20.07.2013 Views

Notes on computational linguistics.pdf - UCLA Department of ...

Notes on computational linguistics.pdf - UCLA Department of ...

Notes on computational linguistics.pdf - UCLA Department of ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Stabler - Lx 185/209 2003<br />

6.4 All the GLC parsing methods (the “stack based” methods)<br />

(23) LC parsing uses a rule after establishing its leftmost element. We can represent how much <strong>of</strong> the right<br />

side is established before the rule is used in the following way:<br />

s:-np][vp<br />

LL parsing uses a rule predictively, without establishing any <strong>of</strong> the right side:<br />

s:-][np , v p<br />

LR parsing uses a rule c<strong>on</strong>servatively, after establishing all <strong>of</strong> the right side:<br />

s:-np, vp][<br />

Let’s call the sequence <strong>on</strong> the right that triggers the use <strong>of</strong> the rule, the trigger.<br />

In a rule like this with 2 c<strong>on</strong>stituents <strong>on</strong> the right side, these 3 opti<strong>on</strong>s are the <strong>on</strong>ly <strong>on</strong>es.<br />

This observati<strong>on</strong> is made in Brosgol (1974), and in Demers (1977).<br />

(24) In general, it is clear that with a rule that has n elements <strong>on</strong> its right side, there are n + 1opti<strong>on</strong>sfor<br />

the parser.<br />

Furthermore, the parser need not treat all rules the same way, so in a grammar like the following, the<br />

number <strong>of</strong> parsing opti<strong>on</strong>s is the product <strong>of</strong> the number <strong>of</strong> ways to parse each rule.<br />

(25) As Demers (1977) points out, the collecti<strong>on</strong> <strong>of</strong> trigger functi<strong>on</strong>s F for any grammar can be naturally<br />

partially ordered by top-downness:<br />

F1 ≤ F2 if and <strong>on</strong>ly if for every producti<strong>on</strong> p, the trigger F1(p) is at least as l<strong>on</strong>g as F2(p).<br />

In other words, a setting <strong>of</strong> triggers F1 is as bottom-up as F2 if and <strong>on</strong>ly if for every producti<strong>on</strong> p, the<br />

triggering point defined by F1 is at least as far to the right as the triggering point defined by F2. It is<br />

easy to see that 〈F, ≤〉 is a lattice, as Demers claims, since for any collecti<strong>on</strong> F <strong>of</strong> trigger functi<strong>on</strong>s<br />

for any grammar, the least upper bound <strong>of</strong> F is just the functi<strong>on</strong> which maps each rule to the trigger<br />

which is the shortest <strong>of</strong> the triggers assigned by any functi<strong>on</strong> in F, and the greatest lower bound <strong>of</strong><br />

F is the functi<strong>on</strong> which maps each rule to the trigger which is the l<strong>on</strong>gest assigned by any functi<strong>on</strong> in<br />

F. Furthermore, the lattice is finite. 25 We call this lattice <strong>of</strong> recogniti<strong>on</strong> strategies the GLC lattice. The<br />

simple lattice structure for a 3 rule grammar can be depicted like this:<br />

86

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!