Reader Polls: Should Regulators Require Flagging of Machine Translated Content?

Regulators Flag Machine Translated Content

Bill Gates once famously said that bad news is a headline. That sage saying may have recently applied to the man, but Covid fatigue — not to mention the shorter news cycle in Gates’ case — may have changed things.

All of a sudden, good news is the headline, and even spiked as a search term worldwide in 2020. Although the numbers have since settled closer to pre-pandemic range (albeit still higher), there remains a general optimism. After a year, the world has learned how to deal.

This positive outlook proliferated at the recent SlatorCon Remote, especially evident in discussions around media localization and e-commerce.

And the theme of the good outweighing the bad seems to be holding up as more language service providers (LSPs) release their latest financial results. Check the good news out of Straker, Stratus, RWS-SDL, ZOO, TransPerfect, Capita TI, Keywords, LanguageLine, Ai-Media. As of press time, only Honyaku (Pro) and Appen (Pro) have reported significant declines.

Slator 2021 Language Industry Market Report

80-pages. Market Size by Vertical, Geo, Intention. Expert-in-Loop Model. M&A. Frontier Tech. Hybrid Future. Outlook 2021-2025.
$680 BUY NOW

Slator also conducted an interim sentiment check as the language industry approaches the half-year mark. The majority of LSPs and freelancers surveyed on May 14 expect “Good growth” (59%) in 2021. Less than a fifth looked forward to “Strong growth” (18%), while an equal number said business would be “Flat.” Only a small number (2.3%) had a “Slightly negative” or “Very negative” business outlook.

On June 7, 2021, subscribers to Slator’s Strategy Package and market report buyers will be able to join the complimentary briefing on Slator’s 2021 Language Industry Market Report to gain deeper insights into the global language market and its growth potential in 2021 and beyond.

Automating Quality Reviews

Another interesting exchange at SlatorCon Remote happened toward the end of the Dell-Welocalize track on what makes for a longstanding, mutually beneficial buyer-LSP relationship. (Dell and Welocalize have been buyer and LSP for 17 years.)

Weighing in on the topic of automating quality reviews for machine translation (MT) output (e.g., BLEU) during the Q&A, Darin Goble, VP Solutions at Welocalize, said, “AI systems have gotten a lot smarter and they are really good at telling how well some other automated system has performed.”

Big tech has indeed been exploring automated QA, going by recent research out of Google and Amazon. Google published a paper on how automatic MT evaluation trained by professional translators can outperform inexperienced crowdsourced (human) evaluators.

Slator 2021 Data-for-AI Market Report

Slator 2021 Data-for-AI Market Report

44-pages on how LSPs enter and scale in AI Data-as-a-service. Market overview, AI use cases, platforms, case studies, sales insights.
$380 BUY NOW

Meanwhile, Amazon is working to further automate subtitle translation reviews and tested their model on both MT and professional (human) subtitlers in the translation pipeline. Even tech-enabled LSP, Lilt, said they are currently working on automated MT QA.

However, research has also backed up what Wayne Bourland, Translation Director at Dell, said; that is, “We’re still gonna need humans to evaluate this stuff […] I don’t see that changing in the near term.” (Check out: How to Improve Automatic Machine Translation Evaluation? Add Humans, Scientists Say)

Slator readers appear to be evenly split between those who think automated quality assessment “Definitely has its use” (37.5%) and those who say getting a machine to do QA will be “Forever futile.” The remaining respondents (25%) admitted that the issue is beyond their expertise.

There Ought to Be a Law…

What appears to be forever futile is legislation catching up with tech advancements. The EU recently published a proposed framework for AI legislation — but then left out machine translation!

The EU draft legal framework addresses deepfakes (not banned, just regulated) and even proposes fines (up to 6% of global turnover or USD 36m, whichever is higher). But it largely stays silent on MT. Should MT output be flagged? And how would the law classify post-edited machine translation (PEMT)?

LocJobs.com I Recruit Talent. Find Jobs

LocJobs is the new language industry talent hub, where candidates connect to new opportunities and employers find the most qualified professionals in the translation and localization industry. Browse new jobs now.

LocJobs.com I Recruit Talent. Find Jobs

Maybe the framework is just a starting point, or maybe the EU aims to leave the nitty-gritty to member countries. Poland, at least, has begun to dip its toes in the water by ruling on the use of Google Translate.

What do Slator readers think? A stunning majority of respondents (70%) said “Yes,” regulators should require MT output to be flagged as such. A fourth said it depends on the application, while the rest replied, “No” (6%).

Treading Water

So how many of us are raring to resume attending conferences IRL? It was somewhat of a draw between safety and cautious optimism in a recent Slator survey. A third of respondents said they preferred to wait and see, and about the same replied, “No, I like remote.”

A fifth, however, were enthusiastic, with “If there’s a way to get there, definitely” garnering 22%.

A little over a tenth, meanwhile, said they would register for an in-person conference, but only in their own country, while the rest (5%) would — if the conference was not very large.

Check out Slator’s list of upcoming conferences for May and June — at least two of which are free to attend!