The year starts with two major trends: a boom in AI, and a dim economic outlook. To stay relevant in a world where globalization processes are slowing down, and machines are becoming more adept at language tasks, localization teams need to reinvent themselves once more. We present a collection of specific ideas that respond to market trends and advances in technology.

1. Roadmap for Large Language Models

ChatGPT is just the beginning. Google is planning to counter OpenAI with Deepmind’s Sparrow, China has had a larger model called Wu Daio 2.0 since May last year, and Inception Institute for Artificial Intelligence (IIAI) in Abu Dhabi released the largest Arabic GPT in January 2023. The race is on. 

While countries and tech giants battle each other for language model dominance, mere mortals in the real economy will race to figure out how to apply these models efficiently. Localization departments possess insider information about their business’ multilingual needs, from personalized education to insurance claims summary. So they are in a position to take on the “internal product owner” role for large language models.

References:

GPT3 demo – a map of use cases

Timeline – models being released

NLP use cases – 30 ideas on implementing LLMs

2. Expand into Authoring & Global Content

Authoring is where the revolution is happening due to the popularity of ChatGPT, and localization leaders have an opportunity to add this growing area to their portfolio. Today’s savvy writers create text drafts with prompts, then get them fact-checked and polished stylistically. Or, they use GPT to rewrite drafts and improve the readability score.

Start a structured program to adopt GPT writing in your business: 

• • Mass-produce content for SEO

• • Automate technical writing of manuals, medical reports, legal contracts, protocols, and statements 

• • Create personalized story reports for clients and application users

• • Generate meeting summaries

References:

Business writing assistants – tools to augment authoring

ChatGPT in medical writing – applied experiment

AI in Content Marketing – how to use generative NNs in SEO

3. Start on Language AI Ethics & Governance

Today’s AI is full of biases inherited from the Internet data used to train it, and it needs to be cleaned up and improved over the next 5-7 years. This is an emerging governance area, relevant to all organizations, from Microsoft to the Vatican. 

Is a bank communicating with its clients in a respectful and friendly way, or is it offending customers and misusing their data, opening the way for potential lawsuits? Does an online store have a compliant and safe product catalog and a review base? Can a voicebot in a telecom call center be posing risks for the mobile operator? 

References:

Fairslator.com – a tool to identify bias in machine translation

Amazon MT-GenEval – dataset for gender bias detection

Translated Wikipedia Biographies – dataset for gender agreement

Microsoft’s Responsible AI principles – guidance material

4. Video: Build Automated Subtitling and Dubbing

In the past year, automated subtitling systems grew in quality, and automated dubbing systems made their first real contact with the market. Content archive holders like Phillips, Siemens, TED, and Adobe have used automated video localization technology in the past year with success, and it’s time to put it to work on all your video archives.

Video localization technology spans over 100 different tools and services. Consumer tools are easy to learn but do not have many features for professional quality assurance. Business tools include workflow components and APIs to be able to process hundreds of videos in a structured way. Professional video localization platforms include heavy-duty customization options to be able to adhere to the strict guidelines of Hollywood studios, streamers, and broadcasters.

References:

A list of video localization tools grouped into consumer, business, and professional

Microsoft AI voice generator – for example voice workbench

5. Optimize Budgets with Machine Translation

The World’s economic growth is slowing down from 6% to just 2.7% (IMF), and investment in tech has dropped by half. Organizations are putting their investments on hold, and are pressuring operations to reduce costs and reduce staff headcount. Localization teams may protect themselves by showing efficiencies. It has become easier to convince your internal stakeholders to switch to machine translation.

• • Move forums, blogs, and catalog content to MT

Where the translation does not carry health or legal hazards and does not penalize your website’s ranking in Google search, you have the option to MT it, preferably with a disclaimer. Automated translation and the human-reviewed translation may be combined efficiently on the same website page, as demonstrated by the travel portal Klook.

• • Optimize with post-editing

After introducing MT in the translation process, it’s reasonable to apply a discount to sentences that have no match in the translation memory. Targeting fuzzy matches of under 85% or even 95% with a 15-50% price reduction, depending on MT accuracy may save a significant amount.

References:

How to Optimize Translation Spend with MT – explains how post-editing metrics become discounts

Freelancer marketplace with listed rates for translation and MT review