Chapter 2. What makes it the same question?

2.1Words and answer alternatives

Chapter 1 closed by asking when two questions are treated as interchangeable. A concrete, if hypothetical, case shows how to begin. Consider a fixed service and date: one person asks, "Will the nine o'clock ferry operate on Monday?" and another asks, "Is Monday's nine o'clock crossing running?" When both speakers mean the same scheduled sailing on the same date, the wording differs but the requested information appears identical.

What has been held constant here matters more than what has been altered. The service, the departure time, the day, and the mode of operation are fixed by context rather than by the sentences themselves. Under those stipulations, both utterances call for a determination of whether the sailing takes place. The wordings vary in syntax and vocabulary—operate versus running, along with a different ordering of the time expression—yet those differences do not obviously shift the point of inquiry. This study's assessment concerns the hypothetical case, not a result about how real interlocutors would actually resolve it.

Wording alone is therefore an insufficient guide. The same sentence can pose different questions if the referent shifts, and different sentences can pose the same question once the relevant conditions are fixed. Roberts supplies a question-centred account of linguistic relevance (Roberts, 2012; earlier manuscript 1996/1998); choosing which question should govern this comparison remains an editorial task. On the editorial account used here, the answer alternatives for the present example concern whether the specified sailing operates. Comparing those contents gives an editor more to examine than a match between strings.

The proposed match remains conditional. An unstated difference in service, date or intended sense of operation could defeat it. This is a local comparison under declared assumptions, not a complete identity test. A further difficulty appears when the same answer word expresses different contents.

Consider two questions about the same hypothetical crossing: "Will the ferry operate on Monday?" and "Should the ferry operate on Monday?" Each admits the same short answers, yes and no, so a transcript of responses might look identical. Yet the answers assert different things. A yes to the first is a forecast about how events will unfold; a yes to the second is an endorsement, a judgment that operation is warranted. The word yes does not carry these contents on its own. Its contribution is fixed by the question it answers, and the two questions fix different contents.

This is why matching answer strings are weaker evidence of a shared question than they appear. What counts as an answer is not merely a string but an assertion relative to a question, and the assertion differs here in kind. The first question invites appeal to plans, schedules, and capacity; the second invites a standard against which operation can be assessed without requiring that a forecast supply that standard. None of that is settled by the forecast. Even a well-supported prediction that the crossing will run supplies no normative conclusion on its own: one would need an additional premise connecting expected operation to the standard that makes operation permissible or desirable. Without that premise, agreement in wording masks a gap in what has been asserted.

Grouping the two questions under one heading nevertheless has limited usefulness. They share a subject, the fate of a crossing, and setting them side by side makes the inferential gap visible rather than hiding it. A reader who answers both can see which claim each response commits her to, and where a further premise is required before one answer can support the other. The point of the comparison is local: identical answer words do not establish that the same question is being asked, and a forecast about operation does not by itself justify the normative judgment. Who should hold the authority to make that judgment is a further matter, which this comparison does not resolve.

2.2Presuppositions that travel silently

Consider a hypothetical case in which an editor recasts “Is the ferry service running?” as “When will the suspended service resume?” The second formulation presents a suspension as background and asks about a return. The original asked about current operation without supplying that premise. If the editor has no further evidence, the change introduces an assumption rather than merely shortening the wording. A reader might answer with a proposed date, reject the assumption, or explain that no return has been scheduled. Those responses do different work; the timing format does not make the background claim true.

Nothing in this construction establishes the status of an actual ferry. Nor does it establish what a real speaker believes: people may repeat someone else’s description, speak hypothetically, or challenge a premise through their answer. In the ordinary literal reading stipulated here, the description “suspended” and the verb “resume” introduce the relevant interruption. The word “when” alone does not supply that entire assumption. A question about a first departure would organize its background differently. The example concerns this particular reformulation, rather than a universal rule for all timing questions.

The practical loss is a change in what remains open. An inquiry about whether an interruption exists becomes an inquiry conducted against an interruption taken for granted. An editor can expose that change by retaining the original question and naming the added premise. This does not settle whether the premise is right. It identifies the point at which evidence or clarification is needed before the reformulation can replace the original.

Continue the hypothetical question about when service will resume. Stipulate that the record gives neither reliable evidence of a suspension nor enough context to identify the interruption being discussed. A cancelled sailing, a seasonal pause and a permanent closure would require different follow-up questions. The editor cannot determine which event is meant merely from the phrase “the suspended service.” The missing context limits this record; it does not show that every question lacking an explicit date is unintelligible.

Neither of two tempting moves is warranted. Treating suspension as established would add an operational fact unsupported by the available material. Silently rewriting the service as continuously running would make a different unsupported assertion. An absence of reliable evidence here establishes neither alternative. The editor can preserve the contributor’s question while withholding endorsement of its background premise. That is a report about what was asked, not a decision that the service was or was not interrupted.

A conditional reformulation can expose the gap: if the service was interrupted during the period the contributor means, what, if any, return has been announced? A request to identify that period and the basis for the suspension claim accompanies the reformulation. The phrase “if any” also leaves open that no return is scheduled. For discovery, the record can remain grouped with service-interruption questions, provided that the grouping does not certify the unverified event. More exact comparison can wait for clarification without making the provisional entry disappear. The useful result is a visible condition that another editor can test, rather than a hidden assertion inherited from a shorter question.

2.3Purpose without mind reading

Consider a hypothetical ferry service in which the printed departure time and the time the vessel actually leaves may come apart. A traveler filling in a reimbursement form may ask when the ferry is scheduled to depart, while a traveler deciding whether to hurry along the quay may ask when the delayed ferry will in fact leave. The purposes diverge, and with them the standard of adequacy. A faithful reading of the timetable satisfies the first request; it fails the second, because the form-filler needs the documented time and the runner needs the best available account of the actual one. Two questions can therefore share subject matter and still demand different answers, because what the answer is for settles what it must deliver.

The contrast does not license a general multiplication of questions by motive. Suppose, as a further hypothetical construction, that one passenger wants the departure time to plan a meeting, another to decide when to eat, and a third merely to pass the time. A single factual answer—say, that the vessel is now expected to leave at 4:40 p.m.—may serve all three. Distinct personal aims do not entail distinct questions whenever the same answer answers each aim as well as any other would. Conversely, the shared aim of two speakers would not by itself make their questions identical if the information each seeks differs. The purpose of an answer grades adequacy against a stated task; it does not, on its own, redraw question boundaries.

This restraint answers a natural objection. If purpose were treated as essential to question identity, any speaker could insulate a question from comparison simply by declaring a private aim, and identity claims would become unverifiable. The ferry example avoids that outcome because the two purposes require different kinds of information: one demands the documented schedule, the other the expected actual departure. Where the declared tasks can be satisfied by the same fact, one answer can serve several motives at once. Adequacy judgments here therefore presuppose a stated practical task; when the task is unclear and the distinction matters, the editor should ask what kind of time is needed.

The objection is fair as far as it goes: an editor cannot open a speaker's mind and read off the purpose that governed an utterance. Yet treating intention as essential need not make comparison impossible. What an editor can rely on is the purpose a speaker expresses, the clarifications offered when challenged, and the practical context surrounding the question. These are fallible evidence of intention, not direct access to it. Editorial judgments drawn from them should therefore be labeled as inferences: the speaker said the ferry timetable question was meant to decide a family's departure day, and the editor infers that a departure-day purpose guided the request. That inference can be wrong even when the speaker is sincere, and it should not silently harden into a claim about what the speaker "really asked."

This evidential discipline also avoids granting speakers an unlimited veto over every comparison. A speaker who later reports a different goal has not thereby demonstrated that the earlier record was misconstrued. Consider a hypothetical case: the speaker originally asks which ferry runs on Tuesday, then a month later wants a comparison of prices across ferries. The new purpose is entirely legitimate, but it does not prove that the initial question secretly concerned pricing. The sound practice is to record the second goal and date it as a later revision, rather than rewriting the original question so that it appears always to have contained that aim. Purpose can shift without either the earlier or the later question being illegitimate.

Bender and Koller similarly urge distinguishing linguistic form from meaning and understanding under their form-only training premise (Bender and Koller, 2020). That position motivates caution about equating surface similarity with shared meaning; it is not, and should not be treated as, an empirical demonstration that current systems cannot compare questions. The modest conclusion is narrower: editors should ground purpose claims in expressed evidence, mark their own inferences, and keep revisions dated. A new goal can be honored without being retroactively inserted into the record.

2.4Questions versus demands

Consider a hypothetical meeting at which a speaker says, "Can the evening ferry be kept open this summer?" and follows it with a follow-up clarifying that no commitment is sought at that point and that only a feasibility assessment is wanted. On its own, the interrogative leaves several readings open. It may seek a factual assessment of feasibility, invite others to explore a proposal, or advance the case for keeping the service running. The sentence form settles none of these. What separates the readings is the surrounding talk: a request for costs and staffing constraints points toward a feasibility inquiry, while an accompanying appeal to support the proposal marks advocacy. If the speaker adds, "Would you be willing to work up an option?", the utterance shifts from asking a question about the world to asking listeners to take an action.

This example shows why intent should not be recovered from punctuation alone, and why intent need not be treated as a hidden essence granting speakers a veto over any paraphrase. The reported utterances supply observable material—questions asked, tasks proposed, commitments sought—from which an editor can classify the practical purpose. Where that material is absent, the honest description is underdetermination rather than a guess about inner aims. The claim here is modest: wording and answer alternatives can be compared independently of intent, but a statement of practical purpose requires textual or contextual support.

For the explicitly described exchange, a heading such as Feasibility of extending evening service preserves the stated inquiry. If the speaker also asks someone to develop an option, a short note should retain that requested task. “Keep the evening ferry open” would instead present a demand the speaker expressly withheld. The heading need not preserve every possible reading after the surrounding utterances have narrowed the interpretation. It should preserve the reading those utterances support, together with any material uncertainty that remains. The purpose is to represent the meeting contribution, not to replace it with a general heading about interrogative sentences.

Consider a hypothetical ferry meeting, with the dialogue invented here and the existing late-service timetable ending in September. A representative asks: "Will you extend the evening service through October?" The manager replies that he understands the request to be for keeping the later crossing running for another month, and that he is saying no this year, because the maintenance budget is already committed. Nothing in this exchange requires treating the manager as having misheard, nor as having agreed. Three distinct things occur. The request was received: the utterance reached the manager and was taken up as directed at him. It was interpreted: he identified what action was being requested and over what period. And it was answered: he declined, supplying a stated budget reason.

Suppose the exchange were logged simply as a failure of understanding, because no service extension followed. That record would mistake a refusal with a stated budget reason for a defect in comprehension, and would fail to record the basis actually offered for the refusal. Suppose instead it were logged as assent, because the manager restated the request before replying. That record would invent an agreement that no one expressed. An accurate restatement supplies evidence of interpretation, but mere repetition is no guarantee of understanding. The explicit refusal still needs to be recorded separately.

The local lesson is that reception, interpretation, and response are separable stages within one exchange, and that a reply may confirm interpretation while rejecting the ask. The case does not establish a universal test for when two wordings express the same question, nor does it settle who may authorize a budget decision. It does show that collapsing the stages destroys exactly the disagreement worth preserving, while the constructive gain is a record that keeps the request, its reading, and the answer apart.

2.5Reformulation and explicit assent

Asking a contributor to confirm one particular reformulation is a modest but concrete way to check whether a question has survived a change of wording. Clark and Brennan treat communication as coordination of content and process, with common ground updated through interaction (Clark and Brennan, 1991). On this study's editorial account—distinct from anything those authors claim about consent—understanding is assessed against what the exchange is for rather than a context-free standard of meaning, and a contributor's confirmation of a reformulation falls short of agreement to a proposed action.

Consider a hypothetical exchange about whether an evening crossing should be rescheduled. An editor mistakenly records the morning crossing. On read-back, the contributor corrects that referent and then explicitly confirms that the revised evening-service question captures the intended inquiry for a monthly review note. This supplies evidence for a specific reformulation at a specific time. It does not establish that rescheduling is warranted, that the contributor endorses it, or that every future use of the wording would preserve the same inquiry. A corrected referent and the ensuing confirmation support a narrower conclusion than permanent identity.

A read-back should therefore be scoped, not sweeping. It yields checkable evidence about one wording and one practical purpose, not a certification that the wording fixes a permanent or universal meaning; a later correction remains possible and need not be evidence of bad faith. By the same token, assent to this wording does not give the contributor a veto over every other paraphrase. The worry that treating intent as essential makes comparison unverifiable is well placed if reformulation is judged by private intention alone; the remedy is to make the purpose and referent explicit and to let the confirmed formulation serve as the reference case for later comparison, while keeping its limits in view.

Consider a historical question whose contributor is dead, unreachable, or otherwise unable to respond to a proposed reformulation. Explicit assent is then unavailable, and that absence has consequences. What remains possible is a different and more modest judgment: that a documented interpretation, grounded in surviving records of what the question was and what it was for, can be assessed against competing readings. Such an interpretation can be defended by appeal to the record. It cannot be reported as the contributor's confirmation of new wording.

The distinction matters because the two statuses license different claims. A reformulation endorsed by the speaker carries commitment; a reformulation adopted by an editor carries the editor's reading, however well evidenced. Silence supplies neither. If the contributor never answers, or cannot be reached, the correct entry is not assent but the basis for the interpretation actually made, together with its authorship. Where the record shows the practical purpose the question served, an editor may argue that a candidate preserves it; that argument is a reasoned inference, not the absent contributor’s confirmation of the editor’s wording.

For discovery, a reversible grouping is useful. Candidate reformulations can be provisionally grouped under the recorded question, with each labeled as contributor-confirmed where that exists and as documented interpretation where it does not. The grouping is a working arrangement: it can be dissolved or revised if further material appears, and it does not by itself settle identity. Nor does support require publishing restricted materials; a description may expose the interpretive steps while leaving some supporting evidence unavailable, and that limit should remain visible. This concerns only the narrower identity judgment — whether a candidate counts as the same question when assent cannot be obtained — and not the ownership, access, or publication rights attached to the underlying record.

2.6Partial equivalence and uncertain matches

Consider a hypothetical construction: twelve monthly counts of completed ferry crossings, each event assigned once to a non-overlapping month within a single calendar year, all drawn from the same population, expressed in the same unit, and obtained under one consistent measurement convention. Under these premises the monthly figures partition the year, so they determine an annual total by addition. The premise set matters: if months overlapped, or if the unit or population changed mid-year, the sum would not express the quantity the annual request has in mind.

A small toy calculation shows the asymmetry. Take twelve monthly counts of ten each; their total is 120. Take instead the pair 12, 8 repeated six times; it also totals 120. The two monthly profiles are different, yet both are consistent with the same annual figure. Aggregation here is many-to-one: many monthly distributions yield 120, so the total cannot be inverted to recover any one of them. The annual value is a function of the monthly values, but the monthly values are not a function of the annual value.

This is what directional answer coverage amounts to. A person holding the twelve monthly counts can answer the annual question without further information, because the annual answer is computed from them. A person holding only the annual answer cannot answer the monthly question, because that answer leaves the distribution undetermined. The finer answer covers the coarser one; the coarser answer does not cover the finer one. The two questions are therefore related but not interchangeable: one asks for a decomposition the other deliberately collapses.

For request merging, the practical consequence is concrete. A request for an annual total and a request for twelve monthly counts can be met by the same underlying data collection when the premises of year, unit, population and method coincide, and a response satisfying the detailed request also satisfies the coarse one. Merging them as duplicates, however, would erase the demand for a monthly breakdown that the coarser request never raises. One response can serve both; one requirement cannot absorb the other.

Consider a hypothetical intake collection containing two questions about safe ferry crossings. The first asks whether a scheduled crossing should be cancelled in severe weather; the second asks whether a vessel carries sufficient rescue equipment. Both concern safe passage across water, and that shared topic is a reasonable basis for placing them in a common navigation group. Yet the answers they call for differ: the first concerns a cancellation decision under given conditions, while the second concerns an inventory of equipment and its adequacy. Similarity of subject matter is therefore real, but it does not settle whether the two questions are one question.

A provisional match label records exactly this middle position. It asserts that the pair is close enough to warrant joint review or grouping, while leaving their identity undecided. The label is a claim about relations worth inspecting, not a license to collapse the entries, and it carries no similarity score and implies no expectation of empirical accuracy. Nor does it presuppose that some unique perfect classification exists and merely remains undiscovered. Two questions can sit in the same topical family indefinitely without becoming interchangeable.

What evidence could move the provisional label in either direction? The answer alternatives matter: if their replies express the same relevant propositions about fixed referents, the case for stronger equivalence improves, whereas incompatible reply types count against it. Presuppositions also matter, since a question that assumes a crossing is running differs from one that assumes a vessel is already at sea. Practical purpose contributes as well; a cancellation policy and an equipment audit may serve different operational ends even within one domain. Finally, the commitment sought from the respondent—advice, a decision, a report, or a forecast—can separate questions that share vocabulary.

This is a local judgment from a synthetic pair, not a measured result about any deployed intake system. Its value is modest but concrete: it shows that an uncertain match can be recorded as a reviewable relation rather than resolved prematurely into a merger.

2.7A usable identity record

Consider a hypothetical case, constructed for this discussion rather than drawn from any observed interaction: a commuter asks a ferry attendant at East Pier, "Does the 7:15 ferry stop at Northport?" A later editor who must judge whether some other wording still asks that question needs more than the sentence itself. A lean record suffices, provided it keeps apart what was said, what the question is about, and what would count as answering it.

The entry keeps the original wording and identifies the sailing: a departure from East Pier at 07:15 on a specified Tuesday in the invented timetable. Its answer requirement concerns whether that sailing calls at Northport. An editor proposes “Does Tuesday’s 07:15 departure from East Pier call at Northport?” and records why the named service and destination match. The record also says whether the contributor confirmed that reading, corrected it, or has not responded. These elements expose different possible errors without requiring an account of the contributor’s whole life.

Contrast a proposed replacement asking which dock the ferry uses. That question may be useful, but its answer need not establish whether the sailing stops at Northport. Recording a reason for the replacement would make the editorial choice inspectable; it would not make the choice sound. A later editor could reject the proposed equivalence or retain the new question as a related entry. The record’s value lies in exposing the claim being made and the evidence offered for it.

This is a lean proposal for comparison, not a fixed record that every question must carry. Its detail should match the consequences of a mistaken match. Detailed provenance machinery can be considered separately; here the immediate aim is to make a disputed reformulation intelligible enough to challenge.

When the recorded differences are exhausted by the comparison criteria themselves, proportionate effort does not require going further. Take two hypothetical ferry entries: the same wording, the same named service and date, the same stated purpose of checking a sailing time, and the same source status, namely an operator timetable rather than a recollection. On the record as given, nothing remains to compare that the criteria can distinguish. Grouping the entries for retrieval is therefore defensible, and defensible without speculating about what either speaker privately intended.

That conclusion is deliberately narrow. It licenses a retrieval decision, not the stronger claim that the two speakers asked the identical question in every respect a philosopher might distinguish. The same record also supports a reversible arrangement: keep the underlying entries intact so that a material difference — a revised service, a second date, a newly disclosed purpose — can reopen the comparison later. Preserving that route matters, because the strongest objection to treating intent as essential is practical rather than semantic. If an unrecorded private motive could always veto a match, no comparison would ever be verifiable, and the speaker would hold an unlimited appeal. Grouping on explicit, inspectable features avoids that while leaving the motive question open rather than answering it.

Nothing here turns retrieval convenience into evidence about the world: two entries grouped in an index are still two entries, and the constructed cases show why the selected comparison criteria need to be stated. The open question is which contextual facts, once added to the record, make a simpler comparison inadequate — how service, date, purpose, or source status move, and what changes when they do. That is the next inquiry, and the present criterion makes no claim in advance about its result.