Computer-Generated Translation and Gender Bias

Did you know that computer-generated translation tools, such as Google Translate and Bing Translate, often provide output that is gender-biased? Consider the following example, shared by Nataly Kelly, VP of Market Development here at Smartling, in this insightful piece written by Neal Ungerleider for Fast Company:

If you type the word “engineer” into Google Translate for Spanish, it defaults to the masculine, “ingeniero.” If you type “female engineer,” you’ll get “ingeniero de sexo femenino,” which means something like “male engineer of feminine sex.” If you type “female engineer” into Bing Translate, you’ll get “ingeniera,” but you still get a default that is male.

How do professional translators handle this issue? Many will include both options if the gender is unknown. For example, if they have very little context to determine gender, they’ll often write, “el/la ingeniero/a” to indicate that the gender could be either male or female. And, if you type “female engineer” into a source text to be translated by a human being, you’ll get “ingeniera” on the other side – definitely not something as confusing as “male engineer of feminine sex.” Where gender is concerned, humans still do a better job than machines.

Are tool providers, like Google Translate and Bing Translate to blame? Statistical machine translation tools mine large amounts of data, so it isn’t that the tools themselves are biased toward one gender or another, but the data they are mining often is. Our take? Tools must get smarter and more sophisticated in order to avoid gender-biased output. Otherwise, their use remains fairly limited – and has definite disadvantages when it comes to accuracy, as this example clearly shows.