- eBL Coding Standards
- Setting up Gitpod with Ona
- How
withDataWorks - Contribution guidelines
- How to run migrations
- Interactive Runner for ATF Importer
- Adding References to eBL
- Rules for Normalizing Names in Cuneiform Texts
- Editorial conventions (Corpus)
- Editorial conventions (Library)
- Adding Dossiers
- TEI Export
- eBL ATF and other ATF Flavors
- Archaeology and Findspot Properties
- Manual for Mapping Cuneiform Tablets
The "Electronic Babylonian Library" (eBL) platform, based at Ludwig Maximilian University of Munich (LMU) and the Bavarian Academy of Sciences (BAdW), involves the digital compilation and analysis of Babylonian texts using AI and computer vision technologies. This project leverages several advanced techniques to identify and match manuscript fragments.
The algorithms and techniques used in this project include:
-
Image Matching and Fragment Identification: The eBL project uses computer vision techniques to digitally reconstruct cuneiform tablets from scattered fragments. This involves capturing high-resolution images of the fragments and using pattern recognition algorithms to identify matching pieces based on the cuneiform signs and their arrangement. This process is enhanced by custom-built software applications specifically designed for Assyriological research.
-
N-gram Matching for Text Overlap: The project employs n-gram matching algorithms to identify overlapping text segments within the corpus of transliterated texts. This method involves breaking down the text into n-grams (substrings of n characters or words) and comparing these across different fragments to find overlaps that indicate they are parts of the same original document.
-
Optical Character Recognition (OCR) for Cuneiform: OCR technology adapted for cuneiform script is used to digitize and transliterate the texts from the high-resolution images of the tablets. This process involves training machine learning models to recognize and interpret the ancient script accurately.
-
Machine Learning and AI: Advanced machine learning algorithms are utilized to analyze the patterns and features of the cuneiform signs, facilitating the identification and classification of fragments. This helps in piecing together the fragmented texts and improving the accuracy of the digital reconstructions.
The project leverages Optical Character Recognition (OCR) and Natural Language Processing (NLP) technologies to read and match the texts. Specifically, using OCR to convert the cuneiform signs from images into machine-readable text. Then apply algorithms to detect and match overlapping segments of different manuscripts, aiding in the reconstruction of fragmented texts.
For more detailed information, you can explore the project’s official website and the related publications from LMU and the International Association for Assyriology, which provide insights into the methodologies and technological innovations employed in the eBL project and IAAssyriology.
To date, thousands of additional cuneiform fragments have been photographed in collaboration with the British Museum in London and the Iraq Museum in Baghdad and CDLI.
eBL GitHub organisation includes various tools, such as:
- ebl-api: The API for accessing the Electronic Babylonian Literature database.
- cuneiform-ocr: Tools for performing OCR on cuneiform texts.
- ngram-matcher: Algorithms for matching n-grams to identify overlapping text segments.
These tools collectively enable the identification and matching of text fragments by comparing pixel patterns and character sequences, allowing researchers to piece together ancient manuscripts from scattered fragments like a text puzzle.
Current cuneiform OCR systems (see e.g. here) achieve limited accuracy (~45%) due to visual complexity, damage, and sign ambiguity. This task aims to incorporate Natural Language Processing (NLP) models into the OCR pipeline to improve recognition by leveraging linguistic context. For example, a language model trained on known Akkadian texts can suggest probable sign sequences, disambiguate visually similar signs, and correct errors based on syntactic and semantic constraints. The integration can be implemented as a post-processing step, where OCR outputs are refined using probabilistic language models or transformer-based architectures. This hybrid approach—combining computer vision with linguistic priors—is expected to significantly boost accuracy, especially for literary texts, and accelerate the digitization and reconstruction of fragmentary tablets.
This task involves creating a phoneme synthesis module capable of generating accurate spoken pronunciations for Akkadian words, based on their IPA transcriptions (e.g., abālu → /a'ba:lu/), already implemented (see e.g. SB Gilgamesh I, with the Setting Phonetic transcription). The module must support glottalized consonants (e.g., ᵵ, ᵴ) and extra-long vowels (e.g., â = /a::/). Possible implementations include adapting existing libraries like espeak-phonemizer or leveraging cloud services such as Amazon Polly (using Arabic voice profiles for closest phonetic approximation) or Azure Cognitive Services (with customizable IPA support). The synthesized audio will be integrated into the eBL dictionary interface, allowing users to hear reconstructed pronunciations, thereby enhancing accessibility and aiding in the study of Akkadian phonology and metrics.
The goal is to develop a computational tool that systematically analyzes the metrical structure of Akkadian verse, with a focus on the so-called clausula accadica, the tendency for verse-final words to have a long, stressed penultimate syllable. The tool will scan annotated poetic texts and quantify adherence to metrical rules across the eBL corpus. It will also detect and categorize exceptions, straddled caesuras, and syllabic patterns within hemistichs. Parallelism, already marked at the line level, will be extended to word-level annotation, enabling the compilation of a Dictionary of Babylonian Parallelism. By processing large volumes of text, the analyzer will identify previously unnoticed rhythmic regularities and strophic patterns, offering empirical insights into Akkadian prosody and supporting philological research.
The texts in the eBL Corpus have been manually scanned (see e.g. SB Gilgamesh I, with the Setting Meter), but the scanning offered is not the only one possible.
For more information about eBL project, read this article on Medium and Prof Enrique Jiménez YouTube speech.