Conclusion
My own interest in natural language processing is generated by the belief
that a successful NL system would provide real insight into the mechanism
of the mind. Computers are fascinating devices, but the human mind is many
orders of magnitude more complex, and therefore more fascinating than any
present-day computer. I believe that computers will one day rival the brain
in the performance of most tasks requiring intelligence; as they do in certain
specific tasks already. But confident predictions ofthe arrival of that day have
been made frequently and just as often have been shown to be over-optimistic.
As each layer of complexity has been penetrated, hidden depths have been revealed.
I hope that this book will have communicated some of my own fascination in the subject.
The early part of the book is intended as an undergraduate study text, and for that reason
is punctuated with suggested exercises. To a large extent it is concerned with syntax, and
the development of algorithms to meet the demands of a parser which takes into account some
aspefts of the semantics. Some of the semantic problems were also explored in the context of
a very concrete application.
The second part of the book discusses some of the outstanding problems of semantic interpretation
and representation, and then describes some approaches to these problems. Some of the methods might
provide a vehicle for a group project, but they are not suitable for an individual student exercise.
To produce a non-trivial system in this field is a major undertaking. The volume of information which
must be supplied to make a system function should not be underestimated.
The third part ventures into areas where there are few guidelines in the public domain.
At some points I have been reduced to speculative suggestions, having been unable to find
any descriptions of systems which tackle these problems. Most of the published work in these
areas is confined to comment on the difficulties, without suggesting solutions.
Within these speculations the reader will perhaps have been able to detect a consistent view
which I have about NL processing. It is not an original view, because many others have proposed
similar schemes. It is not comprehensive as yet, but it might be summarised as follows:
Language is a vehicle for communication (Chapter 18).
Each word in a language is a label for a collection of stored information, which
is pieced together to form an overall representation of the meaning of a statement (section 18.3).
A representation consists of a set of 'states', each with its time-stamp, and representing
an elementary snapshot of the world. (Chapter 4, section 19.2, Chapters 21, 24, 25).
Each state is represented in terms of a 'perception' (Chapter 17).
Causal connections playa crucial role in the formation of representations. These
should be represented explicitly as states in their own right.
The representation of motivation is also crucial to a successful system.
'Truth' in the context of language interpretation is not the same thing as 'truth'
in the context offormal logic (Chapter 22).
The representation of an entity must represent its properties and its role. In representing the
role it must include many other aspects of the scenario in which it normally plays a part (Chapter 25).
Metaphor is an important part of language interpretation, which should not be
regarded as an 'optional extra' (Chapter 26).
The reader will have his/her own view. Even if I have not persuaded the reader that my view is
sound, however, I hope that I have demonstrated that the task of constructing a natural language
processing system is not to be undertaken lightly, and that many currently available systems which
purport to be 'natural language processing systems' are not that at all. The volume and detail of
the information which needs to be dermed (no matter what form of representation is used) is formidable,
and no system which does not tackle the problem of representing motivation can hope to capture the
meaning of a huge range of normal human speech and writing.
The first stage in the process of scientific investigation is that of hypothesis formation.
The next stage is the testing of the hypothesis. In this case that means the attempt to produce
a practical system based on the theoretical ideas of the hypothesis.
Have fun and good hacking!
[NOTE (added in 2004) In 1985, when this text was first published, the term 'hacking' did not
have the connotations of
illegal activity, which it has today. It referred to the process of sitting down at a computer
keyboard and 'hacking'
out a program and getting it to work - similar, in many ways to what was meant when people
referred to a 'hack' reporter.]