‘Algorithm’ for metaphor development could teach computers figurative speech
Image credit: Dreamstime
A University of California (UC) Berkeley study into the cognitive processes unfolding to create and understand metaphors over centuries could eventually help artificial intelligence systems with “creating and comprehending metaphorical language”.
Metaphors are extremely difficult for computers to understand, especially those metaphors which have not yet become commonplace. Voice-activated digital assistants such as Alexa and Siri require commands with clear, literal phrasing.
“Although such systems are capable of understanding many words, they are often tripped up by creative uses of words that go beyond their existing, pre-programmed vocabularies,” said Dr Yang Xu, a UC Berkeley researcher in linguistics and cognitive science.
The researchers suggest, however, that by understanding the “algorithms” that humans naturally use to develop and understand metaphor, these artificial intelligence systems could eventually create their own.
A team of UC Berkeley and Lehigh University, PN, researchers set about unpicking the complex processes giving rise to new metaphors by mapping more than 5,000 examples across 1,100 years of English language, using the Metaphor Map of English database. More than 1,000 participants rated domains such as “water” according to the degree to which they were related to the external world, animate objects or emotions.
These ratings were used to create models to predict which domains had become the sources or targets of metaphors. The researchers found that their models were able to predict 75 per cent of recorded metaphorical language mappings over the past millennium.
They confirmed that words relating to tangible experiences, such as “grasping” (an object) were likely mapped to intangible concepts, such as “grasping” an idea. In particular, they found that words associated with digestion, textiles, water, hardness and plants were common sources for metaphorical extension, while emotional states were the most likely targets.
According to the researchers, their findings suggest that the creation of new metaphors is systematic.
“The use of concrete language to talk about abstract ideas may unlock mysteries about how we are able to communicate and conceptualise things we can never see or touch,” said Professor Mahesh Srinivasan, a UC Berkeley psychologist. “Our results may also pave the way for future advances in artificial intelligence.”
The “algorithm” of metaphor creation could inform the development of natural language processing systems such as Siri, in order to help them understand creativity in the use of language.
“This work brings opportunities toward modelling metaphorical words at a broad scale, ultimately allowing the construction of artificial intelligence systems that are capable of creating and comprehending metaphorical language,” said Dr Xu.