Text summarization remains an elusive goal of automatic summarizers. The state of the art for summarization has begun to reach a limit using extractive techniques. Thus, systems have been given a new task of creating a more responsive summary using guides to help better focus the summary while allowing for detailed diagnostics of its content during evaluations.

Sections 2 and 3 thoroughly explain the purpose of topic-specific summarization and the guided summarization task at the 2010 Text Analysis Conference (TAC), respectively. The excellent analysis of the performances of both human and automatic summarizers on the task clearly shows where the weaknesses and strengths lie for both and how the field can be improved; much evidence is clearly shown within the included tables and figures.

Term frequency- and position-based techniques are still the best available; as a result, the authors’ outlook on the abilities of automatic summarizers to choose the best sentences by selecting something as elusive as “important information” is rather dim: they state that the use of semantic-oriented techniques is futile because “we do not know what to look for.” Although this semantic futility is not well founded within the body of work upon which this paper is based, the paper still elucidates adequate areas of improvement for automatic summarizers, reasons for what works and what does not, and ideas for changes to automatic summarization evaluation.

Although not very advanced, the paper assumes some knowledge of text summarization and its history. It is a must-read for those who already have some understanding of text summarization and are looking to either build a better system or improve the weaker sub-area of automatic summarization evaluation.