Dialogical metaethics

This blog post explains my view on metaethics, which, as I’ve mentioned before, is similar to “discourse ethics” but is different (and it is based on a conception of argumentative dialogue, but the name “argumentation ethics” is taken), and hence I call it “dialogue ethics” (an aptly similar-but-different name) or “dialogical metaethics” (a name that more nicely indicates that it is, after all, a view on metaethics rather than first-order ethics, but loses the parallel with discourse/argumentation ethics).

It is somewhat longwinded because I think if I do not spell things out in a lot of detail, people don’t get it, and misinterpret it. Hence there are four sections of background, to try to teach you some academic theories in case you’ve never heard of them (just skip through them if you have), and then I develop the ideas themselves in the latter four sections.

Basically, I argue that moral and ethical concepts are best understood as dynamic components of a language game. Building on a Wittgensteinian “Use Thesis” as motivation and David Lewis’s “conversational scoreboard” as formal background, I develop a model where ethics is treated as a conduct-defense dialogue (CDD). The ultimate conclusion attempts to ground a narrow, absolute set of ethical norms (the social-pragmatic truth-norm, SPT) in the necessity of preserving the social conditions required for philosophical inquiry itself.

The theory of argumentation will help to develop what a logic of value judgments has tried in vain to provide, namely the justification of the possibility of a human community in the sphere of action when this justification cannot be based on a reality or objective truth. And its starting point, in making this contribution, is an analysis of those forms of reasoning which, though they are indispensable in practice, have from the time of Descartes been neglected by logicians and theoreticians of knowledge. —Perelman & Olbrechts-Tyteca, The New Rhetoric, conclusion

• 1. Background: Why language games?
• 2. Background: Why dialogue games?
• 3. Background: Survey of formal dialogue games
• 4. Background: Introduction to conversational scoreboards
• 5. Arguments as winnable language games (WLGs)
• 6. Ethics as conduct-defense dialogue (CDD)
• 6.1. Formal details of CDDs
• 6.2. CDDs best explain reasons
• 6.3. CDDs best explain vague moral rules
• 6.4. CDDs best explain LYCD and other deontic notions
• 7. Philosophical arguments: ΦDGs
• 8. Philosophical conduct-defense (ΦCD)
• 8.1. Introduction of ΦCDs and the SPT
• 8.2. The insecurity argument and deontological absolutism
• 8.2.1. Exceptions undermine the security conditions of inquiry
• 8.2.2. The ΦCD deontic core is absolutist
• 8.3. Proof in ΦCDs: refutation by SPT-violation
• 8.4. Theoretical strength of SPT
• 8.5. ΦCDs as dialogical metaethics

1. Background: Why language games?

Michael Dummett has suggested in a number of places (The Logical Basis of Metaphysics, 1991, pp. 3–4) that it is a defining dogma of analytic philosophy that “the philosophy of thought can be approached only through the philosophy of language. That is to say, there can be no account of what thought is, independently of its means of expression.” Some philosophers, Dummett remarks, have rejected the doctrine of the priority of language over thought, but “on the face of it, they are overturning the fundamental axiom of all analytic philosophy, and hence have ceased to be analytic philosophers.” —Robert Stalnaker, Context and Content, p. 1

The Use Thesis. The important idea behind language games is that words do not have eternal meanings in heaven, set ahead of time by a perfectly clear formal grammar and semantics written in a perspicuous metalanguage, which underpin our interpretation of each word in every context in which it is used. Far from it: language meaning arises within language use within human activities. Language games are those activities in which language is used, and within which its meaning arises, such as giving orders, describing an object, joking, praying, teaching, reporting, and so on. Each of these activities is governed by its own separate rules, fixed by the social practices and “forms of life” in which the game is played. Language, then, is a family of interrelated practices, each of which may turn out to give a different meaning to the same word. (See this paper for a simple model that easily explains how linguistic meaning can emerge from its usefulness in the context of an activity.)

To imagine a language means to imagine a form of life. (PI §19)

Language use within language games captures all that there is to understanding concepts, since as Ludwig Wittgenstein has proved beyond a shadow of a doubt, there is no such thing as “private language” – there is no prelinguistic “mentalese” in which you think things before you put them into words. Philosopher of mathematics Stewart Shapiro, in his work Foundations without Foundationalism, has called this the “Use Thesis”:

A slogan often associated with such views is ‘meaning is use’, but I believe that this is misleading. For present purposes, it is a thesis about understanding, not a thesis about meaning or semantics, at least not directly. The claim is that understanding should not be ineffable. One understands the concepts embodied in a language to the extent that one knows how to use the language correctly. Call this the Use Thesis.

As Shapiro emphasizes, there is no hope of interpreting human language within a metalanguage, which would then need a metametalanguage for its own interpretation, which would in turn require a metametametalanguage. The buck stops with actual use within human practices.

That there is a misunderstanding here is shown by the mere fact that in this chain of reasoning we place one interpretation behind another, as if each one contented us at least for a moment, until we thought of yet another lying behind it. For what we thereby show is that there is a way of grasping a rule which is not an interpretation, but which, from case to case of application, is exhibited in what we call “following the rule” and “going against it”. (PI §201)

All the same, language games are rule-governed. Just because we cannot create a single formal semantics for all of language and call it a day, doesn’t mean we can’t create formal models of how our words are used and meant (though always with reference to our form of life), and thereby clarify our concepts to ourselves. This is what is especially attempted in semantical models that explicitly invoke the idea of a language game and attempt to keep track of how the linguistic pragmatic context evolves throughout such an activity. Such models can focus, for instance, on the activity of giving orders, or of telling jokes. Of especial interest to us here, however, are those models that attempt to capture the dynamics of cooperative dialogues, or conversations, to which we now turn.

2. Background: Why dialogue games?

So, for the old, limited and theory-ridden programme of analysis, we are to substitute a different aim: that of coming to understand philosophically puzzling concepts by carefully and accurately noting the ways in which the related linguistic expressions are actually used in discourse. Of course, not all features of the use of these expressions will be relevant to the philosopher’s task. It is his special skill to discern which are relevant, and how they are relevant. —P.F. Strawson, Construction and Analysis

Why care about dialogue? Dialogues are of especial philosophical interest because they’re the context in which the established usage of concepts is not just taken for granted, but also interrogated and revised. We may recall Plato’s Socratic dialogues about the meaning of courage, piety, justice, etc. It would be wrong to pull out a dictionary to answer such questions, because the dictionary just describes how people use the word, and the concept in question may still be reviewed for its correspondence with underlying reality, or coherence with other concepts, etc.

Dialogues are especially important to questions that are normative or have immediate deep practical relevance regarding how to conduct one’s life, because such questions are what Agnes Callard, in her book Open Socrates, calls “untimely questions”: they are questions that arrive after we already have an answer to them. Which is not to say that we know the answer, but that the way in which we live already assumes some answer, which may be true or false, but which, without the aid of dialogue, almost certainly goes unexamined, due to the difficulty of distancing ourselves from the life in which we use them. When we do not know the answer, this leads us to wavering about it when it is questioned, and getting led into confusions and inconsistencies.

Dialogues solve the problem of untimely questions because one person can face questions regarding someone else’s life with much more detachment—it is the difference between one of us pondering the question, “will Jones stay married to his wife?”, on the one hand, and asking Jones, “do you think you’ll stay married to your wife?”, on the other hand. The latter is understandably thought by Jones to be an unthinkable question, but this may be precisely what keeps him from facing issues in his relationship. Intriguingly, Callard holds that dialogues allow us to assert what in normal contexts seems paradoxical: that “the belief that I currently hold is false.” This is called Moore’s paradox. Agnes explains that in dialogue, we can be led to such an assertion about our own belief, which is true when stated, although it immediately leads us to change our mind about the belief. That we can so starkly escape what Callard calls our “normative self-blindness” is a testament to the power of dialogues to serve as engines of inquiry.

The point applies to any question regarding which we must have some belief or other, and cannot simply wait and suspend judgment until absolutely conclusive evidence comes our way, as we can do in the case of abstruse questions about the existence of extraterrestrial life, about the solution of famous ancient and distant mysteries such as the Voynich manuscript, or generally theoretical questions about fields that we are not personally involved in researching or working with or investing in. Regarding momentous questions, it is of dire practical importance that we believe the truth, but as William James pointed out, this desire contrasts with the requirement to avoid error. We can believe many truths (together with many errors) if we are credulous, and we can avoid all errors if we form no beliefs and suspend judgment about everything. Only in dialogue can we reach a mean between these demands by letting a different person play the role of sceptic toward our deeply held, loadbearing answers to life’s questions.

Why care about formalizing it? The goal of formalizing dialogue is to achieve part of the aim of interpreting natural language in a formal semantics: to understand the structural, law-governed features of this language game. Formal dialogue games make this structure explicit by itemizing the causal factors that we believe can be relevant to an explanation of dialogue. By doing this, we keep ourselves honest: whenever we believe that a particular case warrants a unique mode of explanation (e.g., from Freudian theory), we can’t be ad hoc about it, we must ask ourselves how this new idea interacts with our previously accepted ideas (our definitions and axioms) and in what cases we can apply it.

Formal dialogue games also make our explanations particularly perspicuous. For instance, in his famous monograph Dialogue Games, Lauri Carlson sets up conversation as a turn-based structure with individual players, their aims and payoffs, discrete moves, and a simple rule (called “D.say”) that allows any player, when it is their turn, to say anything. Carlson then derives many common semantic and pragmatic constructs, such as Paul Grice’s idea of relevance, not from an assumption that the conversation is “cooperative” as Grice had it, but simply from the assumption of the players’ rationality in making their moves according to what will best satisfy their utility function. Even if Carlson’s derivation is wrong, it must certainly be of great scientific interest to the semanticist (or pragmaticist) to find out where and how precisely it is wrong. At least I hope that this example has somewhat shown what is interesting about formal dialogue games. The next section will be a review of the history of dialogue games, as I conceive of it.

3. Background: Survey of formal dialogue games

Like research in many other disciplines, the study of argumentation goes back to classical antiquity. Unlike in most other disciplines, however, knowledge of the ancient literature remains in argumentation theory a necessary condition for a proper exercise of the profession. Certain theoretical insights formulated by classical authors, such as Aristotle and Cicero, still belong to the core of argumentation theory. They are an integral part of the foundations of the hermeneutic and critical tools that are currently available for the analysis and evaluation of argumentative discourse and texts. —Eemeren & Grootendorst, A Systematic Theory of Argumentation, p. 42

The goal of this section is to contextualize what I’m doing. Any gaps in my narrative correspond simply to gaps in my knowledge. This section draws heavily on §3 of “A Systematic Theory of Argumentation” and §§3.2–3.3 of “Strategic Maneuvering in Argumentative Discourse”, as well as the SEP entry on “Argument and Argumentation”.

The scientific study of dialogue is usually traced to the ancient discipline of dialectic: a family of techniques for testing and disputing claims by means of regulated exchanges. This lineage is sometimes pushed back to early Greek philosophy, but the first systematic “dialectical” templates that clearly anticipate later dialogue models are found in Zeno, Plato, and Aristotle—even though none of them present a single, fully explicit model of dialogue in the modern sense.

A useful starting point is the classical division, already explicit in Aristotle, between three related but distinct enterprises:

• Logic (analytica): arguments whose premises are evident and whose validity can be treated in a strictly demonstrative way.
• Dialectic (dialectica): regulated debate in which a thesis is tested through an interaction of moves for and against it, using premises that are generally accepted (endoxa) rather than self-evident truths.
• Rhetoric (rhetorica): the art of persuading a particular audience, where premises need only be plausible to that audience and where ethos and pathos can matter alongside logos.

This tripartition already marks two different “research programs” that later theories inherit:

1. A dialectical program: model the procedures by which a standpoint is critically tested and a difference of opinion is resolved (or a participant is forced to retract).
2. A rhetorical program: model the resources by which a speaker secures assent from an audience.

Zeno of Elea’s (490–430 BC) celebrated style of reasoning is dialectical primarily in aim: it works by refuting an opponent’s standpoint (often by deriving contradictions or absurd consequences) in order to support the opposite. Structurally it is often monological—a single line of reasoning that can sometimes be reconstructed as a dialogue only because an opponent’s thesis is presupposed.

In Plato (427–347 BC), dialectic becomes a dialogue form governed by recognizable norms. Several patterns stand out:

• The Socratic elenchus: a questioner tests whether an answerer’s standpoint is tenable given the answerer’s other commitments; the answerer “defends” largely by avoiding being driven into contradiction by what he has conceded.
• The method of hypothesizing: a hypothesis is tested critically, supported deductively by higher principles, and used to derive conclusions.
• The method of collection and division: conceptual clarification by grouping kindred concepts under a genus and dividing into species and subspecies.

Already here we see “dialogue rules” in embryo: permissions to ask for clarification, expectations of consistency, and a distinctive allocation of roles (questioner vs. answerer) that constrains what counts as a legitimate move.

Aristotle’s (384–322 BC) Topics and Sophistical Refutations describe a game-like procedure that strongly resembles later dialogue games. Its aim is not to “prove” a thesis from self-evident truths, but to test whether it is acceptable in light of generally accepted opinions or the answerer’s own concessions.

Crucially, in Aristotelian dialectic the defender does not typically advance a developed positive case in the way modern debates do. Instead:

• The questioner tries to obtain concessions and then derive a contradiction (or an unacceptable consequence) from them.

• The answerer tries to choose concessions carefully and avoid being refuted on the basis of what he has accepted.

This is already a recognizable dialogue game: a regulated exchange, role-defined obligations, and a win condition (refutation) that turns on commitments made during the interaction.

At the same time, Aristotle’s rhetoric develops a different line: persuasion using logos/ethos/pathos, with rational persuasion often taking the form of enthymemes (rhetorical syllogisms with plausible premises and frequently an implicit premise) and examples (rhetorical induction). Roman rhetoric later elaborates this syllogistic core into richer templates (e.g., the epicheireme with explicit support for both premises).

Medieval universities made dialogue formal in a different way: by institutionalizing debate as disputation. Disputations introduced stable roles, fixed turn-taking, and explicit burdens (who must answer which challenges, what counts as conceding, when a contradiction forces retraction). The most famous variants, called obligational disputations, are valuable historically because they turn argument into a procedure: the debate can be studied as a structured sequence of permissible moves, rather than as free-form persuasion.

A key twentieth-century revival of the medieval–Aristotelian “disputation” idea is Nicholas Rescher’s dialectics (1977). Rescher explicitly opposes a Cartesian picture of isolated reasoners proving conclusions from indubitable truths, and instead models rational inquiry as a controversy-managed exchange in which assertions can be cautious and revisable. Central to this model is a shifting burden of proof—an initial burden attached to the party who asserts, and an evidential burden that can move back and forth as challenges are met—together with a doctrine of presumption grounded in plausibility. This makes Rescher’s work a natural hinge between historical disputation formats and later formal commitment-based dialogue systems.

Hegel’s dialectics can be read as extending the idea of dialectical opposition and development to large-scale domains—metaphysics, social life, and history—treating conflict and resolution as drivers of conceptual and historical change. Relative to dialogue-game traditions, this is less a theory of conversational procedure than a grand analogical application of dialectical patterns.

Modern argumentation theory takes off in the mid-twentieth century, partly by reacting against the idea that reasoning in ordinary language can be fully captured by formal deductive validity.

Stephen Toulmin, in The Uses of Argument (1958), proposes a now-classic schematic: Data → Claim, licensed by a Warrant, optionally supported by Backing, and modulated by qualifiers and rebuttals. The key evaluative idea is that what counts as a good warrant/backing is often field-dependent—legal, scientific, moral, and everyday arguments may be assessed by different standards of adequacy.

Although Toulmin frames his layout as anticipating challenges, the model primarily represents argumentation from the speaker’s side (how to support a claim), rather than as a fully reciprocal procedure in which arguments for and against are systematically weighed in interaction.

Perelman and Olbrechts-Tyteca, in The New Rhetoric (1958), catalog the argumentative techniques by which speakers win adherence from audiences, distinguishing between a particular audience and a universal audience (an idealized representation of reasonableness). The norm of “soundness” is closely tied to what persuades the intended audience—again a strongly rhetorical orientation—along with elaborate classifications of premises and argument schemes (association/dissociation, quasi-logical arguments, arguments based on or grounding the structure of reality).

Another influential attempt to dialectify argument while keeping the traditional premise–conclusion focus is Ralph Johnson’s informal logic, developed in his book Manifest Rationality (2000). Johnson treats the point of arguing as rational persuasion of an Other in a context of controversy. That yields a two-tier conception: an argument has an illative core (reasons supporting a conclusion) but also a dialectical tier in which the arguer anticipates and addresses objections. This makes explicit why argumentative discourse cannot be understood as a static product alone: even apparently monological texts tacitly presuppose a space of criticism that must be managed.

From a dialectical point of view, rhetorical success is not enough. What is wanted is a model that specifies which moves are admissible if the goal is to resolve a difference of opinion “on the merits,” through a regulated exchange of criticism and defense. Frans van Eemeren and Rob Grootendorst develop pragma-dialectics (1984) as a theory that is simultaneously:

• Dialectical: oriented to resolving a difference of opinion through regulated criticism and defense.
• Pragmatic: treating argumentative moves as speech acts performed in real communicative contexts.

They frame argumentation as a purposive, commitment-generating activity embedded in interaction, and build an ideal model of a critical discussion with four analytically distinct stages:

1. Confrontation: a standpoint meets doubt or contradiction; the difference of opinion becomes manifest.
2. Opening: parties establish common ground (starting points and rules) and adopt roles (protagonist/antagonist).
3. Argumentation: the protagonist advances arguments; the antagonist critically tests them; further rounds may follow.
4. Concluding: parties determine whether the standpoint is upheld or must be retracted.

A distinctive feature here—important for dialogue games—is that the model is explicitly tied to types of speech acts and their distribution across stages (assertives to state standpoints and arguments, directives to challenge and request argumentation or clarifications, commissives to accept roles/rules and to accept or reject defenses, plus “usage declaratives” like definitions to prevent ambiguity from derailing the exchange).

This is a normative standard used to reconstruct real discourse and diagnose where it derails into fallacy or unproductive conflict.

Maurice Finocchiaro (2006) usefully systematizes these options by distinguishing increasingly “dialectical” conceptions of argument. On weaker views, an argument is just reasons offered for a conclusion; on stronger views, it also includes (or even primarily consists in) defending the conclusion against objections. In his taxonomy, Johnson’s model is “strongly dialectical” (support + defense), while pragma-dialectics exemplifies a “hyper-dialectical” approach in which argumentation is fundamentally a structured response to actual or potential criticism. Finocchiaro also supports this dialectical shift with historical analysis: studying extended controversies (for example in Galileo, which he had studied in 1980), he argues that much real-life argumentation is predominantly critical—its driving force is rebutting counterarguments and resolving doubts—rather than merely piling up positive supports. This strengthens the motivation for dialogue-based models that treat objection, burden, and reply as constitutive of argumentation.

In later work on Strategic Maneuvering in Argumentative Discourse (2010), pragma-dialectics acknowledges that real arguers simultaneously try to be reasonable and to advance their own case effectively. This creates a bridge to the classical observation that even dialectical exchanges involve tactics—without collapsing the dialectical ideal into mere persuasive success.

In parallel, a more explicitly mathematical tradition treats dialogue as a game with formal rules:

• Arne Naess, in Interpretation and Preciseness (1953) and Communication and Argument (1966), develops a modern dialectical perspective emphasizing agreement on procedures and the systematic weighing of pro and contra considerations (including structured “pro-et-contra” surveys).
• Charles Leonard Hamblin, in his classic Fallacies, coins “formal dialectic” for precisely regimented dialogue systems (dialogue games), arguing that such systems can illuminate real argumentative practices and support the evaluation of fallacies.
• The tradition of dialogical logic (1955) develops dialogical methods for logical consequence: a proponent and opponent attack and defend statements according to rules tied to logical constants.
• Barth & Krabbe systematize these ideas in From Axiom to Dialogue (1982), developing procedures where a proponent defends a thesis ex concessis—by exploiting the opponent’s concessions—under carefully specified rule sets and winning conditions.

This family of work is central to the history of formal dialogue games because it makes explicit what ancient dialectic often left implicit: the move types, commitments, and victory conditions that define a dialogue as a rule-governed structure.

Douglas Walton, in his Informal Logic (1989), develops a complementary line: a pragmatic, normatively oriented analysis of argumentation in everyday contexts, emphasizing dialogue types (persuasion dialogue, inquiry, negotiation, eristic conflict, etc.) and how standards of assessment shift with conversational goals. This preserves the game-like insight of formal dialectic while trying to stay close to real communicative practice.

A variety of additional models intersect with these traditions:

• Commitment-based dialogue systems (J.D. Mackenzie’s 1979 “DC” model and its later developments): formalize how participants’ public commitments evolve as they assert, concede, challenge, and retract.

• Abstract argumentation frameworks (Phan Minh Dung, 1995): represent arguments and attack relations at an abstract level, enabling formal notions of acceptability.

• Dialogue games in formal semantics (Carlson’s 1983 Dialogue Games) and game-theoretical semantics (crystallized in Jaakko Hintikka’s 1973 Logic, Language-Games and Information): treat meaning and logical validity via structured interaction, often not primarily aimed at capturing natural-language debate but at explaining semantic notions through game structure.

• Conversational scoreboards (David Lewis, 1979): model the evolving “context set” and related pragmatic parameters of conversation—what is presupposed, what is at issue, what updates the common ground—often combined with formal semantics (e.g., possible-worlds frameworks) to connect context change to meaning.

Across these threads, the idea of a dialogue game becomes progressively sharper:

• From antiquity: regulated testing of a thesis by concessions and refutation.
• From medieval disputation: institutionalized turn-taking and burdens.
• From modern argumentation theory: richer attention to ordinary language, persuasion, and the normative ideals of reasonable resolution.
• From formal dialectic and dialogue logic: explicit rules, commitment bookkeeping, and formal win conditions.
• From formal pragmatics: explicit models of how conversational context evolves.

Taken together, these developments explain why contemporary dialogue-game research can serve both as a tool for logical and semantic idealization and as a framework for reconstructing and evaluating real argumentative discourse.

4. Background: Introduction to conversational scoreboards

“With this ring I thee wed” is verified by its felicitous use, since the marriage relation is a component of conversational score governed by a rule of accommodation. Is marriage then a linguistic phenomenon? Of course not, but that was not implied. The lesson of performatives, on any theory, is that use of language blends into other social practices. We should not assume that a change of conversational score has its impact only within, or by way of, the realm of language. —David K. Lewis, Scorekeeping in a Language Game

The kind of formal dialogue game model that I’m focusing on is David Lewis’s conversational scoreboard. It arose within philosophy of language, and was a major contribution to formal pragmatics. The model serves as a way to keep explicit track of conversational context in a natural-language conversation; features other than context can be handled formally, if one wishes, by combining the model with a formal semantics such as Cresswell’s, but fine-grained details of such features are, anyhow, indifferent to my present analysis. To serve as further background for what I’m doing, this section explains Lewisian conversational scorekeeping in detail, mainly by copying all the formal features of what Lewis said in his original paper (PDF) and removing the interesting details from the examples that he gave for brevity (hence this section is much less fun to read than the paper), but in compensation for that, enriching the exposition of the model by an added reference to Robert Stalnaker, who invented the idea of context as common ground, as well as various references to the survey paper by Dänzer, Rinner, and Kulakova (referred to here as “DRK”), which is relied on to broaden our horizons by giving some context on how other models than Lewis’s have worked, or what other scorekeeping theorists than Lewis have said in their theories.

Core idea. The core idea of conversational scorekeeping is that speech and context interact bidirectionally. Utterances change the contextual score; that score then shapes the interpretation and acceptability of further utterances.

Conversation-stages as states of the conversational score. Let a conversation stage be indexed by time t. Its conversational score is an n-tuple with the various components that are kept track of by the score; for instance, using the examples from Lewis’s original paper, we may have Sₜ = ⟨P, B, ≼, σ, ρ, π, A, Φ, Λ⟩, where the components are the following:

• P: set of presupposed propositions (common ground).
• B: boundary between permissible and impermissible acts for relevant roles (plus the status of any explicitly permitted/forbidden classes).
• ≼: near-permissibility preorder over impermissible acts (how close they are to being allowed).
• σ: salience ranking over discourse entities (optionally over guises).
• ρ: point of reference for deictic direction (e.g., “come/go”), including its determinacy.
• π: standards of precision governing vagueness (what counts as “true enough”).
• A: modal accessibility (the boundary between relevant vs. ignored possibilities; parameterizable by flavor—epistemic, deontic, practical, etc.).
• Φ: performative registers/social pairings currently in force (e.g., name-of, marriage, institutional statuses) that language can update when felicity conditions hold.
• Λ: plan structure under construction (if the activity is planning), capturing provisional steps/roles/resources.

To repeat, these are just Lewis’s examples of score components. As DRK point out, other models of conversational score have included as components: perspectival / de se information; derived contexts, such as suppositional contexts (“suppose it’s raining…”) created from the common ground and used to model conditionals; probabilistic common ground (shared credences / probability spaces) to handle epistemic modals; Questions Under Discussion (QUDs); goals, imperatives, and norms. None of this matters here.

Acceptability relative to the score. As Lewis says, “what counts as correct play depends on the score.” For an utterance u at t, define an acceptability predicate Acc(u, Sₜ) that composes the ordinary truth-conditions of u with score-dependent parameters:

• Truth/denotation may depend on σ (definite descriptions), ρ (come/go), A (can/must).
• Warranted-assertability/non-triviality may depend on P and π.
• Normative claims depend on B and ≼.
• Performatives depend on Φ and felicity conditions.
• Planning talk depends on Λ.

An utterance is a correct move if, and only if, Acc(u, Sₜ) holds.

Score kinematics. A kinematics function (possibly set-valued due to underdetermination): Sₜ₊₁ ∈ K(Sₜ, uₜ, eₜ), where uₜ is the utterance and eₜ includes salient nonlinguistic events. Baseline constraints on K:

• Conservatism/minimal change: prefer smallest revisions that satisfy constraints introduced by uₜ.
• Coherence: maintain internal consistency and role-gated authority (e.g., only certain speakers can move B by fiat).
• Persistence: absent contrary pressure, components tend to persist (context doesn’t thrash).

As DRK point out, some theorists have analysed the illocutionary force of speech acts (in speech act theory) by means of their effect on the conversational score.

The rule of accommodation (master schema). If uₜ requires some component s ∈ Sₜ to satisfy constraint C for Acc(uₜ, Sₜ) to hold, and C is not met just before t, then—ceteris paribus and within limits—revise Sₜ minimally so that s ⊨ C at t. Instantiations:

• Presupposition P: if uₜ presupposes p and no one objects, add p to P.
• Permissibility B, ≼: master-to-slave style assertions shift the boundary (and, when needed, the near-permissibility order) to make them true.
• Salience σ: make the entity a referential expression needs most salient.
• Reference point ρ: fix/shift or refine its location to fit motion descriptions.
• Precision π: raise/lower standards when an utterance is acceptable only at that setting.
• Modality A: expand/contract the relevant possibility space to render modal claims true.
• Performatives Φ: when felicity conditions hold, update the social fact so the performative is verified by its use.
• Plan Λ: add the needed step/resource/role to the shared plan.

Robert Stalnaker says that he sees no reason to distinguish, as Lewis did, between changes by accommodation and “the simpler kind of change that takes place when ‘something conspicuous happens at the scene of a conversation, and straightway it is presupposed that it happened’” (Context and Content, 1999, p. 103); accordingly, Stalnaker does not take the rule of accommodation as a primitive rule, but sees it simply as “the result of rational responses to events that take place in the course of a conversation” (Context, 2014, p. 6). I believe that Stalnaker is right, so that the rule of accommodation is an unnecessary primitive. Nevertheless for our purposes, we take it as a derived rule, left in the formalism merely for more explicit tracking of presupposition accommodation and, as Stalnaker did, we “have appropriated Lewis’s label for the phenomenon, but without his assumption that to accommodate is to apply a constitutive rule of a language game”, (Context, 2014, p. 58) nor his assumption that accommodation is distinct from other rational inferences that result in changes in presuppositions.

Blocking/defeasibility. Explicit challenge, contrary evidence, or violated felicity conditions prevent accommodation; authority gates apply (e.g., not everyone can move B or Φ by speech).

Ratchet tendencies (asymmetries). Accommodation is not perfectly reversible; some moves stick more than their opposites:

• Precision π: raising standards (being more exacting) tends to stick more than lowering them.
• Modality A: outward expansions (considering more skeptical/remote possibilities) resist re-narrowing. These are conversational inertias, not metaphysical upgrades in truth; what was acceptable under earlier π or A needn’t be impugned.

Regulative aims (what players try to do). These are Lewis’ “regulative rules” (try to make score evolve in favored directions). Alongside constitutive kinematics, participants pursue goals that steer score:

• Cooperative: enlarge P, converge on σ, keep π at a level suited to the task, stabilize ρ, make A fit the inquiry.
• Adversarial/strategic: shift π upward to challenge claims; push A outward (skepticism) or inward (commonsense); maneuver B to permit/forbid actions; seed Λ with favorable steps.

As DRK point out, this idea fits very naturally with Gricean conversational maxims. The idea of accommodation, from earlier, is also related to implicature in this connection.

Analysis of score: the “middle way”. Formally: the score function maps stages to Sₜ and is fixed up to underdetermination by the kinematics plus the scoreboard constraint.

• Constitutive view: take kinematic rules as definition-like determinants of score.
• Operational view: score = whatever trusted mental scoreboards register.
• Middle way (preferred): score is what the best-fitting scoreboards register, where “best-fitting” is measured by how well they track the (partly incomplete) kinematics. Thus rules are partly constitutive but may underdetermine evolution; real conversations can depart from ideal rules.

Lewis frames these two views on the interpretation of the score and his “middle way” between them. As DRK point out, other conversational scorekeeping theorists have taken other views on the score and how it is kept.

Initial conditions and limits. Initial score S₀: thin P; default B, ≼; broad σ with immediate context atop; ρ maximally indeterminate; mid-range π; task-appropriate A; empty/neutral Φ, Λ unless a ceremony/plan is already in progress. Limits on accommodation:

• Impossibility: no consistent revision satisfies C.
• Authority/felicity failures: e.g., performatives without proper procedure.
• Objections: explicit challenge suspends update.
• Cost: excessive oscillation (e.g., rapid σ flips) harms coherence even if it rescues acceptability.

As DRK point out, conversational scoreboards have been used to analyse various phenomena, primarily involving how various underhand rhetorical techniques work by subtly changing the context and relying on the change going unchallenged. We will apply them to an analysis of arguments.

5. Arguments as winnable language games (WLGs)

Argumentation is a communicative and interactional (speech) act complex aimed at resolving a difference of opinion before a reasonable judge by advancing a constellation of reasons the arguer can be held accountable for as justifying the acceptability of the standpoint(s) at issue. —Frans van Eemeren, Strategic Maneuvering in Argumentative Discourse

Informal preamble. The goal is to formalize the idea of “winning an argument” using conversational scorekeeping. It is assumed that different communities may have different “case-evaluative conventions”, which are accepted standards of argument. For instance, in a preindustrial traditional community you may be required to support your argument using “the customs of the elders”, while in Calvinist Geneva you may be required to support your argument using the Bible, whereas in modern society you may be required to use some form of scientific study. The “winnable language game” (WLG) framework is intended to be general enough to work for all of these different conventions and any others. To keep the model anchored on something plausible (at least more plausible than “anything goes”), it is formally assumed that the ten pragma-dialectical rules are usually in force, but nothing hinges on this. Similarly, we assume a third-party judge to have a sharper determination of argument victory; it is easy to simplify the model to have no such judge, and accordingly it becomes less clear by whom an argument has been won (which is realistic, as there are many real-life arguments where both sides believe they have won).

The idea of a case-evaluative convention, possibly varying between arguments, and kept inside conversational context, collapses various ideas from the literature on argumentation. First, clarity is audience-relative, i.e., “what counts as perspicuous and explanatory depends on who you are talking to” (Dialogical Roots of Deduction, p. 66). Second, disputants agree on rules, i.e., “rather than merely speculating on instincts and emotional entanglements, argumentation involves by its constructive nature an appeal to reasonableness that derives its force from the idea of common critical standards” (Strategic Maneuvering in Argumentative Discourse, p. 27); which is to say, “a significant degree of agreement between the parties must be in place (either explicitly or implicitly) on what counts as a legitimate, sufficiently perspicuous inferential step that should not be contested” (Dialogical Roots of Deduction, p. 66). Van Eemeren’s Strategic Maneuvering builds on this via the ideas of dialectical profiles and institutional preconditions, but elaborating on this goes beyond our scope.

The idea of presuppositions kept in conversational context, aside from incorporating Stalnaker’s view of inquiry as seeking to change our acceptance states, also fits in neatly with Toulmin’s idea, developed throughout The Uses of Argument, that the use of modal terms such as can, could, would, might, vary between arguments in different fields, since different things are treated as “possibilities” in different fields, which allows modal terms to stand in different contexts for what’s conceptually coherent, what’s physically practicable, what’s legally proper, etc.

Although this is not required by anything else in the formal model, I claim that case-evaluative conventions are conventions in the Lewisian sense, i.e., it’s true and common knowledge among disputants that everyone conforms to the convention, everyone expects everyone else to conform, and everyone prefers to conform on condition others do (uniform conformity is a coordination equilibrium). This claim is meant mainly to explain why failure to conform to a case-evaluative convention is treated as a reason to end the conversation, but may have other interesting implications that could be explored.

Core formal idea. A winnable language game models arguing as a structured interaction in which (i) participants make moves that both depend on and update a conversational/argumentative score, and (ii) a third-party judge can (under the going case-evaluative convention) correctly ascribe victory to one side—while both the convention and the judge’s application of it are themselves only partly fixed (Lewis’s “middle way”).

Basically: WLGs = Lewisian scorekeeping + a score component Π (case-evaluative convention) + seriousness κ with ratchet accommodation + victory defined by a middle-way convergence of competent judge-scoreboards constrained by Π’s partly constitutive kinematics.

Propositions, standpoints, and arguments. Let ℙ be a space of propositions.

• A standpoint is a pair ⟨i, p⟩ where player i advances p ∈ ℙ.
• A challenge/doubt is a pair ⟨j, p⟩ where player j puts p into dispute.
• A move is an utterance-token with illocutionary force: m = ⟨i, force, content, stage, t⟩.

Examples of force: Assert(p), Doubt(p), Arg(p ⇒ q), RequestDefense(p), Concede(p), Retract(p), AccuseFallacy(φ), Clarify(x), etc.

Roles. A WLG minimally has:

• Pr (protagonist): advances and defends standpoints.
• An (antagonist): casts doubt, tests defenses.
• J (judge/audience proxy): applies the going convention to declare outcome.

Generalization to many parties is straightforward; keep two-sided for clarity. This presentation mirrors the Prover-Skeptic dialectics in Dialogical Roots of Deduction.

The argumentative score. At each time t the game-state is a score tuple: Sₜ = ⟨P, 𝒞, Σ, Δ, 𝓑, Γ, Π, κ, 𝓔, 𝓕⟩, where:

• P ⊆ ℙ: common ground / shared presuppositions.
• 𝒞 : {Pr,An} → 𝒫(ℙ): commitments (what each party is publicly on the hook for).
• Σ ⊆ ℙ: standpoints currently “on the table” (advanced for defense).
• Δ ⊆ ℙ: propositions currently doubted/contested.
• 𝓑 : {Pr,An} → 𝒫(ℙ): burden-of-proof ledger (who must defend what).
• Γ: stage marker of the pragma-dialectical profile (Confrontation / Opening / Argumentation / Concluding).
• Π: the going case-evaluative convention (defined below).
• κ ∈ [0,1]: seriousness level (0 = casual/quarrel; 1 = fully critical discussion).
• 𝓔: admissible epistemic base (e.g., agreed sources, testimony rules, “Bible-only,” “elders’ customs,” scientific standards).
• 𝓕: fallacy ledger (alleged/settled violations with pointers to moves).

Intuition: Π and κ are themselves part of the score, so “what counts as a good case” is context-sensitive and dynamically negotiable.

Case-evaluative conventions Π. As a baseline, we treat pragma-dialectics as a core convention. Let Π_PD be the rule-set consisting of the ten pragma-dialectical rules (Freedom … Usage), with their induced fallacy types as recognizable violation patterns. Formally, treat Π as a normative profile: Π = ⟨R, ≼, gate⟩

• R: a set of rules, each a constraint on permissible moves/updates.
• ≼: a priority/weighting over rules (some conventions treat some rules as “harder”).
• gate: authority/role-gating (e.g., who may issue binding procedural calls; whether J can impose clarifications; whether elders’ testimony overrides).

We assume most culturally specific conventions extend Π_PD: Π = Π_PD ∪ ℝ₊, where ℝ₊ adds domain/culture-specific constraints (e.g., admissible sources, style requirements, honor norms). Some conventions, typically connected to “less serious” argument types, may relax subsets of Π_PD by lowering their priority or by weakening their constraints.

Seriousness κ as “strictness of Π in force”. Model strictness by a monotone map: Strict(Π, κ) = the effective rule-profile at seriousness κ.

• As κ increases, more of Π_PD behaves as “hard constraints” (e.g., strict burden-of-proof, strict clarity, strict closure).
• As κ decreases, some rules become “soft” (violations count against rhetorical quality but don’t defeat a case).

Acceptability and correct play in a WLG. Define a move-acceptability predicate: Acc(m, Sₜ) ∈ {0, 1}. A move is correct play if, and only if:

1. Stage legality: (m) is permitted at stage Γ under the effective convention Strict(Π,κ).
2. Rule compliance: it does not violate any rule in the hard core of Strict(Π,κ), or if it does, it is immediately repaired (clarified, retracted, burden reassigned, etc.) in an allowed way.
3. Content appropriateness: the move’s argumentative function matches its force (e.g., defending a standpoint uses relevant argumentation; rule 4).

Violations update the fallacy ledger: 𝓕ₜ₊₁ ≔ 𝓕ₜ ∪ {⟨m, φ, status⟩}, where φ is a rule-violation type (e.g., StrawMan, ShiftBurden, Ambiguity, …).

Score kinematics (updates). A (possibly set-valued) kinematics function: S_t+1 ∈ K(S_t, m_t, e_t), where (e_t) are relevant nonlinguistic events (new evidence, interruptions, etc.). Baseline update tendencies (Lewis-style):

• Persistence: keep components fixed unless pressured.
• Minimal change: prefer smallest revisions that restore acceptability.
• Coherence: maintain consistency of commitments and burdens where possible.
• Authority sensitivity: gated changes require the right speaker/conditions.

Typical constitutive updates:

• If Pr asserts standpoint p: add (p) to Σ; add commitment (p) to 𝒞(Pr).
• If An doubts p: add (p) to Δ.
• Burden assignment (rule 2): if p ∈ Σ and defense is requested, assign p ∈ 𝓑(Pr) unless Π specifies otherwise.
• Concessions/retractions update commitments and can trigger closure.

Accommodation for conventions (Π) and seriousness (κ). The accommodation schema (WLG version): If a move (m_t) is only acceptable under a stricter effective convention (or higher seriousness) than currently in force, then—ceteris paribus—Π and/or κ shift to make it acceptable. Formally, let the constraints required by the move be C(m_t). If ¬(C(m_t) holds in Strict(Π, κ)) but there exists κ′ ≥ κ (and possibly Π′ with extra rules activated) such that C(m_t holds in Strict(Π′, κ′), then ⟨Π, κ⟩ ⇝ ⟨Π′, κ′⟩, with a minimality preference: choose the smallest upward shift that suffices.

Ratchet asymmetry: “more serious is easier than less”. It is easier to “escalate” to a more rigorous and serious argument than to de-escalate. Encode asymmetry via transition costs:

• Let cost↑(Δκ) < cost↓(Δκ) for comparable magnitude changes.
• Or probabilistically: downward shifts require stronger explicit meta-moves and higher consensus thresholds.

A simple deterministic version:

• Upward accommodation requires no objection by the other party (silence suffices).
• Downward “de-escalation” requires an explicit agreement move by both parties (or an authoritative ruling by J if gate allows).

This captures the “seriousness ratchet”: invoking precision, demanding validity, calling fallacies, requesting explicit burdens, etc. tends to stick.

Examples of convention-shifting moves:

• “That’s irrelevant—answer the claim” pressures κ↑ (rule 4 hardens).
• “Define your terms; that’s ambiguous” pressures κ↑ (rule 10 hardens).
• “You asserted it; defend it” pressures κ↑ (rule 2 hardens).
• “Come on, it’s just banter” pressures κ↓ but typically needs explicit uptake to succeed.

Constitutive victory conditions (idealized). Given final stage Γ = Concluding, define Success(p) for the protagonist’s standpoint p under the effective convention:

• Success(p, Sₜ) = 1 if Pr has defended p with compliant argumentation and An is obligated to retract doubt, 0 otherwise.
• Pr “wins” (ideal constitutive notion) if, and only if: ∃ p ∈ Σ : Success(p, Sₜ) = 1; and An “wins” (in the sense of Pr losing) if, and only if, Pr must retract p by the closure rule.

This mirrors pragma-dialectics: success/failure is partly procedural (rule compliance), not merely rhetorical.

The operational judge. A judge has their own internal scoreboard Ŝᴶₜ = ⟨P̂, 𝒞̂, Σ̂, Δ̂, 𝓑̂, Γ̂, Π̂, κ̂, 𝓔̂, 𝓕̂⟩, and a verdict function Verdictᴶ(H) ∈ {Pr wins, An wins, No winner}, where H is the public history of moves.

• Pure operationalism would say: winning just is whatever J says.
• Pure constitutivism would say: J’s verdict is merely a (fallible) report of the rule-determined winner.

Lewis’s “middle way” for argument victory. We define argument victory as what is registered by the best role-filling judge-scoreboard, where “role-filling” is constrained by the partly constitutive kinematics of Π but may be underdetermined. Let 𝒥* be the class of competent judges for Π (those whose verdict practices sufficiently track Π’s rules across cases). Define a fit measure Fit(J, Π, H), capturing how well J’s scoreboard and rulings conform to the rule-profile (including how J resolves underdeterminations in a stable, publicly learnable way). Then define the going verdict as a selection: GV(Π, H) = the outcome most stably produced by high-fit judges in 𝒥*.

• If high-fit judges converge: the argument has a determinate winner.
• If even high-fit judges diverge due to underdetermination or ambiguous meta-status of Π/κ: outcome is No winner / indeterminate (or “winner relative to a permissible precisification of Π”).

This is Lewisian: the rules are partly constitutive (they constrain competence and permissible score-evolution), but the operational practices of judges help settle what the going convention actually is in a community.

Fallacies as score events (WLG reading). A fallacy is typically one of:

1. an illegal/defective move under Strict(Π,κ), and
2. a move that (if left uncorrected) would distort the score (commitments, burdens, starting points, meaning, etc.).

So in WLG terms:

• Rule violations are events that either force repair (clarification/retraction) or count as defeat conditions for a defense attempt, depending on κ and Π.
• Many fallacies can be modeled as illicit accommodation attempts (e.g., smuggling starting points, forcing a burden shift, manufacturing a fictitious standpoint).

What makes WLGs “winnable”. A language game is winnable when Π and κ make it possible to reach a terminal stage where:

• burdens are assigned,
• defenses can succeed or fail by recognized criteria,
• and closure has normative bite (even if only “in the eyes of” competent judges/audience).

Formally: WLGs are those for which (given typical histories) K tends to drive Γ toward Concluding and makes GV(Π, H) frequently determinate.

6. Ethics as conduct-defense dialogue (CDD)

Understand also, that I tell you what I am about to tell you, not because I have been provoked, abused, calumniated, traduced, assailed with insinuation, innuendo, mispresentation, lies: not because my life has been held up to ridicule, and to most inferior contempt: not because the most preposterous stories to my detriment have been invented, hawked about, believed. No. Please understand that I am not going to speak in my own defence, even to you. —Hadrian the Seventh

The previous sections treated argumentation as a winnable language game (WLG): a regulated activity in which utterances update a public score, and in which “winning” is (i) constrained by an operative convention Π and (ii) settled, Lewis-style, by a middle-way convergence among competent judge-scoreboards. This section applies that framework to the semantics of ethical and deontic expressions by identifying the fundamental evaluation context for such expressions with a particular subtype of WLG, which I will call a conduct-defense dialogue. The idea is that if we want a perspicuous formal model of how deontic expressions get their normative bite in ordinary life, the right “home game” to model is the game in which one’s conduct is called into question and one must answer for it.

Informal picture. In everyday life, ethical evaluation is typically activated by a recognizable pattern:

1. Someone does (or plans to do) something.
2. That conduct becomes saliently contestable—by accusation, doubt, resentment, disappointment, conscience, or institutional scrutiny.
3. A demand is issued (explicitly or implicitly): Why did you do that? / How could you? / What right did you have?
4. The agent responds with considerations offered as a defense: reasons, excuses, justifications, explanations, commitments, values, precedents, rules, exceptions, and so on.
5. An audience (sometimes the interlocutor; sometimes a group; sometimes “the community”; sometimes the agent’s own internalized judge) registers whether the defense stands.

Though no one has formalized it before, this is already a dialogue game: role-defined moves, burdens, admissible supports, and closure conditions (apology, retraction, blame, exoneration, repair). Theories of deontic language, such as “ought,” “permissible,” “wrong,” “excuse,” “justification,” should pay attention to this game, since that’s where these notions come from: they are, in their characteristic use, statuses conferred or denied in a dispute about conduct.

After some formal details of CDDs (6.1), I will claim and show that by treating ethics as grounded in CDDs, we can best explain the prominence of reasons in ethical discussion (6.2), the action-guiding power of vague moral rules (6.3), and the semantics of notions, such as LYCD, that deontic logic struggles to capture (6.4).

6.1. Formal details of CDDs

The Conduct-Defense Thesis. The fundamental evaluation context for deontic expressions is a conduct-defense dialogue in which the agent defends the claim that their conduct was in compliance with the going normative standards. We call these dialogues CDDs, for conduct-defense dialogues.

CDD as a subtype of WLG. We now specialize the WLG apparatus to the ethical case. We keep the Lewisian scorekeeping structure, but we replace the case-evaluative convention Π with a going deontic convention Θ (Theta), and we make explicit what is being defended: not merely a proposition about the world, but a proposition about compliance.

Deontic convention Θ. Aside from specifically argumentative standards, CDDs require a kind of cooperation where all participants seek to do what is good or right by the community’s conduct-evaluative convention. Just as Π encoded “what counts as a good case” for argument evaluation, Θ encodes “what counts as compliance” for conduct evaluation. Formally, let Θ = ⟨N, ≼, gate⟩, where:

• N is a set of normative constraints, including general norms (“don’t lie”), role norms (“as a doctor, disclose risks”), relational norms (“care for your child”), and institutional norms (law, policy, religious rules), plus schemas for exceptions and defeaters.
• ≼ is a priority/weighting structure over norms and norm-sources (e.g., “preventing serious harm outranks etiquette”; “promises matter unless overridden”; “this community treats elder-testimony as decisive”).
• gate specifies authority and standing: who may accuse, who may demand an answer, who counts as a competent judge, what kind of evidence is admissible, etc.

We allow Θ to vary across contexts and communities, and we assume (as with Π) that real practice partly underdetermines Θ: it is a moving target settled by shared expectations, precedent, and “middle-way” convergence among competent scorekeepers.

Score for a CDD. At conversation stage t, let the CDD score be Sₜ = ⟨P, 𝒞, Σ, Δ, 𝓑, Γ, Θ, κ, 𝓔, 𝓕, 𝓗⟩, where most components are as in §5, and we add one crucial new component:

• 𝓗: a conduct record (a structured representation of what the agent did or is planning to do, including salient circumstances, options, and outcomes, to the granularity demanded by κ and Θ).

Intuitively: a CDD is not a dispute about p in the abstract; it is a dispute about p as it bears on conduct.

The defended proposition: compliance. Let dᵢ be the compliance proposition for agent i:

• dᵢ : “Agent i’s conduct (as represented in 𝓗) is in compliance with Θ (at seriousness κ, given evidence base 𝓔).”

In a CDD, the protagonist’s standpoint is typically of the form ⟨i, dᵢ⟩, and the antagonist puts dᵢ into dispute by alleging violation, negligence, disregard, or failure to meet an obligation.

It may be interesting to compare this to the Andersonian-Kangerian-Leibnizian reduction in deontic logic, which represents standard deontic logic inside modal logic by seeing what is strictly implied by a constant that represents a deontically ideal situation, i.e., what necessarily follows from “all obligations are fulfilled.”

Roles. Similar to before:

• Df*(defender): the agent whose conduct is under scrutiny (or their representative).
• Ch (challenger): the party raising doubt/accusation.
• J (judge): audience proxy / competent evaluator; sometimes institutional, sometimes social, sometimes internalized (conscience).

Moves. Example move-types:

• Assert(Comply(i))
• Accuse(Violation(n)) / Doubt(Comply(i))
• RequestDefense(Comply(i))
• OfferReason(r) / CiteNorm(n) / ClaimException(e)
• OfferExcuse(x) / OfferJustification(j)
• ChallengeRelevance(r) / AccuseRationalization
• Clarify(“what counts as ‘harm’ here?”)
• ShiftΘ(“in this situation, the relevant standard is…”) / ContestΘ
• Escalate(κ) / Deescalate(κ)
• Repair(Apologize / Compensate / Retract)

The internal structure of “OfferReason(r)” matters: it can be treated as introducing a candidate consideration into the space of admissible defenses under Θ and 𝓔.

Seriousness κ and the ethical ratchet. In §5, κ measured how “serious” the argument is: how strictly Π is applied and how hard violations bite. The same parameter is even more natural in ethics, because conduct-defense comes in unmistakable grades:

• κ ≈ 0: banter, teasing reproach, low-stakes norm enforcement.
• κ mid: ordinary interpersonal blame/praise, expectation management, minor wrongs.
• κ high: “What you did was wrong” in the full-throated sense—serious moral accounting, institutional scrutiny, integrity-threatening conflict.

A prominent feature of moral confrontation is that escalation is easier than de-escalation. Once certain kinds of considerations are put on the table—serious harm, betrayal, exploitation, cruelty, hypocrisy, disrespect—it is conversationally and normatively hard to return to “come on, it’s not that deep.”

Model this exactly as we did for WLGs:

• Strict(Θ, κ) is the effective deontic profile at seriousness κ.
• As κ increases, Θ’s constraints behave more like hard constraints: more demanding burdens, stricter exception-handling, less tolerance of evasions, more insistence on generalizability, less willingness to ignore remote victims or downstream effects.

And we impose a Lewisian asymmetry:

• Upward shifts (κ↑) can occur by unchallenged invocation of broader values (“but what about the harm?”).
• Downward shifts (κ↓) require explicit mutual uptake (“okay, let’s not moralize; we’re just joking”) or authoritative closure.

This captures, formally, the familiar phenomenon that moral language tends to recruit wider-ranging values once the defense falters, and that such widenings “stick” unless actively blocked.

6.2. CDDs best explain reasons

Why reasons are central in ethics. Ethicists often treat reasons as a basic normative unit. On the present analysis, this is explained by the game form of ethical evaluation. In a CDD, a reason is a move-licensing consideration: something that can be offered, challenged, weighed, defeated, or accepted as part of a defense.

Formally, let 𝓡(Θ, κ, 𝓔) be the set of admissible reason-kinds under the effective convention and evidence base (appeals to harm, rights, promises, virtues, roles, precedents, divine command, consent, etc.). A token reason r is in play at stage t if it has been introduced by some move and not excluded.

Now the familiar philosophical distinctions drop out as role-differences inside the same dialogue:

• Normative reasons correspond to considerations that actually count in favor under Strict(Θ, κ): they are defense-makers in the judge’s scoreboard.
• Motivating reasons correspond to the considerations the agent puts forward as their defense—what they treated as counting in favor (whether or not it really does).
• Explanatory reasons correspond to what the judge or theorist cites to explain why the agent acted (often psychological states), which may diverge from both normative and motivating reasons.

Ethics is “reasons-first,” on this picture, because ethics is (in its live use) a practice of answerability. To be under a norm is, paradigmatically, to be liable to a demand for reasons when your conduct is challenged. That is why ethical discourse is saturated with “because,” “in virtue of,” “you had no right,” “you owed,” “you could have,” “it wasn’t your place,” and the rest of the dialectical machinery.

Weighing, defeat, and the structure of a defense. A conduct-defense almost never turns on a single reason in isolation. It turns on familiar operations:

• weighing (how much does harm matter here?),
• defeat/override (promise-keeping defeated by emergency),
• enablers/disablers (consent enables; coercion disables),
• exclusionary reasons (don’t decide on the basis of prejudice, even if it would benefit you).

These are precisely the operations that the literature on reasons treats as central, and precisely the operations a CDD must model to explain how defenses succeed and fail; so the conduct-defense analysis successfully predicts the prominence of reasons.

6.3. CDDs best explain vague moral rules

Vague moral rules become actionable in defense. Many moral rules do not command a specific action-type. They command something vague, open-textured, and conceptually thick:

• “Love your neighbor as yourself.”
• “Treat people with respect.”
• “Be just.”
• “Do not exploit.”
• “Act with integrity.”

A common skeptical thought is: such norms are too vague to guide action. But in practice they guide, and they guide especially when challenged. The CDD model explains why.

Thick norms as defense-conditions. Let ℍ be a particular conduct record (what you did, in context). A thick norm like “love your neighbor” does not typically determine a unique required action ex ante. Instead, it functions as a constraint on what counts as a defensible portrayal of one’s conduct when “love your neighbor” is invoked as a governing standard at seriousness κ.

In a CDD, the challenger can force the issue by a move like:

• RequestDefense(“How is that loving your neighbor?”)

This raises κ and activates a demand for a certain shape of defense: the agent must show that their conduct coheres with the norm’s core applications (care, non-malice, attention to need, fair concern), and must answer predictable counterconsiderations (self-serving rationalization, indifference, contempt, hypocrisy).

The key is that actionability here is not the provision of a decision procedure; it is the provision of standards of answerability. The norm becomes practically forceful because:

1. It defines what kinds of considerations count as relevant (what must be addressed).
2. It defines what kinds of excuses are disallowed (what will be treated as evasive).
3. It sets up a recognizably defeasible burden: if you cannot produce a credible defense under that standard, you lose the conduct-defense game.

In other words: vague norms become action-guiding because we live in a practice in which conduct is liable to challenge, and the challenge forces the norm to be made concrete in the only way that matters for social life: as a publicly assessable defense.

“But it could mean anything!” Within the CDD model, that complaint is treated as a metalinguistic challenge to Θ itself: a charge that the operative convention is too underdetermined to yield stable verdicts. Sometimes that charge is correct: high-fit judges diverge, so the outcome is indeterminate. But often it is not: communities exhibit stable convergence on paradigms, disallowed rationalizations, and acceptable exception patterns. That convergence is part of what fixes Θ (Lewis’s middle way), and it is part of what gives thick norms their non-trivial content.

6.4. CDDs best explain LYCD and other deontic notions

“The least you could have done” (LYCD) as a conduct-defense notion. Paul McNamara, in his SEP article on deontic logic, flagged the idea of “the least you could do” (LYCD) as a neglected normative concept. His example is as follows:

You ought to have come home on time; the least you could have done was called, and you didn’t do even that.

LYCD is not obviously reducible to obligation or even conditional obligation. It is a complaint that targets a minimum acceptable alternative among what the agent could have done.

The CDD framework makes LYCD look completely at home.

LYCD as “minimal defensible alternative”. Fix a situation represented in 𝓗 with a set of salient options Ω (coarse-grained actions the agent could have taken, as recognized under Θ and κ). We assume Θ (or Strict(Θ, κ)) induces a status structure over Ω—at least:

• Defensible(α): there exists a compliant defense of choosing α against the salient challenges.
• Indefensible(α): any attempted defense collapses under the rules, burdens, and admissible reasons.

But LYCD is not “there existed a defensible option.” It is comparative and reproachful: you failed to meet even the minimum that would have been acceptable by your own lights or by the going convention. To model this, we introduce a near-defensibility preorder ≼ on options (a cousin of Lewis’s “near-permissibility” ranking), where:

• α ≼ β means: α is at least as close as β to being defensible / minimally acceptable under the effective Θ and κ, given the salient burdens and challenges.

Then the “least you could do” picks out an option (or set of tied options) that is minimal among the defensible ones relative to ≼. Informally, LYCD says: there was an option α that would have constituted the minimal acceptable defense (the smallest repair, the smallest courtesy, the smallest mitigation), and you did not even do α.

This explains McNamara’s example about how “the least you could have done was call, and you didn’t do even that.” What is being asserted is not a violated obligation, but a failed defense profile: even after granting that you might have fallen short of the ideal, you failed to take the minimal step that would have secured a tolerable defense against foreseeable criticism.

Supererogation, and permissible suboptimality. Once we have a graded defensibility structure, we can express the cluster of notions standard deontic logic struggled with:

• supererogation: actions that are defensible (indeed admirable) but not required—winning a “praise” subgame without being necessary for avoiding blame.
• permissible suboptimality: options that remain defensible but lose on comparative grounds (“you can do that, but you ought not”), i.e., you can mount a defense, but a higher-fit judge will treat a better alternative as the one demanded by the activated reasons at κ.

In CDD terms, these are predictable byproducts of the fact that conduct-defense is a competitive dialectical activity with (i) thresholds for acceptable defense and (ii) comparative rankings among defensible options.

Deontic expressions as anticipatory moves in a CDD. Finally, we can connect the model back to semantics. In ordinary contexts, we often use deontic terms prospectively:

• “You must do X.”
• “You may do Y.”
• “You ought not do Z.”

On the present view, these utterances function as anticipatory score-moves: they position an agent’s future conduct relative to the defenses that would be available if challenged. A schematic semantics (not a full reduction, but a guide to modeling) is:

• OB_Θ,κ(α) (“α is obligatory”) ≈ in the relevant CDDs under Strict(Θ, κ), failure to do α makes Comply(i) indefensible given the salient burdens and admissible reasons.
• PE_Θ,κ(α) (“α is permitted”) ≈ there exists a compliant defense strategy for doing α under Strict(Θ, κ).
• IM_Θ,κ(α) (“α is impermissible”) ≈ any defense of doing α collapses under Strict(Θ, κ).

This makes explicit why deontic claims are so naturally challengeable by “Why?”: they are, implicitly, claims about the availability of a defense under a shared evaluative convention.

It also explains why ethical discourse so quickly becomes meta-ethical in practice: disputes about what one ought to do often turn into disputes about Θ and κ—what standards are in force, what counts as evidence, whose interests count, how far consequences extend, which norms override which, and whether we are in a “serious” mode or not. Those disputes are not a philosophical add-on; they are built into the scorekeeping dynamics of conduct-defense.

7. Philosophical arguments: ΦDGs

…some attitudes which may indeed be preconditions for a discussion, such as a wish to get to, or nearer to, the truth, and a willingness to share problems or to understand the aims and the problems of somebody else. —Karl Popper, The Myth of the Framework

From WLGs to ΦDGs. A WLG, as defined above, is an argumentative language game whose score contains (among other things) a case-evaluative convention Π and a seriousness parameter κ, and whose moves are “winnable” because competent judges can, often enough, converge on a verdict given Π and the move-history.

A philosophical dialogue game (ΦDG) is a subspecies of WLG distinguished by an additional, specifically philosophical convention: a truth-norm. In slogan form: ΦDGs = WLGs + the truth-norm.

The truth-norm, aside from “do not lie”, includes whatever is implied by the stronger commitment that, whatever else we may be doing—educating, entertaining, saving face, exercising rhetorical skill—we are, in this conversation, subordinating all such values to the pursuit of truth. I claim that this norm is a Lewisian convention tied to philosophical practice, i.e., a coordination equilibrium sustained by mutual expectation and mutual preference to conform given that others conform. The sense in which it is tied to philosophical practice is a claim about the history of philosophy as a tradition, which I will postulate rather than defend, as follows.

The truth-norm as historically constitutive of philosophy. Not every conversation that mentions philosophical topics is a ΦDG. People can “talk philosophy” while aiming at entertainment, seduction, political coalition-building, or rhetorical victory. Such talk can be valuable, but it is not inquiry as I conceive of it here. In this essay, I claim that the central, tradition-setting episodes of philosophy plausibly present themselves as ΦDGs, precisely because the truth-norm is established early as the marker separating philosophy/dialectic from sophistry/eristic, and because later philosophers—often explicitly—situate themselves within that inherited practice. Plato’s dramatization of the difference between dialectical examination and eristic victory-seeking is one salient locus of this self-demarcation, and later philosophical culture repeatedly re-invokes the same boundary (under the names “sophistry,” “merely rhetorical,” “content-free,” “verbal dispute,” and so on).

So, although philosophers have not always succeeded in honoring Τ, I claim in this essay that philosophy, as a historically self-conscious practice, has largely understood itself as answerable to it. In the present framework, that self-understanding will be made explicit by treating Τ as a score component: a convention that can be in force, can be challenged, can be violated, and can collapse—yet which, when it holds, makes a dialogue into the special kind of winnable game characteristic of philosophical inquiry.

That is what ΦDGs are: WLGs whose winnability is fundamentally tied to the enforceability of the truth-norm.

A disagreement can be symbolically organized into a zero-sum game, but it needn’t be. Socrates thinks it shouldn’t be, because what we want to work out is not who wins, but who is right. A fight is a conflict of interest, and a disagreement that has been turned into a fight stands at a symbolic remove from the adjudication of the disagreement. When we are working out who wins, we are, at best, pretending to be working out who is right. —Agnes Callard, Open Socrates

Characterization of the truth-norm. Let Τ be the truth-norm. Again, since T is a convention, Τ holds in a dialogue when (i) it is common ground that all parties are bound by it, (ii) all parties expect all others to be bound by it, and (iii) all parties prefer to remain bound by it on condition that the others do.

What does “being bound” amount to? Minimally, Τ imposes regulative constraints on permissible strategic behavior. It does not abolish strategy (one still chooses which objections to press, which clarifications to demand, which examples to introduce), but it forbids strategy that is indifferent to truth. So, among the regulative commitments characteristic of Τ are norms like these:

• Sincerity / intellectual honesty: do not assert what you take to be false; do not defend what you take to be indefensible.
• Retraction-responsiveness: if you come to see that a claim you advanced is false (or unsupported under the current evidential and inferential standards), retract it rather than “saving” it by ad hoc repair.
• Non-eristic constraint: do not aim at victory as such (victory is permitted only as a byproduct of truth-directed correction).
• Charitable uptake: interpret an opponent’s claim in the strongest form compatible with their explicit commitments, because distorting it is a way of “winning” without learning.

Plato’s contrast between dialectic and eristic is a canonical early illustration of exactly this divide: the eristic credo is “win the argument,” whereas the Socratic elenchus is explicitly framed as examination for the sake of truth, including the striking idea that being refuted is, in a sense, a benefit to the refuted party. This is why the truth-norm is what makes a dialogue philosophical in the sense relevant here: without Τ, what you have may still be “a conversation about philosophy,” but it is not yet (or is no longer) philosophical inquiry.

Formal specification of ΦDG score. Formally, we keep the WLG machinery but add a new score component encoding Τ, and we constrain Π and κ in characteristic ways. Let a ΦDG score at stage t be: Sₜ^Φ = ⟨P, 𝒞, Σ, Δ, 𝓑, Γ, Π, κ, 𝓔, 𝓕, Τ⟩. Here Τ can be modeled in (at least) two equivalent-looking ways:

1. As a rule-profile extension: treat Τ as adding rules to Π (or hardening existing ones at high κ), so that the effective convention is Π^Φ = Π ∪ Τ.
2. As a lexicographic constraint on “utility”: treat each player’s “payoff” as ordered lexicographically, with truth-conduciveness first and all other payoffs second: Uᵢ(outcome) = ⟨TruthScore(outcome), SecondaryScoreᵢ(outcome)⟩, where players may optimize secondary goods (humor, elegance, reputation, pedagogical clarity) only among outcomes tied for TruthScore.

Either way, Τ makes certain otherwise-available WLG equilibria unavailable. In ordinary WLGs (political debates, courtroom advocacy, negotiations), “winning” can coherently mean “securing assent,” “shifting burdens,” “forcing a concession,” or “dominating the conversational score” by manipulating what is taken for granted. ΦDGs explicitly outlaw that as a constitutive aim.

Two further characteristic features usually accompany Τ in philosophy:

• Thin presuppositions. Philosophical dialogues tend to begin from a comparatively thin P: far fewer substantive starting points are granted than in special-science inquiry, where a whole research program, measurement tradition, and background theory may be presupposed as common ground.
• High seriousness default. ΦDGs typically pressure κ upward. Demands like “clarify that,” “distinguish these cases,” “that begs the question,” “you haven’t answered the objection,” in the context of ΦDGs, are the natural enforcement mechanisms of Τ, and so they tend to stick (the seriousness ratchet).

Veritism. My claim that the truth-norm is constitutive of inquiry is a form of veritism, a view defended by Duncan Pritchard and criticized by just about everyone else. Here I briefly address those criticisms by, first, making the generalization about them that they generally claim that truth is not a sufficient goal of inquiry. For instance:

• Jonathan Kvanvig, in his paper on Pointless Truth, points out that some truths appear to have no positive value at all, which pressures any view on which truth is always valuable in the same way.
• Benoit Gaultier, in Epistemic Value: The Insufficiency of Truth, defends that mere true belief may have no epistemic value at all, and that truth alone cannot account for epistemic value
• Catherine Elgin, in her book True Enough, argues that epistemic practice (especially in science) makes indispensable use of idealizations, models, and other “felicitous falsehoods,” and that divergence from truth can sometimes foster epistemic functioning rather than undermine it.

I claim that none of these arguments militate against my claims about T. To be more specific about those claims, the truth-norm of ΦDGs requires only:

1. Truth is a necessary aim of inquiry, and
2. Truth is the highest-order constraint (i.e., other epistemic goods are pursued under it, not instead of it).

This is compatible with the thought that not all truths are equally valuable, and that inquiry may have other, secondary goals such as understanding, explanation, and conceptual unification. Pritchard, the main veritist, argues in Veritism and the Goal of Inquiry (2021) that not all truths are equally valuable; similarly, in Understanding and veritism (2024), Pritchard argues that veritism is compatible with valuing ‘understanding’, which is commonly proposed as an alternative goal of inquiry. So veritism, as we understand it and as Pritchard understands it, applies normatively to inquiry without falling to the objections.

In other words, the ΦDG truth-norm does not mean “only truth matters.” It is, rather: nothing else may be pursued at the cost of truth in the conduct of inquiry, though many things may be legitimate subordinate aims.

Victory in ΦDGs. A distinctive fact about ΦDGs is that, in a ΦDG, the only genuine victories are truth-norm victories. That is: a disputant “wins” only by showing that the opponent, if she persists in her initial opinion, must thereby violate Τ—either because (i) her position is internally inconsistent given her own concessions, (ii) her position contradicts presuppositions that are needed to accommodate the fact of ongoing inquiry (for instance, by implying that inquiry is impossible, that reasons are irrelevant, that contradiction is acceptable, etc.), or (iii) her position entails something which is otherwise presupposed, as common ground, to be clearly false.

This is exactly why the Socratic pattern is so central as an archetype: elenchus aims to bring an interlocutor to a point where maintaining the target claim would require maintaining a contradiction with what she has already granted, thereby forcing either retraction or truth-norm violation.

We can encode this in the ΦDG victory condition by refining the WLG Success predicate. Let Τ impose a constraint set on permissible “final stances,” call it Τ-Consistent(Sₜ^Φ). Then define:

Success^{Φ(p, Sₜ}Φ) = 1 iff
(1) p has been defended in compliance with Strict(Π, κ), and
(2) the antagonist’s continued doubt of p would force the antagonist into a violation of Τ given her commitments (or given the ΦDG’s minimal presuppositions).

Equivalently, ΦDG victory can be characterized as forced retraction under Τ. If the opponent refuses to retract even when the ΦDG score makes it clear that persistence would violate Τ, then two things are true at once:

• Procedurally, the ΦDG cannot close cleanly (the conversation may become indeterminate in the Lewisian middle-way sense).
• Normatively, the opponent has ceased to be playing the ΦDG at all—she has defected into a different game (often: eristic, negotiation, status competition).

This also explains why philosophical discussions can feel so “fragile”: once Τ drops out of common ground, the same sequence of moves can still be performed, but their point changes. Refutations become “gotchas,” clarifications become “time-wasting,” objections become “bad faith,” and the entire scoreboard mechanism begins to model, not inquiry, but conflict-management.

8. Philosophical conduct-defense (ΦCD)

Do not block the way of inquiry. —C.S. Peirce’s “first principle of reason”

Philosophical ethics. In this section, I defend the existence of philosophical ethics. In a broad sense, this is already controversial, since many philosophers are moral non-cognitivists, or error theorists, etc., and hence believe that the true philosophy cannot support the truth of any ethical propositions. However, I aim to defend this in a narrow sense in which it is even more controversial, which is that it is possible for philosophical arguments about moral questions to be won in precisely the same way as philosophical arguments about merely speculative questions, or at least in a closely analogous and related way which, if it is not equivalent to the truth-norm itself, is at least as closely connected with the historical tradition of philosophical practice as the truth-norm itself is, or so I claim. The conduct-evaluative convention in what I’ll call philosophical conduct-defense dialogues (ΦCDs, a subtype of CDDs) is the norm that I’ll call the social-pragmatic truth-norm, or SPT.

8.1. Introduction of ΦCDs and the SPT

ΦCDs as CDDs. In §6, we defined conduct-defense dialogues (CDDs) as those WLGs in which an agent defends a compliance proposition:

dᵢ : “Agent i’s conduct (as represented in 𝓗) is in compliance with the going deontic convention Θ (at seriousness κ, given evidence 𝓔).”

A philosophical conduct-defense dialogue (ΦCD) is a CDD whose deontic convention, aside from any local etiquette or parochial role-norms, is a norm-profile that is explicitly answerable to the continuation of inquiry as such. In a ΦCD, the participants are not merely trying to settle whether “you were allowed to do that,” but whether what was done is consistent with the maintenance of a social world in which the truth-norm Τ can reliably govern future inquiry.

Guiding idea of SPT. The truth-norm Τ, as defined in §7, is a dialogical convention: it governs the conduct of argument within ΦDGs. But we often evaluate actions that occur outside explicit philosophical dialogue—lying in private, intimidation in workplaces, propaganda in politics, falsification in science, deliberate confusion in education, and so on. The crucial question is:

Can it be morally permissible to perform acts whose general permission would make Τ socially insecure—thereby undermining the conditions for future inquiry?

SPT answers: no.

The SPT Thesis. A deontic proposition is defensible in a ΦCD only if the conduct it permits or obligates is consistent with (i) the ongoing enforceability of the truth-norm Τ in actual social life, and (ii) the preservation of the social conditions required for future inquiry.

This is not the same as saying “everything must always be inquiry”, since the conduct of other activities does not violate norms of inquiry. It is the claim that inquiry is a standing practice with standing social prerequisites, and those prerequisites are not morally optional. Formally, treat SPT as a special Θ-profile:

Θ_SPT = ⟨N_SPT, ≼_SPT, gate_SPT⟩

• N_SPT includes constraints on conduct that bear on the reliability of truthful cooperation (roughly: norms governing deception, coercion, epistemic sabotage, and institutional corruption of truth-tracking practices).
• ≼_SPT encodes a very strong lexical priority: when SPT is in force, constraints that secure the possibility of truthful inquiry outrank other practical considerations.
• gate_SPT encodes standing: who may issue challenges, what evidence counts, and (crucially) what counts as a competent judge for inquiry-permitting norms.

And a ΦCD score is a CDD score with Θ = Θ_SPT (or with Θ containing Θ_SPT as its highest-order stratum):

S_t^ΦCD = ⟨P, 𝒞, Σ, Δ, 𝓑, Γ, Θ_SPT, κ, 𝓔, 𝓕, 𝓗⟩.

The defended proposition remains a compliance proposition, but compliance is now compliance with SPT:

d_i^SPT : “Agent i’s conduct (in 𝓗) is in compliance with Θ_SPT at seriousness κ given evidence 𝓔.”

This is the sense in which ΦCDs are “philosophical”: they inherit the ΦDG orientation to what makes inquiry possible, but translate it into a conduct-evaluative register.

SPT is stronger than the truth-norm Τ. If philosophical inquiry is governed by Τ, it might seem that we already have what we need for inquiry, which is simply that we should pursue truth when in philosophical dialogue. But that misses the social-pragmatic point: Τ is a convention, and like all conventions, it depends on a background of mutual expectations. I comply with it in part because I expect you to comply with it, and because the payoff structure of the practice makes that mutual compliance stable.

So the vulnerability is obvious: a convention can be undermined from outside the moment of its explicit enactment. If a society permits (or rewards, or tolerates) systematic acts that erode the reliability of truth-directed cooperation, then the convention Τ becomes fragile even inside explicitly philosophical dialogue. In such a society, “Let’s do inquiry” becomes less of a stable coordination equilibrium and more of a naive request in a hostile environment.

This is where I got the idea of calling SPT “social” and “pragmatic”, anyway:

• Social: inquiry is not an abstract relation between a mind and a proposition; it is a practice sustained by institutions, norms, and expectations across time.
• Pragmatic: the relevant commitments show up in what dialogue participants can treat as stable background conditions—what can be presupposed about sincerity, transparency, and good-faith responsiveness, not just in the present conversation but as part of a standing form of life.

SPT therefore goes beyond Τ. Τ says: within this dialogue game, do not assert what you take to be false; retract when refuted; do not “win” by distortion. SPT says: it is morally impermissible to perform acts which, if permitted in general, make it unsafe for anyone to presuppose that others will keep Τ when it matters.

8.2. The insecurity argument and deontological absolutism

Moral norms as highest-order norms. The plan behind SPT depends on a linguistic datum I will simply assume here:

The Highest-Order Norm Thesis. In any given context, “moral” norms are whatever norms have highest practical authority—norms that cannot be overridden by etiquette, aesthetics, convenience, or mere strategic interest. (This is part of why moral rebuke feels categorically different from “that’s unfashionable” or “that’s tasteless.”)

Given that datum, the following argument motivates the absolutist structure of Θ_SPT.

8.2.1. Exceptions undermine the security conditions of inquiry

Suppose we try to make SPT “moderate” by allowing exceptions:

(E) It is morally permissible to violate Τ (or to perform a Τ-undermining act) in context C.

Even if C is described as “rare,” “emergency,” “private,” “non-argumentative,” or “high-stakes,” the crucial point is that other participants cannot, in general, know whether you (are about to) believe yourself to be in C. Humans cannot read minds. They cannot reliably track the private self-ascription of exception-contexts. If exception-permission is a moral permission, then, by the Highest-Order Norm Thesis, it is permission of the highest practical authority.

So if (E) is in force, then every participant in a potential ΦDG must treat it as a live possibility that their interlocutor is about to regard themselves as in C, and hence as morally licensed to abandon Τ when convenient. But that possibility is precisely what destroys the practical basis for the convention: If I cannot safely presuppose that you will not lie as a matter of moral constraint, then my reliance on your assertion is weakened, my willingness to concede is reduced, my suspicion increases, and the entire scoreboard kinematics shifts: more outward expansion of possibility space, more precision-ratcheting, more adversarial posture, less cooperative convergence. In short: the practice ceases to be the stable truth-directed game that ΦDGs were meant to idealize.

Therefore, if we take seriously the idea that inquiry is a standing social practice whose possibility must be preserved, then the permission (E) must be regarded, not just as “risky”, but as incoherent with the social conditions of inquiry.

8.2.2. The ΦCD deontic core is absolutist

The conclusion is not that all morality is exhausted by inquiry-support. The conclusion is narrower:

In ΦCDs, the only defensible norms are those which make the social enforceability of Τ secure, and this security requires context-free prohibitions on conduct that would license defections from truthful cooperation.

This is why SPT yields rules that are not just deontological, but absolutist, to be held inviolable in every context (hence there is no possibility of grounding a “threshold deontology” on SPT). It’s not that this sort of moral rule is in any way glamorous (it is certainly somewhat unpopular among ethicists), but that social epistemic security is brittle. Once you treat truth-violations as morally negotiable, you dissolve the background trust that makes inquiry possible.

SPT builds a moral firewall around the truth-norm: it converts “truthfulness in inquiry” from a merely local convention into something that cannot be overridden by other concerns at any time—because the social assurance it requires cannot itself be contextually toggled on demand.

8.3. Proof in ΦCDs: refutation by SPT-violation

At this point we can state the promised parallel with ΦDG victory.

In ΦDGs, a claim p is “won” when the opponent’s continued denial of p would force them (given their concessions and the thin presuppositions of inquiry) into violating Τ. In ΦCDs, the analogous structure is:

A deontic claim is won when its negation would, if endorsed as a moral permission/obligation, undermine the social conditions required for truthful inquiry, and hence violate Θ_SPT.

This provides a schematic proof-pattern for certain moral rules. For example, consider the moral proposition:

L: “It is morally impermissible to lie.”

The ΦCD strategy is not to appeal to divine command, aggregate utility, or moral intuition. It is to argue:

1. If ¬L were defensible—if it were morally permissible to lie (even “in some contexts”)—then there exists a live moral license for defections from truthful cooperation.
2. Because mind-reading is impossible and self-ascribed exception-contexts are not publicly trackable, such a license makes compliance with Τ socially insecure.
3. Social insecurity of Τ undermines the very possibility of stable inquiry as a standing practice.
4. But ΦCDs, by definition, are conducted under Θ_SPT, whose constitutive point is to secure that possibility.
5. Therefore ¬L is indefensible in a ΦCD; and so L is the only defensible stance under Θ_SPT.

The result is a philosophically characteristic kind of necessity: it is not the necessity of discovering a moral property “out there,” but the necessity of preserving the practical background without which we cannot even stably play the truth-directed game.

Formally, if we let Indefensible_ΘSPT,κ(α) be the predicate from §6.4 (“no compliant defense strategy exists for α”), then ΦCD refutation takes the form:

• show that adopting ¬L would license a class of actions α such that Indefensible_ΘSPT,κ(α) because α undermines the social enforceability of Τ,
• thereby forcing retraction or defection from the ΦCD itself.

This is exactly analogous to ΦDG refutation, except that the thing “violated” is not the local truth-norm of a single dialogue, but the social-pragmatic norm securing inquiry across dialogues and across time.

8.4. Theoretical strength of SPT

Science cannot deliver norms, but this doesn’t imply that philosophy can’t. Philosophy might, not because it can draw on methods inaccessible for science, but because norms deniable only on pain of losing the distinction between valid and invalid, hence between true and false, prove indispensable for science. —Ulrich Steinvorth, A Plea for Naturalistic Metaphysics

SPT is strong in one way and weak in another.

It is strong in that it treats the pursuit of truth as lexically prior: truth-seeking cannot be overridden by other aims in the sense relevant to moral permission. That is an ambitious presupposition. It says: if some practice requires the systematic moral licensing of Τ-violations, then that practice is not morally defensible in a society that aims to preserve inquiry.

But it is also weak as a full ethical theory: one might think that “don’t sabotage inquiry” is thin content for philosophical ethics. Where are duties of beneficence? justice? sexual ethics? distribution? virtue? etc.

Yes, SPT is not a complete first-order ethics. It is a proposal about what can be established philosophically (in the ΦDG/ΦCD sense) by appeal to the presuppositions and social conditions of inquiry. Its content is limited because inquiry’s social prerequisites are limited.

Still, I claim, though I will not defend in detail, that many influential moral theories can be read as attempts, more or less explicitly, to ground normativity in conditions of intelligibility that are tightly connected to truth. For instance, an Aristotelian line can be interpreted as: to understand humans truthfully is to understand them as having a characteristic form of life; normativity then arises from what it is to flourish as the kind of being one is. Similarly, a Kantian line can be interpreted as: to understand an agent as a rational being is to understand them as answerable to norms of universalizability and consistency; the categorical imperative is bound up with what it is to be a participant in the space of reasons. These ideas, when properly understood and accuirately developed, are tying the moral truths about humans to the conditions that are epistemologically (and metaphysically) required to understand them accurately, and hence to have knowledge of the truth, in line with SPT. They combine SPT with a substantive theory of what it takes to know the truth about humans in order to ground a broader range of norms on purely philosophical, inquiry-supporting, grounds.

My broad idea here is that philosophy has repeatedly tried to anchor ethics in the conditions under which persons are intelligible as reason-responsive beings. That is very close to SPT’s thought that the moral minimum is what makes truthful reason-giving socially possible.

There have been other ethical theories, such as utilitarianism, which are harder to interpret as inquiry-bound in this way, and hence, those theories are not philosophical in the present sense. They may be brilliant proposals for how to run conduct-evaluation under some chosen criterion, but they are not, in their core motivation, attempts to secure the social conditions of inquiry. They are better seen as proposals for special Θ-profiles—ways of structuring CDDs around a chosen value (aggregate welfare, preference satisfaction, etc.). They may be more or less well-motivated. But they are not candidates for the deontic convention of ΦCDs as such, because ΦCDs are defined by their constitutive link to inquiry-permitting social conditions.

8.5. ΦCDs as dialogical metaethics

We can now recapitulate our metaethical moral.

In §7, we said that philosophical inquiry is a special argumentative game whose constitutive rule is the truth-norm Τ. In §6, we said that ethical evaluation is, in its home use, a conduct-defense game: normative terms function as claims about defensibility under a deontic convention Θ. This section says: there is a special conduct-defense game—ΦCD—whose deontic convention Θ_SPT is fixed by the requirement to preserve inquiry as a standing social practice.

So the narrow sense in which philosophical ethics exists is this: some moral norms are defensible (and their negations indefensible) precisely because denying them would undermine the possibility of truth-directed dialogue itself. The norms that fall out of this are the ones that secure the background trust and institutional stability without which ΦDGs cannot function as more than theatrical performances.

That is already enough to explain why philosophical ethics is both austere and profound: austere because it yields a limited deontic core; profound because it identifies that core with the very conditions under which anything like philosophy—anything like shared rational correction—can continue to exist at all.

Illustration for this blog post, drawn by Nano Banana.