The Future of Translation: Collaboration Between Humans and AI

Artificial intelligence (AI) has accelerated over the past decade, transforming everything from creative writing to scientific discovery. Translation is no exception. Neural machine translation (NMT) systems – particularly those powered by large language models (LLMs) such as OpenAI’s GPT‑4, Anthropic’s Claude and Google’s PaLM – are now producing remarkably fluent output. The question for 2025 is no longer whether machines can translate text, but whether they can reliably replace professional human translators and interpreters.

As a researcher in translation studies and a content strategist, I often hear conflicting opinions: some colleagues view AI as a threat to their livelihoods, while others see it as a tool that augments human expertise. To cut through the hype, this article reviews current research and market data on AI translation quality, adoption and economic impact. It also explores the subtler human‑centred aspects – cultural nuance, ethical considerations and the effects on language skills – that remain essential in the translation process.

Takeaways:

• Language services market projected to hit USD 96.1B by 2032.

• AI achieves 60–70% near-human parity in high-resource languages.

• Struggles remain with idioms, nuance, and low-resource languages.

• Bias and inclusivity issues persist in AI systems.

• Human roles shift to quality control and cross-cultural mediation.

How Good Is AI Translation in 2025?

In 2025, AI translation demonstrates strong performance in high-resource languages, often approaching human parity in everyday and technical texts. However, its limitations remain visible in domains requiring cultural nuance, idiomatic expression, or context-dependent interpretation. While neural models supported by large language data have improved fluency and accuracy, challenges persist for low-resource languages and sensitive content. Thus, AI serves as a powerful complement to human translators, rather than a full replacement.

Quality Benchmarks

Quality benchmark for AI translations

Multiple studies confirm that AI translation quality has improved dramatically since 2020. In a 2024 blind comparison study by the localization platform Lokalise, native speakers compared AI translations across more than 600 pairs of sentences in several language pairs. They rated translations as “good” in 56–80% of cases, and LLM‑powered systems like Claude 3.5 Sonnet consistently ranked first or second across languages. Across thousands of live projects, Lokalise reported 84% acceptance rates for AI suggestions – meaning translators or project managers publish the machine output without modifications. Expert evaluators in the same study still preferred human translations 22% of the time, but AI suggestions were accepted in 78% of cases.

Academic comparisons echo these results. A study at the Massachusetts Institute of Technology (MIT) found that professional translators achieved an average COMET score of 0.78, whereas top machine translation engines scored 0.66. COMET is a neural metric that correlates strongly with human judgments of translation quality; a higher score indicates better semantic accuracy and fluency. The researchers concluded that human translators still outperform NMT systems by approximately 18 percentage points in nuanced accuracy, especially for idiomatic or culturally rich texts.

Acceptance Rates Across Language Pairs

Acceptance rates vary by language pair and context. According to Lokalise’s 2025 data, AI translations are accepted without changes in 83% of English→Spanish cases and 84% of English→French projects. Acceptance rises to 88% for English→Portuguese and Polish, while languages such as Chinese and Arabic hover around 74–79%. These numbers show that AI performs best on high‑resource Western languages and still struggles with complex morphology and syntax in some non‑Indo‑European languages.

Accepted AI translation pairs with percentages

Other industry studies report equally impressive gains. Microsoft Translator added a domain‑adaptive layer to its NMT system and boosted financial‑sector translation quality by 15%. Hybrid workflows (machine translation followed by professional post‑editing) can achieve 97% accuracy while lowering costs. And according to the website‑localization provider Weglot, 60% more organizations adopted hybrid translation models in 2024–2025, suggesting that businesses are increasingly comfortable combining automation and human review to balance speed and quality.

Speech and Multimedia Translation

Although this article focuses on text translation, AI is also advancing rapidly in speech and video localization. In KUDO’s 2025 market forecast, industry analysts predict that AI platforms will reach 85% accuracy in translating idioms and emotional context in spoken dialogue by the end of 2025. Real‑time speech translation tools are expected to become a $1.8 billion market, and hybrid interpretation models (AI plus human interpreters) are projected to account for 40% of all interpretation services with a 33% year‑on‑year increase. These figures illustrate how AI is expanding beyond written text into live events, conferences, and multimedia localization, although human involvement remains important for accuracy and audience engagement.

Cost and Efficiency: Where AI Shines

One reason companies are rushing to adopt AI translation is cost savings. Professional human translation typically costs between $0.08 and $0.25 per word and can take days or weeks to complete a large project. In contrast, machine translation tools can produce a rough draft almost instantly, lowering per‑word costs to around $0.05. When combined with post‑editing by qualified linguists, hybrid workflows average $0.08 per word, providing substantial savings while still meeting quality benchmarks.

These cost differences explain why a recent Forbes survey found that 44% of businesses plan to use AI for writing content in other languages. A McKinsey survey also noted that 34% of AI deployments in corporations are focused on marketing and sales, areas that often involve high volumes of text needing rapid translation. Start‑ups and small businesses, which may not be able to afford professional translators for all content, are especially drawn to AI solutions.

Human vs. Machine: Where Gaps Remain

Despite significant advances, clear gaps remain between human and machine translation. AI systems excel at speed and consistency but often falter when texts demand cultural sensitivity, creativity, or pragmatic judgment. Human translators, by contrast, bring contextual awareness and interpretive skills that go beyond word-for-word rendering. These enduring differences highlight the complementary rather than substitutive relationship between humans and machines in translation practice.

Nuance, Cultural Context, and Ethical Considerations

Despite impressive gains, machine translation still struggles with nuance, humor, and culturally bound expressions. Localization specialists note that AI can mistranslate idioms or misinterpret gendered language, leading to awkward or offensive output. Translators.com reports that cultural nuance mistakes in automated translations can reduce user engagement by up to 35%. In fields such as law, medicine, and marketing, these errors can have serious financial or legal consequences. As a result, human oversight remains essential for sensitive and high‑stakes content.

Moreover, AI translation quality depends heavily on context. Lokalise’s research shows that companies that supply detailed glossaries, style guides, and previous translations to their AI systems can achieve acceptance rates above 90%, whereas organizations that feed generic prompts often struggle with accuracy. This suggests that AI may not eliminate the need for linguists; instead, it changes their role into one of curating data and fine‑tuning models.

Impact on Employment and Language Skills

Artificial intelligence is also reshaping the labor market for translators and interpreters. A 2024 survey cited by the Centre for Economic Policy Research (CEPR) reported that over three‑quarters of translators expect generative AI to adversely affect their future incomes. Another CEPR study, using U.S. employment data, found that for each 1 percentage‑point increase in machine translation usage, translator job growth fell by 0.7 percentage points, resulting in about 28,000 fewer translator positions between 2010 and 2023. The same study observed that job postings requiring Spanish, Chinese and German skills grew 1.4, 1.3 and 0.8 percentage points slower in regions with high machine translation adoption.

While these trends might worry human translators, they also highlight new opportunities. As routine translation tasks become automated, the demand for post‑editing, transcreation, multilingual content strategy, and quality assurance is rising. Hybrid translation models not only lower costs but also create new roles for linguists who can refine AI output and ensure cultural and regulatory compliance. Interpreters, particularly those working with rare languages or complex legal/medical content, are also adapting to workflows where AI handles routine terms and they focus on high‑stakes segments.

Adoption Trends and Market Growth

The global language services industry continues to grow. Market research firm Fact.MR valued the sector at US $60.68 billion in 2022 and forecasts it will reach US $96.21 billion by 2032 with a 5.94% compound annual growth rate. Although 2025 projections vary, most analysts agree that AI‑powered translation is driving a significant share of this growth. For example, Weglot notes a 60% year‑over‑year increase in hybrid translation adoption, while KUDO predicts that AI‑driven speech translation tools will represent a $1.8 billion market by 2025.

Adoption is also surging in specialized industries. Lokalise’s 2025 Localization Trends Report recorded a 700% increase in AI translation usage in the finance sector between 2023 and 2024, with human‑only translation decreasing from 67.7% to 35.8%. Healthcare and legal sectors – traditionally cautious about automation – are similarly embracing AI translation for preliminary drafts while relying on human experts for review. These shifts suggest that organisations no longer view AI as a novelty but as a critical component of global expansion strategies.

The Role of Translation Studies in an AI‑Driven Future

In an era where AI is reshaping the landscape of translation, translation studies can offer valuable perspectives on both opportunities and challenges. Rather than prescribing fixed solutions, several avenues of inquiry and potential contributions emerge for scholars and practitioners to consider:

1.     Evaluation frameworks – While metrics such as BLEU and COMET remain widely used, they do not necessarily capture cultural nuance or pragmatic appropriateness. Future research might explore alternative methods of evaluation that reflect human perceptions of meaning and quality more closely.

2.     Multimodal translation – With the increasing convergence of voice, video, and augmented reality, questions arise about how translators and machines can navigate multimodal content. Investigating how humans process multimodal translation, and how AI might support or extend those processes, could become an important research direction.

3.     Bias and inclusivity – Since machine translation systems are shaped by their training data, they inevitably risk replicating or amplifying bias. Continued critical inquiry into how bias operates in AI translation, and how inclusivity of minority languages and dialects can be fostered, remains a pressing concern.

4.     Academic-industry connections – The evolving translation landscape invites dialogue between universities, professional training programs, and technology companies. Collaborations may provide insight into how translators engage with AI in practice, from post-editing to data curation, while also informing curriculum development.

5.     Shifting professional roles – As automation transforms certain aspects of translation work, it may be worthwhile to investigate how roles evolve toward tasks such as quality assurance, cross-cultural mediation, transcreation, or domain-specific expertise. Understanding these shifts can help situate translators not only as linguistic mediators but also as supervisors of AI output.

AI and Humans Are Partners, Not Rivals

The evidence from 2024–2025 suggests that AI translation has reached a level where it can reliably handle routine, high‑volume content – often exceeding 80% acceptance rates across major language pairs and reducing per‑word costs by more than 50%. However, human translators remain indispensable for legal, medical, and culturally nuanced materials, where even small errors can have outsized repercussions. Post‑editing and hybrid workflows show that the most effective approach combines AI efficiency with human judgment, achieving 97% accuracy while preserving cultural sensitivity.

As AI technology continues to evolve, translators and translation scholars should embrace it as a tool that augments rather than replaces their expertise. By focusing on quality assurance, ethical considerations, and nuanced communication, human translators can ensure that the power of AI is harnessed responsibly and effectively – ultimately enabling more people around the world to access information, services and stories in their own languages.

Sources:

1.     Translators.com. “AI vs. Human Translation: What to Expect in 2025.” Translators.com, 13 June 2025.

2.     Pokorny, Elizabeth. “AI vs Human Translation: In‑Depth Comparison and Latest Trends.” Weglot, 2 July 2025.

3.     Wolff, Rachel. “AI Translation Quality Achieves Human Parity: Is This the End of Language Barriers?” Lokalise, 13 Aug. 2025.

4.     Walford‑Delahaye, Chloë. “AI Speech Translation in 2025 & Beyond: Technology, Data, Trends & Predictions.” KUDO, 14 Jan. 2025.

5.     Frey, Carl Benedikt, and Pedro Llanos‑Paredes. “Lost in Translation: AI’s Impact on Translators and Foreign Language Skills.” CEPR, 22 Mar. 2025.

6.     Fact.MR. “Language Services Market Analysis, By Service – Global Market Insights 2022 to 2032.” Fact.MR, 2022.