Image Caption Generation for Low-Resource Assamese Language

Image Caption Generation for Low-Resource Assamese Language Prachurya Nath1, Prottay Kumar Adhikary1, Pankaj Dadure2, Partha Pakray1, Riyanka Manna3, Sivaji Bandyopadhyay1 1National Institute of Technology, Silchar, India 2University of Petroleum & Energy Studies, Dehradun, India 3ADAMAS University, Kolkata, India {prachuryanath00, prottay71@gmail, krdadure, parthapakray, riyankamanna16, sivaji.ju.cse}@gmail.com Abstract Image captioning is a prominent Artificial Intelligence (AI) research area that deals with visual recognition and a linguistic de- scription of the image. It is an interdis- ciplinary field concerning how computers can see and understand digital images & videos, and describe them in a language known to humans. Constructing a mean- ingful sentence needs both structural and semantic information of the language. This paper highlights the contribution of image caption generation for the Assamese lan- guage. The unavailability of an image cap- tion generation system for the Assamese language is an open problem for AI-NLP re- searchers, and it’s just an early stage of the research. To achieve our defined objective, we have used the encoder-decoder frame- work, which combines the Convolutional Neural Networks and the Recurrent Neural Networks. The experiment has been tested on Flickr30k and Coco Captions dataset, which have been originally present in the English language. We have translated these datasets into Assamese language us- ing the state-of-the-art Machine Transla- tion (MT) system for our designed work. Keywords: Caption Generation, Low- resource Language, Attention, Assamese. 1 Introduction Over 24 million native speakers speak the As- samese language in the north-eastern part of India. It is an eastern Indo-Aryan (Indic) language which is the official language of In- dia’s Assam state. Assamese is an indigenous Indo-Aryan language which has been influ- enced in vocabulary, phonetics, and structure by the region’s close association with Tibeto- Burman dialects. Its grammar is notable for its highly inflected forms, different pronouns, plural noun markers, and honorific and non- honorific constructions. The Assamese script is very close to the Bengali script. Assamese, like English, is written from left to right. The Assamese literary tradition can be traced back to the 13th century. In the 16th century, prose texts, most notably bu- ranjis (historical works), began to appear. The Assamese alphabet (Assamese, Oxomiya bornomala) is shown in Fig. 1 which is the Bengali-Assamese script used in the As- samese language. Other north-eastern lan- guages that are using the script include Bodo (now Devanagari), Khasi (now Roman), Mis- ing (now Roman), Jaintia (now Roman), and others. The Kamarupi script was used to cre- ate it. Since Fifth century Umachal/Nagajari- Khanikargaon rock inscriptions written in an eastern variant of the Gupta script, the script has evolved continuously, with significant influ- ences from the Siddha script in the 7th century (Saharia and Konwar, 2012). The current for- mat is identical to the Bengali alphabet with the exception of two letters, (ro) and (vo); and the letter (khya) has progressed into an inde- pendent consonant with its own phonetic qual- ity, however in the Bengali alphabet it is a con- junct of two letters. Attempting to make computers mimic hu- mans’ ability …

Content truncated to 500 words