Natural Language Processing

Text Summarization 287

F

k

i

= max

j

(i<j)

score(S

i

) + (F

j

k–1

)

if S

i

, S

j

have common words

penalty * (score(S

i

) + (F

j

k–1

)) otherwise

Ï

Ì

Ó

Penalty value chosen in the paper in order to inhibit the use of

successive sentences without words in common is 1/10.

Thus, Dynamic Programming algorithm outputs for a sequence

of scored sentences S

1

, . . . , S

n

and a given length X the optimal

summary with the score F

X

1

.

Pure Entailment

The last method presented here relies on textual entailment to reduce

the redundancy of a summary. It is called Pure Entailment method

and is introduced in (Tatar et al. 2008a). The method is more effi cient

if instead of the text {S

1

, S

2

, . . . S

n

} are used the best scored (by one

of the above methods) sentences of the ...