Joining SlatorPod this week is Longyue Wang, a Research Scientist at Tencent AI Lab, where he is involved in the research and practical applications of machine translation (MT) and natural language processing (NLP).

Longyue Longyue expands on Tencent’s approach to language technology where they integrate MT with Tencent Translate (TranSmart). He highlights how Chinese-to-English MT has made significant advancements, thanks to improvements in technology and data size. However, translating Chinese to non-English languages has been more challenging.

Recent research by Longyue explores large language models’ (LLMs) impact on MT, demonstrating their superiority in tasks like document-level translation. He emphasized that GPT-4 outperformed traditional MT engines in translating literary texts like web novels.

Longyue discusses various promising research directions for MT using LLMs, including stylized MT, interactive MT, translation memory-based MT, and a new evaluation paradigm. His research suggests LLMs can enhance personalized MT, adapting translations to users' preferences.

Longyue also sheds light on how Chinese researchers are focusing on building Chinese-centric MT engines, directly translating from Chinese to other languages. There's an effort to reduce reliance on English as a pivot language.

Looking ahead, Longyue's research will address challenges related to LLMs, including handling hallucination and timeless information issues.

Florian and Esther discuss the language industry news from over the past month, where NWN Carousel, a Massachusetts-based cloud communications company, secured a USD 6.5 million contract with Baltimore, Maryland, to provide AI-powered multilingual emergency 9-1-1 services.

Meta has unveiled Llama 2, its latest large language model, trained on 40% more data. Despite its extensive training, the model's focus on English data and limited coverage of other languages might restrict its suitability for non-English language applications.

In a financial results round-up, Straker Translations reported a 2% revenue growth with increased revenues from IBM and strong demand in Europe. After a successful 2022, ZOO Digital is now facing headwinds from Hollywood’s writers and actors strike. The pod also checks in on Keywords Studios’ and Honyaku Center’s financials.

Esther’s M&A corner features Belgian LSP ElaN Languages acquiring Lingua-World, expanding into the German market. Further north, LanguageWire, one of the largest LSPs in the Nordics, acquired Finnish LSP Delingua to further strengthen its position in the region.

Meanwhile, in funding, Phrase raised EUR 12m in debt financing from CIBC Innovation Banking, to support further development of the Phrase Localization Suite. Over in Wyoming, Language I/O has secured USD 8m in a Series A1 funding round to enhance its platform with generative AI technology and for hiring.

In this week’s SlatorPod, we are joined by Florian Stègre, Founder of Checksub, to discuss how the SaaS platform powers subtitling and dubbing for SMEs and enterprise clients.

Florian's entrepreneurial journey began after graduating from business school and founding video production company, ROOM Agency, which laid the foundation for Checksub.

Checksub primarily serves SMEs and enterprises, where they have the option to work with external providers in the video editor.

Florian expands on the platform's unique features, including animated subtitles and voice cloning, which analyzes the emotion and type of voice and provides a better match for the original speaker.

Florian shares why he took a bootstrapping approach to company building, focusing on achieving product market fit with an eye on long-term sustainability rather than rapid scaling.

When it comes to building tech in the fast-evolving age of AI, Florian keeps up to date with recent research, attends conferences and workshops, and collaborates with industry professionals.

Checksub ensures they remain ahead in the competitive AI market by focusing on skills development in their team. On the roadmap, Florian plans to enhance user experience, improve the translation editor, and add more capabilities to their AI dubbing platform.

Rocio Martinez, VP of Language Services at Jellyfish, joins SlatorPod to talk about the role of creative content localization at the digital marketing agency.

Rocio explains that while the language services team at Jellyfish function as an LSP, their positioning in the market is broader, and they act as an internal solution provider that supports the various divisions within Jellyfish. She emphasizes the importance of connecting every part of the creative process, from creation to adaptation, and monitoring performance and optimization.

Rocio provides an overview of J+ Creative, a proprietary set of tools that assist their teams in the production and localization process, including workflow management, deliveries, approvals, and version control.

Rocio talks about the use of back translations in transcreation, primarily for creative content that requires significant adaptation from the source material. Rocio shares how finding the right copywriting talent and linguists can be a challenge, especially for rare languages or specific sectors.

Rocio discusses the potential of large language models and generative AI tools for content generation and localization. She mentions that Jellyfish is actively testing and integrating AI tools, focusing on improving workflows, and considering prompt engineering to enhance content generation.

Regarding the future, Jellyfish's initiatives revolve around AI, with specialist teams being trained to identify opportunities and implement them across marketing capabilities. Rocio also mentions the potential of Pencil, a generative AI tool for content and asset creation.

In this week’s SlatorPod, we are joined by Graham Neubig, Associate Professor of Computer Science at Carnegie Mellon University, to discuss his research on multilingual natural language processing (NLP) and machine translation (MT).

Graham discusses the research at Neulab, where they focus on various areas of NLP, including incorporating broad knowledge bases into NLP models and code generation.

Graham expands on his Zeno GPT-MT Report comparing large language models (LLMs) with special-purpose machine translation models like Google Translate, Microsoft Translate, and DeepL. He revealed that GPT-4 was competitive from English to other languages, but struggled with very long sentences.

When it comes to cost comparison, Graham highlights that GPT-3.5 Turbo (the model behind the free version of ChatGPT) is significantly cheaper than Google Translate and Microsoft Translator, but GPT-4 (available via OpenAI’s subscription) is more expensive.

Graham predicts that companies will likely move towards using general-purpose LLMs and fine-tuning them for specific tasks like translation. The discussion also covers the recent flurry of speech-to-speech machine translation system releases.

Graham talks about his startup, Inspired Cognition, which aims to provide tools for building and improving AI systems, particularly in text and code generation. Graham concludes the pod with advice for new graduates in the NLP field and his plans for Zeno and the Zeno report.

Florian and Esther discuss the language industry news over the past month, with ChatGPT traffic hitting its peak in May 2023 and large language models being released left, right, and center.

The EU Commission (EC) unveiled the launch of a pilot program that will see the EC use its own machine translation tool to translate certain press releases. In mid-June, media localization company Dotsub announced it had ceased operations, due to its current liquidity situation.

In more positive news, Microsoft's beloved Language Portal has made a comeback after the company announced that it would be shut down at the end of June.

Over in M&A, XL8 has acquired AHT to diversify and grow its customer base to US government agencies. Also branching out is RWS as it bought legal and legislative content management system Propylon for a total purchase price of USD 47.3m.

Meanwhile, in funding, Czech localization startup Localazy announced at the end of June that they raised EUR 1m. Voice generation platform, Eleven Labs, announced it had raised USD 19m in a Series A round. 

Deepdub Co-founder and CEO, Ofir Krakowski, joins SlatorPod to talk about the AI dubbing platform’s goal to make content more accessible through automation and human collaboration.

Ofir’s experience in the Israeli Air Force's machine learning and innovation department led him to use his knowledge in deep learning and generative AI to co-found Deepdub with his brother, Nir.

The CEO outlines the importance of human creativity in translation and how they have built a platform that incorporates both AI solutions and human curation. When it comes to AI dubbing, Deepdub prioritizes authentic and natural-sounding voices and incorporates domain-specific technologies to improve translation accuracy.

Ofir recognizes the impact of generative AI, exemplified by ChatGPT, and emphasizes the importance of building a product rather than just a technology. The company actively engages with organizations such as the BFFS, a union for artists, to find ways to combine human talent with technology.

While Deepdub serves enterprise customers in the entertainment industry, they also cater to professional audiovisual creators by making the dubbing process more readily accessible through their recently launched SaaS product Deepdub GO.

Looking ahead, Deepdub plans further language expansion, with a focus on addressing the diverse linguistic landscape of countries like India.

ICON plc’s Senior Vice President of Commercialisations & Outcomes, Brian Mallon, joins SlatorPod to shed light on the crucial role of language services within clinical research organizations (CROs).

Brian recounts ICON's growth journey, from a team of five in Dublin back in 1990 to its present-day workforce of over 41,000 across 50 countries. Brian outlines the clinical cycle for drugs or medical treatments, which consists of several phases aimed at evaluating safety, efficacy, dosage, and potential side effects before making them available to the public.

Brian reveals how language services play a vital role in improving clinical trials by enhancing the quality of data, reaching a more diverse patient population, and increasing retention and access to trials. He shares how medical translation and linguistic validation in clinical research require expertise in medical terminology and cultural nuances of the target language.

Brian talks about how ICON led an industry-wide shift towards insourcing language services instead of relying on external language service providers. Key motivators include cost savings, increased efficiency, faster turnaround times, and the importance of data security and confidentiality.

Brian explains how the cost of failure in translations is high, as errors can impact patient safety, regulatory compliance, and the overall success of the research. To manage these risks, ICON’s Language Services team implements stringent quality assurance processes, audits, and ongoing training.

When it comes to the EU Clinical Trial Regulation and Clinical Trial Information System, Brian believes it has not had a major impact on language services in terms of volumes. However, the expectation is that over time, these regulations will improve the efficiency of conducting trials in Europe and make the EU more attractive for large-scale clinical trials.

The pod rounds off with Brian’s view on the role of AI in the life sciences market and ICON’s  "string of pearls" approach when it comes to broadening its capability in line with industry trends in the life sciences sector.

Florian and Esther discuss the language industry news of the week, covering key takeaways from SlatorCon Remote such as lip sync AI dubbing demos from NeuralGarage and how FIFA and DeepL tackle sports localization.

After unveiling its first-half results for 2023, RWS CEO Ian El-Mokadem emphasized the advantages of utilizing AI and large language models (LLMs) to enhance existing products. In M&A, Unbabel has completed its second acquisition of 2023 by acquiring Bablic, to expand its website and marketing localization solutions.

AI video generation platform, Synthesia, has secured USD 90m in funding at a billion-dollar valuation for its multilingual avatars. One other unicorn, Cohere, has raised USD 270m in a Series C funding round, bringing the total raised amount to USD 445m.

The duo do a SWOT analysis of what LLM capabilities mean for the language industry, specifically for LSPs, TMS providers, machine translation providers, and linguists. 

Unilingo Founder and CEO, Farbod Mansorian, joins SlatorPod to talk about launching the leading dubbing provider for Youtube creators, enabling them to get billions of additional views.

As a regular consumer of YouTube, Farbod came across TED Talks, and after translating a talk for his mother, the idea to bring this to other non-English speaking communities was born. This idea evolved into Unilingo, where Farbod formed a team to work on dubbing YouTube videos so content creators can reach fans across the world.

Farbod reveals how he develops the business by connecting with popular creators in person at YouTube conferences around the world and scaling through word of mouth.

Asked about AI dubbing Farbod emphasizes that AI has not yet been able to capture emotions and intonations effectively, which is evidenced by lower average viewing retention rates than human-dubbed content.

Farbod discusses the pros and cons of YouTube’s multitrack audio feature, with multiple language options increasing engagement and showing dedication to the audience. However, text on screen, thumbnails, and comments pose challenges as they are not currently translated into multiple languages. 

The CEO addresses the challenges faced by voice actors and how Unilingo pays their talent within hours instead of weeks. The pod rounds off with Unilingo's roadmap, where they will continue their human-first AI-assisted approach to developing software tools that enable more voice actors to enter the dubbing space.