LIF at TAC Multiling: Towards a Truly Language Independent Summarizer

摘要

This paper presents the LIF system for the TAC'2011 Multilingual pilot track. We followed a language-independent approach to summarization for this task. In particular, we tried to remove the following dependences to language: sentence segmentation, word segmentation, stop-word lists, and word-level relevance assessment. We applied these modifications to an MMR-based system and observed little degradation on English data. The submitted system had a bug that impeded all official results, therefore we propose in this paper an updated set of results with relevant analysis.