Another year is drawing to a close, and we can hardly believe how fast the time has flown by. Now is a good opportunity to take a look back at all of the important developments that 2024 – the year of AI – has brought us, and give you an insight into what next year has in store for us.

AI has been a hot topic ever since OpenAI stunned the whole world with ChatGPT. Companies are increasingly insisting on using AI wherever this seems possible. From many discussions and exciting customer projects over the course of the year, we have identified key lessons learned and trends in this field.

Five key trends relating to the use of AI in the context of translation

• Expectations for generative AI remain very high.
  However, the purposes for which people want to use it differ greatly, especially in language processes: From the fanciful idea of a wonder machine which produces, translates and optimises texts so they are perfect, through to a clever tool that provides targeted assistance with specific tasks that are usually performed manually at present. The increasing integration of large language models into translation processes makes exactly this possible by providing these with targeted and modular support. This ranges from the bilingual extraction of terminology and the post-editing of machine-translated content, through to assessing the quality of multilingual documents.
  
• If you want to use the terminology efficiently and sustainably, you also need high-quality, well-structured language resources to be able to supply the language models with relevant information.
  This means that years of working with translation memory and terminology management systems now offers double the benefits. If this data is prepared in a structured and sustainable manner, language models can use it to optimise machine-translated content, for instance in the form of retrieval-augmented generation (RAG).
  
• The topic of data protection continues to generate extreme uncertainty despite the adoption of the EU AI Act in May 2024.
  Many companies are looking for ways to use AI in the most secure possible way in order to protect their precious data against misuse.
  
• A lot of businesses are experiencing issues with the scalability of AI solutions, whether this concerns the IT infrastructure, financial resources or further training of staff.
  
• Human in the cockpit. People will increasingly return to the centre of the AI-based translation workflow.
  While translators were previously responsible for the post-editing of predefined machine-translated content, among other tasks, as part of human in the loop concept, the new human in the cockpit principle aims for translators to use modern language technologies – even interactively – in order to exert their own influence on the output and to create efficient design processes.
  The technological transformation is also resulting in changing requirements for current and future language experts. The relevant universities have also recognised these developments and are revising the degrees and courses they offer accordingly. For instance, prompt engineering, language technologies and information management are important focal topics that will feature more often on the curriculum in future.

Are you interested in this subject? Then don’t miss our STAR webinar, which is scheduled for early 2025. There, we will be sharing information about current trends and our latest technological developments.

Can AI help to create high-quality content in any language while adhering to corporate language and specific rules?

Today we’re interviewing David Heider, the owner of a STAR partner sound studio in the Czech Republic, to shed light on this fascinating question – can artificial intelligence be effectively used in the area of video and audio productions?

STAR: David, when did you start offering professional audio productions?

Our recording studio has been providing its services since 1999 and we’ve specialised in the spoken word. We cover two different areas. Firstly, the “corporate world”, with recordings of material for internal purposes, such as e-learning. This also includes localisation of internal company systems and software. This can be either training material or various web-based platforms with voice output or automatic operators on your phone, sat nav, etc.– in short, various applications where we often have to cut the sound word by word or even syllable by syllable and where everything is then put together by a system into sentences and whole messages.

The second area is more artistic in nature and covers advertising and promotional videos, among other content. This area differs from the “corporate world” previously mentioned in that it’s not just about conveying content, but rather about a form that appeals to listeners and attracts them. So we need professionals who can express themselves artistically and use their voice skilfully. To summarise, you might say that our first area of action is to provide information. This is about content where users, to put it more clearly, don’t have much choice, as they generally have to listen. In contrast, artistic productions aim to seduce the “audience” in some way, not only in terms of content but also their form.

STAR: This inevitably leads me on to the next question – can AI be used in your work?

AI is an amazing tool and offers numerous advantages. For example, we don’t need to contact a voice-over artist and make an appointment; the AI is always available.

STAR: Are you already using AI?

Yes. We use AI in some cases for preparing and producing audio material. But there’s also a downside. In most languages, the AI voice seems artificial or boring, especially after listening to it for a long time.

STAR: Can’t AI intonate?

Intonation in itself isn’t usually a problem, but the AI does it in unnatural inflections, which is really inconvenient. Often it doesn’t emphasise the core message, which a person would normally express through a particular emphasis. And when you listen to an AI recording, you get this unnatural inflection on repeat that starts to get annoying after a while, because you can’t shake the feeling that it’s actually just “copy-paste”. In comparison, I find it much better in English than in other languages, where the AI can work with variable intonation and make the voice sound very natural and lively. But in all the other languages, we still have a long way to go before we reach that point. At the moment, the other languages still sound very “plastic”.

STAR: Are there any other disadvantages to AI voices?

There’s a second point that I think is more serious, especially with e-learning. As with any AI, the quality of the output depends on the quality of the input. You also always have to prepare the content correctly for AI voices. Perhaps the AI doesn’t read all the abbreviations correctly, e.g. in the same way as you would read them in a specific corporate culture. Every company has its own corporate jargon and the AI won’t take this into account. This also applies to different product names, place names and foreign words. For example, if French names appear in English text, should it be read in French or English?

STAR: How can this be explained?

Only the employees at a company are really familiar with the corporate language and know why a certain linguistic rule can sometimes be ignored for internal company content or marketing reasons. And the listeners are insiders, i.e. they usually know what the content’s about. Companies also have to be consistent, otherwise it will sound strange to their ears. Sometimes, of course, a term or abbreviation can be misunderstood, either phonetically or for names, but that’s just the way it’s done at the company and we should respect it.

STAR: What other challenges are there?

Abbreviations and other specific features are a major challenge for AI. They usually need a lot of adjustments and corrections, which can result in the final price being similar to that of a traditional voice-over. We need to create pronunciation tips or edit the text so that it’s easy for the AI to read. This is very time-consuming – so AI makes little sense for a one-off project. In addition, we also “proof-listen”, i.e. do a listen-through to check, after the AI.

STAR: Don’t you “proof-listen” for human speakers too?

If there are two of us in addition to the speaker during the recording, we don’t do this any more because we can hear and check everything during the recording. The exceptions are languages that we don’t understand, such as Asian languages. But, in the case of AI, we don’t know beforehand what it knows and what it can read. I’ll give you an example. Let’s take the unit of a “megapascal”. This term has the abbreviation “MPa”, and the AI can read it as “em-pee-ay”, which is complete nonsense to a technical expert. So we’ve got to figure out how to get the AI to read it correctly as “megapascal”.

Sometimes we go through the recording and it seems right to us, but then the customer finds something that doesn’t fit their corporate culture. That’s why, while I think AI is a useful tool in certain informational texts that can make work faster and cheaper, and I’m happy to recommend it, in the hands of an inexperienced user, AI can behave unpredictably, and the end product will cause more disappointment than enthusiasm about the resources saved.

STAR: Is there a financial difference?

Yes, using AI reduces the budget to around half or two-thirds, as the work is mainly done by a machine and no voice professionals are involved in the process.

STAR: What do you do if a recording isn’t suitable for AI?

We are the guarantor of quality, and if we have serious and justified doubts about whether AI will lead to the right result, we’ll inform the customer. But customers also want to have personal experiences of this. I then try to point this out first by saying, “don’t be disappointed, but I don’t think AI is suitable for this particular project.” When I feel that I’ve outlined everything, I leave the decision up to them. But in some cases, customers themselves are unsure and are grateful for our support.

STAR: Thank you, David, for this very interesting discussion about AI in audio recordings.

AI voices aren’t yet perfect, and human voices are still winning the race. They’re able to convey emotions and leave a strong impression. However, AI voices are an inexpensive alternative. Please feel free to contact us for our advice.

David Heider,
owner of a STAR partner sound studio in the Czech Republic

In the fast-paced world of the translation and localisation industry, efficiency is the key to success. One solution that can play an important role in delivering this efficiency is the Common Translation Interface (COTI) standard, particularly in its highly developed form – COTI Level 3. But what exactly does this standard entail and how can it speed up translation processes?

What is the COTI standard?

The Common Translation Interface (COTI) standard was developed specifically for the translation and localisation industry to improve interoperability between different software tools and systems. The COTI standard defines a manufacturer-independent format for exchanging data between translation memory systems (TMS) and editorial systems, such as content management systems (CMS) and other tools used in the industry.

Higher COTI level, more automation

COTI levels build on each other and offer varying degrees of integration and automation:

• Level 1 – core features: Translation data is saved in a defined structure, compressed as a ZIP file with the extension .coti and enhanced with meta information. The data is transferred manually, but the meta information and fixed structure make it easy for the receiving system to interpret the packets.
• Level 2 – extended features: At this level, the transfer of COTI data packets becomes automated. The editorial system generates a package that is automatically recognised and imported by a TMS as soon as it is placed in a shared transfer folder (hotfolder) that is constantly monitored. Meta information enables the receiving system to create an automated order system, for example.
• Level 3 – expert features: The highest level of integration offers fully automated data transfer between the systems. This removes the need to create or monitor packages manually. Instead, translation data and meta information is transferred via an API between the editing system and the TMS. Not only translation data, but also status information such as translation progress can be transmitted.

Benefits of full automation with COTI Level 3

The implementation of COTI Level 3 brings with it several benefits that can dramatically improve the translation process:

• Fast data transfer: Thanks to the fully automated API, translation data is transferred seamlessly between systems without any delay.
• Increased efficiency: Large and complex translation projects can be processed more efficiently, since the workflow no longer has to include any manual steps.
• Round-the-clock operation: Automation facilitates continuous operation without human intervention, resulting in round-the-clock availability of translation data.
• Security: By eliminating manual steps, the risk of human error is minimised, which in turn ensures data transfer is more secure.
• Time and cost savings: Full automation leads to significant time savings, while also reducing the operational effort and costs involved in translation projects.

Conclusion

The introduction of COTI Level 3 signalled a major advancement in the translation industry; one which not only increases efficiency, but also improves the quality and reliability of translation processes. Through seamless integration and automated data transfer, companies are able to expand their global reach while also saving time and resources.

The following editorial systems can currently use COTI packages of various levels:

• • Schema ST4 – Quanos

• • TIM – Fischer

• • COSIMA – DOCUFY

• • docuglobe und XR – gds

• • Smart Media Creator – EC Systems

• • AEM – Adobe

• • GRIPS – STAR

• • and much more besides

With our translation memory system STAR Transit ^NXT; and our workflow solution STAR CLM, we provide links at all three levels – in order to transfer data efficiently, securely and quickly and to speed up translation processes.

We process your COTI packages automatically using STAR CLM! 

Contact us for tailored advice

Large language models (LLMs) could prove to be valuable assets for linguists in the context of language processes.
But what exactly are the advantages of this technology?

Our MT expert Julian Hamm will address this and other important questions in the context of the “LLM Use Cases in Language Services” TechTalk by lingo systems and provide insights into the world of language technologies and CAT tool development.
Curious to learn more?

STAR and lingo systems invite you to a free TechTalk

Secure your free ticket today and follow the discussion on 13^th March from 3.00 to 4.30 pm CET. The TechTalk will be held in English.

We would be delighted to have you join us!

Few words have characterised the year 2023 as much as “AI”.
But what does this buzzword actually mean for translation and language processes? Has the time now come for the relatively new technology of neural machine translation (NMT) to take a step back, and for large language models (LLMs) to take centre stage?

Are you still typing or are you already prompting?

How does this change the way that professional translators work?
A human at the helm, AI as the co-pilot. But what exactly could this look like in day-to-day translation?
This will be the focus of our hour-long webinar.

Opportunities and challenges of large language models

Specifically, this involves the question of how these new language technologies can be used to optimise core processes in the translation industry, including quality assurance, post-editing and terminology management. 

In addition to discovering a strategic approach to prompt engineering , you will also, through a series of practical examples, learn how CAT tools must be set up in the future in order to optimally support language experts in their work.
Prior knowledge: Basic knowledge of CAT tools and machine translation

MT expertise from STAR

The speaker, Julian Hamm, who holds a masters in translation, has been working in the language services since 2018. In his role as Machine Translation Consultant at STAR Deutschland GmbH, he coordinates the implementation of MT-based workflows and delivers expert advise, both within STAR and to external clients, on the exciting topics of MT and language technologies.

Humans at the helm, AI as the co-pilot – curious to see what the future holds? Book your place for the tekom webinar under Veranstaltungen (tekom.de) and get on board on 14th March 2024 at 4.30 p.m. (CET). Please note that this webinar will be held in German only.

See you there! – Sign up today!