NotesFAQContact Us
Search Tips
ERIC Number: ED380996
Record Type: Non-Journal
Publication Date: 1994-Dec-9
Pages: 7
Abstractor: N/A
Reference Count: N/A
An ANN That Applies Pragmatic Decision on Texts.
Aretoulaki, Maria; Tsujii, Jun-ichi
A computer-based artificial neural network (ANN) that learns to classify sentences in a text as important or unimportant is described. The program is designed to select the sentences that are important enough to be included in composition of an abstract of the text. The ANN is embedded in a conventional symbolic environment consisting of lexical/semantic, morphological, syntactic, and pragmatic analyzers and synthesizers. Only certain features are computed by these symbolic modules, those that are more relevant to this sentence classification task. The selected features are translated into vectors of ones, zeros, and intermediate values, and input into the network. In this way, the ANN collectively considers all linguistic and pragmatic levels in making a decision. This is in contrast to most text abstraction systems, which use information retrieval techniques such as keyword extraction. Preliminary testing shows the ANN has a success rate of 86 percent with 10 different sets of novel sentences, after having been trained on disparate sets of 90 sentences. Contains 12 references. (Author/MSE)
Publication Type: Reports - Descriptive
Education Level: N/A
Audience: N/A
Language: English
Sponsor: N/A
Authoring Institution: N/A