Proceedings of the MultiLing 2013 Workshop on Multilingual Multi-document Summarization, pages 13–19, Sofia, Bulgaria, August 9 2013. c 2013 Association for Computational Linguistics Multi-document multilingual summarization corpus preparation, Part 2: Czech, Hebrew and Spanish Michael Elhadad Ben-Gurion Univ. in the Negev, Israel elhadad@cs.bgu.ac.il Sabino Miranda-Jiménez Instituto Politécnico Nacional, Mexico sabino_m@hotmail.com Josef Steinberger Univ. of West Bohemia, Czech Republic jstein@kiv.zcu.cz George Giannakopoulos NCSR Demokritos, Greece SciFY NPC, Greece ggianna@iit.demokritos.gr Abstract This document overviews the strategy, ef- fort and aftermath of the MultiLing 2013 multilingual summarization data collec- tion. We describe how the Data Contrib- utors of MultiLing collected and gener- ated a multilingual multi-document sum- marization corpus on 10 different lan- guages: Arabic, Chinese, Czech, English, French, Greek, Hebrew, Hindi, Romanian and Spanish. We discuss the rationale be- hind the main decisions of the collection, the methodology used to generate the mul- tilingual corpus, as well as challenges and problems faced per language. This paper overviews the work on Czech, Hebrew and Spanish languages. 1 Introduction In this document we present the language- specific problems and challenges faced by Con- tributors during the corpus creation process. To facilitate the reader we repeat some information found in the first part of the overview (Li et al., 2013): the MultiLing tasks and the main steps of the corpus creation process. 2 The MultiLing tasks There are two main tasks (and a single- document multilingual summarization pilot de- scribed in a separate paper) in MultiLing 2013: Summarization Task This MultiLing task aims to evaluate the application of (partially or fully) language-independent summarization algorithms on a variety of languages. Each system participating in the task was called to provide summaries for a range of differ- ent languages, based on corresponding cor- pora. In the MultiLing Pilot of 2011 the lan- guages used were 7, while this year systems were called to summarize texts in 10 differ- ent languages: Arabic, Chinese, Czech, En- glish, French, Greek, Hebrew, Hindi, Roma- nian, Spanish. Participating systems were re- quired to apply their methods to a minimum of two languages. The task was aiming at the real problem of summarizing news topics, parts of which may be described or may happen in different mo- ments in time. We consider, similarly to Mul- tiLing 2011(Giannakopoulos et al., 2011) that news topics can be seen as event sequences: Definition 1 An event sequence is a set of atomic (self-sufficient) event descriptions, se- quenced in time, that share main actors, lo- cation of occurence or some other important factor. Event sequences may refer to topics such as a natural disaster, a crime investiga- tion, a set of negotiations focused on a single political issue, a sports event. The summarization task requires to generate a single, fluent, representative summary from a set of documents describing an event se- quence. The language of the document set will be within the given range of 10 languages and all documents in a set share the same lan- guage. The output summary should be of the same language as its source documents. The output summary should be between 240 and 250 words. Evaluation Task This task aims to examine how well automated systems can evaluate sum- maries from different languages. This task takes as input the summaries generated from automatic systems and humans in the Sum- marization Task. The output should be a grad- ing of the summaries. Ideally, we would want the automatic evaluation to maximally corre- late to human judgement. 13