Demo2: Sentence Reduction for Summarization

Sentence Reduction removes non-essential information from a sentence.

Sample Program Output

You can compare the program output with the reduced sentence by professionals.

Example 1

Example 2

Example 3

Example 4

Example 5

Output from Human Professionals

Example 1 The V-chip will give parents a device to block out programs they don't want their children to see.

Example 2 Som and Hoffman's creation would allow broadcasters to insert multiple ratings into a show.

Example 3 Consumer groups are worried that a pending decision could lead paying per-minute rates for using the Internet.

Example 4 The two men have developed new software that they say would make the V-chip a surgical tool, rather than a sledgehammer.

Example 5 A report being released today by the American Association of University Women's Educational Foundation found that girls lag behind boys in the number and level of computer courses they enroll in, a gap that persists through college and into the lucrative technology career market. (no reduction)


Reduction based on Multiple Types of Knowledge

The reduction program is based on 3 types of knowledge: context information, the probabilities from the training corpus, and grammar knowledge from a lexicon.

Step 1: Determine which components in a sentence are complimentary in order to keep the sentence grammatically correct, using the syntactic knowledge in a large-scale lexicon we combined from English Verb Classes and Alternations, COMLEX, and WordNet.

Step 2: Determine which components in a sentence are important in the context, using the lexical links between words.

Step 3: Derive from the training corpus (which consists of reduced sentences by humans) what components in a sentence are likely to be removed.

Step 4: Combine the evidence from above 3 steps; reduce a sentence when all the evidence show the sentence component is reducible.