Are we on the verge of removing all language barriers with AI?
Olga Beregovaya, VP of AI at Smartling, joins host Conor Bronsdon to tackle this question, discussing the evolution from rule-based NLP to today's powerful LLMs. Together, they confront the persistent challenges that stand in the way, like the English-centric nature of AI, domain-specific inaccuracies, and the unpredictability of model hallucinations. Olga unpacks the difficulties faced when striving for accurate, nuanced translation across all languages, especially under-resourced ones.
Beyond these hurdles, the conversation explores the cutting-edge opportunities and technical innovations driving progress, including RAG, the rise of purpose-built models, agentic AI workflows, and the potential of multilingual multimodality. Olga shares insights into boosting translator productivity, achieving more predictable quality, and the path toward human parity in translation, examining how technology and human expertise will shape the future of global communication.
Chapters
00:00 Introduction and Guest Welcome
01:14 Evolution of NLP: From Rule-Based to Machine Learning
02:40 Challenges in AI Translation
04:21 Biases in Language Models
05:28 Inference Time and Latency
05:44 English-Centric AI Models
08:53 Opportunities in AI Translation
09:14 Industries Benefiting from Language AI
10:36 Human-in-the-Loop Translation
12:06 Architectural Innovations in Language AI
16:20 Success with RAG Architectures
17:58 Multilingual Vectorization
19:54 Agentic AI in Translation
24:35 Data Sets and Data Privacy
28:30 Using Smaller, Purpose-Built Models
32:10 Future of AI in Translation
36:37 Conclusion and Farewell
Follow the hosts
Follow Atin
Follow Conor
Follow Vikram
Follow Yash
Follow Today's Guest(s)
LinkedIn Olga Beregovaya
LinkedIn Smartling
Check out Galileo
Try Galileo
