Consistency Checking vs. Restoring the Knowledge Base

[AdaptIt.Bob]

The following are some excerpts from email conversations between Bruce
Waters and myself:

Hi Bruce,

I am working on the Consistency checking document and in your
comments, requested a note that seems to say that restoring the
knowledge base is a simple way to resolve inconsistencies. I agree
that after doing a restore would make the knowledge base perfectly
consistent with the documents; but, it seems to me all this does is to
reintroduce faulty adaptations back into the knowledge base. In my
mind the primary purpose of consistency checking is to identify faulty
adaptations in the documents so that they can be rectified.

I guess I need you to “improve my mind” on this topic.

Thanks,

Bob

Here is Bruce's Reply:

Hi Bob,

It's a matter of conflicting definitions. Your idea of consistency is
"no faulty adaptations". Unfortunately, there is no algorithmic way
for AI to identify a faulty adaptation amongst all other adaptations.
What is a faulty one anyway? Misspelled maybe? Well, there is no way
yet to determine what is canonical versus wrong spelling, only the
user's eye and brain can do that. So the "faulty" idea has to be
abandoned. Consistency checking doesn't mean that, in Adapt It that
is.

Normal operation of AI means that source text gets adapted to same,
partially different, or totally different target text words, and
phrases. The associations between each source text word and phrase
that was adapted and the resulting target text word or phrase, are to
go into the KB. If that process is done perfectly, every such
association in the documents (at least he adapted parts of them)
visible at one or more places, will be found also in the KB.

Unfortunately, so many complex things happen along the way, that AI
might only temporarily store an association, or in ways not even I
understand it doesn't end up in the KB. To the extent that an expected
association does not appear in the KB, but does appear in one or more
documents, the documents and KB are not consistent. The inconsistency
is ALWAYS of the following kind: an association in a document is not
in the KB (but we reasonably expect it should be).

All that the Inconsistency Check process can do is to identify which
associations in the document(s) don't have a presence in the KB, and
show the user which they are so he can take some corrective action if
he wants.

You are right that if there are misspellings in the adaptations, then
"faulty" (meaning "misspelled") adaptations will be in the KB, and if
a restore of the KB is done using the File menu command for doing
that, it will just restore those misspellings along with all the other
correctly spelled adaptation strings. Yes, we can't avoid that. But
the restoration process doesn't "miss" associations, and it is quick,
so it can be a preferred way to get a good clean KB when the user
wants - more quickly than the Inconsistency Check way. But it won't
give the same KB if the user has also partly populated the KB by
importing associations from elsewhere. If he's done that, after the
restore he'll need to reimport those associations again, so that their
data will get back into the KB.

Hope that makes it clearer.
--Bruce