When an alarming confrontation broke out in the parking lot of a supermarket, a well-meaning passerby attempted to break things up and prevent the fight from escalating. In all the chaos, he brandished his gun in the hopes that those involved would split up and go about their day. Intimidated, the individuals who were fighting contacted the police.
Officers arrived at the scene to question the bystander, only to find he didn’t speak much English. Nonetheless, the officers communicated in English, using a free machine translation (MT) app to render their questions into his native Russian (and vice versa).
“They interrogated him using this translation tool and determined, based on this communication, that they had enough evidence to arrest and prosecute him for intimidating or threatening people with a gun,” said David Utrilla, CEO of the Salt Lake City-based language service provider US Translation Company, who recently made a widely circulated Linkedin post about his experience as an expert witness on the case described above.
MultiLingual spoke with Utrilla last week to learn more about the case and how his testimony on speech technology and MT factored into the bystander’s eventual acquittal. To protect the defendant’s privacy, Utrilla did not disclose identifying information such as his name, location, or timing of the incident, but was able to share some of the problems that may arise when law enforcement officers resort to MT to question individuals with limited English proficiency.
According to Utrilla, the officers used the conversation setting of a free MT tool to question the defendant. Intended for informal use, the conversation setting aims to allow two parties to speak in their preferred language, translating the spoken input into the listener’s preferred language. However, tools like this have a lot of room for error, making them less than ideal in particularly sensitive instances.
In his testimony, Utrilla explained that conversational MT tools have two main steps: transcription of the source language input, followed by a translation into the target language. If a word in the source language isn’t transcribed properly, the translation will, in turn, render the speaker’s intended meaning incorrectly. Moreover, a perfect transcription doesn’t guarantee a perfect translation — while Utrilla acknowledged that MT has made major strides in recent years, it’s still not necessarily 100% accurate. Thanks to Utrilla’s testimony, the defendant was acquitted.
Instead of using an MT tool, Utrilla noted that the police should have contacted a human interpreter to provide language assistance — “Something of this legal magnitude requires a human translation,” he said, noting that in this instance, the police could have accessed a human interpreter if they had thought to contact somebody.
Indeed, advocates for language access have urged law enforcement agencies to better educate officers and other personnel about their local language access policies to prevent episodes like this from occurring as much as possible.
Utrilla told MultiLingual that he ultimately shared his experience on Linkedin to inspire translators, interpreters, linguists, and other professionals in the language service industry to think about the ways they can use their skills to help others and break language barriers. Although some may find the rise of MT intimidating, Utrilla notes that this case is evidence that human translators and interpreters will never truly become obsolete.
“I’ve been doing this for 27 years, and I have seen the evolution of the industry when it comes to translation technologies,” he said. “To translators and interpreters, I want to say: ‘You are not obsolete, you are important, you are needed, and you will be needed always.’ ”