When Is It Safe to Use Machine Translation?

Popularity and myths

Over the past decade, machine translation (MT) has gone from a novelty to an everyday tool. It's built into browsers, office software and corporate document systems. People are used to Google Translate or DeepL being one click away, letting them understand a foreign-language text at no cost.

For businesses it looks attractive too: minimal cost, fast results and the ability to process large volumes of data. That's especially tempting as the amount of multilingual content keeps growing in online stores, web services and global companies.

But this is exactly where the main myth appears: that machine translation can fully replace a professional translator. In practice, it can't. International standards for translation services, such as ISO 17100, make it clear that translation is not just converting text but a whole process: preparation, translation, revision, quality control and data protection. A machine can help, but it doesn't take responsibility.

Professional translators' associations make the same point: translation is an individual service whose quality depends not only on language but on understanding the context, the industry and the culture. A machine can imitate a text, but it won't assess legal risks, protect a company's image or find the right tone.

So the question is not "can machine translation be used?" but "when is it justified, and when does the risk outweigh the benefit?"

Where the risks are

If you treat machine translation as a magic button, the risks for the company and its clients can far exceed the apparent savings.

1. Semantic errors and distorted meaning

Even modern MT systems don't guarantee accurate meaning. The problem is especially visible in legal, medical and technical texts, where one word can change everything. The ASTM F2575 standard stresses that translation quality is measured not by how similar a text looks but by how accurately and completely it conveys the source. Here a machine error becomes critical.

2. Lost style and inconsistent terminology

Even well-trained systems don't always follow a company's house style, terminology and tone. ISO 17100 treats terminology agreed with the client as a mandatory part of the translation process. A machine without glossary integration and without a post-editor doesn't meet that requirement.

3. Confidential data leaks

Many companies use free online services without realizing that the uploaded text may be stored on the provider's servers and used to train the system. Free versions of popular services don't guarantee confidentiality, as their terms of use state. Aware of the risks, the European Commission built its own closed system, eTranslation, for sensitive data.

4. Legal and reputational consequences

Machine translation errors in contracts, medical reports or financial statements can lead to lawsuits and fines. In Russia, the GOST R 57309-2016 standard states that a translation service must meet the client's requirements and that its result has legal significance. Machine translation without a specialist's review doesn't meet these criteria.

When machine translation makes sense

Machine translation can be a genuinely useful tool if you draw clear boundaries around it. International practice identifies several situations where MT is justified and doesn't carry serious risks.

1. To get the gist

If you only need to quickly understand what a text is about, machine translation does the job well. That could be personal correspondence, news, market reviews or documents that won't be published or used in official procedures. The EU's eTranslation service exists for exactly this: it gives public bodies and partners secure automatic translations for internal use, not for publication without human editing.

2. For large volumes of similar content

Online stores, support services, knowledge bases and technical documentation often contain hundreds or thousands of similar text fragments. In such cases machine translation followed by post-editing processes large volumes faster while keeping the terminology consistent. Industry research groups such as TAUS note that MT works especially well in domains with repetitive structures: IT, telecom and retail.

3. When glossaries and term bases exist

If a company has an approved term base, machine translation can be configured to use it. The post-editor then only checks style and consistency rather than fixing every term. ISO 17100 describes this as terminology management, and MT acts as a supporting tool here.

4. For secondary texts with a limited audience

Internal instructions, support tickets or drafts that never go public can be machine translated with light post-editing. This takes the load off specialists and frees resources for the projects that really matter.

When machine translation is not an option

Despite impressive progress, there are areas where machine translation is unacceptable. The cost of an error is simply too high, legally, financially or reputationally.

1. Legally binding documents

Contracts, tender documents, regulations and court decisions require not only precise language but legal accountability. ISO 17100 requires revision by qualified specialists, and GOST R 57309-2016 requires a translation to comply with legal requirements and its intended purpose. Raw machine translation meets neither.

2. Medical and pharmaceutical texts

Drug instructions, medical records and clinical reports: here an error can affect a person's health and life. In Europe and the US such texts fall under strict regulators (EMA, FDA) that require professional translation with mandatory review.

3. Marketing, advertising and creative copy

Slogans, ad campaigns and branding texts call for transcreation. What matters is not a literal translation but adaptation to the culture and perception of the target audience. At best, a machine produces grammatically correct text that doesn't work as a marketing message.

4. Confidential information

Free online services carry a risk of data leaks: the text you enter may be stored and used for training. For confidential material, use only secure corporate systems (like eTranslation in the EU) or professional providers who guarantee compliance with an NDA.

5. Public texts where an error undermines trust

An annual report, press release, executive statement or research paper: in texts like these, any inaccuracy reads as unprofessional. The client risks not only their reputation but the trust of customers and investors.

International and Russian standards

Because machine translation is now part of everyday practice, its use has become a subject of regulation at international and national levels. Standards set out where and how MT may be used, and where only professional translation with a full quality control cycle will do.

ISO 17100:2015, Translation services: Requirements for translation services. The main international standard in the industry. It sets requirements for the qualifications of translators and revisers, the translation process and checks, and terminology and data confidentiality management. ISO 17100 describes translation as a process that includes translation, revision and, where needed, review. Raw machine translation doesn't meet these criteria but can serve as a supporting tool if all other requirements are met.

ISO 18587:2017, Post-editing of machine translation output. A standard dedicated specifically to working with MT. It defines two levels of post-editing:

• light post-editing: the text becomes understandable but not perfect, acceptable for internal use;
• full post-editing: the result must be comparable to a translation from scratch and fit for publication.

ASTM F2575 (USA). An American standard that treats translation as a process with a mandatory quality control system. It covers not only the competence of the people involved but also risk assessment when choosing translation methods, MT included.

GOST R 57309-2016 (Russia). The Russian counterpart of ISO 17100. It sets requirements for the qualifications of translators and editors, mandatory quality control and the legal significance of translation as a service. It explicitly requires taking into account the purpose of the text and the provider's responsibility, so unedited machine output can't count as a translation service under GOST.

Why should a client care about standards?

• Transparency. Standards mean quality can be assessed against defined criteria rather than subjectively.
• Lower risk. If a provider works to ISO or GOST, the client knows that any use of MT is controlled and goes through post-editing.
• Comparable offers. The client can compare providers not just on price but on compliance with standards.

Post-editing as the solution

Raw machine translation rarely meets quality requirements. That's why international standards define a separate procedure, post-editing, which turns a machine draft into a text that meets professional norms.

What is post-editing?

ISO 18587:2017 defines post-editing as editing machine-translated text to bring it to a quality level comparable to human translation.

• A post-editor needs the same competences as a translator under ISO 17100: linguistic qualifications, subject knowledge and terminology skills.
• Machine translation is a supporting tool, while responsibility for the final quality stays with the human.

Two levels of post-editing

1. Light post-editing

   • Goal: make the text understandable quickly.
   • Stylistic rough edges and simplifications are acceptable.
   • Used for internal documents, rough reading and texts with a limited audience.
   • Risk: the text may read as mechanical, so it's not suitable for publication.

2. Full post-editing

   • Goal: bring the translation to the quality of a professional translation from scratch.
   • All errors of meaning, terminology, style and grammar are corrected.
   • The final text is ready for official publication or use in legal processes.
   • This is the format that ISO and GOST standards describe as a full translation service.

Why it matters to the client

• Quality control. Post-editing lets you use MT without risking the result: the machine saves time, the human guarantees accuracy.
• Clear expectations. Standards such as ISO 18587 give the client criteria to agree on the level of post-editing needed and avoid unrealistic expectations.

Practical advice for businesses

Even when a company understands what machine translation can and can't do, the key practical question remains: how do you decide when MT is enough and when you need full professional translation?

1. Define the purpose of the text

• Internal use (analytics, news monitoring, technical correspondence): MT with light post-editing is acceptable.
• External publication (company website, press releases, reports, legal documents): you need a translator, and MT can only be a supporting tool.

2. Assess the risks

• Legal consequences. Any error in a contract or instructions can result in fines or lawsuits.
• Financial risks. Errors in marketing materials can cost sales or require costly image repair.
• Reputational risks. A poor text undermines trust in the brand.

The approach of ASTM F2575 and GOST R 57309-2016 is a useful guide: always consider a translation in light of the text's purpose and the possible consequences of errors.

3. Use standards as a benchmark

• ISO 17100 and GOST R 57309 set requirements for the process and translator qualifications. A provider that works to these standards organizes quality control.
• ISO 18587 helps you agree on the level of post-editing you need: light for internal purposes or full for publication.

4. Agree on a strategy with your provider

The best way to avoid misunderstandings is to agree in advance on:

• the purpose of the text;
• the acceptable level of post-editing;
• the use of glossaries and translation memories;
• confidentiality requirements, such as a ban on uploading to online services.

Put these points in the brief. It protects both sides: the client gets a predictable result, and the provider gets clear quality criteria.

5. Build long-term relationships

Machine translation is effective only with a permanent partner: a translator or agency can accumulate your terminology, train MT on your data and build processes around your industry. Over time this improves quality.

Conclusion

Machine translation has become a familiar tool, and its strength is precisely that it is a tool, not a standalone solution. It helps process large volumes of similar texts faster and speeds up draft work. But where an error can lead to legal, financial or reputational losses, you can't do without a professional translator.

International and Russian standards, including ISO 17100, ISO 18587, ASTM F2575 and GOST R 57309, are clear: translation quality is ensured only through a qualified specialist, quality control and accountability. A machine can be an assistant, but responsibility stays with the human.

For a business, that means a simple strategy:

• use MT where the risk is minimal (internal tasks, rough reading, large volumes of similar content);
• add post-editing when the text goes beyond internal use;
• bring in professionals for anything involving law, medicine, marketing and public communication.

This gives you the right balance: speed from technology, accuracy and reliability from specialists.

Key takeaway: machine translation is not a magic button but part of a well-built process. Those who understand its limits and combine MT with professional control get real benefits. Those who rely on the machine alone risk far more than they save.

ServiceAI translation review: rates, turnaround and examples