Text Summarisation: From Human Activity to Computer Program. The Problem of Tacit Knowledge

Abstract

In this article I discuss whether the human activity of text summarisation can be successfully simulated in a computer. In order to write a computer program that produces high-quality summaries it becomes necessary to specify the cognitive pro-cesses involved when humans summarise text. As texts can be summarised in many different ways, evaluation of summaries becomes an important aspect in the discussion. The article discusses relevant factors in such an evaluation process. It turns out that humans when summarising texts make use of knowledge which is not readily open to scrutiny; it is tacit knowledge. This makes it very difficult to produce computer-generated summaries which are as good as those produced by skilled humans. New developments within artificial intelligence, relying on network processing techniques, may offer solutions to the problem of dealing with tacit knowledge. At present, accept-able computer summaries may be generated by programs combining accessible human knowledge of the summarisation process and knowledge about text.