Sequence Models for Document Structure Identification in an Undeciphered Script

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 9111 - 9121 December 7-11, 2022 ©2022 Association for Computational Linguistics Sequence Models for Document Structure Identification in an Undeciphered Script Logan Born1 loborn@sfu.ca M. Willis Monroe2 willis.monroe@ubc.ca Kathryn Kelley3 kathrynerin.kelley@unibo.it Anoop Sarkar1 anoop@cs.sfu.ca 1Simon Fraser University School of Computing Science 2University of British Columbia Department of Philosophy 3Università di Bologna Dipartimento di Filologia Classica e Italianistica Abstract This work describes the first thorough analysis of “header” signs in proto-Elamite, an undeci- phered script from 3100-2900 BCE. Headers are a category of signs which have been pro- visionally identified through painstaking man- ual analysis of this script by domain experts. We use unsupervised neural and statistical se- quence modeling techniques to provide new and independent evidence for the existence of headers, without supervision from domain ex- perts. Having affirmed the existence of head- ers as a legitimate structural feature, we next arrive at a richer understanding of their pos- sible meaning and purpose by (i) examining which features predict their presence; (ii) iden- tifying correlations between these features and other document properties; and (iii) examining cases where these features predict the presence of a header in texts where domain experts do not expect one (or vice versa). We provide more concrete processes for labeling headers in this corpus and a clearer justification for ex- isting intuitions about document structure in proto-Elamite. 1 Introduction Proto-Elamite (PE) is a largely undeciphered script of the Early Bronze Age, inscribed on clay tablets unearthed in Iran. PE shares certain features, most notably its number systems, with another ancient script called proto-cuneiform: these similarities have allowed for a partial decipherment which in- forms current understandings of the texts as ad- ministrative accounts recording amounts of various goods and personnel. Figure 1 shows a typical text with annotations explaining how it is divided into columns and entries. Dahl (2019) gives a thorough survey of PE from an archaeological perspective; Born et al. (2019) introduce the corpus to technical audiences and describe initial computer-assisted exploratory analyses. Specialists have hypothesized that PE texts fre- quently begin with a “header”, that is, a sign (or string of signs) which “qualifies all transac- tions recorded in a text” by specifying an insti- tution or owner in charge of the associated ac- count (Damerow and Englund, 1989, 14-16). This understanding of headers depends in part on the claim that they correspond to visually demarcated “colophons” in proto-cuneiform accounts (Englund 2004, 144; Damerow and Englund 1989, 15), which are however also largely undeciphered and therefore not certain to consistently convey owner- ship information. Some (but not all) of the signs that occur at the beginning texts have been tentatively labeled as headers by domain specialists. This labeling is recorded using comments in the transliteration of the texts; no explicit list of header signs has been published. The clearest example of this putative category is the ubiquitous sign M157 , which oc- curs at the start of fully one-fifth of …

Content truncated to 500 words