Browse Prior Art Database

Probablistic Model of Grammar Using Dominance And Precedence

IP.com Disclosure Number: IPCOM000119652D
Original Publication Date: 1991-Feb-01
Included in the Prior Art Database: 2005-Apr-02
Document File: 1 page(s) / 37K

Publishing Venue

IBM

Related People

Sharman, RA: AUTHOR

Abstract

Typical grammars of Natural Languages are large, unwieldy, slow and suffer unacceptably from ambiguity. The concepts of precedence and dominance have been used by linguists to explain grammatical relations. A probabilistic model based on the grammatical relations of dominance and precedence, derived from regular context-free phrase-structure grammar (CF-PSG) is disclosed. 1) First assume that every rule of CF-PSG is rewritten as two rules, one showing the dominance of the left-hand side over the symbols of the right-hand side, and the other showing the precedence of the order of symbols on the right-hand side. Thus A->BCD becomes: A->(B, C, D) and (B, C, D) -> BCD (this is well known).

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 85% of the total text.

Probablistic Model of Grammar Using Dominance And Precedence

      Typical grammars of Natural Languages are large,
unwieldy, slow and suffer unacceptably from ambiguity.  The concepts
of precedence and dominance have been used by linguists to explain
grammatical relations.  A probabilistic model based on the
grammatical relations of dominance and precedence, derived from
regular context-free phrase-structure grammar (CF-PSG) is disclosed.
1)   First assume that every rule of CF-PSG is rewritten as two
rules, one showing the dominance of the left-hand side over the
symbols of the right-hand side, and the other showing the precedence
of the order of symbols on the right-hand side.  Thus A->BCD becomes:
A->(B, C, D) and (B, C, D) -> BCD (this is well known).
2)   For each pair of symbols in the grammar, (A, B), calculate (by
observation from the grammar, assuming frequency information is
available for each rule) (i)   The dominance probabilities PD(B/A) of
A dominating B (ii)  The precedence probabilities Pp(B/A) of A
preceding B. (iii) The fertility probability Pf(A) which is the
number of sub- constituents of the left-hand side machine.
3)   Approximate the likelihood of a particular constituent as the
product of all the relevant dominance probabilities and the
precedence probabilities.  This assumes (incorrectly) that the
dominance and precedence is independent, but it is a good
approximation.
      Thus, P(A->BC) Z PD(B/A). PD(C/A). Pp(C/B). Pf(2/A)
...