Translate Score DSL expressions to natural language

[fixtradingcommunity]

Translate Score DSL expressions to natural language

[donmendelson]

It's not hard to substitute words for symbols in a formula like "times" for "*". I have found that the awkward part is how to describe nested message elements in words, e.g. the PartyRole field of the fourth entry of the Parties repeating group. I'm looking for suggestions.

[kleihan]

I use the term "instance" and not "entry" and a shorter expression, i.e. "PartyRole of the fourth instance of Parties". The term "instance of" implies that the name of a repeating group will follow. In writing, one can additionally set the name into angle brackets and add a tag number to imply a field, i.e. "PartyRole(452) of the fourth instance of ".

[donmendelson]

Natural language processing is advancing very rapidly. However, new technologies require very large training databases, which seems highly unlikely for our specialized grammar.

[jimnup]

We should go back and look at natural language rules languages before we reinvent. There have been a few efforts, some of them have been around for quite some time. None have gained wide traction. As far as NLP, my overarching concern is people seem to be turning to robotic process automation (RPA) and machine learning (ML) including NLP in place of building appropriate systems with structured information. When is the use of RPA and NLP putting bandages or in the US vernacular "spit and bailing wire" and when are we expanding into uses of AI that extend beyond traditional systems. Tough calls these. Anyway, I think a natural language simplification that could be dictionary driven as opposed to the full force of NLP would make sense and in keeping with the power we are seeing with tablature.

…

On Sun, Jul 26, 2020 at 2:08 PM Don Mendelson ***@***.***> wrote: Natural language processing is advancing very rapidly. However, new technologies require very large training databases, which seems highly unlikely for our specialized grammar. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#13 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB7Z6SIVFBPN52YXBD6K4L3R5RWJJANCNFSM4DVMUTTQ> .

[donmendelson]

@jimnup I agree that a dictionary and rule based approach is more appropriate, and Orchestra provides a dictionary of message, field, and code names as objects to operate on. What is needed is natural language terms for relationships between the objects, e.g. knowledge that a code name is in the domain a field that is bound to a certain codeset. Useful phrases will never be as quite simple as a word-wise dictionary lookup, but the scope can be much more contained than free-form translation provided by neural networks.