--- title: "Bridge Principles without UG: Projectibility, Crosslinguistic Evidence, and Linguistic Ontology" author: "Brett Reynolds" year: "2026" status: "Preprint" canonical_url: "https://lingbuzz.net/lingbuzz/010104" website_url: "https://brettreynolds.ca/papers/bridge-principles-without-ug/" markdown_url: "https://brettreynolds.ca/papers/bridge-principles-without-ug/paper.md" version: "author-manuscript mirror" version_date: "2026-06-25" keywords: ["Universal Grammar", "projectibility", "linguistic ontology", "crosslinguistic evidence", "bridge principles", "comparative method"] --- # Bridge Principles without UG: Projectibility, Crosslinguistic Evidence, and Linguistic Ontology **Author-manuscript mirror.** This Markdown file is provided for accessibility, search, and machine readability. The canonical public record is linked in the metadata above. ## Abstract Reiss (2026) gives Universal Grammar two roles. In its modest role, UG is a background commitment of Chomskyan I-language inquiry: anyone modelling individual internal grammars with universal primitives is committed, explicitly or implicitly, to such a postulate. In its stronger role, UG is the condition under which data from French, Japanese, or Icelandic can bear on English, so that anyone who denies UG lacks grounds for importing analytic categories across languages. I grant the local stipulation for Chomskyan inquiry, but I deny that inquiry into individual internal grammatical knowledge as such requires UG. I accept the evidential burden but reject the exclusivity. Evidence travels by bridge principles, scoped warrants that specify a projection target, the profile connecting observed and projected features, the diagnostics that locate it, and the defeaters that limit its scope. The test separates three questions the chapter runs together: whether human languages form an inquiry domain, when evidence can travel between them, and whether UG is the only licence for that travel. Some warrants may be Chomskyan; others are causal-historical, phonetic-perceptual, typological, cultural-transmission, or institutional. Applied to Reiss’s French-English *any* case, the French contrast is only a cue until English-internal diagnostics and a support profile are supplied. Historical-comparative reconstruction supplies a worked counterexample to the unscoped claim: common descent and regular correspondence license bounded vertical evidential transfer without UG as the transfer warrant. The same discipline makes public standards, corpus profiles, typological comparanda, and norm-governed varieties legitimate targets when their projections are declared. UG may license some I-language inferences; it doesn’t own the need for warrant. **Keywords:** Universal Grammar; projectibility; linguistic ontology; crosslinguistic evidence; bridge principles; comparative method # Introduction: Reiss’s Challenge I reply to Reiss’s defence of armchair linguistics as empirical theoretical inquiry (Reiss 2026). The chapter’s target isn’t comparison as such, but unlicensed comparison: the use of evidence from one language to analyse another without an account of why the evidence should travel. Reiss’s positive answer is Chomskyan Universal Grammar (UG), understood as the genetically determined Human Language Faculty that supplies the common primitives of human languages. At the centre of the chapter is a Newtonian analogy. Reiss’s thought is that apples, tides, and planets become a common empirical domain when mechanics supplies a shared level of analysis; likewise, English, French, Japanese, and Icelandic become mutually evidential when UG supplies the shared primitives of individual internal grammars, or I-languages. On that view, everyday language names such as *English* and *French* don’t mark the scientific objects directly. The relevant objects are individual internal languages analysed against a universal inventory. Reiss also narrows the people whose views are under discussion. His stipulation itself marks the word typographically: “the views of ‘linguists’ in this chapter” are the views of “Chomsky and those who broadly adopt his philosophical stances” (Reiss 2026, sec. 2). I accept that local stipulation. I too will employ typography. In this paper, I retain Reiss’s ‘linguists’, with quotation marks, to refer to the stipulated Chomskyan group. But linguists will refer to researchers whose work is organized around systematic inquiry into human language phenomena, including historical linguists, typologists, sociolinguists, corpus linguists, descriptive fieldworkers, language philosophers, and Chomskyan syntacticians. I’m targeting the chapter’s written claim, not Reiss’s psychology. I don’t claim that he personally dismisses historical linguistics, phonetics, corpus work, public-language inquiry, or the study of language use, and I don’t deny the local claim that Chomskyan I-language inquiry rests on a UG postulate. The issue is the stronger conclusion the chapter draws from that local frame. If the claim stays local to Chomskyan I-language inquiry, it’s compatible with the projectibility account defended here; if it’s a general claim about linguistic evidential travel, it needs an exclusion argument against rival warrants. My disagreement begins when the stipulation and the UG postulate are made to do global work. Reiss’s conclusion isn’t only that UG can license some crosslinguistic inferences within a Chomskyan programme. It’s unrestricted: “Anyone who denies UG has no grounds for importing analytic categories from one language to the study of another. In this sense, the Chomskyan, UG-based perspective is explicitly more richly empirical than competing theoretical approaches” (Reiss 2026, sec. 6). The quantifier is *anyone*, and “competing theoretical approaches” reaches past Reiss’s stipulated ‘linguists’ to the rest of the field. That’s the monopoly claim at issue here. Reiss’s monopoly claim has two faces. One is evidential: why should French evidence bear on English, or Japanese evidence on Icelandic? The other is classificatory: why are French, Warlpiri, and Swahili in the same domain of inquiry while chess, Python, and tango aren’t? Reiss is right that neither question can be waved away. The mistake is to treat the need for a domain and the need for transfer as if both had already been answered by the word *UG*. Projectibility supplies the alternative: what observing some features licenses us to expect about others (Goodman 1955; Reynolds 2026a). Some features support predictions; others don’t. A projectibility-first linguistics asks what a proposed linguistic kind lets us infer, which profile supports that inference, and where the projection fails. Universal Grammar can be part of a proposed support story. It isn’t the condition of possibility for linguistic comparison. Section 2 grants Reiss’s central methodological point: crosslinguistic evidence needs a warrant. Section 3 separates domain formation, evidential transfer, and exclusion, quotes the chapter’s exclusivity statements, and shows why a local stipulation about ‘linguists’ can’t carry a field-wide conclusion. Section 4 examines the retreat to I-language inquiry and argues that the local claim is either stipulative or unsupported. Section 5 introduces projectibility as the audit framework for bridge principles. Sections 6 and 7 apply the framework, first by demoting Reiss’s French-English *any* example to cue status and then by giving historical comparison as a worked non-UG warrant, vertical within a family, with horizontal transfer between unrelated languages marked as unfinished. Section 8 extends the same discipline to public and distributed linguistic objects. # What Reiss Gets Right Reiss’s best point is methodological. Theoretical linguistics can be empirical without being experimental, corpus-based, or statistical. Good armchair work isn’t mere introspection detached from the world. It’s a form of theory-mediated empirical reasoning in which judgements, contrasts, paradigms, and crosslinguistic patterns are made to bear on abstract structure. Categories such as word, sentence, negative-polarity item, segment, syllable, and interrogative element aren’t read directly from the acoustic signal, the printed page, or a spreadsheet of corpus counts. A statistician still has to know what’s being counted. A corpus linguist still has to decide what counts as a token of the relevant sort. A psycholinguistic experiment still has to decide which contrast the stimuli instantiate. Crosslinguistic evidence needs a warrant. If French data is used to analyse English, or Japanese data is used to analyse English and Icelandic, then the inference isn’t justified by the mere fact that all are called languages. Reiss’s examples of empirical argumentation devices put pressure on any view that treats crosslinguistic comparison as automatic. Nor should the reply pretend that UG-style research never provides content. A Chomskyan account can offer precise hypotheses about locality, binding, polarity, category features, movement, or possible dependencies, and those hypotheses can generate crosslinguistic expectations. The objection is narrower and stronger. Where thick, audited UG theories succeed, they succeed for I-language targets; that success doesn’t establish that UG is the only possible licence, nor that mental-grammar inquiry as such requires UG. Reiss himself grants the distinction: “belief in I-language does not entail belief in UG-object” (Reiss 2026, sec. 3). The transfer-licensing role he assigns attaches to UG-object specifically, not to I-language inquiry, and not to the bare postulate, which by the dilemma below licenses no particular transfer at all. These concessions narrow the dispute. It isn’t about theory, realism, armchair method, or comparison. Two questions remain: whether the evidential-warrant burden has only one possible answer, and whether I-language is the only legitimate target of inquiry. # From Domain Postulate to Transfer Monopoly Call the stronger conclusion the monopoly claim. The label isn’t a claim about Reiss’s intentions; it names the chapter’s written move from UG as a domain-forming postulate for I-language inquiry to UG as the exclusive warrant for importing analytic categories across languages. Reiss’s monopoly claim comes from a chain of roles assigned to UG, not from one local observation. In the opening frame, UG is the innate Human Language Faculty that makes human languages a natural domain. In the discussion of empirical argumentation devices, UG is the postulate under which judgements and crosslinguistic observations count as empirical evidence. In the conclusion, UG is a background assumption, a null hypothesis, and a licence for treating each language as potentially relevant to all others (Reiss 2026, secs. 1, 5–6). Those claims sit close together in the chapter, but they aren’t equivalent. Reiss’s Newton analogy has to be kept in view. In his setup, Newton’s postulate doesn’t treat apples and planets as everyday objects that happen to be similar. It places them under a common theoretical description, with common primitives, measures, equations, counterfactual expectations, and conditions under which the theory would be strained. The proposed linguistic parallel is that Quechua, Hungarian, French, and English aren’t the scientific objects directly; the objects are I-languages analysed with UG-supplied primitives. So Reiss has a powerful argument for a research domain. He doesn’t yet have an argument for a transfer monopoly. Three claims have to be separated: that human languages form a domain, that evidence may sometimes travel across members of that domain, and that only UG can license the travel. The exclusion claim is explicit. Of the Japanese evidence Reiss writes that without UG there’s “no reason to think that data from Japanese should be any more relevant to the choice of which English grammar is correct than data concerning bird migration or soap bubbles” (Reiss 2026, sec. 5.3.1). Of the French-English *any* argument he asks, “What licenses this reasoning?” and answers, “Only the belief in Universal Grammar allows us to construct this kind of argument” (Reiss 2026, sec. 5.3.3). A domain postulate may defend the first claim; the second requires a transfer principle, and the third requires an exclusion argument against rival warrants. Reiss carefully separates UG-object, UG-theory, and I-language; what the chapter doesn’t separate is domain formation, transfer, and exclusion, and at secs. 1 and 5–6 the domain postulate is left to discharge all three. A general postulate of shared human linguistic endowment may define a research domain for ‘linguists’, but it doesn’t by itself say which comparison is licensed, what the projection target is, how far the evidence travels, or when the inference fails. Nor does it explain why conclusions licensed inside an I-language programme should project onto historical linguists, typologists, sociolinguists, corpus linguists, descriptive fieldworkers, language philosophers, or construction grammarians. A licence that never specifies its failure conditions is too permissive to do the work Reiss assigns to it. A familiar retreat doesn’t help. If UG merely makes crosslinguistic evidence potentially relevant, the monopoly conclusion doesn’t follow: potential relevance isn’t a licence for any particular transfer. If, instead, a substantive UG theory licenses a particular transfer, then UG is no longer functioning as a bare domain-defining postulate. It’s a first-order warrant with its own target, support, scope, and failure conditions, answerable to the same audit as any rival. A thin UG postulate can’t license the monopoly. A thick UG theory may well license particular transfers; what being thick doesn’t license is the exclusion of non-UG warrants for other targets. Either way, the exclusion step needs an argument the chapter doesn’t supply. Once the thin/thick dilemma is visible, the Newton comparison turns against the monopoly. Physics transfer principles are answerable to scope, measures, counterfactual expectations, and breakdown conditions. Reiss’s UG postulate is treated as if it also generated jurisdiction: a password that admits preferred transfers and excludes rival licences. It’s asked to do the grouping work, the transfer work, and the exclusionary work at once. A substantive UG theory may do some of that work for a particular comparison. It may predict that a dependency should be constrained by locality, that an apparent lexical item should decompose into two grammatical objects, or that a surface contrast should reflect an unpronounced structural contrast. But then the warrant comes from the specified theory, diagnostics, and failure conditions, not from the bare fact that UG has been postulated. The examples show that Chomskyan ‘linguists’ have found fruitful comparisons. They don’t show that all legitimate comparisons require UG. The empirical argumentation devices don’t need to be rejected. Homophony, hidden structure, distributional contrast, and crosslinguistic prompts are ordinary forms of linguistic reasoning. Their legitimacy depends on the target and on the warrant that connects the prompt to the proposed analysis. A Chomskyan theory may provide such a warrant, but the device itself doesn’t carry a UG credential. If the support comes from crosslinguistic recurrence, diachrony, contact history, or language-internal distribution, the argument remains empirical without becoming Reiss-style UG. Nor does the domain question force the monopoly conclusion. Human languages can be grouped by a recurring profile that supports projections without treating UG as the essence of the group. Section 5 makes profile precise; here the word marks the recurring pattern that supports projection. The pattern includes productive form-interpretation pairings acquired as primary communicative systems, externalized in speech, sign, writing, or other modalities, socially transmitted and norm-governed, historically variable, and available for translation, paraphrase, repair, error, correction, and grammatical judgement. The cluster is stabilized by biological endowment, acquisition and uptake, communicative use, social transmission, norm-governed reproduction, and repair under error. Nothing in that profile smuggles in the Chomskyan conclusion. Form-interpretation pairings and grammatical judgements presuppose grammaticality as a projectible property, not UG as its explanation. Even if a UG theory best explained part of the cluster’s cohesion, that would help individuate the domain; it wouldn’t license transfer from one language to another without a first-order transfer warrant. Chess, Python, and tango share some traits with human languages. They have conventions, learnability, sequencing, and in some cases formal syntax. But they don’t support the same projection package: native-language acquisition, ordinary open-ended communicative uptake, historical sound and morphosyntactic change, indexical register effects, crosslinguistic translation, and grammar-internal acceptability contrasts don’t travel to them as a set. They may be valuable comparanda. They aren’t thereby co-domain targets for the same linguistic projections. Projectibility-first linguistics can ask the missing questions directly. It can distinguish a claim about mental grammars from a claim about a public standard, a corpus profile, a typological comparandum, a diachronic pathway, or an institutional norm. Reiss’s ‘linguists’ can handle the first target on their own terms. They can’t infer from that success that the other targets are illegitimate, or that linguists working on them lack grounds for comparison. Reiss is free to use ‘linguists’ for Chomskyan-internal theorists, and claims about ‘linguists’ may show what follows for that subkind. But before UG is credited with field-wide evidential rights, the chapter has already marked some opponents as scholars who “self-identify as linguists” (Reiss 2026, sec. 2). That phrasing matters only for scope: a conclusion about Reiss’s stipulated ‘linguists’ can be correct without being a conclusion about linguists, linguistic evidence, or linguistic comparison as such. Trouble starts when a local stipulation is made to do global work. Reiss makes the trouble visible himself. Arguing that opponents secretly rely on UG, he writes that “no linguist works in this way, despite their anti-UG protestations” (Reiss 2026, sec. 5.3.1), where working “in this way” means treating each data point as relevant to a single language. Read with the stipulated ‘linguists’, the sentence is incoherent, because Chomskyans make no anti-UG protestations and the clause has no referent. It parses only in the broad sense, with linguists including the deniers, and then it asserts what the chapter never argues: that every linguist covertly depends on UG. A local stipulation can’t carry that conclusion, and the broad reading the sentence forces is the field-wide claim left unsupported. The French case has the same shape. “Only the belief in Universal Grammar allows us to construct this kind of argument” leaves *us* unspecified, and Reiss’s footnote concedes the device is a “common” one (Reiss 2026, sec. 5.3.3), used across the field rather than as a Chomskyan signature. So *us* either picks out the stipulated ‘linguists’, in which case the exclusivity claim doesn’t reach the common practice it’s about, or it picks out everyone who uses the device, in which case it’s again the unsupported universal claim that only UG could license it. Either reading defeats the monopoly: the stipulated one is too narrow to license the field-wide conclusion, and the broad one asserts that conclusion without argument. # The Local Retreat: I-language, Mental Grammar, and UG A retreat to I-language narrows the issue but doesn’t settle it. If *I-language* is a term of art for the attained state of a UG-governed language faculty, then the claim that I-language inquiry presupposes UG is true by stipulation: it says that Chomskyan inquiry uses Chomskyan background commitments. That has no force against inquiry into individual internal grammatical knowledge conducted under other assumptions. If, instead, *I-language* is taken in its less proprietary paraphrase as mental grammar, then UG isn’t built into the target. Reiss grants the distinction himself: one could believe in mental grammars that “arise through ‘general learning mechanisms’ without any language-specific categories”, so “belief in I-language does not entail belief in UG-object” (Reiss 2026, sec. 3). So the substantive question is whether mental grammars can be studied without a genetically specified universal inventory of grammatical primitives. The nativist’s own equation makes the stipulation visible: Reiss endorses Jackendoff’s “Mental Grammar = UG + Experience” (Reiss 2026, sec. 3), on which mental grammar is UG’s output and the dependence is built in by definition. Strip the definition and the question is empirical. Usage-based, constructionist, cognitive-grammar, and processing-oriented work answers it: these approaches model individual internal grammatical knowledge without a Chomskyan inventory (Goldberg 2006; Bybee 2010; Langacker 1987; Kempson et al. 2001). Dąbrowska (2012) documents stable individual differences in attained grammatical knowledge, and Dąbrowska (2010) separates naive from expert acceptability intuitions; neither presupposes UG. They may be right or wrong, but they aren’t unintelligible. These approaches needn’t study *I-language* in Reiss’s proprietary sense; he’s free to reserve the term. They investigate the target the term was meant to secure: individual, internal grammatical knowledge. Even researchers who keep a strong mentalist commitment can drop the specific UG architecture. The Parallel Architecture and Simpler Syntax reject central parts of the mainstream Chomskyan syntax-first picture while staying explicitly mentalistic (Culicover and Jackendoff 2005), which shows that mental grammar doesn’t pin down thick UG even for a nativist-friendly theorist. These approaches don’t escape background assumptions, and they don’t pretend to. They appeal to learning, memory, entrenchment, analogy, categorization, processing, communicative use, and social uptake. But those are bridge principles in the sense developed below: scoped warrants for projecting from observed behaviour, judgement, distribution, acquisition, or processing to claims about internal linguistic knowledge. The need for a warrant doesn’t entail the UG warrant. So the local claim is caught in a dilemma. Read with the proprietary *I-language*, it’s stipulative and carries no weight against other approaches; read with mental grammar as the target, it’s a substantive empirical claim the chapter doesn’t establish. # The Projectibility Framework An evidential warrant has to say why this evidence matters for that claim. A bridge principle names the target of the inference, the pattern that makes the inference plausible, the signs that locate that pattern, and the conditions under which the inference should stop. By evidential transfer I mean the familiar practice of treating evidence from one language, variety, modality, period, or domain as bearing on the analysis of another. By bridge principle I mean the warrant that makes such a move legitimate. The labels are mine; the obligation isn’t. Any account that lets French data bear on English, Japanese data bear on English, or Sanskrit data bear on Proto-Indo-European needs some account of why the evidence travels. Declaring a bridge has a positive payoff before it’s critical: a bridge says what observing some features licenses you to expect about others, and where that expectation stops. Demoting an unwarranted transfer is the corollary, not the headline. First fix the target. A projection is an inference from observed features to expected features: from a diagnostic pattern in the source material to a claim about some target. The target may be an individual mental grammar, a public standard, a corpus profile, a typological comparandum, a diachronic pathway, an institutional norm, or a social practice. The target matters because the same observation can licence one projection and fail to licence another. Call the corresponding worldly pattern a profile. It makes a projection worth trying because it includes properties and relations, not just shared labels. The relevant question is why observing one feature gives reason to expect another. At minimum, the audit needs three pieces. The target says what the inference is about. A diagnostic is evidence that the profile is present. Scope conditions and defeaters say where the projection applies and what would block, localize, or demote it (Reynolds 2026a). Later examples add a further question: what ordering relation keeps the source and target properties connected rather than accidentally associated? An ordering relation is whatever explains their co-availability, causal, compositional, distributional, or developmental, depending on the case. Those pieces support different grades of inference. A diagnostic cue identifies a candidate contrast but doesn’t by itself license transfer. A scoped bridge identifies the target and support profile clearly enough to license a local projection. A robust bridge earns broader use only when it survives independent cases, held-out contrasts, and known defeaters. These grades keep cue, warrant, and scope from collapsing into one another. A corpus frequency difference may cue a register contrast. It becomes a scoped bridge when the conditioning structure is identified and the projection survives held-out material: for instance, knowing the institutional setting predicts likely forms, and hearing the forms helps infer the setting. It becomes robust only if the same profile continues to support projections across genres, speakers, institutions, or periods while respecting declared defeaters. The point isn’t to inflate every coding decision into metaphysics but to keep evidential travel inspectable. Put schematically, a bridge principle is a scoped licence from observed diagnostics to a projected target through a proposed profile: if source features S identify profile P under conditions C, then target expectation Q is licensed until defeaters D appear. Projectibility is a second-order discipline for evaluating such warrants, not a rival language faculty. UG is one possible first-order filling of the schema. Recurrent task structure, diachronic pathway, processing pressure, acquisition bottleneck, social uptake, and institutional maintenance can also fill it when they connect source and target for the declared inquiry. Haspelmath’s comparative concepts and Croft’s Radical Construction Grammar already supply non-UG discipline for crosslinguistic work (Haspelmath 2010; Croft 2001). Comparative concepts make comparison accountable without treating every useful category as an innate universal: a comparative concept licenses comparison and nothing more, never the import of one language’s category into another. Radical Construction Grammar treats categories as language-specific while preserving comparison through shared meaning and typological distribution. That stance meets Reiss’s reading of recurrent categories head-on: where Reiss takes the recurrence of a category such as the negative-polarity item or the interrogative word across languages to be evidence of a shared innate inventory, Croft takes it to be a pattern of meaning and distribution over language-particular categories. These approaches don’t by themselves license evidential transfer, but they show that crosslinguistic discipline doesn’t require a universal inventory of Chomskyan primitives, and that the demotion verdict reached below for the French *any* case follows from standard typological caution, not an ad hoc projectibility manoeuvre. One non-UG ordering relation is cultural transmission. When a system is repeatedly learned from partial evidence and reused for communication, transmission can favour patterns that are easier to learn, remember, extend, and reuse across recurrent tasks; patterns that fail those pressures are more likely to be regularized, lost, or reanalysed (Kirby et al. 2008, 2015, 2014). This makes recurrent structure constrained without making it automatically UG-theoretic. The bridge runs through bottlenecked cultural transmission, ordinary learning biases, and recurrent communicative tasks. An external analogy helps separate projectibility from pre-specification. Old Assyrian trade records from Kanesh support a structural gravity model because long-distance trade had a stable problem profile: goods, routes, distance-sensitive costs, price differences, contracts, credit, courts, taxes, and repeated shipment decisions (Barjamovic et al. 2019). The documents didn’t encode a gravity model, but the recurrent trade ecology generated records structured enough for later investigators to recover trade links and estimate the locations of lost cities. An analogy of this kind doesn’t explain grammar. It shows how recurrence, recordability, and uptake can make public semiotic structure projectible without treating the projected forms as designed or pre-specified. The same discipline separates three operations. Comparison places phenomena under a declared description. Classification assigns a language-internal realization to the comparandum. Evidential transfer treats evidence about one realization as support for an analysis of another (Reynolds 2025). The third begins only when the comparative setup is embedded in a profile with projection, support, scope, and defeaters. Nothing in this framework opposes universal explanation. A bridge principle can be universal without being UG in the Chomskyan grammatical sense. Articulatory and perceptual facts can license crosslinguistic expectations about sound patterns (Ohala 1990); acquisition pressures can license expectations about learnable contrasts; communicative or processing pressures can support some typological projections; common descent can license historical reconstruction; institutional maintenance can license projections about public standards. These warrants differ in target and scope. Some are biological, some social, some historical, some grammatical. The point isn’t that UG never licenses transfer, but that universality and UG aren’t coextensive. # A Worked Audit: French Evidence for English Polarity Reiss’s own French-English example is the best test case because it turns the abstract monopoly claim into an actual inference. He presents it as one of his empirical argumentation devices, under the heading “French data is English data is Japanese data”. The question is whether evidence from French translations can help decide the analysis of English *any* (Reiss 2026, sec. 5.3.3). I don’t treat the example as a full analysis of French polarity or English *any*; Reiss presents it as a schematic argumentation device, and that’s exactly why it’s useful. A schematic argument makes the warrant visible. French can prompt the English split analysis, but the split becomes warranted only when English-internal diagnostics, a support profile, scope conditions, and defeaters are supplied. English *any* occurs in negative-polarity environments[^2] and in free-choice uses. Roughly, *I didn’t eat any cookies* illustrates the polarity-dependent use: *any* depends on a licensing environment such as negation. *Choose any card* illustrates the free-choice use: any member of the relevant set is permitted. French, in the relevant standard examples, uses forms such as *rien* in negative contexts and *n’importe quoi* in free-choice contexts. Reiss uses the French contrast to argue that English has two forms pronounced *any*, and then asks what licenses French evidence for English. His answer is UG. Projectibility-first reconstruction keeps the useful inference while denying the monopoly. The target isn’t “English as such”; it’s the analysis of English *any*: whether a single surface form realizes one lexical-semantic item or two. The comparandum is the semantic contrast between polarity-dependent uses and free-choice uses. The language-internal realizations include English *any*, French *rien*, and French *n’importe quoi*. Classification of English *any* as one item or two is exactly what has to be argued for; it can’t be read off the French split. Diagnostics remain local. Negative contexts, questions, conditionals, downward-entailing contexts, and related environments diagnose the negative-polarity side; downward-entailing contexts are environments where truth over a larger set carries truth over a relevant subset. Free-choice readings, modal environments, and indifference readings diagnose the other side. Translation into French is a diagnostic cue because it shows that the semantic-pragmatic contrast can be lexicalized separately, not because it transfers a French category into English. A simpler case shows why that restriction matters. French expresses age with *avoir*: *j’ai cinquante-sept ans*. English has *have*, numerals, and *years*, and *I have five years* is grammatical when it describes a duration, as in time remaining until retirement. But as a response to *How old are you?*, \**I have fifty-seven years* is ungrammatical on the intended age-predication reading. French evidence can bear on English here only after the target is named: age predication, duration expression, calque, bilingual transfer, or translation practice. The French construction is a cue to a contrast, not a licence to import the French analysis into English. Turkish evidentiality sharpens the reductio. Turkish has grammaticalized evidential morphology in its verbal system, with a direct/witnessed form contrasting with indirect uses associated with report or inference (Sener 2011). English can translate such clauses with a simple past such as *Ali left*, adding *apparently*, *I hear*, or *I saw it happen* only when information source matters. Turkish evidence doesn’t show that English *left* is secretly two past forms, one witnessed and one inferential. It shows that Turkish grammaticalizes a contrast English may express lexically, pragmatically, or not at all. Reiss needn’t bite that bullet. He can say that Turkish evidence becomes relevant only when a UG theory and English-internal diagnostics make the projected evidential contrast independently plausible. That concession is the point. Once target, diagnostics, scope, and defeaters do the licensing, the bare UG postulate isn’t doing the work. A substantive UG theory may be one warrant; it isn’t the monopoly licence. Reiss’s polarity case needs the same caution. Polarity-item classifications aren’t exhausted by a single label or a simple binary contrast. Hoeksema’s survey of negative polarity items shows many distributional classes across questions, comparatives, conditionals, universals, *only*, superlatives, and related environments: the patterns aren’t random, but neither do they reduce to one strong/weak taxonomy (Hoeksema 2012). Reiss’s example needs the sort of item-sensitive distributional profile that a projectibility audit demands. The label “NPI” doesn’t do the transfer work. French creates a second problem. French *rien* is entangled with negative-quantifier and negative concord item behaviour, not simply with the behaviour of English NPI *any* as treated in classic polarity accounts (Ladusaw 1979; Chierchia 2013). If *rien* is made to stand for the polarity-dependent member of an English split, then a classification has already been imported: negative quantification, negative concord, polarity sensitivity, and free choice have been lined up in the way the argument requires. That’s not a bridge principle. It’s the very classificatory shortcut Reiss says his opponents aren’t entitled to make. The inference has been built by assuming the destination classification it’s supposed to license. Mere parsimony can’t be the ordering relation. The ordering relation has to explain why the observed distributions and the projected analysis are co-available. In this case, the candidate relation is compositional and distributional: polarity-dependent readings are tied to licensing environments, while free-choice readings carry modal or indifference force in environments where that licensing relation is absent. If English-internal evidence shows that those two dependency profiles behave differently across environments, a split analysis gains support. If it doesn’t, French remains a prompt rather than a licence. Evidence itself doesn’t stabilize the inference. Overt splits in other languages and English distributional contrasts are diagnostics. Candidate stabilizers would have to be defended separately: a recurrent pressure to keep polarity-dependent indefinites apart from free-choice expressions, a diachronic pathway that repeatedly separates the two, a processing or acquisition pressure, or a grammatical theory that predicts the split. Without such support, the French evidence has cue status. With English-internal diagnostics and a defended recurrence profile, it may become a scoped bridge. It isn’t a robust bridge on Reiss’s example alone. Defeaters matter just as much. The warrant weakens if English-internal diagnostics fail to separate the uses, if a unified semantics predicts both distributions without residue, if the French contrast turns out to reflect idiosyncratic lexicalization rather than the relevant profile, or if dialect and register variation in French undermine the claimed contrast. Reiss himself notes complications involving negative concord, French dialect variation, and the mismatch between standard written French and spoken varieties. A projectibility-first account treats those complications as defeaters or scope conditions, not as footnotes to a universal permission to transfer. That reconstruction gives a verdict different from the bare UG postulate. Reiss treats the French data as licensed for English by UG. Projectibility demotes the French data to a cue until the support profile and failure conditions are supplied. A substantive UG theory may supply some of that missing structure, but then it’s functioning as one first-order warrant under audit. The general postulate alone hasn’t licensed the transfer. Reiss’s worked case needs the very machinery his monopoly conclusion tries to reserve for UG. A horizontal non-UG bridge for this case would look like this. A crosslinguistic study of the forms used for indefinite reference maps the semantic contexts they cover, among them negation, questions, conditionals, and free choice, and finds that the ways languages group or split those contexts are constrained rather than arbitrary (Haspelmath 1997). Those contexts are semantic targets, comparative concepts rather than imported syntactic categories of pronouns, so the comparison stays within the comparison-and-classification discipline of Section 5. Here the warrant is recurrent, independently sampled constraint on how the contexts pattern together, not an innate lexical inventory. To make that sketch a stronger bridge, the recurrence claim would need an ordering relation of the iterated-learning sort: recurrent semantic and communicative tasks, learned through bottlenecked transmission, generating constrained patterns across unrelated languages. And if those constraints prove universal, that still wouldn’t make them UG: a universal constraint can be grounded in conceptual structure, communication, or processing rather than a genetically specified grammatical inventory. On such a profile, an overt French split between polarity-dependent and free-choice expressions makes an English split worth testing; it doesn’t settle the English classification. The map supplies a defeasible expectation, and English-internal diagnostics decide whether it holds. That’s a non-UG support profile of the kind the framework predicts, and it sits closer to Reiss’s own example than the genealogical case does, though it stays a sketch until the recurrence is worked through with declared defeaters. # Historical Comparison as a Non-UG Warrant That audit of French-English *any* is deliberately severe: it shows that Reiss’s own example doesn’t reach robust-bridge status on the evidence supplied. But non-UG warrants needn’t stop at cue level. Historical-comparative linguistics provides a familiar case in which evidence from one language bears on another because the licence is causal-historical rather than UG-theoretic. Here, the target is a causal-historical pattern in a family of languages: an inherited form, a regular sound change, a morphological alternation, a construction, or a subgroup relation. The profile is common descent under transmission and change. That shouldn’t be alien to Reiss’s ‘linguists’. They needn’t infer UG from phylogeny, but they rely on comparative and phylogenetic inference in biology whenever the Human Language Faculty is treated as an evolved, inherited human trait. That makes descent an ordinary inference-supporting relation, not a suspect one. If comparative inference can support claims about inherited biological traits, the same general evidential form can’t be ruled out when the inherited material is linguistic: sounds, forms, morphological patterns, and constructions. The substrate differs, and so do the diagnostics and defeaters; the evidential logic isn’t UG-specific. For historical comparison, diagnostics include recurring cognate sets, systematic sound correspondences, shared irregular morphology, chronology, geography, and independent evidence about borrowing and contact (Beekes 2011). The ordering relation is causal: daughter systems inherit, transform, lose, borrow, analogize, and reanalyse material from earlier systems. Evidence from Sanskrit, Greek, Latin, Gothic, or Old Irish can bear on a reconstruction not because UG says all languages are comparable, but because those languages are historically related in ways that constrain what counts as a good projection. A correspondence set such as Latin *pater*, Greek *patēr*, Sanskrit *pitar-*, and Gothic *fadar* doesn’t merely record resemblance. Together with many comparable sets, it supports a reconstruction of an inherited kinship term and a correspondence profile in which Germanic *f* aligns with non-Germanic *p* under independently motivated sound change, here the familiar Grimm’s Law correspondence (Beekes 2011). Evidence from Gothic can bear on the reconstruction, and the reconstruction can bear on the analysis of the Latin, Greek, and Sanskrit forms, because transfer is mediated by a causal-historical profile. This warrant has explicit scope limits. It licenses evidential travel within a family or contact history; it doesn’t license arbitrary transfer from any language to any other. Its defeaters are also familiar: chance resemblance, borrowing, areal diffusion, analogy, expressive vocabulary, semantic drift, defective attestation, and chronology that makes inheritance impossible. Those defeaters are part of the warrant. They make the inference scientific. A claim about inherited structure can be localized, demoted, or rejected when the correspondence profile fails. Background conditions of several kinds can matter to historical explanation. A UG theory might constrain possible sound systems, learnability, or grammar change; articulatory and perceptual facts may explain why particular sound changes are plausible. Those constraints help shape the space of possible changes, but they don’t license evidential transfer by themselves. The licence still comes from the reconstruction profile. A Reissian objection presses here: stating that Latin *p* corresponds to Gothic *f* presupposes that the daughters share commensurable segmental primitives, and a universalist takes those primitives to be UG-supplied. Reiss maintains exactly this universalist postulate in phonology, following Chomsky and Halle (Reiss 2026). But even if a universal feature vocabulary is granted, it isn’t itself the historical transfer licence. Reiss’s own phonology defends substance-free features (Reiss and Volenec 2022), and substance-free features can’t by themselves ground a phonetically defined correspondence set: they need an additional mapping to phonetic substance, and the inference still runs through the causal-historical correspondence profile. The feature vocabulary describes possible objects of projection. It doesn’t explain why Latin, Gothic, Sanskrit, and Greek evidence should bear on one another rather than on unrelated lookalikes. Historical comparison is bounded, but that bounded success is enough to defeat the written monopoly claim. The conclusion quantifies over *anyone* importing “analytic categories from one language to the study of another” (Reiss 2026, sec. 6). The comparative method does the work that conclusion rules out: it lets evidence from one language constrain the analysis of another, and of a shared ancestor, on a causal-genealogical licence rather than a UG one. So the unscoped claim is false as written. This forces a fork. Reiss can hold the unscoped conclusion, in which case the vertical case refutes it; or he can retreat to the horizontal core his examples actually target, Japanese evidence bearing on English where common descent is unavailable by construction. The retreat narrows the claim to a domain where this paper’s positive counterexample doesn’t reach, but it also concedes the point at issue: non-UG warrants license evidential travel, and I-language inquiry isn’t the only legitimate target. For horizontal transfer between unrelated languages, this paper establishes the negative claim Reiss denies, that UG holds no monopoly, but it doesn’t deliver a worked robust bridge. The indefinite-reference semantic map sketched for the *any* case shows the shape one would take, with crosslinguistic recurrence rather than a universal inventory as the support profile. I mark it as a sketch, not a delivered result: the worked horizontal case is a further project, and the paper’s “crosslinguistic evidence” should be read with that limit in view. # Public and Distributed Linguistic Objects So far, the paper has answered the evidential half of the monopoly claim. The ontological half needs care, because Reiss brackets public languages rather than banning them. He sets Pullum and Scholz’s model-theoretic objections aside as “irrelevant to my goals here”, since “there is no point in comparing theories whose objects are completely different kinds of things”: I-languages are “concrete in the sense of having spatiotemporal locations inside the heads of individuals, whereas it is impossible to say anything useful about where an artificial construct like Standard English exists. In 1969 was it briefly on the moon with Neil Armstrong?” (Reiss 2026, sec. 2). That’s a self-limitation of topic, not a denial that public objects are studiable. But it leans on a concreteness criterion and an individuation challenge worth meeting: if a public language isn’t located in one head, what makes it one object rather than a loose label for many practices? Projectibility individuates public, community-level, and normatively regimented objects by the projections they license under declared defeaters. That’s a genuine individuation criterion, not a relabelling of the norm, because the projections are independently testable and can fail: a wrong call about which features belong to the standard, or a missed prediction on held-out edited texts, refutes the individuation rather than confirming it. Maintenance, uptake, correction, transmission, and institutionalization stabilize that projectible profile and explain why it holds, but the individuation is done by the projections, not by promoting any mechanism to the constitutive essence of a kind. Stainton’s defence of public languages makes this point without denying mental grammars (Stainton 2011). The study of public languages can be viable if the object is understood as a humanly individuated system whose evidence base includes deployment, uptake, and psychological facts. Santana’s pluralist treatment of language also cuts against the idea that one privileged language-object settles all legitimate inquiry; the remaining question is which projections each object licenses, and when evidence about one target is being made to travel to another without warrant (Santana 2016; Reynolds 2026c). Reiss’s remarks about Standard English and public languages depend on a serious individuation worry. But the moon example proves too much. Armstrong didn’t bring NASA command authority, the flight plan, radio protocol, or mission time to the lunar surface as compact objects either. Still, his actions there were partly constituted by those distributed norms and institutions. Without them, planting a flag, reading a checklist, reporting to Houston, and taking a “small step” wouldn’t have been the acts they were. Transport is the wrong test for distributed objects: it turns a trip to the moon into an ontological filter that would erase much of what made the trip intelligible. This defeats the literal transport test, not the concreteness criterion behind it. That criterion is met differently: a distributed object earns its standing not by sitting in one location but by licensing testable projections that can fail, which is the work the rest of this section does. For public standards, the projection question is concrete. Model-theoretic syntax makes room for formally characterizing public constructs without treating them as mental grammars (Pullum and Scholz 2001); sociolinguistic work on standardization and enregisterment, the process by which forms become socially recognized as tied to registers or identities, supplies mechanisms for norm uptake and public recognizability (Milroy and Milroy 1999; Agha 2005). These literatures point to a public target whose projections concern correction, acceptance, institutional uptake, and style-shifting, not individual competence as such. This section answers the ontological half of Reiss’s restriction, and it also feeds the evidential half. The comparanda that would license horizontal transfer between unrelated languages, the comparative concepts and typological generalizations sketched in Section 5, are themselves public, abstracted, non-mental objects. A horizontal non-UG bridge needs exactly the kind of distributed object defended here, so the ontological half is a precondition of completing the evidential half, not a separate enterprise. For non-I-language targets the same evidential demand applies, but the support may be social-practical rather than genetic, and calling only I-language “language” doesn’t make the other targets disappear. Varieties give a compact test case. Register, dialect, and discourse community can be analysed as conditioning structures: participant expectation conditioned primarily on situation, ascription, or identification (Reynolds 2026b). The claim is projective. Knowing enough about the conditioning state should help predict likely forms, and hearing the forms should help infer the conditioning state. When those predictions fail, the variety claim can be localized or demoted rather than disappearing into metaphysical fog. Standard written English fits the same template. Its target is an edited public standard, not every speaker’s mental grammar or every English utterance, and its diagnostics are style guides, editorial intervention, school assessment, dictionaries, usage manuals, and corpora of edited prose. Correction, publication, examination, and professional editing stabilize the profile; the warrant runs through those maintained practices, not a hidden mental mechanism. The scope is edited, institutionally answerable prose, not casual speech, all written English, all Englishes, or any individual’s competence. So evidence from style guides, corrected student prose, and edited corpora can support projections about how *its*/*it’s*, *between you and I*, or *who*/*whom* are treated in academic prose, while licensing no such projection about spontaneous speech, an individual’s I-language, or every digital text. Defeaters include divergent institutional norms, stable acceptance of a once-corrected form, failure to predict held-out edited texts, or evidence that a supposed standard feature tracks genre, medium, or identity rather than the standard. The template licenses some projections and blocks others. It explains why evidence from a style guide may bear on an edited newspaper sentence but not on a child’s spontaneous speech; why corpus frequencies in unedited social media may be diagnostic for one target and misleading for another; and why public-language inquiry doesn’t need to pretend that public standards are located in one head. A distributed standard is disciplined when its target, support profile, scope, and defeaters are declared. Those are projective claims, not slogans. Uptake, correction, transmission, and institutional maintenance stabilize the projectible profile, and the projections, not the mechanisms, do the individuating. A public-language argument still has to declare the target, the support, the scope, and the defeaters for each projection. That demand is a strength of the alternative, because it imposes the same discipline on non-UG objects that Reiss rightly imposes on crosslinguistic inference. # Conclusion Crosslinguistic inference requires bridge principles; it does *not* require Chomskyan Universal Grammar. Reiss is right that categories are theory-mediated and that evidence doesn’t travel for free, and right that, under his stipulated Chomskyan target, I-language inquiry rests on a background commitment to shared grammatical primitives. But those points don’t show that every grounded comparison is either UG-based or ungrounded. Historical reconstruction also shows why the projectibility claim is realist rather than merely instrumental. Reconstructed proto-forms aren’t convenience labels. They’re posits constrained by converging correspondences, causal order, independent chronology, and defeaters. Successful projection is concrete: inherited correspondences are expected under causal-genealogical scope, while French evidence concerning English *any* remains cue-level without independent English-internal support. Public standards are less compact objects, but they answer to the same demand with equally concrete projections: evidence from style guides and edited corpora predicts how *who*/*whom* or *between you and I* are treated in edited prose, while licensing no such prediction about a child’s spontaneous speech. Each bridge cashes out in what it lets you expect and where the expectation stops. Projectibility-first work accepts the burden and raises it. Every proposed evidential transfer has to declare the target, the support profile, and the failure conditions. Chomskyan I-language inquiry can meet that demand for its own targets; so can non-UG inquiries, when their warrants specify what carries the inference and where it stops. Read locally, the claim is either stipulative or too strong. If *I-language* means UG-governed mental grammar, UG is required only because the target was named in UG terms; if the target is individual internal grammatical knowledge, UG is one explanatory framework among others. Either way the methodological lesson survives and generalizes: mental-grammar inquiry, like crosslinguistic inquiry, needs warrants, and those may be grammatical, cognitive, perceptual, usage-based, historical, social, or institutional. That’s the demand Reiss insists on, carried past UG. Defeasibility cuts both ways. The account fails if non-UG warrants keep collapsing into cue-level prompts, if proposed profiles fail on held-out cases, if ordering relations merely rename the data, or if social and typological targets yield no projections beyond their diagnostics. Those are the right risks. They make evidential travel inspectable: what carries the inference, where it stops, and what would break it. UG may answer that demand for some Chomskyan targets, but it doesn’t own it. Held to its written form, the conclusion that anyone who denies UG has no grounds for importing analytic categories across languages is refuted by the comparative method; narrowed to its horizontal core, it survives only by conceding the pluralism defended here. The demand for warrant outruns UG: UG is one way of meeting it, not the thing that makes meeting it possible. # Acknowledgements OpenAI Codex and Anthropic Claude served as drafting, review, and editing aids during the preparation of this paper. I am responsible for all theoretical claims, arguments, errors, and interpretive choices.
Agha, Asif. 2005. “Voice, Footing, Enregisterment.” *Journal of Linguistic Anthropology* 15 (1): 38–59. .
Barjamovic, Gojko, Thomas Chaney, Kerem Coşar, and Ali Hortaçsu. 2019. “Trade, Merchants, and the Lost Cities of the Bronze Age.” *The Quarterly Journal of Economics* 134 (3): 1455–503. .
Beekes, Robert S. P. 2011. *Comparative Indo-European Linguistics: An Introduction*. 2nd ed. John Benjamins.
Bybee, Joan. 2010. *Language, Usage and Cognition*. Cambridge University Press. .
Chierchia, Gennaro. 2013. *Logic in Grammar: Polarity, Free Choice, and Intervention*. Vol. 2. Oxford Studies in Semantics and Pragmatics. Oxford University Press. .
Croft, William. 2001. *Radical Construction Grammar: Syntactic Theory in Typological Perspective*. Oxford University Press. .
Culicover, Peter W., and Ray Jackendoff. 2005. *Simpler Syntax*. Oxford University Press. .
Dąbrowska, Ewa. 2010. “Naive v. Expert Intuitions: An Empirical Study of Acceptability Judgments.” *The Linguistic Review* 27 (1): 1–23. .
Dąbrowska, Ewa. 2012. “Different Speakers, Different Grammars: Individual Differences in Native Language Attainment.” *Linguistic Approaches to Bilingualism* 2 (3): 219–53. .
Goldberg, Adele E. 2006. *Constructions at Work: The Nature of Generalization in Language*. Oxford University Press. .
Goodman, Nelson. 1955. *Fact, Fiction, and Forecast*. Harvard University Press.
Haspelmath, Martin. 1997. *Indefinite Pronouns*. Oxford University Press. .
Haspelmath, Martin. 2010. “Comparative Concepts and Descriptive Categories in Crosslinguistic Studies.” *Language* 86 (3): 663–87. .
Hoeksema, Jack. 2012. “On the Natural History of Negative Polarity Items.” *Linguistic Analysis* 38 (1–2): 3–33.
Huddleston, Rodney, and Geoffrey K. Pullum. 2002. *The Cambridge Grammar of the English Language*. Cambridge University Press. .
Kempson, Ruth, Wilfried Meyer-Viol, and Dov Gabbay. 2001. *Dynamic Syntax: The Flow of Language Understanding*. Blackwell.
Kirby, Simon, Hannah Cornish, and Kenny Smith. 2008. “Cumulative Cultural Evolution in the Laboratory: An Experimental Approach to the Origins of Structure in Human Language.” *Proceedings of the National Academy of Sciences* 105 (31): 10681–86. .
Kirby, Simon, Tom Griffiths, and Kenny Smith. 2014. “Iterated Learning and the Evolution of Language.” *Current Opinion in Neurobiology* 28: 108–14. .
Kirby, Simon, Monica Tamariz, Hannah Cornish, and Kenny Smith. 2015. “Compression and Communication in the Cultural Evolution of Linguistic Structure.” *Cognition* 141: 87–102. .
Ladusaw, William A. 1979. “Polarity Sensitivity as Inherent Scope Relations.” PhD thesis, University of Texas at Austin.
Langacker, Ronald W. 1987. *Foundations of Cognitive Grammar: Theoretical Prerequisites*. Stanford University Press.
Milroy, James, and Lesley Milroy. 1999. *Authority in Language: Investigating Standard English*. 3rd ed. Routledge.
Ohala, John J. 1990. “The Phonetics and Phonology of Aspects of Assimilation.” In *Papers in Laboratory Phonology I: Between the Grammar and Physics of Speech*, edited by John Kingston and Mary E. Beckman. Cambridge University Press. .
Pullum, Geoffrey K., and Barbara C. Scholz. 2001. “On the Distinction Between Model-Theoretic and Generative-Enumerative Syntactic Frameworks.” In *Logical Aspects of Computational Linguistics: 4th International Conference, LACL 2001*, edited by Philippe de Groote, Glyn Morrill, and Christian Retoré, vol. 2099. Lecture Notes in Computer Science. Springer. .
Reiss, Charles. 2026. “Research Methods in Armchair Linguistics.” In *The Oxford Handbook of Philosophy of Linguistics*, edited by Gabriel Dupre, Ryan M. Nefdt, and Kate Stanton. Oxford University Press, forthcoming. .
Reiss, Charles, and Veno Volenec. 2022. “Conquer Primal Fear: Phonological Features Are Innate and Substance Free.” *Canadian Journal of Linguistics* 67: 581–610.
Reynolds, Brett. 2025. “Naturalizing Typological Kinds: Comparanda, Mechanisms, and Measurement.” .
Reynolds, Brett. 2026a. “Kinds as Projectibility Profiles: Support Grades and Demotion Rules.” .
Reynolds, Brett. 2026b. “Varieties as Conditioning Structure: A Homeostatic Account of Register, Dialect, and Discourse Community.” .
Reynolds, Brett. 2026c. “What Do We Mean by Language? A Pluralist Map for the Language Sciences.” .
Santana, Carlos. 2016. “What Is Language?” *Ergo: An Open Access Journal of Philosophy* 3 (19): 501–23. .
Sener, Nilufer. 2011. “Semantics and Pragmatics of Evidentials in Turkish.” PhD thesis, University of Connecticut. .
Stainton, Robert J. 2011. “In Defense of Public Languages.” *Linguistics and Philosophy* 34 (5): 479–88. .
[^1]: Contact: [^2]: I use “negative-polarity” here for the licensing environments, not as a commitment to “NPI” as a settled lexical kind. The *Cambridge Grammar of the English Language* expands “NPI” as “negatively-oriented polarity-sensitive item” and treats the relevant environments as including non-affirmative contexts rather than simply negative ones (Huddleston and Pullum 2002, 822–38).