AI implementation in TBMS system

AI implementation in TBMS system

AI is a hype nowadays, so any software developer is asked if their system uses AI. Hereby we will try to explain how AI can and can’t be used in Translation business management systems.

Basic principle

AI is not magic, it learns from humans. E.g. if you want AI to translate, you have to upload millions of human translations (the more the better), it makes statistical analysis of the uploaded dataset, and then it tried to mimic the human translator. The quality of AI translation output depends on the volume and quality of the dataset it learns from.

And it works the same in every application, If you want to AI to paint pictures, you have to upload huge number of pictures and their descriptions.  If you want AI to make medical diagnosis, you have to upload the health-related data of many millions of patients. And in each case, the quality depends on dataset. If dataset is full of human mistakes, the AI will repeat those mistakes (that’s why machine translation always needs to be checked). If some important data is missing from dataset, AI won’t guess them. You need to feed all variables to AI to teach it to do something.

Limitations in translation business and project management systems

So, what if we teach AI to manage projects, assign jobs to translators, create invoices, track payments, and so on? The idea sounds great. Why not implement AI into Translation Business Management system, let it watch how you are working for a year or two, and then ask it to do the job? Just like we are doing with the translations.

First, you need to agree on being constantly watched by AI of TBMS. Every your action in should be recorded and used in dataset analysis. For some people this alone may be scary, but even if it does not hold you, there is another problem.

There is a huge part of actions done by a project manager outside translation business management system, so AI won’t see them. E.g.:

- You are getting a query or quote request from the customer, review it, assess the cost, check the availability of linguists, and then reply. And then the decision is made depending on that investigation. Most of it is done outside TBMS, so AI won’t have anything to learn from
The client makes decision by himself and sometimes may ask to change the conditions, and you as human make the decision. AI is still out of the game.

- The client sends you a translation project in email or assigns it to you through their portal. So you have to move it manually to TBMS unless you have some automated tool to make it. You still have to analyses if you may complete it in time, check if translators and proofreaders are available, ask the questions, discuss the deadlines, possibly process the files. Most of this is happening in the head of Project Manager, not in TBMS, and different tools are used.

- Watch the deadlines, react to client’s updates, discuss whatever surprises come out, be ready to replace the linguist if he gets unavailable, make sure that interim deliveries are ok and so on.

- Check the files delivered, and deliver them to the customer as they instructed you in the first email.

So many actions are happening outside TBMS. You may use phone/messenger/video calls, emails, spreadsheets, and sometimes even piece of paper for planning, and it requires a lot of thinking and communication. AI in TBMS does not see these additional actions, so it won’t learn from you. So it can’t replace a human project manager if the project is not all predefined and standard, if even minor human decisions outside the scope are needed. We have not come to the point when AI can learn all you do, literally, up to analyzing your brain waves and communication with clients and vendors.

Conclusion

So, AI may be used for some standard and repetitive actions confined in certain environment, like TBMS. If it’s a standard translation project with standard files that is assigned to standard translators and editors, and the ways of delivery are standard. When nothing happens outside TBMS. No emails exchange, no rescheduling, no price discussions, no “what the heck is this?” questions “sorry, translator is not available” problems. Then yes, AI can learn from simple projects. In all other cases, you need a human to make decisions.