As a French minor who studied in Toulouse, France, I was excited that this talk in AI was hosted by Amélie Josselin-Leray from University of Toulouse Jean Jaurés, where I studied. In addition to working as a professor, she trains AI.
Immediately, she clarified that she will not be answering the question whether or not AI will be replacing humans soon. Instead, Josselin-Leray discussed how and why we need better AI technology when it comes to translation. She advocated for both sides of AI; its usefulness and its threat.
Josselin-Leray stressed the need for humans to be part of AI training. She used the deaf community as an example of an already marginalized group, who is concerned they will further be excluded, because of AI.
On the other hand, AI translation has been greatly improving, but there is a downward trend in the marketing of AI translation as of last month. However, this study does not include machine translation that students often use. Josselin-Leray defines machine translation as “any kind of tool that automatically translates from one language to another,” which is different from computer programs such as Google Translate.
Although I do not want to be a translator, I have friends that do, and I still use machine translation to communicate with French friends I made abroad. Josselin-Leray explained that translation is a complex job that mainly deals with meaning. She said that machines can not interpret meaning. Translators must also convey “style, tone and intent of the text,” while keeping in mind the culture and context of the content.
The mid-ground between human and AI translation is computer-aided translation (CAT). Subtitling is one job someone working in CAT could have. Machine translation has been around since the 1940s and 1950s. Rule-based machine translation (RBMT) emerged in the 1950s and continued into the 1990s. RBMT was grounded in bilingual dictionaries and grammar rules, with the end product being 49 sentences. Data-driven MT, based in numbers and patterns, was the new approach in the 1990s. Obviously, there were limitations to both models, because translation involves much more than language, such as real world knowledge.
“A translator cannot do without translation technologies,” said Josselin-Lera. This quote means translators need word processors, spell check, machine translation, etc. to help their job. No one person is expected to know all the grammar rules and spelling of a language, even their native language.
There are 24 languages spoken in the European Union (EU). Josselin-Leray explained that each speaker at the EU has the right to speak and be spoken to in their native language. Even though most people in Europe speak at least two languages, in my experience, representatives of countries need translators, who need AI tools. There are certainly negative opinions of AI, but Josselin-Leray assured students that AI will not steal their jobs, but we need to adapt and change how we use and train AI.
