Natural language

David Jeske
Tue, 29 Dec 1998 09:21:27 -0800

On Tue, Dec 29, 1998 at 01:48:54AM +0100, Laurent Martelli wrote:
> Natural language can be of a great value sometimes, when you have a
> rather complex problem that you want to solve, because of its
> fuzzyness. However, this fuzzyness can be a problem for common actions
> : it can lead to disastruous quiproquos. How shoul the system
> understand a statement like "delete all"? What context should be taken
> into account? What is "all"? 

I don't think this is a big problem for two reasons:

1) everything should be 'undoable'

2) I know this is a specific case, but I don't think 'delete all' in
the unix "rm -rf /" sense is going to be a valid high-level
operation. I don't want the system to ever delete anything, so if I
asked it to 'delete all', it can infer whatever it wants about what I
might want 'removed from the screen in front of me', but ultimately,
that data should always remain. I'm willing to accept that if I really
want to delete everything, I'm going to have to go jump through quite
a few hoops. If we generalize, because of #1 above, I expect that in a
natural language interface, there would be no descructive
operations. Just like communication between two people, I envision a
natural langauge interface more like a negotiation between you and the
computer, where it may take a few iterations to refine your words to
the point that the computer understands what you _mean_ not what you

> Moreover, I think it is a user interface issue. This means that we
> must be able to implement all kinds of UI to interact with the system
> : command line, "graphic" browsing ... All of them should be
> independant from the core design of the system. We must be able to
> build an interface for visually impaired people which requires no
> modification to existing "programs".(1) We should also be able to
> interact with the system seemlessly through various devices (text or
> graphics consoles ...). 


> So my conlusion is : yes it would be great, but I do not place it with
> a very high priority in the todo list. However, once we have a core
> system running, some linguists should be able to experiment with our
> system in a seemless fashion, without the need to "hack" every where.


> (1) I use quotes around program since I consider this as an has-been
>     notion taken from has-been systems like Unix. It fixes arbitrary
>     limits in design/code/pattern reuse.

agreed as well.

David Jeske (N9LCA) + +