Adapting Lexical Chaining to Summarize Conversational Dialogues

We present a system for summarizing transcripts of conversational dialogues based on lexical chaining. The experiments were carried out with twenty Switchboard dialogues (LDC, 1993). We designed and implemented four summarization methods employing lexical chains as their source representation. The summarization task is defined as extracting the most relevant utterances conveying the meaning of the dialogue. We evaluate the methods against lead and random baseline systems and show that lexical chaining outperforms them in terms of precision and recall.