Linguistic annotation of cuneiform texts using treebanks and deep learning.

This article presents a bootstrapping pipeline for morpho-syntactic annotation of ancient language corpora, demonstrated through the creation of a new Akkadian treebank and a spaCy-based language model named AkkParser. Designed for individual scholars working with low-resource ancient languages, the pipeline integrates human annotation with iterative machine learning to efficiently produce richly annotated corpora, using two volumes of Neo-Assyrian letters from the State Archives of Assyria project hosted on Oracc as a case study. The authors discuss linguistic challenges specific to Neo-Assyrian letters and cuneiform data encoding, and provide quantitative evaluation showing that their bootstrapped model improves annotation accuracy over generic transformer models, particularly benefiting morpho-syntactic parsing despite difficulties with syntactic dependency labeling. The pipeline and resulting resources are publicly available and intended to support further research, pedagogy, and advanced linguistic analyses in Akkadian and potentially other ancient languages.