Skip to main content

Language bootstraps cognitive complexity

Periodic Reporting for period 3 - LANGBOOT (Language bootstraps cognitive complexity)

Reporting period: 2020-06-01 to 2021-09-30

Language and cognition have a close, but cryptic, relationship. Is language just another tool in humans' diverse cognitive toolkit; important for communication, but not necessary for complex, high-level thought? Or is it language that allows us to form and manipulate complex thoughts in the first place? Distinguishing between these possibilities is vital to understanding our most fundamental cognitive faculties and the origin of modern human cognition itself.

The current project proposes that language bootstraps the cognitive complexity of the human mind by enhancing its ability to form and manipulate more elaborate mental representations than would otherwise be possible. In an innovative programme of investigation that uses cutting-edge methods from experimental psychology, psycholinguistics, cognitive modelling, and corpus linguistics, we examine how words interact with conceptual knowledge gleaned from perceptual and action experience across a range of fundamental cognitive tasks, including categorisation, memory performance, and abstract thought. We test whether and how language provides indispensable aid to cognitive processing that struggles to complete under time pressure or that strains working memory capacity, and how such aid could have influenced cognitive evolution.

Findings of this project aim to answer whether language provides critical enhancement to the achievable complexity of cognition, and whether language use could have brought about the sudden flowering of art, fine tools and culture that are the hallmarks of complex cognition in modern humans. Our objective is to develop a comprehensive, multidisciplinary perspective on the role of language in cognition that has the potential to reshape how we regard the functioning of the human mind.
In the first half of the project, we have laid the groundwork for the full programme of research. We have collected data from over 5,000 participants to establish the perceptual and action experience underlying the entire conceptual knowledge of an adult speaker of English. These data provide the largest such dataset created for English, and have enhanced our understanding of how different forms of perception and action experience are important to different kinds of concepts. For example, we discovered for the first time that interoception (sensations inside the body) is more important to abstract concepts like “serenity” and “genuine” than to concrete concepts like “chair”, and is particularly important to how we mentally represent emotions.

We have also analysed billions of words of English text in order to model the kind of word-to-word relationships that people pick up via language experience. For instance, words like “cat” and “kitten” often appear together in the same context, and even when they don’t appear together, they often share similar contexts concerning pets, whiskers, purring, and so on. By studying how words are distributed in text, we have been able to map out how well language can capture different forms of conceptual relationship, from simple relations like synonyms (e.g. big = large) to more complex relations like abstract-concrete categories (e.g. “cat” is a concrete concept but “relation” is not).

Using these results, we have running experiments and building computational models to investigate how people mentally represent concepts in different kinds of tasks, such as categorising or remembering objects. So far, our findings suggest that people use language to support their thinking even when there is no need to do so, which is helping us to develop a theory of how and when language is important to performing well in a particular task.
The project has made progress beyond the state of the art in a number of ways. Our work on the perception-action basis of concepts is groundbreaking, resulting in the largest set of semantic norms (i.e. data relating to the meaning of words) ever created for the English language, and we have made it freely available to other researchers. This work has led to a number of discoveries on how perception and action experience is critical to mentally representing all sorts of concepts, even those that are normally considered to be very abstract such as “serenity” and “justice”. In addition, we have carried out the largest study to date of how well different models of word distributions perform in different cognitive tasks. Our findings have resulted in a new framework to understand how language experience can capture the way concepts are related to one another, and how different kinds of language experience capture some relationships better than others.

For the remainder of the project, we aim to complete our series of experiments in order to gain a firmer understanding of how language and perception-action experience work together to help people mentally represent concepts under different circumstances. Simultaneously, we aim to combine perception-action information and language in computational models that we will use to test our theories of how important language is to performing well on a particular task, and how it might have been important in our evolutionary past.