More applications of automatic text summarization are, for example, summaries of news and scientific articles, summaries of electronic mails, summaries of different electronic information which later can be sent as SMS, summaries of found documents and pages returned by a retrieved system.
From one side, there is a single-document summarization which implies to communicate the principal information of one specific document, and from another side--a multi-document summarization which transmits the main ideas of a collection of documents.
According to the classical point of view, there are three stages in automated text summarization [Hov03].
This stage distinguishes extract-type summarization systems from abstract-type systems.
Summary generation is the third stage of text summarization.
That fact confirms the usage of lexical chains in text summarization [Bru01, Zho05, Li07].
For multi-document summarization, passages are retrieved using a language model [Yin07].
Abstractive summarization approaches use information extraction, ontological information, information fusion, and compression.
In this paper, we focus on single-document (1) Text Summarization from an extractive point of view, and we set out two goals for this research.
The paper is structured as follows: Section 2 gives an overview of the Text Summarization task, describing the main criteria that have been used to determine the relevance of a sentence within a document.
2 Determining sentence's relevance in text summarization
Although there has been increased attention to different criteria such as well-formedness, cohesion or coherence when dealing with summarization , , most work in this NLP task is still concerned with detecting relevant elements of text and presenting them together to produce a final summary.
1 The code quantity principle within the text summarization task
Starting from these principles, the approach we suggest here is to study how "The Code Quantity Principle" can be applied in the summarization task, to decide on which sentences of a document may contain more relevant information through its coding, and select these sentences to make up a summary.
Instrinsic evaluation assesses mainly coherence and summary's information content, whereas extrinsic methods focus on determining the effect of summarization on some other task, for instance Question Answering.