GujMORPH - A Dataset for Creating Gujarati Morphological Analyzer

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 7088–7095 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 7088 GujMORPH - A Dataset for Creating Gujarati Morphological Analyzer Jatayu Baxi, Brijesh Bhatt Dharmsinh Desai University Gujarat(India) {jatayubaxi.ce, brij.ce}@ddu.ac.in Abstract Computational morphology deals with the processing of a language at the word level. A morphological analyzer is a key linguistic word level tool that returns all the constituent morphemes and their grammatical categories associated with a particular word form. For the highly inflectional and low resource languages, the creation of computational morphology-related tools is a challenging task due to unavailability of underlying key resources. In this paper, we discuss the creation of an annotated morphological dataset- GujMORPH for the Gujarati - an indo aryan language. For the creation of this dataset, we studied language grammar, word formation rules, suffix attachments in depth. This dataset contains 16527 unique inflected words along with their morphological segmentation and grammatical feature tagging information. It is a first-of-its-kind dataset for the Gujarati language and can be used to develop morphological analyzer and generator models. The dataset is annotated in the standard Unimorph schema and evaluated on the baseline system. We also describe the tool used to annotate the data in the standard format. The dataset is released publicly along with the library. Us- ing this library, the data can be obtained in a format that can be directly used to train any machine learning model. Keywords: Language resources, Morphology,Gujarati 1. Introduction Morphological analysis is a key word level natu- ral language processing task, used to analyze an inflected word. The morph analyzer outputs con- stituent morphemes and grammatical feature sets associated with an inflected word form. It is necessary to have a powerful morphological ana- lyzer system for the downstream NLP tasks such as question answering, machine translation, senti- ment analysis, etc. The morph analyzer usually re- quires knowledge in form of word formation rules, language grammar understanding, and extensive feature engineering in order to achieve decent accu- racy (Goyal and Lehal, 2008; Sharma et al., 2021). As the demand for web-based technologies is in- creasing day by day, language resources have a crucial role to play to prevent the extinction of the natural language. Nearly all European and East Asian languages are resource-rich languages but most of the south and southeast asian Languages including Gujarati can be still considered as low re- sourced.(Ali et al., 2020) Gujarati is an Indo-Aryan language, spoken mainly in the Gujarat state of India. It is the 26th most widely spoken language with approximately 55 million speakers across the world.(Magueresse et al., 2020; Suba et al., 2011) In the literature, there are mainly two methods for building a morphological analyzer. The first is the classic rule-based method and the second is a ma- chine learning-based methods(PJ and KP, 2012). For the rule-based methods, a substantial amount of knowledge is required in form of word formation rules, suffixes, and other language-specific char- …

Content truncated to 500 words