Jane Nemcova

Aimed at providing guidance for human evaluation of the translation output, ISO 5060 can be applied to workflows involving both human and machine translation with or without subsequent post-editing. The standard combines hands-on advice with the strategic considerations and general principles for translation quality evaluation. It is built on the analytic approach that involves a “bilingual examination of target language content against source language content while classifying any errors with respect to translation evaluation specifications and for the purpose of reaching a quality rating.”

2

Have you had a chance to participate in our conversation on the Localization Fireside Chat with Rodrigo Corradi? if not and you don't have the time to watch, here is a blog post that captures in high-level points of our conversation

4

Everything has been said about Globalization and Localization Association (GALA) annual event #gala2024, except of course: #nobodycaresaboutplanktons 😎 Now I'm back home with a ton of things I picked up at the conference: - sessions that will be forever remembered (6-legged horses to the rescue Levan), - interesting, challenging, tough-love and no-AI-BS style talks with people smarter and (probably) better than me, - some more tough-love "let-me-tell-you-how-it-is" talks too, because someone has to "tell-it-like-it-is" and it looks like it was my turn, - making friends new amazing people whom I've never met before, - seeing new amazing scenery and parts of Spain I've never been to before (but Barcelona is still my fav, sorry) - mind-boggling reception of my session (I am your humble servant, and forever grateful to all of you who were there and also gave me a feedback, you made my day (*)), and finally - memories of Star Wars, Los Pollos Hermanos, Motörhead, Biking, Sailing, Grilling, Hungarian Spices, The Wall of Tequila, - paella spice, - chorizo, and fuet of course, - some Spanish chocolate, and - Alhambra Tradicional beer 🍺 OK, now I'm thirsty! It never ceases to amaze me how diverse the Localization industry is, yet we all embrace this uniqueness, friendliness, respect, "multi-culti"-thingie, and tequila (ok, tea is also acceptable) 😂 Jan, thank you for the "now I need to buy a boat" thoughts that are hunting me. And thanks for all the client talks, to "possible new client" talks, to supplier talks, to friend talks, to those who are neither of this, and even to people who didn't quite like talking with me, remember: AI or Robots or what not, we are in the people industry, and sometimes Tapas and Seaside is more important than Dinero! #planktontheory or #locplankton (I can't decide yet) Now it's time to git my bum back to work, which probably means: get ready to receive an e-mail from me in the next couple of days. I'll try not to be too boring. Finally, a promise is a promise, so here's a picture for our "duck-face revolution" Julia Cassidy, sorry, I couldn't make a better duck, and I really-really suck at taking selfies, but I promise to practice, learn, and get better at it! Hugs to all y' all: Nazaret, Rick, Kevin, Jennifer, Mike, Mohamad, Patry, Gökmen, Agnė, Marija, Luz M. Sanchis 小露, Shikha, Jose, Anne Marie, Stephen, Carl, Simon, Albena, Mário, Brahim, Andre, Tayo, Márta, Istvan, Andrew, Asta, Kursat, Arben, Florin, Igor, Daniela, Giada, Ricard, Gergana,... - and sorry if I missed someone, I'm bad with names. And faces. Special thanks to Allison, Isabella and the whole Gala board and staff, both for organizing the event, and for having faith in my speaking abilities :) ) (* nota bene: if you liked my presentation, just imagine the level of how "out of this world great" it would have been if I talked about something I actually have a clue about 🤣 --> my weird way of pre selling you for next year, of course)

118

28 Comments

This is unfortunate, and I disagree. Even "uncomplicated bread-and-butter stuff" like instruction manuals contain specific, context-dependant terminology that can pose challenges for generative AI. Any company depending solely on generative AI for these purposes risks translations containing huge blunders. https://lnkd.in/ex6d4YYS.

13

The localization landscape is witnessing a shift with the advent of Large Language Models (LLMs), which are redefining the benchmarks for quality and efficiency. 🚀 A study by Intento, Inc. reveals that out of 52 systems evaluated, LLMs constitute 55% of the top-tier models, a significant leap from just 17% the previous year. This underscores the rapid development and integration of these sophisticated systems in the translation process. 📈 What's interesting is that bigger isn't always better when it comes to LLMs. Their performance isn't solely dependent on size, which emphasizes the need for a nuanced approach to evaluating translation quality. The evolution of LLMs brings a cost-effective edge, being exponentially less expensive than MT systems. Yet, this comes with a trade-off in speed, as LLMs can be slower. At Global Lingo, we're committed to finding the sweet spot between cost, speed, and quality to ensure our clients receive the best possible localized content. 💡💰 This dynamic interplay between technological innovation and translation efficacy is what keeps us at the forefront of the industry. #GLConnects #MachineTranslation #LargeLanguageModels #AI #TranslationTechnology #ArtificialIntelligence #Localization #Translation

34

1 Comment

The verticals that shrank are those also directly affected by AI (except Engineering, they're low on labor). The most resilient verticals were high-risk ones (aerospace, life sciences, finance, government). There are a lot of other good stats and insights in here. #languageindustry #translationindustry #interpretingindustry #translation #interpreting

4

1 Comment

A recent Slator report highlights that one-third of LSPs are already using LLMs, with close to 20% actively testing them. This rapid adoption is a clear sign that the industry recognizes the immense potential of LLMs to enhance machine translation, quality evaluation, and more. Just 3% of LSPs plan to stay on the sidelines. At Milengo, we've always been at the forefront of adopting new technologies. Personally, I'm fascinated by the possibilities LLMs hold for improving the efficiency and accuracy of our services. However, it's important to remember that human expertise will always be irreplaceable. The future lies in the powerful combination of both😎 #AI #localization #machinelearning #languagetranslation #Milengo https://bit.ly/3ViQs54

7

Who hasn't heard of CSA's Localization (now Globalization) Maturity Model? Few concepts have had such an impact on the language industry. So ingrained I often mistakenly type LMM instead of LLM. A Freudian slip, perhaps? But multilingual or language #AI is adding a totally new dimension — AI maturity — to what would be a more-or-less linear progression on the localization maturity scale. It may be early days, but it's clear that AI is not just one piece of language tech. It touches every element of the traditional maturity model (strategy, process, organization, etc.). Over time, it may also affect the role, position, remit, and actual setup of in-house #localization teams. AI maturity is only loosely related to localization maturity. Some enterprises may be at the optimized or transparent stage, but with AI, they are very conservative (sometimes for a good reason). Others may laugh off being labelled as Level 0 or Level 1 on the localization scale, but are pursuing an AI-first strategy, which makes them fairly AI-mature instead. A lot of research, interviews, and thinking went into designing this AI maturity model (AIMM). There is no right or wrong way. But the goal here is to help measure localization teams' awareness of and readiness to integrate AI capabilities successfully. And show the path (still) ahead. Check it out here! https://lnkd.in/evQK94Jw #AImaturity

58

A new study settles a long-standing debate: BOTH accuracy AND fluency matter for top-notch machine translation! For years, experts have questioned the relationship between these two dimensions. Some believed they were a trade-off, meaning improving one sacrificed the other. Accuracy ensures all information from the source text is preserved in the translated version, while fuency guarantees the translated text reads naturally for the target audience. Focusing on just one hinders effective machine translation. #machinetranslation #translation #communication

9

Highlights from Language Talks, S01E01: Insights from top localization managers https://hubs.la/Q02DJLYP0

3

1 Comment

🧐Some food for thought for localization professionals: a new study shows LLMs can now predict the most suitable translation technique for a specific task. This means AI can help us choose between approaches like back-translation, paraphrasing, or even suggest leaving certain elements in the source language🤔 The study found high accuracy for both from-scratch translation and post-editing scenarios, potentially boosting both efficiency and quality in the translation workflow. #MachineTranslation #AI #LanguageLearning #TranslationTech #TheFutureofTranslation

1

1 Comment

While machine translation has evolved, and multiple language speakers are helpful in providing a rough meaning of some types of content- there are limits to what both options provide. We present to you – and bust!- three of the most common myths people have relating to document translations. https://lnkd.in/ghhWYqhz

1

A common question from prospective and current #NMT customers, "When is it suitable to use machine translation #MT?" This blog by Heather Rossi explores why the more appropriate question to ask regarding #MT use Is "What are our business/intelligence outcomes that we want to achieve with MT?"

7