Thomas Seidl, a computer scientist, is using AI to look for patterns everywhere.Credit: LMU Munich
“Give me your data and we’ll find patterns,” says Thomas Seidl, a computer scientist at LMU Munich. Decoding such patterns reveals secrets hidden in everything from bone fossils to ancient languages. And as academia and industry collect more and more data, there is growing interest in using artificial intelligence (AI) techniques to analyse them.
LMU researchers have been developing machine learning methods for more than 20 years, and now apply them to a broad range of fields. “We work on data mining techniques and machine learning tools to analyse text, networks, multimedia and timeseries from the humanities, life sciences, medicine and engineering,” Seidl explains.
With support from linguists, historians, palaeontologists, mechanical engineers and others, Seidl and his colleagues apply AI to a range of processes, from car manufacturing and language evolution to piecing together artefacts or predicting promising fossil sites. “Strong partnerships with researchers in industry and academia drive our work in different directions,” he explains.
Seidl is also a director of the Munich Centre for Machine Learning, one of five dedicated Machine Learning centres funded by the Federal Ministry of Education and Research. “The centre strengthens our regional, national and international collaborations on fundamental research and we can offer practical training and consultancy services on real-world problems,” he says.
LMU Munich has established an especially strong reputation for digital approaches to the humanities. The Bavarian State Ministry for Science and the Arts recently funded more than 15 professorships in AI across several of the university’s faculties. “LMU encourages researchers in all fields to apply AI approaches to whatever problems they are working on,” says Seidl.
Enrique Jiménez, in the Faculty for the Study of Culture, is using AI to reconstruct Babylonian literature from the first millennium BCE.Credit: LMU Munich
Reading through time
Enrique Jiménez, a professor of Babylonian literature of the first millennium BCE in the Faculty for the Study of Culture, is making the most of this opportunity.
He leads an ambitious AI effort to reconstruct Babylonian literature. The Electronic Babylonian Literature (eBL) Project started in April 2018 and brings together ancient Near Eastern specialists and data scientists to create tools that are transforming the analysis and understanding of ancient literature.
For example, generations of scholars have been trying to put together the works written in cuneiform script — one of the earliest forms of writing in human history. The wedge-shaped characters formed the basis for several languages used from 3200 BCE until the first century AD. Scribes typically pressed a reed stylus into clay tablets. “It is the most durable form of writing ever invented,” says Jiménez.
That durability means hundreds of thousands of tablet fragments survive in museum collections to provide glimpses of everyday life in Ancient Mesopotamia. They form epics, hymns, treaties, private letters, scholarly texts, legal documents, administrative records, and even jokes.
Reconstructing these texts is really difficult. It has taken more than a century for researchers to compile enough pieces to read just four lines from the prologue of the epic of Gilgamesh, regarded as one of the earliest surviving great works of literature. AI is helping them pick up the pace. “Reconstructing ancient literature is very slow and inefficient so we speed it up using machine learning,” Jiménez explains.
Several features of cuneiform script make it particularly difficult to read. Each sign can have different meanings and there are no fixed spelling rules, so every word can be written in a number of different ways, only some of which are explicable or predictable.
Jiménez and his team are assembling a database of fragments, the ‘Fragmentarium’, which so far contains more than16,000 transliterated fragments, equivalent to 200,000 lines of text. Around 60% of these fragments originate from the library of the Assyrian King Assurbanipal in the city of Nineveh, which housed the biggest and most wide-ranging collection of texts in antiquity.
“We are developing algorithms to identify these fragments by comparing them with the Corpus, another database that will contain all the classics of Babylonian literature in the first millennium,” says Jiménez. “The computer learns how words can be written and detects overlaps that are invisible to us.”
Training tomorrow’s data scientists
Students at LMU Munich are heavily involved in research and in the discovery process. “Our students make significant contributions to research and often co-author papers,” says Seidl. “They rarely have a problem finding a good job.”
At LMU Munich, AI is taught at both undergraduate and postgraduate level. And the university offers an elite graduate programme in Data Science in English designed to prepare students to undertake doctoral study or work as data scientists in the public or private sector.
“Here at LMU Munich our students learn and explore how to make AI more robust, reliable and understandable,” Seidl says. “This is vital as the technology is increasingly embedded in decision-making.”