Enhancing Sumerian Lemmatization by Unsupervised Named-Entity Recognition

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 1446–1451, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Enhancing Sumerian Lemmatization by Unsupervised Named-Entity Recognition Yudong Liu, Clinton Burkhart, James Hearne and Liang Luo Computer Science Department Western Washington University Bellingham, Washington 98226 {yudong.liu@-burkhac@students.-james.hearne@-luol@students.}wwu.edu Abstract Lemmatization for the Sumerian language, compared to the modern languages, is much more challenging due to that it is a long dead language, highly skilled language experts are extremely scarce and more and more Sume- rian texts are coming out. This paper describes how our unsupervised Sumerian named-entity recognition (NER) system helps to improve the lemmatization of the Cuneiform Digital Library Initiative (CDLI), a specialist database of cuneiform texts, from the Ur III period. Ex- periments show that a promising improvement in personal name annotation in such texts and a substantial reduction in expert annotation ef- fort can be achieved by leveraging our system with minimal seed annotation. 1 Introduction Because the Sumerian cuneiform writing system is historically the earliest, Sumerian culture is the ear- liest recorded civilization. The large number of clay tablets that have been recovered from Mesopotamia reveal “an almost obsessive concern for the preser- vation of daily events of the time: the digging of ditches, the care of livestock, the storage of grain, and so on. Their survival allows insight into the lives of the city dwellers of remote an- tiquity” (Garfinkle, 2012). Today, most cuneiform texts are held in public institutions, but the texts are widely separated both from each other and of- ten from scholars by great distances and expen- sive journeys. Current projects like the Digital Li- brary Initiative (CDLI, 2014) and the Database of Neo-Sumerian Texts (BDTNS, 2014) aim to provide scholars immediate access to virtual collections of tens of thousands of texts. The Ur III period (2112-2004 BCE) is particularly abundant in surviving texts. Because this era was the specialty of our principle informant, an Assyriolo- gist at our home university, we focus on the tablets that are from this era. The vast majority of these tablets record financial transactions, such as records of cattle deliveries, receipt of metals, repayment of loans, and so forth. Figure 1 shows a tablet from the CDLI repository. For expository purposes, we arranged the original cuneiform drawings on the left (which are not input to our computations), with its transliteration (a one- to-one transcription of signs in a cuneiform text to computer readable text) in the middle, and the mod- ern English translation on the right. The original CDLI data includes transliterations in ASCII format and inline lemmatization markup. More detail about CDLI data will be introduced in Section 2. As we can see in Figure 1, in addition to the provider and recipient of transference, tablets con- sistently enumerate lists of witnesses (“sealed by”). This fact makes the tablet an invaluable resource for the social history of the time since they record, implicitly, on each tablet, lists …

Content truncated to 500 words