The Loracrafft Project

image (c) Antoine Morandi

 


The Loracrafft Project, launched in 2022, aims to enable the translation, using a smartphone or tablet, of texts written in Egyptian hieroglyphs from the Middle Kingdom word by word and not sign by sign, as some tools already do today (read our article of January 21, 2025). But the need to make available to the device software a corpus that will essentially be composed of dictionaries, plus a large number of reference texts containing transliterations and translations, requires the choice of storing all this data on a remote server with large storage capacities, hence the choice we made to set up a client-server architecture, an architecture that we present below:
 


The project is divided into batch jobs, guided by the numerous studies already carried out on the subject:
 

Batch 1: reading, recognition, identification, Gardiner classification and transliteration of signs

Batch 2: translation

Batch 3: documentary research in the corpus to propose extracts of texts containing the translated words
 


Batch 1 would contain 8 phases:

phase 1: availability of source texts

  • mural painting

  • mural engraving

  • stone engraving

  • papyrus painting

  • ostraca painting

  • wood painting

  • photos

phase 2: reading of the source text by the smartphone or tablet

phase 3: recognition of indicator signs for the reading direction

phase 4: recognition of signs in a grouped way (quadrats...)

phase 5: grammatical classification

phase 6: word division

phase 7: conversion of words into Gardiner codes

phase 8: transliteration
 


Batch 2 consists of:

phase 9: translation into French/English/German

phase 10: display and audio reading (optional) of the resulting text
 


Batch 3 consists of:

phase 11: detection of texts from the corpus containing the recognized words

phase 12: display/reading of the text(s) in reference to the request
 


The idea behind the design of this application is to federate already existing "building blocks" to carry out some of the phases above:

For phase 4, we are thinking of the Tomb Reader tool by Morris Franken and Jan van Gemert

For phases 5 and 6, we are thinking of following the work of Serge Rosmorduc

For phases 3 and 7, we are thinking of the Hieroglyphs AI tool by Evgeniy & Alexander Sulimov

For phase 8, several data files exist, including the one provided by Raymond Monfort

For phase 9, we are thinking of using for the French version the Dictionnaire des hiéroglyphes by Yvonne Bonnamy, (c) 2013 Actes Sud, and the appendix "Lexique égyptien-français" of the work by Jean-Pierre Guglielmi, L'égyptien hiéroglyphique, (c) 2021 Méthode Assimil.

Phase 10 will be carried out by the reading equipment.

For phases 11 and 12, a complex reflection is underway because the project intends to allow the server software to consult numerous lexical databases geographically distributed everywhere (TLA, VégA, Ramses, etc.) to train its neural network, and a merger of all these corpora seems unrealistic. Would remote questioning be possible, or even beneficial? We have launched consultations. We'll see.

We are looking for academics who have a complete mastery of Deep Learning, and in particular the programming of convolutional neural networks with attention mechanisms for lots 1 and 3, and computer-assisted translation for lot 2. Should you be interested, please get in touch.
 

[home]

page updated on 2025-02-02 09:31