Introduction

Housekeeping

About AIQE from ModelFront's CEO

TAUS presentation

memoQ demo

Q1: Can AIQE and memoQ AGT be used together? Would we be able to get an AIQE score on translations generated by AGT?

Q2: When this report feature will be avaliable in memoQ?

Q3: Would you limit AIQE to certain content types? We deal with high risk machinery, I am unsure where we stand in terms of indemnity on segments that a human has not checked.

Q4: What are the criteria for evaluating an MT translated segment as “perfect”? Only linguistic, or also other criteria? E.g. terminological consistency, standardized language, regulatory language? Especially, when the engine is not trained.

Q5: Can the setting also be higher than 90%?

Q6: What model is used for AIQE, something proprietary? Or third-party APIs can be hooked up?

Q7: It's interesting to know that AIQE can be used with human translations, too. Would the score be shown live after each segment is confirmed by the translator, or would we have to run a quality estimate check afterwards?

Q8: Is this like the MTQE in Phrase?

Q9: Did I understand this right: to determine that the AIQE is on good quality, you make test evaluations, then you define the threshold of the quality (eg you used is 90%)?

Q10: What about privacy? As LSPs, we work with our clients’ content, which should not be leaked due to NDA and so on.

Q11: When working with a text, are ways to use AI to update and align the text with my editing choices mid-text? In other words, can AI learn my editing choices so I don't need to run through the entire text repeating the same actions?

Q12: Do you think AI estimation of quality will eventually have an impact on the all-over concept of quality in translation? For better or for worse?

Q13: What is your pricing strategy for this? Does it come with a subscription or is it priced per character evaluated?

Q14: Considering the speed at which AI is currently developing, would you still feel confident recommending translation as a field of study for someone starting their academic career?

Q15: If I first pre-translate using my TMs and then I use MT on the remaining segments, will the model also detect terminological inconsistencies across these 2 types of outputs (TMs VS MT)?

Q16: There is a need to evaluate someone else's MT to determine whether it needs to be edited or not. Will AIQE support this?