“Humanizing Machine Translation” (Anthony Perez)

11/23/2016 4:55 PM | Anonymous

Anthony Perez, “Humanizing Machine Translation”

By Abigail Wright

 

Anthony Perez, Vice President of Global Sales at Translate.com, gave a presentation at MATI 13 titled “Humanizing Machine Translation,” dedicated to clearing up misconceptions concerning Machine Translation (MT) and shedding light on how we human translators can make it work for us. Perez affirmed that while MT is here to stay, clients still want a human face, because “people buy from people.” While some translation jobs have been lost to MT, on the whole, demand for human translation has actually increased with the winds of rapid technological change.

 

Perez dubbed 2016 the “Era of Mass Translation,” explaining that in any given minute, millions of posts appear online and billions of messages are sent across the globe. The majority of users of such top Internet properties such as Google, Microsoft, Facebook, and Yahoo! are international. Smartphone use in Asia has skyrocketed. This has all come together to create a global “on-demand economy.” In a world of Snapchat, GrubHub, and Uber, people want translations instantly—just like everything else.

 

In accordance with this demand, Perez explained, MT has evolved from a rules-based approach that generated translations based only on word-to-word matches and naturally required heavy human correction to its present statistics-based model. Current MT works with phrases instead of words, taking its cue from bilingual glossaries, translation memories, and feedback from its users. The future of MT, however, lies in a deep-learning, neural model based on the human brain itself. This model, which is still in development, relies on a main engine which processes an entire sentence, paragraph, or document, while a subnetwork processes source sentences, keywords, grammar, and word meaning.

 

The goal, of course, is to create the best possible raw MT output. Post-editors use their knowledge base for everything from correcting minor punctuation and capitalization errors to retranslating whole words and expressions. The Translation Automation User Society (TAUS) advises post-editors to use all the raw MT they can while aiming for a semantically correct text, never adding or omitting anything, editing offensive or inappropriate content, and leaving their clients with no stylistic worries. Human translation is still necessary, Perez acknowledged, particularly for advertising, legal texts, contracts, marketing materials, and human resources documents.

 

Perez closed out his presentation with a Wizard of Oz analogy and several pieces of advice for human translators in the age of MT. We can choose, he explained, to be the Scarecrow, carrying on as usual; the Cowardly Lion, skeptically (and hopelessly) wishing for MT to die; or Dorothy, pursuing opportunities and learning from the journey. To help make Dorothys out of us, Perez advised the following courses of action: move fast, create new business models around MT, be open to learning new platforms, and give feedback on MT quality to language service providers to help improve MT engines. We should also invest in networking, find new ways to make money, and above all, laugh, because “life is short.”

 

Abigail Wright is a freelance Spanish-to-English translator. She is a 2016 graduate of the Master of Arts in Translation and Interpreting program at the University of Illinois at Urbana-Champaign and recently started her own company, Wright Translations, LLC. She has been a MATI member since 2015.