The lesson that is to be learnt is that should new research get underway it will be very important to have a suitable data-base ready.
I have in mind a natural-language document collection, probably using the full text of each document.
It should be constructed with many applications in mind and then be made universally available.*
Information retrieval systems are likely to play an every increasing part in the community.
They are likely to be on-line and interactive.
The hardware to accomplish this is already available but its universal implementation will only follow after it has been made commercially viable.
One major recent development is that computers and data-bases are becoming linked into networks.
It is foreseeable that individuals will have access to these networks through their private telephones and use normal television sets as output devices.
The main impact of this for IR systems will be that they will have to be simple to communicate with, which means they will have to use ordinary language, and they will have to be competent in their ability to provide relevant information.
The VIEWDATA system provided by the British Post Office is a good example of a system that will need to satisfy these demands.
By extending the user population to include the non-specialist, it is likely that an IR system will be expected to provide not just a citation, but a display of the text, or part of it, and perhaps answer simple questions about the retrieved documents.
Even specialists may well desire of an IR system that it do more than just retrieve citations.
To bring all this about the document retrieval system will have to be interfaced and integrated with data retrieval systems, to give access to facts related to those in the documents. An obvious application lies in a chemical or medical retrieval system.
Suppose a person has retrieved a set of documents about a specific chemical compound, and that perhaps some spectral data was given.
He may like to consult a data retrieval system giving him details about related compounds.
Or he may want to go on-line to, say, DENDRAL which will give him a list of possible compounds consistent with the spectral data.
Finally, he may wish to do some statistical analysis of the data contained in the documents.
For this he will need access to a set of statistical programs.
Another example can be found in the context of computer-aided instruction, where it is clearly a good idea to give a student access to a document retrieval system which will provide him with further reading on a topic of his immediate interest.
The main thrust of these examples is that an important consideration in the design of a retrieval system should be the manner in which it can be interfaced with other systems.
* A study recommending the provision of such an experimental test bed has recently been completed, see Sparck Jones and van Rijsbergen, 'Information retrieval test collections', Journal of Documentation, 32, 59-75 (1976).