This site uses essential browser storage for authentication and preferences. No tracking cookies are used. Privacy Policy
Annotation protocol
Guidelines
The public rulebook for annotating BUDOVA. Every rule is citable by anchor — deep-link individual conventions in papers and discussions.
Version: v0.12Sections: 12Rules: 101CC-BY 4.0
Scope
Guidelines for annotating Ukrainian construction-domain text with named-entity, register, and dialect metadata.
scope-1
What the corpus covers
BUDOVA covers construction-domain Ukrainian across four registers: normative documents (ДБН / ДСТУ), specifications, estimates, and field speech. Out-of-domain text is rejected before annotation.
Монолітні залізобетонні конструкції з класом бетону C25/30.
Учорашній матч чемпіонату України.Out of domain — reject at intake.
scope-2
What is annotated
Every source fragment receives NER labels (9 entity types), a register label, and — for speech samples — dialect region and speaker ID.
Text + NER spans + register=normative + source=ДБН В.1.1-12:2014
scope-3
What is not annotated
Figure captions, formulas, table headers, and numeric-only fragments without surrounding context are excluded. Annotate sentences that carry domain information, not typography. This covers the bare marker only — a caption or a table row that the source text spells out as full sentences is running text and is annotated (see scope-5).
Табл. 3Skip table markers.
ΣM_x = 0Skip isolated formulas.
scope-4
Notation lists — annotate, never skip
A "symbol – definition" row from a Symbols / Notation section is annotated like any other text: the definition on the right is labeled by the normal rules, while the symbol itself (fck, MEd, γC, ⌀, εc, EQU, SLS, ULS) never receives a label — it is notation, not an entity. A row whose definition contains no domain entity is a valid zero-entity task: submit it empty. Do not skip it — those rows are the honest negative examples for a section where neighbouring rows do carry entities, and dropping them would teach the model that every notation row must contain something. Skipping stays reserved for non-text artefacts: table markers, isolated formulas, OCR garbage, truncated fragments.
γC – коефіцієнт надійності для бетонуSymbol and norm-assigned coefficient get nothing; the material does.
m – маса.No domain entity → submit as a zero-entity task, not a skip.
Skip the whole Symbols section as a glossary.The section is running normative text and is annotated row by row.
scope-5
Captions and table rows — marker vs sentence
Draw the line at the sentence. A fragment that is only a bare marker — "Табл. 3", "Рисунок 3.1", "Продовження таблиці 4.1" — is typography and is skipped by scope-3. A caption or a table row that the source spells out as full sentences ("Таблиця 2.1 – Коефіцієнти надійності матеріалів для граничних станів. Таблиця містить колонки: …", "Стрижнева арматура класу А500С діаметром від 8 мм до 22 мм – коефіцієнт надійності …") is running text: annotate it by the normal rules, even when the result is a zero-entity task. Two consequences are absolute. First, "none of the nine labels applies" is NEVER a reason to skip — an unlabelable fragment is submitted with zero spans and escalated (ambiguity-4). Second, "no-domain-content" is not a valid skip category: it contradicts scope-4 and manufactures exactly the selection bias the corpus must avoid, because the empty rows are the honest negatives of a section whose neighbouring rows are full. Figure captions and legends follow the same line — see scope-6.
Продовження таблиці 4.1Bare marker → skip.
Таблиця 4.1 (а) – Марки бетону за морозостійкістю та водонепроникністю…Caption written out as sentences → annotate, never skip.
skip = no-domain-contentNot a valid category — submit zero spans and escalate instead.
scope-6
Figure captions, legends, and sentences that contain formulas
The scope-5 line — marker versus sentence — holds for figures exactly as it does for tables, and no separate skip category exists for them. A bare figure marker ("Рисунок 3.1", "Рисунок 5.4 – продовження") is typography and is skipped. A caption that the source spells out as full sentences, together with the descriptive legend that follows it ("Рисунок 5.3 – … На рисунку показано схему таврової балки з анкером, напруженою арматурою…"), is running text: annotate it by the normal rules. Formulas INSIDE such a sentence do not turn it into an isolated formula. The skip category isolated-formula is reserved for a fragment that IS the formula and carries no predication ("ΣM_x = 0"); a sentence with a subject and a predicate that happens to list design values ("…значення відстані між точками з нульовими моментами: l0 = 0,85l1 для крайнього прольоту…") stays in the corpus, because the elements it names are ordinary spans. Inside such a caption the marker, the formula bodies and the symbols get nothing (scope-4); the domain nouns are labeled as usual.
Рисунок 3.1Bare marker → skip.
Рисунок 5.1 – До визначення відстані l0. На рисунку показано нерозрізну балку(structure) з прольотами…Caption expanded into sentences → annotate; the formulas inside get nothing.
skip = isolated-formula on a sentence that merely contains formulasReserved for a fragment that is the formula itself.
Entity types
Nine NER labels cover the BUDOVA corpus. Definitions below are the ground-truth for every annotator.
entities-material
material
Construction substances, composites, and named products — concrete, rebar, insulation, window units, drywall, cement. Multi-word canonical names are kept whole. Reinforcement products (стрижень, дріт, канат, пучок, прядка, сітка, анкер) are material, not structure: they are manufactured products fed into a member, and the narrower label wins (see ambiguity-1). The same test extends the list: anything supplied as a product and left permanently inside the member is material, not tool and not structure — поперечні в’язі, хомути, спіралі, зв’язки, закладні деталі, монтажні (підйомні) петлі, анкерні пристрої, муфти, пристрої для відхилення, оболонки (футляри) пучків. The word "пристрій" does not pull a permanently embedded product towards tool. The bare hypernym "матеріал" / "вид матеріалу" is not annotated (see ambiguity-17).
монолітний залізобетон
двокамерний склопакет
пучокPrestressing tendon — a reinforcement product → material.
закладні деталіSupplied and embedded forever → material, not tool.
бетонOnly-noun-without-qualifier is borderline; accept if canonical in context.
entities-tool
tool
Machinery, equipment, and instruments used on-site — excavator, crane, drill, trowel, scaffolding, formwork. tool is removable equipment: it performs its job and is taken away. Technological rigs of the tensioning operation belong here — домкрат, напружувальний стенд, упори, форма, опалубка. A device that stays in the member forever is material, not tool (see ambiguity-19).
баштовий кран
зварювальний апарат
натяг на упориRemovable rig → tool; the manner phrase "на" stays outside.
entities-process
process
Construction activities — casting, welding, installation, plastering, drilling, waterproofing. Prefer verbal-noun forms. process is work somebody PERFORMS on the material or the member; a mechanical phenomenon nobody performs (згин, обтиснення, прогин) is load_effect, not process — see entities-load-effect.
бетонування при температурі нижче +5 °C
попереднє напруження(process) — обтиснення(load_effect)The operation is process; the stress state it leaves is load_effect.
роблять бетонColloquial verb form — normalize to "бетонування".
entities-measurement
measurement
Numeric quantities with units, class codes, dimensions, and concentration percentages. Always mark the unit as part of the span.
переріз 400×400 мм
клас бетону C25/30
400Bare number without unit is not a measurement.
entities-structure
structure
Two levels, both structure: (a) structural and architectural elements — foundations, columns, beams, slabs, roofs, stairs, ramps; (b) engineering works and buildings as whole objects — bridge, viaduct, dam, pressure vessel, offshore platform, high-rise building. Keep the typological modifier inside the span. The generic scope cliché "будівель та (і) споруд" and a bare "будівля" / "споруда" without a typological modifier are not annotated.
несуча стіна підвалу
фундаментна плита
висотних будівельWhole object with a typological modifier → structure.
дамбиEngineering work as a whole object — same label as an element.
конструкції будівель та спорудGeneric scope formula — annotate nothing here.
entities-safety
safety
Safety equipment, procedures, and protective measures — PPE, helmets, harnesses, fire extinguishers, evacuation routes, protective railings — and a named field of safety carrying a qualifier: пожежна безпека, промислова безпека, вибухобезпека, біологічний захист. A bare "безпека" / "вимоги безпеки" without a qualifier is not annotated, by analogy with the bare "конструкція" of the structure rule. Standard citations go to regulation; numeric protection classes go to property. safety requires two conditions at once: (a) the phrase names a MEANS, a MEASURE or a NAMED DOMAIN with a qualifier, and (b) its immediate object is a PERSON. A non-human object — the environment, a structure, property — never gives safety (ambiguity-18), so «захист довкілля» takes no span and «[систем очистки та фільтрації](structure)» stays structure. The head need not be the word «безпека»: a named domain of human protection with a qualifier takes safety under another head too — «[санітарно-екологічна безпека](safety)», «[промислова санітарія](safety)», «[виробнича гігієна](safety)». But the BARE hyperonym of the domain — «гігієна», «здоров’я», «охорона здоров’я», «безпека», «вимоги безпеки» — takes no span, on the same principle as a bare «конструкція». An installed system whose function is to warn PEOPLE of danger is safety over its whole name: «[систем аварійної сигналізації](safety)» (ambiguity-39, ambiguity-37).
засоби індивідуального захисту
аварійне освітлення шляхів евакуації
пожежної безпекиNamed field of safety with a qualifier → safety.
вимоги безпекиBare "безпека" with no qualifier — annotate nothing.
ДБН А.3.1-5:2016A standard citation is regulation, not safety.
entities-regulation
regulation
Normative references — ДБН, ДСТУ, СОУ, ISO and EN standards, Eurocodes, technical specifications (ТУ) — and legal acts carrying normative force: laws, Cabinet of Ministers resolutions, technical regulations, ministerial orders. Keep the full code with the year as one span (see boundaries-3); when the document has a code, the span stops at the code and the descriptive title stays outside it (see boundaries-4). regulation always needs an identifier: a code, an official title, or an established proper-name alias ("Єврокод 2"). A document referring to itself — "у цих Нормах", "цих Норм", "цього стандарту", "даного документа" — is not annotated: the phrase is a deictic common noun, it repeats in almost every clause of a ДБН, and labeling it would flood the corpus with identifier-free regulation spans. Three further reference types carry no identifier and are therefore not annotated either: generic references ("згідно з чинними нормативними документами", "за відповідною нормативною та (або) технічною документацією", "відповідні стандарти"); internal cross-references to a structural unit of the same document ("згідно з 3.2.6", "таблиця 2.1", "рисунок 3.6", "формула (4.2)", "додаток А") — the number there is a locator, so it is not a measurement either; and project or as-built documentation with no normative code ("робочі креслення", "проектна документація", "виконавча документація"). A cross-reference to ANOTHER document that does carry a code stays regulation. Legacy and foreign series (СНиП, ГОСТ, СН), the reissue asterisk, bracketed harmonization codes and the classification index of the document itself are handled by boundaries-10.
ДБН В.2.2-40:2018
ДСТУ EN 81-70
відповідно до ДБН В.1.2-14Cross-reference with a code → regulation.
згідно з таблицею 2.1Internal locator — neither regulation nor measurement.
за чинними нормативними документамиGeneric reference with no identifier — annotate nothing.
У цих Нормах застосовуються…Self-reference with no identifier — annotate nothing.
Постановою Кабінету Міністрів України від 20 грудня 2006 р. № 1764A legal act with no ДБН/ДСТУ index — one span, act type + number + date; the date and number inside are not a separate measurement.
Технічному регламенті будівельних виробів, будівель і спорудNo code — the official title itself is the span; the surrounding quotation marks stay outside.
entities-property
property
Named technical properties and performance classes — fire-resistance class, thermal conductivity, sound-insulation index, load-bearing capacity, strength grade. Also the named normed design characteristics on the resistance side: stability, crack resistance, durability, serviceability, stiffness, critical force, effective length, slenderness. A class name with no code ("клас бетону", "клас легкого бетону") is one property span including the material noun; with a code it becomes a measurement ("клас бетону C25/30"). "Строк служби" / "строк експлуатації" is the normed face of durability → property; a value beside it ("50 років") is a separate, adjacent measurement span. "Теплоізоляція" / "звукоізоляція" is property when it names a requirement or a characteristic, and material when it names the product or the layer ("укладання теплоізоляції завтовшки 100 мм"). A numeric value with its unit is a measurement; the property name itself is property. Forces and the states they produce (load, internal force, bending, deflection) sit on the ACTION side and take load_effect, not property — see entities-load-effect and ambiguity-3.
клас вогнестійкості R120
несуча здатність
критична силаA named normed characteristic of the member → property, despite the word "сила".
клас бетонуNo code → one property span; the material inside is absorbed (see boundaries-5).
0.045 Вт/(м·К)Value with unit is measurement, not property.
внутрішні зусилля(property)Action-side effect → load_effect, never property.
entities-load-effect
load_effect
The ACTION side of the Eurocode inequality E_d ≤ R_d: what acts ON the structure, and the mechanical state that action produces. Four families. (a) Actions and loads — навантаження, постійне / змінне / особливе навантаження, власна вага, снігове й вітрове навантаження, сейсмічний вплив, температурний вплив. (b) Internal forces — зусилля, внутрішні зусилля, поздовжня (нормальна) сила, поперечна сила, згинальний момент, крутний момент, тиск, розпір. (c) Stress and strain states — напруження, напруження стиску, дотичні напруження, деформація, відносна деформація, прогин, переміщення, осідання, згин, стиск, розтяг, кручення, зріз, обтиснення. (d) Named design states — граничний стан, граничний стан за несучою здатністю (ULS), граничний стан за придатністю до нормальної експлуатації (SLS), напружено-деформований стан, статична рівновага (EQU). Two contrasts fix the label. Against PROPERTY: property is what the structure HAS or is normed to withstand (несуча здатність, стійкість, міцність, жорсткість, критична сила, розрахункова довжина) — ask "does the structure HAVE this, or does this ACT on it?"; the answer, not the surface word, decides, which is why "критична сила" is property and "поперечна сила" is load_effect. Against PROCESS: process is work somebody PERFORMS that changes the material or the member (бетонування, натяг, попереднє напруження); a mechanical phenomenon nobody performs is load_effect. A number with its unit stays a separate adjacent measurement. Span boundaries: ambiguity-20; fixed turns (nullifiers, limit states, canonical interiors): ambiguity-21.
згинальний момент(load_effect)A force moment — action side.
внутрішні зусилля(load_effect)What acts on the member (v0.4 left this unannotated).
навантаження(load_effect) 5 кН/м(measurement)The action name and its value are separate adjacent spans.
критична сила(load_effect)A named normed characteristic of the member → property.
попереднє напруження(load_effect)A technological operation → process; the state it leaves ("обтиснення") is load_effect.
Span boundaries
Where the annotation starts and ends — the single most common source of disagreement.
boundaries-1
Include modifiers
When a domain adjective or quantifier changes the meaning of the head noun, include it in the span.
монолітний залізобетон"monolithic" is domain-salient.
монолітний залізобетонLosing the modifier changes the sense.
boundaries-2
Exclude articles and fillers
Do not include bare articles, prepositions, or discourse fillers at the edges of a span.
у фундаменті
у фундаменті
boundaries-3
Keep canonical multi-word terms whole
Terms with a fixed collocation (e.g. ДБН + code, class + number) never split across annotations.
ДБН В.1.1-12:2014
ДБН В.1.1-12:2014
boundaries-4
Bibliographic reference — the span stops at the code
In a "code + title" citation the regulation span is the document identifier only: series, number, year. Everything after it — the "Єврокод 2"-style alias that opens the official title, the colon, the descriptive title, the part designation — stays outside the span. A document with no code is spanned by its official title instead. A standalone alias in running text, with no code beside it, is a regulation span of its own.
EN 1992-1-1:2005 Єврокод 2: Проектування залізобетонних конструкцій – Частина 1-1…The span ends at the year.
EN 1992-1-1:2005 Єврокод 2: Проектування залізобетонних конструкційThe title is document metadata, not part of the identifier.
згідно з Єврокодом 2Alias used on its own → regulation.
boundaries-5
A property name split by a material
Spans may not overlap, so a canonical property name interrupted by a material in the genitive ("міцності бетону на стиск") has to be cut. Ask whether the part before the material is a property name on its own. If it is — span that part as property, give the material its own span, and leave the trailing qualifier ("на стиск", "на осьовий розтяг") unannotated. If it is not — keep the whole phrase as one property span and let the material be absorbed. Nothing splits an uninterrupted term: "міцності на розтяг" is one full property span. The trailing qualifier does not become a load_effect either: "на стиск" inside a property name states the kind of resistance, not an action (ambiguity-21).
міцності(property) бетону(material) на стиск"міцність" names a property on its own → cut, keep the material.
густина(property) бетону(material)
клас бетону(property)"клас" alone names no property → one span, material absorbed.
міцності на розтяг(property)Uninterrupted canonical term → full span.
boundaries-6
Coordinate group with a shared head
Two coordinated modifiers sharing one head noun ("арматурної і попередньо напруженої сталі") receive ONE span over the whole group, conjunction included. Splitting would leave a headless "арматурної", which is not a usable entity, and spans cannot overlap on the shared head. Two spans only when the head is repeated for each member ("арматурна сталь і напружувана сталь"). No inner span is placed on a modifier inside such a group — "попередньо напруженої" gets no process span there.
арматурної і попередньо напруженої сталіOne material span over the whole group.
арматурна сталь і напружувана стальHead repeated → two spans.
арматурної і попередньо напруженої сталіHeadless first member.
boundaries-7
Material named by an adjective alone
A relational material adjective is annotated as material when its head noun gets no label at all — the material mention is real domain content and should not be lost with the head. When the head noun IS labeled, the adjective sits inside that span and never gets a span of its own.
бетонного(material) перерізу"переріз" without numbers is not annotated → keep the material.
залізобетонна колона(structure)Head is labeled → the adjective is inside its span.
важкий бетон(material)
бетоннастінаDo not split a labeled head from its adjective.
boundaries-8
Composition qualifier and an elided head
A prepositional group opened by на / з / із that names the VARIETY of a material by its composition or aggregate belongs inside one material span, because it changes which substance is meant: "[бетон на кварцових заповнювачах]", "[заповнювачі із вапняку]". A prepositional group of purpose or location does not: "[арматура] для [фундаментних плит]". When a list of varieties has lost its head noun to the previous clause ("Арматуру поділяють на такі види: гарячекатану гладку; термомеханічно зміцнену…"), every item that names a variety by manufacture or composition takes its own material span although it is headless — the head is recoverable from context, and refusing to annotate would drop the whole enumeration. This does not reopen boundaries-6: when the head noun IS present inside the same noun phrase, the group still takes exactly one span. A purely geometric profile description gets no span of its own but stays inside the span of the product it follows.
бетону на кварцових заповнювачахComposition qualifier → inside one material span.
види: гарячекатану гладку; термомеханічно зміцненуHead elided in a previous clause → each variety is its own span.
арматура періодичного профілюPostpositive profile description stays inside the product span.
бетон на кварцових заповнювачахDo not split a composition qualifier off its head.
boundaries-9
Coordinated property heads, wrappers, and a number inside a term
The mirror image of boundaries-6: when the coordinated members are themselves property NAMES sharing one dependent material, each is a usable entity on its own, so each takes its own span and the material takes one more — "характеристики [міцності](property) і [деформативності](property) [бетону](material)". Quantity wrappers stay outside the span, extending the statistical wrappers of ambiguity-7 to "величина", "значення", "товщина", "розмір": "Мінімальна величина [захисного шару бетону]". A number with its unit that is an inseparable part of a canonical property name stays INSIDE that property span and is not carved out as a measurement (nested-1); a number that merely reports a value stays a separate adjacent measurement.
характеристики міцності і деформативностібетонуCoordinated heads → one span each.
0,1 % умовної границі текучостіThe percentage is part of the canonical term — no inner measurement.
строк служби50 роківReported value → separate adjacent measurement.
мінімальна величина захисного шару бетонуThe wrapper "величина" stays outside the span.
boundaries-10
Bibliographic entry: legacy series, the reissue asterisk, harmonization codes, classification indices
The series list of entities-regulation is OPEN. Any inherited or foreign designation shaped as series + number + year is a regulation span whatever the series — СНиП, ГОСТ, СН, ISO, EN, ДСТУ EN — because the label is defined by the identifier, not by the issuing country. The reissue asterisk is part of the official designation and stays INSIDE the span: "[СНиП 2.03.01-84*]". A harmonization reference in brackets at the end of an entry — "(ГОСТ 31384-2008, NEQ)", "(EN 1992-1-1:2004, IDT)" — identifies ANOTHER document, so nested-3 does not suppress it: the code takes its own regulation span, while the brackets and the degree marker (IDT / MOD / NEQ) stay outside. A classification index of the document itself — Код УКНД, ICS, УДК, ДКПП — is bibliographic metadata: it identifies a shelf, not a normative act, and it is not a class of anything the norm measures. It gets NO span, exactly like the internal locators of entities-regulation.
СНиП 2.03.01-84*Legacy series with the reissue asterisk inside the span.
ДСТУ Б В.2.6-145:2010 … (ГОСТ 31384-2008, NEQ)Bracketed harmonization code → own span; the degree marker stays outside.
Код УКНД 91.080.40Bibliographic index — no span, neither regulation nor measurement.
boundaries-11
A parenthetical synonym inside a canonical term
A span is contiguous, so a canonical term interrupted by a bracketed synonym — "характеристична (нормативна) міцність", "межі (фізичної або умовної) текучості", "деформаційних (жорсткісних) характеристик" — is spanned WHOLE, brackets included. Narrowing the span to the head ("міцність") throws away the qualifier that says WHICH property is meant, and no two annotators narrow it the same way. The rule covers the case where the bracket holds a SYNONYM or a variant of the same term. A bracket holding a different entity keeps its own span, and a bracket holding a unit, a symbol or a code is governed by numeric-4, scope-4 and ambiguity-10 as before.
характеристична (нормативна) міцністьNarrowing to the head loses the property class.
boundaries-12
Inverted term-section headword; a material adjective on "конструкції"
A terminological article inverts the term — "Арматура конструктивна.", "Бетон важкий.", "Конструкції залізобетонні." The postpositive modifier belongs to the canonical term and stays INSIDE the span, by the same reasoning as the postpositive profile description of boundaries-8. The second half of the rule closes a conflict between boundaries-7 and entities-structure: "конструкція" is a bare generic head, but a MATERIAL adjective on it names a CLASS of structures and licenses one structure span with the adjective inside — "[бетонні та залізобетонні конструкції](structure)". boundaries-7 keeps its own scope: a material adjective becomes a material span only when the head noun is not labeled at all ("[бетонного](material) перерізу"). A locative or exposure modifier still licenses nothing (ambiguity-17).
Конструкції залізобетонні(structure)Inverted headword → one span, modifier inside.
Бетон важкий(material)
Конструкції бетонні(material)The head is labeled here, so the adjective sits inside the structure span.
boundaries-13
The kind-adjective of a work name; a coordinated modifier list at one head
Two closures to ambiguity-30 (a), read off the places where all three annotators converged. (a) THE KIND-ADJECTIVE goes INSIDE the process span, and not only in "санітарне очищення". An adjective that says WHICH KIND of work is meant — by technology, by means, or by organization — belongs to the name of the work: "[санітарне очищення]", "[роздільне збирання]", "[механізоване прибирання]", "[ручне прибирання]", "[термічне перероблення]", "[механізоване посипання]", "[технічне обслуговування]", "[планово-регулярне вивезення]". Only an adjective that syntactically belongs to ANOTHER head stays outside: "планово-регулярна система [санітарного очищення](process)", where "планово-регулярна" agrees with "система". The check is mechanical — if the adjective agrees with the deverbal noun in gender, number and case, it is inside; if it agrees with another word, it is outside. This makes the phrase "an adjective that belongs to the canonical name of the work" in ambiguity-30 (a) operational and settles the most frequent boundary disagreement of the genre. (b) A COORDINATED LIST OF MODIFIERS AT ONE HEAD is ONE span, commas and conjunction included: "[твердих, ремонтних, великогабаритних, рідких побутових відходів](material)". The list is never broken up — a modifier without its head is not an entity, and the text does not repeat the head. This does not conflict with ambiguity-10 (3), where the separate spans go to CODES, not to modifiers. The agreement test in (a) is a NECESSARY but not a sufficient condition: it screens out an adjective belonging to another head, and it does not admit a descriptive one. The sufficient condition is CANONICITY: the adjective must name a KIND of work by technology, means or organisation, so that «<adjective> + <work>» is a name the domain uses («[санітарне очищення]», «[роздільне збирання]», «[технічне обслуговування]»). An adjective describing only this instance of the work stays OUTSIDE the span however well it agrees: додатковий, відповідний, необхідний, належний, спеціальний, окремий, подальший — «додаткової [обробки](process)», «відповідному [проектуванню](process)».
роздільного збирання(process) побутових відходів(material)The adjective agrees with "збирання" → inside the span.
планово-регулярна система санітарного очищення(process)The adjective agrees with "система" → outside.
механізованого прибирання(process)The kind of work is not torn off the name of the work.
Nested entities
When one entity appears inside another.
nested-1
Annotate the outer span
If a longer material name contains a numeric class, mark the full material span; the measurement stays inside and is not separately annotated.
бетон класу C25/30One material span; C25/30 is part of it.
бетон класу C25/30Split annotation creates ambiguity.
nested-2
Split at natural boundaries
When two distinct entities abut, give each its own span.
фундаментна плита під колоною
nested-3
No entities inside a document title
Domain words that occur inside a document title, or inside the span of a legal act, are never annotated: a title quotes the name of a document, it is not a domain sentence. Annotating inside titles inflates process / structure counts on every reference list and destroys comparability between documents.
EN 1992-1-1:2005 Єврокод 2: Проектуваннязалізобетонних конструкційDo not annotate inside a title.
Технічному регламенті будівельних виробів, будівель і спорудOne regulation span; "будівель і споруд" inside it gets nothing.
nested-4
What nested-3 does not cover: own-section headings, drawing titles, keyword blocks
nested-3 suppresses annotation inside the title of ANOTHER normative document quoted as an identifier. It does not extend to three things that recur throughout an administrative-planning norm, and all three are annotated by the general rules. (a) A HEADING OF THE DOCUMENT OWN SECTION, quoted in running text — «У підрозділі "Техніко-економічна оцінка існуючого стану [санітарного очищення](process)" наводять …» — is not a quotation of another document but a pointer to a place inside the norm itself. (b) A DRAWING OR SCHEME TITLE — «креслення "Схема [санітарного очищення](process) населеного пункту. Існуючий стан"» — likewise. (c) THE KEYWORD METABLOCK that closes every ДБН is a list of domain terms in dictionary form and is annotated as ordinary text (scope-5); leaving it empty would manufacture exactly the selection bias scope-5 forbids. The same holds for settled legal names of records in running text — «реєстр місць [видалення](process) [відходів](material)», «паспорт місць [видалення](process) [відходів](material)» — which carry no identifier and stand in no bibliographic position. The boundary is simple: nested-3 fires only where a document code, or a regulation span over an official title, stands beside the words. Separately, the STATE-REGISTRATION TAIL of a legal act — «зареєстрований у Мін’юсті 12.02.2004 за № 189/8788» — is a requisite of the same act and not a second document: it stays OUTSIDE the regulation span and takes no label. The act BY WHICH the norms were approved — «затверджені наказом МОЗ України від 17.03.2011 № 145» — does carry its own identifier and takes a SECOND regulation span. The caution concerns zero annotation as a POLICY — exempting the metablock from annotation; a zero RESULT obtained by applying the general rules stays valid and is submitted empty (scope-4). In a framework norm whose keywords are bare hyperonyms of the topic («гігієна», «здоров’я людини», «захист довкілля»), an empty block is the correct answer.
У підрозділі "Техніко-економічна оцінка існуючого стану санітарного очищення(process)"A heading of the document own section is ordinary text.
Державні санітарні норми і правила утримання територій населених місць, затверджені наказом МОЗ України від 17.03.2011 № 145Two regulation spans; the registration tail stays outside both.
ДБН В.2.4-2-2005 Полігони(structure) твердих побутових відходів(material)A bibliographic title of another document — nested-3 applies here.
Dialectal and colloquial terms
Handling regional vocabulary and informal speech.
dialect-1
Annotate the surface form
Use the exact dialectal word as it appears. Canonical mapping is a separate metadata field, not a substitution.
ґіпсокартонWestern variant; keep "ґ".
шпахлівкаDialectal; map to "шпаклівка" in metadata.
dialect-2
Tag dialectal origin
For speech samples, attach region + dialect group (Northern / Southwestern / Southeastern — the standard three-group Ukrainian taxonomy) to the utterance, not to individual spans.
Region tag is an utterance-level field in the speech subset schema.
Recording protocol
How to capture speech samples so they pass the corpus-quality bar. Server-side checks measure SNR and peak for WAV uploads; the in-browser recorder enforces format. Full contributor spec: docs/recording-spec.md.
record-1
Device & permissions
Use a dedicated mic if possible (USB condenser, lavalier, decent phone). Wear headphones to prevent speaker bleed. Grant microphone permission to the browser when prompted — the platform stops the audio track when recording ends, so the indicator should go away.
AGC / noise suppression / echo cancellation are turned off by the in-browser recorder. If you use an external DAW, disable those too.
record-2
Room conditions & SNR
Record in a quiet room (no HVAC blast, no traffic). Take 3–5 seconds of silence first and check the noise floor on your meter. Corpus target is SNR ≥ 20 dB; the upload response includes a warning when a WAV measures below.
Indoor office, doors closed, fan off → typical SNR 30–40 dB.
Construction site without a windscreen → may fall below 15 dB; note the noise level in source_ref.
record-3
Levels — peak −12 to −3 dBFS
Loud enough to be clearly above noise, quiet enough not to clip. Aim for peaks between −12 and −3 dBFS. The server rejects nothing on level alone but stores peak_dbfs so clipped takes can be filtered at export.
Peak ≥ −0.1 dBFS → clipped, distortion baked in.
Peak ≤ −40 dBFS → too quiet, low effective SNR.
record-4
Format & duration
Prefer mono 48 kHz 16-bit PCM WAV for external-DAW uploads. The in-browser recorder produces 48 kHz mono WebM/Opus at 64 kbit/s (Safari falls back to MP4/AAC). One utterance per file, 30 s – 3 min. Max 50 MB.
A 90-second mono WAV at 48 kHz of one speaker explaining a concrete pour.
A 12-minute stereo file with three speakers and background TV.
record-5
One speaker per file
Each recording should have a single speaker. If you are interviewing someone, split into one file per speaker. If a true dialogue is captured, note it in source_ref ("2 speakers, foreman + worker").
speaker_id is auto-assigned per contributor; mixing speakers inside one file breaks that assumption.
record-6
PII & content
No full names, addresses, phone numbers, license plates, or personal medical info. Dialectal vocabulary is welcome — tag your region so dialect researchers can find it. Construction-domain only.
"…шпахлівку накладали в три проходи з шліфуванням між ними."
"Я, Іван Петренко, проживаю за адресою…"PII; either reject the recording or re-record without identifying details.
Brand vs material
Distinguishing product brands from generic substance names.
brand-1
Brand stays with the material
When a brand name qualifies a material, annotate the full span as material and store the brand in a sub-field.
склопакет REHAU 70Material with brand-qualified subtype.
склопакетREHAU 70Do not split into two entities.
brand-2
Standalone brand → skip
A brand name without a construction-material context is not annotated.
REHAUAlone in the text — no material context.
Numeric values
Dimensions, classes, percentages, and compound measurements.
numeric-1
Always include the unit
A number without unit is not a measurement span. Include mm/cm/m/МПа/°C/%/class-code suffix in the annotation.
30 мм
25 МПа
30
numeric-2
Dimension products
For multi-axis dimensions (400×400 mm), keep the entire product in one span. Same for tolerance ranges.
400×400 мм
±5 мм
numeric-3
Class codes
Class labels (C25/30, XC3, B400C) are measurement spans; include the prefix letter and slash.
класу C25/30
клас експозиції XC3
numeric-4
A unit without a number is not a measurement
A measurement span is a number together with its unit. A bare unit closing a definition (", кг/м3", "(%)", "МПа") is not a measurement — symmetrically to the bare-number rule in numeric-1. The same holds for a symbol standing for a quantity: without a value it stays unlabeled (see scope-4).
1000 год
20 °C
густина бетону …, кг/м3Unit alone — no measurement span.
втрат від релаксації (%)Same: percent sign with no value.
numeric-5
Serialized table row — a declared unit does not travel
Flattened table rows declare the unit once and then list bare values: "Характеристична кубикова міцність бетону на стиск fck,cube, МПа, за класами бетону: C8/10 – 10; C12/15 – 15; …". The unit does NOT propagate. The declared unit standing on its own (", МПа,", ", мм,") is not a measurement (numeric-4), and the bare values after the dash are not measurements either (numeric-1). The row does not come out empty: the class codes in it ARE measurements by ambiguity-10, so "[C8/10]", "[C12/15]", "[X0]", "[XC1]" carry the row. Dimensionless table multipliers with footnote markers ("1,3*", "1,5/1,3**") are bare numbers and get no span. The decision is deliberately conservative: binding a value to a unit declared elsewhere in the fragment needs a scope judgement no two annotators make identically, and the same row routinely mixes values in the declared unit with class codes that are not in it.
…, мм, за класами впливу середовища: X0 – 20; XC1 – 35The codes are the measurements; the bare values and the declared unit are not.
…, МПа, за класами бетону: C8/10 – 10A unit declared in the preamble does not license a span on the value.
1,3*Dimensionless multiplier with a footnote marker — bare number, no span.
numeric-6
Intervals and ranges — one span
A range is one measurement span, exactly like the dimension products of numeric-2. The span starts at the first digit (or at "мінус") and ends at the last unit; the prepositions and adverbs framing it stay outside. Never split a range into two measurement spans, and never annotate only its second half.
діаметром від 6 мм до 40 мм
від 7 до 10 %Unit only at the second number — still one span.
понад мінус 20 до мінус 40 °С включно"понад" and "включно" stay outside.
від 6 мм до 40 ммDo not split an interval.
Register labels
Classifying the stylistic register of each source fragment.
register-1
Normative
ДБН / ДСТУ / СОУ text and other standards — formal legal register with explicit numeric thresholds and "must/shall" modals.
Віконний блок ПВХ серії REHAU 70, розміри 1400×1600 мм.
register-3
Estimate
Cost estimates, work-item lines, unit-rate tables. Recognized by work-type verb phrases and per-unit coefficients.
Улаштування покриття з металочерепиці, коефіцієнт 1.08.
register-4
Field
On-site speech and informal field notes — dialectal vocabulary, approximate numbers, discourse fillers.
Стяжку лили в чотири, ну п’ять з хвостиком.
Ambiguity
What to do when a span could fit more than one label.
ambiguity-1
Prefer narrow label
If a term could be a material OR a structure, choose whichever is narrower in the current sentence.
"Стіна" in "бетонна стіна" → structure (the element); in "виготовлення стін" → structure (the element).
ambiguity-2
Flag for review
Tasks where two annotators disagree after the first pass go to the admin-review queue with a "needs adjudication" flag.
The agreement metadata records per-word agreement; items below the threshold are surfaced in the admin dashboard. See the Agreement & merging section.
ambiguity-3
Actions vs resistances (E vs R)
Structural-mechanics vocabulary is split by the Eurocode E ≤ R logic, and since v0.5 BOTH sides carry a label. One question decides: does the structure HAVE this, or does this ACT on it? HAS → property: несуча здатність, стійкість, тріщиностійкість, довговічність, придатність до експлуатації, жорсткість, міцність, критична сила, розрахункова довжина, гнучкість, границя текучості / міцності. ACTS on it, together with the state it produces → load_effect: навантаження, вплив, сила, зусилля, силовий момент, згин, стиск, розтяг, кручення, напруження, прогин, деформація, обтиснення, граничний стан. The temporary v0.3 / v0.4 instruction "the action side is not annotated with any label" is WITHDRAWN: the owner approved the ninth label on 2026-07-20, and every rule in this document now labels the action side load_effect. Physical work performed on the structure (попереднє напруження, бетонування, зварювання) remains process; a mechanical phenomenon nobody performs is load_effect, never process. Entries annotated under v0.3 / v0.4 are not wrong but incomplete: the change is purely additive — no span that was correct then changes its label now — so the action side is filled in by a second pass. Span boundaries are in ambiguity-20, fixed turns in ambiguity-21.
несуча здатність(property)Resistance side → property.
критична сила(property)A named normed characteristic of the member → property.
внутрішні зусилля(load_effect)Action side → load_effect; v0.4 left this unannotated.
попереднє напруження(process)A technological operation on the structure → process.
втрата стійкості(load_effect)One span over the whole turn — see ambiguity-21.
ambiguity-4
Uncovered domain term → leave it and escalate
When a term is plainly domain-central but fits none of the nine labels, do not stretch the nearest label to cover it. Leave it unannotated, stay consistent for the rest of the document, and escalate to the coordinator. A missing entity is recoverable by a later pass; a forced label silently degrades agreement and the trained model, and is far more expensive to undo.
Consistent under-annotation is repairable; inconsistent over-annotation is not.
ambiguity-5
Ability vs event
The named ability — usually the -ість noun — is property; the state or quantity it produces belongs to the action side and takes load_effect. property: міцність, стійкість, тріщиностійкість, довговічність, повзучість, усадка, релаксація, пластичність, деформативність, границя текучості, границя міцності. load_effect: деформації усадки, прогин, напруження — the HEAD noun decides, by the closed list of ambiguity-20. Still annotated with nothing: тріщина, розкриття тріщин — a defect is neither an action nor a resistance, and its head is not on that list (ambiguity-21 d). Two fixed frames still take ONE span over the whole frame with no span inside it, but that span is now load_effect instead of a hole: "[втрата / втрати X]" (втрата стійкості, втрата несучої здатності, втрати від релаксації) and "[граничний стан за X]" (ULS / SLS) — a partial span inside them is still the asymmetry that ruins κ. The wording of ambiguity-3 covers "граничний стан", never "границя текучості", which stays property. "Напруження" as an internal force (напруження стиску, крутні напруження зрізу) is load_effect, while "попереднє напруження" as a technological operation stays process. "Статична рівновага" (EQU) is a named design state → load_effect.
втрата несучої здатності(load_effect)The whole frame is one span; no inner property span.
граничний стан за несучою здатністю(load_effect)ULS is one design situation → one span, nothing inside it.
похила тріщинаA defect: not an element, not an action — annotate nothing.
граничний стан за несучою здатністюNever split the ULS designation into two spans.
ambiguity-6
Raw dimension vs section characteristic
Ask whether the quantity is a plain size you could measure off the drawing, or a named design characteristic computed to describe stiffness, stability or resistance. Plain sizes are not annotated without a number: довжина, висота, ширина, товщина, діаметр, глибина, проліт, площа, периметр, кут, кривизна, плече, ексцентриситет. Named design characteristics are property: момент інерції, статичний момент площі, момент опору, радіус інерції, гнучкість, розрахункова довжина, робоча висота перерізу, жорсткість. A closed list of design modifiers promotes a plain size into this family — see ambiguity-13. This resolves the "момент" clash of ambiguity-3: a force moment (згинальний, крутний) is on the action side → load_effect, while a moment of inertia is a property. The symbol beside it (MEd) stays outside the span and unlabeled, as every symbol does (scope-4). Parts of a cross-section (поличка, стінка, ребро) and "Т-подібний переріз" are not annotated — same reasoning as the bare "переріз".
момент інерціїComputed characteristic of the section → property.
радіус інерціїSame family as slenderness and effective length.
згинальний момент(load_effect) MEdForce moment → load_effect; the symbol stays outside (scope-4).
висотаPlain size with no number → nothing; "[висота 400 мм]" is a measurement.
ambiguity-7
Coefficients — nature-fixed vs norm-assigned
Ask who fixes the value. If the nature of the material or the geometry of the member does, the coefficient is a characteristic the object HAS → property, one span including the word "коефіцієнт": коефіцієнт Пуассона, коефіцієнт тертя, коефіцієнт повзучості, коефіцієнт лінійного температурного розширення, коефіцієнт теплопровідності, коефіцієнт армування. If the norm assigns it as a design multiplier, it is a parameter of the calculation procedure and not a property of anything → not annotated: коефіцієнт надійності (γ), частковий коефіцієнт, коефіцієнт умов роботи, коефіцієнт сполучень (ψ), коефіцієнт надійності за відповідальністю, коефіцієнт прямолінійності, коефіцієнти перерозподілу / збільшення / зменшення. Never carve one word out of a coefficient name ("надійності", "умов роботи") — a canonical term is not split. A material named beside it keeps its own span. Statistical wrappers — "середнє значення", "граничне значення", "характеристичне значення" — stay outside the span, while a domain adjective belonging to the canonical term stays inside ("[характеристична міцність]"). Two further families — coefficients of the action itself and statistical coefficients of test data — also stay unannotated, which closes the rule: a coefficient is a multiplier of the calculation, so it never becomes a load_effect either — see ambiguity-12.
коефіцієнт ПуассонаFixed by the material → property.
коефіцієнт надійності для бетонуNorm-assigned multiplier gets nothing; the material keeps its span.
коефіцієнт надійностіDo not carve the property word out of the term.
граничне значення коефіцієнта повзучостіThe statistical wrapper stays outside: "[коефіцієнта повзучості]".
ambiguity-8
process — construction and design actions only
process covers on-site and technological actions (бетонування, монтаж, зварювання, армування, попереднє напруження, зведення, будівництво) plus a closed list of design actions: розрахунок, проектування, конструювання, моделювання, випробування. General-purpose deverbal nouns performed by nobody in particular are not annotated: аналіз, розгляд, визначення, урахування, забезпечення, оцінювання, застосування. "Експлуатація" is a life-cycle stage, not an action → not annotated; "придатність до експлуатації" remains property as a whole term. Inside a document title nothing is annotated at all, "Проектування" included (nested-3). A mechanical phenomenon nobody performs never takes process: згин, обтиснення, прогин, руйнування are load_effect (ambiguity-3). The pair to memorize is "[попереднє напруження](process)" — the operation — against "[напруження](load_effect)" — the stress it creates.
розрахунок(process) за граничними станами(load_effect)The design action is process; the design situation is load_effect.
у процесі будівництва і експлуатації"будівництво" is an action; "експлуатація" is a stage.
Забезпечення надійностіService deverbal noun — annotate nothing.
ambiguity-9
Environment and calculation abstractions
The environment splits by its head word. A named climatic or seismic ACTION takes load_effect: "[температурний вплив]", "[температурні впливи]", "[снігове навантаження]", "[вітрове навантаження]", "[сейсмічний вплив]". A bare environment PARAMETER is not an action on the member but a description of the ambient condition, so it keeps the v0.4 treatment and gets no span without a value: температура зовнішнього повітря, відносна вологість, агресивність середовища; when a value with a unit is present, the value is a measurement ("мінус 22 °C"). Modelling premises and abstractions are not annotated either: геометричні недосконалості, фізична та геометрична нелінійність, недеформована / деформована схема, розрахункова схема. "Елемент" is a structure only with a constructive-typological modifier (згинальний елемент, залізобетонний елемент); with a purely geometric one (плоский, просторовий, об’ємний) it is not annotated. A bare "будівельний об’єкт" joins the bare "будівля / споруда / конструкція" of the structure rule and is not annotated. The list of environmental parameters is open and covers every named physical factor of the environment: noise, vibration, infrasound, insolation, illuminance, microclimate, air velocity, atmospheric pressure, ionising and electromagnetic radiation. What decides is not the quantity itself but WHOSE characteristic it names: measured in the environment or in a room it is a parameter and takes no span (F3 fires before the property list in ambiguity-34); the same quantity as a normed characteristic of a STRUCTURE or a MATERIAL is property — «[теплова ізоляція](property)», «[звукоізоляція](property)» in a list of structural characteristics, while «шум та вібрація» among environmental factors stay empty. ORDER OF CHECKING. First ask WHOSE characteristic the quantity names, and only then consult the list of factors: measured in the environment or in a room → a parameter, no span; measured in a PRODUCT or a STRUCTURE → property, even if the factor itself is on the list. «Шум та вібрація» among environmental factors are empty, while «[звукоізоляція](property)» and «[сумарна активність природних радіонуклідів](property)» are property.
температури зовнішнього повітря мінус 22 °CAmbient parameter → only the value with its unit is a span.
снігове навантаження(load_effect)Head word from the action list → load_effect.
Resolve a class or grade phrase by COUNTING the codes that follow its head. A code is a designation that contains a digit: C25/30, XC1, XD1/XS1, F75, W6, S6, CC3, А500С, В500, К-7, and the bare ordinal of "клас 1". A ROMAN numeral used as the ordinal of a scale — I, II, III, IV, V — is a code exactly like the Arabic one: the test is whether the designation names a step of the scale, not which alphabet prints it. A code may also PRECEDE its head ("III – IV класів небезпеки", "II категорії"), and position never changes the count. (1) NO code → one property span over the whole scale name, the dependent material noun included: "[клас бетону]", "[класу арматури]", "[клас сталі]", "[марка бетону за морозостійкістю]", "[класи умов експлуатації]", "[класи за релаксацією]", "[клас конструкції]", "[клас небезпеки]". (2) EXACTLY ONE code, directly next to the head — after it or before it → one measurement span from the word клас / марка to the code inclusive: "[клас умов експлуатації ХD1]", "[класу наслідків СС3]", "[клас 1]", "[II категорії]". (3) TWO OR MORE codes at one head → the head takes a property span and EVERY code takes its own measurement span; commas, semicolons and dashes stay outside. Never run one span across a comma list or across a code range. (4) A code separated from its head by an intervening word, a dash or a bracketed synonym is its own measurement span, and the head is labeled by its own rule. (5) When the head is the MATERIAL rather than the word клас / марка, nested-1 governs: a single code is absorbed ("[бетон класу C25/30]"), a list is never absorbed. (6) A letter family that names no step of the scale (А, В і К; В, Вр) is not a code and gets no measurement span, while the scale name keeps its property span; a Roman numeral is an ordinal, not such a letter family. (7) "Клас", "ступінь", "категорія", "раз" used as counting nouns are not units: "збільшення на 2 класи" is not a measurement.
клас умов експлуатації ХD1One code right after the head → merged measurement.
класами впливу середовища(property): X0 – 20; XC1 – 35Code list → property head + one measurement per code.
Класи арматури за міцністю на розтяг(property) А, В і КLetter family without a digit is not a code.
арматурні стрижні(material) класів А600, А600С, А800A material head absorbs one code, never a list.
III(measurement) – IV(measurement) класів небезпеки(property)Two Roman codes at one head → property head, one measurement per code, dash outside.
класами впливу середовища X0, XC1, ХС2/ХС3One span across a comma list destroys IoU.
ambiguity-11
Concrete cover, ducts, sheaths and zones
"Захисний шар бетону" (cmin, cnom) is a named normed design characteristic of the same family as "робоча висота перерізу" — one property span with the dependent "бетону" inside it. The adjectives мінімальний / номінальний belong to the canonical term and stay inside; quantity wrappers stay outside (boundaries-9). A bare "шар" without "захисний" is not annotated, and a term truncated to its modifier ("мінімального захисного" with no "шару") is not annotated at all. A "канал" in the concrete is a void, not an element → not annotated; the "оболонка" / "футляр" of a tendon is a supplied product that stays in the member → material. Named areas — "зона анкерування", "зона передачі зусиль", "стиснута зона", "район будівництва" — are not annotated, and nothing inside them is annotated either: "анкерування" gets no process and "зусиль" gets no load_effect, because an inner word of a canonical area name never becomes a span of its own (ambiguity-21 c). This reading is retroactive over the whole cover chapter: one term, one label, everywhere.
захисний шар бетонуNamed normed characteristic → one property span.
Мінімальна величина захисного шару бетону cmin,durWrapper outside, symbol unlabeled (scope-4).
оболонка пучкаSupplied sheath that stays in the member → material.
зона анкеруванняA named area — annotate nothing, "анкерування" included.
ambiguity-12
Coefficients — action-side and statistical
Two more coefficient families were escalated and both fall on the unannotated side, which closes ambiguity-7: if a coefficient is not in its nature-fixed list, it is NOT annotated. (a) Coefficients of the ACTION itself — асиметрії циклу напружень, динамічності, тривалості дії, сполучень — describe how an action enters the calculation; they are design multipliers and not actions, so the ninth label does not reach them: a coefficient is never a load_effect. The action word inside such a term stays inside it and gets no span of its own. (b) STATISTICAL coefficients of test data — варіації, однорідності, кореляції, забезпеченості, стандартне відхилення — are fixed neither by the nature of the material nor by a table in the norm but by how a test batch was processed: they describe a method, not an object. Dependent names beside a coefficient keep their own spans, exactly as they do beside a norm multiplier. A coefficient quoted with its value is still not a span, and the bare value is not one either (numeric-1).
коефіцієнт варіації міцності(property) бетону(material)The coefficient gets nothing; the dependent names keep their spans.
коефіцієнт асиметрії циклу напруженьDescribes an action but is not one — annotate nothing, "напружень" included.
коефіцієнт надійності γc = 1,3Bare value with no unit — no measurement either.
ambiguity-13
Design modifiers of a size, intensive material quantities
Two extensions of ambiguity-6. (a) A plain size becomes a property when it carries one of a CLOSED list of design modifiers: розрахунковий, робочий, приведений, еквівалентний, умовний, ефективний. The same nouns with a descriptive or locative modifier ("повна висота", "загальна довжина", "внутрішній діаметр") stay unannotated, as does the bare noun. (b) For a non-geometric quantity ask whether the value would change if you took twice as much of the same substance. If it would not, the quantity is intensive and names a material property → property: густина, питома вага, вологість, водопоглинання, пористість, теплопровідність, морозостійкість, водонепроникність. If it would, the quantity is extensive and belongs to a particular item → not annotated without a number: маса, об’єм, вага, кількість, витрата. "Строк служби" / "строк експлуатації" remains property (entities-property); the escalations proposing to drop it are rejected. The closed list of (a) was extended in v0.6 — корисний, зведений, основний — and the treatment of genitive qualifiers and carrier nouns is in ambiguity-22.
еквівалентний діаметрDesign modifier from the closed list → property.
внутрішній діаметрDescriptive modifier does not promote a plain size.
масаExtensive quantity with no number → annotate nothing.
ambiguity-14
Nullifying heads, fatigue, prestress compression
The "втрата X" frame of ambiguity-5 is a CLOSED list of nullifying heads: втрата / втрати, вичерпання, порушення, руйнування, досягнення, перевищення. Since v0.5 the head no longer produces a hole: the WHOLE turn "<nullifier> + <property or state>" is ONE load_effect span, and the property inside still gets no span of its own — "[втрата стійкості]", "[вичерпання несучої здатності]", "[порушення статичної рівноваги]", "[перевищення граничних деформацій]", "[втрати попереднього напруження]". A loss of a resistance is an action-side event, so labelling the frame as a whole is what makes ambiguity-3 self-consistent; a partial span inside it is the same κ-killing asymmetry it always was. What the nullifier suppresses is unchanged: PROPERTY and STATE names inside the turn. A MATERIAL or a STRUCTURE inside it keeps its own span and CUTS the load_effect span at that point, and the trailing qualifier after the cut stays unannotated exactly as in boundaries-5: "[руйнування](load_effect) [бетону](material)", "[досягнення](load_effect) [бетоном](material) проектної міцності". Outside the list the ordinary property rule applies ("забезпечення [несучої здатності](property)"). "Втома" is a named behaviour of the material, the same family as повзучість, усадка and релаксація → property; "[опір втомі]" and "[міцність на втому]" are single undivided property spans, while "[напруження втоми]" is load_effect and "[втомне руйнування]" is one load_effect span under the nullifier list. "[Обтиснення]" is a stress state produced by prestressing and performed by nobody as an operation of its own → load_effect; the operation is "[попереднє напруження](process)". PRECONDITION OF THE CONSTRUCTION: the complement must name a PROPERTY or a MECHANICAL STATE. If it is a norm, a requirement, an indicator or an environmental parameter, the construction does not fire and the phrase stays empty as a whole: «перевищення значень прийнятних концентрацій забруднюючих речовин», «перевищення гранично допустимих викидів», «порушення вимог». This is the same scope that ambiguity-20 and ambiguity-32 set for the whole label. «[Перевищення граничних деформацій](load_effect)» remains a valid example precisely because its complement is a mechanical state.
опір втомі(property)Named behaviour → one undivided property span.
вичерпання несучої здатності(load_effect)Nullifying head → one span over the whole turn.
руйнування(load_effect) бетону(material)The material keeps its span and cuts the turn.
обтиснення(load_effect)A produced stress state, not an operation.
вичерпання несучої здатності(property)No inner property span inside a nullifier turn.
ambiguity-15
Methods, models, diagrams, computational objects
The calculation toolkit is not annotated, extending ambiguity-9: named methods (метод скінченних елементів / МСЕ, метод теорії пружності, деформаційний метод), models (блочна модель, деформаційна модель, розрахункова модель перерізу), diagrams (діаграма стану / деформування, діаграма "напруження-деформації", білінійна та параболічно-прямокутна діаграма), criteria (критерій утворення тріщин) and the computational objects of FEM (матриця жорсткості, матриця мас, вектор навантажень). A method is not a tool — tool is physical equipment — and it is not a process either, because process is the action and not the means of performing it; "[розрахунок]" and "[моделювання]" stay process. Nothing is annotated inside such a name when the inner word belongs to the canonical term: "матриця жорсткості" yields no property, "метод скінченних елементів" no structure, "вектор навантажень" no load_effect, and the diagram name «напруження-деформації» none either. A dependent MATERIAL naming what the model is about does keep its span.
діаграма стану бетону(material)The toolkit name gets nothing; the material keeps its span.
матриця жорсткостіInner word of a canonical computational term — annotate nothing.
метод скінченних елементівNot a tool, not a process — annotate nothing.
моделюванняThe action itself stays process (ambiguity-8).
ambiguity-16
Substances outside the member: agents, ground, subgrade
Ask whether the substance goes INTO the structure or acts on it from outside. Into it → material: цемент, заповнювач, протиморозна добавка, and natural ground as a substance with normed characteristics (ґрунт, скельна порода, насипний ґрунт). Acting on it from outside → not annotated: хлориди, хлорид-іони, сульфати, солі, кислоти, антиобморожувачі, вуглекислий газ. These are SUBSTANCES, not forces, so the load_effect of ambiguity-3 does not reach them either — their head is not on the action list of ambiguity-20, and the aggression they cause is described by the environment-class scale, which is property or measurement by ambiguity-10. Named environment parameters stay as ambiguity-9 left them: no span without a value; the value with its unit is a measurement. "Основа", "ґрунтова основа", "підошва" are not annotated — a natural mass is not a built element — while "[ґрунт]" inside is material and "[паля]", "[фундамент]" are structure. Hyphenated system names ("основа-споруда", "основа-фундаменти-споруда", "ґрунт-конструкція") get no span, and nothing inside them is annotated.
ґрунтA substance with normed characteristics → material.
протиморозна добавкаGoes into the mix → material.
хлоридиActs on the member from outside — annotate nothing.
основа-фундаменти-спорудаSystem name — no span, nothing inside it either.
ambiguity-17
Heads that are not elements: joints, generic nouns, locative modifiers
A named joint is a physical detail of the structure and takes structure with its typological modifier inside: "[зварне з’єднання]", "[стикове з’єднання]", "[з’єднання внапуск]", "[вузол спирання]". A bare "з’єднання" / "стик" / "вузол" with no modifier is not annotated, and the action beside it keeps its own label. A locative or exposure modifier is NOT a typological one: "надвірні / внутрішні / підземні / зовнішні конструкції" stay unannotated exactly like the bare "конструкція"; only a constructive-typological modifier (згинальні, стержневі, плитні, каркасні) or the material adjective of boundaries-7 produces a span. The bare hypernym "матеріал" / "вид матеріалу" joins the bare "будівля / споруда / конструкція / об’єкт": material labels named substances and products, not the word for them. The same principle now covers the action side. A bare "вплив" / "впливи" / "дія" / "стан" with no qualifier, and the function frames built on them ("під дією", "з урахуванням впливу", "залежно від виду впливу"), get no load_effect span; a qualifier naming the action type does produce one ("[сейсмічний вплив]", "[температурні впливи]"). "Навантаження", "зусилля" and "напруження" are NOT hypernyms — they name the action itself and are annotated bare (ambiguity-20). DOMAIN QUALIFIER AGAINST TYPOLOGICAL — the contrast test. A typological qualifier is contrasted with another qualifier in the same classification and gives a span: «[огороджувальні конструкції]» against несучі, «[каркасні]» against стінові. A domain qualifier is contrasted with nothing and gives no span: «будівельні конструкції», «будівельні матеріали», «будівельні вироби», «будівельна продукція», «інженерні системи». Legal definition of a term adds no label: «основна вимога» is defined by statute too and takes no span. «Продукція» and «виріб» join the bare hyperonyms «матеріал» / «вид матеріалу». The test is the same as in ambiguity-39. SCOPE OF THE CONTRAST TEST. The test shows only that the qualifier INDIVIDUATES the head, and it is a necessary, not a sufficient condition; each article adds its own sufficient condition: structure needs a TYPOLOGICAL or FUNCTIONAL qualifier, material a kind by composition or function, safety the name of a DOMAIN of protection. The CASE of the qualifier decides nothing — agreeing or genitive is irrelevant; its ROLE decides. A qualifier of KIND, TYPE or FUNCTION raises the head: «[систем водопостачання](structure)», «[захисні матеріали](material)». A qualifier naming a DOMAIN, the OBJECT of the action, a PURPOSE or a FIELD never raises it: «будівельні конструкції», «інженерні системи», «безпеки здоров’я і життя людини» — the last on the same ground as «протикорозійний захист конструкції» in ambiguity-18: it names WHOM one protects, not which domain. A function-word wrapper is not a qualifier at all (ambiguity-30 c): «засоби забезпечення вентиляції» stays a bare «засоби».
зварне з’єднанняNamed joint with a typological modifier → structure.
зварювання(process) накладок(material)The action and the products keep their own labels.
під дією постійного навантаження(load_effect)The frame stays outside; the named action is the span.
надвірних конструкційLocative modifier is not typological — annotate nothing.
вид матеріалуBare hypernym — annotate nothing.
під дієюBare hypernym of the action side — annotate nothing.
ambiguity-18
Requirement, product or action — and the limits of safety
One term, three readings, resolved by what the sentence makes it name. A REQUIREMENT or a capability → property: теплоізоляція, звукоізоляція, довговічність, зчеплення. A PRODUCT or a layer that stays in the member → material: "укладання [теплоізоляції] завтовшки 100 мм", "[захисне покриття]". An ACTION that is performed → process: "[гідроізоляція]" as the operation, "[катодний захист]" as the applied technology, "[анкерування]" as the operation. Named characteristics of the CONTACT between two materials are property even though they belong to neither alone — зчеплення, адгезія, тертя по контакту — while the phenomena they produce (проковзування, взаємні зміщення) are not annotated. safety keeps the scope of entities-safety: protection of PEOPLE, plus a named field of safety with a qualifier. Corrosion protection of a structure is never safety.
зчеплення арматури з бетономA named characteristic of the contact → property.
укладання теплоізоляції(material) завтовшки 100 ммProduct or layer → material.
катодний захистApplied technology → process.
захисні покриття(safety)Protection of a structure is material or process, never safety.
ambiguity-19
process and tool — control, research, verb forms, rigs
Three closures on ambiguity-8. (a) Organizational and control actions do not change the material or the member and are NOT process: контроль якості, вхідний контроль, приймання, технічний нагляд, сертифікація, обстеження, паспортизація; entities beside them keep their spans. (b) Research actions are not process either — the closed list keeps only "[випробування]", so "дослідження" and "експериментальні дослідження" get nothing. (c) process is placed ONLY on a deverbal noun: finite verbs ("проектують", "встановлюють", "зварюють"), adverbial participles ("Проектуючи") and participles ("запроектований") never receive a span. This hardens the preference stated in entities-process into a rule and removes the largest morphological source of disagreement. (d) Removable technological rigs of the tensioning operation are tool (домкрат, напружувальний стенд, упори, форма), while anything supplied and left in the member forever is material however device-like its name (анкерний пристрій, муфта, пристрій для відхилення).
контроль якості виготовлення(process) бетону(material)The control action gets nothing; the technological action does.
контроль якостіOrganizational action — annotate nothing.
конструкції проектують за граничними станамиFinite verb — never a process span.
проектування конструкційDeverbal noun of the closed design list → process.
ambiguity-20
Where a load_effect span starts and ends
load_effect is decided by the HEAD of the phrase, taken from a CLOSED list: навантаження, вплив, дія, сила, зусилля, момент (силовий), тиск, розпір, напруження, деформація, прогин, переміщення, осідання, згин, стиск, розтяг, кручення, зріз, обтиснення, стан (граничний, напружено-деформований), рівновага. If the head is not on this list the phrase is not a load_effect however mechanical it sounds — that is what keeps the new label from swallowing half the document. Six boundary rules. (1) The span is the canonical term, modifier + head: "[згинальний момент]", "[поперечна сила]", "[постійне навантаження]", "[розрахункове навантаження]". Quantity and statistical wrappers stay OUTSIDE, exactly as in boundaries-9 and ambiguity-7: "величина", "значення", "рівень", "характеристичне значення" — "характеристичне значення [навантаження]". (2) The SYMBOL stays outside the span and unlabeled (scope-4): "[згинальний момент](load_effect) MEd", "[поперечна сила](load_effect) VEd"; a bare symbol on its own never becomes a load_effect. (3) A qualifier naming the KIND of action is INSIDE the span: "[напруження стиску]", "[дотичні напруження зрізу]", "[деформації усадки]", "[двовісний згин]". (4) A MATERIAL or a STRUCTURE inside the phrase keeps its own span and CUTS the load_effect span, by the procedure of boundaries-5 — never absorb a material into an action span: "[напруження стиску](load_effect) у [бетоні](material)", "[зусилля](load_effect) в [арматурі](material)", "[прогин](load_effect) [балки](structure)". (5) A value with its unit is a separate adjacent measurement (numeric-1): "[навантаження](load_effect) [5 кН/м](measurement)". (6) A bare hypernym is not annotated (ambiguity-17): "вплив", "дія", "стан" with no qualifier, and the frames "під дією", "з урахуванням впливу", get nothing — while "навантаження", "зусилля" and "напруження" name the action itself and are annotated even bare. SCOPE OF THE CLOSED LIST. The list of heads applies only where the phrase names an action ON A STRUCTURE or the mechanical state it causes (entities-load-effect). The same head in another role gives no label: the pressure of a working medium in a service system, and atmospheric pressure as a microclimate factor, are parameters and are decided by ambiguity-9 and ambiguity-34 F1-F3 — «падіння тиску», «перепадів тиску», «атмосферного тиску» take no span, while «[тиск ґрунту на підпірну стіну](load_effect)» keeps its own. A quantity derived from a parameter — «падіння», «перепад», «зниження», «приріст» — is neither a head from this list, nor a magnitude wrapper from (1), nor a nullifier from ambiguity-14: the phrase stays empty as a whole, and a partial label on the dependent noun inside it is forbidden (ambiguity-21 c).
згинальний момент(load_effect) MEdCanonical term is the span; the symbol stays outside.
напруження стиску(load_effect) у бетоні(material)The material cuts the span and keeps its own.
характеристичне значення снігового навантаження(load_effect)The statistical wrapper stays outside.
напруження стиску в бетоніNever absorb a material into an action span.
MEdA bare symbol is notation, never a load_effect (scope-4).
ambiguity-21
load_effect in fixed turns: nullifiers, limit states, canonical interiors
Four closures that keep the ninth label from reopening settled rules. (a) NULLIFYING HEADS. The closed list of ambiguity-14 — втрата / втрати, вичерпання, порушення, руйнування, досягнення, перевищення — no longer leaves a hole: the WHOLE turn is ONE load_effect span and the property inside gets no span of its own ("[втрата стійкості]", "[вичерпання несучої здатності]", "[втрати попереднього напруження]", "[втомне руйнування]"). A material or a structure inside still keeps its own span and cuts the turn ("[руйнування](load_effect) [бетону](material)"). (b) LIMIT STATES. The full designation is ONE load_effect span with nothing annotated inside it: "[граничний стан за несучою здатністю]" (ULS), "[граничні стани за придатністю до нормальної експлуатації]" (SLS), "[перший граничний стан]", "[статична рівновага]" (EQU). "Несучою здатністю" inside the ULS turn does NOT take property: spans may not overlap and the designation is one canonical name of a design situation. (c) CANONICAL INTERIORS are untouched — an action word inside the canonical name of something else still gets nothing: "[класами впливу середовища](property)" (ambiguity-10), "вектор навантажень" and «напруження-деформації» (ambiguity-15), "зона передачі зусиль" (ambiguity-11), "коефіцієнт асиметрії циклу напружень" and "коефіцієнт сполучень" (ambiguity-7, ambiguity-12), and the trailing qualifier of a property term, where "на стиск" in "[міцності](property) [бетону](material) на стиск" stays unannotated (boundaries-5). (d) OUT OF SCOPE. load_effect covers actions, internal forces and stress-strain states — not physical DEFECTS or their geometry: "тріщина", "розкриття тріщин", "ширина розкриття тріщин" keep no span, because their head is not on the ambiguity-20 list, while "[тріщиностійкість]" stays property. A coefficient is never a load_effect whatever it describes.
втрата стійкості(load_effect)The whole turn, one span, nothing inside.
розрахунок(process) за граничними станами(load_effect)Design action + design situation.
класами впливу середовища(property): X0 – 20"впливу" inside a scale name gets no load_effect (ambiguity-10).
граничний стан за несучою здатністюNever split the ULS designation into two spans.
розкриття тріщин(load_effect)A defect is not an action — annotate nothing.
ambiguity-22
Sizes: the extended modifier list, a genitive qualifier, the carrier noun
Four closures on ambiguity-6 and ambiguity-13. (a) The CLOSED list of promoting design modifiers gains three synonyms that name the same design quantities and were escalated once per batch: корисний ≡ робочий ("[корисна висота]" is "[робоча висота]"), зведений ≡ приведений, and основний for the normed basic lengths ("[основна довжина зони анкерування]", lb,rqd). The full list is now розрахунковий, робочий, корисний, приведений, зведений, еквівалентний, умовний, ефективний, основний. (b) A GENITIVE qualifier does NOT promote a plain size on its own: "довжина обпирання", "довжина анкерування", "радіус кривизни", "крок каналів", "площа поперечного перерізу", "висота перерізу", "товщина плити", "проліт", "ексцентриситет" get no span without a number, exactly like the bare noun. Only a modifier from the closed list, or a canonical name already registered in these guidelines ("момент інерції", "захисний шар бетону", "робоча висота перерізу"), turns a size into a property. (c) A named computational CROSS-SECTION is not a size and stays unannotated together with the bare "переріз": "переріз брутто", "переріз нетто", "зведений переріз". (d) The CARRIER noun in the genitive stays OUTSIDE the property span — "[робоча ширина](property) полиці" — because a part of a section is not an entity (ambiguity-6) and spans may not overlap. When the property name is uninterrupted, the trailing qualifier stays INSIDE it as boundaries-5 already requires ("[опору на розтяг]", "[опору на зріз при продавлюванні]", "[класами бетону на стиск]"); it is only a MATERIAL cutting the span that leaves the tail outside ("[міцності](property) [бетону](material) на стиск"). THE property SPAN BOUNDARY IN A PREPOSITIONAL PHRASE. The span is continuous and runs to the end of the canonical name; it is cut ONLY by a word that itself takes a label, and an unlabelled word stays inside. Prepositional groups are separated by the same division as in boundaries-8: «до / від / проти <factor>» names WHAT THE PROPERTY IS NORMED AGAINST and falls INSIDE; «при / для / під час <circumstance>» is a purpose or circumstance group, stays OUTSIDE, and a word in it keeps its own label. All four patterns occur in one clause of ДБН В.1.2-8:2021 and were separated identically by all three annotators: «[герметичність](property) [контейнерів](tool) та [трубопроводів](structure)»; «[ефективність](property) [систем очистки та фільтрації](structure)»; «[стійкість до засобів очищення](property)»; «[захищеність приміщень від накопичення радону](property)». A bare head «здатність» with a complement is a named capability under ambiguity-5 and gives ONE property span.
корисна висота(property)Closed-list synonym of "робоча" → property.
радіус кривизниGenitive qualifier does not promote a plain size.
переріз бруттоA computational cross-section is not annotated.
ambiguity-23
Technological operations: an open site list against a closed design list
ambiguity-8 gives two lists and only one of them is closed — the single most escalated question of the fourth round (17 escalations from all three annotators). The DESIGN list is CLOSED: розрахунок, проектування, конструювання, моделювання, випробування. Nothing joins it, so аналіз, визначення, перевірка, обчислення, вимірювання, урахування, забезпечення, оцінювання stay unannotated, and the organizational actions of ambiguity-19 stay out too. The SITE list is OPEN: any deverbal noun naming an operation somebody performs on the material or the member is process whether or not it is printed in these guidelines — догляд (за бетоном), укладання, ущільнення, закладання, замонолічування, звільнення (з упорів), перепуск, транспортування, зберігання, ремонт, технічне обслуговування. The test is unchanged: somebody performs it AND it changes the material or the member; a life-cycle STAGE ("експлуатація") is still not an action. "Розрахунок" is process wherever it heads a phrase, including "при розрахунку", "методи розрахунку", "основи розрахунку"; "розрахункове значення" is a quantity wrapper and stays outside the span it introduces (ambiguity-7), which is why the adjective never carries a process span.
догляд(process) за бетоном(material)Site operation outside the printed list → process.
при розрахунку(process) збірних елементів(structure)
розрахункове значення поперечної сили(load_effect)Wrapper outside; no process span on the adjective.
перевіркаAnalytical action outside the closed design list — annotate nothing.
ambiguity-24
Rebar connections: splice, connector, zone
The chapter on splices mixes three readings and each keeps its own label. (a) The named SPLICE is a physical detail of the member and takes structure, as the named joint of ambiguity-17 does: "[напуск]", "[з’єднання внапуск]", "[з’єднання обтиском]", "[механічне сполучення]", "[зварне з’єднання]". A bare "з’єднання" / "стик" / "вузол" with no modifier still takes nothing, and the operation that produces the splice keeps its own process span when it is named separately ("[зварювання](process) [стрижнів](material)"). (b) A supplied CONNECTOR that stays in the member forever is material and never tool: муфта (гвинтова, заповнена, обтискна), анкер, анкерний пристрій, механічний пристрій, пристрій для відхилення, закладна деталь. The word "пристрій" does not pull it towards tool; tool stays with the removable rig — домкрат, напружувальний стенд, форма, опалубка, упори (entities-tool, ambiguity-19 d). (c) "В’язі" splits by referent: transverse links inside a member are a reinforcement product → material, while the tie SYSTEM against progressive collapse (горизонтальні та вертикальні в’язі, в’язева система) is a structural system → structure.
напуску(structure) стрижнів(material)
обтискні муфти(material)Connector left in the member forever → material.
в’язева система(structure)Ties against progressive collapse → structure.
ambiguity-25
A process noun inside an action head
ambiguity-20 (4) lets a MATERIAL or a STRUCTURE cut a load_effect span; a PROCESS noun does not. "Зусилля попереднього напруження", "сила попереднього напруження", "момент прикладення попереднього напруження" and "втрати попереднього напруження" are canonical names of an action, so each is ONE load_effect span with nothing annotated inside it — the treatment ambiguity-21 (a) already fixed for the nullifier turn, extended to the force heads. "Попереднє напруження" keeps its process span only where it heads the phrase in its own right ("[попереднє напруження](process) виконують домкратом", "втрати від [попереднього напруження]" excepted as a nullifier turn). The reason is agreement, not metaphysics: when half the annotators cut the phrase at the operation and half do not, the merged entry depends on which half was sampled.
зусилля попереднього напруження(load_effect)One span; no inner process.
втрати попереднього напруження(load_effect)
попереднє напруження(process) пучків(material)The operation heads the phrase → process.
зусилля попереднього напруження(process)Never cut an action head at a process noun.
ambiguity-26
A limit expressed as a fraction of a size
A relative limit — "1/150 прогону", "1/75 вильоту консолі", "1/400 довжини елемента", "1/30 висоти перерізу" — is a dimensionless ratio, not a measurement: it has no unit, and numeric-1 and numeric-5 both refuse to let a reference noun act as one. The fraction gets NO span. The reference noun is labeled by its own rule and usually gets nothing (a plain size, ambiguity-22), while an ELEMENT named inside it keeps its own span. Never absorb that element into a measurement span. This leaves untouched a percentage that is an inseparable part of a canonical term ("[0,1 % умовної границі текучості]", boundaries-9) and any value with a real unit ("[30 мм]").
не повинен перевищувати 1/150 прогонуNo span — dimensionless ratio.
1/75 вильоту консолі(structure)The element keeps its own span.
1/150 прогону(measurement)A reference size is not a unit.
ambiguity-27
Age, pore structure, defects
Three small families this document repeats. (a) "Вік" (вік бетону, вік у момент прикладення навантаження) is a point on the time axis, not something the material HAS: no span, while the material beside it keeps one. The pair to memorize is "[строк служби](property)" — a normed durability requirement — against "вік [бетону](material)". (b) "Замкнута структура" / "відкрита структура" of a concrete names a characteristic of its pore system, the same family as пористість and водопоглинання → property; it is not absorbed into the material span: "[бетони](material) [замкнутої структури](property)". (c) DEFECTS and their geometry stay unannotated, extending ambiguity-21 (d) beyond cracks: тріщина, розкриття тріщин, ширина розкриття тріщин, сколювання, відшарування, корозія, знос. A defect is neither an action nor a resistance; the ability not to develop it is the property ("[тріщиностійкість]", "[корозійна стійкість]"). «Життєвий цикл» joins the time-axis family in (a): «життєвий цикл», «стадія життєвого циклу», «протягом усього життєвого циклу» take no span — a stretch of time and a set of stages, not something the object HAS. The contrast stands beside it: «[строк служби](property)» IS property, because it norms durability, while a life cycle norms nothing.
вік бетону(material)Time point — no span on "вік".
бетони(material) замкнутої структури(property)
ширина розкриття тріщинDefect geometry — annotate nothing.
Members of a calculation model: strut-and-tie, schemes, systems
ambiguity-15 covers the calculation toolkit; this covers the fictitious members inside it. A member that exists only in the model gets NO span: "стиснуто-розтягнута модель", "умовні стиснуті елементи", "умовно розтягнуті елементи", "розкоси і затяжки моделі", "розрахункова схема", "конструктивна система". The word "умовний" is the signal — it marks a member the analysis assumes rather than one that is built. A real region of the member keeps its own label under the ordinary rules (a named joint by ambiguity-17, a named zone by ambiguity-11), and a material inside the phrase keeps its span. "Конструктивна система" joins the bare "будівля / споруда / конструкція" of entities-structure: a system is a way of describing a structure, not a structure.
умовних стиснутих елементівMember of the model — annotate nothing.
конструктивна системаA system, not a built object.
стиснута зонаA named zone stays unannotated, and so does everything inside it (ambiguity-11).
залізобетонних елементів(structure)A real member keeps its span.
ambiguity-29
process in an administrative norm: the two-step test
ambiguity-8 and ambiguity-23 were written for a design norm and give no answer in an administrative-planning one, where deverbal nouns of ORGANIZATIONAL action outnumber technological ones and process reaches half of all spans. Two questions, asked in order, decide every case and replace ad-hoc judgement. STEP 0: a process span sits only on a deverbal noun (ambiguity-19 c). STEP 1 — CAN IT BE PERFORMED? Put the noun into the frame "виконують / здійснюють <X> (чого?)". If Ukrainian can only attach the domain object through a preposition — "поводження З відходами", "робота З населенням", "операції З відходами", "діяльність У СФЕРІ …" — or the frame is not idiomatic at all, the noun names a FIELD or an ORDER of activity and not an action: no span, while the dependent material keeps its own. If the frame works with a bare genitive object, go to step 2. STEP 2 — WHAT BECOMES DIFFERENT? Name the ONE thing that is physically different once the action is over. (a) A physical thing of the domain — waste, soil, water, a territory, a street, a yard, plantings, a building, a landfill, a container, a machine → process. (b) A DOCUMENT or its legal content (схема, проєкт, реєстр, ліміт, тариф, паспорт, норматив, план, рішення, договір), or money, legal status, staffing, interaction → NOT process: this extends the organizational exclusion of ambiguity-19 (a) instead of opening the closed design list of ambiguity-23. (c) Nothing, because nobody performs it (утворення, надходження, забруднення) → not process. Under this test process covers санітарне очищення, прибирання, збирання, зберігання, перевезення, вивезення, перероблення, оброблення, сортування, утилізація, знешкодження, захоронення, видалення, розміщення (відходів), благоустрій, утримання, ремонт, рекультивація, миття, дезінфекція, поливання, підмітання, проектування, будівництво, реконструкція, розширення — and does NOT cover поводження, розроблення, складання, ведення, затвердження, погодження, перегляд, формування (тарифів), фінансування, планування, нормування, організація, впровадження, забезпечення, обґрунтування, визначення, утворення, експлуатація. Step 2 works by REFERENT, not by the presence of a word in the text: «водопостачання» with no complement gives process, because it is water that changes (material under ambiguity-38); «освітлення» gives nothing, because only a condition of the environment changes. Adjacent members of one list may therefore take different labels — an expected result, not an inconsistency.
благоустрою(process) населених пунктівIt is performed and the territory ends up different → process, in every syntactic position.
у сфері поводження(process) з побутовими відходами(material)"виконують поводження" does not exist — a field, not an action; the material keeps its span.
Розроблення(process) схемиOnly a document changes → not process, like затвердження and погодження.
утримання(process) в належному стані територіїPhysical upkeep of an object joins ремонт and технічне обслуговування on the open site list (ambiguity-23).
ліміти на утворення(process) відходівNobody performs it → not process.
ambiguity-30
Umbrella nouns, activity hypernyms, and where a process span stops
Three closures that keep ambiguity-29 from drifting. (a) SPAN BOUNDARY. The process span is the deverbal noun plus an adjective that belongs to the canonical name of the work ("[санітарне очищення]"); the DEPENDENT object noun and any prepositional tail stay OUTSIDE and take their own label — "[збирання](process) [побутових відходів](material)", "[санітарне очищення](process) територій населених місць", "[утримання](process) в належному стані території". An object noun is never absorbed into a process span, so "[поводження з побутовими відходами]" is wrong twice over: the head takes no span at all and the material would have been swallowed. A classifying adjective belonging to another head stays outside: "планово-регулярна система [санітарного очищення](process)". (b) ACTIVITY HYPERNYMS extend ambiguity-17 and take no span: захід, робота, послуга, діяльність, порядок, система, метод, етап, вид, and "об’єкт" — which stays a bare head whose dependent keeps its own label, "об’єкти [благоустрою](process)", "об’єкти поводження з [відходами](material)". (c) LIGHT-VERB WRAPPERS take no span and stay outside the span they introduce, exactly as the quantity wrappers of boundaries-9 do: здійснення, проведення, виконання, надання, запровадження — "для здійснення робіт із [прибирання](process) об’єктів [благоустрою](process)". Two adjacent process spans are the expected result and are not merged.
збирання(process) побутових відходів(material)Object outside the action span, with its own label.
поводження з побутовими відходами(process)The umbrella head takes nothing and never absorbs the material.
об’єкти благоустрою(process)The bare head "об’єкт" gets nothing; the dependent keeps its label.
здійснення(process) робітLight-verb wrapper plus an activity hypernym — annotate nothing here.
ambiguity-31
Economic and planning quantities
A planning norm counts money and volumes on every page, and the split follows ambiguity-13 (b) without a new principle. A NAMED cost characteristic is property: вартість, кошторисна вартість, орієнтовна вартість, відновна вартість — the object has it, an entire ДБН (Д.1.1-1) exists to determine it, and a value beside it is a separate adjacent measurement. A PLAIN amount is extensive and takes no span without a number, exactly like the "об’єм / кількість / витрата" already on the ambiguity-13 (b) list: обсяг робіт, обсяг фінансування, кількість, капіталовкладення, витрати, експлуатаційні витрати, тариф. A SCHEDULE or ordering matter is organizational (ambiguity-19 a) and takes no span either: черговість, терміни, строки виконання, графік. "Строк служби" is untouched and stays property (entities-property): it is a normed durability requirement, not a schedule.
вартості(property) будівництва(process)Named cost characteristic → property; the action keeps its own span.
обсягів робіт(property) із санітарного очищенняPlain amount — annotate nothing; "[санітарного очищення](process)" is the span in this phrase.
черговості(property) здійснення заходівOrdering of works — organizational, no span.
капіталовкладення(property)A sum of money is an extensive amount (ambiguity-13 b).
ambiguity-32
Ecological and sanitary impact is not load_effect
load_effect is the E of the Eurocode inequality E ≤ R: it labels what acts ON A STRUCTURE and the mechanical state that action produces (entities-load-effect). An impact on the ENVIRONMENT, on health, on the population or on an activity is not an action on a structure, its head is the bare hypernym "вплив" of ambiguity-17, and it takes NO span: "вплив на навколишнє середовище", "впливу шкідливих факторів на довкілля та здоров’я людини", "зони впливу та поширення шкідливих викидів", "оцінка впливу на навколишнє природне середовище (ОВНС)". The turn is not re-labeled as anything else either — it is a section topic, not an entity. Inside such a turn the ambiguity-16 treatment holds: emissions, pollutants and hazardous substances acting from outside take no span, while a named MATERIAL keeps its own ("[небезпечні відходи](material)"). This is a clarification of scope, not a reversal: nothing that was load_effect in a structural norm changes label.
впливу шкідливих факторів на довкілля(load_effect)Environmental impact — not an action on a structure, no span.
Розділ "Вплив на навколишнє середовище(load_effect)"A chapter topic, not an entity.
рівнів забруднення підземних водNo span: pollution of the environment is neither an action on a structure nor a property of one.
снігове навантаження(load_effect)Unchanged — a mechanical action on the structure.
ambiguity-33
Generic equipment heads are tool
entities-structure and ambiguity-17 refuse a bare "будівля / споруда / конструкція / об’єкт / матеріал", and the fourth round extended that to bare hypernyms of the action side. tool has NO such ladder: "устаткування" names equipment itself, not the word for a tool, and no narrower label competes for it. A generic equipment head is therefore annotated: [машини], [механізми], [устаткування], [обладнання], [інвентар], [технічні засоби], [транспортні засоби], [спецтехніка], [сміттєвози], [контейнери], [урни], [бункери]. A qualifier saying WHICH equipment is meant stays INSIDE the span by boundaries-1, participles included: "[спеціально обладнаних транспортних засобів]", "[поливально-мийних машин]", "[прибиральних машин]". A bare "засоби" with no qualifier is a hypernym of the label and takes nothing. Equipment that is a BUILT facility rather than removable gear stays structure by entities-structure (b): "[контейнерний майданчик]", "[полігон]", "[сміттєперевантажувальна станція]"; a device supplied and left in a member forever is still material (ambiguity-19 d). BOUNDARY. The article applies only under its own precondition — when no narrower label competes with the generic head. A NAMED supplied product installed permanently does have a narrower label and stays material under ambiguity-24 (b), however instrument-like the name sounds: клапан, кран, вентиль, засувка, фітинг, муфта, зворотний затвор. The generic head in the same sentence keeps tool: «[обладнання зворотного потоку](tool), [клапани](material), [крани](material)». THE BOUNDARY IS THREE-WAY, not binary. (1) A generic head of equipment → tool. (2) A named supplied product installed PERMANENTLY → material (ambiguity-24 b): клапан, кран, вентиль, засувка, фітинг, муфта. (3) A named REPLACEABLE element that does its work and is periodically changed → tool by the positive definition of entities-tool: фільтр, картридж, димопоглинач, лампа, контейнер. What decides is not being named, but whether the product stays in the structure permanently.
устаткування(tool) та інвентарю(tool)Generic equipment heads are annotated, unlike a bare "конструкція".
спеціально обладнаних транспортних засобів(tool)The participial qualifier says which vehicles are meant → inside the span.
контейнерний майданчик(structure)A built facility, not removable gear.
технічних засобів(tool)Do not split the qualifier off; the span is "[технічних засобів]".
ambiguity-34
property in a planning norm: four filters on a quantity
In a design norm property almost always names a normed characteristic of resistance; in a planning norm half of the candidates are simply figures to be collected into a table, and this is where the second document scored its lowest agreement (0.684, against 0.987 for process). No new principle is needed: four filters that already exist, applied IN ORDER, decide every case. F1 — A RAW DIMENSION (ambiguity-6, ambiguity-22 b): площа, довжина, ширина, висота, товщина, глибина, розмір, габарит, відстань, периметр take NO span without a number, and a descriptive or averaging modifier does not change that — «загальна площа», «середня відстань», «орієнтовні площі», «висота шару снігу» stay bare exactly as the bare head does; only the closed list of ambiguity-22 (a) promotes them. F2 — AN EXTENSIVE QUANTITY (ambiguity-13 b, ambiguity-31): об’єм, обсяг, маса, кількість, чисельність, витрата (including «витрати води»), тривалість, періодичність, кратність, and the SHARES of those quantities — питома вага, рівень охоплення, відсоток — take no span. «Об’єм контейнера» does not change this: the capacity of a vessel is named by МІСТКІСТЬ, while «об’єм» stays an extensive dimension both in running text and in a table row, because scope-5 grants tables no exception. F3 — AN ENVIRONMENTAL PARAMETER (ambiguity-9): середньорічна температура, кількість опадів, товщина снігового покриву, повторюваність снігопадів, напрямки пануючих вітрів, and a named weather-induced state of a surface — «зимова слизькість», «ожеледиця» — take no span; a value with a unit beside them is a measurement. F4 — A HYPERNYM OF THE MEASURAND (ambiguity-17): показник, параметр, характеристика, «технічні характеристики», дані, відомості, тип are words FOR a property, not a property; they take no span, and the dependent keeps its own label. WHAT SURVIVES THE FOUR FILTERS is property: (a) the CAPACITY an object is designed for — потужність, продуктивність, пропускна здатність, місткість, ємкість; (b) an INTENSIVE quantity (ambiguity-13 b) — щільність, густина, вологість, поверховість, інтенсивність руху, морфологічний / фізико-хімічний / фракційний склад, теплотехнічні властивості; (c) a NAMED COST characteristic (ambiguity-31) — вартість, кошторисна / орієнтовна / відновна вартість, собівартість, строк служби; (d) a SCALE NAME without a code (ambiguity-10) — «клас небезпеки», «ступінь благоустрою». The pair to remember comes from the same document: «[пропускна здатність](property) установок» is the throughput the plant is designed for and is property, while «обсяг робіт» and «площі ділянок» are measured figures and are nothing. EVALUATIVE VERDICT (extending F4). One test separates a characteristic from a word for one: can you name ONE indicator and ONE test method that establishes it? F4 suppresses an EVALUATIVE VERDICT — a phrase saying how GOOD the object is, which therefore presupposes a criterion of goodness: безпечність, якість, комфортність, біонейтральність, біопозитивність. F4 does NOT suppress the name of a physical FAMILY of characteristics, which merely names the field of measurement: «[теплотехнічні властивості]», «[оптичні властивості]», «[акустичні властивості]» — these are the "classes of characteristics" of entities-property and are property together with the adjective (boundaries-1). A single normed characteristic with its own indicator keeps its span all the more: «[герметичність]», «[водонепроникність]», «[енергоефективність]». F2 IS A PHYSICAL TEST, NOT A LEXICAL ONE: the adjective «сумарний / загальний / інтегральний» in the name of an intensive quantity does not make it extensive — the unit decides. Per unit mass, volume or area → intensive → property: «[сумарної активності природних радіонуклідів](property)» (Bq/kg), where «сумарна» means summed over nuclides, not over mass. The lexical caution about «загальний» stays in F1 and only for a bare dimension. F(a) IS ABOUT THE OBJECT ONLY, AND ABOUT THE DESIGN VALUE ONLY. The words "by design" act as a filter: the same quantity is property when it names a DESIGN capacity («[пропускна здатність](property) установок»), and takes no span when it names a MEASURED actual value — then F2 and F3 decide it: «фактичної інтенсивності вентиляції», «швидкості обміну повітря». The list of capacities concerns the built object. The same lexeme about a HUMAN state takes no span: продуктивність праці, працездатність, самопочуття (entities-safety, ambiguity-4). ORDER: FRAME FIRST, THEN FILTERS. When the sentence declares a phrase to be a CHARACTERISTIC of a named product, structure or system, the frame does NOT cancel the filters — it fixes WHOSE characteristic the quantity names, after which the filters apply to the PRODUCT rather than to the world. The consequences differ per filter. F1 is NEVER lifted by the frame: a bare dimension does not become the name of a characteristic just because it was declared one — «Характеристика – товщина» stays empty without a qualifier from the closed list of ambiguity-22 (a). F3 IS lifted, as already recorded in the ORDER OF CHECKING in ambiguity-9. F2 is lifted ONLY for a quantity naming the DESIGN CAPACITY of the product in the sense of (a), not an accounted volume: «[витрати повітря](property)», «[швидкості повітряних потоків](property)», «[перепадів тиску](property)» in a list of ventilation-system characteristics are normed indicators with their own method (m³/h, m/s, Pa); «обсяг робіт», «кількість відходів», «витрати води» of a settlement remain accounted volumes and take no span. The frame is NOT a general promoter — it changes the question, it does not cancel a filter; extending it to F1 would make «Характеристика – загальна площа» property, which is exactly what F1 exists to prevent. A coordinate list of declared characteristics gives a SEPARATE span to each member that is itself the name of a property (boundaries-9).
потужність(property) полігона(structure)The capacity the object is designed for → property.
місткість(property)The nameplate capacity of a vessel or a depot → property, unlike «об’єм».
площі ділянокF1: a raw dimension without a number takes nothing.
об’ємом(property) контейнерівF2: an extensive quantity; a table row is not an exception (scope-5).
технічні характеристики(property)F4: a hypernym of the measurand.
ambiguity-35
Place heads in a municipal norm: specific name, functional slot, absolute hypernym
entities-structure splits heads in two — a name from the list, and a bare «будівля / споруда / конструкція / об’єкт». A municipal norm has many times more place heads, and the split has to be made on THREE levels; without it structure lands at 0.848. LEVEL 1 — A LEXICALLY SPECIFIC HEAD: the noun by itself names ONE kind of built object, and it can be pointed at on a plan without asking «which one?». It is structure even bare — полігон, звалище, сміттєзвалище, снігозвалище, піскобаза, база, гараж, депо, ТЕЦ, котельня, станція, вулиця, дорога, проїзд, під’їзд, тротуар, майдан, площа (as a square), міст, тунель, колодязь, лоток, дощоприймач, гідрант, резервуар, водойма, майданчик, мережа, покриття, вокзал, ринок, пляж, вбиральня / туалет, притулок. LEVEL 2 — A FUNCTIONAL SLOT: the noun names a genus and a TYPOLOGICAL ADJECTIVE says which one; bare it takes nothing, with the adjective it takes a span and the adjective is inside — «[очисні споруди]», «[спортивні споруди]», «[спортивні комплекси]», «[контейнерні майданчики]», «[під’їзні шляхи]», «[зливні станції]», «[перевантажувальні станції]». This level holds споруда, будівля, комплекс, шлях. LEVEL 3 — AN ABSOLUTE HYPERNYM: it NEVER takes a span, whatever the modifier, because any modifier on it is locative, legal or functional rather than typological (ambiguity-17) — об’єкт, конструкція, територія, ділянка, зона, місце, район, пункт, установа, організація. «Прибудинкова територія», «територія загального користування», «пункт заправляння водою» and «місце розміщення» take nothing. TWO ADDITIONAL RULINGS. «Підприємство» takes structure when the modifier names the TECHNOLOGICAL function of the site — «[підприємства](structure) [сортування](process) та [перероблення](process) [побутових відходів](material)» — and takes nothing when it names an ECONOMIC ACTOR («спеціалізованим підприємствам»): the first is built and has a plot area computed for it, the second signs a contract. The taxonomy has no organization label and this round does not propose one. Settlement units — населений пункт, населене місце, місто, селище, село — are NOT structure and take no span anywhere: they name an administrative unit, not a built object. The ruling matters because «населений пункт» heads almost every sentence of this ДБН, and all three annotators reached it independently. «Приміщення» joins level 2. The argument that a room is a void rather than an element is refuted by the level-1 list itself, which already carries колодязь, лоток and тунель: a bounded space formed by construction is already treated as structure in these guidelines. A bare «приміщення» takes no span, because one still has to ask «which one?»; a TYPOLOGICAL qualifier naming a class of use gives one structure span together with the adjective — «[виробничі приміщення]», «[житлові приміщення]», «[підсобні приміщення]» — by the same construction as «[спортивні споруди]» and «[очисні споруди]». A locative qualifier still gives nothing: «всередині приміщень», «у тих зонах приміщення». «Трубопровід» joins the level-1 list: an installed line is a built object, while a supplied length of «[труби](material)» stays a product (ambiguity-24 b, ambiguity-39).
баз(structure) прибиральних машин(tool)Level 1: a lexically specific head is structure even bare.
очисних споруд(structure)Level 2: the typological adjective licenses the span and sits inside it.
на прибудинкових територіях(structure)Level 3: no modifier ever promotes «територія».
населеного пункту(structure)A settlement unit is an administrative unit, not a built object.
ambiguity-36
The structure span boundary: adjective in, entity complement out, canonical compound whole
The mirror of ambiguity-30 (a) for place heads, and the three cases differ. (a) A TYPOLOGICAL ADJECTIVE is INSIDE the span, including one that states legal status or grade of construction: «[несанкціоновані звалища]», «[удосконалене покриття]», «[водні станції]». (b) A GENITIVE OR PREPOSITIONAL COMPLEMENT that names ANOTHER entity of the taxonomy stays OUTSIDE and keeps its own label, exactly as the dependent object of a process span does: «[баз](structure) [спеціально обладнаних транспортних засобів](tool)», «[підприємства](structure) [сортування](process) [побутових відходів](material)», «[баз](structure) для [приготування](process) [спеціальних матеріалів](material)». The head is not stretched over its complement. (c) A SETTLED COMPOUND NAME of an engineering network or facility is ONE span and nothing inside it is annotated, by the same reasoning as the canonical interiors of ambiguity-15: «[вулично-дорожня мережа]», «[водостічна мережа]», «[мережа відведення поверхневих вод]», «[система поверхневого водовідведення]», «[проїзна частина]», «[дорожнє покриття]». The list is closed and grows only by coordinator ruling; the test that separates (c) from (b) is whether the whole phrase is a term the domain uses as a single name — «мережа відведення поверхневих вод» is a synonym of «водостічна мережа», while «база прибиральних машин» is a base with machines kept in it.
несанкціонованих звалищ(structure)A legal-status adjective is typological here → inside.
баз(structure) спеціально обладнаних транспортних засобів(tool)The complement names another entity → outside, with its own label.
мережі відведення поверхневих вод(structure)A settled compound name → one span, nothing inside it.
мережі(structure) відведення(process) поверхневих вод(material)Do not dismantle a settled compound name.
ambiguity-37
Precedence when two rules answer the same phrase
Adjudication rounds keep producing apparent contradictions between an ABSORBING BOUNDARY rule, a CLOSED LEXICAL ARTICLE and a DECISION TEST. They are not contradictions; the rules apply at different moments, and the order is fixed. STEP 1 — AN ABSORBING BOUNDARY RULE runs FIRST, because it decides whether a dependent word gets a span of its own at all: ambiguity-10 (a scale name without a code swallows its dependent), nested-1 (a material name swallows a single code), nested-3 (a document title swallows everything). STEP 2 — A CLOSED LEXICAL ARTICLE runs next and beats any test: «розрахунок» is process wherever it heads a phrase (ambiguity-23) even though the two-step test of ambiguity-29 would stop at step 2 (b), because only a document changes. STEP 3 — A DECISION TEST applies only to words that no list names. TWO ARTICLES AT ONE WORD are resolved by the NARROWER syntactic condition. The two conflicts this order settles in the second document, both reported by all three annotators: (1) «Благоустрій» is process in every syntactic position, but «[ступінь благоустрою]» is ONE property span, because ambiguity-10 runs at step 1 and a scale name absorbs its dependent exactly as «[клас бетону]» absorbs a material; the process article never gets its turn. (2) «Дезінфекція» stands in the process list and also in entities-safety as a protective measure; the safety article carries the narrower condition (a named field or item of safety provision), which a coordinated row of technological operations does not meet, so «прибирання, [миття](process) та [дезінфекції](process) громадських вбиралень» is process — and in this genre дезінфекція, дератизація and дезінсекція are process wherever they act on a named physical object. Step 2 gains «будівництво» and «проектування» beside «розрахунок»: the closed lexical article keeps process in industry usage too — «об’єкта нормування у [будівництві](process)». The «експлуатація» exception stands, but it concerns a STAGE of the life cycle, not an industry.
ступінь благоустрою(property)Step 1: a scale name absorbs its dependent; the process article on «благоустрій» never runs.
об’єкти благоустрою(process)No scale name here, so the lexical article applies as usual.
миття(process) та дезінфекції(process) контейнерів(tool)An operation on a physical object → process, not safety.
розрахунок(process) обсягів фінансуванняStep 2: a closed lexical article beats the two-step test.
entities-material and the waste list of a sanitation norm name only відходи and вторинна сировина, which leaves the substances that the same norm collects, hauls, spends and pays for by the tonne without a label. The test is the one entities-material already implies: a substance is material when the document HANDLES it — collects, transports, stores, spreads or consumes it — and takes no span when it is only the environment the work happens in. MATERIAL: сніг and сніго-льодові утворення (collected, hauled and tipped at a «[снігозвалищі](structure)», with a cost line per tonne), вуличне змітання, бруд, ґрунтові наноси, технічна вода and вода for washing and watering, технологічні / спеціальні / протиожеледні матеріали, стічні води, вторинна сировина. NO SPAN: the environment itself — повітря, ґрунт, підземні та ґрунтові води, водні об’єкти — because ambiguity-32 already keeps the environmental side outside the labels, and «поверхневих вод» inside the settled name «[мережа відведення поверхневих вод]» is absorbed by ambiguity-36 (c). Live animals and animal carcasses are the object of вилов, утилізація, кремація and захоронення, but no label in the nine fits them; they take no span, while the actions keep theirs. The decision is deliberately narrow: it adds substances the norm treats as cargo, not every noun that names matter. The MATERIAL list gains drinking water and water for domestic needs: they are distributed through a pipeline and their quality is normed, so the document handles them as cargo. The NO-SPAN list is unchanged. Both roles stand in one sentence of ДБН В.1.2-8:2021: «виділення небезпечних речовин у ґрунтові води, морські води, поверхневі води або ґрунт» — nothing; «виділення небезпечних речовин у [питну воду](material)» — material.
складування снігу(material)Collected, hauled and tipped like waste → material.
підземні та ґрунтові води(material)The environment, not cargo — see ambiguity-32.
трупів тварин(material)No label in the nine fits; the actions keep theirs.
ambiguity-39
A named building service system; «мережа» against «система»
ambiguity-30 (b) lists «система» among the hyperonyms of ACTIVITY, and ambiguity-28 refuses «конструктивна система» as a way of describing a structure. Neither covers an installed engineering assembly, and the closed list in ambiguity-36 (c) already carries one — «[система поверхневого водовідведення]» beside its synonym «[мережа відведення поверхневих вод]». That pair does NOT imply the head is irrelevant: it shows only that two names WITH A FUNCTION converge. The label is still decided by the level of the head under ambiguity-35. LEVEL OF THE HEAD. «Мережа» and «трубопровід» belong to level 1 beside колодязь, лоток and тунель: one does not ask «which one?», so they are structure even bare, and a domain adjective simply falls inside the span under boundaries-1 — «[інженерні мережі]», «[трубопроводів]». «Система» has no level 1: bare, it is a hyperonym, and it takes structure only as a level-2 FUNCTIONAL SLOT, together with a qualifier naming an engineering function. RULE. «Система» / «комплекс» plus a qualifier naming an engineering FUNCTION of the building or site is ONE structure span over the whole name, with nothing labelled inside: «[системи кондиціонування повітря]», «[вентиляційні системи]», «[систем водопостачання]», «[системами природного і штучного освітлення]», «[систем очистки та фільтрації]». A coordinated run of functions under one head is a single span (boundaries-13 b). WHAT COUNTS AS A FUNCTION. The function names the MEDIUM or SERVICE the system handles: water, sewage, air, heat, gas, electricity, light, smoke, signal. The check is mechanical: is the qualifier contrasted with another qualifier in the same classification? If yes, it is a function. BOUNDARY. A qualifier naming a DOMAIN rather than a function gives no span, because there is nothing to contrast it with: «інженерні системи», «конструктивні та інженерні системи», «системами інженерного забезпечення», «системних рішень». «Інженерне забезпечення» is a collective name for ALL functions rather than one of them, and gives no span either. This boundary does NOT extend to level-1 heads: «[інженерні мережі]» and «[трубопроводів]» keep structure whatever the adjective, because the head itself qualifies them. There is no asymmetry here — there is a difference of level, the same one that separates «[полігон]» from a bare «споруда» in ambiguity-35. A supplied length of «[труби](material)» stays a product (ambiguity-24 b); the installed line «[трубопровід](structure)» is a level-1 object. SPAN COMPOSITION. A genitive complement naming ANOTHER entity of the taxonomy stays OUTSIDE (ambiguity-36 b): «[систем відведення](structure) [стічних вод](material)». VERBAL NOUN OF FUNCTION. Inside a system name, nothing. In the head-of-action position the label follows ambiguity-29 step 2: if the thing physically changed is named as a MATERIAL, process («безпечного [водопостачання](process) [питної води](material)»); if only a condition of the environment changes, nothing («природне освітлення», «вентиляція» standing alone). INTERSECTION WITH safety: where the function protects PEOPLE, entities-safety is narrower and wins (ambiguity-37) — «[систем аварійної сигналізації](safety)»; protecting the environment does not, so «[систем очистки та фільтрації](structure)» (ambiguity-18).
системи кондиціонування повітря(structure)Function qualifier → one structure span; nothing inside.
систем водопостачання, кондиціонування та вентиляції(structure)Coordinated functions under one head — a single span.
систем аварійної сигналізації(safety)The function protects people — entities-safety is narrower.
конструктивні та інженерні системи(structure)Bare hyperonym: «інженерний» is a domain, not a function.
систем відведення стічних вод(structure)The material complement stays outside — ambiguity-36 (b).
ambiguity-40
The span goes on the noun: closing ambiguity-19 (c) over the remaining labels
ambiguity-19 (c) requires a verbal noun for process. property had no such closure, and the genre of requirement lists alternates both forms in adjacent clauses: «труби, фітинги, з’єднання і шви мають бути водонепроникними, повітронепроникними, стійкими до корозії» against «вимогам щодо герметичності, корозійної стійкості» in the preceding subclause. All three annotators independently applied the conservative reading, and all three asked for a ruling. RULE: property, load_effect, structure, tool and safety go on the NOUN; a predicative adjective, a participle or any verb form never takes a span — «мають бути водонепроникними, повітронепроникними, стійкими до корозії» stays empty, while «[герметичності](property)», «[корозійної стійкості](property)» take spans. There is exactly one exception and it is already written: boundaries-7 labels a relational MATERIAL adjective if and only if the head word carries no label at all («[бетонного](material) перерізу»), because otherwise the mention of the substance would vanish. An attributive adjective inside a labelled head still belongs to its span and takes none of its own («[корозійна стійкість]», «[вентиляційні системи]»): this rule concerns only an adjective STANDING ALONE as the name of a property. A descriptive participle names a STATE rather than a type and stays OUTSIDE the span, on the same principle as «загальна площа» in ambiguity-34 F1: «працюючого [устатковання](tool)». The cost is known and accepted — a requirement phrased with an adjective yields no span; the vocabulary is not lost, because the same document gives the same term as a noun, and the boundary stays mechanical.
вимогам щодо герметичності(property), корозійної стійкості(property)Noun form → property.
мають бути водонепроникними(property), стійкими до корозії(property)Predicative adjective — no span, by symmetry with ambiguity-19 (c).
працюючого устатковання(tool)A descriptive participle names a state, not a type — outside the span.
бетонного(material) перерізуThe single exception: boundaries-7, head unlabelled.
ambiguity-41
A verbal noun of a phenomenon as the name of a normed characteristic
ambiguity-29 (c) removes process from a phenomenon nobody performs, and ambiguity-16, ambiguity-32 and ambiguity-38 keep substances acting from outside, and the environment, outside the labels. None of them decides the case where the verbal noun of such a phenomenon IS ITSELF THE NAME of a normed characteristic of a product. The precedent stands in the guidelines twice: ambiguity-13 (b) calls «водопоглинання» property — the same head with the agent fused into it — and ambiguity-9 already gave property to «[сумарній активності природних радіонуклідів](property)», leaving the radionuclides unlabelled inside the span. Deciding by whether the agent has fused into one word is not possible: that is the same lexical accident as the head «мережа» against «система» in ambiguity-39. RULE. A verbal noun of an unperformed physical phenomenon — пропускання, поглинання, проникнення, виділення, викиди, емісія, випромінювання, накопичення — is ONE property span together with its complement if and only if THREE conditions hold: (1) the sentence DECLARES the phrase a characteristic of a named product, structure or system; (2) the quantity is measured IN THE PRODUCT, not in the environment (ambiguity-9); (3) it passes ambiguity-34 F4 — there is one indicator and one test method (transmittance g per EN 410; specific activity in Bq/kg; chamber emission per ДСТУ EN 16516). OUTSIDE THESE THREE CONDITIONS there is no label and the prohibitions apply unchanged: the same phrase in a list of FACTORS, THREATS, EVENTS or CRITICAL STATES stays empty. Positionality is not new here: it works exactly as «[водопостачання](process)» in the head-of-action position against the same word inside «[систем водопостачання](structure)» (ambiguity-39). SPAN BOUNDARY per ambiguity-22: an unlabelled complement stays INSIDE, a labelled word cuts the span. A derived quantity in front — «швидкість», «рівень», «обмеження», «зниження» — is a wrapper and stays OUTSIDE (boundaries-9). Condition (1) applies equally to nouns of QUANTITY, not only to verbal nouns of phenomena — for the order between the characteristic frame and the magnitude filters see the ORDER: FRAME FIRST, THEN FILTERS preamble in ambiguity-34.
пропускання сонячної енергії(property)A row in a list of characteristics: the frame declares it a product characteristic.
викиди летких органічних сполук і інших забруднюючих речовин(property)Same list; the substances inside the span carry no label of their own.
виділення небезпечних речовин у ґрунтові водиIn a list of THREAT FACTORS — condition (1) fails, no span.
обмеження (зниження) швидкості виділення забруднюючих речовинA critical state, not a characteristic; plus wrappers in front.
Agreement & merging
How several annotators’ work becomes one dataset entry, and how the published agreement numbers are computed. Methodology version 2 (June 2026); entries merged earlier carry v1 metadata and are marked as such in exports.
agreement-1
Majority-vote consensus
Each task is annotated independently by N annotators (default 3). Spans are projected onto characters; a character keeps an entity label when at least ⌈N/2⌉ annotators assigned it. Consensus spans whose average per-character support falls below 0.6 are dropped from the merged entry — they stay visible in the per-word agreement table.
2 of 3 annotators mark "ДСТУ EN 81-70" as regulation → the span survives with agreement 0.67.
agreement-2
Fleiss’ κ over the entity window
The headline coefficient is Fleiss’ κ computed only over the entity window — character positions where at least one annotator marked an entity. Unannotated text does not enter the calculation, so the value is comparable between a two-sentence fragment and a multi-page clause: the same labeling pattern yields the same κ regardless of text length.
Full-text character-level κ (v1) is still stored as fleissKappa for reference, but it is length-sensitive: the dominant "outside" class inflates expected agreement on long texts and makes κ unstable on short ones.
agreement-3
Span-level F1
For every annotator pair, spans match when they share a label and overlap with IoU ≥ 0.5. The mean pairwise F1 measures whether annotators find the same entities — complementing κ, which measures positional consistency.
agreement-4
Composite validation score
validation_score = 0.35 · max(0, κ_entity) + 0.65 · F1. Span detection is weighted higher because it is what NER training quality depends on. Entries are colour-coded: ≥ 0.7 high, 0.4–0.7 medium, < 0.4 low (queued for adjudication).
agreement-5
Interpreting κ
κ ≤ 0 — no better than chance; 0.2–0.4 fair; 0.4–0.6 moderate; 0.6–0.8 substantial; > 0.8 almost perfect. κ = 1 with an empty entity window means all annotators agreed the fragment contains no entities.
References
External documents that formalize the domain conventions above.
The nine-label set adapts Ontonotes-style named-entity taxonomy to the construction domain; the ninth label, load_effect, was appended in v0.5 for the action side of the Eurocode inequality E ≤ R. Full rationale in the project paper (v1.0).
ref-3
Cite this document
When citing BUDOVA guidelines: "BUDOVA Annotation Guidelines v0.8 (2026)". Deep-link individual rules with the anchor, e.g. /guidelines#entities-material.
Collaboration
Join BUDOVA
We are looking for researchers, construction professionals, and language specialists to participate in the project.