A Workbench for Rapid Generation of Cross-Lingual Summaries

A Workbench for Rapid Generation of Cross-Lingual Summaries Nisarg Jhaveri, Manish Gupta†, Vasudeva Varma International Institute of Information Technology, Gachibowli, Hyderabad, Telangana - 500 032, India. nisarg.jhaveri@research.iiit.ac.in, {manish.gupta, vv}@iiit.ac.in Abstract The need for cross-lingual information access is more than ever with the easy accessibility to the Internet, especially in vastly multilingual societies like India. Cross-lingual summarization can help minimize human effort needed for achieving publishable articles in multiple languages, while making the most important information available in target language in the form of summaries. We describe a flexible, web-based tool for human editing of cross-lingual summaries to rapidly generate publishable summaries in a number of Indian Languages for news articles originally published in English, and simultaneously collect detailed logs about the process, at both article and sentence level. Similar to translation post-editing logs, such logs can be used to evaluate the automated cross-lingual summaries, in terms of effort needed to make them publishable. The generated summaries along with the logs can be used to train and improve the automatic system over time. Keywords: information access, cross-lingual summarization, machine translation, human evaluation of summaries 1. Introduction In Indian news media, most of the content gets published in English first and then in regional languages, especially for news categories such as national or international news, technology or lifestyle news. The delay can be just a few hours, days or sometimes the news does not appear in the regional languages at all. On the other hand, some content gets generated and consumed in regional languages alone. With the Internet becoming easily accessible and the rise of digital journalism, it is now crucial to make the large amount of news published in English or other popular lan- guages on the Internet available to the readers of other lan- guages having fewer native publications. Advances in Machine Translation (MT) and other fields of Computational Linguistics in recent years make it possi- ble to automate cross-lingual news access. However, the current state of Machine Translation is not able to gener- ate publishable articles in most Indian Languages from En- glish. Although, post-editing MT output has been shown to increase translator’s productivity over translating from scratch (Aziz et al., 2012), it still requires a significant amount of human effort to produce end-user consumable articles. Making the highlights or summaries of articles originally published in English accessible to non-English speaking users helps in making a large amount of critical informa- tion accessible as fast as possible with minimal human ef- fort. We aim to make the process completely or partially automatic so that the gist of the articles can be published in regional languages with minimal delay. Working in this direction, we’ve developed a pluggable sys- tem, implementing a pipeline for cross-lingual summariza- tion of news articles. While summarization and translation are two major modules in the automatic cross-lingual sum- marization pipeline, a number of other modules can be in- cluded, for example, preprocessing, automatic post-editing, etc. After automatic processing, the articles are sent to hu- mans for …

Content truncated to 500 words