I had a play with this new super smart ChatBot this week. Broadly my thoughts on this whole process are generally unchanged but over time I’m coming to think about one extra aspect of automated translation. So here’s my rough idea.
- Communication is based on linguistic laws which are like laws of physics and cannot be broken.
- Due to those laws, the words themselves do not give sufficient information to the translator. We have to refer to our shared common knowledge (eg I did similar training to you in whatever subject area).
- AI seems to be focusing on the words themselves and this will only produce very coarse results.
- However, I guess there’s no reason an AI of some kind couldn’t start to refer to background shared knowledge the same way a human does. But…
- When an AI can do that, then effectively it will also be able to write freely and effectively think for itself, so not just translation, but all creative endeavors will be up in the air.
- There’s a sense that translation will be the thin end of the wedge, and will be easier to do well.
- I don’t accept that though, the same level of shared understanding and knowledge is required as for free form writing. Actually translation is frequently just like free form writing.
However, I’ve had a new thought recently.
- Controlled languages (where the authors are told they have to follow certain clear defined rules) are extremely easy to translate.
- In theory although controlled languages do not happen naturally, they are frequently used in business.
- Real languages and writing are absolutely not controlled languages.
- But…. actually if you look for example at real estate contracts – THIS IS THE NEW THOUGHT – they are actually acting a bit like controlled languages. Lawyers buy templates for even complex legal agreements, virtually all terms can be found in numerous other agreements, and those already exist in bilingual form.
- So actually some business documents may get very good results for automated translation systems.