Advancing linguistic AI: 5 key takeaways from Language Weaver Live

A few weeks ago, Language Weaver hosted an online virtual event bringing together industry experts to share their knowledge and explore the latest machine translation industry trends and technology. We will explore five key takeaways from the event that can help advance your multilingual content. 

1. The dream includes human in the loop to bring epic journeys 

In the session Localization at Flutter Entertainment with Language Weaver Goran Markus, Senior Localisation Manager explained the integration of Language Weaver for improved efficiency and higher quality translation. 

Goran discussed the future of machine translation at Flutter, including his company’s dream to be able to provide more value and excitement for customers and give them more epic journeys. He explained that for this to happen, translators and machine translation need to work together for human-in-the-loop translation.

Goran, Senior Localisation Manager explained: “Gold service will always come from translators. Translators bring the human element, language is human, translators will always know what is culturally acceptable, what is funny, and understand what will bring the epic journey to customer…who will train the machine, and understand the structural issues that machine is producing, machines will not be perfect because language is not perfect, language is not static and evolves and changes, and this is why we need human-in-the-loop.” 

2. The impact of adaptive machine translation to speak your language 

Neural Machine Translation (NMT) is inherently adaptable, but the degree to which the end user or organization has control over the adaptation will depend on your machine provider and the data you have available to you to customize that output. 

Adaptive machine translation can be useful in external, customer-facing documentation or high accuracy required on industry-specific translations. 

Heather gave a scenario of where adaptive MT can be helpful for the precise translation of specific content: 

“I was testing material for a cosmetics company. The term highlighter in terms of make-up appears in their product lines, the generic out-of-box machine translation gave the translation for a highlighter that is used to highlight a page for instance because this is more common in the training data. If the customer used that translation that could potentially be brand damaging for that cosmetics company. A native speaker would possibly be confused as to what product they are selling. Implementing a terminology element into the MT would make a big difference”. 

Language Weaver provides adaptive MT translations that are attuned and tailored to your specific requirements such as domain terminology, specialist industry vocabulary, brand style and tone. Not all adaptive MT models are equal. Language Weaver adaptive capabilities are secure within one easy-to-use UI. 

3. The importance of security when translating with LLMs 

A survey conducted by Slator released in May 2023, revealed that 1 in 10 Language service providers had already run a pilot using LLM for a specific use case. In addition, 90% planned to implement LLM in their workflows. 

Linguists today have various options to use LLMs including for specific capabilities, to dynamically interact with the LLM to extract a different output, and the ability to fine-tune LLMs. 

Security must be top of the list for most organizations. Companies who own the publicly available LLMs will keep your data and are entitled to mine it – a serious data issue. The solution is to instead use enterprise-grade, private LLMs, that keep your data secure and comply with data regulations. 

In the panel session, Elitiza at Coca-Cola Europacific Partners explained their experimentation with LLMs securely: “Being a large enterprise we look first at data security and privacy… "We are not allowed to use open AI ChatGPT, but we do have a testing group and we have a small GPT on premise which is secure…It makes me so much more productive than before having this access.” 

During the produce journey session Arnaud Simon, VP of Product Management, Language Weaver explained: "We believe that the future of Large Language Models is in using private, secure, fine-tuned pre-trained LLM." 

4. The power of using MT in your translation technology stack 

A recent RWS survey Global Understanding Unlocked, showed 82% of consumers would not purchase if the content was not in their language. When companies start implementing translation programmes, the localization process is not always straightforward, it can be siloed, across multiple divisions, and multiple geographies, and turnaround times can be slow, especially without technology like machine translation and without a centralized translation technology. 

In addition, 37% of corporations and 33% LSPs are looking to invest in machine translation. The benefits of integrating machine translation into the technology stack include: 

• Accelerate go-to-market 
• Reduce translation costs 
• Translate content you wouldn’t otherwise translate 
• MT is versatile, can be used for variety of jobs 

Trados can be used by translation teams of any size to improve their processes and meet the growing demand for content localization. It is easy for businesses to work efficiently with their own in-house localization teams, LSPs and individual translators. Trados includes Language Weaver across all editions. 

5. The vision for Language Weaver 

For adaptation: 

• Smarter adaptation algorithm, this will adjust automatically based on things like training data size and live adaptation metrics. 
• More adaptation control – so the user can control how much weight to give to their training data. 
• Bilingual data crawler – this will enable users to retrieve their bilingual data for the auto-adaptive LPs. 

Security improvements 

Being secure is at the core of everything Language Weaver does. It is important that organizations remain up-to-date with the consistently evolving data threats. 

Language Weaver is continuing security improvements including: 

• Bring our own key to Cloud 
• Additional encryption options in Edge 
• Secure Private LLMs 

Watch all sessions from Language Weaver Live 2023 on demand