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

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Stabler - Lx 185/209 2003<br />

(3) Dalrymple and Kaplan (2000), Bayer and Johns<strong>on</strong> (1995), Ingria (1990), and others have pointed out that<br />

agreement seems not always to have the “two way” character <strong>of</strong> unificati<strong>on</strong>. That is, while in English,<br />

an ambiguous word can be resolved <strong>on</strong>ly in <strong>on</strong>e way, this is not always true:<br />

a. The English fish is ambiguous between singular and plural, and cannot be both:<br />

The fish who eats the food gets/*get fat<br />

Thefishwhoeatthefood*gets/getfat<br />

(This is what we expect if fish has a number feature that gets unified with <strong>on</strong>e particular value.)<br />

b. The Polish wh-pr<strong>on</strong>oun kogo is ambiguous between accusative and genitive case, and can be both:<br />

Kogo<br />

who<br />

Janek lubi a Jerzy nienawidzi?<br />

Janek likes and Jerzy hates<br />

c.<br />

(lubi requires acc object and nienawidzi requires gen object.)<br />

The German was is ambiguous between accusative and nominative case, and can be both:<br />

Ich habe gegessen was übrig war.<br />

I have eaten what left was<br />

(The German gegessen requires acc object and übrig war needs a nom subject.)<br />

Dalrymple and Kaplan (2000) propose that what examples like the last two show is that feature values<br />

should not be atoms like sg, pl or nom, acc, gen but (at least in some cases) sets <strong>of</strong> atoms.<br />

3.2 More unificati<strong>on</strong> grammars: case features<br />

(4) We can easily extend the grammar g2.pl to require subjects to have nominative case and objects,<br />

accusative case, just by adding a case argument to dp:<br />

% g3.pl<br />

:- op(1200,xfx,:˜).<br />

ip :˜ [dp(P,N,nom), i1(P,N)]. i1(P,N) :˜ [i0, vp(P,N)]. i0 :˜ [].<br />

dp(1,s,nom) :˜ [’I’]. dp(2,_,_) :˜ [you]. dp(3,s,nom) :˜ [she].<br />

dp(3,s,nom) :˜ [he].<br />

dp(3,s,_) :˜ [it].<br />

dp(3,p,nom) :˜ [they].<br />

dp(1,s,acc) :˜ [’me’]. dp(3,s,acc) :˜ [her].<br />

dp(3,s,acc) :˜ [him].<br />

dp(3,p,acc) :˜ [them].<br />

dp(3,s,_) :˜ [titus]. dp(3,s,_) :˜ [tamora]. dp(3,s,_) :˜ [lavinia].<br />

dp(3,N,_) :˜ [d1(N)]. d1(N) :˜ [d0(N), np(N)]. d0(_) :˜ [the].<br />

d0(p) :˜ [most].<br />

d0(s) :˜ [every]. d0(s) :˜ [some]. d0(p) :˜ [few].<br />

np(N) :˜ [n1(N)]. n1(N) :˜ [n0(N)]. n0(s) :˜ [penguin].<br />

n0(p) :˜ [penguins].<br />

n0(s) :˜ [s<strong>on</strong>g]. n0(p) :˜ [s<strong>on</strong>gs].<br />

vp(P,N) :˜ [v1(P,N)]. v1(P,N) :˜ [v0(intrans,P,N)].<br />

v1(P,N) :˜ [v0(trans,P,N),dp(_,_,acc)].<br />

v0(_,1,_) :˜ [sing]. v0(_,2,_) :˜ [sing]. v0(_,3,s) :˜ [sings].<br />

v0(_,3,p) :˜ [sing].<br />

v0(trans,1,_) :˜ [praise]. v0(trans,2,_) :˜ [praise]. v0(trans,3,s) :˜ [praises].<br />

v0(trans,3,p) :˜ [praise].<br />

v0(intrans,1,_) :˜ [laugh]. v0(intrans,2,_) :˜ [laugh]. v0(intrans,3,s) :˜ [laughs].<br />

v0(intrans,3,p) :˜ [laugh].<br />

(5) The coordinate structure Tamora and Lavinia is plural. We cannot get this kind <strong>of</strong> c<strong>on</strong>structi<strong>on</strong> with<br />

rules like the following because they are left recursive, and so problematic for TD:<br />

dp(_,p,K) :˜ [dp(_,_,K), coord(dp(_,_,K))]. % nb: left recursi<strong>on</strong><br />

vp(P,N) :˜ [vp(P,N), coord(vp(P,N))]. % nb: left recursi<strong>on</strong><br />

coord(Cat) :˜ [and,Cat].<br />

42

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

Saved successfully!

Ooh no, something went wrong!