Skip to content
Archēglyph
Menu

Method and ethics

How Archēglyph works, and why

Archēglyph began with a problem met in the archive, not in a laboratory — Maitrayee's account of it is on the about page. What follows is the method in plain terms: what happens to a page, what you can do with the result, and the principles that decide what the tool will and will not do.

Three questions

Digital preservation is political. Repositories of nineteenth-century print control access through paywalls, their holdings lean heavily towards English, and a great deal of vernacular material — digitised, freely available — sits outside the reach of text recognition, and so outside every search. The project sets out to answer three questions:

  1. What does one do with the expurgations happening in the here and now — the deliberate gap on the shelf, driven by choice, not absence?

  2. How do the echoes of empire inform the way archives are digitised today?

  3. How do those choices bear on researchers who lack sophisticated equipment or institutional support?

From page image to text

It starts with the scans you already have — photographs taken in a reading room, PDFs from a digitisation project, pages in any script. Each file is kept exactly as you gave it, with a digital fingerprint that shows it has not changed since.

  1. Bring your scans

    PDFs or photographs of pages, in any script. Each file is kept exactly as you gave it.

    kept as it was

  2. Each page is read

    Part by part, in reading order. Every transcription notes what read it, and when.

    every line signed

  3. Read, search, find themes

    Read one book closely beside its translation, or a whole run of issues by theme.

    fixed as an edition

  4. Cite and share

    Cite a fixed edition with its own permanent link, and download it whole to keep.

From scans to citation in four steps. What matters is what carries across each one: the original file is kept unaltered, every line of text says what produced it, and what you cite is a fixed edition that will read the same next year.

Each page is then divided into its parts — masthead, headlines, columns, engravings, captions — and read in the order a person would read them, so that one column never runs into the next. For clean print a conventional text-recognition reader is used; for faded, damaged or non-Latin print, a model that reads the page as an image. You choose for the collection, and can change the choice for a single document or a single passage.

Every transcription carries a note of what read it and when. Where the reading is uncertain, it is marked for you rather than guessed. Read a passage again with a different reader and both readings are kept — nothing is silently overwritten.

A periodical page divided into seven numbered parts A two-column newspaper page. Its parts are outlined and numbered in reading order: masthead, headline, the first column, then the second column, an engraving, its caption, and the rest of the second column. A dotted line traces the order, down the first column and back up to the top of the second. Part 4 is highlighted as the passage lifted out; part 7 is tinted because the print there is faint. 1 2 3 4 5 6 7

An illustrative nineteenth-century page

Read the way a person reads

  1. 1 Masthead
  2. 2 Headline
  3. 3 First column
  4. 4 Second column
  5. 5 Engraving
  6. 6 Caption
  7. 7 Second column, continued

4 Passage 4, transcribed

The mails from Calcutta arrived on Tuesday by the steamer, bringing news of the rains in Bengal and of the state of the crops in the eastern districts.

Read by machine on 21 September 2026 at 14:32 — named, dated, and always shown beside the scan it came from.

7 Passage 7, faint print

The reading here is uncertain, so it is marked for you to check against the scan rather than quietly guessed.

Archēglyph divides each page into its parts and reads them in the order a person would — down the first column, then back up to the second — so one column never runs into the next. Every transcription carries a note of what read it and when, and an uncertain reading is marked rather than hidden.

Close reading and distant reading

Once the pages are read, the same text serves two kinds of work. The reader opens a single book page by page; the corpus tools let you search and survey hundreds of issues at once. Neither reads the scans a second time, so they cannot disagree about what a page says.

Once, for everything

The pages are read

Close reading

One book, page by page

Beside its romanisation and translation, with its chapters found and the people and places in each one listed — every mention a click from the passage.

Distant reading

A whole run, as themes

Hundreds of issues searched by word or by meaning, and gathered into themes you read as passages, not as dots on a chart.

Digital humanities has always worked at two scales: the book you read closely and the corpus you cannot read cover to cover. Archēglyph serves both from the same reading of the pages, so the book and the corpus never disagree about what a page says.

In the reader, the original is never alone and never replaced. A romanisation follows a named scholarly scheme — IAST for Indic scripts, ALA-LC for Arabic and Persian, Hepburn for Japanese, Pinyin for Chinese — so it can be cited and compared with other scholarship. A translation goes into the language you work in, not only English, and is labelled as the machine's reading it is.

The opening lines of a Bengali poem in four aligned columns: page image, transcription, romanisation, translation.
The page image, as scanned Original transcribed, in its own script Romanisation · IAST machine-written, signed Translation · English machine-written, signed
আমার সোনার বাংলা, আমি তোমায় ভালোবাসি। আমার সোনার বাংলা, আমি তোমায় ভালোবাসি। āmāra sonāra bāṃlā, āmi tomāẏa bhālobāsi. My golden Bengal, I love you.
চিরদিন তোমার আকাশ, তোমার বাতাস, আমার প্রাণে চিরদিন তোমার আকাশ, তোমার বাতাস, আমার প্রাণে ciradina tomāra ākāśa, tomāra bātāsa, āmāra prāṇe Forever your skies, your air, in my heart —

Rabindranath Tagore, Āmār sonār bāṃlā — opening lines, first published in Bangadarshan, 1905.

Hover or tab to a line to see it in every column.

The reader keeps four aligned columns — the page image, the text in its own script, a romanisation in a named scheme, and a translation into the language you work in — so a translation is always read against its source, never instead of it.

Finding themes across languages

Imagine every passage in a collection placed on a map of meaning, where passages that say similar things sit close together — whatever language they are written in. A theme is a dense neighbourhood on that map. This is what the proposal calls semantic clustering, and it is why a theme can gather a Bengali weekly, an English daily and an Urdu paper without translating any of them first.

Archēglyph shows a theme as the passages at its heart, so you read a theme rather than look at a chart of it. A machine proposes a short title for each; the title is signed, and you can rename it. Search works the same two ways — by the words on the page, and by what a passage is about — and the two are combined by default.

Illustrative passages

  • Bengali

    এ বৎসর বর্ষা বিলম্বে আসিয়াছে; ধান্যের অবস্থা আশাপ্রদ নহে।

    Gloss: This year the rains have come late; the state of the paddy gives no hope.

    A Bengali weekly, 1896 · page 3

  • English

    The late setting-in of the monsoon has occasioned grave anxiety for the rice crop in the eastern districts.

    An English daily, 1896 · page 5

  • Urdu

    اس سال بارش دیر سے ہوئی، دھان کی فصل کی حالت تسلی بخش نہیں۔

    Gloss: The rain came late this year; the condition of the paddy crop is not reassuring.

    An Urdu weekly, 1896 · page 2

A theme, found across languages

Late monsoon and the rice crop

title written by machine · rename it

3 passages · Bengali, English, Urdu

  • A Bengali weekly, 1896 · page 3open on the page
  • An English daily, 1896 · page 5open on the page
  • An Urdu weekly, 1896 · page 2open on the page

These passages share no words — not even a script. They are grouped because they say the same thing.

Passages from three periodicals in three languages. Archēglyph groups passages by what they are about, not by the words they happen to share — without translating them first. The grouping is a proposal: the theme's title is machine-written and signed, you can rename it, and every passage opens on the page it came from, where you judge it for yourself.

A theme is a starting point for reading, not a finding. The tool proposes; the scholar disposes.

Provenance: every line leads back to the page

A footnote is a promise that a reader could go and check. Archēglyph keeps that promise for machine-read text: any passage can be walked back, link by link, to the place on the page it came from and to the scan exactly as it was uploaded — without obscuring provenance at any step.

This is also why Archēglyph does not write summaries or answer questions in its own words. Those would put unsourced sentences between the researcher and the source, and no footnote could lead back from them.

  1. Your footnote

    ¹ An English daily, 14 August 1896, p. 5, col. 2.

    quotes

  2. The passage

    “The late setting-in of the monsoon has occasioned grave anxiety …”

    found at

  3. Its place on the page

    Page 5, part 4 — outlined on the page image, one click away.

    cut from

  4. The scan you uploaded

    Kept unaltered, with a digital fingerprint that shows it has not changed since.

Where a machine writes, it signs its name

  • What read the page is named on every transcription.
  • What romanised or translated it is named on every line.
  • What titled a theme is named on the theme.

None of it replaces the source. All of it sits beside it.

Any passage you quote can be walked back, link by link, to the scan it came from — the chain of custody a footnote depends on. Where a machine has written anything along the way, it says so, and names itself.

Collections you can cite, and keep

A collection changes as you work on it — more issues, better readings, new translations. What you cite must not. So a collection is saved as a series of editions, each fixed when it is made, each with its own permanent link, each downloadable whole to deposit in your institution's repository. The work can outlive a grant, a platform, and us.

An example collection, three editions in

  1. Edition 1

    12 March 2026

    Forty issues, 1896–97, read and searchable.

    fixed · its own link

  2. Edition 2

    3 June 2026

    Twenty-two more issues; three faint pages re-read.

    fixed · its own link

  3. Edition 3

    18 September 2026

    Translations into English added beside the originals.

    fixed · its own link

Download any edition, whole

The text, the index that makes it searchable, the themes, and a list of every page with its fingerprint — ready to deposit in your institution's repository.

next A permanent identifier (DOI) for each edition.

Findable
Every edition has its own permanent link.
Accessible
Published editions are public, and any edition downloads whole.
Interoperable
Open, documented formats — nothing locked inside Archēglyph.
Reusable
Each edition records how it was made: what read, translated and grouped it.
A collection grows; what you cited does not change. Each edition is fixed when it is made, has its own permanent link, and downloads whole — so the work can outlive a grant, a platform, and us. Read against FAIR, it is already findable, accessible, interoperable and reusable; a DOI for each edition is next.

Ethics

Bias, stated plainly

The machines that read text learnt mostly from English and from Latin script, and it shows: clean English print reads best, and faded Bengali or Urdu print — the print of much of the Global South — reads less well. We do not hide the difference. Uncertain readings are marked, every reading names its reader, and a page can be read again with another. Reading non-Latin scripts better is part of the work ahead.

Decolonising the data

The digital record was built largely in and for the Global North. Making the vernacular print of the Global South searchable changes who can be found in it — and who can do the finding, since a web browser and a photograph of a page are enough. The original script is never replaced by its romanisation or its translation; romanisations follow named scholarly schemes rather than improvised spellings; and translation goes into the language the researcher works in, not only into English.

FAIR and CARE — where we stand

The FAIR principles ask whether research data can be found and reused; the CARE principles ask who benefits and who holds authority over it. Here is an honest account of both, including what is not done yet.

FAIR
Principle Today Not yet
Findable Every edition of a collection has its own permanent link; published collections are listed publicly. A persistent identifier (DOI) for each edition.
Accessible Published collections open to anyone. Any edition downloads whole. —
Interoperable Open, documented formats. The text travels with its references back to each page. —
Reusable Each edition records what read, translated and grouped it. Published collections carry a licence chosen by their owner. —
CARE
Principle Today Not yet
Collective benefit Free during the pre-release period, and designed for researchers without institutional access. A long-term model that keeps it free for those who need it most.
Authority to control Collections are private until their owner publishes them. Owners can download, unpublish or delete them at any time. Shared stewardship of a collection by the community it comes from.
Responsibility Every derived line names what produced it; uncertain readings are marked, not smoothed over. Working with the communities whose print is being read — part of the research agenda.
Ethics No machine-written summaries or answers. No trackers. Outside services are named. —

Privacy

Your files are stored on servers we run, and there are no trackers on this site or in the tool. Some work is done by models hosted by an outside service — dividing a page into its parts, and reading difficult print. When it is, the page is sent to that service for that request. The privacy page names the service and its terms, and an offline alternative exists for regular layouts and clean print.

What we chose not to build

No chatbot, no machine-written summaries, no synthesised answers. Each would place unsourced sentences between the researcher and the source. Nor is Archēglyph a replacement for digitisation: it works downstream of the libraries and projects that scan, and depends on their work.

Frequently asked questions

Which languages and scripts can Archēglyph read?
South Asian scripts such as Bengali, Hindi, Marathi, Sanskrit, Tamil, Punjabi, Odia and Gujarati; Arabic, Persian, Urdu and Hebrew; Chinese, Japanese, Korean and Thai; Cyrillic, Greek, Armenian, Georgian and Ethiopic; and Latin-script languages. Clean English print reads best, and uncertain readings are marked for you to check.
Do I need special equipment or an institutional subscription?
No. Archēglyph runs in a web browser and reads PDFs or photographs of pages. It is free during the pre-release period.
Can I cite what Archēglyph gives me?
Yes. Every passage leads back to its place on the page image and to the scan as you uploaded it, and every collection is saved as a series of fixed editions, each with its own permanent link.
Does Archēglyph write summaries or answer questions about my sources?
No. Machines read, translate and group; they do not write about your sources. What a machine does produce — a transcription, a romanisation, a translation, a short theme title — is signed with what produced it and sits beside the source, never in place of it.
Can it translate into languages other than English?
Yes. Translation goes into the language you work in, not only English, and romanisation follows a named scholarly scheme such as IAST, ALA-LC, Hepburn or Pinyin.
What happens to my files?
They are kept unaltered on servers we run. Collections are private until you publish them, and you can download or delete them at any time. Some pages may be sent to an outside service for reading; the privacy page names it.

For the technically curious

Everything above has a technical account, written for the researcher who wants to know exactly what the machine did.