Discovery

Discover more without changing the original

Six ways a reader reaches material that was, until now, effectively invisible — across languages, across media and across time. MediaINFO Digital Library adds each of them as a layer of understanding over the object, while the object itself stays exactly as the institution received it.

A search field types the question “Who was first to step onto the Moon?” and six threads of light run from it to six doors, each a real object from this archive. The first is a French front page of 21 July 1969, its headline “L'Homme a marché sur la Lune” outlined and rendered in English as “Man has walked on the Moon”. The second is the crew portrait of May 1969, with three faces boxed and named: Neil Armstrong, Michael Collins and Buzz Aldrin. The third is the recording of the first words on the surface, drawn as its own waveform with the moment “one small step” marked at nought minutes and a quarter of a second. The fourth is the same French page again, this time with the name of the sea, Tranquillité, lit where it sits in the standfirst. The fifth is three papers — French, German and Spanish — each printing the same photograph, AS11-40-5903, and each dated 21 July 1969. The sixth is a pair of photographs from the same magazine, shown side by side as a picture match. The six threads then gather into one answer card: Neil Armstrong · 21 July 1969 · 02:56 UTC, with one evidence mark for each door. Each door is a link to the section of this page that demonstrates it.

Search beyond the language of the archive

A query typed in one language reaches material written in another script — and the passage comes back marked as a translation. The box below is the real thing, working on the eight objects this page is built from.

  • Where your administrators enable it for a content type, every page carries machine-translated text in one or more target languages alongside the original recognised text.
  • A query in any supported language is matched against the original text and every translated rendition — an Arabic newspaper surfaces for an English query, a Hungarian leaflet for a French one.
  • Hits are highlighted on the original page layout, in their original positions, whatever language you searched in.
  • A translation overlay swaps the page into the language you chose while line breaks, headlines, blocks and images stay exactly where they were.
  • Translation is for discovery and sense-making, not for citation: the original text is preserved and remains the canonical record.
Arabic daily, front page, 3 December 1971: the announcement of the state of the United Arab Emirates. The line quoted beside it is outlined on the page.
Arabic daily, front page, 3 December 1971: the announcement of the state of the United Arab Emirates. A page in the public domain, shown whole.

Arabic · transcribed and translated

زايد رئيس للإتحاد

“Zayed, president of the union”

Found by the English search “president of the union”: it is matched against the translation stored beside the page, and the hit is lit on the Arabic line it came from. The translation is a layer over the page: the original text is preserved and remains the canonical record.

Try

All 8 items, unfiltered.

  • Map · 1926

    Sketch map of the Trucial Coast and the southern Persian Gulf, cropped to the coast

  • Photograph · 2 December 1971

    The declaration of the establishment of the state read at Union House, Dubai

  • Photograph · 2 December 1971

    The flag of the Union raised at Union House, Dubai

  • Photograph · 2 December 1971

    The Rulers of the Emirates after the proclamation of the Union, Guest Palace, Dubai

  • Front page · Dated 3 December 1971

    “The Trucial Courier”, front page of 3 December 1971

  • Front page · Dated 3 December 1971

    “Le Courrier du Golfe”, front page of 3 December 1971

  • Front page · 3 December 1971

    Arabic daily, front page of 3 December 1971: the announcement of the state of the United Arab Emirates

  • Sound · 12 March 1933

    Fireside Chat 1, on the banking crisis — radio address from the White House

A small local index of the 8 items on this page. The platform searches across the whole archive.

Faces, names and film, searchable together

A photograph, a printed page and a reel of film are three different problems. Below: a face matched in a photograph, a name found in recognised text, and appearances listed per segment in a film — three kinds of media, each searchable.

  • Faces are located in photographs and in video, and where a person record exists they are matched and labelled.
  • A person named once is recognised in later material, and every appearance is listed on that person's record.
  • In video, each appearance is outlined at its own timestamp, and continuous appearances are grouped into one entry per segment instead of one per frame.
  • Names your team enters always take precedence. Where the platform proposes that two person records are the same individual, an administrator reviews the pair side by side, with the similarity shown, before they are combined.
  • Recognition is subject to your own privacy policy: administrators control whether results are displayed at all, and which users may see them.
Two people seated on an official dais in 1932, each face outlined by the recognition layer.Face matched to a person your archive already holdsSecond face detected · unnamed, and ready for a curator
Photograph — faces located and matched to the people your archive already knows. Sam Hood · State Library of New South Wales (public domain).
The front page of a 1932 newspaper, with one line of the lead paragraph — the line naming the Australian premier — outlined as a match in the recognised text.A name in the recognised text of the page — searchable from it
Printed page — a name in the recognised text (here the Australian premier, not either of the people in the photograph beside it), searchable exactly like the catalogued metadata around it. The China Mail, 19 March 1932 (public domain).
A frame of Apollo 11 film: an astronaut climbing down the ladder of the lunar module to the surface, 20 July 1969.

An appearance listed at 00:41–01:12

Moving image — each appearance is listed at its own timestamp and continuous appearances are grouped, so a result opens at the appearance rather than at the top of the reel. Apollo 11, 20 July 1969 (NASA, public domain); timeline illustrative.

Jump to the spoken word

Speech becomes text with timings, so a result is a moment rather than a file. Below is a real recording with its real transcript: press play and the words arrive as they are said — or click any one of them to hear it.

  • Video segments are listed beside the player, so a result opens at the moment it was spoken rather than at the top of an hour of footage.
  • Audio tracks from video and audio files are transcribed automatically, producing searchable captions and subtitles.
  • Transcribed segments appear as timed regions on the waveform; clicking one seeks the player to that point.
  • The full text of a transcription is indexed, so an ordinary search returns the recording — and the viewer highlights the matching phrase in the transcript panel.

0:00.00:13.5Paused at the start of a 13.5-second excerpt.

39 words, each timed to the recording.

Click a word to hear it. With the player held, space plays and pauses, the left and right arrows move a second, and Home and End go to the beginning and the end. Enter in the search box plays from the first word it found.

A 13.5-second excerpt of a real recording made in 1933, transcribed word by word, so a search lands on the moment a word was spoken. The platform does the same across whole collections, of any length.

Ask in your own words

Describe what you are looking for in your own words, and the question becomes a search you can see and correct. The results are the archive's own items, in the order search puts them. It is a search, not a story to take on trust.

  • The filters it applied appear as chips in the active-filters bar — including ones you cannot build by hand, such as excluding a value or choosing either of two — and each can be removed like any other filter.
  • If the question is ambiguous, it asks back with options to click; each turn refines the last, so a follow-up narrows what came before instead of starting over.
  • Names are matched against your own people register, aliases included; dates are read in either calendar, Hijri or Gregorian; words for kinds of material — newspapers, photographs, documents — are matched to the content types your installation actually has.
  • What it does not do is write you an answer. Nothing is summarised at you: the results are the archive's own items, ranked as search ranks them, each one openable and citable.
  • Available where the licence includes conversational search and the role carries the permission; keyword search continues regardless.

Then, on the page it found: every hit lit where it sits.

Try
1.0×

The whole page. 461 words on it are indexed.

An English-language front page, dated Dubai, Friday, 3 December 1971, its headline reading “Six Emirates Proclaim a Union”.

Drag the page to move it; double-click a word to fly to it. With the viewer held, the arrow keys pan, plus and minus zoom, the scroll wheel zooms, Home returns to the whole page and Escape lets it go.

A typeset page, every word of it indexed. On a scanned original the platform does the same from its OCR.

More like this — and the copies

One click from any result to the items that resemble it, and one switch that turns a listing into its duplicates: the same file imported twice, the same photograph cropped and re-scanned, the same passage printed in two places. The control below is the real thing, working on a seven-picture archive and two documents.

  • Every result carries a More like this button that finds the items most like it, judging metadata, picture and text together and showing how close each match is — and a Similar to … chip that takes you straight back.
  • Point at a photograph, a region of a page or a single article and ask for more like that — not only more like the item it happens to sit in.
  • Your search and your filters keep working inside the neighbours: narrow to one year or one collection and the counts describe the matches, not the archive.
  • Find duplicates inside whatever you have searched: the same file imported twice needs no AI at all; the rescans, re-exports and crops of a picture — including a photograph sitting inside a scanned page — are found at the strictness you choose, with mastheads and standing headers set aside and counted.
  • Copied text: documents that share the same words, with the passages and the share of each side marked, so a reviewer reads the match on both originals — a submitted thesis or diploma paper checked for plagiarism against everything the library holds. Available when enabled for an installation.
  • Every member of a group carries its date, so a group of the same story across titles also settles who published it first.
  • Two further ways to compare text, the same story in different words and the same catalogue record filed twice, are in preview.

Find the copies in a small archive

7 pictures · 2 documents
The flag of the Union on its mast at Union House, a crowd in white below it.

The flag of the Union raised at Union House2 December 1971 · flag-1971.webp

The same photograph of the flag at Union House.

Flag raised, Union House, Dubai2 December 1971 · flag-1971 (1).webp

Seven Rulers standing on a carpet before the sea, a flag flying, photographers in front.

The Rulers of the Emirates after the proclamation2 December 1971 · rulers-1971.webp

The same photograph cropped to the Rulers on the carpet.

The Rulers on the carpet, Guest Palace2 December 1971 · rulers-1971-detail.webp

The same photograph re-scanned from a print, warmer in tone.

Rulers of the Emirates, Guest Palace, Dubai1971 · scan_0412.webp

Men at a long table, one reading from a sheet, others crowding behind.

The declaration of the establishment of the state is read2 December 1971 · reading-1971.webp

A 1926 sketch map of the Trucial Coast.

Sketch map of the Trucial Coast1926 · trucial-coast-1926.webp

A working demonstration on the seven pictures and two documents this page holds; the platform does the same across a whole archive.

A page becomes its articles

Layout analysis reads a page into its articles. It answers a different question from text recognition: not what the page says but how it is put together — which block is a headline, which picture belongs to which story, where one article ends and the next begins, and in what order a human reads it. Below are two real front pages, read into their articles.

  • Layout analysis names the regions of a page — heading, text block, byline, caption, picture — and groups them into segments: on a newspaper page a segment is an article, its headline, its columns, its photograph and that photograph's caption held together as one thing.
  • A page that arrives as one large image comes out of import as its articles, each with its own headline, its own text and its own pictures, rather than as one undivided slab of recognised words.
  • Reading order is recorded on the article and preserved everywhere its text is used: in the text panel, in what you copy, in what is indexed, and in what is handed on. It is determined from the page itself, so either reading direction is assembled as written.
  • An article is a unit of its own: Select segment picks the whole article rather than the block under the pointer, the Segments tab lists every article on the item, and search can return the article rather than the whole issue.
  • A page also says what it is: the newspaper's name, the issue date and the issue number are read off the masthead and the dateline and recorded on the item, so a bulk of scanned issues sorts itself into titles, years and numbers without anyone typing a date per page.
  • The date is read in whichever calendar the page carries — Hijri or Gregorian, with Arabic and English month names and Arabic-Indic figures — and the reading in the other calendar is worked out and kept beside it, so an issue dated in one is found by a search in either.
  • Structure that arrives with an import is kept rather than recomputed, and an additional article-segmentation pass is available when enabled for a job.

A front page, read into its articles

3 articles · 1 pictureReading order · left to rightIssue date and issue number read off the pageSelect segment · the whole article
A 1919 front page: a two-line display headline over a standfirst and a pull-quote, then three columns of type, with a studio portrait at the head of the third column.

A working demonstration on two front pages this site sets; in MediaINFO this happens at import, on every page: the zones are found, grouped into articles and given the order a reader reads them in, in either reading direction, and the newspaper's name, the issue date and the issue number are read off the masthead and the dateline, in whichever calendar the page carries, with the other calendar's date worked out and kept beside it. An additional article-segmentation pass is available when enabled.

Enrichment you can audit

An institution cannot stand behind what it cannot check. So everything the platform adds says that it was added.

  • Text recognition, printed or handwritten, in either reading direction, with every word placed on the page.
  • The fields automatic analysis populates — people, places, tags, word clouds — are shown as what they are in the metadata panel, and every one of them can be edited or augmented by an administrator.
  • Saving an item re-runs validation but does not re-trigger analysis. Reprocessing is always a deliberate action, and can be re-run on a single page rather than a whole volume.
  • Automated results aid discovery; they do not replace editorial review. Administrators configure which analysis features are active and which results are published at all.
  • Usage and job history are visible in the admin panel, so an institution can see what was processed, and how much.

Metadata

Title
The Sydney Bridge opened — Wonder of the world
Date
19 March 1932
Description AI
A newspaper front page reporting the opening ceremony of a steel arch bridge.
Tags AI
bridge · ceremony · harbour · engineering
People AI
2 recognised · reviewable
Full text AI
Recognised from the page · searchable and selectable

Illustrative fields; the tag marks a value an automatic source populated.

The stance

AI can interpret the archive. The archive remains the authority.

  1. The original object stays intact.

    Originals can be retained alongside every derivative, and images stored losslessly for preservation-grade holdings.

  2. Intelligence is a layer on top.

    Recognised text, faces, transcripts, translations and descriptions sit beside the object, never inside it.

  3. What was derived is marked as derived.

    AI-populated fields are visible as such, remain editable, and a name a curator has entered is never overwritten by an automatic source.

  4. A translation is a translation.

    The overlay never modifies the asset, and the original text remains the canonical record for citation.

  5. Nothing is published until you publish it.

    Which collections are exposed, which fields are mapped to which standard, and what may be downloaded are all your decisions.

Where the analysis runs is your decision

Configured for on-premise processing, analysis happens on infrastructure you control and nothing — no media, no metadata, no result — leaves your network; configured to use a hosted service, only the material a given operation needs is sent, and the results come back to be stored on your own systems. Each capability is set independently, and for collections under data-residency obligations every one of them, face recognition included, can be pinned to on-premise processing. Your own people stay private to your installation; only public figures are shared.

Ask it a question about your own collection.

A guided demonstration on our demo server, with a few items from your material if you wish.