Exploiting Cross-Dialectal Gold Syntax for Low-Resource Historical Languages: Towards a Generic Parser for Pre-Modern Slavic

TL;DR

This paper demonstrates that training a generic parser on cross-dialectal data from related pre-modern Slavic languages significantly improves dependency parsing accuracy, achieving new state-of-the-art results for Old Church Slavonic and Old East Slavic.

Contribution

It introduces a cross-dialectal training approach for low-resource historical languages, creating a generic parser that outperforms specialized models on pre-modern Slavic data.

Findings

Achieved state-of-the-art UAS and LAS scores for OCS and OES.

Cross-dialectal data improves parser performance on low-resource languages.

Neural network model effectively handles linguistic heterogeneity.

Abstract

This paper explores the possibility of improving the performance of specialized parsers for pre-modern Slavic by training them on data from different related varieties. Because of their linguistic heterogeneity, pre-modern Slavic varieties are treated as low-resource historical languages, whereby cross-dialectal treebank data may be exploited to overcome data scarcity and attempt the training of a variety-agnostic parser. Previous experiments on early Slavic dependency parsing are discussed, particularly with regard to their ability to tackle different orthographic, regional and stylistic features. A generic pre-modern Slavic parser and two specialized parsers -- one for East Slavic and one for South Slavic -- are trained using jPTDP (Nguyen & Verspoor 2018), a neural network model for joint part-of-speech (POS) tagging and dependency parsing which had shown promising results on a number of Universal Dependency (UD) treebanks, including Old Church Slavonic (OCS). With these experiments, a new state of the art is obtained for both OCS (83.79\% unlabelled attachment score (UAS) and 78.43\% labelled attachement score (LAS)) and Old East Slavic (OES) (85.7\% UAS and 80.16\% LAS).