Abstraction in CREATIVITY

How do we create new meanings by combining familiar words and concepts in unfamiliar ways? What makes a metaphor effective, surprising, or funny? And can artificial intelligence understand creative expressions in the same way humans do?

Metaphors are one of the most familiar forms of linguistic creativity. We use them every day to explain difficult ideas, express emotions, and describe experiences in unexpected ways.

For example, during the Covid-19 pandemic, the virus was often described as an invisible enemy that we had to fight. Of course, no actual soldiers or battles were involved. Instead, familiar concepts from warfare were used to make a complex and invisible threat easier to understand.

Research has long investigated how the concreteness of words influences metaphor comprehension. But what about specificity? Does it matter whether a metaphor uses broad, general concepts or highly detailed ones?

Consider a wonderfully unusual description by food critic A. A. Gill, who compared a dish served in an Egyptian restaurant to:

“belly buttons in a soup of fake tan”

The metaphor is vivid, unexpected, and humorous. Imagine replacing it with the much more generic description “round things in a liquid”. The basic idea might remain, but much of the humour and vividness would disappear.

On the other hand, when a metaphor is intended to explain something unfamiliar, a more general expression may sometimes work better. Describing a virus to a child as a tiny bug, for instance, may be more helpful than comparing it to a highly specific type of insect the child has never encountered.

This suggests that the effectiveness of a metaphor may depend not only on how concrete its words are, but also on how specific they are, and on what the metaphor is intended to achieve.

Through the ABSTRACTION project, we have investigated this question by asking people to evaluate metaphorical and literal expressions such as:

  • Alice fixed her shirt with care (general, literal)

  • Alice mended her shirt with care (specific, literal)

  • Alice fixed her relation with John (general, metaphoric)

  • Alice mended her relation with John (specific, metaphoric)

Our findings indicate that within the metaphoric expressions, those containing more specific words tend to be perceived as less familiar, more difficult and more metaphorical than expressions where the verb is more general. A similar pattern is observed within the literal sentences, where specificity lowers comprehensibility, but in our data there was a variable (frequency) that was hard to disentangle from specificity.

Overall, we interpreted these findings suggesting that specificity can make a metaphor feel more unusual or unexpected. However, this does not automatically make it better: a surprising metaphor might be particularly effective for humour or artistic expression, while a more familiar one might be preferable when the goal is to explain a difficult concept.

But linguistic creativity is not limited to metaphors. We also create new meanings simply by combining words.

Consider the expression carrot cake. We immediately understand that it refers to a cake made with carrots. But what about birthday cake? In this case we are talking about a cake made for a birthday, not made with birthdays. Based on these examples we are likely to understand the intended meaning of expressions that may sound new to us, like eggplant cake (likely: made with eggplants) and graduation cake (likely: made for graduation).

Understanding these combinations requires more than knowing what individual words mean. We must also work out how the concepts are connected, often by drawing analogies with expressions we already know.

Can artificial intelligence do this?

We investigated whether large language models, the technology behind systems such as ChatGPT, can interpret familiar combinations of nouns and extend their interpretations to new ones.

Our research revealed an interesting pattern. The models were better at identifying some relationships than others, and generally performed better when the combinations involved concrete concepts. However, they encountered substantial difficulties when asked to transfer familiar patterns to novel combinations.

This suggests that recognizing familiar expressions and creatively extending their meanings are not necessarily the same ability. Even systems trained on enormous amounts of language may struggle when they need to apply a known relationship to an unfamiliar combination.

Another way to investigate how people use and transform language is to observe what happens when they play with words.

In the ABSTRACTION project, we have collected linguistic data through gamified activities, where participants interact with words in playful settings. These activities allow us to observe not only conventional language use, but also unexpected forms, including unusual spellings.

For example, someone might write definately instead of definitely, or produce a non-standard spelling while responding quickly during a game. Such forms can emerge for different reasons, from typing mistakes to uncertainty about spelling conventions. Our research has examined how patterns of misspellings in gamified data vary in relation to participants' sociodemographic characteristics.

Importantly, not every unconventional word form is a creative invention. Some are simply errors, while others may reflect deliberate experimentation with language. Distinguishing between these possibilities helps us understand what linguistic data collected through games can, and cannot, tell us about creativity.

What do these findings tell us about the relationship between abstraction and creativity?

They suggest that linguistic creativity involves more than inventing entirely new words or expressions. It often relies on our ability to recognize familiar patterns, select relevant features of concepts, and apply existing knowledge in unfamiliar situations.

Concreteness and specificity contribute to this process in different ways. Concreteness connects creative expressions to recognizable experiences and images, while specificity influences how familiar, unexpected, or semantically related different combinations of concepts appear.

Ultimately, creativity involves moving flexibly between what we already know and what we have never encountered before: using familiar experiences to imagine new possibilities, and adjusting the level of detail at which concepts are represented and combined to produce new meanings.

Related publications