International digital library of artifacts inscribed with cuneiform writing
Showing 2 of 2 projects. Click any project card for scope, mentors, and proposal studio.
Mentors: Student: Bakhtiyar Syed
<p>Semantic role labeling (SRL) is a task in Natural Language Processing which helps in detecting the semantic arguments of the predicate/s of a sentence, and then classifies them into various pre-defined semantic categories thus assigning a semantic role to the syntactic constituents.</p> <p>The larger goal of the project is to develop a supervised semantic role labeling system for Sumerian, a Mesopotamian language spoken in the 3rd millennium B.C.</p> <p>Advancing in this direction, we propose to first use projections of English annotations onto Sumerian thus creating gold-standard SRL Sumerian data for the research community to use. We will then evaluate our system using existing architectures available for semantic role labeling with the projected SRL data. The final system also involves the production of word embeddings for Sumerian which can be documented and used for improving other downstream tasks like POS tagging, dependency parsing, etc. The developed SRL system will have many potential applications, viz. in the fields of document summarization, machine translation and also towards a better abstract semantic representation of the originally sparse textual data.</p>
Mentors: Student: Sameer Sengar
<p>The Cuneiform Digital Library Initiative (CDLI) aims to collect, maintain and make available images, text, and metadata of all artifacts inscribed with the cuneiform script. Lots of temporal information can be gathered from cuneiform artifacts, from the dating of the object to specific dates at which events happened in the text. In concert with the ongoing rewrite of the CDLI codebase (“CDLI Framework update”), this project will result in the extension and improvement of the current data model for temporal information, by structuring the data in a new, richer model which will enable the full utilization of relationship among these data. The new structure will be exploited so artifacts can be searched and displayed based on this temporal information and in a granular manner.</p>