“Researchers used machine learning to create the first large-scale, data-driven study to illuminate how culture affects the meanings of words.”
Human language does great injustice in representing the mysterious capabilities of the brain. Nevertheless, it serves as an inexpensive yardstick to measure thoughts and intelligence. If this is remotely true and if we are serious about our pursuits of AGI then it is almost inevitable to end up with natural language processing. But, how well have we understood the language itself?
In a new paper published in Nature, Willian Thompson of Princeton University and his colleagues harnessed machine learning to analyse over 1,000 words in 41 languages.
The researchers investigated the meaning of words in relation to culture, history and geography. They found that this is even true for some concepts such as emotions, landscape features and body parts.
Semantics Of Beauty
“The way we interpret the world through words is part of our culture inheritance.”
Linguists and anthropologists have been trying to decipher the way we communicate. However, conducting these studies is time taking. Anthropologists have to sit down with people from different cultures, record interviews and then weave their findings, if any, with the help of linguists. The process is tedious, to say the least.
Modern-day computers are well equipped to transcribe and accurately translate many universal languages. So, the researchers at Princeton developed a machine learning algorithm and leveraged its high dimensionality to find the link between culture and language.
Talking about the traditional process, one of the authors said that it might take years to document a specific pair of languages and their differences. Whereas, machine learning models offer a new level of precision to process this vast information.
As illustrated above, the algorithm translated the semantic associates of a particular word. For example, the semantic neighbours of the word “beautiful” were translated into French, and then those of “beau” were translated into English. We can see that the respective lists were substantially different because of different cultural associations of the French and the English.
The researchers have also applied another algorithm to compare the similarity of cultures based on language. For this experiment, they have used an anthropological dataset comparing things like marriage practices, legal systems and political organisation of given language’s speakers.
The datasets that were fed into these models were created by 20th century anthropologists as well as more recent linguistic and psychological studies. According to the researchers, the results from their experiments indicates that the meanings of common words reflect the cultural, historical and geographical aspect of their users.
Key Findings
- Universally translatable word types that referred to animals, food and emotions, were much less well-matched in meaning.
- Algorithms could correctly predict how easily two languages could be translated based on how similar the two cultures that speak them are.
Alexa, Think In Chinese
“If we accept that a book has no mind of its own, we cannot then endow a computer with intelligence and remain consistent.”
John Searle
In the infamous Chinese room experiment, John Searle, an American philosopher, questioned the concept of intelligence in the context of computers. When can we attribute intelligence to machines? Searle’s Chinese room conundrum is a thought experiment where a non-Chinese speaker hidden on one side of the room fools a native Chinese speaker into thinking that messenger is of Chinese origin. The non-Chinese speaker communicates just by following a bunch of Mandarin instructions. He is not Chinese. So, Searle asked, can we, on similar lines, call a computer to be intelligent when it is just following a bunch of instructions?
Be it the Turing test or the Chinese room experiment, the intelligence in machines had been assessed through language. Think: chatbots. And what constitutes a language? Demography, cultural cues and other intricacies. The overlapping of culture, language and the measure of intelligence does make this grand experiment by Princeton University researchers a serious contender for the next AI revolution. Or, at least, the Alexas of the future shall behave in French, Indian and Japanese!
The post Alexa, Think In French: ML Finds A Crucial Link Between Culture And Language appeared first on Analytics India Magazine.