150+ LANGUAGES · ISO 18587 CERTIFIED · 24/7 SUPPORT
Machine Translation and AI Services
Machine translation and AI services combine neural MT engines with ISO 18587 post-editing, cutting translation cost 40–60% vs pure human work while keeping ISO 17100 quality.
- 150+ Languages
- ISO 18587 Certified
- 24/7 Support
Get a free, no-obligation quote
Complete guide
Everything you need to know
Machine translation and AI services sit inside the wider remit of Professional Translation Services in London UK, blending neural engines and large language models with human post-editing to serve UK businesses across 150+ languages.
What are machine translation and AI services?
Machine translation and AI services are language services that convert text between languages using neural machine translation engines and large language models, with an optional human post-editing layer under ISO 18587 that lifts raw output to publish-ready quality for business content. The offer runs on two tracks: raw output for internal comprehension, and full post-edited output for publication. Translation memories, glossaries and quality assurance sit around both tracks.
The service covers 6 core deliverables:
- Raw neural machine translation for internal comprehension.
- Light post-editing for internal knowledge bases and catalogues.
- Full post-editing to ISO 18587 for publish-ready copy.
- LLM-based AI translation steered by prompt engineering.
- Translation API integration into a CMS or repository.
- Quality assurance and terminology governance across all four.
How is machine translation different from AI translation?
Machine translation is the automated conversion of text between languages by dedicated MT systems, while AI translation is a broader label covering large language models and generative AI tools that also translate text but were not built solely for that task. Neural machine translation engines are trained end-to-end on aligned bilingual corpora. Generative AI tools translate as one skill among many. For a deeper comparison against human work, see What is the difference between machine translation and human translation?.
Does machine translation use AI?
Machine translation uses AI: modern neural machine translation engines are built on deep learning and neural networks, and the newest tools layer large language models on top for context handling, tone control and terminology consistency. Rule-based and statistical systems from the 1990s used no AI in the modern sense. Today’s engines do.
How have machine translation and AI translation evolved?
Machine translation evolved through 3 paradigms: rule-based and statistical systems in the 1990s–2000s, neural machine translation from 2016 onward, and generative AI translation from 2022 with large language models such as GPT and specialised models like DeepL. Each paradigm changed the workflow, the quality ceiling and the role of the human translator.
| Paradigm | Era | Core technology | Quality ceiling |
|---|---|---|---|
| Rule-based and statistical systems | 1990s–2000s | Hand-written rules, phrase alignment | Literal, brittle |
| Neural machine translation | 2016 onward | Sequence-to-sequence neural networks | Fluent, contextual |
| Generative AI translation | 2022 onward | Large language models (GPT-class) | Instruction-following, context-aware |
What were rule-based and statistical machine translation systems?
Rule-based and statistical systems generated translations from hand-written linguistic rules and phrase alignment probabilities mined from bilingual corpora, producing brittle, literal output that struggled with idiom and long-distance grammar. Statistical Machine Translation is largely superseded by neural machine translation, though it remains useful for research on low-resource language pairs.
What is neural machine translation?
Neural machine translation uses sequence-to-sequence neural networks trained on billions of aligned sentences to produce fluent, contextual output. It powers Google Translate, DeepL, Microsoft Translator and agency-tuned NMT engines. Deep learning gives the engine the ability to model context beyond the phrase. Read the technical detail on Neural Machine Translation.
How do large language models change AI translation?
Large language models change AI translation by handling context across full documents, following stylistic instructions in a prompt, and translating rare language pairs through pivoting — but they hallucinate confidently, which is why post-editing is required for publication. LLMs also enable a two-step workflow: NMT for the first draft, LLM for stylistic refinement.
Is ChatGPT considered a machine translation tool?
ChatGPT is a general-purpose large language model that performs machine translation as one of many tasks, but it is not a purpose-built machine translation service: it lacks translation memories, quality assurance workflow, terminology enforcement and the audit trail required for ISO-compliant delivery. It has no signatory, no confidentiality contract by default, and no ISO 17100 accountability. For the document-level view, see Can Chatgpt translate documents?.
Can ChatGPT translate better than Google Translate?
ChatGPT translates better than Google Translate on stylistic register, context handling and instruction-following for marketing copy, while Google Translate and DeepL remain stronger on rare language pairs and on the raw throughput needed for high-volume translation tasks. The choice depends on the translation project, not the tool’s popularity. See Is Google Translate good enough for professional use in the UK? for the UK-specific view.
| Tool | Register | Domain terminology | Data residency | CMS integration |
|---|---|---|---|---|
| ChatGPT | Strong with prompt | Only with glossary in prompt | API tier controllable | Via API |
| DeepL | Strong | Glossary support | EU option | Native connectors |
| Google Translate | Neutral | Limited | Google Cloud regions | Cloud Translation API |
| Agency-tuned NMT | Tuned to brand | Enforced via TM and glossary | UK and EU private tenant | API + full workflow |
When is ChatGPT the wrong tool for a UK business translation?
ChatGPT is the wrong tool for certified translation, legal translation, medical translation and any document going to a UK court, the Home Office, HMRC, Companies House or a UK regulator: the model has no signatory, no ISO 17100 accountability and no confidentiality agreement covering the client’s data. Regulated content stays on the human ISO 17100 TEP path.
What are the different types of machine translation and AI translation services?
There are 5 distinct types of machine translation and AI translation services: raw neural machine translation, MT with light post-editing, MT with full post-editing to ISO 18587, LLM-based AI translation with prompt engineering, and translation API integrations plugged into a CMS or repository.
| Type | Human intervention | Best fit | Quality |
|---|---|---|---|
| Raw NMT | None | Comprehension of large volumes | Internal only |
| Light post-editing | Critical errors fixed | Knowledge bases, catalogues | Comprehensible |
| Full post-editing (ISO 18587) | Full human review | Publish-ready business content | Human-parity |
| LLM with prompt engineering | Prompt design + review | Marketing, tone-sensitive copy | Depends on prompt |
| Translation API integration | Automated pipeline + QA | CMS, product database, repo | Scales with workflow |
What is raw neural machine translation output?
Raw neural machine translation output is the direct result of an NMT engine with no human intervention, fit only for internal comprehension of large volumes such as email triage, competitor monitoring or forum research. It should never be published. It should never be signed.
What is light post-editing of machine translation?
Light post-editing corrects only critical errors — mistranslations, terminology mistakes and offensive output — while accepting stylistic imperfection. It suits internal knowledge bases and product catalogues where readability, not polish, is the goal. Learn more on Light Post-Editing.
What is full post-editing to ISO 18587?
Full post-editing to ISO 18587 is a comprehensive human review that produces quality indistinguishable from human translation, carried out by the same expert translators used for certified work so terminology and tone stay consistent. See Full Post-Editing for the service specification and What is post-editing machine translation and do London agencies offer it? for the London context.
What is AI translation with prompt engineering?
AI translation with prompt engineering uses large language models steered by structured prompts, glossaries and few-shot examples to control register, style and domain terminology at translation time. Details of the method are on Prompt Engineering for Translation.
How does a translation API fit into a CMS or repo workflow?
A translation API plugs directly into a CMS, code repository or product database, sending source strings out and returning translations without staff copying text into a chat window. It is the automation layer that turns MT and AI from an experiment into a repeatable workflow. See Translation API for connector detail.
Pricing
How much do AI translation services cost in the UK?
AI translation services in the UK cost from £0.06 per source word for machine translation with full post-editing, against £0.12/word for human legal translation, £0.13/word for technical and £0.14/word for medical. MTPE delivers a 40–60% cost saving versus pure human translation on high-volume, low-risk content.
| Service | Price per source word | Turnaround benchmark |
|---|---|---|
| MT with full post-editing (ISO 18587) | From £0.06 | 1,500–2,000 words/day per linguist |
| Human legal translation (ISO 17100 TEP) | £0.12 | 2,000–2,500 words/day |
| Human technical translation (ISO 17100 TEP) | £0.13 | 2,000–2,500 words/day |
| Human medical translation (ISO 17100 TEP) | £0.14 | 2,000–2,500 words/day |
What is the MTPE price per source word?
Machine translation with full post-editing starts at £0.06 per source word for internal comms and product catalogues, rising with domain complexity, language pair rarity and turnaround urgency. The engine cost, glossary import and post-editor rate all sit inside that per-word price.
What factors influence the cost of AI translation?
There are 6 factors that influence the cost of AI translation:
- Source word volume and repetition rate.
- Language pair rarity.
- Subject-matter domain (legal, medical, technical, marketing).
- Post-editing tier (light versus full).
- File format and CMS integration effort.
- Turnaround urgency, including the 25–50% same-day surcharge on rush work.
When does MT with AI post-editing become economically justified?
MT with AI post-editing becomes economically justified when source volume exceeds 20,000 words, the content is repetitive, and distribution is internal-only or low-risk. Below that threshold the setup cost of engine tuning and glossary import outweighs the per-word saving.
Does Google Translate charge a fee for business use?
Google Translate is free for consumer web use but charges through the Google Cloud Translation API for programmatic business use, billed per million characters. Agency post-editing sits on top of that engine cost and is where publish-ready quality is added. Full detail on Does Google Translate charge a fee?.
Why us
What are the benefits and challenges of using AI for translation?
What are the pros of AI translation for UK businesses?
The pros of AI translation for UK businesses are 4:
- 40–60% lower cost on high-volume content.
- 24/7 throughput on tight campaign deadlines.
- Consistent terminology across large document sets.
- The ability to break down language barriers on internal comms in near real-time.
What are the cons of AI translation?
The cons of AI translation are 5:
- Systematic errors in idiom, register and domain terminology.
- Hallucinated content in generative AI tools.
- Weak performance on rare language pairs.
- Data-security exposure when free tools train on user inputs.
- Legal inadmissibility of unsigned output.
Read the deeper analysis on What are the limitations of machine translation for business documents?.
Is my translation data safe with AI tools?
Translation data is safe only with secure tools that contractually exclude client content from AI model training, encrypt data in transit and at rest, and sign a confidentiality agreement. Free consumer LLMs and translation apps typically do none of these. Private-tenant NMT engines, signed NDAs and UK/EU data residency close that gap.
How can I improve the quality of machine translation?
There are 5 proven ways to improve the quality of machine translation:
- Apply full post-editing under ISO 18587.
- Feed the engine domain-tuned translation memories and glossaries.
- Use prompt engineering on LLM-based translation.
- Run quality assurance checks against source.
- Route regulated content off MT entirely to full human ISO 17100 TEP.
How do translation memories and glossaries improve AI output?
Translation memories reuse previously approved sentences and glossaries lock domain terminology, so the AI engine produces translations that already match the client’s house style, brand voice and regulatory wording. TMs cut cost and raise consistency at the same time.
What does ISO 18587 full post-editing actually change?
ISO 18587 full post-editing requires a qualified human post-editor to compare each machine output against the source, correct errors of meaning, terminology, register and grammar, and sign off on a fully accurate translation. The output quality is indistinguishable from human translation.
How does prompt engineering improve LLM translation?
Prompt engineering improves LLM translation by injecting a role, glossary, tone rules and few-shot examples into the request, so the model translates in the correct register the first time instead of producing generic output that a post-editor must rewrite. It compresses two workflow steps into one.
When must content skip MT and go straight to human TEP?
Content must skip MT and go straight to human ISO 17100 TEP when it is certified, legal, medical, court-bound or Welsh public-facing: the signed certification statement, legal accountability of the translator and independent reviser stage cannot be replicated by a post-editor.
| Content type | MT fitness | Required workflow |
|---|---|---|
| Certified translation | Not suitable | Human ISO 17100 TEP + certification |
| Legal translation | Not suitable | Human ISO 17100 TEP |
| Medical translation | Not suitable | Human ISO 17100 TEP |
| Welsh public-facing content | Not suitable | Human ISO 17100 TEP |
| Engineering specs for regulated output | Internal drafts only | Human TEP for release — see Engineering Translation Services |
| Internal documentation | Suitable | MT + light post-editing |
| Product catalogues, e-commerce copy | Suitable | MT + full post-editing |
| Low-risk website localisation | Suitable | MT + full post-editing |
How does AI impact the translation industry and will AI replace human translators?
AI is reshaping the translation industry by shifting linguists from typing translations to post-editing and prompt-engineering AI output, but it is not replacing human translators. Certified, legal, medical and creative translation still require human expertise, accountability and an ISO 17100 signature. The London perspective is covered in Can machine translation replace professional translators in London?.
How is the translator’s role changing with AI?
The translator’s role is changing from typist to editor and terminologist: post-editing machine output, curating translation memories, engineering prompts, and running quality assurance on AI translation agents that generate first drafts at scale. The linguistic judgement stays human.
Which translation tasks stay fully human?
Certified, sworn, legal, medical, high-stakes marketing transcreation, literary and Welsh public-facing translation stay fully human under ISO 17100 TEP, because they require signed accountability, cultural judgement and creativity that generative AI tools cannot supply.
How should a UK business choose between AI translation tools and a translation service?
A UK business should choose AI translation tools for internal, low-risk, high-volume content and a professional translation service for published, regulated or certified work. Routing decisions are best made on a fitness-for-purpose matrix, not on tool marketing.
Which content types suit MT with AI post-editing?
MT with AI post-editing suits 6 content types: internal documentation, product catalogues with consistent terminology, e-learning modules, knowledge bases, forum monitoring and website localisation of low-risk marketing pages. Retail catalogues at scale are a strong fit — see Retail Translation Services.
Which content types require a human ISO 17100 workflow?
Certified translation, legal translation, medical translation, contracts, court bundles, patient records and Welsh public-facing content require a human ISO 17100 workflow with translator, independent reviser and proofreader, each contributing a distinct quality-control stage.
How does an AI translation agent plug into a CMS?
An AI translation agent plugs into a CMS through a translation API or connector that pulls source strings, sends them to the tuned engine, applies the client’s translation memory and glossary, and returns translations for human post-editing. Native connectors exist for WordPress, Drupal, Adobe Experience Manager, Contentful and Git-based repositories.
What quality assurance runs on top of MT and AI output?
Quality assurance on top of MT and AI output covers 5 checks:
- Automated terminology checks against the client glossary.
- Tag and formatting verification to preserve layout.
- Spell and grammar checks in the target language.
- Source-target completeness (no dropped segments).
- Final human sign-off by an ISO 18587-qualified post-editor.
Services
How are machine translation and AI services delivered by a London translation agency?
A London translation agency delivers machine translation and AI services through a two-track workflow: a tuned neural MT engine plus large language model layer for first-draft output, then full post-editing to ISO 18587 by the same expert linguists used for ISO 17100 certified work. Throughput runs at 1,500–2,000 words per linguist per day with 24/7 project management. The end-to-end service is documented on MT Post-Editing.
- Source analysis and TM/glossary import.
- Engine and prompt configuration.
- Machine translation generation.
- ISO 18587 full post-editing.
- Quality assurance checks.
- Delivery in the source file format.
The 6 most common questions on machine translation and AI services cover accuracy, ChatGPT, cost, CMS integration, translation memory reuse and data safety. The answers below draw on our ISO 17100 workflow for UK clients.
Frequently asked questions
How does neural machine translation work?
Neural machine translation (NMT) works by using deep learning models trained on vast bilingual datasets to predict the most probable translation of a text, word by word and sentence by sentence. Unlike older rule-based or statistical systems, NMT models — such as those powering Google Translate or DeepL — analyse entire sentences for context before generating output. They continuously improve through exposure to more data, but still struggle with ambiguity, domain-specific terminology, cultural nuance, and low-resource language pairs, which is why human review remains essential for professional use.
When should you use AI translation vs a professional translation service?
AI translation is suitable for low-stakes, internal, or gist-reading purposes, while a professional translation service is required for any content where accuracy, legal validity, or brand reputation is at stake. Machine translation can handle quick internal communications, rough drafts, or high-volume content that needs only a general understanding. However, certified documents (e.g. birth certificates, contracts, immigration papers), marketing copy, technical manuals, and client-facing content demand a qualified human translator — and in many UK legal and official contexts, only a certified human translation is accepted.
What industries should not rely on machine translation?
Legal, medical, pharmaceutical, and financial industries should not rely solely on machine translation due to the high risk of costly or dangerous errors. A mistranslated drug dosage, contract clause, or compliance document can result in patient harm, legal liability, or regulatory penalties. Other sectors where machine translation alone is insufficient include immigration and government services (which require certified translations), technical engineering documentation, and marketing or brand communications where tone and cultural nuance are critical. Professional translation agencies in the UK apply specialist translators with subject-matter expertise to these high-risk fields.
How do professional translation agencies in London use AI tools?
Professional translation agencies in London use AI tools as productivity aids within a human-led workflow, not as a replacement for qualified translators. Common applications include AI-assisted translation memory software (e.g. SDL Trados, memoQ), terminology management databases, and machine translation post-editing (MTPE), where a human linguist reviews and corrects AI-generated output to professional standards. This hybrid approach can reduce turnaround times and costs on large-volume projects while maintaining the accuracy, consistency, and certified quality that businesses and official bodies require across the UK.