Sentence Extraction as a Classification Task

Abstract

A useful first step m document summan - sation m the selectton of a small number of 'meamngful' sentences frOm a larger text Kupiec et al (1995) describe tins as a classfi caton task on the bass of a corpus of technical papers with summaries written by professional abstractors, their system dentLqes thoze sentences m the text winch also occur m the summary, ud then acquires a model of the 'a]stract-wort]nness' of a sentence as a combination of a bzmted numbel of properties of that sentence We report on a rephcatlon of tins exper- lment wth different dat summaries for our documents were not written by professwnal abstractors, but by the authors themselves TIn produced fewer ff, nable sentences to tram on We use alternative 'meamn' ntence (selected by a humanjudge) as tremng and evaluation material, Because tins has advantages for the subsequent automaCac generatxon of more qe.ble abstracts We quantitatively compare the two dfferent strategies for ud evaluation (vz ahsrnment vz human judgement), we a dincuss quahtatlve chf- ferences and consequences for the genera- tion of abstracts 1