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 />

1 Setting the stage: logics, prolog, theories<br />

1.1 Summary<br />

(1) We will use the programming language prolog to describe our language processing methods.<br />

Prolog is a logic.<br />

(2) We propose, following M<strong>on</strong>tague and many others:<br />

Each human language is a logic.<br />

(3) We also propose:<br />

a. Standard sentence recognizers can be naturally represented as logics.<br />

(A “recognizer” is something that tells whether an input is a sentence or not.<br />

Abstracting away from the “c<strong>on</strong>trol” aspect <strong>of</strong> the problem, we see a recognizer as taking the input<br />

as an axiom, and deducing the category <strong>of</strong> the input (if any).<br />

We can implement the deducti<strong>on</strong> relati<strong>on</strong> in this logic in the logic <strong>of</strong> prolog. Then prolog acts as a<br />

“metalanguage” for calculating pro<strong>of</strong>s in the “object language” <strong>of</strong> the grammar.)<br />

b. Standard sentence parsers can be naturally represented as logics.<br />

(A “parser” is something that outputs a structural representati<strong>on</strong> for each input that is a sentence,<br />

and otherwise it tells us the input is not a sentence.<br />

As a logic, we see a parser as taking the input as an axiom from which it deduces the structure <strong>of</strong><br />

the input (if any).)<br />

All <strong>of</strong> the standard language processing methods can be properly understood from this very simple<br />

perspective. 1<br />

(4) What is a logic? A logic has three parts:<br />

i. a language (a set <strong>of</strong> expressi<strong>on</strong>s) that has<br />

ii. a “derives” relati<strong>on</strong> ⊢ defined for it (a syntactic relati<strong>on</strong> <strong>on</strong> expressi<strong>on</strong>s), and<br />

iii. a semantics: expressi<strong>on</strong>s <strong>of</strong> the language have meanings.<br />

a. The meaning <strong>of</strong> an expressi<strong>on</strong> is usually specified with a “model” that c<strong>on</strong>tains a semantic<br />

valuati<strong>on</strong> fucti<strong>on</strong> that is <strong>of</strong>ten written with double brackets. So instead <strong>of</strong> writing<br />

semantic_value(socrates_is_mortal)=true<br />

we write<br />

[[socrates_is_mortal]] = 1<br />

b. Once the meanings are given, we can usually define an “entails” relati<strong>on</strong> ⊨, sothatforanyset<br />

<strong>of</strong> expressi<strong>on</strong>s Γ and any expressi<strong>on</strong> A, Γ ⊨ A means that every model that makes all sentences<br />

in Γ true also makes A true.<br />

And we expect the derives relati<strong>on</strong> ⊢ should corresp<strong>on</strong>d to the ⊨ relati<strong>on</strong> in some way: for<br />

example, the logic might be sound and complete in the sense that, given any set <strong>of</strong> axioms Γ we<br />

might be able to derive all and <strong>on</strong>ly the expressi<strong>on</strong>s that are entailed by Γ .<br />

So a logic has three parts: it’s (i) a language, with (ii) a derives relati<strong>on</strong> ⊢, and with (iii) meanings.<br />

1 Cf. Shieber, Schabes, and Pereira (1993), Sikkel and Nijholt (1997). Recent work develops this perspective in the light <strong>of</strong> resource logical<br />

treatments <strong>of</strong> language (Moortgat, 1996, for example), and seems to be leading towards a deeper and more principled understanding<br />

<strong>of</strong> parsing and other linguistic processes. More <strong>on</strong> these ideas later.<br />

4

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

Saved successfully!

Ooh no, something went wrong!