Abstract
Thousands of first-millennium BCE ivory carvings have been excavated from Neo-Assyrian sites in Mesopotamia (primarily Nimrud, Khorsabad, and Arslan Tash) hundreds of miles from their Levantine production contexts. At present, their specific manufacture dates and workshop localities are unknown. Relying on subjective, visual methods, scholars have grappled with their classification and regional attribution for over a century. This study combines visual approaches with machine-learning techniques to offer data-driven perspectives on the classification and attribution of this early Iron Age corpus. The study sample consisted of 162 sculptures of female figures. We have developed an algorithm that clusters the ivories based on a combination of descriptive and anthropometric data. The resulting categories, which are based on purely statistical criteria, show good agreement with conventional art historical classifications, while revealing new perspectives, especially with regard to the contested Syrian/South Syrian/Intermediate tradition. Specifically, we have identified that objects of the Syrian/South Syrian/Intermediate tradition may be more closely related to Phoenician objects than to North Syrian objects; we offer a reconsideration of a subset of Phoenician objects, and we confirm Syrian/South Syrian/Intermediate stylistic subgroups that might distinguish networks of acquisition among the sites of Nimrud, Khorsabad, Arslan Tash and the Levant. We have also identified which features are most significant in our cluster assignments and might thereby be most diagnostic of regional carving traditions. In short, our study both corroborates traditional visual classification methods and demonstrates how machine-learning techniques may be employed to reveal complementary information not accessible through the exclusively visual analysis of an archaeological corpus.
The Classification Challenge
For over a century, archaeologists have struggled to classify thousands of ivory carvings found in Neo-Assyrian palaces in modern-day Iraq. These luxury items were imported from the Levant (modern-day Syria, Lebanon, and Israel), but their exact origins and specific workshop traditions remain a subject of intense debate. Traditionally, scholars have relied on subjective visual analysis to sort these artifacts into three primary traditions: Phoenician, North Syrian, and a contested middle ground known as Syrian/South Syrian (or "Intermediate").
Because traditional visual analysis is prone to observer bias and often results in "fuzzy" boundaries between styles, this study sought a data-driven, objective method to test these classifications and discover hidden relationships within the corpus. By applying unsupervised machine learning, the researchers aimed to validate existing intuition while uncovering patterns invisible to the naked eye.
Methodology: A Mixed-Mode Approach
The researchers utilized a specialized dataset of 162 ivory sculptures of female figures (including sphinxes), primarily from Nimrud, Arslan Tash, and Khorsabad. To analyze these objects, they employed a "mixed-mode" unsupervised clustering approach:
- Feature Engineering: 33 features were defined for each sculpture—21 categorical (e.g., hairstyle, headdress type, eye shape) and 12 anthropometric (ratios of facial measurements like lip thickness to length).
- Probabilistic Mixture Model: An Expectation-Maximization (EM) algorithm was used to group the objects. This "soft" assignment provides the probability of an object belonging to a specific cluster, acknowledging the inherent ambiguity in ancient art.
- Handling Sparsity: Given that ancient artifacts are often damaged, the dataset had approximately 37% missing values. The authors used a simplified model assuming feature independence to prevent overfitting on this sparse data.
- Model Selection via "Oracle": To determine the optimal number of clusters ($K$), the team created a hand-curated "oracle"—a set of ivory pairs known to belong together (e.g., figures carved on the same physical object). They found that $K=4$ provided the best balance between statistical likelihood and archaeological relevance.
Key Insights and Results
The algorithm’s clusters showed high agreement with traditional art historical style-groups, effectively validating decades of subjective scholarship. However, the study also produced several "aha moments":
- The Phoenician-Intermediate Link: A major discovery was that Syrian/South Syrian (Intermediate) objects clustered more closely with Phoenician works than with North Syrian ones. This suggests their production origins might be geographically closer to Phoenicia (South Syria) than previously hypothesized.
- Diagnostic Features: The study quantified which features are most "informative" for classification. The shape of the hairline and the form of the eyes emerged as the top predictors. Interestingly, ancient Near Eastern literature frequently praised these specific features as markers of beauty, suggesting the algorithm captured the same aesthetic priorities as the original carvers.
- Regional Workshops: Cluster 4 was found to consist almost exclusively of Syrian/South Syrian ivories from Arslan Tash, confirming a distinct regional "hand" or workshop tradition at that site.
Conclusion and Future Directions
While the sample size is relatively small for machine learning, the study demonstrates that statistical clustering can both corroborate and extend traditional archaeological methods. Future work may involve applying Bayesian non-parametric models to automatically infer cluster numbers and expanding the dataset to include male figures and maker's marks. By integrating this data with other media, such as metalwork, researchers hope to create a broader "social network analysis" of ancient Near Eastern craftsmanship.
Summary updated