Is it possible to add vocabulary word lists to this implementation of the model?

[HugoPfeffer]

title.

I'm not very experienced, meaning this question may be really stupid, but can anyone help me?

[xmontero]

My use case is the following:

• I own a travel agency and sometimes we speak about names of providers, name of our own company, etc. We use many times "Catalan = ca" as our internal language for meetings.
• It usually gets "converted" into dictionary words. Say a provider is named "T. E. Ball" and the phrase is "Call the T. E. Ball guys" it might be transcribed as "Call the table guys". (But in catalan, just translated the use case into English).

So I want to pass a list of "own words" that if sounded "similar to that written spelling" then it's spelled as in the "custom dictionary".

I also think that provided that the LLMs use "proximity" in the vectors, maybe a list of words along with their definition could help decode the context. Something like:

definitions:
    T. E. Ball: "The name of a provider in Japan"
    Mr. Watanabe: "The boss of T. E. Ball"
    [...]

Does it make sense?