Read this question: “Which report did Aisha edit and Ben approve?” The opening phrase which report belongs to two relations at once: Aisha edited that report, and Ben approved that report. In syntax, this can illustrate across-the-board extraction, abbreviated ATB extraction, in which one displaced expression is interpreted in more than one position across a coordination. To translate it accurately, keep the shared referent and both actions. Do not turn one report into two unrelated reports or attach the report only to the first verb.
Readers searching for across-the-board extraction examples, ATB movement in English, coordinated wh-questions, relative clause extraction or how to translate shared wh-dependencies need a way to verify the hidden relations without getting trapped in formal diagrams. A practical method is to expand the question into a pair of coordinated propositions: Aisha edited which report; Ben approved which report? Both questions seek the same value of the report variable. The final target sentence may use one fronted question phrase, two repeated objects, a relative-clause pattern or a fully explicit construction.
This guide explains how to recognize and translate across-the-board dependencies in questions and relative clauses, why coordination matters, how to distinguish shared-reference interpretations from separate object lists, and what to do when the target language does not readily support the English surface form. We will work through publishing, scientific review, museum records, business processes, education and conversation, then practise with examples and an accuracy checklist. The goal is faithful meaning, not a requirement that every language display English wh-movement.
The reader-first rule: one expression, two relationships
Start with a simple pair: “Lina translated the article. Omar checked the article.” A compact question is “Which article did Lina translate and Omar check?” The article is linked to translated and checked. In a typical across-the-board reading, both conjuncts are missing the same interpreted object, and the fronted phrase provides it for both. This relationship is much easier to see after expansion.
An inaccurate translation could produce “Which article did Lina translate, and what did Omar check?” That asks for two potentially different values. Another might produce “Which article did Lina translate while Omar checked something?” That drops the specific relation between the article and Omar’s checking. Even if the grammar is smooth, the target no longer asks the same question.
The same mechanism can appear in a relative clause: “the article that Lina translated and Omar checked.” The relative expression connects the noun article to both predicates. A translation like “the article Lina translated and the material Omar checked” could introduce two items and make the reader think there were separate documents.
The safest expansion is not necessarily the final wording. It is an intermediate semantic test. Write the referent once, then list every action that targets it. If both conjuncts target the same expression, preserve that relation. If the original actually contains two different expressions, do not force them together just because the verbs happen to be coordinated.
What counts as across-the-board extraction?
In familiar English questions, a wh-expression can appear at the beginning of a clause even though it is understood as an object or other constituent of a verb later in the sentence. “Which folder did May open?” asks about a folder that fills the object position of opened. When two coordinated clauses both depend on a single fronted wh-expression—“Which folder did May open and Dan index?”—linguists describe a possible across-the-board pattern.
The formal significance concerns coordination and the constraints that typically limit extraction from one conjunct alone. The classic idea is that an element can be related to corresponding positions in both conjuncts, rather than extracted from just one while leaving the other unrelated. Different theories propose different mechanisms for this shared interpretation. Some use parallel movement; others invoke deletion, sharing or distinct dependency analyses. The observable meaning is the translator’s starting point.
The phrase across the board does not mean every participant in the world or every item in a document. It refers to the structure of the coordination: relevant conjuncts participate in a linked dependency. Nor does ATB mean a sentence must contain two separate gaps written visibly on paper. A source may simply have natural English punctuation and order, while its syntactic analysis represents positions for the shared expression in each conjunct.
A concise reference is the Utrecht lexical entry on across-the-board movement. For theoretical alternatives and subtle exceptions, see research on the limits of ATB movement and Cambridge research on coordination and subject gaps. A translator can use the structural insight without declaring one contested theory uniquely correct.
A two-conjunct reconstruction
Take “Which design did the engineer test and the client approve?” There are two distinct predications: the engineer tested a design, and the client approved a design. The leading phrase which design seeks the shared value for the object of tested and approved. If the answer is Design C, the intended answer says Design C was tested by the engineer and approved by the client. It does not say the engineer approved Design C or that the client tested it.
A simple diagram for working purposes is: SHARED OBJECT = design; CONJUNCT 1 = engineer tested that design; CONJUNCT 2 = client approved that design. This shows the fact structure without assuming a particular derivation. The translator can now choose a target-language construction that preserves the same pair of relations.
Next mark tense and modality. If the source says “Which design could the engineer test and the client approve?” it may ask about a potential design suitable for both actions, not a design already tested and approved. If the source says “Which design did the engineer test but the client reject?” it uses contrast with but and different outcomes. The shared object does not erase the verbs’ distinct meanings.
Finally check whether the source is a question seeking information, a rhetorical question suggesting a criticism, or a quoted question from a meeting. The grammar may be the same across these discourse uses. A translator should preserve both the dependency and the speaker’s communicative purpose.
Why coordination can mislead an otherwise fluent translation
Coordination is a compact way of combining content, but its surface form does not always show every participant’s complete phrase. Readers often process “Aisha edited and Ben approved” as a natural sequence of workplace actions without noticing that an earlier question phrase is doing grammatical work in both halves. That is exactly where a translation engine may decide to repeat the object for one verb but leave it implicit for the other.
There are three common possibilities that a translator should keep apart. One item participates in both actions; separate items participate in separate actions; or the sentence is structurally ambiguous between multiple analyses that converge on a similar practical reading. The target must not decide among them on the basis of habitual business workflow alone. Formal coordination and real-world plausibility are different sources of information.
A useful counterexample is a printer that produced two versions of a report. “Which report did Aisha edit and Ben approve?” ordinarily seeks one identifying report that satisfies both predicates. “Which reports did Aisha edit and which did Ben approve?” asks two possibly different questions. The visible repetition of which creates a different information structure, even when the broad topic is the same.
If a target language requires repeating a noun phrase in each coordinated clause, that repetition can still express one shared item by using matching reference, definite markers or an explicit same expression as appropriate. But adding same unnecessarily can sound stronger or more contrastive than the source. Prefer the natural local mechanism for shared reference and verify its interpretation with a bilingual reader.
Worked example: articles, editing and approval
Source: “Which article did the researcher draft and the editor publish?” Reconstruct the relations: researcher drafted the article; editor published that article. The source does not assert that the editor drafted it or the researcher published it, and it does not supply a date or reason for publication. A precise translation asks which article satisfied both relations.
What if a target version says “Which article did the researcher draft, and which one did the editor publish?” In everyday use, the pair of questions may invite two different article answers. The original has one leading wh-expression. A translator may repeat the question phrase for grammatical reasons, but then must ensure that the intended same-item reading survives. Without a linking device, the new wording may no longer pose a single shared-referent question.
Now imagine the source’s next sentence: “The editor rejected three other drafts.” This context does not license a translation that adds “only one article” unless the question’s answer or another sentence establishes that count. The wh-expression seeks identification and does not itself assert uniqueness in every discourse.
For publication metadata, articles may be drafts, submitted manuscripts, accepted manuscripts or published records. Keep the source’s status verbs exact. Drafted and published identify very different stages; an accurate translation does not collapse them into “produced.” The syntactic relationship can be perfectly preserved while the workflow semantics is lost if both verbs are replaced by a generic word.
Worked example: research data with two verification stages
Source: “Which dataset did the biologist compile and the statistician validate?” One dataset is linked to compilation by a biologist and validation by a statistician. The source does not say the biologist validated it, nor does it claim that validation proves the dataset contains no error. The noun dataset identifies the shared object; the two verbs describe different operations.
Suppose a translator writes “Which dataset did the biologist compile before the statistician validated it?” The chronological before relation may be plausible, but the English source with and does not necessarily encode that sequence. It is often inferable from the workflow, yet the translator should not add it as a fact unless the surrounding text warrants it. Linking one dataset to both actions is enough.
For an annotated quality check, create fields: shared referent = dataset; first agent = biologist; action 1 = compiled; second agent = statistician; action 2 = validated; chronology = not expressed; validation criteria = not expressed. If the target adds peer-review status, number of samples or a date, it exceeds the source.
If the target language permits relative clauses more naturally than direct coordination of two gaps, a paraphrase such as “Which dataset was compiled by the biologist and validated by the statistician?” may work, although it changes voice and topical emphasis. Such restructuring is acceptable when it keeps the participant-action assignments and the shared referent.
Worked example: museum collections and a single object
Source: “Which sculpture did the curator catalogue and the conservator restore?” The same sculpture is the object of cataloguing and restoration. A translation that writes “the sculptures catalogued by the curator and those restored by the conservator” can suggest two groups, whereas the question is about a shared object meeting both descriptions.
Catalogue and restore are not synonyms. Cataloguing can involve identification, descriptive records and indexing. Restoration concerns physical intervention and conservation judgments; it is not guaranteed to improve condition and may be subject to strict professional standards. The source does not tell us how extensive either act was. Translators should retain the verbs’ different functions.
A museum label might instead use a relative clause: “the sculpture that the curator catalogued and the conservator restored.” This is one of the clearest demonstrations of shared relative reference. If the target has special markers for restrictive relative clauses, preserve the intended identifying function rather than turning the phrase into a detached aside about two unrelated activities.
The semantic audit is to ask a museum reader to point to the sculpture that qualifies, then identify both actions. If the reader selects a pair of sculptures, the translation has likely broken the original dependency. If the reader understands a single sculpture linked to both roles, the central meaning survives even if the grammar looks nothing like the English question.
Worked example: invoices in compliance records
Source: “Which invoices did Finance approve and Compliance archive?” The fronted plural expression invoices is understood with both approved and archived. The question requests the set of invoices for which both actions hold. It does not ask separately for all Finance-approved invoices and all Compliance-archived invoices; the overlap is the central information.
The distinction becomes visible with sets. Suppose Finance approved invoices A, B and C, while Compliance archived B, C and D. The answer to the shared-dependency question is B and C, not all four invoices. A translation that asks “Which invoices were approved, and which were archived?” may elicit A, B, C for the first question and B, C, D for the second. That is a different inquiry.
Do not infer that approval preceded archiving unless the text or workflow defines it. Do not infer that archiving signals agreement with the payment amount. A document can be archived without being approved, and a record of approval can exist without an archived copy. The source deliberately asks about invoices satisfying both properties, not about the meaning of either process.
For a bilingual audit, compare the target against a tiny matrix with rows A through D and columns approved and archived. If a reader cannot identify the intersection from the target question, the structural relationship has been weakened. This simple set test is often clearer than a diagram of syntactic movement.
Worked example: candidates and two recruitment actions
Source: “Which applicant did the panel interview and the board appoint?” The same applicant is the object of interview and appoint. The source does not state that the panel appointed anyone or that the board conducted an interview. It also does not say how many applications were received or whether the interview caused the appointment.
A translation that writes “Which applicant did the panel interview before the board appointed them?” adds a temporal or procedural relationship that may be contextually natural but is not explicit in the conjunction and. If an official hiring process does establish this sequence, the surrounding text can support an elaborated rendering. Without that evidence, retain the separate actions.
In languages where the object must be repeated, ensure that the two repetitions identify the same applicant. A translation with indefinite forms meaning “some applicant” in each conjunct could accidentally introduce two applicants. A definite or anaphoric mechanism may be needed, depending on the target grammar.
This distinction matters in recruitment records. An interview and an appointment are separate actions by separate bodies. A target that makes the panel the appointing authority may misreport governance arrangements. ATB analysis helps catch the grammatical reassignment; accurate institutional terminology prevents a second, domain-specific error.
Worked example: laboratory instruments and shared samples
Source: “Which sample did Ana measure and Ben photograph?” The sample is common to the two acts, but the measurement and photograph are different outputs. If a translation says Ana measured the sample and Ben measured the photograph, it has attached the second predicate to the wrong thing. If it asks which sample Ana measured and which object Ben photographed, it may cease to ask for a common sample.
Imagine a lab with samples S1, S2 and S3. Ana measured S1 and S2; Ben photographed S2 and S3. The question’s answer is S2. The target must allow readers to recover that intersection. It should not require the same measurements to be visible in the photograph or imply that photography validated the measured values.
Precision in scientific writing includes not upgrading methods. Measured is not assessed, photographed is not documented completely, and neither action proves a result. The syntactic issue is reference sharing; the scientific issue is method identity. Both must be checked separately.
This example also shows that a single source noun may be translated differently according to a controlled domain vocabulary, but its identity should remain stable across clauses. A sample number or specimen label may be preferable to a broad noun in a highly technical document. Such labels must come from the source, never from the translator’s imagination.
Worked example: school learning and two teachers’ actions
Source: “Which essay did Ms Lee mark and Mr Khan discuss?” The essay is the shared object of marked and discussed. This does not say Mr Khan awarded a grade or Ms Lee conducted the discussion. In an educational setting, those distinctions matter because marking, feedback, moderation and discussion each refer to different teaching activities.
A target rendering such as “Which essay was marked by Ms Lee and discussed by Mr Khan?” is an acceptable paraphrase when the same essay remains identifiable. But if the original is a direct question from a student, a passive reformulation may sound more formal. That stylistic change can be acceptable in an institutional report but awkward in dialogue. Accuracy and voice deserve separate attention.
Try adding context: “The students wrote four essays, and two were chosen for additional feedback.” The source question can refer to one of those essays, but it does not tell us which until answered. A translation that assumes the most recent essay was the one marked and discussed is reading an answer into the question.
A useful classroom exercise is to draw four essay cards and show which teacher performed which action. Ask students to identify the essay that belongs to both sets. The visual model makes the ATB dependency comprehensible without advanced terminology and reveals the failure of target wording that accidentally asks two independent questions.
Worked example: oral histories and one speaker
Source: “Which witness did the historian interview and the reporter quote?” The target answer identifies a witness who was interviewed by the historian and quoted by the reporter. It does not imply the reporter attended the historian’s interview, that the quotation was taken from it, or that the witness confirmed every claim in the historian’s notes.
In historical reporting, the source of a quotation is a distinct evidential matter. The grammar links the witness to two actions; it does not prove the relationship between the historian’s interview and the reporter’s published text. A translation that writes “Which witness did the reporter quote from the historian’s interview?” would add a chain of provenance not established by the original.
The question may also matter ethically. Names, pseudonyms and anonymised designations should remain consistent. If one witness is referred to as W4 in the source, the translated clauses must not introduce separate referents W4 and W5. At the same time, do not expand the designator to an identity absent from the record.
A clean target should let a reader answer “the same witness was interviewed by one person and quoted by another” while leaving the venue, timing and source of the quotation unknown unless specified. That is an example of preserving structure without inventing evidence.
Worked example: a shared topic with two different prepositions
Source: “Which proposal did Lina speak about and Omar write about?” The sentence is deliberately constructed to show a shared topic object in two prepositional relationships. Lina spoke about the proposal, and Omar wrote about that same proposal. The sentence does not require both people to hold the same opinion or to discuss identical aspects of it.
Many target languages do not like repeating a preposition at the end of coordinated clauses. An idiomatic translation may instead use a relative construction, repeat the noun after each preposition, or move a phrase to a different position. That is fine as long as the single proposal remains the subject matter of both actions. But if the target asks which proposal Lina spoke about and what Omar wrote, two objects become possible.
The subject matter also should not be changed into the object of action. Speaking about a proposal differs from supporting, submitting or implementing it. Writing about a proposal differs from drafting the proposal itself. A translation that collapses about into a direct-object relation may make the speakers participants in policy creation when they only commented on it.
This is a useful example of why lexical valency matters. Different verbs select different grammatical arrangements. The shared referent stays constant, but its syntactic relationship to each predicate must be respected. The goal is not symmetrical surface punctuation; it is faithful thematic roles.
Worked example: public transport incidents and shared reference
Source: “Which incident did the station manager report and the inspector investigate?” The incident is linked to reporting by one participant and investigation by another. The source does not imply that the manager investigated or that the inspector originally reported the incident. It also does not establish whether the investigation concluded or verified the report.
If the target says “Which report did the manager make and the inspector investigate?” incident has become report. That may sound plausible, but investigating an incident and investigating a report differ. One concerns an event; the other concerns a document or claim about an event. Keep the head noun stable unless the source provides a basis for the shift.
The same-item reading can be tested with incident IDs. Suppose the manager reported I-21 and I-22, while the inspector investigated I-22 and I-23. The original question seeks I-22. A translation that produces two separate lists has missed the central information need.
Operational language often contains status terms: reported, opened, referred, investigated, closed. Translate each state distinctly. The ATB dependency connects an incident to more than one state or action; it does not equate those states or guarantee completion. That distinction can affect accountability and must survive natural-language localization.
Worked example: product testing and customer approval
Source: “Which prototype did the engineers evaluate and the customers prefer?” One prototype must fit two predicates: evaluated by engineers and preferred by customers. Evaluation may have technical criteria, while preference may reflect user experience. The sentence does not assert that engineers preferred it, that customers tested it scientifically, or that the prototype was selected for release.
A target that writes “Which prototype did the engineers recommend and the customers approve?” changes both verbs and turns evaluation and preference into institutional recommendations or approvals. The sentence may still sound like a product meeting, but it asks a different question.
Now consider numbers. If five prototypes were evaluated and customers preferred two, the answer to the original question depends on which preferred prototypes were evaluated; it is an intersection, not necessarily a unique winner. Do not add the word best or one unless context establishes uniqueness. Wh-questions ask for identity, not always a single item.
For localization of a research questionnaire, the target must preserve the same response space. If the source lets a participant name more than one prototype, a translation that demands exactly one answer can affect survey results. Syntactic faithfulness and research-method faithfulness meet at the point where a shared wh-expression identifies the relevant set.
Questions, relative clauses and other surface shapes
ATB-style shared dependencies are not limited to direct questions. A relative clause can connect one head noun to multiple predicates: “the essay that Priya summarized and Mateo translated.” A topic construction can likewise link a foregrounded expression to several clauses, depending on the language and syntax. The public-facing translation should preserve the relation between the head or topic and the coordinated predicates.
But not every sentence with and contains across-the-board extraction. “Which novel did Priya buy and read?” might be analyzed as coordination of verbs with one shared object or through another relevant structure; the surface alone does not force one theoretical account. If the intended reading is one novel both bought and read, translation should preserve that interpretation without insisting on a particular syntax-tree explanation.
Similarly, “The editor copied the figure and the caption” is plain object coordination, not a question phrase extracted across two conjuncts. “Which figures did Aisha copy and which captions did Ben correct?” has two independent wh-expressions. “Which figures did Aisha copy and Ben correct?” ordinarily asks for one shared set of figures. These differences should be checked before compressing or expanding the text.
A good translator tests both possibilities by asking whether the source could be answered with one shared item set or with two unrelated lists. If a rewrite alters that answer format, it likely changed the information structure. This is a useful test even when the target language has no familiar term for ATB.
Asymmetry, islands and why a grammatical rule is not a slogan
Introductory syntax accounts often teach that extraction from one part of a coordination is constrained and that across-the-board dependencies provide a special pattern in which the relevant extraction relation is shared. That is a useful baseline, but actual linguistic research discusses different coordination structures, exceptional patterns and variation in judgments. Some apparently asymmetric examples become acceptable under particular interpretations or analyses.
For translation, avoid turning the baseline into a claim that every sentence not fitting the diagram is ungrammatical in every dialect and language. The goal is to read the source as it stands. If a nonstandard passage makes sense in a character’s speech, a target may preserve the meaningful irregularity. If a formal legal draft is ambiguous because an object seems linked to only one conjunct, ask for clarification rather than correcting it automatically.
Likewise, “ATB” should not become a universal rule that one referent is always physically identical in every conceivable reading. A question about which book two people bought may refer to the same title rather than the same copy, depending on context. Structural reference sharing and real-world identity at the level of individual physical objects must be kept distinct.
Scholarly discussions such as On the Limits of Across-the-Board Movement illustrate this complexity. Translators benefit from the insight that linked gaps can be analyzed in different ways, but they should not let theoretical disagreement obscure ordinary, testable meaning in a clear sentence.
When the shared item is a title, type or physical token
Consider “Which book did Maya read and Leo review?” The shared noun phrase may denote a title or work rather than the same printed copy. Both actions can be true if Maya reads one copy and Leo reviews another copy of the same work. If a translation says “the same physical volume,” it adds a level of identity that the source may not support.
Now switch to “Which parcel did Maya weigh and Leo seal?” In a logistics register, parcel often refers to an individually tracked package. The intended same-item reading likely concerns a particular parcel, perhaps identified by a tracking number. But still do not invent a number or claim that the parcel was shipped. Weighing and sealing do not logically entail dispatch.
A third example is “Which software version did the tester install and the reviewer approve?” Here version identifies a release or build, not one physical device. A correct target should not suggest that the tester and reviewer used the same hardware unless stated. The ATB relation links both actions to the software version.
This distinction matters because a translation may preserve the grammar and still change what counts as one referent. Before choosing a definite pronoun or repetition, ask whether the source noun denotes a work, a type, an instance, a collection or a labeled record. Domain knowledge helps interpret reference; it does not supply permission to invent one narrower identity without evidence.
Focus, emphasis and answer expectations
A fronted wh-expression signals the information the speaker seeks. In “Which report did Aisha edit and Ben approve?” the report’s identity is unknown or being solicited. The two action descriptions establish criteria for an answer. If the translator instead writes “Aisha edited the report and Ben approved it,” the sentence becomes a declarative assertion rather than a question. That changes communicative function even if the participant roles survive.
Questions can also be rhetorical. “What rule did the school publish and then fail to enforce?” may communicate criticism as well as request identification. The shared object remains a rule, but the second predicate carries a negative evaluation in context. A target should neither erase the implied criticism nor intensify it into a factual accusation beyond what the question asserts. Context and tone matter.
Focus can be altered by moving the object or switching to passive voice. Such changes are acceptable when required by target-language grammar, but assess whether they keep the intended response type. A survey item that asks for which report must not become a yes/no question about whether any report exists.
For reader-facing instructions, write a test answer to the target question. If the expected answer is a report title or identifier and the target naturally elicits that, the focus is likely correct. If it elicits two names, two actions or a simple yes/no response, review the interrogative construction.
A bilingual translation workflow for shared dependencies
Begin by isolating the coordination. Identify conjunct one, conjunct two and the fronted or relative expression. Write a full version of each conjunct containing the expression. Use exactly the source verbs: edited, approved; measured, photographed; catalogued, restored. Do not replace them with workflow synonyms.
Next ask whether the expression is shared across the two relations. If yes, mark a single referent variable X and map X to the appropriate argument position in each conjunct. For “Which policy did they explain and the community reject?” the map is explain(X) by they and reject(X) by the community. The same policy is at issue. Then identify negation, aspect, time, modality and other modifiers in each clause.
Draft a target-language version with the two relations unambiguously available. If the language requires repeating the object or using an overt pronoun, do so with reference tracking. If it allows a compact coordinated relative clause or shared question phrase, use that when natural. Do not force fronting where the target grammar does not support it.
Finally test the target against a mini dataset. Make three items that satisfy only the first action, only the second and both. The target must select or ask about the both category when that is the source’s meaning. Also check for unintended additions such as before, therefore, jointly, unanimously or successfully, which are not necessarily encoded by and. This workflow makes a subtle dependency concrete.
An audit of frequent mistranslations
Two answers instead of one shared answer. The target asks what each participant handled separately. One action dropped. The target retains the first predicate but leaves the second without the shared object. Agent swap. The second participant is assigned the first participant’s task. Predicate flattening. Edited and approved become “worked on,” obscuring workflow states.
Invented chronology. A conjunction becomes an explicit before/after relation. Invented causation. An action by the first agent is said to have caused the second. Unwarranted unanimity or collective success. A shared reference is mistaken for agreement among agents. Wrong noun ontology. A book title becomes one physical copy; a software version becomes one device.
Premature unique answer. A question about which invoices becomes a statement about one invoice even when the source permits several. Loss of interrogation. A question turns into a factual assertion. Misplaced negation. “Which bill did the committee approve but the minister reject?” becomes a target in which the minister approved or the committee rejected. Inaccurate repetition. The target repeats a pronoun or noun with a different referent in one conjunct.
These failures can be caught systematically. A good reviewer creates a small table containing head expression, first predicate, first agent, second predicate, second agent and expected answer set. Read the target without consulting the source. If the same table cannot be reconstructed, inspect the specific row that changed. That is more efficient than generic requests to make the translation clearer.
Practice questions and worked answers
Problem 1: “Which painting did Jo purchase and Mei restore?” Answer: the painting that Jo purchased and Mei restored. Do not say Mei purchased it or Jo restored it.
Problem 2: “Which database did the analyst query and the engineer back up?” Answer: the database involved in both actions. Querying does not mean modifying, and backup does not mean validation.
Problem 3: “Which author did Lina invite and Omar interview?” Answer: the same author is the invitee and interviewee. The source does not report whether the author accepted the invitation.
Problem 4: “Which regulation did the ministry announce and the court review?” Answer: one regulation qualifies under both relations. Announced does not imply enacted or upheld; review does not imply approval.
Problem 5: “Which photographs did the curator identify and the archivist scan?” Answer: the shared set of photographs identified by one person and scanned by the other. Do not turn identify into discover or scan into publish.
Problem 6: “Which files did the secretary prepare and the director sign?” Answer: the same files were prepared and signed by the relevant parties. Sign does not automatically mean the director checked every factual claim inside them.
Problem 7: “Which essays did the teacher mark and the moderator select for discussion?” Answer: the intersection of marked and selected essays. Not every marked essay need be selected.
Problem 8: “Which complaint did the support agent escalate and the manager resolve?” Answer: a complaint that underwent both actions. Escalation does not guarantee resolution unless the second act separately asserts it for that complaint.
Problem 9: Convert “the product that the technician tested and the customer returned” into a question. One possible source-language question is “Which product did the technician test and the customer return?” The target must preserve one product variable with both predicates.
Problem 10: Compare “Which contracts did Legal approve and Finance archive?” with “Which contracts did Legal approve, and which did Finance archive?” The first naturally asks for the shared set; the second can ask for two sets. The translator should preserve the source’s intended answer structure.
For each problem, mark X as the common referent and write two full propositions about X. Translate those propositions into your target language, then try recombining them. A recombined sentence is successful only when a reader can still determine that both predicates apply to X.
Frequently asked questions
What is across-the-board extraction? It is a dependency in which one displaced expression is interpreted with positions in multiple coordinated parts, as in “Which book did one person read and another review?”
Is there always exactly one physical item? Not necessarily. The shared expression can denote a title, type, set or individual token according to context. Preserve the reference level the source supports.
Is this the same as gapping? No. Gapping removes verbal material in a coordinated structure, while ATB extraction involves a displaced expression associated with more than one position. Both can make sentences compact, but their missing relationships differ.
What is the difference from a shared object after two verbs? A sentence such as “Jo bought and read the book” can have one object governed by coordinated verbs without necessarily needing an across-the-board analysis. Translation should preserve the common object, not force a theoretical label.
Must the target language front a wh-word? No. Languages have different strategies for interrogatives, relativization and coordination. A target may use an in-situ question phrase, relative marker, repeated object or other natural structure.
Does and mean both actions happened at the same time? No. It coordinates propositions without always fixing their temporal relationship. Do not insert simultaneity, chronology or causation unless context supports it.
Are asymmetrical extractions always impossible? No universal statement of that sort is safe. Syntactic acceptability varies with construction, interpretation, discourse and language. Analyses of exceptions remain active research topics.
Can a relative clause show this pattern? Yes. “The report that one editor revised and another approved” links the head report to two predicates.
How can an AI translation be checked? Ask it to identify the shared expression and list the two participant-action relations. Compare those with a source-grounded expansion and a simple three-item intersection test.
When does domain review matter? When verbs such as approved, certified, diagnosed, reported or authorized carry specialized consequences. Syntactic accuracy is necessary but does not replace legal, scientific or technical terminology review.
Related constructions and a route to further learning
ATB extraction sits near several other syntactic puzzles but should not consume their territory. Right-node raising concerns material shared at the right edge of coordinated expressions and has its own interpretive and theoretical debates. Parasitic gaps involve a dependency with an additional gap licensed in particular environments. Gapping and stripping concern omission and remnants. Relative clauses more broadly involve how an identifying expression relates to its clause. The reader should keep these distinct tools available rather than flatten every compression into “ellipsis.”
Explore the dedicated Right-Node Raising translation guide, Parasitic Gaps guide, Gapping in Coordination guide, and the Master Art of Translation for the larger meaning-preservation system. Each offers a different way to ask where an expression belongs and what the target reader must understand.
What connects them is a rigorous habit: expand the meaning before compressing it again. In ATB constructions, reconstruct every relation held by the one displaced expression. Then choose the natural target-language form that preserves them. Translation becomes more dependable once invisible dependencies are treated as information that must be accounted for.
Final source-to-target quality check
Before publishing a translation of “Which report did Aisha edit and Ben approve?”, verify five things. The target remains a question if the source is a question. The same report variable is available to edited and approved. Aisha remains the editor and Ben the approver. No completion status, chronology or causation beyond those predicates has been invented. The target is natural enough that a reader can answer with a report identifier.
Apply the same test to relative clauses by asking which head noun both actions describe. If the target leads readers toward two independent items where the source identifies one shared item or set, revise it. If the target preserves a single item but changes the action verbs, revise that too. Referential alignment and predicate identity are separate requirements.
The guiding rule is compact: one shared expression can participate in two different relationships; translate both relationships without splitting the referent. That is the practical value of across-the-board extraction for multilingual writing, translation and editorial review.