Piecing together puzzles from the past
CSE researchers are leveraging mathematics to study ancient animal bones
Most jigsaw puzzles include a picture of the finished product on the box to guide you in piecing them together. Now, imagine not having those pictures. Imagine that the puzzles are over a million years old, and the different pieces are all jumbled together.
Such is the process of piecing together broken artifacts from the past. But, a University of Minnesota team—that includes two College of Science and Engineering (CSE) faculty—has come together to speed up that process.
Anthropology Ph.D. student Katrina Yezzi-Woodley has been analyzing and 3D scanning broken animal bones since she was an undergraduate in 2010. Her goal was to find new ways of extracting data that could be applied to samples from a 1.8 million-year-old early human site in Dmanisi, Georgia—which her advisor, College of Liberal Arts professor Martha Tappen, has been studying for more than 20 years in collaboration with the National Museum of Georgia and researcher Maia Bukhsianidze.
After years of work, Yezzi-Woodley struggled to find effective new methods to analyze and reassemble the bones. In 2016, she began working with professor Peter Olver and assistant professor Jeff Calder, both from the School of Mathematics. Their connection came courtesy of her brother, Anthony Yezzi, Jr., who met Olver when earning his Ph.D. in Electrical and Computer Engineering with a doctoral minor in math from CSE.
The CSE researchers are using advanced mathematical techniques like differential geometry and image processing algorithms to help Yezzi-Woodley and her team deduce how the bones were broken, who or what broke them, and how early humans interacted with animals and each other.
Their collaboration is somewhat of a first for anthropologists working with mathematicians. Olver, Calder, and Yezzi-Woodley co-founded the U of M Anthropological and Mathematical Analysis of Archaeological and Zooarchaeological Evidence (AMAAZE) consortium. The group facilitates international collaboration between anthropologists, mathematicians, and computer scientists to further the study of fossils, lithics, pottery, and other remnants of the past.
Math lends a hand
After anthropologists gather specimens from historical sites, it typically takes at least thirty minutes—per artifact—to scan them by hand. The U of M team has innovated a new, much faster way to mass scan bone fragments using computed tomography (CT) scanners at hospitals or imaging facilities. Then, with the help of code written by Calder, the data are automatically organized into separate files.
“Currently, what people do is take a white light or laser scanner and scan something individually,” Yezzi-Woodley explained.
“We are able to scan between 300-800 bone fragments in two hours," she said.
The next step is developing a machine learning algorithm that can identify how the bones were broken. Were they shattered with stone-age tools? Chomped on by a carnivore’s teeth? Did dynamic geological processes break down the bones?
Olver and Calder are working on training the algorithm. Analogy is key here—by purposefully breaking other bones, the researchers have something to compare to when analyzing bone fragments from the Dmanisi site. In order to see how bones break in different settings, Yezzi-Woodley and several anthropology and math students have been breaking them with real stone tools with the help of the anthropology department’s Human Evolution Laboratory. And, the team even feeds bones to spotted hyenas from two different Wisconsin zoos—the Milwaukee County Zoo and Irvine Park Zoo in Chippewa Falls—to imitate how now-extinct large carnivores broke bones, a process anthropologists have been using for years.
The researchers have now amassed 7,000 samples using these methods.
“We have all this data where we know what the answer is,” said Olver, who is the mathematics department head. “The goal is to develop those and test those on the known specimens before we can see what they say about the fossil specimens.”
Once the algorithm is refined, the team can start analyzing the fossils from the site in the country of Georgia. And, they’ll get to work on the second facet of their research—reassembling the broken bones.
Cultivating new collaborations
According to Yezzi-Woodley, the goal of this research is to find out what life was like for very early humans, particularly Homo erectus.
By analyzing surface marks, bone shape, and how the fragments are distributed across the Dmanisi site, anthropologists can learn about the social organization of our ancestors. When did humans begin working together and caring for each other? When did we acquire meat into our diets, and what did that do for brain and social development?
Most anthropologists aren’t usually trained in using highly sophisticated mathematical models, Yezzi-Woodley noted.
“When you start talking about surfaces and advanced mathematics, so much more is available to you, including ways in which you can mess it up,” she said with a laugh.
Because of this, working with the math department is crucial to applying mathematical image processing tools like geometric analysis and machine learning to the anthropology and archaeology fields.
“This is a really big opportunity for mathematics to make a real impact,” Olver said.
“We need the experts in the field, just like when mathematicians started to venture into biology," he said. "Without biologists or health scientists, you’re not going to get anywhere. It’s the same way with archaeology or anthropology. Without the context in that area, you can’t do it as just a mathematician.”
The University of Minnesota team has invented several new processes for data gathering. In addition to the more efficient 3D scanning process, the researchers created a “virtual goniometer,” which uses algorithms and image processing software—instead of the physical protractor-like instrument—to measure the angles of specimens.
The researchers have published one paper on mathematical techniques use for detecting ridges, which is useful for identifying the boundary between the natural surfaces of a bone fragment and the broken surfaces. They are working on several more about data extraction from 3D models. Ultimately, these techniques will be applied to the Dmanisi bones. This work has been funded by a National Science Foundation grant and several U of M grants for the past three years, and the team plans to apply for another grant to support future research.
Learn more about this project and the consortium on the AMAAZE website.
Story by Olivia Hultgren