If you’re a driver, you’ve probably heard of the two-second rule. Staying at least two seconds behind any vehicle is considered a rule of thumb for drivers wanting to maintain a safe following distance at any speed. The two seconds don’t represent safe stopping distance but rather safe reaction time.
Reaction time depends on several factors, including age and experience. For most people, it’s in the 0.2 to 2 second range. The same reaction time approach can be applied to machine translation post-editing (MTPE).
Source: “What Is a Safe Following Distance?,” Smart Motorist
When linguists deal with raw MT engine output, they sometimes spend too much time analyzing it and deciding whether or not it’s usable at all. Here’s where imposing some limits on decision-making time could be of help: if you spend two seconds looking at an MT segment (after familiarizing yourself with both the source and target text), and see that you cannot easily edit it, discard it and translate it from scratch (or use a lower fuzzy match from the translation memory).
Research on efficient post-editing shows that it might be difficult to determine whether you should correct bad MT output or if it would be faster to delete and re-translate any segments of borderline quality. But if you want to be efficient, you shouldn’t spend more than two to three seconds determining this. As with driving, some experts consider two seconds to be the minimum time you should allow, but recommend applying a three-second rule instead. This means an extra margin of safety and confidence (both when driving and post-editing).
Before the Neural Machine Translation era began in 2014-2017 there was a lack of post-editing guidelines online. In 2009, TAUS addressed the post-editing dilemma with the recommendation: “In a customer support application, for example, MT users should avoid PE where possible, or limit it to error items that can be evaluated as correctable within 2 seconds. Otherwise PE becomes too expensive for the possible end user benefits.”
In 2013, quoting the rules of Microsoft, Mesa-Lao (as quoted in a Comparative Study of Post-Editing Guidelines, 2016) provided suggestions on how to decide whether MT output should be recycled. Those included a “5-10 second evaluation” recommendation on making such decisions.
Talking to MTPE practitioners in 2020, you’ll hear some of them consider five seconds to be an actual norm.
Indeed, there are times when the two-second rule doesn’t apply (you could argue that leaving just two seconds of distance between your car and the one in front of you is still dangerous). It was designed for making decisions in a normal traffic situation under normal circumstances. Some MT segments may be particularly challenging to process and therefore require more time to decide whether ‘tis nobler “to MTPE or not to MTPE.” At the same time, as with driving, thinking for too long about every decision makes you lose precious time. And what’s MT here for? Why, to save time and effort.
We recently introduced you to the two- (or five-) second rule, which is essentially the reaction or decision-making time a linguist should spend judging whether to post-edit a segment of machine translation (MT) output or to retranslate it. This rule of thumb aims to help increase the linguist’s productivity when working with MT.
Do you remember the last time when people were NOT talking about machine translation (MT)? We don't. Wherever you go, there’s someone talking about MT. With few exceptions, it seems like the only major disruptors in our industry over the past few decades have been breakthroughs in language technology.
Some machine translation providers are holding out hope for MT systems that adapt to document context. Could this development eliminate the need for custom MT engines? Will context-enabled MT help MT achieve human parity? Will we still need to customize a few years from now? Let’s discuss further.