One of the things that sets the human apart from other species in the animal world, is his ability to communicate. Without the tool of language the development of our species to the stage where we can begin to comprehend the science beyond our existence, would, in all liklihood, be beyond our limitations.
The key areas of communication are:
* To give information
* For that information to be understood
* For indication to be given that the communicated information has been understood.
The ``Natural Language Problem'' is the name given to that area of AI development which deals exclusively with making machines interact on the level of conversation with their human users.
The machine itself has the language of binary code. Strings of 0's and 1's, representing a switch in an on or off state, will enable the computer to perform logical calculations and operations. By teaching the computer to recognise a set pattern of these binary strings as representing some highly specific instruction and matching these strings to key words or symbols in a programming language, a computer programmer can use a defined set of instructions to build more complex problems for the machine to deal with.
For the mere mortal, it might be nice to be able to talk to a machine as one might to an assistant or employee, to deliver instructions in ones native language, without the bother of learning one or maybe several intricate programming languages and then utilising them, or worse still sitting down and individually setting several billion switches to their appropriate binary settings. Here, then, is the task of the AI programmer dealing with the Natural Language Problem; to teach a machine to learn the intricacies of human language rather than vice versa.
The machine may be taught to recognise nouns, verbs, conjunctions, an so on. To give meaning to each word, or to search only for keywords, and discarding all others, to follow the instructions it interprets. Different machines may be taught to recognise written instructions, or to match oral patterns to a memory of pre-designated commands.
It can take years for a human to master just one language, and longer to teach a computer to do the same. How long then might it take to teach a computer to recognise so many languages, each with its own dialect, that any individual, the whole world over, could sit before it and instruct it with as much ease and simplicity as one might with a fellow human?
An example of a primitive ``talking'' machine, designed in 1966 by Weizenbaum and Colby, is Eliza, the algorithm for which has been used and abused by many people over the ensuing decades. An example of a much degraded version of the program is attached in Appendix A.
For the time being let us examine the algorithm to understand how, based loosely on human intuition, a machine can be taught to think and talk.
As can be seen from the flowchart below, this program works on quite a simple algorithm. The conversation may not be the most stimulating the user might hope, but it will be relatively coherent. The greater the expansion and development of library information afforded to the machine the more realistic the program will seem.
A machine can also be taught to recognise sentence and
grammatical structures. By matching more than one aspect of the
sentence, for example all the nouns and the verbs joining them,
and working from a greatly expanded library, or even a limited
library with a different algorithm to construct sentences from
several clauses, conversation may become more developed.
While this program is used mainly for entertainment, with appropriate adjustment it could be used as a reference tool, working off vast encyclopediae of information.
Seeing this simple ``talking'' program, it is apparent how the concept could be applied to other areas. By instructing a computer to perform a task and asking it to mirror the information it thinks it has been given for verification before proceeding, processes can be carried out without the necessity of the human learning the language of the machine, sometimes a highly arduous task.