Concepts and Semantic Memory
Media
Part of Concepts and Semantic Memory
- Title
- Concepts and Semantic Memory
- extracted text
-
Concepts and Semantic Memory
BARBARA C. MALT
Abstract
Humans accumulate vast amounts of knowledge over the life span. Much work
aimed at understanding this knowledge store has been called either concepts
or semantic memory research. This essay reviews early research on the nature of
concrete concepts (concepts of concrete objects) and their organization in memory.
It then raises considerations of abstract and relational concepts and of how action
affects representation and vice versa. Additional advances discussed come from
statistically based views of semantics, connectionist modeling, and neuroscientific
evidence, all showing how distributed sources of information can be integrated to
create semantic or conceptual content. Cross-cultural and cross-linguistic evidence
indicate, though, that models based on evidence from any one cultural or language
group may not apply well to others. The essay concludes by arguing that key issues
for future research include broadening the kinds of knowledge structures that are
studied and clarifying how language and nonlinguistic representations are related.
INTRODUCTION
The flexibility in how humans respond to their environment is unmatched by
any other species. This flexibility is due, in part, to the vast amount of knowledge that each person accumulates over the life span. Much work aimed
at understanding this mental database has been called either “concepts” or
“semantic memory” research.
The dual terminology raises the questions of what each term means and
what their relation is. Semantic memory is often said to be general world
knowledge, contrasted with autobiographical memory (memory for specific
events). Concepts are generally taken to be stable units of knowledge in
long-term memory that pick out meaningful sets of entities in the world
and that provide the elements out of which more complex thoughts are
constructed. Practically speaking, the concepts of interest are typically the
knowledge associated with words such as apple, bird, and chair. Given these
characterizations, concepts would be one kind of knowledge within a larger
system. That would leave much else to investigate about the nature and
contents of general world knowledge.
Emerging Trends in the Social and Behavioral Sciences. Edited by Robert Scott and Stephen Kosslyn.
© 2015 John Wiley & Sons, Inc. ISBN 978-1-118-90077-2.
1
2
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
In reality, though, researchers who do “semantic memory” research and
those who do “concepts” research have traditionally addressed similar
issues. Early on, those interested in semantic memory pursued a line of
work inspired by Collins and Quillian (1969), investigating the organization in memory and processing of knowledge associated with common
nouns such as those just mentioned. Meanwhile, “concepts” research
emerged from a line of work focusing on human acquisition of artificial (experimenter-generated) categories. With the advent of Eleanor
Rosch’s work (e.g., Rosch & Mervis, 1975; Rosch, Mervis, Gray, Johnson, &
Boyes-Braem, 1976) on natural categories (those named by common nouns)
and Smith and Medin’s (1981) book Concepts and categories, the modern era of
research on concepts that people acquire from the real world was launched.
The early convergence of the “semantic memory” and “concepts” endeavors is illustrated in work examining the relation in memory of representations
such as APPLE to FRUIT and the processes involved in making judgments
about them (such as answering Is an apple a fruit?) (e.g., Smith, Shoben, &
Rips, 1974). This research appears reliably in discussions of both concepts
and semantic memory. Indeed, to some extent, early “semantic memory”
researchers simply relabeled themselves as “concepts” researchers as work
by Rosch et al. came to the forefront (Rips, Smith, & Medin, 2012).
To the extent that a distinction persisted in the early decades, it might
be said that researchers associated with concepts were more interested in
the structure and content of individual concepts, and those associated with
semantic memory were more interested in how concepts were organized
in memory and how retrieval processes operated on them. However, the
boundary between the two research areas is fuzzy and not consistently
drawn. The following discussion does not attempt to label individual bodies
of research as one or the other. An important question we will return to,
though, is what remains to be understood about general world knowledge
beyond what currently is studied under either rubric.
FOUNDATIONAL RESEARCH
THE STRUCTURE AND CONTENT OF CONCEPTS
Rosch’s work in the 1970s was inspired in part by observations by the
philosopher Wittgenstein, who had provided an important analysis of the
meaning of common nouns such as game. Wittgenstein pointed out the
impossibility of finding a set of features that are necessary and sufficient for
some activity to be called a game. For instance, some games involve balls
but others do not, some have a winner and loser but some do not, and some
are not even fun although most may be. Wittgenstein argued that the things
called game have a family resemblance to each other: They overlap with one
Concepts and Semantic Memory
3
another in varied ways rather than sharing a single set of features. Rosch
and Mervis (1975) proposed that the family resemblance analysis applied to
many common nouns, and they provided evidence in the form of the feature
distributions people produced to exemplars of many common nouns.
There are several important corollaries of Rosch and Mervis’ analysis. One
is that the things called by a given name vary in their typicality with respect to
the name. The more an entity shares features with other things called by that
name, the more typical it is. Typicality is an important predictor of behavior
in many tasks, ranging from speed to verify an instance as having a name
(e.g., answering Is a robin a bird?) to how early children learn that relationship. Second, the most typical examples of categories serve as a prototype
associated with the name, and, as such, might provide a summary mental
representation of the category. Third, properties in the world occur in clusters. For instance, creatures with feathers tend to have wings and beaks, and
creatures with fur tend not to. These property clusters might provide natural
“break points” for perception. If so, natural categories are natural not only
in the sense of contrasting with artificial, experimenter-generated ones, but
in the sense that they are the result of a natural segmentation of the world
easily perceived by human observers.
Studies using artificial categories supported the idea that prototypes might
constitute the mental representation of sets of exemplars. However, models
assuming that people accumulate traces of individual exemplars of a category as their representation can account for the same results. These exemplar
models have received much less vetting for their ability to account for performance on tasks with real-world concepts, but some work suggests that
they are useful (e.g., Storms, 2004). Knowledge of instances does not preclude
also having a summary representation in the form of a prototype, however,
and other work has supported the possibility of people maintaining both
(Smith & Minda, 1998).
Later work elaborated on the contents of the mental representations.
Knowledge of BIRD cannot be only a set of features. The relation between
having wings and flying (causal) is understood as different from the relation
between having wings and eating seed (mere co-occurrence). Feature
knowledge is embedded in a rich set of intuitive theories about how the
world works (see Murphy, 2002, ch. 6). Theories and beliefs about causation
may be particularly important. For instance, features that are causes of other
features are seen as particularly central to category membership (Ahn &
Kim, 2001) Another overarching folk theory is that many categories have
an unseen, underlying essence shared by category members (e.g., Medin &
Ortony, 1989). Even if it is impossible in reality to find such essences, people
may still believe they exist and appeal to them in their judgments (Keil,
1989).
4
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
Given these enriched ideas about the content of concepts, the possibility
is also opened up that the contents will differ across domains (such as
naturally occurring things—water, gold, cats, and dogs—vs human-made
ones—chairs, tables, and pens and pencils). People tend to believe that
naturally occurring things have essences but artifacts do not (e.g., Malt,
1990). Furthermore, knowledge of living things and artifacts is observed
to be differentially affected by brain damage. Patients can have degraded
ability to understand one without harm to the other (see Yee, Chrysikou, &
Thompson-Schill, 2013). This suggests some domain-specificity in their
representation.
The early research was dominated by investigations of concepts having a
taxonomic or “kind” relation to one another (e.g., ROBIN is a kind of BIRD),
but an early departure from this tradition was made by Barsalou (1991). He
pointed out that people also use groupings that are created on the spot to
serve particular purposes. For instance, when faced with a fire in the house,
a person might construct a grouping of the things that should be rescued
first (including, say, pets, photos, and jewelry). What unites the things in this
category is their importance in the person’s life, not overlapping properties
such as fur or feathers.
ORGANIZATION OF CONCEPTS IN MEMORY AND RETRIEVAL FROM MEMORY
Several insights emerged from this era on how concepts are related to one
another and retrieved from memory. Rosch et al. (1976) noted that common
objects can be labeled at multiple levels of abstraction: rocking chair, chair,
or furniture, for instance. They observed that the middle level is the level at
which objects are most commonly named: chair rather than rocking chair or
furniture. This “basic” level is also the level first learned by children, and the
level at which objects are most quickly identified, suggesting that it is the
most salient and useful. Rosch et al. proposed that it is the level at which
things within a category are substantially similar while still being distinct
from things in contrasting categories.
Collins and Quillian (1969) suggested that concept hierarchies are embedded in a semantic network linking CHAIR not only to ROCKING CHAIR and
FURNITURE but also to BENCH and TABLE, and each of these to properties
such as WOOD and HAS LEGS. Collins and Loftus (1975) further suggested
that once one concept has been activated, activation spreads along the links
within the network, activating other nearby concepts. Given the network
structure and spreading activation, predictions can be made about how easy
it should be to answer questions such as Is a robin a bird? versus Is a robin
an animal?, as well as about whether activating a concept such as DOCTOR
should speed responses to others such as NURSE versus BUTTER. The work
Concepts and Semantic Memory
5
by Smith et al. (1974) mentioned earlier suggested the need for some refinements (such as allowing that ROBIN might be closer to BIRD than DUCK
based on typicality). However, the general notion that concepts are interconnected in a massive network of associations, and that retrieving some
information can activate related material, has remained in place.
CUTTING-EDGE RESEARCH
LINKING CONTENT TO ORGANIZATION AND PROCESS
Recent threads of research have broadened the range of concepts studied
and promoted new ways of thinking about the content of individual concepts. In doing so, ideas about the representation of individual concepts have
become more integrated with ones about processing and the relations among
concepts.
Attention to abstract concepts (e.g., FAITH, BEAUTY) has led to asking
whether representations of abstract and concrete words differ only in
content, or whether a crucial difference lies in their connections to other
concepts within the semantic network. For instance, abstract concepts may
have more emotion-related features and fewer sensorimotor ones (e.g.,
Kousta, Vigliocco, Vinson, Andrews, & Del Campo, 2011). However, abstract
ones may also tend to have connections dominated by associations (e.g.,
FAITH is connected to GOD and FIDELITY), while concrete concepts have
ones dominated by categorical relations (e.g., DOG is connected to CAT and
MAMMAL) (Crutch & Warrington, 2005). Separately, the notion of relational
knowledge has also become more prominent. In some cases, important
relations are captured by individual words. Gentner and Kurtz (2005), for
instance, describe the meanings of gift and predator as centering on relations
rather than properties intrinsic to the entity. In other cases, what is of interest
is thematic links among different entities within the semantic network, such
as the relations between a cow and milk or a boat and an anchor (e.g., Estes,
Golonka, & Jones, 2011), again blurring the line between representation and
organization.
Work on how concept use affects concept contents, and vice versa, more
explicitly links processing to representation. Concepts are shaped by input
conditions. The focus of learning (whether to learn to distinguish categories
per se or to learn about the properties that entities have) influences the
knowledge that is acquired about instances of a category. Other types of
interactions with instances do, too, as when people put knowledge to use.
For instance, making treatment decisions given symptoms (Markman &
Ross, 2003), affects disease category representations. The other side of the
coin is how knowledge is translated into action. Representation affects
6
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
how humans interact with each other. For instance, physician beliefs about
a child’s home, school, and peers contexts affects diagnosis of mental
disorders (De Los Reyes & Marsh, 2011).
Even more closely integrating action and representation is an approach to
the nature of concepts known as the embodiment perspective. Traditionally,
the conceptual system was said to contain amodal knowledge, meaning
that experiences taken in through the senses are converted to a more
abstract representational format. Counter to such assumptions, an explosion
of recent research has demonstrated that sensorimotor variables affect
performance in diverse tasks. For instance, verifying a property in the
auditory modality (e.g., BLENDER-loud) is slower after verifying a taste
property (e.g., CRANBERRIES-tart) than after verifying another auditory
property (e.g., LEAVES-rustling). Activating stored knowledge of a word is
said to entail mentally simulating the actual experience of encountering its
referent in the world. In this view, the representations themselves are rooted
in systems for perception, action, and emotion (see Barsalou, Simmons,
Barbey, & Wilson, 2003).
COMPUTATIONAL AND NEUROSCIENTIFIC ADVANCES: DISTRIBUTIONAL STATISTICS AND
DISTRIBUTED REPRESENTATIONS
New methodologies have brought further insights into both representation
and processing. In doing so, they have helped show how the boundaries
between these two components of cognition are blurred.
Computation-intensive distributional models of semantics assume that
words appearing in similar linguistic contexts have similar meanings, and
that meaning can be deduced from the distribution of uses. From massive
inputs of discourse in the form of electronic corpora, word meanings are
constructed using information about word co-occurrences in the data.
The models differ substantially in detail, but all have the advantage of
demonstrating how semantic representations might be acquired from environmental input. Although they do not necessarily account well for basic
patterns of word choice (Murphy, 2002, ch. 11), recent models integrating
feature-based representations with corpus-derived co-occurrence data may
help address such issues (see McRae & Jones, 2013.)
A different computational approach lies in connectionist modeling. While
the Collins and Quillian-type semantic network represented each concept
as a unitary node in memory, connectionism opened the possibility of
re-envisioning them in terms of patterns of activations across nodes that
individually have no conceptual content (e.g., Rogers & McClelland, 2004).
The same nodes participate in the representation of multiple concepts via different patterns of activation across them. The patterns of activation capture
the similarities among concepts, with concepts having more shared features
Concepts and Semantic Memory
7
producing more similar patterns of activation. Connectionist models can
model the acquisition of concept knowledge through gradual adaptation of
the connection weights between nodes, sharing with distributional models
the advantage of showing how representations may emerge from input.
They can also simulate the selective loss of knowledge through distortions of
weights, and they demonstrate how behavioral results (for instance, speed
to verify properties or class-inclusion relations) can be modeled even if the
association-based network as a literal map of how knowledge is laid out in
the brain is false.
The other major methodological advance has been in the area of cognitive
neuroscience. The most important technique has been functional neuroimaging looking at what brain regions are active in processing words and pictures
of objects. Results support the embodiment perspective on conceptual representation. Activating object concepts elicits patterns of neural activity in
regions close to, if not identical to, those involved in perceiving and interacting with the actual objects. The same is true for perceiving color and thinking
about the colors of objects, and for carrying out actions and thinking about
the meaning of action words. For instance, thinking about pianos and typewriters create similar patterns of activation in motor regions of the brain, and
reading a word strongly associated with a sound, such as telephone, activates
regions involved in auditory perception and processing (see Martin, 2007;
Yee et al., 2013). In short, stored knowledge about objects entails a great deal
of sensorimotor knowledge, and activating the knowledge engages relevant
sensorimotor regions of the brain.
Such evidence also highlights new possibilities for thinking about the overall organization of conceptual knowledge in the brain. If object knowledge is
represented in terms of sensorimotor and other attributes distributed across
different brain areas, it is less localized. A representation about telephones
would involve sound, but it would also involve purpose, shape, and several
types of motor activity (holding, dialing, and speaking), distributed across
different regions. Theories differ in specifics, but all are in agreement that
conceptual representations are distributed in some manner. A central hub
or convergence zone may integrate the information across modalities to create the experience of a unified thought about an entity (see Yee et al., 2013,
for review). By moving away from notions of unitary concept nodes and
localized representation, this perspective is broadly compatible with connectionist models of semantic knowledge.
Although studies of brain damage had shown that people can have
deficits for some types of concepts but not others, a difficulty for the idea
of domain-specific neural representation has been that the deficits do
not always respect the living thing versus artifact domain boundary. An
appealing aspect of the newer approach is that it can explain differential
8
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
degradation of knowledge in terms of the features involved. Although
artifact representations may tend to involve more function-related features
and living things more sensory attributes, there can be variation across
individual entities. This variation may account for variable loss within a
domain (see Yee et al., 2013).
ANGLOS ARE NOT PEOPLE
Yes, English-speaking people of European descent are people in an important sense. But when researchers talk about how “people” represent or use
knowledge, English speakers of European descent are not necessarily a good
gauge of “people” in general. Cognitive psychology spent its first decades
focused on this population. In many areas of psychology there is now interest
in the generalizability of findings and theories to other cultures (e.g., Henrich,
Heine, & Norenzayan, 2010). Under the umbrella of concepts and semantic memory, however, work on other cultural groups has been limited. Pioneering research comes from Douglas Medin’s group (e.g., Atran & Medin,
2008), which has examined cultural differences in intuitive theories about
nature and the human place within nature. This work demonstrates that differences can be pronounced and impact decisions and behaviors toward the
environment.
Other researchers have been motivated by observations about language.
It has become apparent that languages have different ways of dividing
up domains. For instance, English, Chinese, and Spanish group common
household containers differently by name. Such differences have been
documented for domains including color, number, plants and animals,
drinking vessels and household containers, body parts, spatial relations,
locomotion, acts of cutting and breaking, acts of carrying and holding (see
Malt & Majid, 2013, for review). These observations mean that when words
of one language (e.g., robin, chair) are used to identify concepts, the concepts
may be language-specific. Some researchers have begun to look for shared
components of word meaning underlying cross-linguistic variation (see
Malt & Majid, 2013) as a way to identify aspects of understanding of the
world that are language-general. Work on animal cognition and concepts
(e.g., Phillips & Santos, 2007) also helps identify aspects of conceptual
thought that occur across species and so are not language-dependent.
KEY ISSUES FOR FUTURE RESEARCH
THE REST OF SEMANTIC MEMORY
Semantic memory as “general world knowledge” in principle encompasses
much more than units of knowledge labeled by single words. Existing
Concepts and Semantic Memory
9
research on concepts and semantic memory by no means provides a
complete picture of general world knowledge.
Some decades ago, researchers identified other kinds of knowledge structures that are important in guiding thought and behavior. For instance, work
on “scripts” (e.g., Schank & Abelson, 1977) illustrated how people use knowledge of standard event sequences (e.g., having a restaurant meal) to interpret
events and select behaviors. Other larger scale knowledge structures such as
culture-specific formulas for telling a story (e.g., Mandler & Johnson, 1977)
and the “mental models” that people have of complex phenomena such as
the solar system and electricity (Gentner & Stevens, 1983) also are critical to
how people understand the world and act on it.
More recent related work does exist, such as that by Medin and colleagues
on how different cultural groups understand the natural world. Also in a
related vein is work on intuitive theories, explanatory schemes, and causal
understanding (e.g., Keil, 2006). An intriguing observation in some of this
work is how fragmentary the knowledge of some domains can be despite a
person’s sense that they understand the domain, and how inconsistent people can be in the judgments they make and conclusions they draw, depending
on what elements of knowledge are retrieved in a given context (Shtulman
& Valcarcel, 2012). Understanding these forms of knowledge and how they
are used is essential to addressing societal problems such as threats to the
environment and achieving success in education.
None of these threads of research is well integrated with work labeled
“semantic memory.” The challenge will be to work out how knowledge
at different scales and levels of complexity are related to each other and
interact in determining thought and behavior. This integration is necessary
for an understanding of general world knowledge.
THE LANGUAGE–CONCEPTS INTERFACE
A persistent haze has lingered over research in concepts and semantic memory about what exactly is the topic of investigation. Is it general-purpose
mental representations of the world, or is it words and their meanings as
part of the linguistic system? Some researchers talk about concepts, some
about word meanings, and some use the terms interchangeably. The confusion over concepts versus word meanings also applies to the study of “conceptual combination,” which is said to be about how simple concepts are
combined to create more complex ones (see Murphy, 2002, ch. 12), but which
in practice is about how noun phrases (such as chocolate bee or ocean bird) are
interpreted. Meanwhile, “polysemy” or the multiple senses associated with
a single word is routinely treated by cognitive psychologists as a feature of
language, although the units involved are presumably exactly the same as
those called concepts in other contexts.
10
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
A commitment to the idea that lexicalized concepts are the ones used in
general nonlinguistic cognitive tasks is in line with the Whorfian hypothesis
that language strongly shapes an individual’s understanding of the world.
However, much recent research on the Whorfian hypothesis favors weaker
interpretations of linguistic effects on thought, such as that they are momentary, online effects (e.g., Athanasopoulos & Bylund, 2013). Figuring out how
nonlinguistic and linguistic knowledge interface and interact will be an
important task for the future. Answers may help with gaining traction on
“conceptual combination,” which remains without a satisfactory account.
Similarly, clarifying when deficits due to brain damage reflect disruption
to conceptual knowledge, language processing, or the interface between
the two may help with models to account for such deficits. Traditionally
separate literatures on language production and on bilingual word learning
and lexical access also stand to benefit from clarifying the language–thought
interface.
One possible resolution is that no fixed concepts or word meanings exist. If
conceptual knowledge is stored in a distributed manner across the brain and,
metaphorically, in the mind, stimuli or tasks can selectively recruit portions
of this knowledge. Some stimuli and tasks may activate linguistic knowledge and some may not. Online construction of concepts has been proposed
in the past (e.g., Barsalou, 1987). Online construction of word meaning also
has been proposed before and helps solve the problems of how senses vary
with context and how a single word can have indefinitely many senses (e.g.,
Clark, 1983). Most recently, online construction of word meaning has been
suggested in the context of distributional models of semantics (Kintsch &
Mangalath, 2011). Proposals of this sort have been the minority view to date.
In light of the recent advances in modeling and neuroscience, they may be
poised to become more dominant.
ACKNOWLEDGMENTS
Preparation of this essay was supported in part by National Science Foundation Grant Number 1057885 to Barbara Malt and Ping Li. I thank Gordon
Bower and Doug Medin for discussion and Jessecae Marsh and Gregory Murphy for comments on a previous draft.
REFERENCES
Ahn, W.-k., & Kim, N. S. (2001). The causal status effect in categorization: An
overview. In D. L. Medin (Ed.), The psychology of learning and motivation: Advances
in research and theory (Vol. 40, pp. 23–65). San Diego, CA: Academic Press.
Concepts and Semantic Memory
11
Athanasopoulos, P., & Bylund, E. (2013). Does grammatical aspect affect motion
event cognition? A cross-linguistic comparison of English and Swedish speakers.
Cognitive Science, 37, 286–309.
Atran, S., & Medin, D. L. (2008). The native mind and the cultural construction of nature.
Boston, MA: MIT Press.
Barsalou, L. W. (1987). The instability of graded structure: Implications for the nature
of concepts. In U. Neisser (Ed.), Concepts and conceptual development: Ecological and
intellectual factors in categorization (pp. 101–140). Cambridge, England: Cambridge
University Press.
Barsalou, L. W. (1991). Deriving categories to achieve goals. In G. H. Bower (Ed.),
The psychology of learning and motivation: Advances in research and theory (Vol. 27, pp.
1–64). San Diego, CA: Academic Press.
Barsalou, L. W., Simmons, W. K., Barbey, A. K., & Wilson, C. D. (2003). Grounding
conceptual knowledge in modality-specific systems. Trends in Cognitive Sciences, 7,
84–91.
Clark, H. H. (1983). Making sense of nonce sense. In G. B. Flores d’Arcais & R. J.
Jarvella (Eds.), The process of language understanding (pp. 297–331). London, England: Wiley.
Collins, A. M., & Loftus, E. F. (1975). A spreading activation theory of semantic processing. Psychological Review, 82, 407–428.
Collins, A. M., & Quillian, M. R. (1969). Retrieval time from semantic memory. Journal
of Verbal Learning and Verbal Behavior, 8, 241–248.
Crutch, S., & Warrington, E. K. (2005). Abstract and concrete concepts have structurally different representational frameworks. Brain, 128, 615–627.
De Los Reyes, A., & Marsh, J. K. (2011). Patients’ contexts and their effects on clinicians’ impressions of conduct disorder symptoms. Journal of Clinical Child and
Adolescent Psychology, 40, 479–485.
Estes, Z., Golonka, S., & Jones, L. L. (2011). Thematic thinking: The apprehension
and consequences of thematic relations. In B. Ross (Ed.), Psychology of learning and
motivation (Vol. 54, pp. 249–294). Burlington, Canada: Academic Press.
Gentner, D., & Kurtz, K. J. (2005). Relational categories. In W.-k. Ahn, R. L. Goldstone, B. C. Love, A. B. Markman & P. Wolff (Eds.), Categorization inside and outside
the laboratory: Essays in honor of Douglas L. Medin (pp. 151–175). Washington, DC:
American Psychological Association.
Gentner, D., & Stevens, A. L. (Eds.) (1983). Mental models. Hillsdale, NJ: Lawrence
Erlbaum Associates.
Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world?
Behavioral and Brain Sciences, 33, 1–23.
Keil, F. C. (1989). Concepts, kinds, and cognitive development. Cambridge, MA: MIT
Press.
Keil, F. C. (2006). Explanation and understanding. Annual Review of Psychology, 57,
227–254.
Kintsch, W., & Mangalath, P. (2011). The construction of meaning. Topics in Cognitive
Science, 3, 346–370.
12
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
Kousta, S.-T., Vigliocco, G., Vinson, D. P., Andrews, M., & Del Campo, E. (2011). The
representation of abstract words: Why emotion matters. Journal of Experimental
Psychology: General, 140, 14–34.
Malt, B. C. (1990). Features and beliefs in the mental representation of categories.
Journal of Memory and Language, 29, 289–315.
Malt, B. C., & Majid, A. (2013). How thought is mapped into words. WIREs Cognitive
Science, 4, 583–597.
Mandler, J. M., & Johnson, N. S. (1977). Remembrance of things parsed: Story structure and recall. Cognitive Psychology, 9, 111–151.
Markman, A. B., & Ross, B. H. (2003). Category use and category learning. Psychological Bulletin, 129, 592–613.
Martin, A. (2007). The representation of object concepts in the brain. Annual Review
of Psychology, 58, 25–45.
McRae, K., & Jones, M. N. (2013). Semantic memory. In D. Reisberg (Ed.), The Oxford
handbook of cognitive psychology (pp. 206–219). Oxford, England: Oxford University
Press.
Medin, D. L., & Ortony, A. (1989). Psychological essentialism. In S. Vosniadou & A.
Ortony (Eds.), Similarity and analogical reasoning (pp. 179–195). Cambridge, England: Cambridge University Press.
Murphy, G. L. (2002). The big book of concepts. Cambridge, MA: The MIT Press.
Phillips, W., & Santos, L. R. (2007). Evidence for kind representations in the absence
of language: Experiments with rhesus monkeys (Macaca mulatta). Cognition, 102,
455–463.
Rips, L. J., Smith, E. E., & Medin, D. L. (2012). Concepts and categories: Memory,
meaning, and metaphysics. In K. J. Holyoak & R. G. Morrison (Eds.), The Oxford
handbook of thinking and reasoning (pp. 177–209). Oxford, England: Oxford University Press.
Rogers, T. T., & McClelland, J. L. (2004). Semantic cognition: A parallel distributed processing approach. Cambridge, MA: MIT Press.
Rosch, E., & Mervis, C. B. (1975). Family resemblances: Studies in internal structure
of categories. Cognitive Psychology, 7, 573–605.
Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D., & Boyes-Braem, P. (1976). Basic
objects in natural categories. Cognitive Psychology, 8, 382–439.
Schank, R. C., & Abelson, R. P. (1977). Scripts, plans, goals, and understanding: An
inquiry into human knowledge structures. Hillsdale, NJ: Erlbaum.
Shtulman, A., & Valcarcel, J. (2012). Scientific knowledge suppresses but does not
supplant earlier intuitions. Cognition, 124, 209–215.
Smith, E. E., & Medin, D. L. (1981). Categories and concepts. Cambridge, MA: Harvard
University Press.
Smith, E. E., Shoben, E. J., & Rips, L. J. (1974). Structure and process in semantic memory: A featural model for semantic decisions. Psychological Review, 81, 214–241.
Smith, J. D., & Minda, J. P. (1998). Prototypes in the mist: The early epochs of category
learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24,
1411–1436.
Concepts and Semantic Memory
13
Storms, G. (2004). Exemplar models in the study of natural language concepts. Psychology of Learning and Motivation, 45, 1–39.
Yee, E., Chrysikou, E. G., & Thompson-Schill, S. L. (2013). Semantic memory. In K.
Ochsner & S. Kosslyn (Eds.), The Oxford handbook of cognitive neuroscience, Vol. 1:
Core topics (pp. 353–374). Oxford, England: Oxford University Press.
FURTHER READING
Atran, S., & Medin, D. L. (2008). The native mind and the cultural construction of nature.
Boston, MA: MIT Press.
Barsalou, L. B. (2008). Grounded cognition. Annual Review of Psychology, 59, 617–645.
Keil, F. C. (2006). Explanation and understanding. Annual Review of Psychology, 57,
227–254.
Malt, B. C., & Majid, A. (2013). How thought is mapped into words. WIREs Cognitive
Science, 4, 583–597.
McRae, K., & Jones, M. N. (2013). Semantic memory. In D. Reisberg (Ed.), The Oxford
handbook of cognitive psychology (pp. 206–219). Oxford, England: Oxford University
Press.
Murphy, G. L. (2002). The big book of concepts. Cambridge, MA: The MIT Press.
Rogers, T. T., & McClelland, J. L. (2004). Semantic cognition: A parallel distributed processing approach. Cambridge, MA: MIT Press.
Yee, E., Chrysikou, E. G., & Thompson-Schill, S. L. (2013). Semantic memory. In K.
Ochsner & S. Kosslyn (Eds.), The Oxford handbook of cognitive neuroscience, Vol. 1:
Core topics (pp. 353–374). Oxford, England: Oxford University Press.
BARBARA C. MALT SHORT BIOGRAPHY
Barbara C. Malt is a Professor of Psychology at Lehigh University. Her
research focuses on thought, language, and the relation between the two. She
is especially interested in how objects and actions are mentally represented,
how people (including monolinguals and bilinguals, both children and
adults) talk about these objects and actions using the tools available in their
language(s), and what influence, if any, the different ways of talking have
on nonlinguistic representations. She is an associate editor for the Journal of
Experimental Psychology: Learning, Memory, and Cognition. More information
can be found at http://www.lehigh.edu/∼bcm0/bcm0/index.html
RELATED ESSAYS
Language, Perspective, and Memory (Psychology), Rachel A. Ryskin et al.
Misinformation and How to Correct It (Psychology), John Cook et al.
Language and Thought (Psychology), Susan Goldin-Meadow
Social Aspects of Memory (Psychology), William Hirst and Charles B. Stone
14
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
Implicit Memory (Psychology), Dawn M. McBride
Memory Gaps and Memory Errors (Psychology), Jeffrey S. Neuschatz et al.
Theory of Mind (Psychology), Henry Wellman
-
Concepts and Semantic Memory
BARBARA C. MALT
Abstract
Humans accumulate vast amounts of knowledge over the life span. Much work
aimed at understanding this knowledge store has been called either concepts
or semantic memory research. This essay reviews early research on the nature of
concrete concepts (concepts of concrete objects) and their organization in memory.
It then raises considerations of abstract and relational concepts and of how action
affects representation and vice versa. Additional advances discussed come from
statistically based views of semantics, connectionist modeling, and neuroscientific
evidence, all showing how distributed sources of information can be integrated to
create semantic or conceptual content. Cross-cultural and cross-linguistic evidence
indicate, though, that models based on evidence from any one cultural or language
group may not apply well to others. The essay concludes by arguing that key issues
for future research include broadening the kinds of knowledge structures that are
studied and clarifying how language and nonlinguistic representations are related.
INTRODUCTION
The flexibility in how humans respond to their environment is unmatched by
any other species. This flexibility is due, in part, to the vast amount of knowledge that each person accumulates over the life span. Much work aimed
at understanding this mental database has been called either “concepts” or
“semantic memory” research.
The dual terminology raises the questions of what each term means and
what their relation is. Semantic memory is often said to be general world
knowledge, contrasted with autobiographical memory (memory for specific
events). Concepts are generally taken to be stable units of knowledge in
long-term memory that pick out meaningful sets of entities in the world
and that provide the elements out of which more complex thoughts are
constructed. Practically speaking, the concepts of interest are typically the
knowledge associated with words such as apple, bird, and chair. Given these
characterizations, concepts would be one kind of knowledge within a larger
system. That would leave much else to investigate about the nature and
contents of general world knowledge.
Emerging Trends in the Social and Behavioral Sciences. Edited by Robert Scott and Stephen Kosslyn.
© 2015 John Wiley & Sons, Inc. ISBN 978-1-118-90077-2.
1
2
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
In reality, though, researchers who do “semantic memory” research and
those who do “concepts” research have traditionally addressed similar
issues. Early on, those interested in semantic memory pursued a line of
work inspired by Collins and Quillian (1969), investigating the organization in memory and processing of knowledge associated with common
nouns such as those just mentioned. Meanwhile, “concepts” research
emerged from a line of work focusing on human acquisition of artificial (experimenter-generated) categories. With the advent of Eleanor
Rosch’s work (e.g., Rosch & Mervis, 1975; Rosch, Mervis, Gray, Johnson, &
Boyes-Braem, 1976) on natural categories (those named by common nouns)
and Smith and Medin’s (1981) book Concepts and categories, the modern era of
research on concepts that people acquire from the real world was launched.
The early convergence of the “semantic memory” and “concepts” endeavors is illustrated in work examining the relation in memory of representations
such as APPLE to FRUIT and the processes involved in making judgments
about them (such as answering Is an apple a fruit?) (e.g., Smith, Shoben, &
Rips, 1974). This research appears reliably in discussions of both concepts
and semantic memory. Indeed, to some extent, early “semantic memory”
researchers simply relabeled themselves as “concepts” researchers as work
by Rosch et al. came to the forefront (Rips, Smith, & Medin, 2012).
To the extent that a distinction persisted in the early decades, it might
be said that researchers associated with concepts were more interested in
the structure and content of individual concepts, and those associated with
semantic memory were more interested in how concepts were organized
in memory and how retrieval processes operated on them. However, the
boundary between the two research areas is fuzzy and not consistently
drawn. The following discussion does not attempt to label individual bodies
of research as one or the other. An important question we will return to,
though, is what remains to be understood about general world knowledge
beyond what currently is studied under either rubric.
FOUNDATIONAL RESEARCH
THE STRUCTURE AND CONTENT OF CONCEPTS
Rosch’s work in the 1970s was inspired in part by observations by the
philosopher Wittgenstein, who had provided an important analysis of the
meaning of common nouns such as game. Wittgenstein pointed out the
impossibility of finding a set of features that are necessary and sufficient for
some activity to be called a game. For instance, some games involve balls
but others do not, some have a winner and loser but some do not, and some
are not even fun although most may be. Wittgenstein argued that the things
called game have a family resemblance to each other: They overlap with one
Concepts and Semantic Memory
3
another in varied ways rather than sharing a single set of features. Rosch
and Mervis (1975) proposed that the family resemblance analysis applied to
many common nouns, and they provided evidence in the form of the feature
distributions people produced to exemplars of many common nouns.
There are several important corollaries of Rosch and Mervis’ analysis. One
is that the things called by a given name vary in their typicality with respect to
the name. The more an entity shares features with other things called by that
name, the more typical it is. Typicality is an important predictor of behavior
in many tasks, ranging from speed to verify an instance as having a name
(e.g., answering Is a robin a bird?) to how early children learn that relationship. Second, the most typical examples of categories serve as a prototype
associated with the name, and, as such, might provide a summary mental
representation of the category. Third, properties in the world occur in clusters. For instance, creatures with feathers tend to have wings and beaks, and
creatures with fur tend not to. These property clusters might provide natural
“break points” for perception. If so, natural categories are natural not only
in the sense of contrasting with artificial, experimenter-generated ones, but
in the sense that they are the result of a natural segmentation of the world
easily perceived by human observers.
Studies using artificial categories supported the idea that prototypes might
constitute the mental representation of sets of exemplars. However, models
assuming that people accumulate traces of individual exemplars of a category as their representation can account for the same results. These exemplar
models have received much less vetting for their ability to account for performance on tasks with real-world concepts, but some work suggests that
they are useful (e.g., Storms, 2004). Knowledge of instances does not preclude
also having a summary representation in the form of a prototype, however,
and other work has supported the possibility of people maintaining both
(Smith & Minda, 1998).
Later work elaborated on the contents of the mental representations.
Knowledge of BIRD cannot be only a set of features. The relation between
having wings and flying (causal) is understood as different from the relation
between having wings and eating seed (mere co-occurrence). Feature
knowledge is embedded in a rich set of intuitive theories about how the
world works (see Murphy, 2002, ch. 6). Theories and beliefs about causation
may be particularly important. For instance, features that are causes of other
features are seen as particularly central to category membership (Ahn &
Kim, 2001) Another overarching folk theory is that many categories have
an unseen, underlying essence shared by category members (e.g., Medin &
Ortony, 1989). Even if it is impossible in reality to find such essences, people
may still believe they exist and appeal to them in their judgments (Keil,
1989).
4
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
Given these enriched ideas about the content of concepts, the possibility
is also opened up that the contents will differ across domains (such as
naturally occurring things—water, gold, cats, and dogs—vs human-made
ones—chairs, tables, and pens and pencils). People tend to believe that
naturally occurring things have essences but artifacts do not (e.g., Malt,
1990). Furthermore, knowledge of living things and artifacts is observed
to be differentially affected by brain damage. Patients can have degraded
ability to understand one without harm to the other (see Yee, Chrysikou, &
Thompson-Schill, 2013). This suggests some domain-specificity in their
representation.
The early research was dominated by investigations of concepts having a
taxonomic or “kind” relation to one another (e.g., ROBIN is a kind of BIRD),
but an early departure from this tradition was made by Barsalou (1991). He
pointed out that people also use groupings that are created on the spot to
serve particular purposes. For instance, when faced with a fire in the house,
a person might construct a grouping of the things that should be rescued
first (including, say, pets, photos, and jewelry). What unites the things in this
category is their importance in the person’s life, not overlapping properties
such as fur or feathers.
ORGANIZATION OF CONCEPTS IN MEMORY AND RETRIEVAL FROM MEMORY
Several insights emerged from this era on how concepts are related to one
another and retrieved from memory. Rosch et al. (1976) noted that common
objects can be labeled at multiple levels of abstraction: rocking chair, chair,
or furniture, for instance. They observed that the middle level is the level at
which objects are most commonly named: chair rather than rocking chair or
furniture. This “basic” level is also the level first learned by children, and the
level at which objects are most quickly identified, suggesting that it is the
most salient and useful. Rosch et al. proposed that it is the level at which
things within a category are substantially similar while still being distinct
from things in contrasting categories.
Collins and Quillian (1969) suggested that concept hierarchies are embedded in a semantic network linking CHAIR not only to ROCKING CHAIR and
FURNITURE but also to BENCH and TABLE, and each of these to properties
such as WOOD and HAS LEGS. Collins and Loftus (1975) further suggested
that once one concept has been activated, activation spreads along the links
within the network, activating other nearby concepts. Given the network
structure and spreading activation, predictions can be made about how easy
it should be to answer questions such as Is a robin a bird? versus Is a robin
an animal?, as well as about whether activating a concept such as DOCTOR
should speed responses to others such as NURSE versus BUTTER. The work
Concepts and Semantic Memory
5
by Smith et al. (1974) mentioned earlier suggested the need for some refinements (such as allowing that ROBIN might be closer to BIRD than DUCK
based on typicality). However, the general notion that concepts are interconnected in a massive network of associations, and that retrieving some
information can activate related material, has remained in place.
CUTTING-EDGE RESEARCH
LINKING CONTENT TO ORGANIZATION AND PROCESS
Recent threads of research have broadened the range of concepts studied
and promoted new ways of thinking about the content of individual concepts. In doing so, ideas about the representation of individual concepts have
become more integrated with ones about processing and the relations among
concepts.
Attention to abstract concepts (e.g., FAITH, BEAUTY) has led to asking
whether representations of abstract and concrete words differ only in
content, or whether a crucial difference lies in their connections to other
concepts within the semantic network. For instance, abstract concepts may
have more emotion-related features and fewer sensorimotor ones (e.g.,
Kousta, Vigliocco, Vinson, Andrews, & Del Campo, 2011). However, abstract
ones may also tend to have connections dominated by associations (e.g.,
FAITH is connected to GOD and FIDELITY), while concrete concepts have
ones dominated by categorical relations (e.g., DOG is connected to CAT and
MAMMAL) (Crutch & Warrington, 2005). Separately, the notion of relational
knowledge has also become more prominent. In some cases, important
relations are captured by individual words. Gentner and Kurtz (2005), for
instance, describe the meanings of gift and predator as centering on relations
rather than properties intrinsic to the entity. In other cases, what is of interest
is thematic links among different entities within the semantic network, such
as the relations between a cow and milk or a boat and an anchor (e.g., Estes,
Golonka, & Jones, 2011), again blurring the line between representation and
organization.
Work on how concept use affects concept contents, and vice versa, more
explicitly links processing to representation. Concepts are shaped by input
conditions. The focus of learning (whether to learn to distinguish categories
per se or to learn about the properties that entities have) influences the
knowledge that is acquired about instances of a category. Other types of
interactions with instances do, too, as when people put knowledge to use.
For instance, making treatment decisions given symptoms (Markman &
Ross, 2003), affects disease category representations. The other side of the
coin is how knowledge is translated into action. Representation affects
6
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
how humans interact with each other. For instance, physician beliefs about
a child’s home, school, and peers contexts affects diagnosis of mental
disorders (De Los Reyes & Marsh, 2011).
Even more closely integrating action and representation is an approach to
the nature of concepts known as the embodiment perspective. Traditionally,
the conceptual system was said to contain amodal knowledge, meaning
that experiences taken in through the senses are converted to a more
abstract representational format. Counter to such assumptions, an explosion
of recent research has demonstrated that sensorimotor variables affect
performance in diverse tasks. For instance, verifying a property in the
auditory modality (e.g., BLENDER-loud) is slower after verifying a taste
property (e.g., CRANBERRIES-tart) than after verifying another auditory
property (e.g., LEAVES-rustling). Activating stored knowledge of a word is
said to entail mentally simulating the actual experience of encountering its
referent in the world. In this view, the representations themselves are rooted
in systems for perception, action, and emotion (see Barsalou, Simmons,
Barbey, & Wilson, 2003).
COMPUTATIONAL AND NEUROSCIENTIFIC ADVANCES: DISTRIBUTIONAL STATISTICS AND
DISTRIBUTED REPRESENTATIONS
New methodologies have brought further insights into both representation
and processing. In doing so, they have helped show how the boundaries
between these two components of cognition are blurred.
Computation-intensive distributional models of semantics assume that
words appearing in similar linguistic contexts have similar meanings, and
that meaning can be deduced from the distribution of uses. From massive
inputs of discourse in the form of electronic corpora, word meanings are
constructed using information about word co-occurrences in the data.
The models differ substantially in detail, but all have the advantage of
demonstrating how semantic representations might be acquired from environmental input. Although they do not necessarily account well for basic
patterns of word choice (Murphy, 2002, ch. 11), recent models integrating
feature-based representations with corpus-derived co-occurrence data may
help address such issues (see McRae & Jones, 2013.)
A different computational approach lies in connectionist modeling. While
the Collins and Quillian-type semantic network represented each concept
as a unitary node in memory, connectionism opened the possibility of
re-envisioning them in terms of patterns of activations across nodes that
individually have no conceptual content (e.g., Rogers & McClelland, 2004).
The same nodes participate in the representation of multiple concepts via different patterns of activation across them. The patterns of activation capture
the similarities among concepts, with concepts having more shared features
Concepts and Semantic Memory
7
producing more similar patterns of activation. Connectionist models can
model the acquisition of concept knowledge through gradual adaptation of
the connection weights between nodes, sharing with distributional models
the advantage of showing how representations may emerge from input.
They can also simulate the selective loss of knowledge through distortions of
weights, and they demonstrate how behavioral results (for instance, speed
to verify properties or class-inclusion relations) can be modeled even if the
association-based network as a literal map of how knowledge is laid out in
the brain is false.
The other major methodological advance has been in the area of cognitive
neuroscience. The most important technique has been functional neuroimaging looking at what brain regions are active in processing words and pictures
of objects. Results support the embodiment perspective on conceptual representation. Activating object concepts elicits patterns of neural activity in
regions close to, if not identical to, those involved in perceiving and interacting with the actual objects. The same is true for perceiving color and thinking
about the colors of objects, and for carrying out actions and thinking about
the meaning of action words. For instance, thinking about pianos and typewriters create similar patterns of activation in motor regions of the brain, and
reading a word strongly associated with a sound, such as telephone, activates
regions involved in auditory perception and processing (see Martin, 2007;
Yee et al., 2013). In short, stored knowledge about objects entails a great deal
of sensorimotor knowledge, and activating the knowledge engages relevant
sensorimotor regions of the brain.
Such evidence also highlights new possibilities for thinking about the overall organization of conceptual knowledge in the brain. If object knowledge is
represented in terms of sensorimotor and other attributes distributed across
different brain areas, it is less localized. A representation about telephones
would involve sound, but it would also involve purpose, shape, and several
types of motor activity (holding, dialing, and speaking), distributed across
different regions. Theories differ in specifics, but all are in agreement that
conceptual representations are distributed in some manner. A central hub
or convergence zone may integrate the information across modalities to create the experience of a unified thought about an entity (see Yee et al., 2013,
for review). By moving away from notions of unitary concept nodes and
localized representation, this perspective is broadly compatible with connectionist models of semantic knowledge.
Although studies of brain damage had shown that people can have
deficits for some types of concepts but not others, a difficulty for the idea
of domain-specific neural representation has been that the deficits do
not always respect the living thing versus artifact domain boundary. An
appealing aspect of the newer approach is that it can explain differential
8
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
degradation of knowledge in terms of the features involved. Although
artifact representations may tend to involve more function-related features
and living things more sensory attributes, there can be variation across
individual entities. This variation may account for variable loss within a
domain (see Yee et al., 2013).
ANGLOS ARE NOT PEOPLE
Yes, English-speaking people of European descent are people in an important sense. But when researchers talk about how “people” represent or use
knowledge, English speakers of European descent are not necessarily a good
gauge of “people” in general. Cognitive psychology spent its first decades
focused on this population. In many areas of psychology there is now interest
in the generalizability of findings and theories to other cultures (e.g., Henrich,
Heine, & Norenzayan, 2010). Under the umbrella of concepts and semantic memory, however, work on other cultural groups has been limited. Pioneering research comes from Douglas Medin’s group (e.g., Atran & Medin,
2008), which has examined cultural differences in intuitive theories about
nature and the human place within nature. This work demonstrates that differences can be pronounced and impact decisions and behaviors toward the
environment.
Other researchers have been motivated by observations about language.
It has become apparent that languages have different ways of dividing
up domains. For instance, English, Chinese, and Spanish group common
household containers differently by name. Such differences have been
documented for domains including color, number, plants and animals,
drinking vessels and household containers, body parts, spatial relations,
locomotion, acts of cutting and breaking, acts of carrying and holding (see
Malt & Majid, 2013, for review). These observations mean that when words
of one language (e.g., robin, chair) are used to identify concepts, the concepts
may be language-specific. Some researchers have begun to look for shared
components of word meaning underlying cross-linguistic variation (see
Malt & Majid, 2013) as a way to identify aspects of understanding of the
world that are language-general. Work on animal cognition and concepts
(e.g., Phillips & Santos, 2007) also helps identify aspects of conceptual
thought that occur across species and so are not language-dependent.
KEY ISSUES FOR FUTURE RESEARCH
THE REST OF SEMANTIC MEMORY
Semantic memory as “general world knowledge” in principle encompasses
much more than units of knowledge labeled by single words. Existing
Concepts and Semantic Memory
9
research on concepts and semantic memory by no means provides a
complete picture of general world knowledge.
Some decades ago, researchers identified other kinds of knowledge structures that are important in guiding thought and behavior. For instance, work
on “scripts” (e.g., Schank & Abelson, 1977) illustrated how people use knowledge of standard event sequences (e.g., having a restaurant meal) to interpret
events and select behaviors. Other larger scale knowledge structures such as
culture-specific formulas for telling a story (e.g., Mandler & Johnson, 1977)
and the “mental models” that people have of complex phenomena such as
the solar system and electricity (Gentner & Stevens, 1983) also are critical to
how people understand the world and act on it.
More recent related work does exist, such as that by Medin and colleagues
on how different cultural groups understand the natural world. Also in a
related vein is work on intuitive theories, explanatory schemes, and causal
understanding (e.g., Keil, 2006). An intriguing observation in some of this
work is how fragmentary the knowledge of some domains can be despite a
person’s sense that they understand the domain, and how inconsistent people can be in the judgments they make and conclusions they draw, depending
on what elements of knowledge are retrieved in a given context (Shtulman
& Valcarcel, 2012). Understanding these forms of knowledge and how they
are used is essential to addressing societal problems such as threats to the
environment and achieving success in education.
None of these threads of research is well integrated with work labeled
“semantic memory.” The challenge will be to work out how knowledge
at different scales and levels of complexity are related to each other and
interact in determining thought and behavior. This integration is necessary
for an understanding of general world knowledge.
THE LANGUAGE–CONCEPTS INTERFACE
A persistent haze has lingered over research in concepts and semantic memory about what exactly is the topic of investigation. Is it general-purpose
mental representations of the world, or is it words and their meanings as
part of the linguistic system? Some researchers talk about concepts, some
about word meanings, and some use the terms interchangeably. The confusion over concepts versus word meanings also applies to the study of “conceptual combination,” which is said to be about how simple concepts are
combined to create more complex ones (see Murphy, 2002, ch. 12), but which
in practice is about how noun phrases (such as chocolate bee or ocean bird) are
interpreted. Meanwhile, “polysemy” or the multiple senses associated with
a single word is routinely treated by cognitive psychologists as a feature of
language, although the units involved are presumably exactly the same as
those called concepts in other contexts.
10
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
A commitment to the idea that lexicalized concepts are the ones used in
general nonlinguistic cognitive tasks is in line with the Whorfian hypothesis
that language strongly shapes an individual’s understanding of the world.
However, much recent research on the Whorfian hypothesis favors weaker
interpretations of linguistic effects on thought, such as that they are momentary, online effects (e.g., Athanasopoulos & Bylund, 2013). Figuring out how
nonlinguistic and linguistic knowledge interface and interact will be an
important task for the future. Answers may help with gaining traction on
“conceptual combination,” which remains without a satisfactory account.
Similarly, clarifying when deficits due to brain damage reflect disruption
to conceptual knowledge, language processing, or the interface between
the two may help with models to account for such deficits. Traditionally
separate literatures on language production and on bilingual word learning
and lexical access also stand to benefit from clarifying the language–thought
interface.
One possible resolution is that no fixed concepts or word meanings exist. If
conceptual knowledge is stored in a distributed manner across the brain and,
metaphorically, in the mind, stimuli or tasks can selectively recruit portions
of this knowledge. Some stimuli and tasks may activate linguistic knowledge and some may not. Online construction of concepts has been proposed
in the past (e.g., Barsalou, 1987). Online construction of word meaning also
has been proposed before and helps solve the problems of how senses vary
with context and how a single word can have indefinitely many senses (e.g.,
Clark, 1983). Most recently, online construction of word meaning has been
suggested in the context of distributional models of semantics (Kintsch &
Mangalath, 2011). Proposals of this sort have been the minority view to date.
In light of the recent advances in modeling and neuroscience, they may be
poised to become more dominant.
ACKNOWLEDGMENTS
Preparation of this essay was supported in part by National Science Foundation Grant Number 1057885 to Barbara Malt and Ping Li. I thank Gordon
Bower and Doug Medin for discussion and Jessecae Marsh and Gregory Murphy for comments on a previous draft.
REFERENCES
Ahn, W.-k., & Kim, N. S. (2001). The causal status effect in categorization: An
overview. In D. L. Medin (Ed.), The psychology of learning and motivation: Advances
in research and theory (Vol. 40, pp. 23–65). San Diego, CA: Academic Press.
Concepts and Semantic Memory
11
Athanasopoulos, P., & Bylund, E. (2013). Does grammatical aspect affect motion
event cognition? A cross-linguistic comparison of English and Swedish speakers.
Cognitive Science, 37, 286–309.
Atran, S., & Medin, D. L. (2008). The native mind and the cultural construction of nature.
Boston, MA: MIT Press.
Barsalou, L. W. (1987). The instability of graded structure: Implications for the nature
of concepts. In U. Neisser (Ed.), Concepts and conceptual development: Ecological and
intellectual factors in categorization (pp. 101–140). Cambridge, England: Cambridge
University Press.
Barsalou, L. W. (1991). Deriving categories to achieve goals. In G. H. Bower (Ed.),
The psychology of learning and motivation: Advances in research and theory (Vol. 27, pp.
1–64). San Diego, CA: Academic Press.
Barsalou, L. W., Simmons, W. K., Barbey, A. K., & Wilson, C. D. (2003). Grounding
conceptual knowledge in modality-specific systems. Trends in Cognitive Sciences, 7,
84–91.
Clark, H. H. (1983). Making sense of nonce sense. In G. B. Flores d’Arcais & R. J.
Jarvella (Eds.), The process of language understanding (pp. 297–331). London, England: Wiley.
Collins, A. M., & Loftus, E. F. (1975). A spreading activation theory of semantic processing. Psychological Review, 82, 407–428.
Collins, A. M., & Quillian, M. R. (1969). Retrieval time from semantic memory. Journal
of Verbal Learning and Verbal Behavior, 8, 241–248.
Crutch, S., & Warrington, E. K. (2005). Abstract and concrete concepts have structurally different representational frameworks. Brain, 128, 615–627.
De Los Reyes, A., & Marsh, J. K. (2011). Patients’ contexts and their effects on clinicians’ impressions of conduct disorder symptoms. Journal of Clinical Child and
Adolescent Psychology, 40, 479–485.
Estes, Z., Golonka, S., & Jones, L. L. (2011). Thematic thinking: The apprehension
and consequences of thematic relations. In B. Ross (Ed.), Psychology of learning and
motivation (Vol. 54, pp. 249–294). Burlington, Canada: Academic Press.
Gentner, D., & Kurtz, K. J. (2005). Relational categories. In W.-k. Ahn, R. L. Goldstone, B. C. Love, A. B. Markman & P. Wolff (Eds.), Categorization inside and outside
the laboratory: Essays in honor of Douglas L. Medin (pp. 151–175). Washington, DC:
American Psychological Association.
Gentner, D., & Stevens, A. L. (Eds.) (1983). Mental models. Hillsdale, NJ: Lawrence
Erlbaum Associates.
Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world?
Behavioral and Brain Sciences, 33, 1–23.
Keil, F. C. (1989). Concepts, kinds, and cognitive development. Cambridge, MA: MIT
Press.
Keil, F. C. (2006). Explanation and understanding. Annual Review of Psychology, 57,
227–254.
Kintsch, W., & Mangalath, P. (2011). The construction of meaning. Topics in Cognitive
Science, 3, 346–370.
12
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
Kousta, S.-T., Vigliocco, G., Vinson, D. P., Andrews, M., & Del Campo, E. (2011). The
representation of abstract words: Why emotion matters. Journal of Experimental
Psychology: General, 140, 14–34.
Malt, B. C. (1990). Features and beliefs in the mental representation of categories.
Journal of Memory and Language, 29, 289–315.
Malt, B. C., & Majid, A. (2013). How thought is mapped into words. WIREs Cognitive
Science, 4, 583–597.
Mandler, J. M., & Johnson, N. S. (1977). Remembrance of things parsed: Story structure and recall. Cognitive Psychology, 9, 111–151.
Markman, A. B., & Ross, B. H. (2003). Category use and category learning. Psychological Bulletin, 129, 592–613.
Martin, A. (2007). The representation of object concepts in the brain. Annual Review
of Psychology, 58, 25–45.
McRae, K., & Jones, M. N. (2013). Semantic memory. In D. Reisberg (Ed.), The Oxford
handbook of cognitive psychology (pp. 206–219). Oxford, England: Oxford University
Press.
Medin, D. L., & Ortony, A. (1989). Psychological essentialism. In S. Vosniadou & A.
Ortony (Eds.), Similarity and analogical reasoning (pp. 179–195). Cambridge, England: Cambridge University Press.
Murphy, G. L. (2002). The big book of concepts. Cambridge, MA: The MIT Press.
Phillips, W., & Santos, L. R. (2007). Evidence for kind representations in the absence
of language: Experiments with rhesus monkeys (Macaca mulatta). Cognition, 102,
455–463.
Rips, L. J., Smith, E. E., & Medin, D. L. (2012). Concepts and categories: Memory,
meaning, and metaphysics. In K. J. Holyoak & R. G. Morrison (Eds.), The Oxford
handbook of thinking and reasoning (pp. 177–209). Oxford, England: Oxford University Press.
Rogers, T. T., & McClelland, J. L. (2004). Semantic cognition: A parallel distributed processing approach. Cambridge, MA: MIT Press.
Rosch, E., & Mervis, C. B. (1975). Family resemblances: Studies in internal structure
of categories. Cognitive Psychology, 7, 573–605.
Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D., & Boyes-Braem, P. (1976). Basic
objects in natural categories. Cognitive Psychology, 8, 382–439.
Schank, R. C., & Abelson, R. P. (1977). Scripts, plans, goals, and understanding: An
inquiry into human knowledge structures. Hillsdale, NJ: Erlbaum.
Shtulman, A., & Valcarcel, J. (2012). Scientific knowledge suppresses but does not
supplant earlier intuitions. Cognition, 124, 209–215.
Smith, E. E., & Medin, D. L. (1981). Categories and concepts. Cambridge, MA: Harvard
University Press.
Smith, E. E., Shoben, E. J., & Rips, L. J. (1974). Structure and process in semantic memory: A featural model for semantic decisions. Psychological Review, 81, 214–241.
Smith, J. D., & Minda, J. P. (1998). Prototypes in the mist: The early epochs of category
learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24,
1411–1436.
Concepts and Semantic Memory
13
Storms, G. (2004). Exemplar models in the study of natural language concepts. Psychology of Learning and Motivation, 45, 1–39.
Yee, E., Chrysikou, E. G., & Thompson-Schill, S. L. (2013). Semantic memory. In K.
Ochsner & S. Kosslyn (Eds.), The Oxford handbook of cognitive neuroscience, Vol. 1:
Core topics (pp. 353–374). Oxford, England: Oxford University Press.
FURTHER READING
Atran, S., & Medin, D. L. (2008). The native mind and the cultural construction of nature.
Boston, MA: MIT Press.
Barsalou, L. B. (2008). Grounded cognition. Annual Review of Psychology, 59, 617–645.
Keil, F. C. (2006). Explanation and understanding. Annual Review of Psychology, 57,
227–254.
Malt, B. C., & Majid, A. (2013). How thought is mapped into words. WIREs Cognitive
Science, 4, 583–597.
McRae, K., & Jones, M. N. (2013). Semantic memory. In D. Reisberg (Ed.), The Oxford
handbook of cognitive psychology (pp. 206–219). Oxford, England: Oxford University
Press.
Murphy, G. L. (2002). The big book of concepts. Cambridge, MA: The MIT Press.
Rogers, T. T., & McClelland, J. L. (2004). Semantic cognition: A parallel distributed processing approach. Cambridge, MA: MIT Press.
Yee, E., Chrysikou, E. G., & Thompson-Schill, S. L. (2013). Semantic memory. In K.
Ochsner & S. Kosslyn (Eds.), The Oxford handbook of cognitive neuroscience, Vol. 1:
Core topics (pp. 353–374). Oxford, England: Oxford University Press.
BARBARA C. MALT SHORT BIOGRAPHY
Barbara C. Malt is a Professor of Psychology at Lehigh University. Her
research focuses on thought, language, and the relation between the two. She
is especially interested in how objects and actions are mentally represented,
how people (including monolinguals and bilinguals, both children and
adults) talk about these objects and actions using the tools available in their
language(s), and what influence, if any, the different ways of talking have
on nonlinguistic representations. She is an associate editor for the Journal of
Experimental Psychology: Learning, Memory, and Cognition. More information
can be found at http://www.lehigh.edu/∼bcm0/bcm0/index.html
RELATED ESSAYS
Language, Perspective, and Memory (Psychology), Rachel A. Ryskin et al.
Misinformation and How to Correct It (Psychology), John Cook et al.
Language and Thought (Psychology), Susan Goldin-Meadow
Social Aspects of Memory (Psychology), William Hirst and Charles B. Stone
14
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
Implicit Memory (Psychology), Dawn M. McBride
Memory Gaps and Memory Errors (Psychology), Jeffrey S. Neuschatz et al.
Theory of Mind (Psychology), Henry Wellman