- Get link
- X
- Other Apps
In the myth about the tower of Babel, people decided to build a tower-city, which would reach to the skies. And then the Creator realized that nothing more would restrain people and they think they know something about themselves. Then God created different languages to stop people and that they could no longer work together easily. In our time, thanks to technology, we feel an unprecedented connection. However, we still live in the shadow of the Tower of Babel. Language remains a barrier in business and marketing. Despite the fact that technological devices can easily and quickly connect, people from different parts of the world often can not.
Translation bureaus try to keep up: they make presentations, contracts, instructions on outsourcing and advertisements for all comers. Some agencies also offer so-called "localization". For example, if a company enters the market in Quebec, it needs advertising in Quebec French, and not in European French. Companies can be seriously affected by incorrect translation.
Global markets are waiting, but language translation by artificial intelligence is not yet ready, despite recent advances in the field of natural language processing and mood analysis. AI still has difficulties with processing queries even in one language, not to mention the translation. In November 2016, Google added a neural network to its translator. But some of her translations are still socially and grammatically strange. Why?
"To the credit of Google, the company introduced quite a few improvements that appeared almost overnight. But I'm not really using them. Language is difficult, "says Michael Hausman, Principal Research Scientist at RapportBoost.AI and a teacher at Singularity University.
He explains that the ideal scenario for machine learning and artificial intelligence will be the fixed rules and clear criteria for success or failure. Chess is an obvious example, and with them and go. The computer very quickly mastered these games, because their rules are clear and precise, and the set of moves is limited.
"The language is almost exactly the opposite. There are no clear and verified rules. Conversation can go in an endless number of different directions. And, of course, you also need labeled data. You need to tell the machine what it does right and what does not. "
Hausman noted that it is fundamentally difficult to label information labels in a language. "Two translators can not agree on the correctness of the translation," he says. "Language is the Wild West in terms of data."
Google's technologies are now able to understand the whole sentence, without trying to translate individual words. But glitches still happen. Jorg Mayfood, associate professor of the Spanish language department, a specialist in Latin literature at the University of Jacksonville explains why accurate translations are not yet given to artificial intelligence:
"The problem is that understanding the proposal as a whole is not enough. Just as the meaning of a single word depends on the rest of the sentence (mostly in English), the meaning of the sentence depends on the rest of the paragraph and the text as a whole, and the meaning of the text depends on the culture, intentions of the speaker and others. Sarcasm and irony, for example, make sense only in a broad context. Idioms can also be problematic for automated translation. "
"Google translation is a great tool if you use it as a tool, that is not trying to replace human learning or understanding," he says. "A few months ago I went to buy a drill in Home Depot and read the inscription under the car:" Saw machine ". (Machine saw). Below was the Spanish translation of 'La máquina vió,' which means "The machine saw it". "Saw" was translated not as a noun, but as a verb of the past tense. "
Dr. Mayfood cautions: "We must be aware of the fragility of this interpretation. Because translating is essentially an interpretation, not just an idea, but a feeling. Human feelings and ideas that only people can understand - and sometimes even we, people, can not understand other people. "
He noted that culture, gender and even age can create obstacles to this understanding, and excessive dependence on technology leads to our cultural and political decline. Dr. Maifud mentioned that the Argentine writer Julio Cortázar called the dictionaries "cemeteries." Automatic translators could be called a "zombie".
Eric Cambria, an academician who researches AI, and a professor at Nanyang University of Technology in Singapore, deals mostly with the processing of the natural language that underlies the interpreters based on AI. Like Dr. Mayfood, he sees the complexity and the associated risks in this direction. "There are so many things that we do unconsciously when we read the text." Reading requires the execution of many unrelated tasks that are beyond the power of automatic translators.
"The biggest problem of machine translation for today is that we tend to move from the syntactic form of the sentence in the input language to the syntactic form of this sentence in the target language. We, people, do not do that. We first decode the value of the sentence in the input language, and then we encode this value in the target language. "
In addition, there are cultural risks associated with these transfers. Dr. Ramesh Srinivasan, director of the Digital Culture Laboratory at the University of California, Los Angeles, says that new technological tools sometimes reflect underlying prejudices.
"There must be two parameters that determine how we design" intelligent systems. " One is the values and, so to speak, the prejudices of the one who creates the systems. The second is the world in which the system will learn. If you create AI systems that reflect the prejudices of your creator and the wider world, sometimes there are very impressive failures. "
Dr. Srivanisan says that the translation tools should be transparent in terms of opportunities and limitations. "You see, the idea that one system can take languages (which are very diverse semantically and syntactically) and unite them, or to some extent generalize, or even make them whole, is ridiculous."
Mary Cochran, cofounder Launching Labs Marketing, sees commercial growth potential. She noted that lists in online markets like Amazon can in theory automatically translate and optimize for buyers in other countries.
"I believe that we have now touched only the tip of the iceberg, so to speak, about what AI can do with marketing. And with improved translation and globalization around the world, AI can not but lead to explosive growth of the market. "
The article is based on materials .
Comments
Post a Comment