Can we train a machine to translate Egyptian hieroglyphs? This is the question the independent production company Psycle explored in a Ubisoft research project. Taking the facsimiles created by Egyptologists as our starting point, we created a processing pipeline that begins with photographs of hieroglyphs in situ, and ends with possible translations in a modern language. The pipeline uses computer vision techniques to find candidate glyphs and glyph groups in an automatically generated facsimile. Next, neural nets are used to recognise the candidate glyphs and, finally, sequences of glyphs are compared to dictionaries of glyph words in order to find possible translations.
The lecture discusses not only the technical challenges of the project, but also highlights the limitations of machine learning, and the critical importance of the amount and consistency of data for enterprises of this type. The engagement and support of the academic community, which was crucial in giving direction and substance to the project, is also discussed.
The project has generated a research and teaching tool for Egyptologists, an open-source repository of code, and several avenues for further research and collaboration.
Please register until 07.02.2019 under the following Link