A Novel Multi-Task Learning Approach for Context-Sensitive Compound Type Identification in Sanskrit

Proceedings of the 29th International Conference on Computational Linguistics, pages 4071–4083 October 12–17, 2022. 4071 A Novel Multi-Task Learning Approach for Context-Sensitive Compound Type Identification in Sanskrit Jivnesh Sandhan1, Ashish Gupta2, Hrishikesh Terdalkar1, Tushar Sandhan1, Suvendu Samanta1, Laxmidhar Behera1,3 and Pawan Goyal2 1IIT Kanpur, 2IIT Kharagpur, 3IIT Mandi jivnesh@iitk.ac.in, ashishgupta2598@gmail.com, pawang@cse.iitkgp.ac.in Abstract The phenomenon of compounding is ubiqui- tous in Sanskrit. It serves for achieving brevity in expressing thoughts, while simultaneously enriching the lexical and structural formation of the language. In this work, we focus on the San- skrit Compound Type Identification (SaCTI) task, where we consider the problem of iden- tifying semantic relations between the compo- nents of a compound word. Earlier approaches solely rely on the lexical information obtained from the components and ignore the most cru- cial contextual and syntactic information useful for SaCTI. However, the SaCTI task is chal- lenging primarily due to the implicitly encoded context-sensitive semantic relation between the compound components. Thus, we propose a novel multi-task learning architecture which incorporates the contextual information and enriches the complementary syntactic information using morphological tag- ging and dependency parsing as two auxiliary tasks. Experiments on the benchmark datasets for SaCTI show 6.1 points (Accuracy) and 7.7 points (F1-score) absolute gain compared to the state-of-the-art system. Further, our multi- lingual experiments demonstrate the efficacy of the proposed architecture in English and Marathi languages.1 1 Introduction A compound is defined as a collection of one or more entities that act as a single meaningful entity. The process of decoding an implicit semantic re- lation between the components of a compound in Sanskrit is called Sanskrit Compound Type Iden- tification (SaCTI). Alternatively, it is also termed as Noun Compound Interpretation (NCI) (Ponkiya et al., 2021, 2020). In the literature, the NCI prob- lem has been formulated in two related but different ways. Let’s take mango juice as an example. In the 1The code and datasets are publicly available at: https: //github.com/ashishgupta2598/SaCTI first formulation, the relation between the two com- ponents is labeled from a set of semantic relations (MADEOF) (Dima and Hinrichs, 2015; Fares et al., 2018; Ponkiya et al., 2021). The second formu- lation uses paraphrasing to illustrate the semantic relations (a juice made from mango) (Lapata and Keller, 2004; Ponkiya et al., 2018a, 2020). In this work, we use the first formulation that frames the task as a multi-class classification prob- lem. The task is challenging and often depends upon the context or world knowledge about the en- tities involved (Krishna et al., 2016). For instance, the semantic type of the compound r¯ama-¯ı´svarah. can be classified into one of the following semantic types depending on the context: Karmadh¯araya2, Bahuvr¯ıhi and Tatpurus.a. Although the compound has the same components as well as the final form, the implicit relationship between the components can be decoded only with the help of available con- textual information (Kulkarni and Kumar, 2013; Krishna et al., 2016). Due to such instances, the downstream Natural Language Processing (NLP) applications for Sanskrit such as question answer- …

Content truncated to 500 words