Ksuravar writeup of Jansche and Abney

This is review of Jansche_2002_information_extraction_from_voicemail_transcripts by user:ksuravar.

My previous review is at Information_extraction_from_voicemail_transcripts,_by_M._Jansche,_S._P_Abney._In_Proceedings_of_the_ACL-02_conference_on_Empirical_methods_in_natural_language_processing-Volume_10,_2002.. copied as suspecting a problem with long names in the wiki.

Summary: The paper is about information extraction (extracting caller information and phone numbers) applied to voice mail transcripts. The paper suggests features (feature selection) to be considered for caller information ,(both caller phrases and caller names) like position of the name (or name phrases) from the beginning of the voce mail and the length of the phrase or name, and for phone numbers ,like phone number length and the position from the end of the voice transcript. For both the names and phone numbers the paper then uses simple grammer rules to select possible candates and then uses binary classifiers to do the classification to prune the false candidates. The paper finally presents their results manual automatically generated voice mails.

I liked the simple approach of the paper in solving the problem and yet getting good results.