lr,6-2-I05-5004,ak preliminary experiments on building a <term> paraphrase corpus </term> have so far been producing promising
measure(ment),22-2-I05-5004,ak which we have evaluated according to <term> cost-efficiency </term> , <term> exhaustiveness </term> , and
measure(ment),24-2-I05-5004,ak according to <term> cost-efficiency </term> , <term> exhaustiveness </term> , and <term> reliability </term> . We
measure(ment),27-2-I05-5004,ak </term> , <term> exhaustiveness </term> , and <term> reliability </term> . We propose a method that automatically
other,17-1-I05-5004,ak class-oriented framework </term> for collecting <term> paraphrase examples </term> , in which <term> sentential paraphrases
other,22-1-I05-5004,ak paraphrase examples </term> , in which <term> sentential paraphrases </term> are collected for each <term> paraphrase
other,28-1-I05-5004,ak paraphrases </term> are collected for each <term> paraphrase class </term> separately by means of <term> automatic
other,4-1-I05-5004,ak Towards <term> deep analysis </term> of <term> compositional classes </term> of <term> paraphrases </term> , we have
other,7-1-I05-5004,ak <term> compositional classes </term> of <term> paraphrases </term> , we have examined a <term> class-oriented
tech,1-1-I05-5004,ak used in the experiments . Towards <term> deep analysis </term> of <term> compositional classes </term>
tech,13-1-I05-5004,ak paraphrases </term> , we have examined a <term> class-oriented framework </term> for collecting <term> paraphrase examples
tech,34-1-I05-5004,ak class </term> separately by means of <term> automatic candidate generation </term> and <term> manual judgement </term> .
tech,38-1-I05-5004,ak automatic candidate generation </term> and <term> manual judgement </term> . Our preliminary experiments on
hide detail