- Hello World Conference: 12 lessons from a poignant event
- Native is the new standard
- NMT is not suitable for all types of text
- NMT works well needs lots of data
- Errors in NMT are harder to detect than in SMT
- NMT is the best choice for corruptible content
- Anticipate differences between language localization needs
- Everyone at the Hello World conference agreed that language localization has gone far beyond the translation industry
- Centralize language localization needs within the company
- Language localization managers are heroes
- Launch a product and effectively solve the problems that come with it
- Translation quality avoids risk
- AI and Automatic Translation are shaping the future of the language industry
Hello World Conference: 12 lessons from a poignant event
The Hello World conference that PhraseApp organized in Hamburg, Germany, has just ended. After 7 hours of thought-provoking presentations, we felt we had learned a lot.
Representatives from Xing, Deliveroo, AWS, Jimdo, Auto1, Beluga, Lengoo, Wieners&Wieners, University of Hamburg, University Dublin, Otovo, Glossa, and PhraseApp were present and provided invaluable insights into the copy process. what localization means to them and how they do it.
Discussions include use cases, success stories, business advice, best practices, in-depth research, future challenges and more. Read on to find out the most valuable lessons we've taken home!
Native is the new standard
Davide Gallo from AWS shared with us about localization at Amazon. He referenced among other things the CSA's Can't Read Won't Buy study. He also told us that 51% of people prefer to express content in their own language even if they speak English!
So how do you effectively localize your content now when your brochure includes billions of products with thousands of them being added every minute?
Amazon is one of the best examples of why Neural Automatic Translation (NMT) is sometimes the most viable way to localize content.
Davide highlighted the significant advantages of NMT in terms of speed and cost, and showed us that the output of Amazon Translate sounded much more natural than the output of other open automated translation solutions. We didn't know their software was actually in the top 5% of the Auto Translation market!
NMT is not suitable for all types of text
In terms of quality and accuracy, Davide explained that while not perfect, their output can be up to 85% better than human translation. At first hearing, you will doubt this ratio, but it is true.
Davide has also expressed that companies should only rely on Auto Translation for content that doesn't need 100% accuracy. Therefore, it is best to leave marketing documents and sensitive information in the hands of a translator.
Ultimately, we learned that NMT is useful for a wide range of activities beyond localizing product descriptions on the Amazon platform. That proves useful for other tasks like media analysis, e-discovery, content moderation, understanding customer issues at contact centers, converting documents by voice, etc
NMT works well needs lots of data
The Hello World conference also includes research insights from undergraduate and PhD professors. Postdoctoral researcher Meriem Beloucif did his research on Automatic Translation from a low-resource language perspective that took us by surprise. She suggested that the key to making translation useful lies in semantics, which is keeping meaning when rephrasing the message in the target language.
Did you know that neural automatic translation and statistical translation systems can be trained? So you will need to provide them with a lot of data for the specific language combination and area in question? Jonathan Wuermeling from Lengoo has also suggested that the minimum data input should be between 200k and 400k words.
Imagine the challenge of training the system for low-resource languages like Hausa, Uzbek, Tigrinya, Oromo, etc. How would you improve and evaluate the output of automatic translations? when you don't have enough data?
Meriem Beloucif proposed the first model that successfully inserted semantics instead of the BLEU method when training automatic translation systems and she showed that if the tendency to insert semantics into the target language (English), then Translation quality is significantly improved.
After all, in the words of Beloucif, what we all want to get from a translation is who did what to whom, to whom, when, where, why and how?
Errors in NMT are harder to detect than in SMT
Professor Joss Moorkens from University College Dublin brought a very comprehensive study of NMT to the Hello World Conference. We analyzed the different ways to use NTM and compared the outputs of Statistical Autotranslation versus Neural Automatic Translation.
We also looked at how much time editors spent editing each output type. Furthermore, we made a strange finding. It is the errors in NTM that are more difficult to detect than in SMT because the former output shows higher accuracy and fluency and fewer errors in word order and morphology.
NMT is the best choice for corruptible content
Another valuable lesson from the Hello World Conference is that the “shelf life” of a given text can determine the choice between human translators versus NMT. For example, think of a review on TripAdvisor and about a website's landing page title.
So which one will be more "broken"? Which is high quality that is more important? NMT is a viable solution for temporary content like a review. Because precision is not as important for that type of text as keeping the gist of the source text message.
However, using the same approach for content marketing can be a very bad idea! So using NMT for your promotional content is not a smart decision.
Anticipate differences between language localization needs
Tilman Büttner from Xing talks about the mistakes product managers should avoid. We really got back to when Xing was OpenBC! He came to an important conclusion after analyzing the bad decisions the company made at the time. We love this decision: “You need to think beyond your control”, yet many brands fail there.
The most important lesson from Tilman, however, is that developing a product with the potential to reach international markets requires predictability. Localization needs to be integrated into the product or feature development timeline from the outset. This will help you prevent half-translated products and undervalue your customer's need for user experience.
+ Note: Translation of Bid Documents
Everyone at the Hello World conference agreed that language localization has gone far beyond the translation industry
As many speakers Tilman Büttner, Anne-Sophie Delafosse (Deliveroo), Antonella Zagaria (Auto1) and Eike-Marie Eiting (Jimdo) have pointed out, localization takes more attention than text. Aspects like colors, navigation flow, placement of CTA buttons among many others need to be taken into account when localizing. Especially when choosing product names! We laughed and laughed when Antonella showed us the following example:
In many Spanish-speaking countries, “pajero” is translated as incompetent! Imagine how much money Mitsubishi wasted on product rebranding and they could have avoided it with some initial research.
This also reminds us of the fact that Honda did not change the name of the Fitta in 2001 when releasing the car in Sweden. In Swedish, "fitta" is a vulgar way of referring to a woman's genitals. Therefore, localization from the very beginning of the product development timeline is a necessity.
Centralize language localization needs within the company
Eike-Marie Eiting, Anne-Sophie Delafosse and Antonella Zagaria gave us a lot to think about. The first two are localization managers at Jimdo and Deliveroo, and Antonella is a Product Manager at Auto1.
All three of them show us that language localization can become a challenge. Businesses often neglect to localize the language or handle it in a decentralized way, which has caused chaos. Who has heard the typical saying “person X in department X is bilingual, can they translate this?”
Centralizing all efforts for language localization is key to attracting foreign markets and then generating more revenue. Consider Auto1's operations in 30 European countries and 25 languages: that is not centralized would be a disaster! A common challenge for Eike-Marie and Anne-Sophie is translation requests coming in from all over.
They had a lot to do when they started in their current roles! However, their role did not exist a year ago. In addition, all three speakers agreed that centralization produces reliable data. You can analyze how much the company spends on translation and what the ROI is.
Language localization managers are heroes
The language localization manager's impressive work includes creating and maintaining style guides, terminology, translation memory, among other resources. They also act as intermediaries between the groups of people who need localization and the linguists who provide it.
When someone needs a translation, they fill out a request and send it to the localization manager. At that point, this person would assign the task to a linguist or a group of linguists, and then hand it over to whoever submits the request.
The benefits of focusing on language localization are that translation makes a lot of progress in each department, thus freeing up their time. This way, they can focus on their actual work and what they do best. In addition, the speed and quality of translation is greatly improved and affects the performance of the product in the target market.
Launch a product and effectively solve the problems that come with it
Rikard Eide, a Software Engineer from Norwegian solar panel startup Otovo, demonstrated with his experience report how efficient the expansion for the company was.
His presentation took us through the brand's journey and their expansion from Norway to Sweden and then France. However, Rikard has found that investing in language localization before entering the Spanish market leads to higher risk avoidance and conversions.
Translation quality avoids risk
For us, this is not a new concept. However, David Benotmane from the Glossa Group made this extremely clear in his presentation. After Meriem Beloucif's definition of semantic quality, David took a qualitative approach to the potential negative consequences of a bad translation.
To reduce the risks associated with liability and product safety, and for regulatory compliance purposes, companies need to translate their technical documents. Especially in cases where a faulty translation could result in injury or death.
He goes on to describe their Q&A model and system, where the core risk management processes according to ISO 31000 are applied to processes in translation projects. A risk matrix enables risk analysis of documents to be translated in order to develop such processes for translation projects. Implementing comprehensive risk management for translations leads to better regulatory compliance and translation quality.
AI and Automatic Translation are shaping the future of the language industry
This was the subject of a discussion at the end of the conference. Jan Hinrichs from Beluga, Ann Huels from Wieners&Wieners and Jonathan Wuermeling from Lengoo discussed their own experiences with AI and NMT both positive and negative.
It was a great way to end the conference as we could all realize that AI and Automatic Translation are still here and reshaping the translation industry while still in need of improvement. In our own opinion, contrary to many linguists who see a threat in these technologies, we see an opportunity.
That is not going to be a few jobs for linguists, but instead the nature of our work is changing and we need to adapt to survive.
We'd love to hear your thoughts on Neural Automatic Translation! So feel free to leave your comments or get in touch with us directly through social media.
Bai viet share translation experience Most read: