2.2 - Database Access

In the course of improving the user interface's capabilities, introduction of speech-controlled systems is only a logical step. It is much more convenient to provide a way to speak to the computer or at least to pose questions in natural language to the machine (which is equivalent in the scope of our study) than applying any kind of programming language to get information about stored data,. Therefore, we need a system that can extract the user's wishes from the linear string that is given and subsequently give answers to the questions.
Again, we face the restriction that also applied to translation: restricted domain of the text in scrutiny. But, sticking to this rule, we can achieve somewhat satisfying results.

For example, covering only one season's results of American League Baseball, one of the first natural language processing front-ends to a database, named Baseball, was capable of answering simple questions posed by the user. 'Simple' means restricted vocabulary and complexity, but nevertheless this proved that systems of this kind were possible.
A similar application was the LUNAR system which allowed researchers to investigate the chemical properties of lunar rock samples returned with the Apollo 10 mission.
More impressive was the SHRDLU study [DHOF]. This program provided a front-end to a simple albeit practical task: controlling physical entities in space. Here, blocks, pyramids and similar objects were placed in a room and subsequently manipulated and interrogated. The physical part of all this was only a simulation, but it is easy to imagine how this could be implemented in hardware. This program was running on many different levels (lexical, syntactic, semantic and practical) which were closely interconnected. This attempt relied heavily on the notion that the sender wants to evoke certain impressions in the hearer's mind, and thus it worked by translating the user's input into procedures acting upon internal representations of the mentioned objects.
(Interestingly, recent studies [VGAL] suggest that the linguistic capabilities of humans might have evolved out of motoric predecessors. There seem to be centers in the ape brain that are activated both when the ape performs a certain action and when it only observes another ape doing the same thing. This "mirroring" could be an early form of a procedural representation of the world that reflects the environment in the brain - a very similar concept to SHRDLU's mechanisms.)

All of these examples have a limited range of vocabulary and grammar, which render it slightly impractical for the user to benefit from the system as it may always fall short from understanding certain constructs of the query.
This is a basic problem of all contemporary language processors: You never know what is accepted because you would need total insight for total comprehension, which is quite impossible for the time being. Besides, sometimes it is highly impractical to communicate one's wishes by language. When you need to specify certain options it may be faster to use graphic interfaces, and speech recognition interfaces are also inappropriate when it comes to entering confidential data like passwords; too easily one could overhear the input and abuse the gained information. Hence, secure transmission channels must be maintained in spite of more natural (and sophisticated) means of access.

back to main page...