Poetry to Prose Conversion in Sanskrit as a Linearisation Task: A Case for Low-Resource Languages

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1160–1166 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 1160 Poetry to Prose Conversion in Sanskrit as a Linearisation Task: A case for Low-Resource Languages Amrith Krishna1, Vishnu Dutt Sharma2, Bishal Santra1, Aishik Chakraborty3∗, Pavankumar Satuluri4 and Pawan Goyal1 1Dept. of Computer Science and Engineering, IIT Kharagpur 2American Express India Pvt Ltd 3School of Computer Science, McGill University 4Chinmaya Vishwavidyapeeth {amrith,bishal.santra}@iitkgp.ac.in, pawang@cse.iitkgp.ac.in chakraborty.aishik@gmail.com Abstract The word ordering in a Sanskrit verse is of- ten not aligned with its corresponding prose order. Conversion of the verse to its cor- responding prose helps in better comprehen- sion of the construction. Owing to the re- source constraints, we formulate this task as a word ordering (linearisation) task. In do- ing so, we completely ignore the word ar- rangement at the verse side. k¯avya guru, the approach we propose, essentially consists of a pipeline of two pretraining steps followed by a seq2seq model. The first pretraining step learns task specific token embeddings from pretrained embeddings. In the next step, we generate multiple hypotheses for possible word arrangements of the input (Wang et al., 2018). We then use them as inputs to a neu- ral seq2seq model for the final prediction. We empirically show that the hypotheses gener- ated by our pretraining step result in predic- tions that consistently outperform predictions based on the original order in the verse. Over- all, k¯avya guru outperforms current state of the art models in linearisation for the poetry to prose conversion task in Sanskrit. 1 Introduction Prosody plays a key role in the word arrange- ment in Sanskrit Poetry. The word arrangement in a verse should result in a sequence of syllables which adhere to one of the prescribed meters in Sanskrit Prosody (Scharf et al., 2015). As a re- sult, the configurational information of the words in a verse is not aligned with its verbal cogni- tion (Bhatta, 1990; Dennis, 2005). Obtaining the proper word ordering, called as the prose order- ing, from a verse is often considered a task which requires linguistic expertise (Shukla et al., 2016; Kulkarni et al., 2015). ∗Work done while the author was at IIT Kharagpur In this work, we use neural sequence genera- tion models for automatic conversion of poetry to prose. Lack of sufficient poetry-prose parallel data is an impediment in framing the problem as a seq2seq task (Gu et al., 2018).1 Hence, we formu- late our task as that of a word linearisation task (He et al., 2009). In linearisation, we arrange a bag of words into a grammatical and fluent sentence (Liu et al., 2015). This eliminates the need for paral- lel data, as the poetry order is not anymore rele- vant at the input. A neural-LM based model from Schmaltz et al. (2016) and a seq2seq model form Wiseman and Rush (2016) are the current state of the art (SOTA) models in the linearisation task. We first show that …

Content truncated to 500 words