ATA Law Division Professional Forum: Using Neural Machine Translation Engines for Translating Legal Documents

Takeaways from participants

Around 60 linguists attended our forum that took place in Miami during ATA64. They were sitting at seven tables, and each table discussed the following prompts: Our Future, Operations, Marketing, and Pricing. Please find here a summary of the participants’ takeaways.

1. Our Future

ATA gave a virtual conference in May 2023 on MT and AI called “Translating and Interpreting the Future.” In one of the sessions, Jonathan Downie mentioned three strategies to adapt to the new reality of our field: systematize, specialize, or diversify.

“Systematize” refers to the idea of increasing your productivity to cover a large volume of translation. “Specialize” is working in a niche market as an expert in one or several fields, whereas “diversify” refers to finding streams of revenue other than translation. How do you see your future? If you have applied one of those three strategies in legal translation (or perhaps another strategy), can you share your current practices with your group?

Takeaways from participants:

• Many participants noted a combination of all three strategies is needed.
• Specialization is considered as the main strategy among all participants since the MT technology is getting better at generating “general translations.” It is also a career path for those who are not in favor of post-editing. The field of specialization varies:
  • Terminology management
  • Some specific fields in law, e.g., immigration, securities, WIPO, patents, litigation, data protection
• The translators who choose systematization as a strategy have several sub-strategies to this end:
  • Master their CAT tool to optimize their use of it.
  • Outsource work to cater for expansion purposes.
  • Some participants also report using ChatGPT in legal translation jobs.
• Several participants have chosen to diversify their streams of revenue in different ways. Here are some examples:
  • Teaching Legal English
  • Interpretation (and finding a transition path to it)
  • Subtitling
  • Proofreading services
  • Paralegal services
  • Some note MT will lead to new job careers in the future
  • For participants who are not used to MT jobs, accepting post-editing jobs is also viewed as a diversification strategy
  • Unrelated work, such as dog sitting
• As a final note, some emphasize it is important we enjoy what we are doing and that our future is bright!

2. Operations

For this second prompt, we’ll focus on the operational side of post-editing in a legal context. If you work with direct clients, how did you choose your MT engine? If you work with translation agencies mostly, what specific techniques do you apply to cope with the caveats of MT output: non-consistent terminology, trademarks, quotes from precedent court cases, or specific statutes/laws/regulations? E.g., have you set up your CAT tool in some specific ways to make sure those aspects were taken care of, or do you follow a specific workflow to ensure you’re covering all your bases?

Takeaways from participants:

• Participants are using different MT engine tools:
  • DeepL
  • Cloud AutoML (used e.g., when DeepL is underperforming in some language pairs)
• Some participants use the following tools or methods for quality assurance. The choice of tools is also based on the sensitivity or confidentiality of the data processed:
  • Linguee
  • CAT integrated (manual) glossaries
  • Specific management tools
  • Specific QA platforms or in-house QA tools
  • Hiring translators who have a certification (e.g., lawyer-linguist)
• Some tools to ensure consistency throughout a project:
  • Filtering tool of the CAT tool
  • ChatGPT
• For several translators, terminology management and consistency are time-consuming tasks.
• Some participants note it is important to listen to the client’s wishes. In some cases, the use of MT tools is prohibited and could lead to a contract breach. (In other words, they choose the tools they use based on client demands and confidentiality requirements.) In other cases, some clients are very adamant about using MT technology, and the only way to keep those clients is to perform post-editing.
• Some participants still point to the need to check sources on official websites (e.g., GDPR, EU law…) as the output delivered by the MT engine is defective.

3. Marketing

How do you market post-editing work? Do you discuss the pros and cons of using MT/AI engines in your pre-negotiating talks to manage your prospect’s expectations? Have you tailored your website, social media profiles, and service agreements to offer this new service? On the flip side, how do you market your work as a human translator/reviewer and differentiate your services from what MT offers, especially in the legal field?

Takeaways from participants:

• Several groups discussed client education (advising the client on when to use MT or not) and the importance of understanding the difference between an MT-generated translation and the added value a human translator brings.
• In court interpretation, the human element is key. For now, interpreting is less affected than translation when it comes to new technologies.
• Some translators refuse to post-edit, as they find they deliver better work and are more efficient using their CAT tool.
• Some translators point to the importance of marketing your speed in legal translation, where time is often of the essence, and clients might be less sensitive to price than in other industries.
• A participant reported they shared with the client the new investment they made in an MT technology and showed the client the time and money saved. They think they landed the project because they put the client’s interest first.
• Some translators adapt their pricing strategies to the type of services requested by the client: some charge per page, per hour, per hour. Some use a combination of these pricing strategies based on the client’s wishes.
• Some translators itemize their invoices to show all the research done and are transparent about the tools used. They also annotate their work to show their reference sources. This is a way to show the added value in your work compared to a translation generated by a machine.

4. Pricing

Since the quality of machine translation output is so variable, how do you evaluate the time needed to perform a post-editing project? If your output varies based on the MT quality output, what strategies have you put in place to manage your client’s expectations in terms of billing?

[We kindly remind you not to discuss specific dollar amounts or price recommendations]

Takeaways from participants:

• Some translators refer to the need to charge for our expertise (linguists are the “conductor of the orchestra”).
• Some translators advise to first review the MT output before accepting a post-editing job.
• In terms of how they set their prices for MT post-editing, some translators charge by the hour, others by the word, and others by the page.
• Other linguists charge the same price for post-editing jobs or human translations: the underlying philosophy is to offer the same quality rendition of a job from scratch. At the same time, MT allows you to accept a bigger volume or charge no extra fee for quicker service.

We hope these takeaways spark ideas and help you navigate the deep changes AI has in our industry!