Fiverr revisited: can German translators do any better?

Following on from the experiment I conducted with Natalie Soper a few weeks ago, in which we tasked four translators from Fiverr.com with the translation of a French travel news piece, I decided to readjust the parameters and try again. “What if there are actually a few good German translators trying to get their career off the ground on that platform?” I thought to myself.

 

So I took this text and went about my search. This time, the text is shorter – our last text was around 500 words and not all of the translators we asked were happy to be paid $5 for so much work. Rather than trawling through the many profiles listed on the website, this time I decided to send out a request. It read as follows: “I need a text translating. It’s German & I need it in English. 234 words total.” No text attached. I could have given more information but I wanted to see whether any of the applicants asked questions, for example the subject of the text or the target audience.

 

They didn’t.

 

Within 24 hours, I had 22 requests. All but a handful were completely out of the question. Headlines such as “Translate your documents from english to french[sic]” (um, how about German to English?!), “translate any language to any other required language” (omg), “translate to any known language from english and hindi” (wrong direction for starters) and “html ccs bugs, and front end designing for you” (err, what?!) grabbed my attention… for all the wrong reasons.

 

Once I had deleted these inappropriate requests, I narrowed it down to the following four:

Translator A

Had more than 600 reviews on Fiverr, and a five-star rating. I was definitely ordering from him, even though he said he was a native German – technically translators should translate into their native language. (This can be a grey area, especially with German.) He even claimed to be specialised in marketing, SEO and business. Excellent!

Translator B

Had 21 reviews and a 4.5-star average rating. Not bad! Native English speaker, even better. Claimed to have lived in Germany and had experience of DE-EN translation. Excellent! I’d be a fool not to book her. Although, when I clicked through to her profile, there was very little mention of translation. Oh…

Translator C

Offered “Top QUALITY” German to English translation. Once I clicked on her profile, I saw 61 five-star ratings. She also claimed to have 12 years’ experience as a translator in English, French and German. Seemed promising. This translator is also a “holder of a masters in modern letters”. I wasn’t quite sure what this meant, but I was willing to overlook it.

Translator D

Promised a high level of accuracy and said he was a native German and English speaker. He has 50 five-star reviews. Although, once again, when I clicked on the profile there was no mention of translation. There wasn’t even a mention of him speaking anything other than English, even though he claimed to speak German and English. Maybe I made a mistake with this one…

 

The ordering process

I decided to order these translations bright and breezy at 9:45am on a Sunday morning. The first translation was delivered precisely two hours later by Translator D. I could certainly translate 234 words in two hours. I wouldn’t be delivering my best quality work, though.

Next to deliver was Translator A. At 5:38 am on a Monday morning. Translator C followed two hours later. Translator B messaged me on Monday morning: “Hello, Thanks for your order, I’ve been really busy but I’ll get on it right away and I’ll send it immediately I’m done. [Last name], [First name]”. She claimed to be a native English speaker, but there are several things in that message that set alarm bells off: immediately I’m done? And signing with your last name first? Hmm… She delivered the next day, at 2am. So I got all translations back from my original translators, that’s already progress compared to the last time!

 

Analysis

General comments

It is once again clear to see that all four translators heavily depended on Google Translate, with all four texts looking very similar to the machine translation version with a few tweaks here and there. I could go through the entire text with a fine-tooth comb in this analysis, but I’ve chosen to focus on four issues with the translations.

 

Michael Kerkloh and Reuters

In the first paragraph, Google Translate inadequately translated “[…] sagte Flughafen-Chef Michael Kerkloh der Nachrichtenagentur “Reuters”” as “[…] said airport chief Michael Kerkloh of the news agency “Reuters””. As a result, three of the four translators mistranslated this simple phrase, and Translator A, B and C decided to keep Reuters in quotation marks. The only reason why Translator D didn’t was because he put the quotation marks in the wrong place. Translator A went with: “[…] as told the news agency “Reuters” was told by the airport’s manager Michael Kerkloh” (yes, with two mentions of ‘told’ in one sentence). Translator C wrote: “[…] said airport Chief Michael Kerkloh of the news agency “Reuters”” (thus making it sound like Kerkloh was talking about Reuters, not to them) and Translator D opted for “[…] “said Michael Kökloh, CEO of the airport,” Reuters.” (thus quoting text that did not need to be quoted, and spelling the man’s name wrong).

 

My solution was to flip the sentence order and have this phrase at the start rather than the end. So my sentence begins with: “Munich Airport CEO Michael Kerkloh told news agency Reuters that […]”.

 

Breaking tents

The following sentence was pretty tough to translate: “Auch den überraschenden Abzug der Air France-KLM-Günstigairline Transavia, die ihre Zelte in München nach nur einem Jahr abbricht, erklärt der Flughafen mit fehlenden Start- und Landefenstern.”

Translators B, C and D  all decided to translate the “ihre Zelte in Münschen […] abbricht” literally with “breaks its/their tents”, which is not an idiom we use in English, and every native speaker should know this. I was also shocked by the sloppy translations of “Start- und Landefenstern”, especially as the word “slots” had been used in the German text. Translator B opted for “starting and landing windows”, C for “start and country Windows” and D for “, And land windows.” (yes, omission of a translation for “Start-“ and a random capitalisation of “and”.)

 

My solution was to split the sentence into two: “After just one year, Air France/KLM’s low-cost carrier Transavia unexpectedly withdrew from Munich. The airport blames this on a lack of take-off and landing slots.” I did this to make the text flow better, but my phrasing meant that I could omit the translation of “breaking tents” as it became superfluous. I also translated “Fenster” as “slots” to keep it consistent with the rest of the text.

 

Airport owner(s) and referendum (results)

The following section also proved to be tricky, and several issues arose in the translations. I want to focus on the text in bold:

“Noch im Frühjahr wollen Staatsregierung, Bund und Landeshauptstadt als Flughafen-Eigentümer entscheiden, ob sie den Bau einer dritten Startbahn vorantreiben und dazu einen neuen Bürgerentscheid abhalten. Das Projekt liegt seit dem ablehnenden Münchner Bürgerentscheid von 2012 auf Eis.”

 

Translator A: The state government, federal government and state capital as airport owner will decide whether to campaign and conduct a new referendum for a third runway in Spring. The project had already been halted with the negative referendum result of 2012.

Translator B: In the spring, the state government, federal government and the state capital as an airport owner will meet to decide whether to push forward the construction of a third runway and to hold a new referendum. The project has been on ice since the negative Munich referendum of 2012.

Translator C: In the spring, State Government, Federal Government and the city as the airport owner wants to decide whether they should promote the construction of a third runway and to hold a new referendum. The project has been on the ice since the disapproving Munich referendum of 2012.

Translator D: In the spring, the state government, the federal government and the state capital will decide as owners of the airport whether they are pushing for the construction of a third runway and make a decision on it. The project has been on the ice since the disapproving Munich citizenship of 2012.

 

Listing the three entities and adding “as the/an airport owner” is not technically incorrect, but it is very clunky in English due to verb used in the sentence. Do you go with plural as there are three owners or singular referring back to “airport owner”?

 

My solution: “In their capacity as the airport’s owners, the state and federal government and the city of Munich plan to decide on whether to proceed with the construction of a third runway and hold another referendum on this matter in spring. The project has been on ice since the residents of Munich voted against it in a referendum in 2012.” Although the first sentence is still long, I’ve avoided repeating the words “state” and “government” (see Translator A, B and D) and made everything plural so there is no confusion with the verb. I was also careful with the wording of the next sentence to avoid the phrasing used by the Fiverr translators.

 

The subjunctive

In German, authors can use the subjunctive mood to express doubt or to distance themselves from the opinions stated in the text. This is frequently used in journalism to indirectly quote speech. There is an example of this in the text:

“Kerkloh wirbt daher für den Bau einer dritten Start- und Landebahn. Ohne deren Kapazität werde München Wachstum verlieren.“

All of the Fiverr translators failed to spot it, translating this sentence as fact rather than Kerkloh’s opinion.

A: Kerkloh therefore campaigns for the construction of a third runway. Without it Munich will lose out on growth.

B: Kerkloh is therefore promoting the construction of a third runway. Without their full capacity, Munich will not continue to grow.

C: Kerkloh therefore promotes the construction of a third runway. Without their capacity, Munich will lose growth.

D: Kerkloh is therefore promoting the construction of a third runway. Without their capacity, Munich will lose growth.

As you can see, all of these translators have made it look like Munich will definitely lose out on growth if a third runway is not constructed. What the text actually says, though, is that Kerkloh thinks Munich will not be able to see growth without the construction of a third runway. Hence my translation: “He [Kerkloh] is therefore canvassing for the construction of a third runway as he believes Munich would not see growth without this capacity.”

 

This is something that Google Translate always struggles with because it needs to be translated differently depending on the context. For example, you can use allegedly/apparently to create distance or you can add in the speaker’s name with believes/reckons/thinks (or another synonym) to make it clear that you are referring to something they have said rather than a fact. The latter is particularly difficult for a machine to replicate because it requires understanding of the text, not just an ability to string words together based on translation patterns detected in texts already translated by humans and fed into the machine’s database.

Conclusion

All four Fiverr translators used Google Translate, and none of the texts are fit for purpose. Every single one of them had me scratching my head at least once, and if I couldn’t speak German, I would not be able to figure out what they were trying to say. The quality of these texts is sub-par as these texts do not accurately convey all of the information written in the German.

 

This is the second experiment I have conducted on Fiverr, the first one being a French translation in collaboration with Natalie Soper. The results of both experiments are the same: the translators use Google Translate and tweak the output slightly in an attempt to iron out Google’s mistakes. But it’s still not good enough. But as somebody pointed out on the Fiverr forums, what do I expect for $5?!

 

Once again, here are the translations and Google Translate’s output for comparison:

Translator A Translator B Translator C Translator D

Google Translate

Advertisements

Fiverr: Do you get what you pay for? (Part 2)

A friend and translation colleague, Natalie Soper, recently messaged me to tell me about a friend of hers who was considering using Fiverr to translate professional texts. The text needed to be translated from English into several languages and Natalie’s friend didn’t know where to start. We were both horrified that anybody would think Fiverr is a viable option for finding translators. Personally, if I can’t recommend a trusted colleague, then I refer them to the directory of a professional translation association, depending on the languages required.

 

So we decided to put Fiverr to the test. We selected a text from a newspaper website relating to one of our common specialisms: tourism. This is a field that is sometimes considered “easy” in comparison to other fields, say legal or medical. Yet it has its unique challenges, and this text was tricky, ready to trip up our translators in every paragraph. And it did. The least we’d expect from a professional translation is a text that reads fluently in English, that adapts to the new audience (with clear research and extra explanations where needed), has taken numerical formats into account and has creative solutions to the uniquely Francophone complexities… and we got none of that.

 

You can read more about how we selected the translators and the intricacies of Fiverr on Natalie’s blog.

 

Natalie and I also translated the text as if we were delivering for a client. We met to discuss our versions plus the versions delivered to us. I am going to outline a few of the parts we found difficult, and compare the translations.

Analysis

General

startup-photosFirst of all, it was plain to see that every translator used a machine translation tool and made slight edits to the output. We suspect they used Google Translate. The texts delivered by Translator C and Translator E were so similar in parts that I thought they may have been delivered by the same person. This is despite the fact that, as Natalie mentioned in her blog post, each translator claimed to translate their texts “manually”. (One translator also claimed not to use “spinning software”, whatever that means!) When we compared our own texts, they were vastly different but still accurate, which is what we expected from our Fiverr translators.

 

Two of our translators, A and C, used US dates and spelling, whereas B and E adapted the date to UK English but used US spelling conventions. As Natalie mentioned in her blog, none of them asked if we’d prefer US or British English, but we hoped that they would at least stick one style consistently.

 

For some reason, Google Translate added in a space at the end of the third paragraph:

The director general of Atout France, the French tourism promotion agency, confirmed that France “really started a recovery from September / October”, and that the end of 2016 had been better than expected “.

Miraculously, this erroneous space turned up in every single translation we purchased.

 

La France séduit toujours autant.

For me, the very first sentence was one of the toughest in the text. Each translator (including Natalie and me) dealt with it differently.

Translator A: France still seduces as much.

Translator B: France is still very attractive, […]

Translator C: France is still quite seductive.

Translator E: France still remains attractive as before.

These are all pretty literal translations, and some are less effective than others. Translator A’s solution is an outright word-for-word translation, and it doesn’t work. Nor does Translator C’s version as the French author did not mean seductive here. The verb ‘seduce’ has sexual undertones in English, which is certainly not what we were going for here. Natalie and I thought that ‘attractive’ was a better option, although Translator E’s version seems to be missing an ‘as’ (“remains *as* attractive as before”. Yet this still makes the reader wonder: “Before what?”) I used ‘attractive’ in my version: “France remains an attractive destination.” Natalie, on the other hand, went for a much more stylish: “France hasn’t lost its charm”.

 

Jean-Marc Ayrault

Another challenge came in the second sentence: “C’est le constat tiré par Jean-Marc Ayrault, qui a annoncé, le 10 février, que pour l’année 2016, le pays conservait sa place de première destination touristique mondiale.”

Being published on a French news website, the French text is mainly aimed at a French audience who is expected to know who Ayrault is. Yet a generalist English-speaking audience probably would not know who he is. Natalie and I both added his title in this sentence, but none of our Fiverr translators did.

Natalie: This was the conclusion made by the French Minister of Foreign Affairs Jean-Marc Aryault, who announced on 10 February that the country has held onto its crown as the world’s top tourist destination in 2016, […]

Me: This conclusion was drawn by French Minister of Foreign Affairs Jean-Marc Ayrault, who, on 10th February, said that the country remained the number one destination in the world in 2016, […]

 

Le directeur général d’Atout France

About halfway through the text, the author writes:

Le directeur général d’Atout France, l’organisme de promotion du tourisme dans l’Hexagone, a confirmé que la France “a vraiment amorcé un redressement à partir de septembre/octobre”, et que la fin 2016 avait été “meilleure que prévu”.

All of our translators came up with a different title for this man. Translator A chose ‘general director’, Translators B and C opted for ‘general manager’ and Translator E went for ‘executive director’. A simple Google search reveals two viable options for his title, CEO or Director General. I plumped for the former, whereas Natalie chose the latter. It’s important to get the person’s title right as he could be mistaken for somebody else, especially as his name was not mentioned in the text.

 

Umih

In the last paragraph, all of the translators stumbled on the following sentence:

2016 a été une année difficile pour nos entreprises, surtout à Paris et sur la Côte d’Azur”, a confirmé Roland Héguy, président de la principale organisation hôtelière, l’Umih.

In fact, Translators A, B and C all stuck to Google Translate’s version:

confirmed Roland Héguy, president of the main hotel organization, the Umih.

And Translator E wasn’t far off this with:

affirmed by Roland Héguy, president of the main hotel organization, the UMIH, simply changing “confirmed” to “affirmed” and capitalising the name.

The problem here is that Umih is a union, not a hotel organisation. A little bit of research would have helped our Fiverr translators here, as both Natalie and I managed to spot this:

Natalie: says Roland Héguy, President of the French hoteliers union, UMIH.

Me: confirmed Roland Héguy, President of UMIH, the country’s main union for the hospitality sector.

Conclusion

As highlighted throughout this post, the Fiverr translators depended on Google Translate, adjusting it slightly where they thought appropriate. None of them researched any part of the text or added glosses for aspects that English-language readers would not understand or be aware of, which a professional translator will do as a matter of course. Although these translations only cost $5 each, we got neither speed, quality nor good customer service – we basically paid for Google Translate with a few tweaks, some of which weren’t even appropriate. Was this a good investment? No. If anything, this exercise demonstrates that it takes much more than language skills to craft a good translation: knowledge of the subject of the text and target audience are also crucial factors, and although Natalie and I are clearly biased, we feel that hiring an expert translator is still the only option for translating professional texts.

 

Edit: here is the French text plus the translations we received.

Text for translation

Translator ATranslator BTranslator CTranslator E