Proceedings of the Twenty-Ninth International Florida Artificial Intelligence Research Society Conference Recognizing Proper Names in UR III Texts through Supervised Learning Yudong Liu, James Hearne and Bryan Conrad Computer Science Department Western Washington University Bellingham, Washington 98226 {
[email protected]@-bconrad@students.}wwu.edu Abstract names on the tablets suggests the possibility of reconstruct- This paper reports on an ongoing effort to provide computa- ing social networks of actors in the mercantile class and also, tional linguistic support to scholars making use of the writ- given the overlap, their social network connections to roy- ings from the Third Dynasty of Ur, especially those trying alty. Use of the tablets for such purposes is impeded in three to link reports of financial transactions together for the pur- ways. First, they are by and large written in Sumerian, a lan- pose of social networking. The computational experiments guage already going out of use as the language of the streets presented are especially addressed to the problem of identify- at the time the tablets were being written and which, even ing proper names for the ultimate purpose of reconstructing after nearly two centuries of scholarly investigation, is ill- a social network of UR III society. We describe the applica- understood. A casual exploration conducted by us showed tion of established supervised learning algorithms, compare that roughly half of the lexical items found in the UR III Cor- its results to previous work using unsupervised methods and pus are hapax-legomena, words having but a single attesta- propose future work based upon these comparative results.