If your answer is “they're round objects”, you have just performed an act of abstraction.

You identified something that three very different objects share, ignored their many differences, and created a broader category based on a common feature: roundness.

In doing so, you climbed one step up the ladder of abstraction. You could climb even higher, moving from round objects to geometric shapes. Or you could move down the ladder, from pizza to Margherita pizza, or from wheel to bicycle wheel.

Our ability to abstract allows us to move from individual experiences to broader ideas, recognize patterns, learn new things, communicate with others, and create something entirely new. And one of the most powerful tools supporting this ability is human language.

Words help us organize the world into categories. Some refer to things we can see, touch, or experience directly, such as cat or table. Others refer to abstract ideas, such as freedom, empathy, or justice.

Some words describe broad categories, such as vehicle or art, while others identify more specific ones, such as tandem bicycle or Impressionism.

These examples illustrate two fundamental dimensions of meaning: concreteness, which concerns how closely a concept is connected to our sensory experiences, and specificity, which concerns how general or detailed a category is.

Although these dimensions are related, they are not the same. A concept can be concrete but general, like furniture, or abstract but highly specific, like freedom of speech.

Understanding how these two dimensions work together is at the heart of the ERC-funded project ABSTRACTION – Unlocking Meaning from Experience through Language.

What have we discovered?

One of the project's main contributions is showing that concreteness and specificity play different but complementary roles in how we think, communicate, and create meaning.

For a long time, researchers studying how language connects to human experience have mainly focused on whether words are concrete or abstract. Other researchers, particularly in linguistics and computer science, have investigated how concepts are organized into general and specific categories.

ABSTRACTION brings these perspectives together.

Our research has shown that specificity influences how quickly we process words, independently of their concreteness. More specific words can be easier to recognize and categorize than more general ones. We have also found that abstract concepts are not necessarily vague or generic: just like concrete concepts, they can be organized at different levels of specificity.

We have investigated how these abilities develop in children, and how people learn to move between specific examples and broader categories.

But abstraction is not only something that happens inside our minds. It also shapes the way we communicate.

Imagine explaining a virus to a child and to a medical specialist. You might tell the child that a virus is a tiny bug that can make people sick, while describing its biological structure in much greater detail to the specialist. Both explanations refer to the same phenomenon, but they differ in their concreteness and specificity.

Our experiments with children have shown that texts containing more concrete words are generally perceived as clearer, while more specific words contribute to how informative a text seems. We have also found that discussing abstract concepts requires greater effort to establish a shared understanding between speakers.

In other words, successful communication depends not only on what we say, but also on how we adjust the level of abstraction and detail to the people we are talking to.

Abstraction is equally important for creativity.

Think of a metaphor such as “time is money”. We understand an abstract concept, time, through something more familiar from everyday experience. Or consider how we interpret a new combination of words that we have never encountered before: we must connect existing concepts and work out a plausible relationship between them.

Our research on metaphors and novel word combinations shows that the concreteness and specificity of concepts influence how familiar, related, or unexpected creative expressions appear. It also reveals how challenging it can be, even for advanced artificial intelligence systems, to transfer familiar patterns of meaning to entirely new combinations.

These three interconnected areas — Thought, Language, and Creativity — form the core of ABSTRACTION.

Playing with words to understand meaning

To investigate abstraction, we also needed new ways to measure how people organize concepts.

While researchers already had extensive information about how concrete or abstract words are perceived, much less was known about their specificity.

We therefore developed new linguistic resources and experimental tools, collecting judgments from Italian and English speakers about how specific different words are.

One of our most innovative tools is Word Ladders, a language-based game that turns the process of abstraction into a playful activity.

Imagine starting with the word poodle. You might continue with dog, then animal, and finally living being. With every step, you move towards a more general category.

By collecting thousands of these connections, we can investigate how people organize knowledge and how conceptual categories relate to one another.

Our research has also shown that gamification can be a valuable way to collect linguistic data, making participation more engaging while opening new possibilities for studying meaning.

Can artificial intelligence abstract like humans?

Today, systems such as ChatGPT can produce remarkably fluent explanations, recognize categories, and generate generalizations. But does producing language that sounds meaningful imply organizing knowledge in the same way humans do?

This is another question explored by ABSTRACTION.

We have compared how humans and large language models organize conceptual categories, interpret general statements, and move between specific and general meanings.

For example, if we ask people to name a more specific type of bird, they might answer sparrow, eagle, or penguin. An artificial intelligence system can generate similar answers. But does this mean that its underlying organization of categories corresponds to ours?

Our findings suggest that the answer is not always yes. Language models can produce plausible associations while differing from humans in how they organize specific and general concepts. Their performance also varies depending on the types of concepts and relationships involved.

Together with researchers from cognitive science and artificial intelligence, we have investigated what it really means for humans and machines to generalize, highlighting both common challenges and important differences.

Our research also raises questions about the social perspectives represented in AI-generated language. When language models are asked to speak from the perspective of people with different social identities, they may reproduce stereotypes rather than authentically represent the diversity of human experiences.

These findings remind us that generating convincing language is not necessarily the same as sharing human understanding, knowledge, or experience.

Why does abstraction matter?

Abstraction is everywhere.

We use it when we recognize a new animal as a member of a familiar category, when we explain a complex idea to a child, when we understand a metaphor, or when we interpret an unfamiliar expression.

We also rely on abstraction when we disagree about concepts such as justice or freedom, when we decide how much detail to include in an explanation, and when we try to understand what artificial intelligence systems actually know.

The findings of ABSTRACTION suggest that meaning emerges from the interplay between our experiences, the categories we construct, and the language we use to communicate them.

Concreteness helps connect our concepts to perceptual experience. Specificity allows us to organize those concepts at different levels of generality and adjust the information we communicate to different people and situations. Creativity allows us to recombine and reinterpret familiar concepts in novel ways.

Understanding abstraction, therefore, means understanding how we move from experience to meaning, from individual examples to general ideas, and from our own thoughts to the shared knowledge we construct through language.