Re: what AI tool can we use to generate the AI KR CG vocabulary

Thanks Paola!

If you can point me to the document or documents that the vocabulary 
should be generated from, I can create an initial draft. A collection of 
documents with overlapping material is a good starting point. It can be 
interesting to see which terms come up in which documents.

It may also help if you can indicate what kinds of term the vocabulary 
should contain. For example, I'm collecting key terms for some reference 
architecture standards, and am finding that asking the LLM to find 
things like systems, components, designs etc. can give better results 
than just asking it for key terms.

There are many possible output formats. The simplest is probably HTML 
dl/dt/dd. How far does the group want to formalise things like "see 
also" and allow for a term to have different definitions in different 
contexts? This can get quite complicated.

Pete's question about how the vocabularies are to be used is a good one. 
For example, will the people who use them have a technical background, 
or are they intended for the general public?

On 03/12/2024 21:41, Paola Di Maio wrote:
> @chris  thanks for the offer- yes i expect some manual input to complete 
> the process but maybe with some training..
> could become fully automated. let me know how to proceed any too to 
> support the effort is welcome
> 
> @Peter Rivett <mailto:pete.rivett@federatedknowledge.com>
> 
>     It's not clear to me in all this what format(s) existing glossaries/
>     vocabularies in step 3 (do we have a list?) 
> 
> <Milton pointed to two (ISO and another) in a related email, but we 
> could identify more  I entered the link the in wiki page
> 
>     re expected to be in, and what the required generated output format
>     should be.
> 
> a nice plain language to start with, with any encoding of choice,  could 
> be any ML
> 
>     And, for step 4, what degree of difference should lead to a new
>     vocabulary entry. 
> 
> as defined by the user - that is: are you satisfied that the current AI 
> standards so far (as pointed by Milton and possibly to be expand) 
> represent the AI KR domain correctly? I am not- current AI standard 
> vocabularies do not represent the AI KR domain adequately
> There may be KR terms  already in existing AI standards that need to be 
> disambiguated/defined further
> that is where we can make an additional contribution
>   1st.contribution is to identify the terms that are overlooked in 
> existing AI standards 2nd.nd contribution to provide alternative KR 
> definitions  if we are not satisfied with the existing ones
> 
>     And how the new entry should reference the existing one (e.g. as
>     some sort of "similar" or specialized term).
> 
> good question- I think It could simply list the existing term as defined 
> in existing standards as an attribute
> (previous definitions?)  In fact, I may add another field to the entry 
> form 'overlapping domain' as some AI KR terms could be defined elsewhere 
> * in medicine vs law for example
> 
>     And, in the resultant ecosystem, how the whole family of
>     vocabularies should be represented; both the existing ones that can
>     be reused and the new differentiated terms. 
> 
> Uhm, that is a very big question- ecosystems tend to sort themselves 
> out, so as long as they are published on the internet, but suggestions 
> welcome
> 
>     It would also seem desirable to be able to indicate the specific
>     terms in the existing vocabularies that have been deemed reusable.
> 
> think about a way of doing that and share it here, perhaps
> 
> 
>     To answer my own question somewhat, there is the OMG Multiple
>     Vocabulary Facility https://www.omg.org/spec/MVF <https://
>     www.omg.org/spec/MVF> that provides for different vocabularies
>     (terms and definitions related to Communities and with mutual import
>     relationships) with the terms mapped to their meaning as a concept.
> 
> THANKS that s great, can you add to the wiki? i can do it if I remember it
> 
>     However a lot depends on how we expect the set of vocabularies to be
>     used and by whom (type/role of person or machine)?
> 
> for now, I feel that I would be satisfied if we could, after about six 
> years of looking into AI KR
> can point out with some efficiency and precision the concepts/terms not 
> yet covered in AI standards
> Nice to be able to put our finger into some open wound with clarity and 
> precision
> i guess the end result would be used by both humans and machines in 
> whatever way other resources are used, in the same way that we find use 
> any directory
> I am itching to identify some terms and concepts and invite everyone to 
> do the same
> 
> Ponder one or two terms close to your heart and mind
> I am starting with:
> misrepresentation (justification is:fake AI)
> malicious confounding (justification is:deliberate use of 
> misrepresentation to mislead)
> 
> 
>     Pete
> 
>     Pete Rivett (pete.rivett@federatedknowledge.com
>     <mailto:pete.rivett@federatedknowledge.com>)
>     Federated Knowledge, LLC (LEI 98450013F6D4AFE18E67)
>     tel: +1-701-566-9534
>     Schedule a meeting at https://calendly.com/rivettp <https://
>     calendly.com/rivettp>
> 
>     ------------------------------------------------------------------------
>     *From:* Chris Harding <chris@lacibus.net <mailto:chris@lacibus.net>>
>     *Sent:* Tuesday, December 3, 2024 9:37 AM
>     *To:* paoladimaio10@googlemail.com
>     <mailto:paoladimaio10@googlemail.com> <paoladimaio10@googlemail.com
>     <mailto:paoladimaio10@googlemail.com>>
>     *Cc:* W3C AIKR CG <public-aikr@w3.org <mailto:public-aikr@w3.org>>
>     *Subject:* Re: what AI tool can we use to generate the AI KR CG
>     vocabulary
>     Hi Paola
> 
>     I am working on a commercial tool in this area. I can do 1 and 2: they
>     are no problem with modern NLP technology but, as with anything produced
>     by AI, need final human review. 3, 4 and 5 should be fairly
>     straightforward also, but would need some work on the mechanism for
>     importing existing glossaries/vocabularies.
> 
>     On 03/12/2024 03:05, Paola Di Maio wrote:
>      > Knowledgeable CG members
>      >
>      > Since the future is here, suggestions as to how to generate
>      > a vocabulary, would be great (preferably open, free, online tool or
>      > python script?)
>      > Process
>      > 1.select source of K (upload or point to source doc)
>      > 2. extract key terms and definitions
>      > 3. compare key terms with existing glossaries/vocabularies already
>      > published (input URLs)
>      > 4. if not included in existing resources in 3 OR
>      >          if definition is different from the same term in existing
>      > resource in 3
>      > THEN
>      > 5. include term in a list (to be discussed, evaluated, refined)
>      >
>      > I definitely buy drinks is someone can do
> 
>     --
>     Regards,
>     Chris
>     ++++
> 
>     Chris Harding
>     Chief Executive, Lacibus Ltd
> 
> 

-- 
Regards,
Chris
++++

Chris Harding
Chief Executive, Lacibus Ltd

Received on Wednesday, 4 December 2024 14:51:58 UTC