Two archive posts may both mention “speech synthesis in the 2000s” while answering different questions. One could explain why statistical models changed synthesis methods; the other could examine how researchers evaluated systems beyond a convincing audio demo. Merging them because the titles overlap would erase that distinction. But if two short posts both define hidden Markov model synthesis, discuss the same conference papers, and end with the same limitations, readers may have to piece together a single explanation.
The question is whether the pages serve the same reader need with the same evidence, not whether they share keywords. A stronger page gives readers a coherent answer without blurring differences between methods, dates, datasets, or claims.
Identify the overlap before editing
Make a comparison sheet for each candidate article: its central question, intended reader, time period, main examples, sources, and distinctive claims. Compare the sections, not just the titles or search results. A broad post on document analysis and a focused account of early optical character recognition may mention the same systems, while the focused account supplies technical detail the broader post cannot carry.
Try writing a one-sentence promise for each page: “After reading this, a visitor can explain…” If the promises and the evidence needed to fulfill them are much the same, consider a merge. If one page explains a mechanism and the other evaluates an application, give them clearer boundaries instead.
- Strong consolidation candidate: Both pages answer the same question, repeat most of the explanation, and lack distinct bodies of evidence.
- Possible consolidation candidate: Their introductions overlap, but each has a substantial section that would strengthen a unified account.
- Keep separate: Shared terms conceal different methods, historical periods, research settings, or reader tasks.
An article about language identification in short speech clips, for instance, is not interchangeable with one about identifying a document’s language. Both may discuss low-resource languages, but audio duration, transcription, writing systems, and evaluation data pose different problems. A generic “language identification” page would make the archive less precise.
Look for evidence that readers encounter the same answer twice
Editorial inspection should lead; usage evidence can test the judgment. Queries that bring readers to both pages, similar questions in correspondence, and repeated paths between the pages may suggest that visitors see them as alternatives. None proves a merge is needed. Someone moving from an overview to a technical case study may be following exactly the reading path you intended.
Search performance needs the same caution. Two pages appearing for a similar query may serve different levels of expertise. A broad introduction can be useful even when a specialist page gives the deeper answer. Nor does weak traffic make historical material expendable: an obscure conference report may contain the archive’s clearest account of a particular dataset or evaluation decision.
Ask whether readers find what they came for, rather than treating page views as a verdict. How to Measure a Historical-Technical Archive Beyond Page Views examines the difference between traffic and more informative archive measures. For a merge decision, look at whether visitors move to another page for a missing explanation, search again for a term the first page covered, or cite both pages for the same claim.
Choose the right relationship between the pages
Merge when the material forms one answer
Merge when readers need the best material from both articles to understand one subject and the split creates needless repetition. Suppose one short post explains the premise of early stereo reconstruction, while another repeats that premise before discussing failures in poorly textured scenes. A single page could move from the matching problem to the limits of matching without making readers assemble the argument themselves.
Keep the new scope in check. “Stereo reconstruction research in the mid-2000s” may support a coherent account; “the history of computer vision” is too broad a destination for two narrow posts.
Keep separate when the distinction teaches something
Related articles sometimes earn their separate places by distinguishing concepts readers might otherwise conflate. In medical imaging research, computer-aided detection is not necessarily autonomous diagnosis. Clinical roles, evaluation measures, and claims differ. If one article explains a system’s research performance and another addresses clinical use, that boundary helps readers interpret the evidence. It is also central to Medical Imaging AI in the 2000s: What CAD Could—and Couldn’t—Do.
Articles about successive conference years may need to stay separate if terminology, benchmark data, or reported results changed. A continuous narrative can accidentally suggest that researchers knew in an earlier year what became clear only later. When chronology explains the difference, retain the dated accounts and show how they relate.
Revise without merging when the problem is unclear positioning
Sometimes two worthwhile pages sound alike because their titles and openings are vague. Give each a specific title, establish its scope early, and cut background that does not serve its main question. An overview can define a method briefly and point to the technical account; the technical account need not repeat the whole overview. That may work better than making one page too long for either purpose.
Build the stronger page from claims, not paragraphs
Once a merge is justified, do not paste article A above article B. Duplicate definitions and awkward transitions usually survive. Outline a new answer to the reader’s question, then place individual claims where they belong. Note the source behind each claim and the date or research setting it describes. In historical-technical writing, “worked well” might mean only that a system performed well on a limited test corpus.
- Inventory the distinctive material. List explanations, examples, figures, caveats, and source references found in only one article.
- Set a narrow central question. Answer it in a logical order instead of collecting every related fact on one page.
- Check the evidence again. Keep quotations, dates, system names, and descriptions of results attached to the right studies.
- Remove duplicated setup. Retain one accurate definition and use the freed space for distinctions, limitations, or a better example.
- Read for chronological consistency. Do not put later terminology in an earlier researcher’s mouth unless you identify it as a retrospective description.
One workable structure starts with the research problem, moves through competing approaches and their evaluation, and ends with what the evidence did not establish. That may serve readers better than the original posts’ publication order. It also reveals gaps: if neither post explains why a benchmark mattered, the new page still needs that context.
Preserve context that would otherwise disappear
Small details can change the meaning of a result. A post might specify that a finding came from clean laboratory speech rather than noisy recordings; another might distinguish scanned document images from born-digital text. Keep such details when they affect how readers should interpret a claim, even if you remove the paragraphs around them.
Keep attributed disagreements visible as well. If contemporary researchers meant different things by “reliable,” say whose criterion you are describing and in what setting. If an older post records what was known at a particular moment, it may be better to retain it as a dated account and publish a separate synthesis. The archive should not appear to have held a single, unchanging view all along.
Before retiring a page, check whether other posts cite one of its distinctive passages. Readers following that reference should reach the corresponding evidence in the new version. A brief note on the surviving page can say when the material was combined and what period the revised account covers, rather than presenting the synthesis as though it had always existed.
Handle the retired page deliberately
Publishing the combined article is only part of the job. Keep the existing address that best describes the new scope, not automatically the oldest or busiest one. If neither fits, a new descriptive address may be clearer. Where site tools permit, redirect retired addresses to the most relevant surviving page and update internal references so readers need not depend on redirects.
Redirect only when the destination covers the retired page’s central subject. If a post about a particular speech corpus disappears but its replacement barely mentions that corpus, the redirect hides a loss. Preserve the specialized material, expand the destination, or leave the original post available with a clearer scope.
Finally, inspect the result as a reader would. Does the surviving page answer the original questions? Are dated claims still in context, and do references from other articles lead to the right section? Record the old and new titles in an editorial log. Then pick one distinctive claim from each retired article and find its supported, properly dated counterpart on the surviving page. If either is missing, the merge is not finished.