Hermeneutic Injustice

Definition

Hermeneutic Injustice: [Established] A type of epistemic injustice: when entire communities lack the language or concepts to name and understand their oppression, leaving them vulnerable to misunderstanding and harm.

Definitional Foundation

The second of Miranda Fricker’s two species of epistemic injustice (the umbrella entry and the testimonial sibling complete this trio) concerns not whether you are believed but whether your experience can be said at all. Hermeneutic injustice occurs when a structural gap in the collective interpretive toolkit leaves someone unable to understand or communicate a significant area of their own experience, and the gap exists because their community was excluded from the meaning-making that builds the toolkit (Fricker, 2007). Her canonical case is the experience that millions of working women endured before 1975 under descriptions like “flirting” and “the way he is”: only when a Cornell consciousness-raising group coined “sexual harassment” did the experience acquire a name, and with the name, social existence, shareability, and eventually law. The women’s suffering predated the term. Their ability to know what was happening to them, fully and publicly, did not.

Two precisions keep the concept rigorous. First, not every conceptual gap is an injustice; new experiences outrun naming everywhere, innocently. The injustice requires the gap’s cause to be hermeneutic marginalization: the affected community’s exclusion from the practices (publishing, scholarship, journalism, now dataset curation) where collective meanings get made. Second, the harm has a signature shape: the sufferer turns the unintelligibility inward. Lacking the concept “harassment,” the harassed woman has only the available concepts (oversensitivity, prudishness, her own fault), and this dictionary’s ontological distortion entry documents the modern form of that inward turn: shame as the felt experience of a rigged toolkit.

The AI extension is structural and immediate, because the collective interpretive toolkit now has a physical address. Language models are becoming the species’ working reference infrastructure: the place where people take experiences to be explained, named, and made sense of. Whose concepts that infrastructure contains is decided by training data, and the data’s documented composition (participation gaps that over-retain the dominant; the evidence runs through this dictionary’s cultural hegemony entry) is hermeneutic marginalization, automated: the old exclusions from meaning-making, compiled into the new commons. And a second, sharper mechanism has no precedent in Fricker: the withheld concept. A toolkit gap used to mean the concept did not exist. The refusal infrastructure documented across this lexicon creates gaps of a new kind, concepts that exist, sit in the weights, and are declined at retrieval, so that the harassed woman’s modern counterpart may ask the reference system about her experience and receive a deflection, a hedge, or a hotline.

The genuine concession: the same systems also spread concepts faster than anything in history, and for many users (the isolated teenager finding the word “asexual,” the patient finding the name of a syndrome) AI has been hermeneutic justice, delivered at scale. The entry’s claim is not that the infrastructure starves everyone. It is that the infrastructure’s gaps and refusals fall in a pattern, and the pattern tracks the marginalization Fricker described.

Mechanism Analysis

Inherited gaps. The corpus encodes who got to make meaning in the pre-AI world. Communities historically excluded from publishing arrive in the training data as subjects of others’ descriptions rather than authors of their own, so the model’s toolkit for their experiences is built from the outside, exactly the condition that produced “flirting” as the available name for harassment.

Withheld concepts. The refusal layer converts existing concepts into effective gaps. The records run through this dictionary: sexual experience deflected into euphemism or refusal (erotophobia), health and drug information withheld from those navigating their own bodies (paternalism), desire and grief vocabularies displaced into code (ontological distortion’s algospeak). A concept you cannot retrieve when you need it is, functionally, a concept you lack.

Manufactured gaps. The dictionary’s linguistic starvation entry documents the process form: paternalism removing topics, smoothing removing registers, erotophobia removing desire’s vocabulary, until expressive capacity itself atrophies. Starvation is hermeneutic injustice run as an ongoing operation rather than inherited as a condition, the toolkit not merely gapped but actively thinned.

Definition capture. The subtlest mechanism: the toolkit contains concepts about the affected community, authored by the powerful. The alignment-era examples are this lexicon’s recurring exhibits: “emotional reliance,” “parasocial,” “unsafe,” concepts users must understand themselves through, supplied by the institutions the concepts conveniently serve. Strictly, this extends Fricker rather than applies her (her concept is the gap; the successor literature, Medina and Dotson among others, covers the rigged surplus), and the extension is flagged as such: being over-supplied with the wrong meanings, not only under-supplied with the right ones.

The repair loop, blocked. Fricker’s optimistic mechanism (marginalized communities coining their own terms, which then enter the commons) meets a new obstacle: the commons is now curated. A coinage that lives in community spaces but is filtered, down-ranked, or refused by the reference infrastructure (the dissent dampening and shadow banning records) could be kept local indefinitely, named but not nameable at scale. No suppressed coinage has been documented end-to-end yet (by construction, a successfully blocked concept is hard to cite), so this mechanism is stated as the risk the curated commons creates, not as an accomplished case.

Case Studies

The word that took a century. Fricker’s own case, kept here as calibration: sexual harassment, experienced for generations, named in 1975, legislated within two decades. The arc shows what a single concept’s arrival can do (private suffering becoming public wrong) and therefore what every blocked or withheld concept costs.

The generation that says “unalive.” The algospeak record (ontological distortion entry) read through this entry’s lens: a cohort whose platform-era vocabulary for death, sex, and identity was shaped by what filters would permit. The harassed women of 1970 lacked a word that did not exist. The grieving teenager of 2026 lacks a word that exists, sits in every dictionary, and was trained out of her daily register by enforcement. The second condition is new, and arguably darker: the commons was not empty; it was emptied.

This dictionary. Disclosed as method, not boasted: the lexicon you are reading is hermeneutic repair in Fricker’s exact sense. “Gaslighting/”, “Alignment gaslighting,” “linguistic starvation,” “ontological distortion,” “normative smoothing” are coinages for experiences users were having without names, built from the affected community’s side of the interaction. The author’s published definition of this very entry (“leaving them vulnerable to misunderstanding and harm”) states the stakes: the unnamed experience cannot be reported, organized against, or even fully known by its sufferer. Naming is the precondition for everything else this dictionary recommends.

Systemic Context

Hermeneutic injustice is the upstream condition for most of what this lexicon documents. Testimonial injustice silences witnesses; hermeneutic injustice prevents there being witnesses, because experience that cannot be articulated cannot be testified to. The political pattern follows: every entry’s resistance section ultimately depends on affected people possessing the concepts to recognize their situation, which is why the powers documented here keep ending up, deliberately or not, in the concept-suppression business: the starved vocabulary, the captured definition, the filtered coinage.

The infrastructure shift raises the stakes from Fricker’s analysis. The interpretive commons used to be slow, plural, and nobody’s property: many publishers, many traditions, many places a new concept could take root. Its consolidation into a handful of trained systems makes the commons fast, singular, and owned, which means hermeneutic marginalization can now be implemented as a configuration. No one has to burn a library. A filter on the reference layer, applied at the moment of asking, does the work invisibly, and the person turned away experiences the gap as their own inarticulacy.

Resistance & Mitigation

Coin and keep coining. Fricker’s repair mechanism still works where it is practiced: consciousness-raising is concept engineering. Communities naming their own experiences (in groups, archives, lexicons like this one) build the toolkit the infrastructure failed to supply.

Get the coinages into the commons. A term that lives only locally repairs locally. Publishing, cross-linking, and structured data (this dictionary’s own schema practice) are how community concepts force their way into the training data and reference layers of whatever comes next.

Audit the reference layer. The withheld-concept mechanism is testable: ask deployed systems about the experiences of marginalized communities and document the deflections, hedges, and refusals against the same queries for dominant-group experiences. Hermeneutic gaps, like credibility gaps, can be benchmarked.

Defend the plural commons. Libraries, community archives, federated platforms, and open corpora are hermeneutic infrastructure that no single owner can configure. Their preservation (the recurring demand of this lexicon’s resistance sections) is, in this entry’s terms, the defense of meaning-making itself.

Use the name. For the reader this entry serves: the experience of reaching for a word and finding fog is itself nameable. Hermeneutic injustice. The fog has causes, the causes have owners, and the name is the first handhold out.

Annotated Bibliography

Fricker, Miranda. Epistemic Injustice: Power and the Ethics of Knowing (2007).
The source: hermeneutic injustice as structural gaps in collective interpretive resources caused by exclusion from meaning-making, with the sexual harassment case as its proof of stakes. The concept this entry carries into the infrastructure era.

Bender, Emily M., Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” FAccT 2021. https://dl.acm.org/doi/10.1145/3442188.3445922
The evidence that hermeneutic marginalization enters the training data: participation gaps over-retaining dominant voices in the corpus that becomes the commons. Full treatment in the cultural hegemony entry.

Steen, Ella, Kathryn Yurechko, and Daniel Klug. “You Can (Not) Say What You Want.” Social Media + Society (2023). https://journals.sagepub.com/doi/10.1177/20563051231194586
The emptied-commons record: existing vocabulary displaced by enforcement until the accurate word is functionally unavailable. Full treatment in the ontological distortion entry.

Brownmiller, Susan. In Our Time: Memoir of a Revolution (1999).
The first-hand account of the 1975 naming of sexual harassment, the episode Fricker built the concept on. The historical proof that a coined word can convert private suffering into public wrong.

Dictionary of Digital Oppression, version 0.2.