Liuliu writeup of Jansche and Abney

This is a review of Jansche_2002_information_extraction_from_voicemail_transcripts by user:Liuliu.

This paper is about extracting call names and telephone numbers from transcripts (manual and automatic) of voicemails.

I like

• The structure of the paper is very clear: first introduce the challenges in this field and review related work; then, describes how they solve the two problems; concludes at last.

• To achieve both high precision and high recall, they followed a two phases approach and each phase focus on one objective. By doing so, it also became easier to achieve each objective. This reminds me of divide and conquer, which I like.

• They didn't use a lot of sophisticated lexical features or syntactic features.

I don't like

The evaluation part in this paper is very solid in my opinion.

• Are results of different methods on different data sets comparable?

• Although they argued that because they didn't have annotated automatic transcripts so that they didn't report JA strict on it, I still want to know it and I think to make the research solid, they should do that