How is meaning represented and processed in the brain? To address this fundamental question, we develop cognitive models of conceptual knowledge, based on semantic feature statistics, and use these models to investigate how conceptual knowledge is processed. Through combining cognitive theory and neurobiological models, our research endeavour reflects a neurocognitive approach. Our current research builds upon the hierarchical model of object processing in the ventral stream where increasingly anterior regions represent increasingly complex information about objects. While activity in the posterior fusiform is sufficient to support coarse semantic representations, anteromedial aspects of the stream are required for the integration of more fine-grained conceptual knowledge supporting subtle distinctions between similar objects.
The evolution of meaningful object knowledge
Recognising and understanding what visual objects are is a critical ability if we are to relate and interact with the world, and requires the ability to extract meaningful semantic information from our visual perceptions. While it's generally appreciated that our knowledge about visual objects progresses along a course-to-fine time-course, exactly how the brain transforms this visual information into a more abstract, meaningful representations is unclear. Our research aims to uncover how such meaning evolves from vision over time as neural signals propagate and reverberate through the cortex in a dynamic and interactive manner. To address these issues we utilise a variety of behavioural and neuroimaging techniques - including MEG and fMRI, coupled with multivariate statistical analyses and connectivity measures.
Recurrent interactions between the left anterior temporal and posterior fusiform increase when more specific semantic information is required. This is shown through increased phase-locking between these regions during basic (e.g. tiger) compared to domain naming (i.e. living or nonliving; left), and increased activity in the anterior temporal peaking ~200 ms and posterior fusiform peaking ~250 ms (right). Adapted from Clarke, Taylor & Tyler (2011) Journal of Cognitive Neuroscience, 23(8), 1887-1899.
Processing of objects and categories
We tested our hypothesis that conceptual knowledge is processed hierarchically in the ventral stream by using fMRI to measure brain activity during picture naming. We then investigated where and how brain activity related to the properties of the pictures as defined using feature statistics based on property norms. For example, the extent to which an object’s features are shared with other objects can be used to process general information about the object, whereas to identify the object uniquely, its distinctive features must be processed in combination with its shared features.
We found that objects with many shared features (typical of living things) were associated with activity in the lateral fusiform gyri whereas objects with fewer shared features (typical of nonliving things) were associated with activity in the medial fusiform gyri. Significantly, a feature statistic reflecting differentiation between highly similar objects, enabling object-specific representations, was associated with activity in the bilateral perirhinal cortex, at the apex of our hypothesized ventral stream hierarchy.
These findings unified findings from fMRI in healthy individuals and studies of lesions in brain damaged patients. Although fMRI studies show that different regions in the fusiform gyri respond to different object categories, patients with category-specific deficits (especially for living things) typically have damage in the perirhinal cortex. Our study suggests that the fMRI results reflect earlier, category-level processing in fusiform regions, whereas category-specific deficits following perirhinal damage reflect impaired ability to differentiate highly similar objects, such as animals.
Results from the fMRI picture naming experiment. Top: regions in the fusiform gyri showing contrasting responses to items with more shared features (lateral fusiform, orange) and fewer shared features (medial fusiform, blue). Bottom: bilateral regions, including the perirhinal cortex, showed greater activity to concepts whose distinctive features were more weakly correlated, i.e. concepts that were harder to differentiate.
Conceptual knowledge as measured by property norms
To facilitate the development of models of conceptual knowledge, we have released a new set of property norms from which we can calculate various statistics relevant to conceptual representation. In particular, we are interested in the relationship between features of concepts that occur rarely or often and the co-occurrence of features. We believe that these norms are the most extensive collected with British English participants. The norms are also unique in that they list all the features generated by at least 2 participants (out of 30 participants per concept) as well as showing all the variants that make up a norm feature (e.g.,the feature is big may have variants such as is large, is huge, etc). By publishing this additional information, we believe that the norms will have wide applicability across of variety of academic disciplines, such as cognitive psychology, computational linguistics and computer science. We also make available semantic similarity scores for every pair of concepts (cosines between feature vectors).
The norms show a high degree of comparability with previous norms (e.g. McRae et al, 2005), and show clear category structure:
Perceptual and conceptual processing in brain-damaged populations
This research project focuses on relating behavioural measures of perceptual and conceptual processes to patterns of neural integrity in brain-damaged populations, with a particular focus on the ventral stream. The ventral stream is a network of regions believed to be crucial for translating visual input into a meaningful representation that can be understood and acted upon. Our aim is to better understand how regions along the ventral stream support perceptual and conceptual processes.
Functional neuroimaging studies using healthy individuals have shown that regions in the anteromedial aspects of the ventral stream, such as the peri- and entorhinal cortices, are increasingly involved when making fine-grained distinctions between objects. These regions are particularly responsive when naming living things, which have a relatively high number of shared features (e.g. like eyes and legs), but relatively few distinguishing features (e.g. a tiger's stripes), making the recognition of living things more reliant on these fine-grained processes supported by the anteromedial temporal cortex. Further, damage to the anteriomedial temporal cortex selectively impairs (1) the recognition of living things and (2) the ability to integrate information from different sensory modalities (i.e. vision and hearing).
This research project aims to extend this work through collecting behavioural data using an extensive test battery from brain-damaged volunteers, that assesses a range of perceptual (e.g. sorting shapes by size or identifying objects at odd angles) and conceptual abilities (e.g. deciding whether certain objects belonged in groups together or answering questions about the features of an object). We aim to answer a number of specific questions by relating our volunteers' test scores to patterns of brain damage:
- Does damage to different parts of the ventral stream differentially impair perceptual and conceptual processing?
- Does damage to the anteromedial temporal cortex leave perceptual processing intact while selectively impairing
- the more fine-grained processing of objects?
- recognition of living vs. nonliving things?
- Do perceptual and conceptual abilities depend specifically upon the left or right hemisphere?
- Which abilities require both hemispheres, and which can be supported by either hemisphere?
Regions in anteromedial temporal lobe where damage impairs integration of a concept's features across visual and auditory modalities. The graph shows that damage has a greater influence on performance for living things (red) than nonliving things (black). From Taylor et al. (2009) Brain, 132, 671-683.