Stockhead: Small language models could be the next big disruptor in AI translation
- Specialised small language models (SLMs) outperform large, general-purpose AI in translation accuracy, notes Straker
- SLMs deliver faster turnaround and lower costs by focusing on specific languages and industries
- Straker’s Tiri model suite uses domain-specific training and human feedback to continually improve results
Special Report: Small language models promise faster, cheaper and more accurate AI translation, notes Straker.
Over the past five years, the translation industry has ridden a wave of AI adoption. Large language models (LLMs) have moved from niche tools to near-standard in content localisation but their generalist nature leaves room for a new contender.
The co-founder and CEO of ASX-listed language tech company Straker (ASX:STG), Grant Straker, argues that small language Models (SLMs) – purpose-built for specific industries and language pairs – are the next leap forward.
“SLMs are designed to do one thing extremely well: understand the meaning, terminology and context behind content in a specific domain and language,” he says.
By training SLMs on curated translation memories and industry-specific data, the company notes its Tiri model family delivers higher accuracy and contextual understanding, reducing the need for costly human post-editing.
These smaller, more focused models also require less computing power, meaning faster turnaround times and lower costs for clients in sectors like finance, healthcare, and law.
‘SLMs will finish the job’ started by LLMs
Tiri’s integration of human feedback directly into the workflow – via what’s known as reinforcement learning from human feedback (RLHF) – ensures the models evolve in line with client needs, says the company.
For organisations managing high volumes of multilingual content, that can translate into better quality, higher trust, and shorter delivery times.
Straker believes the next chapter in AI translation won’t be defined by scale but by specialisation delivering translations that not only sound fluent but are precisely understood in context.
“LLMs lit the fuse for AI translation,” says Straker. “But SLMs will finish the job delivering high-impact, high-trust, and high-speed localisation at scale.”
High-powered integrations
Straker has extended its reach into the fast-growing enterprise automation market, announcing a new integration with AI workflow platform n8n.
The integration introduces Straker’s Verify product – its AI-powered translation quality and compliance tool – into n8n’s ecosystem, giving over 230,000 active users and 3,000 enterprise clients the ability to automate translations, receive real-time quality scores, and seamlessly escalate content to human linguists when needed.
This platform strategy already includes integrations with Slack, and with Microsoft Teams on the horizon.
That news came just days after the company reported record profitability for the year ending March 31 2025 and winning a price target upgrade from broker Ord Minnett.