in Language Technology, Technology

AI, or Artificial Intelligence, is the quickly increasing intelligence of machines. Though AI has often been associated with machines beginning to have human-like qualities, as is often shown in popular movies and TV shows, AI has also become an increasingly helpful resource to the language field, being used to help translation between languages on various websites and in real-life language barriers.  In fact, AI is becoming progressively aware of language and has begun to communicate in a more natural, human-like manner. In the past, users have experienced many issues with AI language systems, including an inaccurate “understanding” of words and phrases. Today, AI systems show marked improvement and have become a part of the core foundation for machine translation.

Out with the old and in with the new

AI systems have been slow to keep up with the increasing evolution of human language, often times resulting in “unnatural” translations when it comes to machine translation. These systems used to learn through Phrase-Based Machine Translation (PBMT), which is learning translation through analyzing human-translated documents and making a statistical decision on which translation version would be best.  Online translation systems, such as Google Translate, used to use frequently-translated material, like the Bible and European Union documents, to get the ‘best’ translation for you. This is why some words chosen in the translations may not be technically incorrect, but they may not be the most natural fit either. This past September, Google introduced the Google Neural Machine Translation (GNMT) to replace PBMT. Through this new process, Google claims that it has managed “to get its system close to human levels of translation by using bilingual people to train the system.”

Related:  The Challenges of Crowdsourcing Translation

Facebook also recently introduced a new system to help translate the information on their website. Facebook is using a new concept developed by researcher Yann LeCun called ‘convolutional neural networks’ that simultaneously analyze multiple parts of a sentence and organizes them into a more coherent sentence, as opposed to analyzing one part of a sentence at a time. The introduction of this technique could mean Facebook will be using less computing power when training translation models. Like Google, Facebook is making the code for this system freely available for other companies to use which allows companies to contribute to AI data and research and quickly increase the linguistic ‘data pool’. Because of data sharing, translation on the internet is growing and evolving at a much faster rate than ever before.

Growth of AI and language

Despite the recent rapid advancements to this field, the idea of AI was only created in the 1950s and online translation tools are only a little over 10 years old. Language is about 50,000-80,000 years old and is always changing; scientists have not yet been able to create a system that adapts as quickly as language evolves.  Because of this, it has taken much longer than expected for AI to become more commonplace than originally assumed.  Siri, Alexa, Cortana, and others will still face many problems communicating with the consumer until the day comes when scientists develop a technique that allows machines to learn and evolve language at the rate a human does. Though this day is inevitable, ruling out the importance of humans for translation still underestimates the importance humans have in establishing and improving strong AI to begin with.

Related:  The Fate of Language Diversity in a Neural MT World

Recently, scientists may have had a breakthrough in language AI. Scientists have been developing a computer that, when holding conversations with humans, starts with the language skills of a four-year-old child and continues to “learn”. This computer is called ANNABELL, or the Artificial Neural Network with Adaptive Behaviour Exploited for Language Learning, and has been developed to have conversations similar to a young child. So far, the system has learned to count, understands the concepts of friendship and age, and uses personal pronouns to describe ‘herself’. These scientists created the system specifically so that ANNABELL could learn in a more human-like way, through a series of neural networks. These neural networks activate neurons simultaneously, similar to synaptic plasticity, to allow the computer to learn at a faster rate.

With the rapid advancements of AI and translation currently underway, it is hard not to picture a world dominated solely by machine intelligence. One thing that all of the more successful AIs have in common is that they learn through human interaction. Without human translators providing a solid base for these machines, the progress for AI translation would not be as fast as it is currently. Language and translation are old and evolving concepts that prove quite a challenge for even the largest, well-known websites and online translation tools, such as Google and Facebook. Although the implementation of code-sharing for these websites has encouraged faster and stronger growth for these companies, only time will tell how all of society will be impacted by the advancements of artificial intelligence. In the meantime, us humans still hold the upper hand.

 

Write a Comment

Comment