Intento Releases 9th Annual “State of Translation Automation 2025” Report

ByRicardo

Intento has launched its 9th annual trade report, The State of Translation Automation 2025 (previously State of Machine Translation).

The report analyzes the most recent advances in translation automation and exhibits how AI improves translations to fulfill particular enterprise and technical necessities. It offers world enterprises sensible steerage to elevate end-user satisfaction, drive adoption of multilingual programs, and align translation technique and tooling with their particular language necessities.

Intento analyzed 46 machine translation engines and LLMs (giant language fashions) throughout 11 language pairs, conducting full-text evaluations towards 5 enterprise necessities: normal translation high quality, terminology, tone of voice, formatting (tag dealing with), and full-text consistency.

The report compares:

  • Off-the-shelf fashions—each NMT programs (e.g., Amazon, Baidu, DeepL, Google NMT, Microsoft, Oracle, Tencent) and LLMs (e.g., Anthropic, Cohere, Gemini, LLaMA, OpenAI),
  • The similar fashions adjusted to fulfill particular necessities utilizing out there customization choices (excluding fine-tuning),
  • A multi-agent workflow (Translator → Reviewer → Post-Editor),
  • Human translation.

What the information exhibits

  • Requirements-based translation outperforms generic engines: Solutions custom-made to particular necessities outperformed off-the-shelf fashions. Human evaluators usually couldn’t distinguish AI-generated from human translations—and in some instances rated the AI increased.
  • Multi-agent workflows ship one of the best outcomes: A multi-agent resolution (with separate brokers to confirm necessities and take a look at outputs, designed to keep away from compounding hallucinations) produced the very best common efficiency, incomes “greatest” scores in 9 of 11 language pairs. These workflows mix brokers for terminology integration, tone adjustment, and post-editing to persistently elevate high quality.
  • Clear necessities, fewer errors: In testing, baseline programs averaged 10–15 errors per textual content. Requirements-based options lower that to 0–2—at the very least 80% fewer errors, and in some instances eliminating them completely. This hole makes requirements-based customization important for skilled translation.

    “Automatic translation is now not about selecting the ‘engine’—it’s about constructing options that meet particular translation necessities,” stated Konstantin Savenkov, CEO & Co-founder of Intento. “The most telling indicator emerged from our human analysis: reviewers usually couldn’t distinguish AI from human translation—and typically rated AI translations increased. The multi-agent method, using a number of AI brokers to confirm and take a look at necessities, delivered one of the best common efficiency, as anticipated. However, these brokers require language-specific customization, and standardizing this course of is important to stop extreme customized engineering and debugging overhead that presently limits adoption to solely large-scale purposes.”

The put up Intento Releases 9th Annual “State of Translation Automation 2025” Report first appeared on AI-Tech Park.

Similar Posts