Apappu writeup of Jansche and Abney

From Cohen Courses
Jump to navigationJump to search

This is a review of Jansche_2002_information_extraction_from_voicemail_transcripts by user:apappu.

Task: IE from Voice mail transcripts This paper aims at extracting a small subset of named entites (as opposed to NE extraction from broadcast news) from voice-mails.

-Authors claim that they could leverage from data for rich set of features for candidate classification.

-Authors are using lexical indicators for beginning and ending of the caller phrase.

-and they report that results in table 1 are worse because of not adapting the generic NEtagger to the task.

Authors also state that their approach and Huang's is similar except for ASR component and the hand-rules component.


Comments :

- This paper good enough for the task at hand but I am skeptical about how would they deal with In-Vocabulary and foreign origin names whose length could be similar to "call-phrases" but not "names".

- This task may involve some long distant dependencies between entities extracted, since this paper addresses the problem of exploiting local features without a global view of the entities.