1990 noun classification from predicate argument

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Posi-ve Pointwise Mutual Informa-on (PPMI) is common 62 Dan Jurafsky Pointwise Mutual Informa1on •  Pointwise mutual informa1on: •  Do events x and y co ­occur more than if they were independent? PMI( X, Y ) = log 2 P( x, y) P ( x )P ( y ) •  PMI between two words: (Church & Hanks 1989) •  Do words x and y co ­occur more than if they were independent? PMI(word1, word2 ) = log 2 P(word1,word2 ) P(word1)P(word2 ) •  Posi1ve PMI between two words (Niwa & Nina 1994) •  Replace all PMI values less than 0 with zero Dan Jurafsky Compu1ng PPMI on a term ­context matrix •  Matrix F with W rows (words) and C columns (contexts) •  fij is # of -mes wi occurs in context cj W C pij = fij WC !! fij pi* = i=1 j =1 64 ! fij ! fij p pmiij = log 2 ij pi* p* j j =1 WC p* j = !! fij i=1 j =1 i=1 WC !! fij i=1 j =1 ! pmi # ij ppmiij = " #0 $ if pmiij > 0 otherwise Dan Jurafsky pij = fij WC !! fij i=1 j =1 W C ! fij ! fij p(w=informa-on,c=data) = 6/19 = .32 p(wi ) = j=1 11/19 = .58 p(w=informa-on) = N p(c=data) = 7/19 = .37 !"#$%&'()*(+ !"#$%&'( *$(+!"& 1211 $+,'*$$/' 1211 )+0+&*/ 1244 +,6"(#*&+", 1213 65 )*&* 1211 1211 1213 1275 !"%&'()*(+ 127: 1249 $+,!- ('.%/& 1213 1211 1213 1211 1211 1213 1211 1254 1244 1259 p( c j ) = i=1 N !"#+ .%0*( 1213 1213 1211 1211 1244 1244 1244 1254 1238 Dan Jurafsky pij pmiij = log 2 pi* p* j !"#$%&'( *$(+!"& 1211 $+,'*$$/' 1211 )+0+&*/ 1244 +,6"(#*&+", 1213 !"%&'()*(+ 1249 !"#$%&'()*(+ )*&* $+,!- ('.%/& 1211 1213 1211 1211 1213 1211 1213 1211 1213 1275 1211 1254 127: 1244 1259 !"#+ .%0*( 1213 1213...
View Full Document

This document was uploaded on 02/14/2014.

Ask a homework question - tutors are online