Summarizing Text Conversations - Methods for Mining and Summarizing Text Conversations - page 102

4.6 IMPORTANT POINTS

￿ We discuss summarization systems according to three aspects: assumptions and inputs, mea-
sures of informativeness, and outputs and interfaces.

￿ The assumptions and inputs aspect includes the nature of the corpus to be summarized, rep-
resentations of the data, and the presence of upstream processes on which summarization
depends.

￿ The measures of informativeness aspect is concerned with how salience is determined within
the summarization system.

￿ The outputs and interfaces aspect concerns the modality and structure of the final summary.

￿ We surveyed summarization work applied to meetings, emails, blogs and chats. In each of these
areas, domain-specific summarization approaches have been developed, incorporating features
and techniques specific to the particular characteristics of the data.

￿ Multi-modal conversation summarization techniques have been developed and can be applied to
conversations in any modality. General findings indicate that these approaches are competitive
with domain-specific techniques.

￿ We provided a detailed case study of one abstractive conversation summarization system, illus-
trating how such a system differs from more common extractive systems.

4.7 FURTHER READING

Several topics have been published on general automatic summarization [ Endres-Niggemeyer , 1998 ,
Mani , 2001a , Mani and Maybury , 1999 ]. While none of these topics are current, each provides an
overview of the basic tasks and distinctions. Within Mani and Maybury [ 1999 ], Karen Spärck-Jones
has an influential paper on factors and directions for summarization [ Jones , 1999 ].
A more recent discussion of automatic summarization can be found in Jurafsky and Martin
[ 2008 ]. This includes summarization case studies from prior to 2008.
A forthcoming topic will discuss speech summarization specifically [ Penn and Zhu ,
Forthcoming ]. Whereas we concentrate on textual conversations, including spoken conversations
with written transcripts, Penn and Zhu describe summarization systems that exploit speech-specific
characteristics such as prosody.