TALKING POINT: NEURAL MACHINE TECHNOLOGY
“How can museums join the Post-Language Economy?”
A major road block in implementing multilingual exhibitions is the high cost of professional translation services. In order to keep up with the demands of a global economy, major corporations are able to invest into setting up international offices, translating their websites, and releasing packaging to fit different cultural sensibilities. It takes an enormous amount of resources to retain an international audience - something that non-profit, overextended museums rarely have room for in the budget. A common complaint is that professional translation services are expensive* and waiting for the final drafts of bilingual signage can confuse the flow of exhibition design process. A promising advancement in machine translation technology could change this imbalance in our lifetime and open up a post-language economy.
Neural Machine Translation (NMT) is “an end-to-end learning approach for automated translation, with the potential to overcome many of the weaknesses of conventional phrase-based translation systems.” To put it simply, NMT translates entire phrases instead of individual words, thus improving understanding of the meaning of the translation. It is an end-to-end learning system so it picks up on language patterns through continued use. It is aimed at translating massive amounts of text for businesses or government agencies. There has been a lot of NMT “hype” in the technology industry recently, which prompted machine translation researcher Phillip Koehn to caution patience with optimism: “while the technology is promising we still have some way to go to commercial implementations that can rival Rule-Based Machine Translation (RBMT) and Statistical Machine Translation (SMT) in all use cases.” Another set back to implementing this new evolution in machine intelligence is that it is costly and time-consuming to train.
Even so, the past few years saw an impressive jump in improvements to the technology. Google’s NMT project (GNMT) announced in September 2016 that their own service “reduces translation errors by an average of 60% compared to Google's phrase-based production system.” GNMT’s team is also cutting down the computational time it takes to train this language and recognize rare words. Between Google and Systran’s NMT systems, this revolutionary technology now covers 130 language pairs. Facebook has started implementing NMT services in their infrastructure and the European Union has invested in the technology in the hopes of creating a more equitable linguistic field for their members.
In terms of museum design, if machine translation services were available to curatorial and design teams from the beginning, it would go a long way to improving the speed of the notoriously slow-moving process of producing a bilingual exhibition. Designers would have a working bilingual draft in minutes to start making typography choices and measuring space on the wall signs. Curatorial teams could visualize every project as linguistically accessible if the added pressure of coordinating with a translation team was eliminated or simplified. Museum websites could be offered in different languages with less concern for the language limitations of Google Translate. While I believe that a machine can never replace the superior understanding of human translators, paying humans to edit translations for clarity rather than translate from scratch would open up the ability to host bilingual exhibits to even small gallery spaces.
While NMT is useful for replacing the traditional process of translating physical exhibitions, it is even better equipped to use with digital innovations such as RFID technology. I was inspired by the tour of the new U.S. Olympics and Paralympics Museum in Colorado Springs that successfully paired RFID screens into their museum experience to create greater accessibility for all visitors. When you first walk in you are directed to screens where you can register your RFID name tags that you wear around the museum. The name tag will record your accessibility customizations and adjust the exhibition screens throughout the museum. For example, a person in a wheelchair could ask for the label text to appear lower, an older visitor could ask for larger font size, and an international tourist could change the language of the text. The museum experience could be different for every visitor according to their individual needs, but still equitable. If you could integrate NMT into this accessibility set up, it could be easily edited and the translation would continue to improve as the machine intelligence learns more.
There is a long list of concerns that stop museums from taking a leap to produce bilingual exhibitions and resources, and in my experience, the expense attached to translation process always rises to the top. NMT innovations could open up a new avenue to linguistic accessibility in the post-language economy.
Neural Machine Translation (NMT) is “an end-to-end learning approach for automated translation, with the potential to overcome many of the weaknesses of conventional phrase-based translation systems.” To put it simply, NMT translates entire phrases instead of individual words, thus improving understanding of the meaning of the translation. It is an end-to-end learning system so it picks up on language patterns through continued use. It is aimed at translating massive amounts of text for businesses or government agencies. There has been a lot of NMT “hype” in the technology industry recently, which prompted machine translation researcher Phillip Koehn to caution patience with optimism: “while the technology is promising we still have some way to go to commercial implementations that can rival Rule-Based Machine Translation (RBMT) and Statistical Machine Translation (SMT) in all use cases.” Another set back to implementing this new evolution in machine intelligence is that it is costly and time-consuming to train.
Even so, the past few years saw an impressive jump in improvements to the technology. Google’s NMT project (GNMT) announced in September 2016 that their own service “reduces translation errors by an average of 60% compared to Google's phrase-based production system.” GNMT’s team is also cutting down the computational time it takes to train this language and recognize rare words. Between Google and Systran’s NMT systems, this revolutionary technology now covers 130 language pairs. Facebook has started implementing NMT services in their infrastructure and the European Union has invested in the technology in the hopes of creating a more equitable linguistic field for their members.
In terms of museum design, if machine translation services were available to curatorial and design teams from the beginning, it would go a long way to improving the speed of the notoriously slow-moving process of producing a bilingual exhibition. Designers would have a working bilingual draft in minutes to start making typography choices and measuring space on the wall signs. Curatorial teams could visualize every project as linguistically accessible if the added pressure of coordinating with a translation team was eliminated or simplified. Museum websites could be offered in different languages with less concern for the language limitations of Google Translate. While I believe that a machine can never replace the superior understanding of human translators, paying humans to edit translations for clarity rather than translate from scratch would open up the ability to host bilingual exhibits to even small gallery spaces.
While NMT is useful for replacing the traditional process of translating physical exhibitions, it is even better equipped to use with digital innovations such as RFID technology. I was inspired by the tour of the new U.S. Olympics and Paralympics Museum in Colorado Springs that successfully paired RFID screens into their museum experience to create greater accessibility for all visitors. When you first walk in you are directed to screens where you can register your RFID name tags that you wear around the museum. The name tag will record your accessibility customizations and adjust the exhibition screens throughout the museum. For example, a person in a wheelchair could ask for the label text to appear lower, an older visitor could ask for larger font size, and an international tourist could change the language of the text. The museum experience could be different for every visitor according to their individual needs, but still equitable. If you could integrate NMT into this accessibility set up, it could be easily edited and the translation would continue to improve as the machine intelligence learns more.
There is a long list of concerns that stop museums from taking a leap to produce bilingual exhibitions and resources, and in my experience, the expense attached to translation process always rises to the top. NMT innovations could open up a new avenue to linguistic accessibility in the post-language economy.