{"path":"research/a-cut-through-a-concept-is-an-argument.md","content":"# A Cut Through a Concept Is an Argument — the outside reading\n\n**Date**: 2026-09-21 · **Status**: research, against a canon line approved the same day ·\n**Origin**: founder, on the line's approval — *\"Argument-not-a-split is thought-provoking though —\nthink deeply about it and search online for perspectives about this from different academic and\nphilosophical angles. Ultrathink.\"* · **Home of the claim**:\n[core-periphery-and-the-concept-layer.md](core-periphery-and-the-concept-layer.md) § 5a–5a-ii\n\n**The line under examination, now canon:**\n\n> A cut through a concept is an argument, not a split — *causes suffering* plus *causing suffering\n> is wrong* does not entail *cruel*, since a dentist satisfies both, so the pieces fail to\n> reconstitute the parent and are premises rather than constituents.\n\nFive traditions have a view on this and none of them was in the corpus when the line was minted.\nThree support it, and **two complicate it in ways worth a ruling**. The complications are the\npayload; the support is filed so nobody re-finds it.\n\n---\n\n## 1. Dummett's analysis-versus-decomposition is our two kinds of cut, named in 1981\n\nThe sharpest find, and it supplies vocabulary the corpus lacked. Reading Frege, Dummett separated\ntwo operations that both look like taking a thought apart:\n\n- **Decomposition** — forming a complex predicate by treating one or more occurrences of a term in\n  a sentence as replaceable. **There are alternative decompositions of one thought, and none is\n  privileged.**\n- **Analysis** — recovering the thought's *constituents*. **There is at most one correct analysis.**\n\nThat is our distinction exactly, and it explains why the two felt like one act. A **cut between\npropositions** is a decomposition: several cuts of one claim can all be right, which is why the\ncorpus keeps finding that *whether* two things are separable is logic while *how far down to go* is\na function of purpose. A **cut through a concept** claims to give the concept's constituents, which\nis an analysis, and an analysis is the kind of thing that has one right answer or none.\n\n**What this adds to the canon line**: not merely that the two pieces fail to reconstitute the\nparent, but that they were playing a *different game* — one where plurality is a defect rather than\na feature. It also gives the recomposition check its principled reading: the check is asking\nwhether the pieces are constituents, and a decomposition is not obliged to pass it.\n\nSources: [SEP, *Analysis* § 6](https://plato.stanford.edu/entries/analysis/s6.html) ·\n[Bermúdez, *Frege on Thoughts and Their Structure*](https://philpapers.org/archive/BERFOT-5) ·\n[Levine, *Decomposition and Analysis in Frege's Grundgesetze*](https://www.tandfonline.com/doi/abs/10.1080/01445349608837260)\n\n## 2. The paradox of analysis: the same tension, stated in 1942\n\nMoore raised it and Langford named it. Take an analysis *A is C*. Either *A* and *C* mean the same,\nin which case the analysis is a trivial identity; or they do not, in which case it is wrong. **No\nanalysis can be both correct and informative.**\n\nThis is the split-versus-argument problem from the other end. If *cruel* really did mean\n*suffering-causing* + *wrong*, cutting it would tell a reader nothing they did not have; the cut is\ninteresting precisely to the degree that it says something the parent did not — which is the\ncondition under which it is a contribution rather than a separation. **The corpus's own coverage\ninstrument reports that condition as `invented`**, and the paradox says this is not an occasional\nfailure of thick terms but the general shape of conceptual analysis.\n\nSources: [SEP, *Analysis*](https://plato.stanford.edu/entries/analysis/) ·\n[PhilPapers bibliography](https://philpapers.org/browse/paradox-of-analysis) ·\n[a semantic resolution](https://link.springer.com/article/10.1007/s12136-007-0010-0)\n\n## 3. Argumentation theory already treats classification as an inference\n\nWalton's **argument from verbal classification** has the form: *a* has property *F*; for all *x*, if\n*x* has *F* then *x* can be classified as *G*; therefore *a* is *G*. Its sibling, **argument from\ndefinition to verbal classification**, runs the same move with a definition as the licensing\npremise, and Walton's worked case is the redefinition of *planet* to exclude Pluto.\n\nTwo consequences for the build, and both are concrete.\n\n**First, the filing is not novel — it is the field's default.** Calling *the policy is cruel* a\nconclusion supported by *it causes suffering* plus a classification premise is precisely this\nscheme. The canon line is therefore conservative rather than adventurous, which is worth knowing\nbefore defending it.\n\n**Second, the scheme brings critical questions with it**, which is the machinery we already have:\n*is it a matter of degree?* · *is the classification premise based on a definition that is\nadequate?* · *is there room for doubt that the definition fits this case?* A cut through a concept\nre-filed as an argument does not just avoid a bad `DECOMPOSES_INTO`; it lands on a link with its\nown standard challenges. **This is a small build**, since the scheme is already in the set.\n\nSources: [Walton & Macagno, *Reasoning from Classifications and Definitions*](https://link.springer.com/article/10.1007/s10503-008-9110-2) ·\n[Walton, *Arguing from Definition to Verbal Classification: Redefining 'Planet'*](https://philpapers.org/rec/WALAFD-2) ·\n[Walton & Macagno, *A classification system for argumentation schemes*](https://journals.sagepub.com/doi/10.1080/19462166.2015.1123772)\n\n## 4. Definition as a move: the adversarial reading\n\nStevenson's **persuasive definition** gives a familiar word a new descriptive meaning while keeping\nits emotive force, and is typically flagged by *true* or *real*. Zarefsky and Schiappa's **argument\nby definition** is the quieter version: the speaker simply *names* reality and leaves the\ndefinitional premise unexpressed, so there is no proposition to attack.\n\n**This is the adversarial case for the canon line, and it is the strongest one.** A cut through a\nconcept that is stored as a decomposition is an argument by definition with the argument deleted:\nthe definitional premise has become graph structure, which nothing can attack, and the strength\nmachinery then treats it as constitutive. Re-filing it as an argument does not merely classify it\ncorrectly — **it restores the attackable premise that the manoeuvre works by hiding.**\n\nDirectly continuous with the motte-and-bailey reading already in the corpus\n([the-motte-and-the-periphery.md](the-motte-and-the-periphery.md)), where the same move is run live\nagainst a challenger rather than written into a graph.\n\n**The other half of Stevenson's pair (2026-10-07).** He also named the move that keeps what a word\ncovers and changes its standing, a *quasi-definition*; reclaiming a word is the common case. The\ncorpus's kinds of difference between two senses had no place for it: [prime-real-estate-words.md](prime-real-estate-words.md) § 4.\n\nSources: [Macagno & Walton, *The Argumentative Structure of Persuasive Definitions*](https://www.researchgate.net/publication/225817244_The_Argumentative_Structure_of_Persuasive_Definitions) ·\n[*Persuasive Definitions*, Informal Logic](https://informallogic.ca/index.php/informal_logic/article/view/594/527) ·\n[Aberdein, *Persuasive Definition*](https://philpapers.org/rec/ABEPD)\n\n## 5. Parts versus premises has no settled formal test — the first complication\n\nThe corpus treats the recomposition check as *the* discriminator. Argumentation theory has been\narguing about the same boundary for forty years under the name **linked versus convergent**: in a\nlinked argument the premises work together and none supports the conclusion alone; in a convergent\none each gives an independent reason. Walton sorts the rival tests along two axes\n(falsity-versus-suspension, no-support-versus-insufficient-proof), and the honest summary from the\nfield is that **the criterion is highly controversial with close to a dozen proposals**, and at\nleast one respected paper argues for **not making the distinction at all**.\n\n**What this costs us**: the colimit check is one criterion among many, not the discovery of a fact.\nIt is a good one — it is mechanical, it runs unanchored *(⚠ 2026-10-10: it does not. The rebuilding is\nthe first answer of a call whose prompt also holds the original:\n[quality-control-for-deeper-steps.md](quality-control-for-deeper-steps.md) § 1.3)*, and it is cheap — but a proposal it re-files\nhas been re-filed *by our criterion*, and a careful reader may hold another. Two consequences worth\ncarrying: the re-filing should be visible as a judgement rather than presented as a correction, and\n**reviewer disagreement about a re-filing is the expected case rather than a defect**, which is the\nsame conclusion § 3a of `earning-the-right-to-write.md` reached about how far to cut.\n\nSources: [Walton, *The Linked-Convergent Distinction — Refining the Criterion*](https://link.springer.com/chapter/10.1007/978-94-007-0357-5_6) ·\n[Goddu, *Against Making the Linked-Convergent Distinction*](https://www.researchgate.net/publication/226710280_Against_Making_the_Linked-Convergent_Distinction) ·\n[*Identifying Linked and Convergent Argument Structures*](https://informallogic.ca/index.php/informal_logic/article/download/7133/5419)\n\n## 6. The thick-concept debate is live, and one 2021 result pushes back — the second complication\n\nThe canon line leans on McDowell's **shapelessness**: the descriptive extension of a thick term\ncannot be specified without the evaluation, so you cannot peel one off the other. That is already in\nthe corpus (§ 2 here, and `philosophical-foundations-stress-test.md` § 6) and it is contested from\ntwo directions.\n\n**Väyrynen** has argued at length that shapelessness does not deliver what its users want from it,\nand that thick terms' evaluative content may be pragmatically conveyed rather than semantically\nencoded — which would make the evaluative side *detachable* after all. And **Willemsen et al.\n(2021)** ran the experiment: *Separating the Evaluative from the Descriptive* reports that subjects\ndo separate the two components of thick terms under test conditions.\n\n**What this costs us**: the canon line's *usually* is load-bearing and should stay. Where a thick\nterm's evaluation turns out to be conveyed rather than encoded, a cut through it may genuinely be a\ndecomposition, and our rule would then re-file a legitimate split as an argument. The\nrecomposition check handles this correctly without needing the philosophy settled — it will simply\npass — which is a point in favour of a mechanical test over a doctrine about thick terms.\n\nSources: [SEP, *Thick Ethical Concepts*](https://plato.stanford.edu/entries/thick-ethical-concepts/) ·\nVäyrynen on shapelessness (cited by name; the URL was not verified in the search that found it) ·\nWillemsen, *Separating the Evaluative from the Descriptive* (2021), same caveat — **both are read\nfrom secondary description rather than from the papers, and neither should be cited outward until\nsomeone opens them.**\n\n## 7. The unasked question: is a cut through a concept a *report* or a *proposal*?\n\nEverything above assumes the cut is a claim about what a word **does** mean. Conceptual engineering\nis the position that the interesting act is saying what a word **should** mean — Chalmers's three\nstages are designing a concept, evaluating it, and implementing it — and its proponents argue the\nnormative element was always there in analysis too, with engineering's virtue being that it\n**makes the normativity explicit**.\n\n**This is a third option the canon line does not have a slot for**, and it is the one worth a\nruling. *Cruel means suffering-causing and wrong* can be:\n\n| Reading | What it is | Where it belongs |\n|---|---|---|\n| a **report** about usage | an empirical claim about how the word is used | claim layer, checkable against use |\n| an **argument** | a premise licensing the classification | claim layer as support — the canon line |\n| a **proposal** | *let us use the word this way* | the concept layer, as a sense with an author |\n\nThe corpus already has the machinery for the third: a sense is somebody's delta from a shown\nreference, attributed (§ 5b). And the three-layer table in § 5a-ii already sends *cruel means\nsuffering-causing and wrong* to the concept layer as a criterion — so the canon line and the\nproposal reading may be describing **different pieces of the same cut** rather than competing, which\nis exactly the resolution § 5a-ii reached for the other two pieces.\n\n**Unruled, and the sharpest thing this reading surfaces**: whether a proposed criterion should be\nmarked as report-or-proposal when it lands in the concept layer, the way a horizon reading is marked\nas a reading. A criterion nobody marked reads as a report of usage, which is the argument-by-\ndefinition move at the concept layer.\n\nSources: [Chalmers, *What is Conceptual Engineering and What Should It Be?*](https://consc.net/papers/engineering.pdf) ·\n[Isaac, *Conceptual engineering: a road map to practice*](https://compass.onlinelibrary.wiley.com/doi/10.1111/phc3.12879) ·\n[*Conceptual engineering is old news*](https://academic.oup.com/pq/article/76/1/320/7724643)\n\n---\n\n## What survives, and what changes\n\n**Survives, and is better supported than when it was minted**: the line is the field's default\nfiling rather than an invention (§ 3), it restores an attackable premise that a known rhetorical\nmanoeuvre works by deleting (§ 4), and Dummett gives it a principled reading in which the two cuts\nare different operations rather than a good and a bad version of one (§ 1).\n\n**Changes, and both are small:**\n\n1. **State the re-filing as a judgement by a named criterion, never as a correction.** The\n   parts-versus-premises boundary is unsettled in the field (§ 5), so our recomposition check is one\n   criterion and should say so wherever a re-filing is shown.\n2. **Keep *usually*.** One 2021 experiment separates thick terms' components under test conditions\n   (§ 6), so the mechanical check should decide each case rather than a doctrine deciding all of\n   them in advance.\n\n**Three buildables, cheapest first:**\n\n- **A re-filed cut should land on `argument_from_verbal_classification`**, which is already in the\n  scheme set and brings its own critical questions — so the re-filing produces open questions rather\n  than merely avoiding a wrong edge.\n- **The failed-reconstruction re-filing itself** is still proposed and unruled\n  (§ 5a-ii): nothing acts on `CoverageReport.reconstructed` today.\n- **Mark a concept-layer criterion as report or proposal**, per § 7.\n\n## Cross-references\n\n[core-periphery-and-the-concept-layer.md](core-periphery-and-the-concept-layer.md) § 5a–5a-ii (the\nclaim's home) · [the-motte-and-the-periphery.md](the-motte-and-the-periphery.md) (the same move run\nlive) · [decomposition-axes.md](decomposition-axes.md) (the eight axes; losslessness as a colimit) ·\n[the-jhu-decomposition-line.md](the-jhu-decomposition-line.md) (coverage, coherence, atomicity) ·\n[earning-the-right-to-write.md](earning-the-right-to-write.md) § 3a (why reviewer divergence marks\nan unstated purpose) · [philosophical-foundations-stress-test.md](philosophical-foundations-stress-test.md)\n§ 6 (thick predicates, the July ruling) · [scheme-set-exhaustiveness.md](scheme-set-exhaustiveness.md)\n(the open curated set the classification scheme sits in)\n"}