Automatic Summary Generation

Every day the number of electronic articles, books, reports and web pages published increases. As professionals we are expected to keep up to date but on many occasions we drown under a sea of information. Professional researchers, journalists and doctors are the target audiences for much of the research carried out on the automatic generation of document summaries.
The MultiReader project has identified a very different audience who would benefit from regularly having information summarised for them. This group of users are people who are visually impaired, deaf and dyslexic. Whilst some members of these groups have to deal with large amounts of information in their working capacity, many print disabled readers would benefit from the summarisation of everyday information such as newspapers or web sites.

It may be obvious that people with a visual impairment would benefit from having a Braille or audio summary of a text that would otherwise take a considerable length of time to read and digest. The deaf however, are often forgotten when reading strategies and assistive technologies are discussed as their reading impairment is not as obvious a disability as that of people who are visually impaired. Deaf people very often have to contend with poor reading skills which make the assimilation and comprehension, particularly of long documents, very difficult. People with dyslexia are also likely to benefit from having documents summarised for them. Being given a summary of a document will allow the person with a visual impairment, deafness or dyslexia to choose whether to persevere with reading the whole document or if the summary is sufficient for their needs.

Accurate summarisation of text depends largely on the accuracy of the writer and the algorithms used in the package. The automatic summarisation technology devised by Copernic and Sinope are based upon statistical models and "knowledge intensive" processes. The statistical model is based upon Bayesian estimates of the specific vocabulary and the grammatical rule systems of the language. The knowledge intensive process aims to replicate the factors that a human summariser uses when summarising documents. These factors are the concepts involved with the documents topic and the relative importance of each sentence. The process of summarisation involves the application of linguistic algorithms which pick out the most frequently used words and looks for the key concept words in the document.
Packages such as Copernic and Sinope and the auto summarise function in later versions of Microsoft Word warrant further investigation and evaluation with users who have a visual impairment, are deaf or dyslexic.