Multiple alternative sentence compressions for automatic text summarization

We perform multi-document summariza-tion by generating compressed versions
of source sentences as summary candi-
dates and using weighted features of these
candidates to construct summaries. We
combine a parse-and-trim approach with
a novel technique for producing multiple
alternative compressions for source sen-
tences. In addition, we use a novel method
for tuning the feature weights that maxi-
mizes the change in the ROUGE-2 score
(∆ROUGE) between the already existing
summary state and the new state that re-
sults from the addition of the candidate
under consideration. We also describe ex-
periments using a new paraphrase-based
feature for redundancy checking. Finally,
we present the results of our DUC2007
submissions and some ideas for future
work.