TAC2010 Update Summarization and AESOP of ICL

For the guided summarization task of TAC 2010, we submitted two runs using the combination of manifold ranking score and aspects information. The only difference between these two runs is that one computes a score according to aspect information while the other one considers aspect coverage in each sentence. For the update summarization task, we adopt similar methods and simultaneously penalize the information overlap between docset B and docset A. For the AESOP task, we focused on No Models evaluation case and submitted three runs. The first one uses a simple linear combination of three ROUGE scores, including ROUGE-1, ROUGE-2 and ROUGE-SU4, with four human summaries as models. The other two methods extract term information from four human summaries, such as term frequency and model frequency, based on which we evaluate each automatic summary. The difference between these two runs is that one of them introduces more term information, such as bigram information.