IPP Program Description

IPP PROORAM DESCRIPTIOO

Michael Lebowitz

Colunbia University, Deparbnent of Computer Science

406 Mudd Buildin:J, New York, NY 10027

IPP (the Integrated Partial Parser) is a complete urrlerstaooi~ system that reads news stories and records them in a lo~-tenn episodic memory. IPP learns about the 100rld by readi~ stories taken from newspapers and the UPI news wire, addi~ infonnation from these stories to memory, aoo makin:J generalizations that describe specific situations. Its primary domain is international terrorism. IPP uses the generalizations that it has made to help in l.l'derstandi~ future stories. A complete description of IPP can be found in [1].

As IPP ad1s stories to memory, it looks for similar events aoo attempts to make generalizations that describe the 100rld. SUch generalizations fonn the basis for organizi~ events in memory and urrlerstaooi~ later stories.

As a story uooerstan:'ler, IPP makes extensive use of to~wn, predictive processi~. As it processes a story, IPP accesses memory in an attempt to identify generalizations describi~ stereotypical situations that can provide predictions to be used in Lllderstaooi~, yieldi~ a a robust and efficient l.l'derstandi~ system that has been used to Lllderstaoo over 300 news stories.

'ftle overall design of integration of update/generalization, and memory is shown in Figure 1. IPP, involvi~ the parsi~, memory generalization-based Figure 2 illustrates IPP readin:J a typical terrorism story aoo generalizin:J that terrorist attacks in Spain are often Basque a!Dushes (havi~ read a similar story previously) ·

Figure 2: An IPP ~ample

*(PARSE EXl)

Story: EX1 (7 29 80)

(BASQUE TERRCRISTS No!BUSHED A CAR CARRYit«> AN ARMY GENERAL THRQlX;H I:XliNTCWN HAmiD TUESil\Y SLIGifl'LY WOUNDII«> THE GENERAL 8111' KILLIIIG HIS BOD'iGIJARD AND SERIOUSLY INJUR!t«> HIS !:RIVER *(.'(Mo!A* POLICE SAID)

Creatin:J more specific S-ATI'ACK-PERSON with features:

1. Lebowitz, M. Generalization an:) memory in an integrated urrlerstaooin:J system. Tech. Rept. 186, Yale University Deparbnent of Computer Science, 1980. PhD 'ftlesis