Notes on computational linguistics.pdf - UCLA Department of ...
Notes on computational linguistics.pdf - UCLA Department of ...
Notes on computational linguistics.pdf - UCLA Department of ...
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 />
11 Semantics, discourse, inference<br />
A logic has three comp<strong>on</strong>ents: a language, a semantics, and an inference relati<strong>on</strong>. As discussed in §1, a<br />
computati<strong>on</strong>al device may be able to recognize a language and compute the inferences, but it does not even<br />
make sense to say that it would compute the semantics. The semantics relates expressi<strong>on</strong>s to things in the<br />
world, and those things are <strong>on</strong>ly relevant to a computati<strong>on</strong> to the extent that they are represented. For example,<br />
when the bank computes the balance in your account, the actual dollars do not matter to the computati<strong>on</strong>; all<br />
that matters is the representati<strong>on</strong>s that are in the bank’s computer. The interpretati<strong>on</strong> functi<strong>on</strong> that maps the<br />
numbers to your dollars is not computed. So typically when “semantics” is discussed in models <strong>of</strong> language<br />
processing, what is really discussed is the computati<strong>on</strong> <strong>of</strong> representati<strong>on</strong>s for reas<strong>on</strong>ing. The semantics is<br />
relevant when we are thinking about what the reas<strong>on</strong>ing is about, and more fundamentally, when we are<br />
deciding whether the state changes in a machine should be regarded as reas<strong>on</strong>ing at all.<br />
Standard logics are designed to have no structural ambiguity, but as we have seen, human language allows<br />
extensive ambiguity. (In fact, S6.6.3 shows that the number <strong>of</strong> different derivati<strong>on</strong>s cannot be bounded by any<br />
polynomial functi<strong>on</strong> <strong>of</strong> the number <strong>of</strong> morphemes in the input.) The different derivati<strong>on</strong>s <strong>of</strong>ten corresp<strong>on</strong>d<br />
to different semantic values, and so linguists have adopted the strategy <strong>of</strong> interpreting the derivati<strong>on</strong>s (or<br />
sometimes, the derived structures). But it is not the interpretati<strong>on</strong> that matters in the computati<strong>on</strong>al model;<br />
rather it is the syntactic analysis itself that matters.<br />
With this model <strong>of</strong> human language use, if we call the representati<strong>on</strong> <strong>of</strong> percieved sounds PF (for ‘ph<strong>on</strong>etic’<br />
or ‘ph<strong>on</strong>ological form’) and the representati<strong>on</strong> <strong>of</strong> a completed syntactic LF (for ‘logical form’), the basic picture<br />
<strong>of</strong> the task <strong>of</strong> the grammar is to define the LF-PF relati<strong>on</strong>. The simplest idea, and the hypothesis adopted here,<br />
is that LF simply is the syntactic analysis. We find closely related views in passages like these:<br />
PF and LF c<strong>on</strong>stitute the ‘interface’ between language and other cognitive systems, yielding direct<br />
representati<strong>on</strong>s <strong>of</strong> sound, <strong>on</strong> the <strong>on</strong>e hand, and meaning <strong>on</strong> the other as language and other systems<br />
interact, including perceptual and producti<strong>on</strong> systems, c<strong>on</strong>ceptual and pragmatic systems.<br />
(Chomsky, 1986, p68)<br />
The output <strong>of</strong> the sentence comprehensi<strong>on</strong> system…provides a domain for such further transformati<strong>on</strong>s<br />
as logical and inductive inferences, comparis<strong>on</strong> with informati<strong>on</strong> in memory, comparis<strong>on</strong> with<br />
informati<strong>on</strong> available from other perceptual channels, etc...[These] extra-linguistic transformati<strong>on</strong>s<br />
are defined directly over the grammatical form <strong>of</strong> the sentence, roughly, over its syntactic structural<br />
descripti<strong>on</strong> (which, <strong>of</strong> course, includes a specificati<strong>on</strong> <strong>of</strong> its lexical items). (Fodor et al., 1980)<br />
…the picture <strong>of</strong> meaning to be developed here is inspired by Wittgenstein’s idea that the meaning <strong>of</strong><br />
a word is c<strong>on</strong>stituted from its use – from the regularities governing our deployment <strong>of</strong> the sentences<br />
in which it appears…understanding a sentence c<strong>on</strong>sists, by definiti<strong>on</strong>, in nothing over and above<br />
understanding its c<strong>on</strong>stituents and appreciating how they are combined with <strong>on</strong>e another. Thus the<br />
meaning <strong>of</strong> the sentence does not have to be worked out <strong>on</strong> the basis <strong>of</strong> what is known about how<br />
it is c<strong>on</strong>structed; for that knowledge by itself c<strong>on</strong>stitutes the sentence’s meaning. If this is so, then<br />
compositi<strong>on</strong>ality is a trivial c<strong>on</strong>sequence <strong>of</strong> what we mean by “understanding” in c<strong>on</strong>necti<strong>on</strong> with<br />
complex sentences. (Horwich, 1998, pp3,9)<br />
In these passages, the idea is that reas<strong>on</strong>ing is defined “directly” over the syntactic analyses <strong>of</strong> the perceived<br />
language. Understanding an expressi<strong>on</strong> is nothing more than having the ability to obtain a structural analysis<br />
over basic elements whose meanings are understood.<br />
It might seem that this makes the account <strong>of</strong> LF very simple. After all, we already have our syntactic analyses.<br />
For example, the grammar from the previous chapter, gh5.pl provides an analysis <strong>of</strong> Titus be -s human:<br />
231