Download presentation
Presentation is loading. Please wait.
1
Prof. Pushpak Bhattacharyya, IIT Bombay
CS460/IT632 Natural Language Processing/Language Technology for the Web Lecture 20 (28/03/06) Prof. Pushpak Bhattacharyya IIT Bombay Probabilistic Parsing 28/03/06 Prof. Pushpak Bhattacharyya, IIT Bombay
2
Example Grammar – Syntax Items
S NP VP 0.8 VP 0.2 NP n 0.4 n PP 0.4 n NP 0.2 VP v 0.3 v NP 0.3 v PP 0.2 v NP PP 0.2 PP p NP 1.0 28/03/06 Prof. Pushpak Bhattacharyya, IIT Bombay
3
Example Grammar - Lexical Items
p like 1.0 v swat 0.2 flies 0.4 like 0.4 n swat 0.05 flies 0.45 ants 0.5 28/03/06 Prof. Pushpak Bhattacharyya, IIT Bombay
4
Prof. Pushpak Bhattacharyya, IIT Bombay
Language Model Language Model can be categorized into two types – N-grams Probabilistic context free grammar (PCFG) A Language Model either accepts or generates a piece of text. 28/03/06 Prof. Pushpak Bhattacharyya, IIT Bombay
5
Input Sentence & Interpretations
Let the input sentence be – “Swat flies like ants” (in the sense like – “Time flies like an arrow”) Swat means – n : a violent blow v : to render a violent blow Interpretations – Swat is the subject of fly. Swat is a qualifier for fly (some kind of fly). Swat is a verb with ‘fly’ as an object. 28/03/06 Prof. Pushpak Bhattacharyya, IIT Bombay
6
Parse Tree for Interpretation-1
NP VP v ants PP p flies like swat 28/03/06 Prof. Pushpak Bhattacharyya, IIT Bombay
7
Parse Tree for Interpretation-2
NP VP v ants like swat flies 28/03/06 Prof. Pushpak Bhattacharyya, IIT Bombay
8
Parse Tree for Interpretation-3
PP VP v flies NP swat n like ants 28/03/06 Prof. Pushpak Bhattacharyya, IIT Bombay
9
Prof. Pushpak Bhattacharyya, IIT Bombay
Comparison – CFG & PCFG CFG Corpus insensitive Does not deal with ambiguity effectively Fails in case of noise Does not scale up PCFG Corpus sensitive Principled way of dealing with ambiguity Deals with noise Scales up (have good learning algorithms) 28/03/06 Prof. Pushpak Bhattacharyya, IIT Bombay
10
Meaning of the Probability of a Rule
P( A α : p ) Probability of the rule A α is p Definition- We say that in a derivation sequence D, Nj dominates tkl where w1 w2 …… wn tkl is a segment of the sequence w1, w2,…wn iff tkl is generated (directly or indirectly) from Nj by a set of application of production rules. 28/03/06 Prof. Pushpak Bhattacharyya, IIT Bombay
11
Meaning of the Probability of a Rule (Contd.)
Then the probability of a parse tree t given the sentence s ( = w1 , w2 ,…wn) is defined as P(w1 w2 … wn | s) = P(X1k1, Xk1k2,… Xkm,n , w1, w2, … wn | s1,n) 28/03/06 Prof. Pushpak Bhattacharyya, IIT Bombay
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.