Machines learning human languages: How Computers Are Catching Up With Humans in Learning Languages
By Samantha Wickramasinghe
Charles Darwin, in his book Decent of Man, wrote that human beings have an innate tendency to speak words other than doing things like baking, brewing or writing. This is why when children are born, speaking a language or uttering words become one of the first things they do .
This primary tendency is not an accident and it shows the importance of language as a vital instrument for the survival of the species. This also means that the early social, cognitive and psychological developments of children go hand in hand with learning a language (the mother tongue). Nevertheless, many of us do not stop by learning one language. Apart from learning our mother tongue, we go on to learn other languages as well. But as a human we are not the only ones who are learning languages these days. Machines and computers are also learning human languages in their own ways.
This new development of machines learning human languages should not be confused with programming languages. In order to programme machines by given commands, software engineers and computer scientists use programming languages such as Python and C. But the machines themselves learn human languages in deferent ways. One of the ways that machines learn languages is by analysing parsers, which is, broken-down information that are learning-friendly for the computers.
With the help of technologies like the ASR (Automatic Speech Recognition Technology) and the TTS ( Text to Speech) technology we are able to get a glimpse of how machines are processing human languages, in our day to lives. For example, take a look at your phone.
When you are using your voice on the phone to do a Google search, your voice gets converted to text and the text appears in the search bar. Finally, the text becomes the search item on the screen and the search results appear eventually. The developments of these technologies are quite promising although, there are shortcomings. Still these technologies need supervised human assistance and their work can be time-consuming and cumbersome.
Another way that machines are learning human languages is through Artificial Intelligence (AI) that mimic human learning behaviour. In 2018, the world-famous engineering and technological university, Massachusetts Institute of Technology (MIT) announced that it has developed an artificial intelligent programme that learns languages similarly to a child.
This programme, also called as a ‘semantic parser’, that learns language through observation, and it mimics the children’s language acquisition process. According to MIT, this technology does not need a lot of ‘training data’ which means that the technology works in an efficient way. In fact, the way it works is quite interesting to think about. First, the AI observes videos that have captions, which means that the videos could have a dialogue between two people in a drama.
But the AI does not process the video or audio, it only looks at the caption, or the subtitles of the videos. After observing captions, then the AI, can predict what is meant by the data collected through the captions. According to MIT, this method allows AI to learn language in a natural way, like children which means that the AI can become much more conversational-friendly than it already is as we find in devices like Amazon Alexa and Google Speaker.
One of the challenges that scientists and linguists have is teaching machines the grammatical structures or languages which do not always mimic real life conversations. People do not always speak grammatically correct phrases or sentences. Sometimes, people don’t always talk to each other in full sentences and they used idioms and jargon. Through the AI designed by the MIT, these problems could be addressed, which could eventually lead to much more ‘natural’ conversations between human beings and machines. If this process is too difficult to understand, think about like this. It is like computers are now able to connect between captions and videos. By observing caption, they can accurately predict what is in the videos. This ability to make connections, must look simple in human terms. As human beings, we take it for granted how children make connections with pictures and sentences. What we forget is that it has taken so many generations may be millions of years to make such progress for human beings although it comes naturally to us. What machines are doing with AI is somewhat similar. Perhaps, the difference is, machines are learning things in the speed of light. But for us humans, it is almost like looking back on our own evolution as a species manifested by the learning of machines.