ISC Digital Humanities
A research-tools directory for the ISC Digital Humanities platform — what's ready to use today, what's partially working, and what's planned.
14 capabilities ready to use nowGrouped by the kind of work each capability supports. Status reflects what is actually implemented and configured on this deployment, not a roadmap promise.
The tools below are stages of one workflow, not separate products. Each stage hands its output to the next.
Upload
Add a manuscript scan, PDF or image to a collection you can edit.
Extract text
Turn the page image into text with OCR. Handwriting recognition (HTR) is a separate step and runs only where a deployment has enabled it; the capability card states the current status.
Review & correct
Check the extracted text against the page and fix it. Nothing is published until a human has read it.
Translate & analyse
Translate the corrected text, or run frequency and readability analysis over it.
Extraction, translation and analysis run from the page of an individual document. Open a collection to reach one.
Extract text from images using the selected free or paid mode. Preserve the original image and machine reading; review uncertain words and source-supported corrections.
Historical OCR benchmarks belong to the exact sample, model and version that produced them. They do not establish the accuracy of the currently selected API. Review this page against its image; uncertain readings remain drafts.
Handwritten and calligraphic text requires review against the image. No accuracy guarantee applies across scripts or historical hands; unsupported pages must be reported explicitly.
Handwriting recognition is switched off for this site.
Real, local, deterministic computational analysis of transcribed texts — word frequency, n-grams, keyword-in-context concordance, and collocations, plus corpus-level statistics (token/type counts, lexical diversity). No AI provider or credential involved; open any item's Text Analysis tab to run it against its real transcription text.
Vector/hybrid ranking layered on top of keyword search. One pgvector index serves both the AI Research Assistant's retrieval and Explore: once an item has been indexed — via the “Index for semantic search” action on the item itself — the first page of Explore's results is reordered by meaning rather than keyword score alone. Items that have not been indexed still surface normally through keyword search.
On-demand translation of item transcriptions between English, Persian, and Arabic, through DeepL, a self-hosted LibreTranslate, or an AI-chat fallback. A translation is stored against the exact version of the original it was made from, so a later correction to the transcription marks the translation out of date instead of leaving a silent mismatch.
Named-entity recognition, sentiment classification, and key-phrase extraction over item transcriptions in English, Persian, or Arabic. Analysis runs only on transcriptions a reviewer has verified, and where the configured analyser cannot handle the text's script it reports that it did not run rather than returning an empty or default result.
A graph of relationships between researchers, institutions, projects, and labs. The interactive force-directed graph (pan, zoom, drag, hover, click-to-focus) is already built and renders live on the Network page from real database records — it just needs published ISC records with linked entities to show more than an empty graph.
Mapping historical places, routes, and spatial relationships. An interactive map (OpenStreetMap tiles, no API key required) is built and renders live on the Network page, plotting every published place with real coordinates from the real GeoJSON API — it just needs more published places with coordinates to show more than a handful of markers.
A conversational assistant that grounds its answers in retrieved ISC sources and says so when it has none. It needs the AI capability enabled and a provider configured. A deterministic offline demo mode is also available, always labelled as a demo and never presented as a live provider response.
These tools are the stages of one research workflow — from a page image to readable text, then translation and analysis. Every status above is read from this deployment's real configuration, the same capability matrix /api/health reports.
