Difference between revisions of "Mnduong writeup of Jansche and Abney"

From Cohen Courses
Jump to navigationJump to search
m (1 revision)
 
(No difference)

Latest revision as of 10:42, 3 September 2010

This is a review of Jansche_2002_information_extraction_from_voicemail_transcripts by user:mnduong.

  • This paper discusses an approach to extracting person names and telephone numbers from manual transcription of voicemail.
  • In the caller name extraction task, the approach relies heavily on positional and length features, taking advantage of the relatively homogeneous format of voicemail messages.
  • In the caller number extraction task, the method has two phases: the first phase uses a handcrafted grammar to propose candidate phone numbers, which aims at getting a high recall. The second phase uses a learner to classify these candidates as actual phone numbers or not.
  • I did not like the way the authors evaluated their method and compared it with others. For example, I think reporting results using the strict criterion for HZP but not for JA (even though they said it's not possible to do so for JA) is a little misleading.