Semantic Memory: Cognitive Effects
Researchers originally conceptualized semantic memory as human knowledge of language (Tulving, 1972). The term now generally refers to one's everyday knowledge of the world, as contrasted with episodic memory, one's knowledge of personal experience. Semantic memory includes such facts as, "Robins are birds," "Chairs have legs," and "Fire-works are dangerous." Semantic memory has been thought by some researchers (Tulving and Schacter, 1990) to be a functionally separate memory system, distinct from episodic memory. Neuroimaging techniques implicate differential brain region activity in semantic and episodic tasks (Wheeler, Suss, and Tulving, 1997; Knowlton, 1998). There is some suggestion in scalp-based recordings of electrical brain activity following stimulus presentation (event related potentials) that a wave of negative amplitude may accompany. Nevertheless, the interdependence of semantic and episodic information (McKoon, Ratcliff, and Dell, 1986; Shoben and Ross, 1986) urges caution in accepting the claim that these are separate systems. Indeed, the best-known computational model of memory continues to treat the two as part of the same (declarative) memory system (Anderson, and Lebiere, 1998).
Semantic and episodic recall may be distinguished by whether one believes one knows something, or whether one personally remembers it (Knowlton and Squire, 1995). Mark Wheeler and colleagues (1997) proposed that semantic memory is associated with the absence of the phenomenological experience of remembering, a distinction that may prove more relevant than one based on content. Thus, semantic memory may include autobiographical content that is known, but not remembered.
Research in semantic memory can be divided into three categories concerning human knowledge of concepts: First, one knows that concepts belong to various categories. Second, one knows that concepts have certain properties and bear certain relations to other concepts. Third, one knows that concepts can be combined with other concepts.
Categorical knowledge is the most studied area in semantic memory. One of the robust findings in this area is the typicality effect (Rosch, 1975). Items that are good examples of a category are more readily verified as members than are poor examples. For example, people decide that "Robins are birds" more rapidly than "Chickens are birds."
The typicality effect has its counterpart for false statements (the relatedness effect: Smith, Shoben, and Rips, 1974). These are easier to disconfirm if subject and predicate are unrelated. For example, "A goose is a mammal" is more difficult to disconfirm than "A goose is a tool."
There are many proposed explanations of the typicality effect. It has been ascribed to computing similarity to a prototype or to multiple members of a category, and to feature overlap between an item and a category description. Other factors such as frequency (Barsalou, 1985; Heit and Barsalou, 1996) appear to moderate typicality, although twenty-first-century evidence does not strongly support an earlier claim that familiarity is one of these (Hampton, 1997). While typicality is often discussed in terms of similarity of semantic memory representations, there is a growing acknowledgment among the scientific community that the notion of similarity is itself a complex construct requiring some basis for comparison (Medin, Goldstone, and Gentner, 1993). For example, similarity may sometimes involve abstract decisions based on common goals or instrumentalities. Typical items to bring to school (books, a calculator, a pen) may be similar by virtue of accomplishing related goals, but will be unlikely to share much physical resemblance.
Context effects are common in semantic memory research. For example, early findings indicated that the size of the typicality effect varies as a function of the proportion of related false statements among the test stimuli (McCloskey and Glucksberg, 1979). Other studies have shown that the same categories may exhibit different typicality and similarity relationships in different contexts. Thus, what is a typical bird in southern Florida may depend in part upon whether the informant is Hispanic (Schwanenflugel and Rey, 1986). Similarly, North Americans will alter their judgments of the typicality of various birds when asked to take the perspective of a South American (Barsalou, 1987). These effects extend to similarity. Black and grey will be more similar than grey and white in the context of cloud color, but not in the context of hair color (Medin and Shoben, 1988).
Other effects are more controversial (see Chang, 1986, for a review). For example, it has often been argued that there is a category size effect in semantic memory in which it is harder to confirm membership in a large category than it is in a smaller one. However, reversals of this finding do occur. The finding may in part be confounded with the fact that larger categories tend to be at a higher level of abstraction than smaller categories (for example, "tool" and "implement").
The question of how categorical information is represented in memory remains unresolved. Some early theorists suggested that semantic memory could be viewed as a hierarchical tree or semantic network (Collins and Quillian, 1969; Holyoak and Glass, 1975) in which concepts such as "robin" and "canary" would be represented at the bottom of the tree with links from both extending up to "bird." Others espoused a componential approach in which concepts were represented as sets of properties or semantic features. Semantic decisions were made by a comparison of these features or properties (Smith et al., 1974; McCloskey and Glucksberg, 1979). Yet other representations and mechanisms have been proposed to account for categorical knowledge. A partial listing of representations includes prototypes, images, exemplars, schemas, situational models, and hyperdimensional vectors that represent semantic similarity by tracking the contexts in which different concepts may be found. None of these approaches by itself is likely to account for all of the available data, and different semantic units subject to different processes may be required (Brewer, 1993). For example, production frequency (the frequency with which an item is generated as an example of a category) and typicality ratings appear to provide different contributions to classification times. The suggestion has been made that the former relies on retrieval of network links whereas the latter involves a similarity computation operating over features (Hampton, 1997). Sophisticated techniques suggest that retrieval of information occurs continuously in time as opposed to in discrete quanta (Kounios, 1997), but whether such results are inconsistent with network models, as researchers sometimes claim, will be sensitive to the specific structural and processing assumptions adopted by such models.
Judgments of Relative Magnitude
Although there have been some direct investigations of how people process information about properties of concepts, most of the work on this question has concerned how people make judgments of relative magnitude such as, "Are rabbits larger than mice?"
The oldest and most robust finding in the literature is the symbolic distance effect. Objects that are further apart on some dimension are more readily discriminated than objects that are closer together (Moyer, 1973). For example, it is easier to determine that "Desks are larger than strawberries" than "Desks are larger than dogs."
A second finding, the congruity effect, is that judgments are easier when the question matches the magnitude of the objects, small or large (Banks, 1977). If one asks which is larger, a person's response will be faster with large than with small things. But if one asks which is smaller, a person's response will be faster with small things.
A third finding is the bowed serial position effect: Holding symbolic distance constant, objects of intermediate magnitude are more difficult to discriminate than objects of extreme magnitude (Shoben, Čech, Schwanenflugel, and Sailor, 1989). For example, it is more difficult for a person to select the larger of "wolf" and "pig" than to select the larger of either "toad" and "snail" or of "bull" and "elephant."
Like decisions about categorical information, judgments of relative magnitude are subject to context effects. Whether pairs such as "rabbit-beaver" are discriminated more readily under the instruction to select the smaller or the larger item will depend on the presence of large pairs. If there are no larger pairs, then "rabbit" and "beaver" suddenly behave as if they are large items (Čech and Shoben, 2001). Moreover, in appropriate contexts, multiple congruity effects may ensue, each corresponding to a different group of items.
There are also effects of categorization difficulty. Some items are difficult to classify as large or small, and others are relatively easy. In general, items of extreme magnitude are more readily classified than items of intermediate magnitude. This finding helps explain the serial position effect. When items across the magnitude scale are equated for categorization difficulty, no bowed serial position effect occurs (Shoben and Wilson, 1998).
Finally, there are reference point effects. If people are asked to determine which of two objects is closer to a third, the task is harder if the items are far from the reference point (Holyoak, 1978). Thus, for example, it is easier to determine whether 4 or 5 is closer to 3 than whether 6 or 7 is closer to 3.
Theories of Relational Information
Many theories have been proposed to account for relational judgments, although most have problems explaining at least some of the findings. Priming or expectancy-based theories claim that asking which is larger or smaller will bias people toward the specified size. However, there appear to be congruity effects that are not due to such priming; moreover, these models wrongly predict that congruity should occur early in processing, rather than late (Čech, 1995). Traditionally, theories fall into two broad camps: those like the discrete code model (Banks, 1977), which posit comparison of discrete, propositional information, and analogical models like the reference point model (Holyoak, 1978), which posit direct comparison of stored analogical magnitude information. Models used by twenty-first-century researchers include the recoding model (Čech and Shoben, 2001), evidence accrual models (Petrusic, 1992; Birnbaum and Jou, 1990), and the connectionist model of Leth-Steensen and Marley (2002).
The recoding model claims that people use range and co-occurrence information to cluster items into groups. People assign relative, contrastive magnitude codes to items within a group ("this is large," "that is small") to facilitate comparison. Comparison thus necessarily relies on attentional and categorization processes. However, in evidence accrual models, item magnitudes are sampled until there is sufficient evidence regarding relative size. Finally, the Leth-Steensen and Marley model also includes an evidence accrual component, but enables responding on the basis of positional information (whether an item is fourth or third largest, for example). This latter model employs learned associations and effectively combines aspects of the other approaches. William Petrusic has suggested that models under which people operate may differ: Some people seem to act in accord with the recoding model, and others seem to use evidence accrual.
A domain of particular interest concerns numerosity judgments. Researchers in this area study the relative merits of the single-format and the multiple-format assumptions (Blankenberger and Vorberg, 1997). The single-format assumption asserts that different number representations ("4," "four," "IV") project onto the same semantic representations, whereas the multiformat assumption claims multiple semantic representations for numbers (Dehaene, 1997). Some numerical comparison studies examine these positions by focusing on the brain regions active during the comparison of different number types, and others study how format influences the distance effect. For example, is the average time required to determine the larger of the pairs "two-four" and "2-4" the same as the time required to compare "two-4" and "2-four"? The answer appears to be no: Different distance effects are found with mixed-format numbers (Vorberg and Blankenberger, 1993). Although it is premature to favor the multiple-format assumption from results such as these, the domain of numerical comparisons does hold out the promise of helping scientists determine how rich semantic representations are.
A problem in categorization concerns how concepts are combined. Some of these combinations appear superficially simple, while others are clearly more complex. For example, "red ball" can readily be paraphrased as "a ball that is red," but "malarial mosquitoes" must instead be paraphrased as "mosquitoes that cause malaria." "Red ball" uses a predicating adjective, "red," in which a property of the modifier is ascribed to the head noun, while "malarial mosquitoes" employs the nonpredicating adjective "malarial." Phrases involving nonpredicating adjectives are sometimes more difficult to comprehend (Murphy, 1990), although faster comprehension times have also been reported for nonpredicating nominal compounds (Gagné, 2000). In nonpredicating compounds, the interpretation appears to rely on establishing some relationship between the two components (a relational interpretation).
Most research on conceptual combinations has used predicating adjectives. One fundamental question is how people determine membership in combined categories. For example, "A cardinal is a red bird" is clearly true, because a cardinal is clearly both a red thing and a bird. At the same time, "A female cardinal is a red bird" is only partly true, because female cardinals are only somewhat red. Such concerns might lead researchers to propose a min rule: An exemplar is a member of a combined category only to the minimum degree that it is a member of the category defined by the adjective and the category specified by the noun. Despite the intuitive appeal of such a rule, it cannot handle many cases: Guppies are perhaps the paradigmatic example of the category "pet fish," but they are good examples of neither pets nor fish. Empirical judgments of subjects show clear violations of the min rule (Smith and Osherson, 1984).
In natural language corpora, conceptual combinations that involve an initial predicating noun are less frequent than combinations that require a relational interpretation, although there is disagreement about the exact disproportion in the two types. Most contemporary work in conceptual combination centers around the dual-process model (Wisniewski, 1997) and the CARIN model (Gagné and Shoben, 1997). In the dual-process model, relational interpretations require a process of scenario construction in which specified roles may be assigned the two concepts. Property interpretations, in contrast, arise from a process of alignment in which alignable differences (differing values on a common dimension) help drive the decision regarding which property to move to the complex conceptual schema of the head noun. An alignable difference between "dog" and "man," for example, is that each locomotes using legs, although a dog moves on four legs. Thus, a potential interpretation of "dog man" is a man who travels on all fours, whereas a potential interpretation of "man dog" is a dog walking upright. By way of contrast, a relational interpretation of "dog man" might include some scenario in which an individual prefers dogs to cats. This model has recently been extended to include nonalignable differences that display spatial correspondences (Wisniewski and Middleton, 2002): Kangaroos and dogs do not have a readily retrievable common dimension on which "pouch" can serve as a contrasting value that helps distinguish the two, but the presence of equivalent spatial areas on dogs and kangaroos may allow the interpretation that a "kangaroo dog" is a dog with a pouch on the lower front of its body.
CARIN does not rely on an alignment process, or a privileged attempt to modify the representation of the head noun. Instead, the model posits that the repertoire of relations associated with the modifier will guide finding an appropriate interpretation. For example, because "mountain" has a high frequency of relations involving location, "mountain man" will likely be interpreted as a man who lives on a mountain. Accordingly, the relative rarity of property interpretations ought to result in long interpretation times.
The interpretation process for conceptual combinations has also been modeled as a constraint satisfaction process that requires simultaneous solutions to the constraints of diagnosticity, plausibility, and informativeness (Costello and Keane, 2000). Moreover, like categorization and magnitude comparisons, conceptual combinations are also subject to context effects (Gerrig and Bortfeld, 1999). The work on conceptual combinations illustrates the reliance on a much richer and more complex semantic memory than researchers assumed in the early work on network and feature models.
This review has necessarily been selective, with many topics omitted. In closing, however, it should be noted that one's knowledge of the world will influence most of the cognitive things that one does. Solving a problem, following directions, or simply reading involves one's semantic memory. Semantic memory is fundamental to cognition.
Anderson, J. R., and Lebiere, C. (1998). The atomic components of thought. Mahwah, NJ: Erlbaum.
Banks, W. P. (1977). Encoding and processing of symbolic information in comparative judgments. In G. H. Bower, ed., The psychology of learning and motivation, Vol. 11. New York: Academic Press.
Barsalou, L. W. (1985). Ideals, central tendency, and frequency of instantiation as determinants of graded structure in categories. Journal of Experimental Psychology: Learning, Memory, and Cognition 11 (4), 629-654.
—— (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. Cambridge, UK: Cambridge University Press.
Birnbaum, M. H., and Jou, J. (1990). A theory of comparative response times and "difference" judgments. Cognitive Psychology 22 (2), 184-210.
Blankenberger, S., and Vorberg, D. (1997). The single-format assumption in arithmetic fact retrieval. Journal of Experimental Psychology: Learning, Memory, and Cognition 23 (3), 721-738.
Brewer, W. F. (1993). What are concepts? Issues of representation and ontology. In G. V. Nakamura, R. M. Taraban, and D. L. Medin, eds., The psychology of learning and motivation, Vol. 29: Categorization by humans and machines. San Diego, CA: Academic Press.
Čech, C. G. (1995). Is congruity due to encoding? Journal of Experimental Psychology: Learning, Memory, and Cognition 21 (5), 1,275-1,288.
Čech, C. G., and Shoben, E. J. (2001). Categorization processes in mental comparisons. Journal of Experimental Psychology: Learning, Memory, and Cognition 27 (3), 800-816.
Chang, T. M. (1986). Semantic memory: Facts and models. Psychological Bulletin 99 (2), 199-220.
Collins, A., and Quillian, M. R. (1969). Retrieval time from semantic memory. Journal of Verbal Learning and Verbal Behavior 8, 240-247.
Costello, F. J., and Keane, M. T. (2000). Efficient creativity: Constraint-guided conceptual combination. Cognitive Science 24(2), 299-349.
Gagné, C. L. (2000). Relation-based combinations versus property-based combinations: A test of the CARIN theory and the dual-process theory of conceptual combination. Journal of Memory and Language 42 (3), 365-389.
Gagné, C. L., and Shoben, E. J. (1997). Influence of thematic relations on the comprehension of modifier-noun combinations. Journal of Experimental Psychology: Learning, Memory, and Cognition 23 (1), 71-87.
Gerrig, R. J., and Bortfeld, H. (1999). Sense creation in and out of discourse contexts. Journal of Memory and Language 41 (4), 457-468.
Hampton, J. A. (1997). Associative and similarity-based processes in categorization decisions. Memory & Cognition 25 (5), 625-640.
Heit, E., and Barsalou, L. W. (1996). The instantiation principle in natural categories. Memory 4 (4), 413-451.
Holyoak, K. J. (1978). Comparative judgments with numerical reference points. Cognitive Psychology 10 (2), 203-243.
Holyoak, K. J., and Glass, A. L. (1975). The role of contradictions and counterexamples in the rejection of false sentences. Journal of Verbal Learning and Verbal Behavior 14 (2), 215-239.
Knowlton, B. J. (1998). The relationship between remembering and knowing: A cognitive neuroscience perspective. Acta Psychologica 98 (2-3), 253-265.
Knowlton, B. J., and Squire, L. R. (1995). Remembering and knowing: Two different expressions of declarative memory. Journal of Experimental Psychology: Learning, Memory, and Cognition 21(3), 699-710.
Kounios, J. (1996). On the continuity of thought and the representation of knowledge: Electrophysiological and behavioral time-course measures reveal levels of structure in semantic memory. Psychonomic Bulletin & Review 3 (3), 265-286.
Leth-Steensen, C., and Marley, A. A. J. (2000). A model of response time effects in symbolic comparison. Psychological Review 107(1), 62-100.
McCloskey, M., and Glucksberg, S. (1979). Decision processes in verifying category membership statements: Implications for models of semantic memory. Cognitive Psychology 11 (1), 1-37.
McKoon, G., Ratcliff, R., and Dell, G. S. (1986). A critical evaluation of the semantic-episodic distinction. Journal of Experimental Psychology: Learning, Memory, and Cognition 12 (2), 295-306.
Medin, D. L., Goldstone, R. L., and Gentner, D. (1993). Respects for similarity. Psychological Review 100 (2), 278.
Medin, D. L., and Shoben, E. J. (1988). Context and structure in conceptual combination. Cognitive Psychology 20 (2), 158-190.
Moyer, R. S. (1973). Comparing objects in memory: Evidence suggesting an internal psychophysics. Perception and Psychophysics 13 (2), 180-184.
Murphy, G. L. (1990). Noun phrase interpretation and conceptual combination. Journal of Memory and Language 29 (3), 259-288.
Petrusic, W. M. (1992). Semantic congruity effects and theories of the comparison process. Journal of Experimental Psychology: Human Perception and Performance 18 (4), 962-986.
Rosch, E. H. (1975). Cognitive representations of semantic categories. Journal of Experimental Psychology: General 104 (3), 192-233.
Schwanenflugel, P. J., and Rey, M. (1986). The relationship between category typicality and concept familiarity: Evidence from Spanish-and English-speaking monolinguals. Memory & Cognition 14 (2), 150-163.
Shoben, E. J., Čech, C. G., Schwanenflugel, P. J., and Sailor, K. M. (1989). Serial position effects in comparative judgments. Journal of Experimental Psychology: Human Perception and Performance 15 (2), 273-286.
Shoben, E. J., and Ross, B. H. (1986). The crucial role of dissociations. Behavioral and Brain Sciences 9 (3), 568-571.
Shoben, E. J., and Wilson, T. L. (1998). Categorization in judgments of relative magnitude. Journal of Memory and Language 38 (1), 94-111.
Smith, E. E., and Osherson, D. N. (1984). Conceptual combination with prototype concepts. Cognitive Science 12 (4), 485-527.
Smith, E. E., Shoben, E. J., and Rips, L. J. (1974). Structure and process in semantic memory: A featural model for semantic decisions. Psychological Review 81 (3), 214-241.
Tulving, E. (1972). Episodic and semantic memory. In E. Tulving and W. Donaldson, eds., Organization of memory. New York: Academic Press.
Tulving, E., and Schacter, D. L. (1990). Priming and human memory systems. Science 247, 301-306.
Vorberg, D., and Blankenberger, S. (1993). Mentale repräsentation von zahlen. Sprache and Kognition 12 (2), 98-114.
Wheeler, M. A., Suss, D. T., and Tulving, E. (1997). Toward a theory of episodic memory: The frontal lobes and autonoetic consciousness. Psychological Bulletin 121 (3), 331-354.
Wisniewski, E. J. (1997). When concepts combine. Psychonomic Bulletin & Review 4 (2), 167-183.
Wisniewski, E. J., and Middleton, E. L. (2002). Of bucket bowls and coffee cup bowls: Spatial alignment in conceptual combination. Journal of Memory and Language 46 (1), 1-23.
Revised byClaude G.Čech