Conceptual structure and the structure of concepts: a distributed account of category-specific deficits

Can a Hebbian-like learning rule be avoiding the curse of dimensionality in sparse distributed data?

It is generally assumed that the brain uses something akin to sparse distributed representations. These representations, however, are high-dimensional and consequently they affect classification performance of traditional Machine Learning models due to the "curse of dimensionality". In tasks for which there is a vast amount of labeled data, Deep Networks seem to solve this issue with many layers and a non-Hebbian backpropagation algorithm. The brain, however, seems to be able to solve the problem with few layers. In this work, we hypothesize that this happens by using Hebbian learning. Actually, the Hebbian-like learning rule of Restricted Boltzmann Machines learns the input patterns asymmetrically. It exclusively learns the correlation between non-zero values and ignores the zeros, which represent the vast majority of the input dimensionality. By ignoring the zeros the "curse of dimensionality" problem can be avoided. To test our hypothesis, we generated several sparse datasets and compared the performance of a Restricted Boltzmann Machine classifier with some Backprop-trained networks. The experiments using these codes confirm our initial intuition as the Restricted Boltzmann Machine shows a good generalization performance, while the Neural Networks trained with the backpropagation algorithm overfit the training data.

What we mean when we say semantic: Toward a multidisciplinary semantic glossary

Tulving characterized semantic memory as a vast repository of meaning that underlies language and many other cognitive processes. This perspective on lexical and conceptual knowledge galvanized a new era of research undertaken by numerous fields, each with their own idiosyncratic methods and terminology. For example, "concept" has different meanings in philosophy, linguistics, and psychology. As such, many fundamental constructs used to delineate semantic theories remain underspecified and/or opaque. Weak construct specificity is among the leading causes of the replication crisis now facing psychology and related fields. Term ambiguity hinders cross-disciplinary communication, falsifiability, and incremental theory-building. Numerous cognitive subdisciplines (e.g., vision, affective neuroscience) have recently addressed these limitations via the development of consensus-based guidelines and definitions. The project to follow represents our effort to produce a multidisciplinary semantic glossary consisting of succinct definitions, background, principled dissenting views, ratings of agreement, and subjective confidence for 17 target constructs (e.g., abstractness, abstraction, concreteness, concept, embodied cognition, event semantics, lexical-semantic, modality, representation, semantic control, semantic feature, simulation, semantic distance, semantic dimension). We discuss potential benefits and pitfalls (e.g., implicit bias, prescriptiveness) of these efforts to specify a common nomenclature that other researchers might index in specifying their own theoretical perspectives (e.g., They said X, but I mean Y).

Helpless infants are learning a foundation model

Humans have a protracted postnatal helplessness period, typically attributed to human-specific maternal constraints causing an early birth when the brain is highly immature. By aligning neurodevelopmental events across species, however, it has been found that humans are not born with especially immature brains compared with animal species with a shorter helpless period. Consistent with this, the rapidly growing field of infant neuroimaging has found that brain connectivity and functional activation at birth share many similarities with the mature brain. Inspired by machine learning, where deep neural networks also benefit from a 'helpless period' of pre-training, we propose that human infants are learning a foundation model: a set of fundamental representations that underpin later cognition with high performance and rapid generalisation.

The effects of sensorimotor and linguistic information on the basic-level advantage

The basic-level advantage is one of the best-known effects in human categorisation. Traditional accounts argue that basic-level categories present a maximally informative or entry level into a taxonomic organisation of concepts in semantic memory. However, these explanations are not fully compatible with most recent views on the structure of the conceptual system such as linguistic-simulation accounts, which emphasise the dual role of sensorimotor (i.e., perception-action experience of the world) and linguistic distributional information (i.e., statistical distribution of words in language) in conceptual processing. In four preregistered word→picture categorisation studies, we examined whether novel measures of sensorimotor and linguistic distance contribute to the basic level-advantage in categorical decision-making. Results showed that overlap in sensorimotor experience between category concept and member concept (e.g., animal→dog) predicted RT and accuracy at least as well as a traditional division into discrete subordinate, basic, and superordinate taxonomic levels. Furthermore, linguistic distributional information contributed to capturing effects of graded category structure where typicality ratings did not. Finally, when image label production frequency was taken into account (i.e., how often people actually produced specific labels for images), linguistic distributional information predicted RT and accuracy above and beyond sensorimotor information. These findings add to our understanding of how sensorimotor-linguistic theories of the conceptual system can explain categorisation behaviour.

Transcranial Direct Current Stimulation Effects on the Neural Substrate of Conceptual Representations

The aim of this study was to shed light on the neural substrate of conceptual representations starting from the construct of higher-order convergence zones and trying to evaluate the unitary or non-unitary nature of this construct. We used the 'Thematic and Taxonomic Semantic (TTS) task' to investigate (a) the neural substrate of stimuli belonging to biological and artifact categories, (b) the format of stimuli presentation, i.e., verbal or pictorial, and (c) the relation between stimuli, i.e., categorial or contextual. We administered anodal transcranial direct current stimulation (tDCS) to different brain structures during the execution of the TTS task. Twenty healthy participants were enrolled and divided into two groups, one investigating the role of the anterior temporal lobes (ATL) and the other the temporo-parietal junctions (TPJ). Each participant underwent three sessions of stimulation to facilitate a control condition and to investigate the role of both hemispheres. Results showed that ATL stimulation influenced all conceptual representations in relation to the format of presentation (i.e., left-verbal and right-pictorial). Moreover, ATL stimulation modulated living categories and taxonomic relations specifically, whereas TPJ stimulation did not influence semantic task performances.

Why are human animacy judgments continuous rather than categorical? A computational modeling approach

Introduction

The concept of animacy is often taken as a basic natural concept, in part I because most cases seem unambiguous. Most entities either are or are not animate. However, human animacy judgments do not reflect this binary classification. They suggest that there are borderline cases, such as virus, amoeba, fly, and imaginary beings (giant, dragon, god). Moreover, human roles (professor, mother, girlfriend) are consistently recognized as animate by far less than 100% of human judges.

Method

In this paper, I use computational modeling to identify features associated with human animacy judgments, modeling human animacy and living/non-living judgments using both bottom-up predictors (the principal components from a word embedding model) and top-down predictors (cosine distances from the names of animate categories).

Results

The results suggest that human animacy judgments may be relying on information obtained from imperfect estimates of category membership that are reflected in the word embedding models. Models using cosine distance from category names mirror human judgments in distinguishing strongly between humans (estimated lower animacy by the measure) and other animals (estimated higher animacy by the measure).

Discussion

These results are consistent with a family resemblance approach to the apparently categorical concept of animacy.

Semantic feature production norms for manipulable objects

Feature generation tasks and feature databases are important for understanding how knowledge is organized in semantic memory, as they reflect not only the kinds of information that individuals hold about objects but also how objects are conceptually represented. Traditionally, semantic norms focus on a variety of object categories and, as a result, have a small number of concepts per semantic category. Here, our main goal is to create a more fine-grained feature database exclusively for one category of objects-manipulable objects. This database contributes to the understanding of within-category, content-specific processing. To achieve this, we asked 130 participants to freely generate features for 80 manipulable objects and another group of 32 participants to generate action features for the same objects. We then compared our databases with other published semantic norms and found high similarity between them. In our databases, we calculated the similarity between objects in terms of visual, functional, encyclopaedic, and action feature types using Spearman correlation, Baker's gamma index, and cophenetic correlation. We discovered that objects were grouped in a distinctive and meaningful way according to feature type. Finally, we tested the validity of our databases by asking three groups of participants to perform a feature verification experiment while manipulating production frequency. Our results demonstrate that participants can recognize and associate the features of our databases with specific manipulable objects. Participants were faster to verify high-frequency features than low-frequency features. Overall, our data provide important insights into how we process manipulable objects and can be used to further inform cognitive and neural theories of object processing and identification.

Decoding semantic representations in mind and brain

A key goal for cognitive neuroscience is to understand the neurocognitive systems that support semantic memory. Recent multivariate analyses of neuroimaging data have contributed greatly to this effort, but the rapid development of these novel approaches has made it difficult to track the diversity of findings and to understand how and why they sometimes lead to contradictory conclusions. We address this challenge by reviewing cognitive theories of semantic representation and their neural instantiation. We then consider contemporary approaches to neural decoding and assess which types of representation each can possibly detect. The analysis suggests why the results are heterogeneous and identifies crucial links between cognitive theory, data collection, and analysis that can help to better connect neuroimaging to mechanistic theories of semantic cognition.

Investigating the influence of semantic factors on word retrieval: Reservations, results and recommendations

There is consensus that word retrieval starts with activation of semantic representations. However, in adults without language impairment, relatively little attention has been paid to the effects of the semantic attributes of to-be-retrieved words. This paper, therefore, addresses the question of which item-inherent semantic factors influence word retrieval. Specifically, it reviews the literature on a selection of these factors: imageability, concreteness, number of semantic features, typicality, intercorrelational density, featural distinctiveness, concept distinctiveness, animacy, semantic neighbourhood density, semantic similarity, operativity, valence, and arousal. It highlights several methodological challenges in this field, and has a focus on the insights from studies with people with aphasia where the effects of these variables are more prevalent. The paper concludes that further research simultaneously examining the effects of different semantic factors that are likely to affect lexical co-activation, and the interaction of these variables, would be fruitful, as would suitably scaled computational modelling of these effects in unimpaired language processing and in language impairment. Such research would enable the refinement of theories of semantic processing and word production, and potentially have implications for diagnosis and treatment of semantic and lexical impairments.

Moving in Semantic Space in Prodromal and Very Early Alzheimer's Disease: An Item-Level Characterization of the Semantic Fluency Task

The semantic fluency task is a widely used clinical tool in the diagnostic process of Alzheimer's disease. The task requires efficient mapping of the semantic space to produce as many items as possible within a semantic category. We examined whether healthy volunteers (n = 42) and patients with early Alzheimer's disease (24 diagnosed with amnestic Mild Cognitive Impairment and 18 with early Alzheimer's dementia) take advantage of and travel in the semantic space differently. With focus on the animal fluency task, we sought to emulate the detailed structure of the multidimensional semantic space by utilizing word2vec-method from the natural language processing domain. To render the resulting multidimensional semantic space visually comprehensible, we applied a dimensionality reduction algorithm (t-SNE), which enabled a straightforward division of the semantic space into sub-categories. Moving in semantic space was quantified with the number of items created, sub-categories visited, and switches and returns to these sub-categories. Multinomial logistic regression models were used to predict the diagnostic group with these independent variables. We found that returning to a sub-category provided additional information, besides the number of words produced in the task, to differentiate patients with Alzheimer's dementia from both amnestic Mild Cognitive Impairment patients and healthy controls. The results suggest that the frequency of returning to a sub-category may serve as an additional aid for clinicians in diagnosing early Alzheimer's disease. Moreover, our results imply that the combination of word2vec and subsequent t-SNE-visualization may offer a valuable tool for examining the semantic space and its sub-categories.

Linguistic Distributional Knowledge and Sensorimotor Grounding both Contribute to Semantic Category Production

The human conceptual system comprises simulated information of sensorimotor experience and linguistic distributional information of how words are used in language. Moreover, the linguistic shortcut hypothesis predicts that people will use computationally cheaper linguistic distributional information where it is sufficient to inform a task response. In a pre-registered category production study, we asked participants to verbally name members of concrete and abstract categories and tested whether performance could be predicted by a novel measure of sensorimotor similarity (based on an 11-dimensional representation of sensorimotor strength) and linguistic proximity (based on word co-occurrence derived from a large corpus). As predicted, both measures predicted the order and frequency of category production but, critically, linguistic proximity had an effect above and beyond sensorimotor similarity. A follow-up study using typicality ratings as an additional predictor found that typicality was often the strongest predictor of category production variables, but it did not subsume sensorimotor and linguistic effects. Finally, we created a novel, fully grounded computational model of conceptual activation during category production, which best approximated typical human performance when conceptual activation was allowed to spread indirectly between concepts, and when candidate category members came from both sensorimotor and linguistic distributional representations. Critically, model performance was indistinguishable from typical human performance. Results support the linguistic shortcut hypothesis in semantic processing and provide strong evidence that both linguistic and grounded representations are inherent to the functioning of the conceptual system. All materials, data, and code are available at https://osf.io/vaq56/.

Visual features drive the category-specific impairments on categorization tasks in a patient with object agnosia

Object and scene recognition both require mapping of incoming sensory information to existing conceptual knowledge about the world. A notable finding in brain-damaged patients is that they may show differentially impaired performance for specific categories, such as for "living exemplars". While numerous patients with category-specific impairments have been reported, the explanations for these deficits remain controversial. In the current study, we investigate the ability of a brain injured patient with a well-established category-specific impairment of semantic memory to perform two categorization experiments: 'natural' vs. 'manmade' scenes (experiment 1) and objects (experiment 2). Our findings show that the pattern of categorical impairment does not respect the natural versus manmade distinction. This suggests that the impairments may be better explained by differences in visual features, rather than by category membership. Using Deep Convolutional Neural Networks (DCNNs) as 'artificial animal models' we further explored this idea. Results indicated that DCNNs with 'lesions' in higher order layers showed similar response patterns, with decreased relative performance for manmade scenes (experiment 1) and natural objects (experiment 2), even though they have no semantic category knowledge, apart from a mapping between pictures and labels. Collectively, these results suggest that the direction of category-effects to a large extent depends, at least in MS' case, on the degree of perceptual differentiation called for, and not semantic knowledge.

Representation of semantic typicality in brain activation in healthy adults and individuals with aphasia: A multi-voxel pattern analysis

This study aimed to investigate brain regions that show different activation patterns between semantically typical and atypical items in both healthy adults and individuals with aphasia (PWA). Eighteen neurologically healthy adults and twenty-one PWA participated in an fMRI semantic feature verification task that included typical and atypical stimuli from five different semantic categories. A whole-brain searchlight multi-voxel pattern analysis (MVPA) was conducted to classify brain activation patterns between typical and atypical conditions in each participant group separately. Behavioral responses were faster and more accurate for typical vs. atypical items across both groups. The searchlight MVPA identified two significant clusters in healthy adults: left middle occipital gyrus and right calcarine cortex, but no significant clusters were found in PWA. A follow-up analysis in PWA revealed a significant association between neural classification of semantic typicality in the left middle occipital gyrus and reaction times in the fMRI task. When the typicality effect was examined for each semantic category at the univariate level, significance was identified in the visual cortex for fruits in both groups of participants. These findings suggest that semantic typicality was modulated in the visual cortex in healthy individuals, but to a lesser extent in the same region in PWA.

On Simulating Neural Damage in Connectionist Networks

A key strength of connectionist modelling is its ability to simulate both intact cognition and the behavioural effects of neural damage. We survey the literature, showing that models have been damaged in a variety of ways, e.g. by removing connections, by adding noise to connection weights, by scaling weights, by removing units and by adding noise to unit activations. While these different implementations of damage have often been assumed to be behaviourally equivalent, some theorists have made aetiological claims that rest on nonequivalence. They suggest that related deficits with different aetiologies might be accounted for by different forms of damage within a single model. We present two case studies that explore the effects of different forms of damage in two influential connectionist models, each of which has been applied to explain neuropsychological deficits. Our results indicate that the effect of simulated damage can indeed be sensitive to the way in which damage is implemented, particularly when the environment comprises subsets of items that differ in their statistical properties, but such effects are sensitive to relatively subtle aspects of the model's training environment. We argue that, as a consequence, substantial methodological care is required if aetiological claims about simulated neural damage are to be justified, and conclude more generally that implementation assumptions, including those concerning simulated damage, must be fully explored when evaluating models of neurological deficits, both to avoid over-extending the explanatory power of specific implementations and to ensure that reported results are replicable.

ELECTRONIC SUPPLEMENTARY MATERIAL

The online version of this article (10.1007/s42113-020-00081-z) contains supplementary material, which is available to authorized users.

Feature distinctiveness effects in language acquisition and lexical processing: Insights from megastudies

Semantic features are central to many influential theories of word meaning and semantic memory, but new methods of quantifying the information embedded in feature production norms are needed to advance our understanding of semantic processing and language acquisition. This paper capitalized on databases of semantic feature production norms and age-of-acquisition ratings, and megastudies including the English Lexicon Project and the Calgary Semantic Decision Project, to examine the influence of feature distinctiveness on language acquisition, visual lexical decision, and semantic decision. A feature network of English words was constructed such that edges in the network represented feature distance, or dissimilarity, between words (i.e., Jaccard and Manhattan distances of probability distributions of features elicited for each pair of words), enabling us to quantify the relative feature distinctiveness of individual words relative to other words in the network. Words with greater feature distinctiveness tended to be acquired earlier. Regression analyses of megastudy data revealed that Manhattan feature distinctiveness inhibited performance on the visual lexical decision task, facilitated semantic decision performance for concrete concepts, and inhibited semantic decision performance for abstract concepts. These results demonstrate the importance of considering the structural properties of words embedded in a semantic feature space in order to increase our understanding of semantic processing and language acquisition.

Measures of Neural Similarity

One fundamental question is what makes two brain states similar. For example, what makes the activity in visual cortex elicited from viewing a robin similar to a sparrow? One common assumption in fMRI analysis is that neural similarity is described by Pearson correlation. However, there are a host of other possibilities, including Minkowski and Mahalanobis measures, with each differing in its mathematical, theoretical, and neural computational assumptions. Moreover, the operable measures may vary across brain regions and tasks. Here, we evaluated which of several competing similarity measures best captured neural similarity. Our technique uses a decoding approach to assess the information present in a brain region, and the similarity measures that best correspond to the classifier’s confusion matrix are preferred. Across two published fMRI datasets, we found the preferred neural similarity measures were common across brain regions but differed across tasks. Moreover, Pearson correlation was consistently surpassed by alternatives.

The Neural Representations of Movement across Semantic Categories

Previous evidence from neuropsychological and neuroimaging studies suggests functional specialization for tools and related semantic knowledge in a left frontoparietal network. It is still debated whether these areas are involved in the representation of rudimentary movement-relevant knowledge regardless of semantic domains (animate vs. inanimate) or categories (tools vs. nontool objects). Here, we used fMRI to record brain activity while 13 volunteers performed two semantic judgment tasks on visually presented items from three different categories: animals, tools, and nontool objects. Participants had to judge two distinct semantic features: whether two items typically move in a similar way (e.g., a fan and a windmill move in circular motion) or whether they are usually found in the same environment (e.g., a seesaw and a swing are found in a playground). We investigated differences in overall activation (which areas are involved) as well as representational content (which information is encoded) across semantic features and categories. Results of voxel-wise mass univariate analysis showed that, regardless of semantic category, a dissociation emerges between processing information on prototypical location (involving the anterior temporal cortex and the angular gyrus) and movement (linked to left inferior parietal and frontal activation). Multivoxel pattern correlation analyses confirmed the representational segregation of networks encoding task- and category-related aspects of semantic processing. Taken together, these findings suggest that the left frontoparietal network is recruited to process movement properties of items (including both biological and nonbiological motion) regardless of their semantic category.

Reconstructing meaning from bits of information

Modern theories of semantics posit that the meaning of words can be decomposed into a unique combination of semantic features (e.g., "dog" would include "barks"). Here, we demonstrate using functional MRI (fMRI) that the brain combines bits of information into meaningful object representations. Participants receive clues of individual objects in form of three isolated semantic features, given as verbal descriptions. We use machine-learning-based neural decoding to learn a mapping between individual semantic features and BOLD activation patterns. The recorded brain patterns are best decoded using a combination of not only the three semantic features that were in fact presented as clues, but a far richer set of semantic features typically linked to the target object. We conclude that our experimental protocol allowed us to demonstrate that fragmented information is combined into a complete semantic representation of an object and to identify brain regions associated with object meaning.

Concepts, control, and context: A connectionist account of normal and disordered semantic cognition

Semantic cognition requires conceptual representations shaped by verbal and nonverbal experience and executive control processes that regulate activation of knowledge to meet current situational demands. A complete model must also account for the representation of concrete and abstract words, of taxonomic and associative relationships, and for the role of context in shaping meaning. We present the first major attempt to assimilate all of these elements within a unified, implemented computational framework. Our model combines a hub-and-spoke architecture with a buffer that allows its state to be influenced by prior context. This hybrid structure integrates the view, from cognitive neuroscience, that concepts are grounded in sensory-motor representation with the view, from computational linguistics, that knowledge is shaped by patterns of lexical co-occurrence. The model successfully codes knowledge for abstract and concrete words, associative and taxonomic relationships, and the multiple meanings of homonyms, within a single representational space. Knowledge of abstract words is acquired through (a) their patterns of co-occurrence with other words and (b) acquired embodiment, whereby they become indirectly associated with the perceptual features of co-occurring concrete words. The model accounts for executive influences on semantics by including a controlled retrieval mechanism that provides top-down input to amplify weak semantic relationships. The representational and control elements of the model can be damaged independently, and the consequences of such damage closely replicate effects seen in neuropsychological patients with loss of semantic representation versus control processes. Thus, the model provides a wide-ranging and neurally plausible account of normal and impaired semantic cognition.

A feature-based neurocomputational model of semantic memory

According with a featural organization of semantic memory, this work is aimed at investigating, through an attractor network, the role of different kinds of features in the representation of concepts, both in normal and neurodegenerative conditions. We implemented new synaptic learning rules in order to take into account the role of partially shared features and of distinctive features with different saliency. The model includes semantic and lexical layers, coding, respectively for object features and word-forms. Connections among nodes are strongly asymmetrical. To account for the feature saliency, asymmetrical synapses were created using Hebbian rules of potentiation and depotentiation, setting different pre-synaptic and post-synaptic thresholds. A variable post-synaptic threshold, which automatically changed to reflect the feature frequency in different concepts (i.e., how many concepts share a feature), was used to account for partially shared features. The trained network solved naming tasks and word recognition tasks very well, exploiting the different role of salient versus marginal features in concept identification. In the case of damage, superordinate concepts were preserved better than the subordinate ones. Interestingly, the degradation of salient features, but not of marginal ones, prevented object identification. The model suggests that Hebbian rules, with adjustable post-synaptic thresholds, can provide a reliable semantic representation of objects exploiting the statistics of input features.

The attribute priming effect in patients with Alzheimer's disease

Experiments with semantic priming (SP) paradigms have documented early hypopriming in patients with AD when concepts are used as primes and attribute concept features as targets, suggesting that concept attributes are vulnerable to damage very early in the disease course. The aims of this study were to confirm early priming reduction in the attribute condition in patients with AD and to determine which of several semantic indexes (such as the level of distinctiveness, correlation or feature dominance of concept features) best predicts the priming effect size in AD. We administered an SP attribute condition paradigm to 20 mildly demented patients with AD and to 10 NCs. We used concept-attribute pairs for which normative data of semantic indexes relative to both concept primes (i.e., number, type, mean level of dominance, distinctiveness and correlation of features constituting the concepts) and target features (i.e., level of feature dominance, correlation and distinctiveness) were available. Results showed that compared to NCs, the AD group obtained very reduced priming facilitation. Furthermore, the item regression analyses showed that the priming decrement in the AD group was predicted by the feature dominance of the target in the related pairs; that is, the lower the target feature dominance, the lower the priming effect elicited. These results confirmed hypopriming in the attribute condition from the very early phase of AD and support the view that attributes which are more salient for the identification of a given concept are also those most resistant to semantic memory degradation in AD pathology.

Animacy and real-world size shape object representations in the human medial temporal lobes

Identifying what an object is, and whether an object has been encountered before, is a crucial aspect of human behavior. Despite this importance, we do not yet have a complete understanding of the neural basis of these abilities. Investigations into the neural organization of human object representations have revealed category specific organization in the ventral visual stream in perceptual tasks. Interestingly, these categories fall within broader domains of organization, with reported distinctions between animate, inanimate large, and inanimate small objects. While there is some evidence for category specific effects in the medial temporal lobe (MTL), in particular in perirhinal and parahippocampal cortex, it is currently unclear whether domain level organization is also present across these structures. To this end, we used fMRI with a continuous recognition memory task. Stimuli were images of objects from several different categories, which were either animate or inanimate, or large or small within the inanimate domain. We employed representational similarity analysis (RSA) to test the hypothesis that object-evoked responses in MTL structures during recognition-memory judgments also show evidence for domain-level organization along both dimensions. Our data support this hypothesis. Specifically, object representations were shaped by either animacy, real-world size, or both, in perirhinal and parahippocampal cortex, and the hippocampus. While sensitivity to these dimensions differed across structures when probed individually, hinting at interesting links to functional differentiation, similarities in organization across MTL structures were more prominent overall. These results argue for continuity in the organization of object representations in the ventral visual stream and the MTL.

Primacy of Functional Knowledge in Semantic Representations: The Case of Living and Nonliving Things

In 3 experiments, participants decided whether sensory and functional features were true of living and nonliving concepts. In Experiments 1 and 2, concepts were presented twice: test phase followed study phase after either 3 min (Experiment 1) or 3 s (Experiment 2). At test, concepts were paired with the same feature as that at study, or a different feature from either the same modality (within-modality priming) or another modality (cross-modality priming). In both experiments functional decisions were faster than sensory decisions for living and nonliving concepts. Whilst no semantic priming occurred between study and test in Experiment 1, the shorter study–test interval of Experiment 2 did lead to test phase semantic priming. Here there was greater within- than cross-modality priming for sensory decisions, but equivalent within- and cross-modality priming for functional decisions owing to significantly greater facilitation of functional decisions from prior sensory decisions than vice versa. Experiment 3 involved a single verification phase: For half the participants the feature name preceded the concept name, and for half the concept name preceded the feature name. The functional processing advantage persisted irrespective of presentation order. Results suggest that functional information is central to the representation of all concepts: Function is processed faster than sensory information and is activated obligatorily.

Age of acquisition for naming and knowing: A new hypothesis

This paper reports an investigation into the age of acquisition of object names and object knowledge in a cross-sectional study of 288 children aged between 3 years 7 months and 11 years 6 months, comprising equal numbers of boys and girls. The objects belonged to four categories: animals, fruit and vegetables, implements, and vehicles. They were presented in three image types: line drawings, black-and-white photographs, and coloured photographs. In the knowledge test, five probe questions were asked for each object given the spoken name. Results showed that line drawings were more difficult to name than either black-and-white photographs or coloured photographs, which did not differ. The boys significantly out-performed the girls at naming and knowing, both overall and specifically for the category of vehicles. Naming and knowledge increased steadily with age but while young children below about 6 years 6 months showed an advantage to naming, older children showed an advantage to knowing. Similarly, age-of-acquisition measures for each item revealed a significant shift in the relationship between naming and knowing at around 80 months. We argue that differences in learning experience lead younger and older children to associate object names with different types of information, and we suggest that this difference probably accounts for the age-of-acquisition effects reported in adult object naming.

The Anatomy of Object Processing: The Role of Anteromedial Temporal Cortex

How objects are represented and processed in the brain remains a key issue in cognitive neuroscience. We have developed a conceptual structure account in which category-specific semantic deficits emerge due to differences in the structure and content of concepts rather than from explicit divisions of conceptual knowledge in separate stores. The primary claim is that concepts associated with particular categories (e.g., animals, tools) differ in the number and type of properties and the extent to which these properties are correlated with each other. In this review, we describe recent neuropsychological and neuroimaging studies in which we have extended our theoretical account by incorporating recent claims about the neuroanatomical basis of feature integration and differentiation that arise from research into hierarchical object processing streams in nonhuman primates and humans. A clear picture has emerged in which the human perirhinal cortex and neighbouring anteromedial temporal structures appear to provide the neural infrastructure for making fine-grained discriminations among objects, suggesting that damage within the perirhinal cortex may underlie the emergence of category-specific semantic deficits in brain-damaged patients.

Incremental learning of perceptual and conceptual representations and the puzzle of neural repetition suppression

Incremental learning models of long-term perceptual and conceptual knowledge hold that neural representations are gradually acquired over many individual experiences via Hebbian-like activity-dependent synaptic plasticity across cortical connections of the brain. In such models, variation in task relevance of information, anatomic constraints, and the statistics of sensory inputs and motor outputs lead to qualitative alterations in the nature of representations that are acquired. Here, the proposal that behavioral repetition priming and neural repetition suppression effects are empirical markers of incremental learning in the cortex is discussed, and research results that both support and challenge this position are reviewed. Discussion is focused on a recent fMRI-adaptation study from our laboratory that shows decoupling of experience-dependent changes in neural tuning, priming, and repetition suppression, with representational changes that appear to work counter to the explicit task demands. Finally, critical experiments that may help to clarify and resolve current challenges are outlined.

Finding features, figuratively

Object concepts refer to unique clusters of properties that can be selectively activated or inhibited depending on what information is currently relevant. This conceptual "stretching" enables limitless new meanings to be generated, and figurative language provides a useful framework in which to study this conceptual flexibility. Here we probe the cognitive and neural mechanisms underlying the comprehension of novel metaphors as a means of understanding the conceptual flexibility inherent to language processing more generally. We show that novel metaphor comprehension involves the activation or inhibition of conceptual properties that are either relevant or irrelevant to the metaphor, and that left inferior frontal gyrus is recruited in this process, supporting a role for this region in the fine-tuning of conceptual meaning. Our results are consistent with a flexible, compositional account of conceptual structure in which semantic control mechanisms operate over conceptual properties during figurative language comprehension in order to create context-dependent meaning.

Defining a Conceptual Topography of Word Concreteness: Clustering Properties of Emotion, Sensation, and Magnitude among 750 English Words

Cognitive science has a longstanding interest in the ways that people acquire and use abstract vs. concrete words (e.g., truth vs. piano). One dominant theory holds that abstract and concrete words are subserved by two parallel semantic systems. We recently proposed an alternative account of abstract-concrete word representation premised upon a unitary, high dimensional semantic space wherein word meaning is nested. We hypothesize that a range of cognitive and perceptual dimensions (e.g., emotion, time, space, color, size, visual form) bound this space, forming a conceptual topography. Here we report a normative study where we examined the clustering properties of a sample of English words (N = 750) spanning a spectrum of concreteness in a continuous manner from highly abstract to highly concrete. Participants (N = 328) rated each target word on a range of 14 cognitive dimensions (e.g., color, emotion, valence, polarity, motion, space). The dimensions reduced to three factors: Endogenous factor, Exogenous factor, and Magnitude factor. Concepts were plotted in a unified, multimodal space with concrete and abstract concepts along a continuous continuum. We discuss theoretical implications and practical applications of this dataset. These word norms are freely available for download and use at http://www.reilly-coglab.com/data/.

The neuro-cognitive representations of symbols: the case of concrete words

We live our lives surrounded by symbols (e.g., road signs, logos, but especially words and numbers), and throughout our life we use them to evoke, communicate and reflect upon ideas and things that are not currently present to our senses. Symbols are represented in our brains at different levels of complexity: at the first and most simple level, as physical entities, in the corresponding primary and secondary sensory cortices. The crucial property of symbols, however, is that, despite the simplicity of their surface forms, they have the power of evoking higher order multifaceted representations that are implemented in distributed neural networks spanning a large portion of the cortex. The rich internal states that reflect our knowledge of the meaning of symbols are what we call semantic representations. In this review paper, we summarize our current knowledge of both the cognitive and neural substrates of semantic representations, focusing on concrete words (i.e., nouns or verbs referring to concrete objects and actions), which, together with numbers, are the most-studied and well defined classes of symbols. Following a systematic descriptive approach, we will organize this literature review around two key questions: what is the content of semantic representations? And, how are semantic representations implemented in the brain, in terms of localization and dynamics? While highlighting the main current opposing perspectives on these topics, we propose that a fruitful way to make substantial progress in this domain would be to adopt a geometrical view of semantic representations as points in high dimensional space, and to operationally partition the space of concrete word meaning into motor-perceptual and conceptual dimensions. By giving concrete examples of the kinds of research that can be done within this perspective, we illustrate how we believe this framework will foster theoretical speculations as well as empirical research.

A unified model of human semantic knowledge and its disorders

How is knowledge about the meanings of words and objects represented in the human brain? Current theories embrace two radically different proposals: either distinct cortical systems have evolved to represent different kinds of things, or knowledge for all kinds is encoded within a single domain-general network. Neither view explains the full scope of relevant evidence from neuroimaging and neuropsychology. Here we propose that graded category-specificity emerges in some components of the semantic network through joint effects of learning and network connectivity. We test the proposal by measuring connectivity amongst cortical regions implicated in semantic representation, then simulating healthy and disordered semantic processing in a deep neural network whose architecture mirrors this structure. The resulting neuro-computational model explains the full complement of neuroimaging and patient evidence adduced in support of both domain-specific and domain-general approaches, reconciling long-standing disputes about the nature and origins of this uniquely human cognitive faculty.

Past, present, and prospects: Reflections 40 years on from the selective impairment of semantic memory (Warrington, 1975)

We summarize the main findings and conclusions of Warrington's (1975) paper, The Selective Impairment of Semantic memory, a neuropsychological paper that described three cases with degenerative neurological conditions [Warrington, E. K. (1975). The selective impairment of semantic memory. The Quarterly Journal of Experimental Psychology, 27, 635–657]. We consider the developments that have followed from its publication and give a selective overview of the field in 2014. The initial impact of the paper was on neuropsychological investigations of semantic loss followed some 14 years later by the identification of Semantic Dementia (the condition shown by the original cases) as a distinctive form of degenerative disease with unique clinical and pathological characteristics. We discuss the distinction between disorders of semantic storage and refractory semantic access, the evidence for category- and modality-specific impairments of semantics, and the light that has been shed on the structure and organization of semantic memory. Finally we consider the relationship between semantic memory and the skills of reading and writing, phonological processing, and autobiographical memory.

The meaning of 'life' and other abstract words: Insights from neuropsychology

There are a number of long-standing theories on how the cognitive processing of abstract words, like 'life', differs from that of concrete words, like 'knife'. This review considers current perspectives on this debate, focusing particularly on insights obtained from patients with language disorders and integrating these with evidence from functional neuroimaging studies. The evidence supports three distinct and mutually compatible hypotheses. (1) Concrete and abstract words differ in their representational substrates, with concrete words depending particularly on sensory experiences and abstract words on linguistic, emotional, and magnitude-based information. Differential dependence on visual versus verbal experience is supported by the evidence for graded specialization in the anterior temporal lobes for concrete versus abstract words. In addition, concrete words have richer representations, in line with better processing of these words in most aphasic patients and, in particular, patients with semantic dementia. (2) Abstract words place greater demands on executive regulation processes because they have variable meanings that change with context. This theory explains abstract word impairments in patients with semantic-executive deficits and is supported by neuroimaging studies showing greater response to abstract words in inferior prefrontal cortex. (3) The relationships between concrete words are governed primarily by conceptual similarity, while those of abstract words depend on association to a greater degree. This theory, based primarily on interference and priming effects in aphasic patients, is the most recent to emerge and the least well understood. I present analyses indicating that patterns of lexical co-occurrence may be important in understanding these effects.

Relational categories are more mutable than entity categories

Across three experiments, we explore differences between relational categories-whose members share common relational patterns-and entity categories, whose members share common intrinsic properties. Specifically, we test the claim that relational concepts are more semantically mutable in context, and therefore less stable in memory, than entity concepts. We compared memory for entity nouns and relational nouns, tested either in the same context as at encoding or in a different context. We found that (a) participants show better recognition accuracy for entity nouns than for relational nouns, and (b) recognition of relational nouns is more impaired by a change in context than is recognition of entity nouns. We replicated these findings even when controlling for factors highly correlated with relationality, such as abstractness-concreteness. This suggests that the contextual mutability of relational concepts is due to the core semantic property of conveying relational structure and not simply to accompanying characteristics such as abstractness. We note parallels with the distinction between nouns and verbs and suggest implications for lexical and conceptual structure. Finally, we relate these patterns to proposals that a deep distinction exists between words with an essentially referential function and those with a predicate function.

Feature Statistics Modulate the Activation of Meaning During Spoken Word Processing

Understanding spoken words involves a rapid mapping from speech to conceptual representations. One distributed feature-based conceptual account assumes that the statistical characteristics of concepts' features--the number of concepts they occur in (distinctiveness/sharedness) and likelihood of co-occurrence (correlational strength)--determine conceptual activation. To test these claims, we investigated the role of distinctiveness/sharedness and correlational strength in speech-to-meaning mapping, using a lexical decision task and computational simulations. Responses were faster for concepts with higher sharedness, suggesting that shared features are facilitatory in tasks like lexical decision that require access to them. Correlational strength facilitated responses for slower participants, suggesting a time-sensitive co-occurrence-driven settling mechanism. The computational simulation showed similar effects, with early effects of shared features and later effects of correlational strength. These results support a general-to-specific account of conceptual processing, whereby early activation of shared features is followed by the gradual emergence of a specific target representation.

Verbal Description of Concrete Objects: A Method for Assessing Semantic Circumlocution in Persons With Aphasia

PURPOSE

We investigated from a theoretically motivated perspective what information differentiated sufficient from insufficient descriptions of objects provided by persons with aphasia.

METHOD

Twenty-one adults with aphasia consequent to single left-hemisphere stroke verbally described 9 living and 9 nonliving objects. Responses were scored for accuracy (i.e., sufficiency) and tallied for type and quantity of semantic feature information provided. Main effects and interactions were identified using repeated measures analyses of variance, with significant findings followed up with planned comparisons.

RESULTS

Differences between correct and incorrect descriptions were identified with respect to both feature type and feature distinctiveness for living and nonliving items, in particular highlighting the importance of distinctive features in descriptions of both domains.

CONCLUSIONS

These findings add to the relatively small body of literature investigating semantic feature processing in adults with aphasia. This is a critical gap to close when considered in light of the preponderance of semantically based treatments for word-retrieval impairment in stroke-aphasia. Our findings provide preliminary support for the notion that semantically guided treatments for word-retrieval impairment in stroke-aphasia may be geared toward increasing specificity of semantic circumlocution to increase semantic self-cueing and to improve communication of information to conversation partners.

A Model of Emergent Category-specific Activation in the Posterior Fusiform Gyrus of Sighted and Congenitally Blind Populations

Theories about the neural bases of semantic knowledge tend between two poles, one proposing that distinct brain regions are innately dedicated to different conceptual domains and the other suggesting that all concepts are encoded within a single network. Category-sensitive functional activations in the fusiform cortex of the congenitally blind have been taken to support the former view but also raise several puzzles. We use neural network models to assess a hypothesis that spans the two poles: The interesting functional activation patterns reflect the base connectivity of a domain-general semantic network. Both similarities and differences between sighted and congenitally blind groups can emerge through learning in a neural network, but only in architectures adopting real anatomical constraints. Surprisingly, the same constraints suggest a novel account of a quite different phenomenon: the dyspraxia observed in patients with semantic impairments from anterior temporal pathology. From this work, we suggest that the cortical semantic network is wired not to encode knowledge of distinct conceptual domains but to promote learning about both conceptual and affordance structure in the environment.

You shall know an object by the company it keeps: An investigation of semantic representations derived from object co-occurrence in visual scenes

An influential position in lexical semantics holds that semantic representations for words can be derived through analysis of patterns of lexical co-occurrence in large language corpora. Firth (1957) famously summarised this principle as "you shall know a word by the company it keeps". We explored whether the same principle could be applied to non-verbal patterns of object co-occurrence in natural scenes. We performed latent semantic analysis (LSA) on a set of photographed scenes in which all of the objects present had been manually labelled. This resulted in a representation of objects in a high-dimensional space in which similarity between two objects indicated the degree to which they appeared in similar scenes. These representations revealed similarities among objects belonging to the same taxonomic category (e.g., items of clothing) as well as cross-category associations (e.g., between fruits and kitchen utensils). We also compared representations generated from this scene dataset with two established methods for elucidating semantic representations: (a) a published database of semantic features generated verbally by participants and (b) LSA applied to a linguistic corpus in the usual fashion. Statistical comparisons of the three methods indicated significant association between the structures revealed by each method, with the scene dataset displaying greater convergence with feature-based representations than did LSA applied to linguistic data. The results indicate that information about the conceptual significance of objects can be extracted from their patterns of co-occurrence in natural environments, opening the possibility for such data to be incorporated into existing models of conceptual representation.

Semantic feature degradation and naming performance. Evidence from neurodegenerative disorders

The failure to name an object in Alzheimer's disease (AD) and in the semantic variant of the primary progressive aphasia (sv-PPA) has been generally attributed to semantic memory loss, with a progressive degradation of semantic features. Not all features, however, may have the same relevance in picture naming. We analyzed the relationship between picture naming performance and the loss of semantic features in patients with AD with or without naming impairment, with sv-PPA and in matched controls, assessing the role of distinctiveness, semantic relevance and feature type (sensorial versus non-sensorial) with a sentence verification task. The results showed that distinctive features with high values of semantic relevance were lost only in all patients with naming impairment. The performance on the sensorial distinctive features with high relevance was the best predictor of naming performance only in sv-PPA, while no difference between sensorial and non-sensorial features was found in AD patients.

Action verbal fluency in Parkinson’s patients

We compared the performance of 31 non-demented Parkinson´s disease (PD) patients to 61 healthy controls in an action verbal fluency task. Semantic and phonemic fluencies, cognitive impairment and behavioural dysfunction were also assessed. The mean disease duration of PD was 9.8 years (standard deviation (SD) = 6.13). There were no age (U = 899.5, p = 0.616), gender(chi-square = 0.00, p = 1.00) or literacy (U = 956, p = 0.96) differences between the two groups. A significant difference was observed between the two groups in the action verbal fluency task (U = 406.5, p < 0.01) that was not found in the other fluency tasks. The education level was the only biographical variable that influenced the action (verb) fluency outcomes, irrespective of disease duration. Our findings suggest a correlation between the disease mechanisms in PD and a specific verb deficit, support the validity of the action (verb) fluency as an executive function measure and suggest that this task provides unique information not captured with traditional executive function tasks.

Hemispheric lateralization of semantic feature distinctiveness

Recent models of semantic memory propose that the semantic representation of concepts is based, in part, on a network of features. In this view, a feature that is distinctive for an object (a zebra has stripes) is processed differently from a feature that is shared across many objects (a zebra has four legs). The goal of this paper is to determine whether there are hemispheric differences in such processing. In a feature verification task, participants responded 'yes' or 'no' following concepts which were presented to a single visual field (left or right) paired with a shared or distinctive feature. Both hemispheres showed faster reaction times to shared features than to distinctive features, although right hemisphere responses were significantly slower overall and particularly in the processing of distinctive features. These findings support models of semantic processing in which the dominant left hemisphere more efficiently performs highly discriminating 'fine' encoding, in contrast to the right hemisphere which performs less discriminating 'coarse' encoding.

The neural correlates of semantic richness: evidence from an fMRI study of word learning

We investigated the neural correlates of concrete nouns with either many or few semantic features. A group of 21 participants underwent two days of training and were then asked to categorize 40 newly learned words and a set of matched familiar words as living or nonliving in an MRI scanner. Our results showed that the most reliable effects of semantic richness were located in the left angular gyrus (AG) and middle temporal gyrus (MTG), where activation was higher for semantically rich than poor words. Other areas showing the same pattern included bilateral precuneus and posterior cingulate gyrus. Our findings support the view that AG and anterior MTG, as part of the multimodal network, play a significant role in representing and integrating semantic features from different input modalities. We propose that activation in bilateral precuneus and posterior cingulate gyrus reflects interplay between AG and episodic memory systems during semantic retrieval.

Automatic extraction of property norm-like data from large text corpora

Traditional methods for deriving property-based representations of concepts from text have focused on either extracting only a subset of possible relation types, such as hyponymy/hypernymy (e.g., car is-a vehicle) or meronymy/metonymy (e.g., car has wheels), or unspecified relations (e.g., car--petrol). We propose a system for the challenging task of automatic, large-scale acquisition of unconstrained, human-like property norms from large text corpora, and discuss the theoretical implications of such a system. We employ syntactic, semantic, and encyclopedic information to guide our extraction, yielding concept-relation-feature triples (e.g., car be fast, car require petrol, car cause pollution), which approximate property-based conceptual representations. Our novel method extracts candidate triples from parsed corpora (Wikipedia and the British National Corpus) using syntactically and grammatically motivated rules, then reweights triples with a linear combination of their frequency and four statistical metrics. We assess our system output in three ways: lexical comparison with norms derived from human-generated property norm data, direct evaluation by four human judges, and a semantic distance comparison with both WordNet similarity data and human-judged concept similarity ratings. Our system offers a viable and performant method of plausible triple extraction: Our lexical comparison shows comparable performance to the current state-of-the-art, while subsequent evaluations exhibit the human-like character of our generated properties.

Inborn and experience-dependent models of categorical brain organization. A position paper

The present review aims to summarize the debate in contemporary neuroscience between inborn and experience-dependent models of conceptual representations that goes back to the description of category-specific semantic disorders for biological and artifact categories. Experience-dependent models suggest that categorical disorders are the by-product of the differential weighting of different sources of knowledge in the representation of biological and artifact categories. These models maintain that semantic disorders are not really category-specific, because they do not respect the boundaries between different categories. They also argue that the brain structures which are disrupted in a given type of category-specific semantic disorder should correspond to the areas of convergence of the sensory-motor information which play a major role in the construction of that category. Furthermore, they provide a simple interpretation of gender-related categorical effects and are supported by studies assessing the importance of prior experience in the cortical representation of objects On the other hand, inborn models maintain that category-specific semantic disorders reflect the disruption of innate brain networks, which are shaped by natural selection to allow rapid identification of objects that are very relevant for survival. From the empirical point of view, these models are mainly supported by observations of blind subjects, which suggest that visual experience is not necessary for the emergence of category-specificity in the ventral stream of visual processing. The weight of the data supporting experience-dependent and inborn models is thoroughly discussed, stressing the fact observations made in blind subjects are still the subject of intense debate. It is concluded that at the present state of knowledge it is not possible to choose between experience-dependent and inborn models of conceptual representations.

Scene categorization in Alzheimer's disease: a saccadic choice task

AIMS

We investigated the performance in scene categorization of patients with Alzheimer's disease (AD) using a saccadic choice task.

METHOD

24 patients with mild AD, 28 age-matched controls and 26 young people participated in the study. The participants were presented pairs of coloured photographs and were asked to make a saccadic eye movement to the picture corresponding to the target scene (natural vs. urban, indoor vs. outdoor).

RESULTS

The patients' performance did not differ from chance for natural scenes. Differences between young and older controls and patients with AD were found in accuracy but not saccadic latency.

CONCLUSIONS

The results are interpreted in terms of cerebral reorganization in the prefrontal and temporo-occipital cortex of patients with AD, but also in terms of impaired processing of visual global properties of scenes.

What matters in semantic feature processing for persons with stroke-aphasia: Evidence from an auditory concept-feature verification task

BACKGROUND

The relationship between object concept domains (living vs. nonliving) and their underlying feature structures is a frequent area of investigation regarding semantic processing in healthy individuals and some individuals with neuropsychological impairment resulting from herpes simplex encephalitis, semantic dementia and Alzheimer's disease. However, this relationship has been less well-investigated in persons with stroke-aphasia (PWA), even though many treatments for anomia following stroke are predicated on the use of semantic feature cues.

AIMS

As part of a larger investigation into the influence of semantic feature processing on naming for PWA, this study examined the ability of PWA to confirm the relations between object concepts and associated semantic features.

METHODS & PROCEDURES

15 native English-speaking, right handed individuals with post-stroke-aphasia responded yes or no via button press to feature-verification questions designed to probe the relationships between concept domain and feature type and distinctiveness.

OUTCOMES & RESULTS

PWA were more accurate and quicker to confirm concept-feature relationships when features contained function/action, rather than visual-perceptual, information about concepts and when features were distinctive to concepts rather than shared. The truthfulness (i.e., veracity) of concept-feature pairings was demonstrated to differentially affect living versus nonliving concepts. Within domain, only nonliving concepts were verified more accurately and more quickly when pairings were true (rather than false). Between domains, true nonliving concept-feature pairings were more accurately and more quickly verified than true living concept-feature pairings. Also with respect to veracity, correlations were observed between aphasia severity and accuracy and speed of response to false concept-feature pairings.

CONCLUSIONS

Findings have implications for the way in which semantic processing is probed with PWA, as well as providing preliminary information regarding responsivity of PWA to differing types of semantic information for living versus nonliving concepts. The fact that PWA demonstrated disproportionate difficulty responding to certain types of semantic information also suggests preliminary implications for the utility of different types of semantic cues in semantically-based treatments for word retrieval impairment.

Revisiting domain-general accounts of category specificity in mind and brain

Theories about the neural basis of semantic knowledge have been strongly influenced by reports that particular semantic categories can be differentially impaired by neuropathology and can differentially activate particular regions of cortex in brain imaging studies. One well-known interpretation of these data is that the brain has evolved distinct functional and anatomical modules for storing and retrieving knowledge about different kinds of things. We review the evidence supporting an alternative view: that category specificity arises from many heterogeneous factors and so tells us little directly about the cognitive and neural architecture of semantic memory. We consider four general hypotheses about domain-general causes of category-specific patterns, their roots in early work, and their reemergence in contemporary research. We argue that there is compelling evidence supporting each hypothesis, and that the different hypotheses together can explain most of the interesting data. We further suggest that such a multifactor domain-general approach to category specificity is appealing partly because it explains the important findings with reference to theoretical claims that are already widely accepted, and partly because it resolves several puzzles that arise under the alternative view. For further resources related to this article, please visit the WIREs website.

CONFLICT OF INTEREST

The authors have declared no conflicts of interest for this article.