Machine Translation Post Edition (MTPE): your feedback as linguists/LSPs Thread poster: BabelOn-line
|
|---|
BabelOn-line
United Kingdom
Local time: 23:56
English to French
+ ...
MTPE is likely to be an unpleasant topic for many linguists.
[Hence, a quick preamble] Yes, we are a LSP (a smallish one at that) but I am first and foremost a translator. I would like to invite you to check our BlueBoard, please do not confuse us with a "pay as little as you can get away with" kind of agency – or people who do not care about quality. In a word: we are not the enemy. [end of preamble].... See more MTPE is likely to be an unpleasant topic for many linguists.
[Hence, a quick preamble] Yes, we are a LSP (a smallish one at that) but I am first and foremost a translator. I would like to invite you to check our BlueBoard, please do not confuse us with a "pay as little as you can get away with" kind of agency – or people who do not care about quality. In a word: we are not the enemy.[end of preamble].
The context: one of our French clients produces quite a volume of internal emails, reminders, etc that they translate into English using MT (for their staff and users, 20% of which are not French-speaking) – and they simply slap the unedited result after the French version. They also produce more client-facing/official documents that they translate internally (poorly).
I have no doubt that we could do a better job with the important, official translations using our network of qualified, motivated linguists and client is on board with this. So far, so good.
We are however wondering what to suggest regarding the large volume of small, machine translated (MT) emails, reminders etc, that they churn out.
Even at our lowest possible price point, the client would simply not have the budget to afford human translation. Chances are, at normal market price, they'd stick to unedited MT output. At this point, I'd like to pinpoint the fact we are not robbing any linguist's work here, as the client would NOT pay for a full translation anyway.
At present, we consider offering this simple approach:
1/ Official, client-facing, large English readership documents: 100% human translation (with CAT and own internal glossaries for consistency)
2/ Internal, small English readership, less important communications: MT translation plus very basic "sanity check" (TAUS calls it "light editing"). Better than raw output, avoiding major mistranslations.
3/ Internal, larger English readership, semi-official stuff: MT plus in-depth editing (TAUS: "full post -editing").
Option 2/ sounds simple: it does not need to be pretty, it just needs to be "good enough" as TAUS guidelines put it. Avoiding disasters created by MT.
Option 3/ sounds good in principle, but it is more tricky: depending on many factors (type of topic, quality of the source, etc), we may end up having to rewrite the whole MT-produced text for it to be better than "just good enough". Or we could be in luck and the MT output only requires so much editing to bring it to something closer to human translation.
Questions for LSPs: how do you go about it? Does the above look like a solution? how do you handle option 3 and charge it with regards to your usual per word rate?
Questions for Linguists: do you have any experience of the above? How do you feel about MTPE? Client is likely to ask for a "per word" rate, but we understand that the time spent on a 100 words can vary enormously. Any solution that would be fair to both parties?
Thanks in advance for your input.
Kind regards.
Jean-Louis ▲ Collapse
| | | | John Fossey 
Canada
Local time: 18:56
Member (2008)
French to English
+ ...
| One approach | Apr 21, 2020 |
In my opinion as a translator, your proposed approach seems reasonable on 2 conditions:
1. Using a good quality neural-based MT engine. Some MT engines consistently produce garbage, requiring a lot of work to get a sensible output, while others (generally more expensive) are more fit for the purpose.
2. Structuring the pricing so the translator/reviser does not suffer reduced income. It might take some real-life testing with the proposed source material and MT engine to... See more In my opinion as a translator, your proposed approach seems reasonable on 2 conditions:
1. Using a good quality neural-based MT engine. Some MT engines consistently produce garbage, requiring a lot of work to get a sensible output, while others (generally more expensive) are more fit for the purpose.
2. Structuring the pricing so the translator/reviser does not suffer reduced income. It might take some real-life testing with the proposed source material and MT engine to establish reasonable price points. ▲ Collapse
| | | | BabelOn-line
United Kingdom
Local time: 23:56
English to French
+ ...
TOPIC STARTER | Thanks for this input | Apr 21, 2020 |
Thanks, John.
"Using a good quality neural-based MT engine."
We were thinking of DeepL. Not tried many others, truth be said, but so far this is clearly the best we tried.
"Structuring the pricing so the translator/reviser does not suffer reduced income. It might take some real-life testing."
Yes, that's the rub.
Ideally, we would have some margin to tweak our prices within an agreed bracket depending on the quality of the MT output, but client will demand a one size fits all, per word rate.
So we end up in an uncomfortable situation:
• either we quote a high per word MTPE rate and cover our backside - but risk scaring the client,
• or we offer the client a rate based on a guesstimate and we tell our linguist: this text has a budget of (words x rate), this equates to xx hours of your normal hourly rate, so please spend exactly xx hours on this job. If I was asked to do this, i'd find it difficult. I'd have to do a rough calculation and spend no more than that many minutes per line/word count.
I think we will need to take a sample of source text and try our hand at it. May be a first 15 days period with a low introductory offer followed by a revised permanent rate.
The fact we are using the same linguists long term will probably help a bit: some text will take longer, some will be shorter, they will average out on the long term.
| | | | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Machine Translation Post Edition (MTPE): your feedback as linguists/LSPs | Wordfast Pro | Translation Memory Software for Any Platform
Exclusive discount for ProZ.com users!
Save over 13% when purchasing Wordfast Pro through ProZ.com. Wordfast is the world's #1 provider of platform-independent Translation Memory software. Consistently ranked the most user-friendly and highest value
Buy now! » |
| | Anycount & Translation Office 3000 | Translation Office 3000
Translation Office 3000 is an advanced accounting tool for freelance translators and small agencies. TO3000 easily and seamlessly integrates with the business life of professional freelance translators.
More info » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
