AI 2 Exam
			(September Resit)
Date : 17.09.86					Time : 14.00-17.00

				Examiners : Chairman : J A M Howe
				            External : Alan Turing
				                     : Charles Babbage
				                     : ELIZA

1. Candidates in the 3rd or later year of the degrees MA(General),
LLB, MSC, LBW or GCHQ should write :
	"I've got this far, just try and stop me now."
after their names in the script book.

2. Answer four questions. Do not attempt to answer more than two questions

3. All questions carry equal marks. 25 marks will be awarded for each
question attempted. PhDs will be awarded for any question correctly


1) Natural Language.
	(a) Construct an RTN for the English language, including
subordinate clauses, relative pronouns, passive sentences and all possible
regional and dialect variations, including fragmentary utterances.
Indicate how each sentence could be converted into 5th-order predicate
calculus, with mechanisms to represent time and causality.

	(b) Convert the RTN above to an ATN, and add a speech-understanding
mechanism of your own design. The ATN should generate as well as
parse all possible sentences which could be understood by an English

2) Natural Language
	(a) Describe 'War & Peace' in terms of predicate calculus, explaining
the meaning of each predicate that you introduce.
	(b) Translate Proust's 'A la recherche du temps perdu' into
Schank's conceptual dependency notation.

3) Expert Systems
	Implement an expert system in PROLOG to diagnose all possible
diseases of the higher primates. The system should be capable of
conversing in English with the user, and after identifying the disease,
should explain its reasoning in terms which could be understood by a
naive user, and recommend appropriate therapy. The system should be
aware of all drugs currently listed in the British National Formulary,
and should have the option of recommending surgery if necessary.

4) Expert Systems
	The major problem in expert systems design is knowledge acquisition.
Design a system to acquire knowledge from an expert telepathically,
formalise this knowledge, and implement it as an expert system based on
production rules.
5) Planning & Search
	(a) Represent the politics of the Middle East as a state space, taking
into account the affect of Reagan, political assassination and the value
of the shekel, together with any other factors you feel may be relevant.
An example format for operators is given below :

Operator        Preconditions          Delete List     Add List
bomb_libya      crazy(reagan)          tripoli         a huge crater

Search this state space to find a formula for peace.
	(b) Repeat the above question using problem reduction.

6) Planning & Search
	The monkey and bananas problem has been complicated by the following
additional factors :
	  (i) The bananas will not be ripe for two weeks.
	 (ii) The monkey is blind.
	(iii) The bananas are locked in a safe.
	 (iv) There are no suitable boxes present.
	  (v) The monkey is afraid of heights.
	 (vi) The floor of the room has been electrified.
	(vii) The monkey doesn't like bananas anyway.
Describe how STRIPS could be used to solve the problem given these
conditions, noting any operators you introduce.

7) Vision
	(a) Implement a complete vision system. You should give attention to
both the low-level and high-level aspects of this system, as follows :
	Low-level : - resolution should be sufficient to read the small print
on Insurance Policies, and the system should represent a full range of
	High-level :- the system should be capable of analysing a scene in
terms of edges, vertices, faces, regions, bodies and support relations. It
should also identify any humans present in the scene by name, age and sex,
and provide a reasonable guess at their shoe size and waist measurement.
	(b) Extend the system to make use of information from beyond the normal
visual spectrum, paying special attention to the Far Ultra-Violet, and
Extremely Low Frequency radio transmissions. With respect to the latter, the
system should be screened against possible interference from passing
nuclear submarines.

8) Vision
	Apply Waltz's line-labelling technique to the 'Mona Lisa', and
produce a list of all bodies present. Why does it fail ? How could it
be extended to take into account features present in this scene, such
as the smile ?