PII: S0911-6044(00)00002-6 Copyright © 2000 Elsevier Science Ltd. All rights reserved.
Word association norms for two cohorts of British adults Katherine W. Hirsh, , a and Jeremy J. Tree, b a School of Psychology, Cardiff University, PO Box 901, Cardiff CF10 3YG, UKb Department of Psychology, University of Plymouth, Drake Circus, Plymouth PL4 8AA, UK Available online 18 December 2000.
Katherine W. Hirsh, , a and Jeremy J. Tree, b
Word association data were obtained from two cohorts of British adults. Young adults (21¯30 years of age) and older adults (66¯81) responded to 90 words in a discrete word association task. An associative frequency measure was calculated by counting how many participants produced a particular word and then converting this number into a proportion. The degree of overlap between the cohorts in terms of dominant responses, the responses with the highest association frequencies, was moderate. Dominant responses were common to the two cohorts for only 36 of the 90 items. When the top three responses were considered the degree of overlap increased to approximately 60%. Four measures of response heterogeneity were calculated for each stimulus item. Comparison of the responses of the younger and older adults indicates that there was less response heterogeneity amongst the older cohort. These norms should be of use to investigators interested in developmental changes in the structure of semantic memory across the adult lifespan as well as to researchers interested in comparing results from neurologically impaired older adults to a normative sample from the same age cohort.
Author Keywords: Ageing; Word association; Semantic memory
There has been a recent upsurge of interest in developmental changes in semantic memory and other linguistic functions (see Burke [1], for a review with respect to language production). For example, the number of studies investigating semantic priming in older adults has multiplied rapidly over the past 10¯15 years [2, 3 and 4]. In part this increase stems from an interest in individuals with suspected Alzheimer's type dementia, a disorder where semantic memory is thought to be compromised ( [5] although see [6] for dissenting view). While some studies have compared the responses of dementing and non-dementing older adults on word association and related tasks [7, 8 and 9], as far as we are aware there are no normative data available on word association in older British adults. We set out therefore to collect data on 90 words from both a young and an older cohort of British adults.
Previous work by Burke and Peters [10] suggests that the level of response variability in word association is not influenced by age. Burke and Peters measured between-participant variability in two ways, firstly by examining the proportion of each participant's responses that were the first or second most popular response to each item and secondly by counting the number of unique responses each participant gave (unique in that no other individual in the cohort gave that response to that stimulus). They found no contribution of participant age to either measure. They did find differences however in the words that were given as the most popular (or dominant) responses. There was only 60.5% overlap in the three most popular responses to stimulus items across the two cohorts.
Burke and Peters [10] also examined variability across cohorts in terms of the type of response given. Following Deese's [11] criteria, they classified responses as either paradigmatic or syntagmatic. Paradigmatic responses were those that shared form class with the stimulus item as well as sharing features in terms of meaning (e.g., boy¯girl, wise¯clever, carrot¯vegetable). Syntagmatic responses were those from a different form class than the stimulus item and as such they were words that could co-occur with the stimulus in a sentence (e.g.,phone¯ring, formal¯dress, blue¯sky). Burke and Peters found that the majority of responses were paradigmatic, and that the proportion of paradigmatic responses was not influenced by age. This finding is important as several studies have revealed a reduction in the proportion of paradigmatic responses in older adults with dementia of the Alzheimer's type (DAT; [7, 8 and 9]) suggesting that such a reduction might be a marker of a breakdown in semantic memory.
In addition to providing normative data we report data on between-participant agreement in our two cohorts. We were interested to determine whether there were age-related changes in response variability. We utilised four agreement measures to assess between-participant variability. We also examined variability in terms of type of response produced. A change in the degree of variability or the type of response given would suggest that there are age-related changes in the structure of semantic memory. If DAT is indeed a function of changes in semantic memory, finding an increase in the degree of variability or a change in the type of response given in older adults would be consistent with the idea that there is a continuum between old age and DAT [12 and 13].
The participants were 90 adult volunteers. Forty-five of them were between 66 and 81 years of age (the older cohort) and the remainder were between 21 and 30 years of age (the young cohort). The educational attainment of the older cohort was: secondary school to age 14 ¯¯ 10 Ps; secondary school to age 15 ¯¯ 13 Ps; secondary school to age 16 ¯¯ 13 Ps; secondary school to age 17 ¯¯ 2 Ps; secondary school to age 18 ¯¯ 1 P; and post-secondary education ¯¯ 6 Ps. All participants in the young cohort were first year postgraduate students in the School of Psychology, Cardiff University. All participants were native speakers of British English.
The materials were 90 written words. The stimulus words were selected either because they were the names of concrete, picturable objects or because it was hoped that they would elicit the name of a concrete object as a frequent response. The norms should therefore be particularly useful to researchers interested in picture comprehension or picture naming. The norms may also be of use to individuals looking to design semantically-based therapy programmes.
Each word appeared on a separate page in 36-point New York font. The first letter of each stimulus word was capitalised. Forty-one of the stimulus words were unequivocally nouns. The remaining 49 items could stand as members of more than one form class. For example, the word blue is primarily thought of as an adjective but it may also be used as a noun or a verb. The frequency of the words ranged from 0 to 312 per million [14] and the length ranged from 3 to 10 letters. Frequency values are available in Appendix B.
Participants were tested individually. They were instructed to write down only their first response to each stimulus word. Items were ordered pseudo-randomly such that neither semantically nor phonologically related words occurred consecutively. No time limit was placed on the participants, however they were encouraged to respond quickly. The instructions to participants were taken from Moss and Older [15]:
In this booklet you will find a list of simple words. Please read each word and then write down the first word it brings to mind. Write your response in the space provided. For example, if the word is butter, the first word you think of might be bread or milk or cup. Please work through the booklet quickly. Remember, we are interested in the word that comes to mind immediately, not after thinking about it for a while. If you can't think of anything at all, leave the space blank. Don't go back and change your mind about any of the words after you have written your first response. Thank you very much for taking part in the experiment.
Participants recorded their responses in writing next to the stimulus words. A typical participant took 15 min to complete the task.
Responses were tabulated and a full list of the stimuli, their associated responses and the proportion of participants who produced each response can be found in Appendix A (unless overwise noted N=45). Following Moss and Older [15] we treated morphological variants of a word as separate responses except in the case of plurals.
Four measures of response heterogeneity (between-participants variability) were computed for each stimulus item. The first of these was the association frequency of the dominant response (ADOM). The dominant response is that produced by the largest number of participants. Association frequency of a response word is the proportion of participants who produced that particular word in response to the stimulus word. Thus ADOM is the association frequency of the word produced most often by participants in response to a target stimulus. Dominant responses and their ADOM values can be found in Appendix B. The second heterogeneity measure was the number of unique responses (NUR) that were given to individual stimulus words, that is the number of responses there were that were produced by only a single participant. The third measure was the number of different responses that were given by more than one participant (NMR), that is it is the number of responses given by multiple participants as opposed to by only a single participant. The final heterogeneity measure was the information statistic H which was calculated according to the following formula:
The H statistic offers information about the distribution of responses. If the responses were uniform the H statistic would be zero. Increasing values of H signal decreasing response agreement and typically a smaller percentage of overlapping responses (i.e., an increase in the number of unique responses). For example consider two items where the dominant response was produced by 30 people and in one case one additional response was produced by the other 15 participants while in the other case three responses were given each by five people. These two items will have an identical dominant response proportion (ADOM) but the latter item will have a higher H statistic than the former (1.45 vs 0.92). Number of unique responses (NUR), Number of responses given by multiple participants (NMR) and H values for each stimulus item may be found in Appendix B.
As can be seen from Table 1 the proportion of young adults contributing dominant responses was lower than the proportion of older adults producing dominant responses. A Wilcoxon Signed Ranks test comparing stimulus ADOM values for the two groups confirmed the significance of this difference (=-3.25, p<0.002). The NUR values were also greater for the young adults than for the old (=-4.58, p<0.0001). The NMR values were greater for the young than for the old (=-2.45, p<0.02). As NUR+NMR is equal to the total number of responses made, these two results indicate that the young adults produced more responses on average to a stimulus word than did the older adults. Table 1 also shows that the H statistic values were higher for the younger adults indicating yet again that there was significantly less agreement amongst this cohort in terms of their associative responses (=-5.24, p<0.0001). Overall the findings from the four agreement measures show that younger adults were less consistent in their responses to the stimulus words than were the older adults, that is, between-participant variability decreased with age.
Table 1. Mean agreement values for the young and the older cohorta (<1K)
We utilised linear regression to examine whether the frequency of occurrence of the stimulus word affected any of the agreement measures. We found no significant relationship between stimulus frequency and any of the agreement measures for the younger cohort. However in the older cohort, log stimulus frequency was related to ADOM (r2=0.14, F(1,88)=15.09, p<0.0005), NUR (r2=0.12, F(1,88)=12.68, p<0.001) and the H statistic (r2=0.13, F(1,88)=14.34, p<0.0005). As the stimulus frequency increased so did the number of unique responses and the H statistic; the association frequency of the dominant response decreased. Thus high frequency items resulted in lower levels of agreement than low frequency items. The dominant responses of both groups were significantly more frequent than the stimulus items that evoked them (older cohort z=-2.58, p<0.01; young cohort z=-3.02, p<0.005). There was no significant difference however between the frequency of the dominant responses produced by the two groups (z=-1.13, p=0.26).
In addition to assessing agreement in a cohort internal fashion, we also examined the degree of agreement across cohorts in terms of the dominant response. There were a total of 36 stimulus items where the two age cohorts had the same dominant response (40%). Of the remaining 54 items there were 13 (14% of total) where the dominant response for one cohort was not produced by at least one member of the other cohort. When the top three responses are considered the degree of overlap between the cohorts increases to 57% (cf. 60.5% for [10]).
Finally, it is worth noting that some responses appeared as the dominant response for more than one stimulus item, that is some responses were repeatedly produced. The older adult cohort produced ache, clothes, dress, fire, pet(s), pond, ring, sky and vegetable as the dominant response to two different stimulus items and flower(s) was the dominant response given to three different stimulus items. Thus although there were 90 stimulus items only 79 unique dominant responses were made by the older cohort. The young adult cohort produced ball, fire and sky as the dominant response to two stimulus items and cat was the dominant response given to three of the stimuli. Thus there were 85 unique responses made by the young participants.
We adopted Bandera et al.'s [7] classification of responses as hierarchical¯categorical or propositional¯relational rather than the more commonly used paradigmatic or syntagmatic. Hierarchical¯categorical (corresponding to paradigmatic) responses were those where the response and the stimulus were from the same form class and included coordinates, subordinates, superordinates, synonyms, antonyms and metonyms of the stimulus. A response was considered a propositional¯relational (syntagmatic) response when the stimulus and response were "only syntactically contiguous within the frame of a sentence...[the response was] an action related to the stimulus, or an attribute of the stimulus, or a noun in a typical sentence containing the stimulus" [7, p.296]. Note that in this system responses classified as syntagmatic may be from the same form class as the stimulus. Classification of the dominant responses was performed individually by each of the authors and then any discrepancies (of which there were only 10 out of a possible 180) were resolved through discussion. Participants in this study produced predominantly propositional¯relational responses. The young cohort produced slightly more propositional¯relational responses than did the older cohort (70 and 60% respectively). The reversal of the typical predominance of paradigmatic-type over syntagmatic-type responses is due to classification without regard to form class. When form class is the decisive factor for classification, the dominance of paradigmatic responses re-emerges: 14% syntagmatic responses for the older cohort and 13% for the young cohort.
We also categorised the propositional¯relational stimulus-response pairs in terms of whether the relationship was one of phrasal collocation (e.g., cave¯man versus grill¯bacon). Twenty-two of the fifty-five propositional¯relational dominant responses made by the older cohort were phrasal collocations. Twenty-nine of the sixty-three propositional¯relational responses made by the young cohort were phrasal collocations. Seven of the phrasal collocations were common to both cohorts: cloak¯dagger, daisy¯chain, drain¯pipe, gas¯fire, peanut¯butter, sly¯fox and tummy¯ache. It is worth noting that a large number of the collocational responses would have been classified as paradigmatic responses because the stimulus and response were from the same form class: alley and cat are not related in the same way as alley and lane however both pairs are composed of two nouns. It was because we felt that phrasal collocations were properly considered "syntagmatic" that we selected Bandera et al.'s classification scheme [7] rather than that of Deese [11].
In Appendix A we have provided a set of nouns that may be used by researchers interested in studying semantic memory in participants who are speakers of British English. Appendix A contains the word association responses of an older and a young cohort to 90 stimulus words and as such it offers age-appropriate normative data for the construction of experimental materials. The importance of utilising age-appropriate normative data is demonstrated through our analysis of cohort-based differences in between-participant variability.
We have shown that there are cohort effects in word association. The degree of overlap between the two groups in terms of their dominant responses showed this very clearly. The same word was produced as the most popular response by both young and old for only 36 of the 90 stimulus items presented. Even when the top three responses to each item were considered the degree of overlap still only reached 57%. When we look at the between-participant variability in responding we also find differences between the cohorts. Younger adults produced a wider variety of responses ¯¯ fewer participants contributed to the dominant response proportion (ADOM), more unique responses were produced (NUR), more non-unique responses were produced (NMR) and as a result the average H agreement statistic for the younger cohort was higher. If we look instead at the response types we see that the two cohorts were quite similar. Whether responses are classified as hierarchical¯categorical/propositional¯relational or paradigmatic/syntagmatic, the two cohorts produced very similar proportions of each response type. Taken together, the lack of agreement in terms of actual responses produced and the similarity in terms of the type of responses produced suggest that cohort effects are not due to changes in the structure of semantic memory; rather, they can be related to differences in the content stored in similarly structured semantic memory systems. The lack of evidence for structural changes to semantic memory in old age suggests that the changes in word association performance seen in dementia are qualitatively different to what is seen in normal ageing.
One implication of these results is that word association data collected from young adults may not be the ideal source of experimental materials for research involving older adults. Cohort differences were most apparent in the collocational stimulus-response pairs. Here only seven dominant responses were common to both cohorts. Some of the differences appear to result from the introduction into the language of new collocations: the dominant response to the word couch was potato in the young cohort but settee in the older cohort; for the word basket it was ball for the young participants and shopping for the older participants. Other differences are more difficult to explain in this way: the dominant response to the word skirt in the young cohort was dress whereas in the older cohort it was blouse. Moreover, even when the same response was dominant the level of agreement differed across cohorts: 47% of the younger cohort produced the word fork in response to the stimulus knife while in the older cohort fork was produced by 73% of respondents.
One intriguing idea is that the higher level of between-participant agreement seen in the older cohort may account for the finding in a meta-analysis that "semantic priming effects are reliably larger for older than for younger adults" (hyper-priming; [3, p.34]). If older participants are in general more likely to produce the dominant response and dominance is a reasonable measure of association strength, then the same item may be a potent prime for 40% of older participants to whom it is presented but only 20% of young participants. In a given experiment then, a larger proportion of older participants would be receiving a highly associated prime and therefore the effects of prime presentation appear to be augmented. This hypothesis, although rather speculative, is amenable to experimental investigation. It would be possible to select from these norms a set of stimuli where the same response was the dominant one for both cohorts but the associative frequency of that response was greater for one of the cohorts. If the hypothesis is correct one would expect to see reliably larger associative priming effects in a sample from the age-cohort that produced the dominant response more frequently be they old or young. These norms will allow experimenters interested in the effects of ageing on language to select materials in such a way as to equate associative frequency for the two age-cohorts. It is possible that such matching could eliminate hyperpriming.
The finding that for the older cohort there was an inverse relationship between frequency of occurrence and three of the agreement measures also deserves some discussion. In a paper examining the effects of frequency and imageability on word association (in young adults), de Groot [16] argued that conceptual representations of high frequency words may be linked to a larger number of other conceptual representations than are the representations of low frequency words. A wider range of links could account for the decrease in proportion of participants providing the dominant response as well as the increase in the number of unique responses: as the number of links increases so too does the number of possible responses and arguably thereby the amount of overlap between choices will decrease if selection is somewhat random. What is difficult to explain however is why this effect occurred only for the older cohort.
Finally, we must note that in addition to differing in age our two cohorts also differed in level of educational attainment. All of the young participants were university graduates, however only one of the older participants was a graduate. This does present a difficulty in interpreting the cohort effects, in that they could be due to age differences or to educational differences. We do not feel however that this lessens the value of the norms we have provided. The level of educational attainment found in our older sample is typical of this age-cohort and therefore the norms provided here are likely to be more appropriate to this group whatever the underlying cause of the cohort differences.
1. D.M. Burke, Language production and aging. In: S. Kemper and R. Kliegel, Editors, Constraints on language: Aging, grammar, and memory, Kluwer, London (1999), pp. 3¯28.
2. N.L. Bowles, Age and semantic inhibition in word retrieval. Journal of Gerontology: Psychological Sciences 44 (1989), pp. P88¯P90. MEDLINE EMBASE
3. G.D. Laver and D.M. Burke, Why do automatic semantic priming effects increase in old age? A meta-analysis. Psychology and Aging 8 (1993), pp. 34¯43. MEDLINE
4. B.A. Ober, K. Shenaut, W.J. Jagust and R.C. Stillman, Automatic semantic priming with various category relations in Alzheimer's disease and normal aging. Psychology and Aging 6 (1991), pp. 647¯660. MEDLINE
5. A. Martin, Degraded knowledge representations in patients with Alzheimer's disease: Implications for models of semantic memory and repetition priming. In: L.R. Squire and N. Butters, Editors, Neuropsychology of Memory (2nd ed.),, Guildford Press, London (1992), pp. 230¯232.
6. R.D. Nebes, Semantic memory dysfunction in Alzheimer's disease: Disruption of semantic knowledge or information-processing limitation?. In: L.R. Squire and N. Butters, Editors, Neuropsychology of Memory (2nd ed.),, Guildford Press, London (1992), pp. 233¯240.
7. L. Bandera, Salla S. Della, M. Laiacona, C. Luzzatti and H. Spinnler, Generative associative naming in dementia of the Alzheimer's type. Neuropsychologia 29 (1991), pp. 291¯304. MEDLINE EMBASE
8. L.R. Gewirth, A.G. Shindler and D.B. Hier, Altered patterns of word association in dementia and aphasia. Brain and Language 21 (1984), pp. 307¯317. MEDLINE EMBASE
9. Pietro M.J. Santo and R. Goldfarb, Characteristic patterns of word association responses in institutionalized elderly with and without senile dementia. Brain and Language 26 (1985), pp. 230¯243. EMBASE
10. Burke DM, Peters L. Word associations in old age: Evidence for consistency in semantic encoding during adulthood, Psychology and Aging 4:283¯92.
11. J. Deese, Form class and the determinants of association. Journal of Verbal Learning and Verbal Behavior 1 (1962), pp. 79¯84.
12. F.A. Huppert and C. Brayne, What is the relationship between dementia and normal aging?. In: F.A. Huppert and D.W. O'Connor, Editors, Dementia and normal aging, Cambridge University Press, Cambridge (1994), pp. 3¯11.
13. E. LaBarge, D.A. Balota, M. Storandt and D.S. Smith, An analysis of confrontation naming errors in senile dementia of the Alzheimer type. Neuropsychology 6 (1992), pp. 77¯95.
14. for Lexical Information Centre. The Celex lexical database, The Max Planck Institute for Psycholinguistics, Nijmegen (1993).
15. H. Moss and L. Older. [bj]Birbeck word association norms, Psychology Press, Hove, UK (1996).
16. A.M.B. de Groot, Representational aspects of word imageability and word frequency as assessed through word association. Journal of Experimental Psychology: Learning, Memory, and Cognition 15 (1989), pp. 824¯845.
Corresponding author. Fax: +44-29-20874858; email: hirsh@cf.ac.uk
Send feedback to ScienceDirectSoftware and compilation © 2001 ScienceDirect. All rights reserved.ScienceDirect® is an Elsevier Science B.V. registered trademark.