Hidden Label¶

A string attached to a controlled-vocabulary concept that is indexed for retrieval but withheld from display — so a user typing a typo, deprecated synonym, or embarrassing prior name finds the concept while the result still shows only the sanctioned preferred term.

Core Idea¶

A hidden label is a string attached to a concept in a controlled vocabulary that is indexed for retrieval but deliberately withheld from public display. A user who types the typo, the deprecated synonym, the dialect variant, or the embarrassing prior name will find the concept, but the label that surfaces on the result is the sanctioned preferred term. The device separates two functions of a name that ordinary vocabularies conflate: recognition during search and authority during presentation.

In SKOS the mechanism is skos:hiddenLabel, the third tier below prefLabel (the displayed canonical name) and altLabel (additional displayed synonyms). Hidden labels carry the forms a system wants to accept as input — misspellings, retired terms, offensive historical names, purely phonetic guesses — without endorsing them as output. The asymmetry is structural: the retrieval index is wider than the display layer, absorbing the long tail of how users actually type while keeping the presentation surface clean and authoritative. Unlike an altLabel, a hidden label carries no display obligation and can be silently added or removed without altering what the system says back to the user. PubMed's MeSH thesaurus indexes "heart attack" as a hidden entry term for the preferred descriptor "Myocardial Infarction": the lay term is findable but not sanctioned, and the cited record shows only the preferred heading.

Structural Signature¶

Sig role-phrases:

the concept — the controlled-vocabulary entry that strings attach to
the displayed name set — the preferred label plus any sanctioned synonyms, the concept's authoritative public face
the hidden string inventory — typos, deprecated descriptors, offensive prior names, phonetic guesses attached to the concept but withheld from display
the retrieval index — the lookup surface that consumes both the displayed and hidden sets, so any string in either lands the user on the concept
the display layer — the rendered output that consumes only the displayed set, never surfacing a hidden string
the width asymmetry — the index is deliberately wider than the display, the gap between them being exactly the absorbed long tail of how users type
the silent-update property — hidden strings can be added or removed without changing anything the system says back, since they carry no display obligation
the per-variant routing call — each incoming form is sorted to one of three tiers: accept-as-input-only (hidden), accept-and-display (alt), or exclude

What It Is Not¶

Not a bug or a stale index. "The search finds it but the page won't call it that" is the device working, not a broken redirect to be reconciled. A hidden string deliberately grants recognition during search while withholding authority during presentation; the mismatch is the intended consequence of a label sitting on the hidden tier.
Not an endorsement of the variant. Indexing "heart attack" or a retired or offensive prior name as a hidden entry term makes it findable, not sanctioned. The device accepts the form as input precisely so it need not appear as output; the surfaced label stays the preferred term, so absorbing an embarrassing name for retrieval does not put it on the concept's public face.
Not an altLabel. A displayed synonym carries a display obligation — adding it changes what the vocabulary says back to the user — whereas a hidden label carries none and can be added or dropped silently. Treating the two as interchangeable collapses the input surface into the output surface, which is exactly the conflation the third tier exists to undo.
Not a redirect that rewrites the display. Routing a user to the concept is all the hidden string does; it never overrides the rendered name. The display layer consumes only the displayed set, so a hidden label moves where the user lands, not what the result is called.
Not a genuinely cross-substrate pattern. Product search, wiki redirects, DNS aliases, and MeSH entry terms are not distinct substrates but one — a retrieval index over a catalog — in different wrappers. What recurs across truly different substrates is the more general markedness + aliasing insight (recognition-canonical and display-canonical coming apart), not the SKOS convention itself.

Scope of Application¶

The hidden label lives across the retrieval and controlled-vocabulary subfields of information science; it applies wherever one substrate recurs — a retrieval index run over a catalog, with a separable index layer and display layer — and the genuinely cross-substrate version of the insight (recognition-canonical vs. display-canonical coming apart) belongs to its parents markedness and aliasing, not to this SKOS convention.

Thesauri and subject vocabularies — the canonical use: MeSH entry terms, retired ERIC descriptors, and SKOS hiddenLabel strings indexed for retrieval but suppressed from the displayed heading, so "heart attack" finds "Myocardial Infarction" without endorsing the lay term.
Product search — misspellings and competitor names mapped onto catalog items so a mistyped query lands on the product, while neither string appears on the product page.
Wiki redirects — old or easily-confused article titles route to the current article while being kept off its visible name (the {{distinguish}} case).
DNS and email aliasing — addresses that route a message to the right destination without appearing in the rendered signature or display name.

Clarity¶

Naming the hidden label makes legible a confusion that looks, from the outside, like a defect: the search bar resolves a term the result page refuses to call itself by. Without the concept, that mismatch reads as a stale index or a broken redirect — something to be reconciled. With it, the same behavior is recognized as a deliberate split between two jobs a vocabulary asks of a string, recognition during search and authority during presentation, which a single-tier name list silently fuses. Once a vocabulary maintainer holds those apart, the sharper question becomes available: for any given variant — the typo, the retired descriptor, the offensive prior name, the phonetic guess — should this form be accepted as input, displayed as a sanctioned synonym, or neither? That is a three-way decision (hiddenLabel vs. altLabel vs. excluded) that the flat preferred-term-plus-synonyms model cannot even pose.

The distinction also clarifies what an entry term commits the system to. An altLabel carries a display obligation — adding it changes what the vocabulary says back to the user — whereas a hidden label carries none and can be added or removed silently as misspellings and dialect variants accumulate. Seeing this lets the curator treat the retrieval index as legitimately wider than the display layer rather than as an inconsistency to be flattened, and turns the long tail of how users actually type into a maintainable input surface kept cleanly separate from the authoritative output surface — so that absorbing a retired or embarrassing name for findability no longer risks endorsing it as the concept's public face.

Manages Complexity¶

A controlled vocabulary's recurring headache is the unbounded variety of strings real users send at a concept: every misspelling of "receive," every regional name for a disease, every retired descriptor a decade-old citation still uses, every offensive prior name no curator wants on the result page. Trying to manage that variety on a single flat list of names forces an impossible trade — admit each variant as a displayed synonym and the public face of the concept bloats with typos and slurs, or refuse it and the concept becomes unfindable to anyone who types it the wrong way. The hidden label tames this by splitting the one quantity the maintainer was overloading — "the names of this concept" — into two separately sized surfaces governed by a single per-variant decision. The maintainer no longer reasons about each string against the whole tangle of search-and-display obligations at once; for any incoming variant the question collapses to a three-way routing call read straight off the variant's character: is this form one to accept as input only (hiddenLabel), to accept and also display as sanctioned (altLabel), or to exclude entirely — and the typo, the dialect form, the embarrassing legacy name each land in the first bin without any further deliberation about how they interact with the presentation layer.

What the maintainer reads off, having made that call, is that the retrieval index is legitimately wider than the display layer, and the gap between the two widths is exactly the absorbed long tail. This is the compression: the display surface stays small, clean, and authoritative no matter how many variants accumulate, because every variant added on the hidden tier changes the index width without touching the display width at all — the asymmetric-update property means the maintainer can pour the entire messy input distribution into one surface while the other holds constant. Two qualitative outcomes then follow without case-by-case analysis. A reported "the search finds it but the page won't call it that" mismatch is read not as a stale-index defect to reconcile but as the intended consequence of a string sitting on the hidden tier — recognition without authority, by design. And the commitment of any name becomes a one-bit fact: a hidden label binds the system to nothing it says back and can be silently dropped, whereas an altLabel binds the public output and cannot. The sprawling "how do we handle every way users type this concept while keeping its public name clean?" reduces to: sort each variant into one of three tiers, let the hidden tier grow freely, and read the two surface widths as deliberately unequal.

Abstract Reasoning¶

The device's foundational move is a per-variant routing inference read off the character of an incoming string. For any form a user might send at a concept — a typo, a deprecated descriptor, an offensive prior name, a phonetic guess, a sanctioned synonym — the curator reasons to one of three placements: accept as input only (hiddenLabel), accept and also display as sanctioned (altLabel), or exclude entirely. The inference runs from what the variant is to what the system should do with it: a misspelling or an embarrassing legacy name is findable-but-not-endorsable, so it routes to the hidden tier; a legitimate alternate name is endorsable, so it routes to display; a form the vocabulary refuses to accept at all is excluded. This is a three-way decision the flat preferred-term-plus-synonyms model cannot even pose, and it is licensed by separating the two jobs a name does — recognition during search and authority during presentation — that a single-tier list silently fuses.

A diagnostic move reclassifies an apparent defect. Facing the behaviour "the search bar finds it but the result page won't call it that," the analyst infers not a stale index or a broken redirect to be reconciled but a string deliberately sitting on the hidden tier — recognition without authority, by design. The reasoning runs from the observed search/display mismatch to its intended cause, and it inverts the naive reading: the mismatch is the device working, not failing. This pairs with a commitment inference that reduces a name's obligations to one bit: a hidden label binds the system to nothing it says back and can therefore be added or removed silently, whereas an altLabel binds the public output and cannot be changed without altering what the vocabulary tells the user. From a label's tier the analyst thus predicts whether editing it is a silent maintenance act or a change to the concept's public face.

A predictive / boundary move follows from the asymmetry between the two surfaces. Because the retrieval index consumes both the displayed and the hidden sets while the display layer consumes only the displayed set, the index is legitimately wider than the display, and the gap between the two widths is exactly the absorbed long tail of how users type. The analyst predicts that pouring more variants onto the hidden tier widens findability while holding the display surface constant — clean and authoritative no matter how many misspellings and dialect forms accumulate — so the move is to read the two surface widths as deliberately unequal rather than as an inconsistency to flatten. The boundary on all this is the substrate: the inferences presuppose a retrieval system over a controlled vocabulary or catalog with a separable index and display layer, so the routing, mismatch-diagnosis, commitment, and width-asymmetry reasoning apply to thesauri, product search, wiki redirects, and routing aliases alike, and do not extend to settings lacking the index-versus-display split the device exploits.

Knowledge Transfer¶

Within information science the hidden label transfers as mechanism across every system that runs a retrieval index over a controlled vocabulary or catalog. The three-tier prefLabel / altLabel / hiddenLabel structure, the per-variant routing call, the index-wider-than-display asymmetry, and the silent-update property all carry intact from one such system to the next. In thesauri and subject vocabularies it is the canonical use — MeSH entry terms, retired ERIC descriptors, SKOS hiddenLabel strings indexed but suppressed from the displayed heading. In product search the same device maps misspellings and competitor names onto catalog items without showing them on the product page; in wiki redirects old or distinguishable titles route to the current article while being kept off its visible name; in DNS and email aliasing addresses route without appearing in signatures. A maintainer who learns the device in one of these reads "the search bar finds it but the page won't call it that" the same way in all of them — as the hidden tier working by design, not a stale-index defect — and applies the same intervention (route the variant to the hidden tier rather than the display tier). The vocabulary and diagnostics travel without translation because these are not distinct substrates but one substrate in different wrappers: an indexed string that does not appear in the rendered output of the indexed object.

That observation is exactly why the case beyond the home substrate is (B), shared abstract mechanism rather than the named device travelling. The hidden label's apparent cross-domain spread (library catalog, e-commerce, wiki, DNS) is not genuine substrate-bridging — every example is the same retrieval-over-catalog structure — so what would carry to a truly different substrate (a cognitive, social, or formal system) is not "hidden label" but the more general patterns it instantiates: markedness (a canonical form and non-canonical variants, asymmetrically privileged) and aliasing (multiple names for one referent), here specialized to the case where some aliases are deliberately non-displaying. The portable lesson is that recognition-canonical and display-canonical can come apart — the form a system accepts as input need not be the form it endorses as output — and that lesson is available in substrate-neutral terms from those parent primes. The home-bound cargo is the SKOS-specific apparatus: the named third tier, the skos:hiddenLabel semantics, the entry-term curation workflow, the treatment of the index as a maintainable input surface distinct from the display surface. Invoking "a hidden label" outside a retrieval index renames the parts and borrows the input-versus-output split while dropping the index column that gives the device its meaning — illuminating by resemblance, but the mechanism that actually recurs across domains is markedness + aliasing, not the library-science convention (see Structural Core vs. Domain Accent).

Examples¶

Canonical¶

The defining construction is a SKOS concept carrying strings on three tiers. Take the MeSH concept for heart attack, expressed in SKOS:

skos:prefLabel → "Myocardial Infarction" (the one displayed heading)
skos:altLabel → "Cardiac Infarction" (a sanctioned synonym, also displayable)
skos:hiddenLabel → "heart attack", "myocardial infraction" (the lay term and a common typo)

The retrieval index is built from all three tiers, so a user who types the lay term "heart attack" or even the misspelling "infraction" lands on the concept. But the display layer draws only from the preferred (and alternate) labels, so the record that surfaces is titled "Myocardial Infarction." The lay term did its job — recognition — without ever appearing as the concept's public name. A curator could add a hundred more misspellings on the hidden tier and the displayed heading would not change at all.

Mapped back: "Myocardial Infarction" plus "Cardiac Infarction" are the displayed name set; "heart attack" and the typo are the hidden string inventory. The lookup drawing on all tiers is the retrieval index, the titled record is the display layer, and that the first is broader than the second is the width asymmetry.

Applied / In Practice¶

E-commerce and enterprise search engines deploy exactly this device through their synonym configuration. A retailer running a platform such as Elasticsearch or Solr maintains synonym lists that map the messy reality of how shoppers type — misspellings ("headphnes"), regional variants ("jumper" for a sweater), abbreviations, and even competitors' product names — onto the correct catalog items, so that a mistyped or off-brand query still returns the right product. Critically, these mapped strings are indexed for matching only; they never render on the product page, which continues to show the clean, brand-sanctioned product name. Merchandisers add and prune these entries continuously as new query variants appear in the logs, without any of it touching what customers see on the listing.

Mapped back: The misspellings, regional variants, and competitor names in the synonym list are the hidden string inventory feeding the retrieval index while kept out of the display layer. Sorting each incoming query variant into "match-only" rather than "show on page" is the per-variant routing call, and adding or pruning them without altering the listing is the silent-update property.

Structural Tensions¶

T1: Findability without endorsement versus acceptance as complicity (matching is behavior, not display). The device's central promise is that a system can accept an offensive prior name, a slur, or a deprecated term as input without endorsing it as output — recognition during search, authority withheld at presentation. But withholding a string from display does not make the system neutral toward it: a user who types the disavowed term is still matched, ranked, and delivered to the concept, so the system is quietly responsive to language it publicly refuses to show. The clean output surface conceals a behavioral fact — the index acts on the hidden string — and that responsiveness can be surfaced (a probe reveals the search accepts the slur), embarrassing a maintainer who believed hiding the label discharged the concern. Non-display is not the same as non-participation. Diagnostic: Is the worry here only that the bad term not appear (hidden label suffices), or that the system not act on it at all (which the hidden tier does not address)?

T2: Silent maintainability versus an unaudited input surface (invisible is ungovernable). The hidden tier's great convenience is that its strings carry no display obligation and can be added or pruned silently as query logs evolve — the input surface grows freely without touching what users see. But the same invisibility removes the scrutiny display would have imposed: because hidden labels never surface, no reviewer routinely inspects what the system quietly matches on, so erroneous mappings, biased associations, or even adversarial aliases (a competitor's name captured onto your product, a misleading term routed to a concept) can accumulate unseen. The property that makes the tier cheap to maintain is exactly the property that makes it hard to audit, and an input surface no one looks at is an input surface no one governs. Diagnostic: Is there any review process over what sits on the hidden tier, or does its silence mean variants are matched on strings no human has vetted?

T3: Index-wider-than-display as pure gain versus the precision cost of a wide index (false matches). The concept frames the width asymmetry as free: pour the whole long tail onto the hidden tier and findability widens while the display stays clean "no matter how many variants accumulate." But a wider retrieval index is not costless on the matching side. Typos are sometimes valid different words, phonetic guesses collide across concepts, and mapping a competitor's name onto your item hijacks a query whose intent may have been genuinely elsewhere — so absorbing more variants raises false matches, ambiguity, and mis-routing that the display-constancy framing renders invisible. The device holds the output surface constant but quietly trades away retrieval precision, and a hidden tier stuffed to maximize recall can degrade exactly the relevance it was meant to protect. Diagnostic: Does adding this variant to the hidden tier only widen recall, or does it also introduce a collision or intent-hijack that lowers precision for other queries?

T4: A clean authoritative face versus concealed provenance (hiding the alias hides information too). Suppressing the retired descriptor, the lay term, or the embarrassing prior name from display keeps the concept's public face sanctioned and uncluttered — the device's headline benefit. But the alias is sometimes information the user legitimately wants: that a term was renamed, that their word maps to a different official heading, that a concept once carried another name. By granting recognition while hiding the matched string entirely, the device can leave a user unsure why their query resolved to an unfamiliar preferred term, and can erase the naming history a researcher or auditor needs. The authoritative output is bought by concealing the messy real-world naming, and clean is not always the same as transparent. Diagnostic: Does the user benefit from never seeing the matched variant, or do they need to know which alias resolved and why — provenance the hidden tier deliberately withholds?

T5: Autonomy versus reduction (a SKOS convention or markedness plus aliasing). The hidden label is a specific, well-specified information-science device — the skos:hiddenLabel third tier, the entry-term curation workflow, the index-as-maintainable-input-surface treatment — and within retrieval-over-a-catalog it transfers as mechanism intact. But the entry concedes its own limited autonomy: its apparent cross-domain spread (product search, wiki redirects, DNS aliases, MeSH) is one substrate in different wrappers, and what recurs across genuinely different substrates is not "hidden label" but markedness (a privileged canonical form against non-canonical variants) and aliasing (many names, one referent), here specialized to non-displaying aliases. The portable lesson — recognition-canonical and display-canonical can come apart — is available in substrate-neutral terms from those parents. Diagnostic: Resolve toward markedness + aliasing when carrying the input-versus-output-canonical insight to a cognitive, social, or formal system; toward the named hidden label when curating an actual controlled vocabulary's tiers in situ.

Structural–Framed Character¶

Hidden label sits in the mixed band of the structural–framed spectrum, leaning framed — an evaluatively neutral curation device that is nonetheless a standards-born convention of information-science practice. On evaluative_weight it scores structural: a string indexed but withheld from display is neither good nor bad, and the device renders no verdict — even its handling of slurs and embarrassing prior names is a neutral input/output routing, not a moral stance the concept itself takes. But the remaining criteria pull framed. Human_practice_bound is high: the device presupposes a retrieval system over a controlled vocabulary with a separable index and display layer, curated by a maintainer — remove that curation practice and there is no hidden tier, only strings. Institutional_origin is pronounced: the skos:hiddenLabel third tier, the prefLabel/altLabel/hiddenLabel structure, MeSH entry terms, and the entry-term curation workflow are artefacts of information-science standards. Vocab_travels fails at the named level, since that SKOS apparatus has no referent off the retrieval substrate, though the abstract split (index wider than display, per-variant routing) does lift. On import_vs_recognize the entry's own concession is the tell: its apparent cross-domain spread (product search, wiki redirects, DNS) is "one substrate in different wrappers," recognized as the same mechanism, while genuinely cross-substrate transfer rides the parents, not the named device.

The portable structure is a conjunction of parent primes — markedness (a privileged canonical form against asymmetrically-subordinated variants) and aliasing (many names, one referent), here specialized to the case where some aliases are deliberately non-displaying. Their joint lesson — that recognition-canonical and display-canonical can come apart, the form a system accepts as input need not be the form it endorses as output — is genuinely substrate-neutral and is exactly what the hidden label instantiates, keyed to a controlled vocabulary; the cross-domain reach belongs to markedness + aliasing, while the SKOS tier-and-workflow apparatus stays home. Its character: an evaluatively-neutral but practice-constituted and standards-born retrieval convention, structural only as the markedness-plus-aliasing insight it specializes to non-displaying aliases, and otherwise pinned to SKOS vocabulary and the index-versus-display curation it presupposes.

Structural Core vs. Domain Accent¶

This section decides why the hidden label is a domain-specific abstraction and not a prime — a case with a revealing twist, because even its apparent cross-domain spread turns out to be one substrate in different wrappers, and the surviving skeleton is a conjunction of two parents.

What is skeletal (could lift toward a cross-domain prime). Strip the retrieval system and a thin relational structure survives: one referent has a privileged canonical form and a set of asymmetrically-subordinated variant forms, and the form a system accepts for recognition need not be the form it endorses for presentation — recognition-canonical and display-canonical can come apart. The abstract pieces are a canonical form set against non-canonical variants (markedness), multiple names bound to one referent (aliasing), and the further split in which some of those aliases are deliberately non-displaying. That skeleton is genuinely substrate-portable — it is the conjunction of markedness and aliasing the hidden label instantiates, and their joint lesson recurs wherever an accepted input form and an endorsed output form diverge across cognitive, social, or formal systems. But it is the core the hidden label shares with those parents, not what makes it the distinctive thing it is.

What is domain-bound. What makes it the hidden label in particular is controlled-vocabulary furniture that does not survive extraction. The tiers are the SKOS prefLabel / altLabel / hiddenLabel structure; the mechanism is skos:hiddenLabel semantics; the variants are typos, deprecated descriptors, offensive prior names, and phonetic guesses attached as entry terms; the two surfaces are a retrieval index and a display layer; the workflow is entry-term curation against query logs; the exemplar is MeSH indexing "heart attack" for "Myocardial Infarction." The decisive test: remove the retrieval system over a controlled vocabulary — the separable index and display layer and the maintainer curating them — and there is no hidden label, only strings, with nothing to be indexed-but-not-displayed. Notably, this substrate is more replayed than crossed: product search, wiki redirects, and DNS aliasing look like different domains but are all "a retrieval index over a catalog" in different wrappers, so the term's in-domain breadth is one substrate, not cross-substrate reach.

Why this does not clear the prime bar. A prime's vocabulary travels and its cross-domain transfer is recognition of the same mechanism, not analogy. The hidden label's transfer is bimodal, with the wrinkle just noted. Within information science it travels intact as full mechanism — the three-tier structure, the per-variant routing call, the index-wider-than-display asymmetry, and the silent-update property apply identically to thesauri, product search, wiki redirects, and routing aliases, but those are the same retrieval-over-catalog substrate in different wrappers, so this is mechanism-recognition within one home, not travel. Beyond that substrate the named device does not go: a genuinely different substrate (a cognitive, social, or formal system) is reached by recognising the parents' mechanism, and calling such a case "a hidden label" renames the parts and borrows the input-versus-output split while dropping the index column that gives the device its meaning. So when the bare structural lesson genuinely is needed cross-domain — that recognition-canonical and display-canonical can come apart — it is already carried, in more general form, by markedness + aliasing. The cross-domain reach belongs to those parents; "the hidden label," as named, is the controlled-vocabulary instance whose SKOS tier-and-workflow apparatus should stay home.

Relationships to Other Abstractions¶

Current abstraction Hidden Label Domain-specific

Parents (2) — more general patterns this builds on

Hidden Label presupposes Preferred label Domain-specific

Hidden status requires a separate sanctioned Preferred Label that the result displays after the hidden string has performed only its recognition role.
Hidden Label is a decomposition of Alias-to-Authority Mapping Prime

Hidden Label is Alias-to-Authority Mapping specialized so a variant is accepted by the resolution layer but withheld from every public name surface.

Hierarchy paths (9) — routes to 4 parentless roots

Hidden Label → Preferred label → Arbitrariness of Symbolic Conventions → Signifier–Signified Duality → Representation → Abstraction

Not to Be Confused With¶

altLabel (displayed synonym). The adjacent SKOS tier: a sanctioned alternate name that is both indexed and displayed. The whole point of the hidden tier is that it carries no display obligation — a hidden label can be added or dropped silently, while adding an altLabel changes what the vocabulary says back to the user. Collapsing the two fuses the input surface into the output surface, exactly the conflation the third tier undoes. Tell: does the string appear as a sanctioned synonym on the result (altLabel), or match queries while never rendering (hiddenLabel)?
A redirect that rewrites the display name. A routing device that both sends the user to the target and changes the name shown. A hidden label moves only where the user lands, never what the result is called — the display layer keeps drawing on the preferred term alone. Tell: after resolving, does the shown name change to reflect the matched form (renaming redirect), or stay the sanctioned preferred label while the matched variant vanishes (hidden label)?
A stale index or broken redirect (a bug). The failure mode a hidden label is mistaken for: "the search finds it but the page won't call it that" read as an inconsistency to reconcile. That mismatch is the device working — recognition granted, authority withheld — not a defect. Tell: is the search/display divergence an accidental artifact of an out-of-date index (bug), or the intended consequence of a string deliberately placed on the non-displaying tier (hidden label)?
Synonym ring / symmetric equivalence set. A set of terms treated as fully interchangeable, any of which may surface as the label. The hidden label is asymmetric: the variant is accepted for input only and is deliberately subordinated, never eligible for display, with one privileged preferred form. Tell: are all the names co-equal and any displayable (synonym ring), or is there a canonical form plus non-displaying input-only variants (hidden label)?
Query expansion / synonym query-rewriting. An IR technique that transforms the query at search time — adding related or corrected terms before matching. A hidden label instead attaches non-displaying strings to the concept in the vocabulary. One rewrites the incoming query; the other widens the indexed target. Tell: is the mechanism expanding the user's query terms at run time (query expansion), or a curated string bound to the concept and indexed but suppressed from display (hidden label)?
Markedness plus aliasing (the parents). The substrate-neutral conjunction the hidden label instantiates — a privileged canonical form against asymmetrically-subordinated variants (markedness) and many names for one referent (aliasing), here specialized to non-displaying aliases. Their joint lesson — recognition-canonical and display-canonical can come apart — is what carries to genuinely different (cognitive, social, formal) substrates. Tell: off the retrieval-index substrate the recurring content is markedness + aliasing; "hidden label" applies only where a controlled vocabulary has a separable index and display layer with SKOS-style tiers. (Treated fully in an earlier section.)

Neighborhood in Abstraction Space¶

Hidden Label sits in a crowded region of the domain-specific corpus (13^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Surface Form & Underlying Structure (23 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12