Journal of Experimental Psychology:
Learning, Memory, and Cognition
1984, Vol. 10, No. 2, 198-202
Copyright 1984 by the
American Psychological Association, Inc.
Recognition and Recall of Words with a Single MeaningAddress correspondence to Paul Muter, Department of Psychology, University of Toronto, Toronto, Ont., M5S 3G3, Canadamuter@psych.utoronto.ca
University of Toronto, Toronto, Ontario, Canada
Twenty-four subjects were tested for recognition of famous surnames and then were tested for cued recall of the same surnames. For common names (e.g., Cooper), the usual pattern of recognition failure of recallable words (Tulving & Wiseman, 1975) was found. For unique, "single-node" names (e.g., Kierkegaard), virtually no recognition failure (1.4%) of recallable words was obtained. These results fit well with generation-recognition theory.
The relationship between recognition and recall has been a focus of much theory and experimentation in the study of human memory. An important phenomenon bearing on the nature of this relationship is recognition failure of recallable words. Substantial recognition failure of recallable words has been demonstrated in both episodic memory (e.g., Begg, 1979; Tulving & Thomson, 1973) and semantic memory (Muter, 1978; Neely & Payne, 1983). In the episodic memory experiments, subjects typically study word-pair episodes such as glue-chair, are tested for their ability to recognize chair as having been recently studied, and finally are tested on their ability to recall chair given the cue glue. In the semantic memory experiments, subjects indicate which of a list of surnames (e.g., Cooper) they recognize as being famous, then they attempt to recall famous surnames given appropriate cues (e.g., "author of The Last of the Mohicans, James Fenimore:").
According to some theorists (e.g., Gillund & Shiffrin, 1984; Reder, Anderson, & Bjork, 1974), the probability of recognition failure of recallable words should be lower for words that are likely to have only one meaning than for words with several meanings. For example, under generation-recognition theory (Anderson & Bower, 1974; Martin, 1975; Reder et al., 1974), words with several shades of meaning, such as water, are represented in permanent memory by several concept nodes, one for each shade of meaning. If the word water is presented as a probe in a recognition test, one of the water nodes is automatically accessed, and if the relevant information, an occurrence tag, for example, is attached to the node, a positive response is emitted. For a recall test, a subset of all nodes is accessed and examined, and those nodes bearing the relevant information are output. Different retrieval cues in the recognition test and the recall test may result in different water nodes being examined in the two tests. In this case, recognition failure of a recallable word can occur. Words with only one meaning, in contrast, have only one node to be examined, and if this node bears the information necessary for successful recall, successful recognition should also result.
Reder et al. (1974) studied memory for low- and high-frequency words. It was presumed that low-frequency words would have fewer senses than high-frequency words. Reder et al. found that recall was superior to recognition for high-frequency words, but that the result was reversed for low-frequency words. Reder et al. argued that these data supported generation-recognition theory. However, if the data of Reder et al. are analyzed in a different way (Tulving & Wiseman, 1975), the recognition failure rates of recallable words for the high- and low-frequency conditions are comparable. Tulving and Wiseman summarized results from 40 conditions from 12 experiments and found a consistent relationship between overall recognition rate and amount of recognition failure of recallable words. This relationship is demonstrated by the curve in Figure 1, which fits the data quite well. The data of Reder et al., for both high- and low-frequency words, are consistent with this relationship. In this sense, the pattern of recognition failure was not affected by the frequency of the word.
The assumption of Reder et al. (1974) that low-frequency words tend to have only one sense may have been unwarranted (Tulving & Watkins, 1977). In a more concerted attempt to study words with a single meaning, Tulving and Watkins found the usual pattern of recognition failure of recallable words using words with only a single dictionary meaning. Examples were cactus and handshake. Tulving and Watkins argued that this result was incompatible with generation-recognition theory, since words like cactus are maximally likely to be represented by a single node in memory.
The number of meanings listed in a dictionary is a clean operational definition of the number of senses of a word, but there remains some doubt regarding whether the words used by Tulving and Watkins (1977) would be represented by a single node. For example, there is more than one kind of handshake, and there is more than one species of cactus. A more convincing example of a word with a single node in memory would be a unique name such as Kierkegaard, since people in North America in the 1980s are likely to have acquired information about at most one Kierkegaard.
In the present experiment, subjects were tested on their ability to recognize and recall famous names. Half of the targets were common (e.g., Cooper) and half were unique (e.g., Kierkegaard). For common names, the usual pattern of recognition failure (i.e., a replication of the results of Muter, 1978, and Neely & Payne, 1983) was expected. Of greater interest was whether the unique names would show the usual pattern of recognition failure, or whether, as generation-recognition theory predicts, these unique names would be immune to the effect.
MethodTwenty-four students in an undergraduate cognitive psychology course received credit for participating in the experiment. They were run in small groups.
The 24 unique target surnames used in the experiment are given in column 1 of Table 1. These surnames do not appear in the 1982 Toronto telephone directory, and they are names of people who are listed in the Random House Dictionary (1969). For the recall test, given names and a brief description were provided as cues. These cues are listed in column 2 of Table 1.
Of the 24 common target surnames, 16 were taken from Muter (1978, Table 1). These names are listed in the Random House Dictionary (1969) and are sufficiently common to take up at least half a column in the Toronto telephone directory. The remaining eight common surnames were taken from Neely and Payne (1983), who used similar criteria in selecting names. The following are examples of common targets and cues: Cooper, Author of The Last of the Mohicans: James Fenimore; Davis, President of the Confederacy during the U.S. Civil War: Jefferson; and Ross, Maker of the first U.S. flag: Betsy.
All 48 of the famous people achieved their fame before 1950.
There were 48 lures for the recognition test, 24 common and 24 unique. None of the lures are listed in the Random House Dictionary. The common lures take up at least half a column in the Toronto telephone directory, and the unique lures appear exactly once in the Toronto telephone directory. Examples of common lures are Bedford, Ferguson, and Phillips. Examples of unique lures are: Albazi, Holstock, and Pucirius.
The recognition test sheets consisted of all 96 surnames, 48 targets and 48 lures, in a different random order for each subject. The recall test booklets consisted of all 48 cues in a different random order for each subject. Following each cue was a space in angle brackets.
For the recognition test, subjects were given the following instructions.
"Some of the surnames on the sheet in front of you belong to people who are or were famous (i.e., they are in the Encyclopedia Britannica), and some do not. Please circle those names which you recognize as belonging to a famous person (real, not fictional) who would be in the Encyclopedia Britannica). Please do not try to guess what proportion are intended to be famous; simply call them as you see them."
Subjects were allowed as much time as they needed to complete the recognition test. This took approximately 15 min. The recognition sheets were then collected, and the recall booklets were passed out. For the recall test, instructions were the following.
"Please try to fill in the spaces between the angle brackets with the appropriate surnames. Do not worry about correct spelling."
Subjects took approximately 20 min to complete the recall test.
ResultsThe results for each condition collapsed across subjects and words are presented in Table 2. Table 2 shows substantial recognition failure of recallable words for common famous names, as in Muter (1978) and Neely and Payne (1983) and virtually no recognition failure of recallable words for unique famous names.
The relationship between overall recognition rate and recognition rate of recallable words is portrayed in Figure 1. Each point represents the data for six names collapsed across subjects. For each condition, names were ranked according to overall recognition rate and then were blocked in groups of six. The curve in Figure 1 is the function used by Tulving and Wiseman (1975): P(Rn given Rc) = P(Rn) + 0.5[P(Rn) - P(Rn)**2], where Rn is successful recognition and Rc is successful recall.
The overall recognition rates were almost identical in the two conditions: .458 in the common condition and .474 in the unique condition. A z test for the difference between two proportions revealed that the recognition rates for recallable words were significantly different in the two conditions: z = 7.7, p < .000001. The correlation between recognition and recall was .34 in the common condition and .60 in the unique condition. The difference between these two correlations is significant: z = 5.7, p < .000001.
In the unique condition, there were only two instances of recognition failure of recallable words: One subject recalled but failed to recognize Ataturk and another subject recalled but failed to recognize Curie.
In the common condition, though a good fit to the Tulving and Wiseman (1975) function is suggested in Figure 1, the overall recognition rate of recallable words was slightly higher, .65, than the .58 rate "predicted" by the Tulving and Wiseman function. However, in the present context, the important point regarding the common condition is that there was substantial recognition failure of recallable words.
The false-alarm rate was 5.4% for unique lures and 14.9% for common lures. On the recall test, for unique cues, the rate of intrusions was 4.7% (2.4% extralist, 0.9% lures from the recognition test, and 1.4% inappropriate targets). For common cues, the rate of intrusions was 4.3% (2.8% extralist, 1.0% lures from the recognition test, and 0.5% inappropriate targets).
DiscussionThe present results fit well with generation-recognition theory. According to this theory, which was modified (Martin, 1975; Reder et al., 1974) to accommodate the finding of recognition failure of recallable words, in recognition a node in memory is automatically accessed, and if it bears the relevant information a positive response is emitted. In recall, a subset of all nodes is accessed and examined, and those that bear the relevant information are output. For a common name, it is possible that a nonfamous node will be accessed at the time of the recognition test and a famous node will be accessed at the time of the recall test. If this happens, recognition failure of a recallable word occurs. For a unique name, on the other hand, the number of nodes in memory would be at most one. Successful recall would imply that this node had been accessed in the generation phase and that the relevant information had been found; therefore recognition alone would necessarily be possible. The only difficulty for generation-recognition theory in the present results would be in accounting for the 1.4% rate of recognition failure of recallable words that did occur, but it is customary to regard such small rates as theoretically unimportant (e.g., Tulving, 1983, p. 280).
An alternative interpretation is in terms of the amount of overlap between the encoded features in the retrieval cue and the encode features in the memory trace (Flexser & Tulving, 1978; Tulving, 1983). According to this theoretical framework, which was developed for episodic memory, the extent of the overlap between cue features and trace features determines whether successful retrieval from memory will occur. It could be argued that in the unique condition, the encoded features in the recall cue either were unhelpful or were a subset of the encoded features in the recognition cue, whereas in the common condition sometimes encoded features in the recall cue were helpful and were not among the encoded features in the recognition cue. This interpretation is unsatisfying in that it begs the question: Why do unique and common names differ in this way? It is surprising, for example, that the information "founder of existentialism" was always either unhelpful or part of the the encoded retrieval information in recognition as well as in recall. From the demonstration of recognition failure of recallable words in semantic memory, Tulving (1983) concluded that "much of the information in semantic memory is encoded as specifically for access as is the information in episodic memory" (p. 288). By the same reasoning, the present results suggest that, for some unknown reason, unique names in semantic memory are not encoded specifically for access.
Tulving and associates assume that there are at least two memory systems, episodic memory and semantic memory. Generation-recognition theorists assume that "it is basically all one big memory" (Anderson & Ross, 1980, p. 442) consisting of concept nodes, connections among nodes, and episodic tags. A third general possibility (e.g., Jacoby, 1983) is that memory consists of nothing but traces of episodes, and that consultation of episodic memory traces underlies all memory tasks, including tests of world knowledge. The present results may be interpreted within this framework in the following way if the assumption of automatic access (in recognition) to at least one relevant trace is incorporated. If a subject's memory includes any episodic traces involving Kierkegaard, it is reasonable to expect that all of these traces will contain information indicating, directly or indirectly, that Kierkegaard is famous. Therefore, when any of the traces involving Kierkegaard is accessed in the recognition test, the subject will emit a positive response; recognition failure will not occur. On the other hand, a subject's memory will include many traces of episodes involving Cooper, some of which will contain no information indicating fame. If only traces involving nonfamous Coopers happen to be accessed in the recognition test, a negative response will be emitted; recognition failure will occur.
ReferencesAnderson, J. R., & Bower, G. H. (1974). A propositional theory of recognition memory. Memory & Cognition, 2, 406-412.
Anderson, J. R., & Ross, B. H. (1980). Evidence against a semantic-episodic distinction. Journal of Experimental Psychology: Human Learning and Memory, 6, 441-465.
Begg, I. (1979). Trace loss and the recognition failure of unrecalled words. Memory & Cognition, 7, 113-123.
Flexser, A. J., & Tulving E. (1978). Retrieval independence inrecognition and recall. Psychological Review, 85, 153-171.
Gillund, G., & Shiffrin, R. M. (1984). A retrieval model for both recognition and recall. Psychological Review, 91, 1-67.
Jacoby, L. L. (1983). Perceptual enhancement: Persistent effects of an experience. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9, 21-38.
Martin, E. (1975). Generation-recognition theory and the encoding specificity principle. Psychological Review, 82, 150-153.
Muter, P. (1978). Recognition failure of recallable words in semantic memory. Memory & Cognition, 6, 9-12.
Neely, J. H., & Payne, D. G. (1983). A direct comparison of recognition failure rates for recallable names in episodic and semantic memory tests. Memory & Cognition, 11, 161-171.
Random House dictionary (unabridged ed.). (1969). New York: Random House.
Reder, L. M., Anderson, J. R., & Bjork, R. A. (1974). A semantic interpretation of encoding specificity. Journal of Experimental Psychology, 102, 648-656.
Tulving, E. (1983). Elements of episodic memory. New York: OxfordUniversity Press.
Tulving, E., & Thomson, D. M. (1973). Encoding specificity and retrieval processes in episodic memory. Psychological Review, 80, 352-373.
Tulving, E., & Watkins, O. C. (1977). Recognition failure of words with a single meaning. Memory & Cognition, 5, 513-522.
Tulving, E., & Wiseman, S. (1975). Relation between recognition and recognition failure of recallable words. Bulletin of the Psychonomic Society, 6, 79-82.
Received March 25, 1983.
This research was supported by Grant UO149 from the Natural Sciences and Engineering Research Council of Canada to the author.
I thank Irene Rukavina and David Schloen for valuable assistance, and Mary L.A. Lamon, Colin M. Macleod, Dan L. Schacter, an anonymous reviewer, and especially Christian W. Mueller for helpful comments.