A Graph-Based Framework for Structured Prediction Tasks in Sanskrit

Abstract

We propose a framework using Energy Based Models for multiple structured prediction tasks in Sanskrit. Ours is an arc-factored model, similar to the graph based parsing approaches, and we consider the tasks of word-segmentation, morphological parsing, dependency parsing, syntactic linearisation and prosodification, a prosody level task we introduce in this work. Ours is a search based structured prediction framework, which expects a graph as input, where relevant linguistic information is encoded in the nodes, and the edges are then used to indicate the association between these nodes. Typically the state of the art models for morphosyntactic tasks in morphologically rich languages still rely on hand-crafted features for their performance. But here, we automate the learning of the feature function. The feature function so learnt along with the search space we construct, encode relevant linguistic information for the tasks we consider. This enables us to substantially reduce the training data requirements to as low as 10 % as compared to the data requirements for the neural state of the art models. Our experiments in Czech and Sanskrit show the language agnostic nature of the framework, where we train highly competitive models for both the languages. Moreover, our framework enables to incorporate language specific constraints to prune the search space and to filter the candidates during inference. We obtain significant improvements in morphosyntactic tasks for Sanskrit by incorporating language specific constraints into the model. In all the tasks we discuss for Sanskrit, we either achieve state of the art results or ours is the only data driven solution for those tasks.

Systematic Investigation of Strategies Tailored for Low-Resource Settings for Low-Resource Dependency Parsing

“…The STBC follows annotations based on Kāraka [Kulkarni et al 2010;Kulkarni and Sharma 2019], the grammatical tradition of Sanskrit, while the VST uses Universal Dependency (UD). Following Krishna et al [2020c], we use sentence level macro averaged Unlabelled and Labelled Attachment Scores (UAS, LAS) and t-test for statistical significance [Dror et al 2018].…”

“…[BiAFF] is a graph-based approach with BiAFFINE attention mechanism. Krishna et al [2020b][MG-EBM] extends Krishna et al [2020c][Tree-EBM-F] using multi-graph formulation. We report their standalone numbers for fair comparison.…”

“…We do these investigations on the dev set to find out the best strategy from each category. Following Krishna et al [2020c], we use sentence level macro averaged Unlabelled and Labelled Attachment Scores (UAS, LAS). [2018] introduce Cropping: deletes some parts of a sentence to create multiple short meaningful sentences and Rotation: permutes the siblings of headword restricted to a set of relations.…”

Naive Bayesian Prediction of Japanese Annotated Corpus for Textual Semantic Word Formation Classification

“…In order to solve these problems, this paper attempts to introduce ontology, using ontology hierarchical structure and attribute constraints to match keywords with domain ontology concepts, and establish a concept vector space model for Japanese text classification. It aims to solve the multisense and conceptual hierarchical problems in Japanese text classification, overcome the shortcomings of keyword-based classification methods, and improve the accuracy of classification [12][13][14]. At the same time, this paper also studies the relationship between Japanese text classification and personalized information retrieval, analyzes the text interest model, and proposes a text interest model establishment and adjustment algorithm to make the classification result more in line with the text intent.…”

Evaluating Neural Word Embeddings for Sanskrit

“…The availability of such resources triggered the emergence of the data-driven neural-based approaches into the SCL field. The wide success of neural-based data-driven approaches is greatly justified by the state of the art performance for many downstream tasks for Sanskrit, namely, segmentation (Hellwig and Nehrdich, 2018;Reddy et al, 2018;, dependency parsing (Krishna et al, 2020c;Krishna et al, 2020a;Sandhan et al, 2021;Krishna et al, 2020b), semantic type identification (Sandhan et al, 2019;Krishna et al, 2016), word order linearisation Krishna et al, 2020c) and morphological parsing (Gupta et al, 2020;.…”