Minds and Computers : An Introduction to the Philosophy of Artificial Intelligence

Figure 14.1
Phrase structure tree.
Figure 14.2
Phrase structure tree.

---

14.4 COMPUTING LANGUAGE
Determining the grammaticality of sentences of language according to
a generative grammar is clearly a computational procedure. It should
also be clear that given a phrase structure grammar and an arbitrary
string of its symbols, we can conduct a bottom-up search to determine
whether or not the string is generated by a phrase structure tree.
The example phrase structure grammar we constructed in the pre-
vious section is, of course, greatly simplified and considers only a
small fragment of the lexical items and syntactic structures of
English. Constructing a full generative grammar for a natural lan-
guage involves not just specifying the rules by which phrase structure
trees are constructed, but also specifying the various syntactic trans-
formations on terminal strings of phrase structure trees which
account for the myriad sentences native speakers produce.
This further element of syntactic transformation need not concern
us here. As always, the interested reader can follow the suggestions for
further reading or take an introductory course in generative grammar.
It su
ffices for our purposes to make the following observations.
It seems that the mechanisms which facilitate grammaticality
judgements in native speakers are computationally implementable.
Given that this is one of the functions implicated in the various
152
  
Figure 14.3
Phrase structure tree.

---

representational transformations involved in comprehending an
utterance of natural language, we have made some small progress
towards a computational account of the linguistic facility.
It should be clear to you now, however, just how complicated lin-
guistic behaviour is to account for. Although producing and compre-
hending written and spoken language is so natural to us as to appear
to be the most simple of processes, there are, in fact, a large number
of mediating procedures facilitating linguistic activity.
While we have seen – at least in part – how one of these procedures
might be accounted for in computational terms, there are still numer-
ous mechanisms we are in want of an account for. We will return to
some of these in later chapters. In Chapter 16 we are going to draw
out further evidence that linguistic behaviour is rule governed and,
hence, computationally implementable. In Chapter 19 we are going to
examine how we might model some of the mechanisms implicated in
reading written language.
For the moment, however, let’s finish this chapter with one final
problem. It appears that one of the most di
fficult elements of lan-
guage comprehension to account for is the determination of seman-
tics – meaning – from syntactic structures. Consider, for instance, the
fact that many English sentences are amphibolous.
Amphiboly is a property of sentences such that they admit of more
than one semantic interpretation as a result of their syntactic struc-
ture. This is distinct from cases of lexical ambiguity where a homonym
introduces the potential for multiple interpretations. Sentence [8]
below is lexically ambiguous, whereas sentence [9] is amphibolous:
[8]
The bank provided the pilot with a challenge.
[9]
I saw the man on the hill with the telescope.
While sentence [8] can be interpreted three ways, depending on the
semantic interpretation of ‘bank’, sentence [9] admits of multiple
interpretations but this ambiguity is not parasitic on the ambiguity of
any particular word in the sentence. Rather, there are a number of
ways in which we can interpret the syntactic structure of the sentence
and these give rise to distinct meanings.
This problem of the determination of semantics from syntax is one
we will return to at length in Chapter 17.
  
153

---

C H A P T E R 1 5
HUMAN REASONING
In the previous two chapters, we approached the rational and lin-
guistic faculties with a view to analysing their constituent mechan-
isms and accounting for these mechanisms in computational terms.
In this chapter and the next, we are going to examine evidence from
psychology and linguistics that bears on the question of whether
human rationality and linguistic competence are e
ffectively rule gov-
erned and, hence, computationally implementable.
You will be aided in our brief examination of empirical data con-
cerning human rationality in this chapter if you first answer the fol-
lowing reasoning problems:
1. If Mike is married then he is happy. Mike is married. Does it
follow that he is happy?
2. No hippies are financial advisors. No financial advisors are nuclear
protestors. Does it follow that some hippies are nuclear protestors?
3. There are four cards in front of you. Each card has a letter on one
side and a number on the other. The cards are lying with one face
up such that you can see the following on the showing faces:
A
K
4
7
You are told that this rule holds of these cards – If there is an A