Project roadmap

Archēglyph is a research project, not only a piece of software. It began with a humanities problem — the gaps in the digitised archive of South Asia and the wider Global South — and its roadmap is organised around that problem. The work runs in five strands:

• Research — the humanities questions: what has been digitised, for whom, and what the gaps mean.
• Collections and partnerships — the archives we read, and the institutions that hold them.
• Technology research — reading the page better, above all in the scripts machines read worst, and new ways of finding what a collection holds. It serves the other four strands, not the other way round.
• Community and training — researchers, librarians and archivists, across the Global South and beyond.
• Open infrastructure — keeping the work findable, citable and lasting.

What is in use today works now. Everything under under way, next and later is a plan, and plans change; this page is updated when they do. For the case behind the plan, see research and impact.

In use today

Technology research

• Reading a page in any script. Scanned pages are divided into their parts and read in reading order, with a note of what read each part and when, and uncertain readings marked. See how a page is read.
• Themes across languages. Passages are grouped by what they are about, whatever language they are in, and read as passages rather than as charts. Search works by the words on the page or by what a passage is about. See finding themes.
• The reader. A book, page by page, with the original, a romanisation in a named scholarly scheme and a translation side by side; chapters found, and the people and places in each one listed. See the reader.

Open infrastructure

• Provenance on every line. Every transcription, romanisation, translation and theme title names what produced it. See transparency is a feature.
• Citable editions. Collections are saved as fixed editions, each with its own permanent link, and any edition downloads whole in open formats. See exporting and archiving a collection.
• Public collections and published books, readable by anyone.

Community and training

• Writing for researchers. Essays on method and a glossary of every term, in plain language.

Under way

Research

• Bengali periodicals. The first sustained case study, drawn from Maitrayee Roychoudhury’s ongoing research — the test of whether the approach answers the questions a periodicals scholar actually asks.

Technology research

• Each script on its own terms. Where sentences end, where words break and which way text runs, handled language by language across 34 languages, so that search in Bengali or Urdu works as well as search in English.

Next

Research

• Mapping the archive gap. Which collections of South Asian vernacular print have been digitised, by whom, on what terms and for whom — and what the deliberate gaps on the shelf say about the politics of digitisation. See the research agenda.

Collections and partnerships

• Open repositories. Approaching open-access repositories of multilingual periodicals — such as Heidelberg University Library’s specialised information service for South Asia (FID4SA), the archives of the Centre for Studies in Social Sciences, Calcutta, and British Library collections — to read their holdings where they live. These are aims; no partnership is implied.

Technology research

Each item is a question a researcher asks of a collection, paired with the technique that would answer it.

• Reading the scripts machines read worst. An open evaluation of how well each script is read, starting with Bengali and Urdu, published as it goes — and the basis for improving the weakest.
• Who appears in this collection, and where? People, places and organisations across a whole collection, each mention leading back to its passage. See named-entity recognition.
• Where am I reading the same thing twice? Reprinted, repeated and near-identical passages, flagged. See near-duplicate detection.
• What changes over time? How words, names and themes rise and fall across dated issues — and whether the meaning of a key term drifts across decades. See diachronic analysis.
• Which names keep company? People and places that appear together, theme by theme. See co-occurrence.
• Are there other ways to group this? Alternative views of a collection’s themes, to set beside the ones found today. See topic modelling.
• Who wrote this? Authorship signals, where a collection allows them. See stylometry.
• What does the source actually say about this? Answers given only as quotations from the text, never paraphrased. See extractive question answering.
• How far can I trust this collection? The quality of the reading, set out openly: how confident each reading is, and which language each passage is in. See confidence score.

Community and training

• Workshops and training for researchers, librarians and archivists, beginning with institutions that hold South Asian vernacular collections.

Open infrastructure

• A persistent identifier (DOI) for each edition, so that a collection can be cited like any other scholarly output.

Later

Technology research

• Better reading for non-Latin and historic scripts. Dedicated paths for Arabic-script, Cyrillic, Chinese and historic scripts, where reading is weakest today.
• Rescuing low-resolution scans. An optional step that sharpens pages scanned at too low a resolution — used only where a page genuinely needs it, because sharpening can invent letters, and always recorded in the page’s provenance.
• The language of every passage. Detecting the language of each passage in a mixed-language collection. See language detection.
• Citations as data. Footnotes and bibliographies drawn out, so that the citation network of a collection can be followed. See citation extraction.

Community and training

• Shared collections. Collections that researchers can share, build on and annotate together.
• An international community of researchers across the Global South and beyond, sharing collections, methods and findings.
• Stewardship with source communities. Shared governance of collections with the communities whose print is being read, in the spirit of the CARE principles.

Open infrastructure

• A sustainable model that keeps Archēglyph free for researchers without institutional support.

What we have chosen not to build

We say no to these deliberately.

• Machine-written summaries. A claim a researcher cites must lead back to a passage in the source.
• Free-form chat over a collection. A chat reply hides which passages informed it.
• Synthesised answers to questions. Any answer a researcher cites must point at a passage, not at a paraphrase.
• Replacing digitisation. Scanning is the archive’s work; Archēglyph works downstream of it, on what the archive has already digitised. See downstream of Trove.