"... Natural language query technology is underused ..."
Because language itself is plastic, evolves and changes meaning. Users may ask a clear logical question and phrase it with language that may totally miss the restricted, non-plastic language domain of the data they want to query. [0]
[0] Bower, Bruce, Shifty Talk: Probing the process of word evolution
And this is why techniques for rapid authoring of NLIs are so crucial. And why the approach should not be based on idealizations of languages, but rather on ad hoc phrasal patterns.
The actual idea is simple. Natural language as you correctly point out is beyond any clean/perfect theory. Besides it's constantly changing.
An admin needs to constantly patch an NLI system to achieve high recall (e.g. over 90% of the reasonable questions posed can be answered) and high precision (e.g. over 90% of those questions that are answered are answered correctly).
In all cases in which the system answers, it must paraphrase the query back to the user so that they understand if they were misunderstood.
Of course any user will always be able to ask a question that the NLI will fail on. For example asking an inventory database "what's the meaning of life" will not lead to an intelligent response.
Users, vendors and analysts agree that though little training is required, the problem with English language query tools is the users, not the software.
This quote in the end tells how great the current products are.
6 comments
[ 3.4 ms ] story [ 29.6 ms ] threadBecause language itself is plastic, evolves and changes meaning. Users may ask a clear logical question and phrase it with language that may totally miss the restricted, non-plastic language domain of the data they want to query. [0]
[0] Bower, Bruce, Shifty Talk: Probing the process of word evolution
http://www.sciencenews.org/articles/20071013/fob2.asp
Haven't heard of this. How do you do this? Any cites?
The actual idea is simple. Natural language as you correctly point out is beyond any clean/perfect theory. Besides it's constantly changing.
An admin needs to constantly patch an NLI system to achieve high recall (e.g. over 90% of the reasonable questions posed can be answered) and high precision (e.g. over 90% of those questions that are answered are answered correctly).
In all cases in which the system answers, it must paraphrase the query back to the user so that they understand if they were misunderstood. Of course any user will always be able to ask a question that the NLI will fail on. For example asking an inventory database "what's the meaning of life" will not lead to an intelligent response.
This quote in the end tells how great the current products are.
http://www.youtube.com/watch?v=fWio8bHq4wQ http://www.anaphoric.com