This paper describes the systems of PKUTM in Text Analysis Conference (TAC) 2010. We participated in the Recognizing Textual Entailment (RTE) track and the Summarization track. For the RTE track, we propose a method to map every node in the hypothesis to one or more nodes in the text. With the help of named-entities tools, MINIPAR relationships, and regular patterns to recognize temporal and numeric expressions, some nodes are merged into one node. We transform the hypothesis by using semantic knowledge from sources like WordNet, VerbOcean, and LingPipe. In the Summarization track, we propose a unified framework for both kinds of summarization. We employ a manifold-ranking model to select sentences and a novel sentence ordering method to generate final summaries. The underlying idea of the proposed approach is that a good summary is expected to include the sentences with both high biased information richness and high information novelty. The evaluation results show that our proposed two frameworks are very effective for RTE and Summarization tasks, respectively.
[1]
Christiane Fellbaum,et al.
Book Reviews: WordNet: An Electronic Lexical Database
,
1999,
CL.
[2]
Patrick Pantel,et al.
VerbOcean: Mining the Web for Fine-Grained Semantic Verb Relations
,
2004,
EMNLP.
[3]
Dekang Lin,et al.
Dependency-Based Evaluation of Minipar
,
2003
.
[4]
Xiaojun Wan,et al.
Manifold-Ranking Based Topic-Focused Multi-Document Summarization
,
2007,
IJCAI.
[5]
Ido Dagan,et al.
Addressing Discourse and Document Structure in the RTE Search Task
,
2009,
TAC.
[6]
Adrian Iftene,et al.
UAIC Participation at RTE4
,
2008,
TAC.
[7]
Hua Li,et al.
Improving web search results using affinity graph
,
2005,
SIGIR '05.
[8]
Vladimir I. Levenshtein,et al.
Binary codes capable of correcting deletions, insertions, and reversals
,
1965
.
[9]
M. Felisa Verdejo,et al.
Textual Entailment Recognition Based on Dependency Analysis and WordNet
,
2005,
MLCW.