Journal of Psycholinguistic Research, Vol. 1, No. 4, 1972
Defineability
as a n I n d e x o f W o r d M e a n i n g
Brian J. O'Neill 1 Received September 9, 1971
A sample o f 277 nouns was scaled for rated defineability and rated pronunciability. lntercorrelations were computed among the following variables: defineability (D), pronunciability (P), imagery (I), concreteness (C), associative meaningfulness (m), familiarity (F), and Thorndike-Lorge frequencies (TLF). Rated D was substantially related to all the other var~bles, notably to rated I. Correlations were highest among D, I, C, and m, on the one hand, and among F, TLF, and P, on the other. The pattern o f correlations was interpreted in terms o f Paivio's (1970] distinction between higher-order and lower-order meaning. Two experiments were conducted to test the relationship between the word attributes and recall o f word labels with dictionary definitions provided as retrieval cue~ The positive effect o f D on label retrieval was independent o f the I values o f the defined words but the effectiveness of I on retrieval was dependent on word defineability.
INTRODUCTION Despite ~he fact that meaning is identified with definition in commonsense thinking, there has been no explicit attempt to scale words in terms of their defineability. The failure to investigate this aspect of meaning is the more surprising in view of the fact that there is no word attribute available which indicates the extent to which subjects (Ss) know what a word refers to. The development of an index of the "referential meaningfulness" of words, or the extent to which Ss know what a word denotes, is important for at least two reasons. In the first place, it should permit an assessment of the degree to which knowledge of referential meaning is implicated in responses This research was supported in part by grants to Dr. A. Paivio from the National Research Council of Canada (Grant 0087). 1Department of Psychology, University of Western Ontario, London, Ontario, Canada. 287 O 1972 Plenum Publishing Corporation, 227 West 17th Street, New York, N.Y. 10011.
288
O'Neill
that are the basis of available attributes of word meaning. These attributes include various indices of associative meaningfulness (e.g., Noble, 1952), rated familiarity, and rated concreteness and imagery (e.g., Paivio, Yuille, and Madigan, 1968) as well as the dimensions that have been identified using the semantic differential rating technique (e.g., Osgood, Suci, and Tannenbaum, 1957). Knowledge of what a word denotes may mediate the responses that contribute to any or all of these attribute values. A correlational analysis of the relationships that exist between an index of word defmeability and other word attributes should suggest the extent to which knowledge of referential meaning is involved in the latter. A second justification for an index of defineability concerns the functional significance of referential meaningfulness. It is possible that the learning difficulty or memory for a word may be related to its referential meaningfulness. That is, whether or not Ss know what a word means may have significant consequences for learning and memory performance. This possibility has not been examined experimentally nor does there seem to have been any attempt to control referential meaningfulness in traditional verbal learning and memory experiments. In the present research a sample of words was scaled in terms of the rated defmeability of the words. It was assumed that a S's evaluation of a word in terms of its defmeability should be directly related to his knowledge of what the word refers to. Thus the referential meanings of words rated high in defineability (D) may be assumed to be dearly known and words rated as low in D may be assumed to be unclear with respect to S's knowledge of their referential meanings. Average D ratings were taken as an index of referential meaningfulness. In order to estimate the implication of referential meaning in available indices of meaning, the relationships between D and several other word attributes were investigated. Of particular interest was the word imagery variable which has been linked to referential meaning by Paivio (Paivio, 1970, 1971; Paivio and O'Neill, 1970). The notion of imagery as reference is not unique (see also Ausubel, 1965; Bugelski, 1970; Mowrer, 1960) and it has a long history. James (1890), for example, has proposed that the static meaning of concrete words consists of sensory images reawakened; the meaning of abstract words, on the other hand, is linked to verbal associative processes. Variables of interest other than rated imagery (/) included rated concreteness (C), verbal associative meaningfulness (Noble's 1952 production m), Thorndike-Lorge frequencies (TLF), rated familiarity (F), and rated pronunei. ability (P). Most of these variables are incorporated in a single theory of meaning in which higher-order meaning is distinguished from lower-order meaning (Paivio,
Defmeability as an Index of Word Meaning
289
1970, 1971). Referential processes, coordinated with the I measure, and associative processes, defined by m, constitute higher-order meaning, the former referring to the symbolic imaginal transformations that are initiated by verbal stimuli, and the latter to primarily verbal symbolic associative reactions. Lower-order or representational meaning is identified with indices of word availability such as frequency counts ( T L F ) and familiarity ratings (F). Rated P was included in the study since pronunciability may be expected to be related to the availability of a verbal response and, therefore, to lower.order meaning. On the basis of Paivio's theory of meaning the defineability measure might be expected to be related more to higher-order meaning attributes rather than to attributes that are coordinated with lower-order meaning, since a defineability rating, presumably, involves a higher-order cognitive judgment.
METHOD Materials and Procedure
The stimulus materials were 277 nouns from the Paivio, YuiUe, and Madigan (1968) norms which contain rated I and C values, production m data (see Nobel, 1952), and T L F values. Mean attribute values and standard deviations for these variables, as well as for rated F, 2 are reported in Table I. Values of 100 and 50 respectively were assigned to AA and A frequencies. Defineability and pronunciability ratings were obtained using essentially the same procedures as were used by Paivio et al. to scale I and C. The nouns were randomized and reproduced in booklets each of which contained a page of instructions and six pages of stimulus words. Each page contained three 18-word columns with a short line printed beside each word. The order in which the page occurred in the booklet was random. The total number of 2Familiarity ratings have been obtained in Dr. Paivio's laboratories at the University of Western Ontario.
Table I. Means and Standard Deviations for Imagery (I), Concreteness (6"), Meaningfulness (m), Familiarity (F), and Frequencies ( T L F ) II
$.D.
iiii
I
C
m
F
TLF
4.31 1.28
4.11 1.82
5.56 1.049
5.15 1.I0
35.72 34.89
290
O'Neill
stimulus items included the original 277 nouns and duplicates of 47 of them. Subjects received one of two sets o f rating instructions. Define.ability Instructions The purpose of this experiment is to rate a list of words on how easy or difficult it is to define their meaning. To do this you are to rate each word as relatively easy or difficult to define using a seven-point scale. A word that is very easy for you to define would get a rating of 7 and a word that is very difficult to define would get a rating of 1. Words that are of intermediate value in terms of ease or difficulty of definition should be rated appropriately between the two extremes. To give you some idea of the procedure, read the first column of words to yourself, giving each word a rating in your mind. When you feel that you can accurately judge the relative difficulty of defining the meanings of the words, go back to the beginning and start writing down your ratings. Write on the line beside each word the number from 1 to 7 that best indicates your ability to define the meaning of the word. Feel free to use the entire range o f numbers, from 1 to 7; at the same time don't be concerned about how often you use a particular number as long as it is your true judgment. Work fairly quickly but do not be careless in your ratings.
Pronunciability Insemctions The purpose of this experiment is to determine the ease or difficulty of pronouncing a number of words. To do this you are to rate each word as relatively easy or difficult to pronounce using a seven-point scale. A word that is very easy to pronounce would get a rating of 7 and a word that is difficult to pronounce would get a rating of 1. Words that are intermediate in pronunciability should be rated appropriately between the two extremes. To give you some idea of the range of pronunciabiIity of the words, read the first column of words to yourself, giving each word a rating in your mind. When you feel that you can judge the relative pronunciability of the words accurately, go back to the beginning and start writing down your ratings. Write on the line beside each word the number from 1 to 7 that best indicates your judgment of the pronunciability of the word. Feel free to use the entire range o f numbers, from 1 to 7; at the same time don't be concerned about how often you use a particular number as long as it is your true ]udgment. Work fairly quickly but do not be careless in your ratings. Six sample items (coffee, dawn, cranium, unreality, tree, emporium) were printed below a rating scale on the instruction page to give Ss practice in making rating judgments and to ensure that they understood the instructions.
Defmeability as an Index of Word Meaning
291
Subjects
The Ss were 40 undergraduate students from introductory psychology classes at the University of Western Ontario. Twenty Ss (eight females and 12 males) rated the words according to the defmeability instructions and 20 (eight females and 12 males) according to the pronunciability instructions. The number of Ss tested at any given time ranged from one to ten.
RESULTS AND DISCUSSION Ease of definition ratings were averaged across Ss to yield mean defineability (D) values for each word) The overall D mean and standard deviation are 4.95 and 0.81 respectively (N= 277). The two sets of means for word duplicates (N = 47) were intercorrelated; the magnitude of the correlation coefficient (r = 0.90) suggests that within.subject ratings are stable. The reliability of the ratings was also estimated by randomly assigning rating booklets to one of two groups of equivalent size (n = 10) and by calculating the degree of correlation' between the mean word values of the two groups of raters. With N = 227, between-group reliability is substantial (r = 0.83). Pronunciability scores were subjected to the same statistical treatment as the D variable. Overall mean and standard deviation values are 5.38 and 0.76 respectively. Within-subject (r= 0.91 for 47 pairs of duplicate values) and between-group (r = 0.84) reliability correlation coefficients suggest that the P ratings are stable. The interrelationships among the variables, defineability (D), pronunciability (P), imagery (/), concreteness (C), meaningfulness (m), familiarity (F), and frequency (TLF), are indicated by the correlation coefficients in Table !I. 3A list of the stimulus words together with def'meabflity (D) and pronunciabflity (P) mean and standard deviation values for each word are available on request. Table lI. Intercorrelations Between D, 1, C, m, F, TLF, and pa
D D I C
m F TLF i
aro.o~s = 0.12.
1
C
0.703
0.640 0.816
rn 0.522 0.561 0.372
F
TLF
P
0.495 0.202 --0.027
0.279 0.164 0.040
0.402 0.308 0.140
0.436
0.256 0.651
0.342 0.613 0.540
292
O'Neill
As might be expected, the intercorrelations involving L C, m, and TLF, are approximately equivalent to the correlations reported by Paivio, Yuille, and Madigan (1968). Rated D correlates more substantially with all the other variables than does any other attribute, suggesting that defineability may be taken as a general index of meaning. A correlation of 0.70 between D and I indicates that imagery is a major correlate of defineability. In order to clarify the interrelationships among the variables, the coefficients were ranked in order of magnitude. Two clusters of correlations were apparent. Correlation values were highest among the variables D, I, C, or rn, on the one hand, and among F, TLF, and P, on the other, whereas the values of the coefficients tended to be lower if one variable was D, L C, or m, and the other was F, TLF, or P. This pattern of correlations suggests two factors and is in accord with Paivio's theoretical distinction between higherorder meaning and lower-order meaning. It also indicates that, although it is correlated with attributes that vary with word availability, the variable D is more highly related to attributes of higher-order meaning. Rated P, on the other hand, is more directly related to lower-order representational meaning. A final consideration is the degree of relationship between D and L Although the magnitude of the correlation involving the two variables is consistent with the notion that both are affected by common referential processes, the value of the correlation is not unity, suggesting that the psychological processes that underly the two rating measures are not identical. Concreteness-abstractness differences may be related to this discrepancy, imagery, perhaps, constituting the referential meaning of concrete words but not of abstract words. Intraverbal cognitive processes, on the other hand, may correspond to the referential meaning of an abstract word. This interpretation is supported by a significant correlation of verbal associative meaning, indexed by m, with D even when the influence of I was partialled out (r = 0.22, df = 274, P < 0.001). EXPERIMENT 1
So far the assumption has been that Ss make defineability ratings on the basis of knowledge of what words refer to. The assumption was examined by testing the effects of D in a paradigm that explicitly involves knowledge of reference. Dictionary definitions may be considered t o be verbal descriptions of rdatively invar/ant word-referent relations. The following deduction was tested. If D is based on Ss' knowledge of. these relations, then definitions of words rated high in defmeability should be better cues for the retrieval of the word labels than definitions of words rated low in D.
Def'meability as an Index of Word Meaning
293
Method
Sub/ects Thirty-six undergraduate students (16 males, 20 females) attending summer school courses at the University of Western Ontario served as Ss. All Ss were tested in a single group, the testing session lasting 40 rain.
Materials Every fifth item from the original set of 277 items was selected to yield a random sample of 55 nouns. Mean attribute values for the sample are 4.48 (S.D. = 1.32) for I, 4.38 (S.D. = 1.90) for C, 5.52 (S.D. = 0.85) for m, 5.40 (S.D. = 0.78) for F, 37.64 (S.D. = 33.73) for TLF, 5.16 (S.D. = 0.74) for D, and 5.59 (S.D. = 0.58) for P. A definition from one of three dictionaries (The Concise Oxford Dictionary, 1964; Webster's Seventh Collegiate Dictionary, 1965; Everyman's English Dictionary, 1967) was selected for each of the 55 nouns according to its appropriateness as determined by two judges. Booklets containing the definitions, but not their verbal labels, were reproduced. Each booklet consisted of one page of instructions and five pages of definitions with 11 definitions to a page. The definitions were printed on the left side of the page, and adjoining each definition on the right of the page, were four lines numbered 1 to 4. The occurrence of a def'mition on a given page was randomly determined, as was the order of test pages in a booklet.
Procedure The instructions indicated that S was to print beside each definition the word that he considered to be most appropriately described by the definition. If S considered a definition to be appropriate to more than one word he was to list the alternatives, in order of appropriateness, on the numbered lines beside the definition. It was specified that all of the definitions applied to words that were nouns. Results
The percentage of Ss who provided the appropriate word as a response to a definition was calculated for every test item. The percentage of correct responses per word ranged from zero to 97% with an overall average of 36% for the word sample. Correlations between percentages correct and D, I, C, m, F, 7LF, and P were, respectively, 0.52, 0.56, 0.44, 0.43, 0.46, 0.40, and 0.36. All of the
294
O'Neill
correlation coefficients were statistically significant (P
Defineability
as a n I n d e x o f W o r d M e a n i n g
Brian J. O'Neill 1 Received September 9, 1971
A sample o f 277 nouns was scaled for rated defineability and rated pronunciability. lntercorrelations were computed among the following variables: defineability (D), pronunciability (P), imagery (I), concreteness (C), associative meaningfulness (m), familiarity (F), and Thorndike-Lorge frequencies (TLF). Rated D was substantially related to all the other var~bles, notably to rated I. Correlations were highest among D, I, C, and m, on the one hand, and among F, TLF, and P, on the other. The pattern o f correlations was interpreted in terms o f Paivio's (1970] distinction between higher-order and lower-order meaning. Two experiments were conducted to test the relationship between the word attributes and recall o f word labels with dictionary definitions provided as retrieval cue~ The positive effect o f D on label retrieval was independent o f the I values o f the defined words but the effectiveness of I on retrieval was dependent on word defineability.
INTRODUCTION Despite ~he fact that meaning is identified with definition in commonsense thinking, there has been no explicit attempt to scale words in terms of their defineability. The failure to investigate this aspect of meaning is the more surprising in view of the fact that there is no word attribute available which indicates the extent to which subjects (Ss) know what a word refers to. The development of an index of the "referential meaningfulness" of words, or the extent to which Ss know what a word denotes, is important for at least two reasons. In the first place, it should permit an assessment of the degree to which knowledge of referential meaning is implicated in responses This research was supported in part by grants to Dr. A. Paivio from the National Research Council of Canada (Grant 0087). 1Department of Psychology, University of Western Ontario, London, Ontario, Canada. 287 O 1972 Plenum Publishing Corporation, 227 West 17th Street, New York, N.Y. 10011.
288
O'Neill
that are the basis of available attributes of word meaning. These attributes include various indices of associative meaningfulness (e.g., Noble, 1952), rated familiarity, and rated concreteness and imagery (e.g., Paivio, Yuille, and Madigan, 1968) as well as the dimensions that have been identified using the semantic differential rating technique (e.g., Osgood, Suci, and Tannenbaum, 1957). Knowledge of what a word denotes may mediate the responses that contribute to any or all of these attribute values. A correlational analysis of the relationships that exist between an index of word defmeability and other word attributes should suggest the extent to which knowledge of referential meaning is involved in the latter. A second justification for an index of defineability concerns the functional significance of referential meaningfulness. It is possible that the learning difficulty or memory for a word may be related to its referential meaningfulness. That is, whether or not Ss know what a word means may have significant consequences for learning and memory performance. This possibility has not been examined experimentally nor does there seem to have been any attempt to control referential meaningfulness in traditional verbal learning and memory experiments. In the present research a sample of words was scaled in terms of the rated defmeability of the words. It was assumed that a S's evaluation of a word in terms of its defmeability should be directly related to his knowledge of what the word refers to. Thus the referential meanings of words rated high in defineability (D) may be assumed to be dearly known and words rated as low in D may be assumed to be unclear with respect to S's knowledge of their referential meanings. Average D ratings were taken as an index of referential meaningfulness. In order to estimate the implication of referential meaning in available indices of meaning, the relationships between D and several other word attributes were investigated. Of particular interest was the word imagery variable which has been linked to referential meaning by Paivio (Paivio, 1970, 1971; Paivio and O'Neill, 1970). The notion of imagery as reference is not unique (see also Ausubel, 1965; Bugelski, 1970; Mowrer, 1960) and it has a long history. James (1890), for example, has proposed that the static meaning of concrete words consists of sensory images reawakened; the meaning of abstract words, on the other hand, is linked to verbal associative processes. Variables of interest other than rated imagery (/) included rated concreteness (C), verbal associative meaningfulness (Noble's 1952 production m), Thorndike-Lorge frequencies (TLF), rated familiarity (F), and rated pronunei. ability (P). Most of these variables are incorporated in a single theory of meaning in which higher-order meaning is distinguished from lower-order meaning (Paivio,
Defmeability as an Index of Word Meaning
289
1970, 1971). Referential processes, coordinated with the I measure, and associative processes, defined by m, constitute higher-order meaning, the former referring to the symbolic imaginal transformations that are initiated by verbal stimuli, and the latter to primarily verbal symbolic associative reactions. Lower-order or representational meaning is identified with indices of word availability such as frequency counts ( T L F ) and familiarity ratings (F). Rated P was included in the study since pronunciability may be expected to be related to the availability of a verbal response and, therefore, to lower.order meaning. On the basis of Paivio's theory of meaning the defineability measure might be expected to be related more to higher-order meaning attributes rather than to attributes that are coordinated with lower-order meaning, since a defineability rating, presumably, involves a higher-order cognitive judgment.
METHOD Materials and Procedure
The stimulus materials were 277 nouns from the Paivio, YuiUe, and Madigan (1968) norms which contain rated I and C values, production m data (see Nobel, 1952), and T L F values. Mean attribute values and standard deviations for these variables, as well as for rated F, 2 are reported in Table I. Values of 100 and 50 respectively were assigned to AA and A frequencies. Defineability and pronunciability ratings were obtained using essentially the same procedures as were used by Paivio et al. to scale I and C. The nouns were randomized and reproduced in booklets each of which contained a page of instructions and six pages of stimulus words. Each page contained three 18-word columns with a short line printed beside each word. The order in which the page occurred in the booklet was random. The total number of 2Familiarity ratings have been obtained in Dr. Paivio's laboratories at the University of Western Ontario.
Table I. Means and Standard Deviations for Imagery (I), Concreteness (6"), Meaningfulness (m), Familiarity (F), and Frequencies ( T L F ) II
$.D.
iiii
I
C
m
F
TLF
4.31 1.28
4.11 1.82
5.56 1.049
5.15 1.I0
35.72 34.89
290
O'Neill
stimulus items included the original 277 nouns and duplicates of 47 of them. Subjects received one of two sets o f rating instructions. Define.ability Instructions The purpose of this experiment is to rate a list of words on how easy or difficult it is to define their meaning. To do this you are to rate each word as relatively easy or difficult to define using a seven-point scale. A word that is very easy for you to define would get a rating of 7 and a word that is very difficult to define would get a rating of 1. Words that are of intermediate value in terms of ease or difficulty of definition should be rated appropriately between the two extremes. To give you some idea of the procedure, read the first column of words to yourself, giving each word a rating in your mind. When you feel that you can accurately judge the relative difficulty of defining the meanings of the words, go back to the beginning and start writing down your ratings. Write on the line beside each word the number from 1 to 7 that best indicates your ability to define the meaning of the word. Feel free to use the entire range o f numbers, from 1 to 7; at the same time don't be concerned about how often you use a particular number as long as it is your true judgment. Work fairly quickly but do not be careless in your ratings.
Pronunciability Insemctions The purpose of this experiment is to determine the ease or difficulty of pronouncing a number of words. To do this you are to rate each word as relatively easy or difficult to pronounce using a seven-point scale. A word that is very easy to pronounce would get a rating of 7 and a word that is difficult to pronounce would get a rating of 1. Words that are intermediate in pronunciability should be rated appropriately between the two extremes. To give you some idea of the range of pronunciabiIity of the words, read the first column of words to yourself, giving each word a rating in your mind. When you feel that you can judge the relative pronunciability of the words accurately, go back to the beginning and start writing down your ratings. Write on the line beside each word the number from 1 to 7 that best indicates your judgment of the pronunciability of the word. Feel free to use the entire range o f numbers, from 1 to 7; at the same time don't be concerned about how often you use a particular number as long as it is your true ]udgment. Work fairly quickly but do not be careless in your ratings. Six sample items (coffee, dawn, cranium, unreality, tree, emporium) were printed below a rating scale on the instruction page to give Ss practice in making rating judgments and to ensure that they understood the instructions.
Defmeability as an Index of Word Meaning
291
Subjects
The Ss were 40 undergraduate students from introductory psychology classes at the University of Western Ontario. Twenty Ss (eight females and 12 males) rated the words according to the defmeability instructions and 20 (eight females and 12 males) according to the pronunciability instructions. The number of Ss tested at any given time ranged from one to ten.
RESULTS AND DISCUSSION Ease of definition ratings were averaged across Ss to yield mean defineability (D) values for each word) The overall D mean and standard deviation are 4.95 and 0.81 respectively (N= 277). The two sets of means for word duplicates (N = 47) were intercorrelated; the magnitude of the correlation coefficient (r = 0.90) suggests that within.subject ratings are stable. The reliability of the ratings was also estimated by randomly assigning rating booklets to one of two groups of equivalent size (n = 10) and by calculating the degree of correlation' between the mean word values of the two groups of raters. With N = 227, between-group reliability is substantial (r = 0.83). Pronunciability scores were subjected to the same statistical treatment as the D variable. Overall mean and standard deviation values are 5.38 and 0.76 respectively. Within-subject (r= 0.91 for 47 pairs of duplicate values) and between-group (r = 0.84) reliability correlation coefficients suggest that the P ratings are stable. The interrelationships among the variables, defineability (D), pronunciability (P), imagery (/), concreteness (C), meaningfulness (m), familiarity (F), and frequency (TLF), are indicated by the correlation coefficients in Table !I. 3A list of the stimulus words together with def'meabflity (D) and pronunciabflity (P) mean and standard deviation values for each word are available on request. Table lI. Intercorrelations Between D, 1, C, m, F, TLF, and pa
D D I C
m F TLF i
aro.o~s = 0.12.
1
C
0.703
0.640 0.816
rn 0.522 0.561 0.372
F
TLF
P
0.495 0.202 --0.027
0.279 0.164 0.040
0.402 0.308 0.140
0.436
0.256 0.651
0.342 0.613 0.540
292
O'Neill
As might be expected, the intercorrelations involving L C, m, and TLF, are approximately equivalent to the correlations reported by Paivio, Yuille, and Madigan (1968). Rated D correlates more substantially with all the other variables than does any other attribute, suggesting that defineability may be taken as a general index of meaning. A correlation of 0.70 between D and I indicates that imagery is a major correlate of defineability. In order to clarify the interrelationships among the variables, the coefficients were ranked in order of magnitude. Two clusters of correlations were apparent. Correlation values were highest among the variables D, I, C, or rn, on the one hand, and among F, TLF, and P, on the other, whereas the values of the coefficients tended to be lower if one variable was D, L C, or m, and the other was F, TLF, or P. This pattern of correlations suggests two factors and is in accord with Paivio's theoretical distinction between higherorder meaning and lower-order meaning. It also indicates that, although it is correlated with attributes that vary with word availability, the variable D is more highly related to attributes of higher-order meaning. Rated P, on the other hand, is more directly related to lower-order representational meaning. A final consideration is the degree of relationship between D and L Although the magnitude of the correlation involving the two variables is consistent with the notion that both are affected by common referential processes, the value of the correlation is not unity, suggesting that the psychological processes that underly the two rating measures are not identical. Concreteness-abstractness differences may be related to this discrepancy, imagery, perhaps, constituting the referential meaning of concrete words but not of abstract words. Intraverbal cognitive processes, on the other hand, may correspond to the referential meaning of an abstract word. This interpretation is supported by a significant correlation of verbal associative meaning, indexed by m, with D even when the influence of I was partialled out (r = 0.22, df = 274, P < 0.001). EXPERIMENT 1
So far the assumption has been that Ss make defineability ratings on the basis of knowledge of what words refer to. The assumption was examined by testing the effects of D in a paradigm that explicitly involves knowledge of reference. Dictionary definitions may be considered t o be verbal descriptions of rdatively invar/ant word-referent relations. The following deduction was tested. If D is based on Ss' knowledge of. these relations, then definitions of words rated high in defmeability should be better cues for the retrieval of the word labels than definitions of words rated low in D.
Def'meability as an Index of Word Meaning
293
Method
Sub/ects Thirty-six undergraduate students (16 males, 20 females) attending summer school courses at the University of Western Ontario served as Ss. All Ss were tested in a single group, the testing session lasting 40 rain.
Materials Every fifth item from the original set of 277 items was selected to yield a random sample of 55 nouns. Mean attribute values for the sample are 4.48 (S.D. = 1.32) for I, 4.38 (S.D. = 1.90) for C, 5.52 (S.D. = 0.85) for m, 5.40 (S.D. = 0.78) for F, 37.64 (S.D. = 33.73) for TLF, 5.16 (S.D. = 0.74) for D, and 5.59 (S.D. = 0.58) for P. A definition from one of three dictionaries (The Concise Oxford Dictionary, 1964; Webster's Seventh Collegiate Dictionary, 1965; Everyman's English Dictionary, 1967) was selected for each of the 55 nouns according to its appropriateness as determined by two judges. Booklets containing the definitions, but not their verbal labels, were reproduced. Each booklet consisted of one page of instructions and five pages of definitions with 11 definitions to a page. The definitions were printed on the left side of the page, and adjoining each definition on the right of the page, were four lines numbered 1 to 4. The occurrence of a def'mition on a given page was randomly determined, as was the order of test pages in a booklet.
Procedure The instructions indicated that S was to print beside each definition the word that he considered to be most appropriately described by the definition. If S considered a definition to be appropriate to more than one word he was to list the alternatives, in order of appropriateness, on the numbered lines beside the definition. It was specified that all of the definitions applied to words that were nouns. Results
The percentage of Ss who provided the appropriate word as a response to a definition was calculated for every test item. The percentage of correct responses per word ranged from zero to 97% with an overall average of 36% for the word sample. Correlations between percentages correct and D, I, C, m, F, 7LF, and P were, respectively, 0.52, 0.56, 0.44, 0.43, 0.46, 0.40, and 0.36. All of the
294
O'Neill
correlation coefficients were statistically significant (P
