Information extraction from voicemail transcripts

This is a review of Jansche_2002_information_extraction_from_voicemail_transcripts by by user:sgopal1.

• The paper describes some methods to extract named entities out of voicemail transcripts. They have use hand-crafted rules, as well as certain other rules that applies to this task ( such as finding maximal sequence of digits etc ).

• To extract the caller names, they perform two prediction tasks, one to predict the start of the caller name and next one to predict the length of the name. A two phase approach is applied to extract the phone numbers - A set of handcrafted rules and decision tree learner.

• I'm not sure I understand their containment based evaluation. It seems to me like I can achieve 100% performance if I simply output the entire text, as that it is going to have the required text anyway.
  • Good observation!