in scientific theory. This is a mistake.
It may be objected that this discussion completely omits consideration of what may be thought to be the crucial matter, which is why the inhabitants of the model would value what I've labelled 'Q-sentences', and why we are apt to 'hold true' our real counterpart sentences. I hold true "I have a pain sensation in my toe" precisely because I feel it, because of the qualia! What about the model's agents? It's all very well to posit these Q-sentence place-holders in the model, what matters is what would explain their valuations. That is the subject matter, properly-speaking, of investigations into consciousness, and it's where this treatment completely whiffs. So it may be objected.
I hope it will be clear on consideration that the conception of the reference of 'a pain sensation in my toe' presupposed by this objection simply begs the question against what I am proposing. To try to buttress my position, I will say three things.
A first response, which requires no commitment to the model, is simply to shift the burden regarding the mooted explanatory lacuna. What mode of explanation, exactly, is in question? If it's meant to be a causal 'because', then the objector has the very hard work ahead of her of explaining (A) how purely subjective (objectively undetectable) qualia can be shoe-horned into a scientifically respectable framework, and -what is maybe even more challenging- (B) how the subjective 'I' which experiences them is meant to be made scientifically intelligible; to make good the claim that it (the 'I') is physically changed in being causally impacted by the qualia. What are the physical boundaries and constitution of the self? Or, on the flip side, how does a properly scientific treatment of the subject avoid collapsing down to concepts in whose terms the crucial subjective element just evaporates? One allure of the approach recommended here is that it completely dismantles these apparently intractable questions. Alternately, if we're meant to be understanding a justifying 'because' in the objection, then the objector seems to be equally at sea. How exactly does the justification work? Wittgenstein's injunctions immediately bubble-up, here. And if the 'because' is neither causal nor justificatory, then what sort of a 'because' is it, exactly?
The second point concerns the role of explanation here in general terms. It's all very well for me to throw the explanatory challenge back in the objector's face, as I just have, but it's not much progress if there's reasonable agreement that an explanation of some sort is called for. A main component of the model is the case that the whole explanatory idiom to which the objector's mooted explanation would belong, is misbegotten. It emerges from a conception of rationality as individualistic, according to which truth and meaning can be made sense of only if language can be, in the relevant way, theoretically tethered to non-linguistic reality. The model's point of departure is that sentence valuations - valuations such as the objector insists need explanation - are what's given in experience, and that this experience is fundamentally social (it depends on experience of other people uttering the sentences in question). The thought that intellectual responsibility compels us to try to account in terms of some finite, scientific theory for this admittedly infinite domain of facts originates in a failure to grasp the nature of rationality and our relation to the world. The case that rationality and our relation to the world are as the model maintains is, in the main, simply that the picture of these things it recommends works and is complete. The model does allow that the relevant physical facts can be explained by a finite theory, just not a theory formulated in the terms of sub-personal states and faculties philosophy usually seeks in this context.
The third, related point concerns the putative objects of experience, including the qualia which so vex us. The objector's protest is premised on a conception of the objects of experience as things which can be got hold of independently of language, and which can serve somehow in substantive explanations of our judgements. Some people will not be budged from this position - they will always prefer dismissing philosophical positions such as I am defending to what they will insist are the undeniable facts of experience. To those who would dig their heels in here, I would urge consideration of the profusion of books and papers under whose weight university library shelves groan, on the subject of how perceptual experience figures in judgement. What sort of a thing is a perceptual experience, even of something as pedestrian as a dog on a rug nearby in good lighting, that it can be a basis for judging, in the example, "The dog is on the rug." - how exactly does that work? I should repeat here my position that this is quite a different problem than explaining how it is that light reflecting off the dog causes retinal and many other neural actuations, ultimately eventuating in the production of vocal chord movements. It is (many think) about how a judgement like this can be justified. It is only the beginning of your problems, if you are going to insist that your privately experienced feeling is the theoretically explicable, more than trivial basis for your claim 'I have a tickle in my left big toe'.
Robert Kermode, 2026
(Note: this explanation follows the leading-'_' convention for the model's terms, explained in 'The model' -> 'The Idea'.)
The starting point for the model, and by extension this whole project, is the recognition that our having a feeling of assent or dissent on hearing a token sentence is properly understood as a primitive fact, not amenable to systematic explanation , and the attendant realization that if we allow that speakers are looking to maximize this feeling of assent, then -adding in a few plausible assumptions- sentences acquire a property with exactly the contours of truth.
What was initially missing from the model was a motivation for speakers to maximize the feeling of assent or agreement, as we require them to have. This feeling of assent we are now leaning on is comparable to a sense of harmony or dissonance, but only in a weak sense. I did not intend that this whole account of language should rest on the idea that the feeling of assent is sufficiently intrinsically appealing that people would seek it out (through conversation) for its own sake. To fill the gap, the model introduced the (admittedly contrived) concept of '_pleasure'. _Pleasure was stipulated to be something we get from certain _valuable token _sentences only - not all - and in different degrees (it was later amended to be associated with _valuable _propositions not already _believed). It was meant to be a model-specific feeling, distinct from any we actually have, and sufficiently intrinsically worthy to motivate maximizing _sentential _value (the thought, again, being that an _agent's only known way to maximize _pleasure would be simply to maximize the number of _valuable _propositions encountered). One benefit of basing the model on _pleasure, rather than on some range of real pleasures, as one might seek to do, is that it thwarts any temptation to try to ground the diversity of word-meanings in the diversity of our real pleasures. It's a key point that only a single pleasure is needed to get to a full treatment of truth and meaning.
This being said, _pleasure has always been a contrivance which needs ultimately to be cashed-out.
Let us now remove _pleasure altogether from the model. What real thing should be modelled in its place to provide _agents with a motivation to '_converse'? More simply, bracketing for a moment our theoretical ambitions, what is the practical value of truth? This latter question is not difficult. If I want tea (clumsily, want that I have tea), and I have the ability to get tea (make it the case that I have tea) only if certain things are the case such as that "There is tea in the cupboard." is true, then learning that "There is tea in the cupboard." is true will be of value to me. Similarly, if I dislike being rained-on (dislike that I am rained-on), and I have the ability to avoid being rained-on (make it the case that I am not rained-on) only if other certain things are the case, such as that "The rain will stop in the next five minutes." is true, then learning that "The rain will stop in the next five minutes." is true will be of value to me.
I submit that to a first approximation, the full litany of conditional, contextually situated ('token') opportunities and coordinate wants like this wholly exhausts the benefit of truth . Our having many such abilities conditional on the truth of sentences is what makes learning truth valuable. Adding correlates to the model would do the work it needs to be done.
In place of the singular _pleasure, the model is now augmented with a set of potential, particular, dated token _goods for each _agent ai,
Gi = {gi1, gi2, ..., gin }.
These are meant to be thought of as correlates of all the particular, dated goods and evils a person has the potential to experience or avoid - tasting an aged cheddar, avoiding paying a parking ticket, hearing a Jaco Pastorius solo - whatever. I take this addition to be realistic and intuitive. Additionally, there is now included in the model for each ai a set of token '_actions' they may potentially take,
Mi = {mi1, mi2, ..., min }
This addition, again, I take to be intuitive. Note that the model doesn't need to care about what any of these _goods/_evils-avoided or _actions is. It needs only that they should exist.
With these elements in place, the way is clear to define for each ai a total expected '_utility' function:
Ei : Gi, Mi, Bi → e
where Bi is the set of _sentences _believed by ai and e is a numeric measure of _utility. If _actions are assumed to have some associated cost, Ei can reasonably be taken to schematize ai's expected payout should they act optimally. The important point for present purposes is that ai's expected _utility is in part a function of their _beliefs. To round-out the changes to the model, the condition is imposed on Ei that ai's expected _utility increases roughly as the number of _true _sentences in Bi increases.
As with the _value function, the point of this function is solely to make plain the relevant conceptual components. It emphatically is not to encourage investigation of the algorithm needed to compute it. To think there is something useful to say in the vocabulary of pleasures, actions and beliefs about how such a function might be calculated would be precisely to miss (or reject) the present point.
These additions, I propose, are both realistic and sufficient to motivate _agents to maximize their inventories of _true _beliefs. They are realistic insofar as people in fact do have potential goods as described, and some more or less efficacious range of actions available to them, at any given time. And, crucially, it is realistic that having fewer or more true beliefs can impact the available utility which peoples' scope for action puts at their disposal.
Having _true _beliefs be in this way instrumental in the getting of mundane _goods I think does exactly the job I've been labouring to get done, which is to provide _agents a motivation to accumulate _truths.
Ahmed arrives home and finds Boris on the couch, drinking a cup of tea. He asks, how did you get that cup of tea? Boris answers (improbably pedantically), "Carla said 'There is tea in the cupboard.' I have the ability to get a cup of tea only if what Carla said is true. I wanted tea and I had learned that the conditions for my getting it were fulfilled, so I got myself a cup of tea."
What do we need, to satisfy ourselves that we fully understand this explanation? If we are Ahmed and have no special agenda, and nothing special about the circumstances calls out for clarification, the answer is, "Nothing". Common sense is self-sufficient.
If we are scientists, intrigued by the bewildering responsiveness of animals in general and humans in particular to transitory changes in their bodies and surroundings, we will feel the need for more. We will want to understand the fine details of how it is that the impingement of the sounds "There is tea in the cupboard." on Boris's ears can lead to the sequence of bodily movements which culminates in his being seated on the couch, moving his arm and wrist at intervals to cause tea to pour down his throat. None of this explanation, carefully formulated, it should be emphasized, should contain any semantic terms. The explanation sought is purely naturalistic, non-normative, non-teleological and non-intentional .
A second layer of scientific study might be undertaken to understand the efficacy of our ability-invoking explanations. This layer would seek to understand how it is that ascriptions of abilities, wants and the like effectively predict the movements of agents. Such a study would plausibly show that attributions of abilities correlate, no doubt in some complicated way, to the presence in the brain and body of complex structures with complex correlations to the exterior world. This study, again, would be purely scientific, and quite unsuitable for those professionally armchair-bound .
What is there left, one might ask, for the philosopher to do? What about the small forests eradicated to print the debates over whether reasons for actions are their causes, or what it is to have an ability?
The philosopher might insist that there remains a further task. For any explanation to be adequate, they might say, it must be the case that it can ultimately be formulated as a logical inference. It may be pressed that our job as responsible thinkers is to explicate the terms of Boris's explanation so as to show that it translates to a valid deduction from fully general principles. The model would have to be reconcilable with this explication, making its absence a failure of the model as-is.
This thought, that any fully-adequate justification must look like a scientific justification, is grounded in an individualistic conception of rationality according to which our knowledge of the world must be atomizable in the way the physical world itself is. I am fully aware that this conception is so thoroughly ingrained that questioning it looks to be utterly flaky. What I think the model shows is that the alternate, public conception of rationality is fully, responsibly defensible, free of the problems and paradoxes which beset the individualistic conception and much better aligned to ascriptions of rationality outside of philosophy . This latter conception substantiates a picture according to which justifications can be fully adequate which 'come to an end' as they do in everyday talk, with their requisite normative force intact.
Zooming out to a broader view of the work of philosophy, a possible rejoinder to all this is that the activity of looking to eke out a fully general, logically robust understanding of the present and other explanations, and of concepts generally, is a normative undertaking, meant solely to refine our understanding of what we are saying and to help us to come to a more precise discrimination of the distinctions we ordinarily make. This provides for philosophical inquiries' being consistent with what the model espouses, while remaining opposed to the model's effective quietism .
I am not wholly averse to the basic point of this response. I do not claim the model, if accurate, puts philosophers completely out of business - only that some philosophical undertakings are rendered moot. It's not impossible that there is a deeper understanding to be had of our thinking about abilities and capacities in particular and philosophically challenging concepts in general.
The main point is simply that there is not, as philosophers often imply there is, any substantial problem with our talk in these domains. They are in order as they are. Generally speaking, there are not, in our everyday talk, lurking paradoxes, internal inconsistencies or explanatory gaps which it is the philosopher's job to resolve. The thought that there are originates in the conception of rationality which the model replaces. Where there is work for philosophers to do, it will look in the main like what Ryle, Austin and Wittgenstein undertook in their writing (in their quite different ways).