Zero-shot reductive paraphrasing for digitally semi-literate
Prawaal Sharma, Navneet Goyal · Forum for Information Retrieval Evaluation · 2021
People in developing countries with restricted schooling, face hurdles with their digital enablement. Constrained education creates issues with comprehensibility of information on internet and other digital platforms. Most content on digital platforms use enriched vocabulary for more accomplished users and hence does not seem to be very useful for digitally semi-literate users. Artificial Intelligence (AI), Information Retrieval (IR) and Natural Language Processing (NLP) can be applied for text simplification to bridge the digital divide and empower these users. In this paper we propose to achieve reductive paraphrasing using Neural Machine Translation (NMT) framework along with encoder-decoder model and Gated Recurring Unit (GRU). Our approach combines Zero-shot Learning (ZSL) using multi-pivot method to execute our experiment. We have considered English as the base language and three pivot languages (French, German and Spanish) to verify our claims. We have designed the simplified vocabulary from movies for younger audience. It has been observed that using the approach as described in our paper, an improvement of an average of 25% in ease of comprehension can be achieved.