Semantic fluency task with different lexical-semantic categories: performance factors and the problem of generality of the retrieval model

SEMANTIC FLUENCY TASK WITH VARIOUS LEXICAL-SEMANTIC CATEGORIES: PERFORMANCE FACTORS AND THE PROBLEM OF SEARCH MODEL GENERALITY

 

The article presents the results of a study of the role of psycholinguistic variables (age of word acquisition, word frequency, word figurativeness, and familiarity of the corresponding objects) as factors in the performance of a semantic fluency task—the task of listing objects included in a lexical-semantic category (category members) within a certain time frame—and an assessment of the generality of the information retrieval model in the performance of this task, tested in the studies of other authors on a small number of categories. The results are as follows. (1) It was shown that earlier acquired, more frequent, figurative, and familiar category members are mentioned more often in the semantic fluency task. (2) It was not possible to identify a contribution of psycholinguistic variables to the prediction of category productivity—the average number of category members listed by respondents during the semantic fluency task—that is independent of the degree of category clustering. This result can be explained by the small sample of categories, due to the fact that the necessary data were available only for a small number of categories, as well as by the high correlation between productivity and the degree of category clustering. (3) By analyzing individual data on the number of listed members of each category, evidence was obtained in favor of the generality of the information retrieval model, tested on a small number of lexical-semantic categories, when performing a semantic fluency task.

Key words: semantic fluency, psycholinguistic variables, lexical-semantic categories, categorical norms, information search.

A. Grigorjev, E. Lapteva

SEMANTIC FLUENCY TASK WITH DIFFERENT SEMANTIC CATEGORIES: FACTORS OF FULFILMENT AND THE PROBLEM OF GENERALITY OF THE SEARCH MODEL

 

In the article, the results of the study of the role of the psycholinguistic variables (age of word acquisition, word frequency, word imageability and object familiarity) as factors of fulfillment of semantic fluency task, in which objects belonging to a semantic category are generated during certain time, and of estimating of generality of the model of information search when this task is fulfilled, which was tested by other researchers on few categories, are presented. These results are as follows. (1) It has been shown, that more early acquired, more frequent, imageable and familiar category members are generated more often in semantic fluency task. (2) We failed to reveal contribution of psycholinguistic variables over and above the degree of category clustering into the prediction of category potency (an average number of category members generated in fulfillment of category fluency task). This result can be explained by a small sample of categories, which is due to the fact that only for a few categories the necessary data were available, and by strong correlation of potency and the degree of category clustering. (3) By means of analysis of individual data about numbers of generated members for each category, an evidence for generality of the model of information search when semantic fluency task is fulfilled, which has been tested on few semantic categories, has been obtained.

Keywords: semantic fluency, psycholinguistic variables, semantic categories, category norms, information search.

 

The editorial in the first issue of this journal outlines the problems of organizational psycholinguistics [4]. E.F. Tarasov's article in the same issue examines issues related to the development of the theory of this discipline [5]. The research unfolding within it reveals applied problems that can be solved by its means.

In our opinion, one such task is the development of tools for assessing professional suitability for professions that require verbal abilities. This development, of course, must be based on knowledge of the factors influencing the solution of the corresponding test tasks and the mechanisms underlying their execution. This article examines from this perspective one such test task, currently used primarily (unfortunately!) in clinical settings: the semantic fluency task.

In this task, subjects are asked to list, within a certain time limit, objects that belong to a certain lexical-semantic category (category members), for example, to list fruits. In numerous studies with this task, the same lexical-semantic categories are usually used. Most often, only one category is used – «animal» [e.g., 11; 12; 13; 14; 15]. Sometimes, some others are added to it: «fruits and berries» [1; 3]; «birds», «food», «cold dishes» [9]; «fruits and vegetables», «vehicles» [10], etc.

The authors of such studies believe that the obtained results allow us to draw conclusions about the structure of semantic memory and the semantic search process, the structure of the representation of lexical-semantic categories in health and disease, and so on. Thus, the following model of the structure of the representation of lexical-semantic categories and information retrieval during the semantic fluency task was proposed. Category members are grouped in memory into subcategories united by some basis (for example, in the case of the category "animal," these might be "pets" and "large animals"); semantic search consists of finding a subcategory (switching from one subcategory to another), which usually takes time, and listing the category members included in the subcategory, which is accomplished quickly. We will henceforth refer to this model as the two-stage search model.

It turns out that broad conclusions are being drawn based on results obtained in a small number of categories, often just one. How justified is this?

This question seems quite reasonable in light of the data presented in categorical norms, the collection of which actually uses the task of semantic fluency [1]. These norms typically provide indicators for several dozen lexical-semantic categories and their members [see, for example, 2; 7; 17]. One such indicator is the average number of objects listed by respondents per unit of time (we will call this indicator the productivity of a category; in English-language categorical norms, the term potency is used). Categories differ in this indicator, and these differences are characterized by geographical and temporal stability. For example, the productivity of 22 categories in categorical norms for the English language, collected in the 1960s, correlates with the productivity of the corresponding categories in categorical norms for the Russian language, collected in 1999-2001 [2].

The categorical norms for the Russian language [Ibid.] also present data on the degree of category clustering. According to this data, categories differ in the number of clusters (subcategories) into which their members are grouped (these numbers were determined by analyzing respondents' responses using a special algorithm). Calculations show that the number of subcategories in a category is closely related to its productivity. The degree of clustering appears to be an important factor in category productivity, consistent with the two-stage model.

The degree of clustering explains approximately 70% of the variation in productivity categories. What factors might explain some of the remaining 30% of variation? Assessing the effects of some of these possible factors is one of the objectives of this study.

The second objective is to evaluate the generality of the two-stage retrieval model, tested on a small number of lexical-semantic categories. According to this model, retrieving category members from semantic memory consists of finding a subcategory, which requires time, and quickly "scanning" the contents of this subcategory. There are structural differences between categories [Ibid.]. Can we assume that the process of retrieving category members is the same when listing members of structurally different categories? The generality of this model can be tested as follows. If the processes of information retrieval differ for different categories, one can expect interindividual variation associated with the type of category structure: the differences in the retrieval process caused by this variation may provide advantages to individuals who have well-developed relevant skills. This will be expressed in the identification of specific factors in a factor analysis of individual data. If such factors are identified, this will indicate differences in the mechanisms of semantic retrieval for structurally different categories and, accordingly, will not be consistent with the assumption of the generality of the semantic retrieval model. The absence of specific factors in the factor solution will allow the assumption about the generality of the semantic search model to remain in force.

 

Category productivity factors

Psycholinguistic variables such as word frequency and age of acquisition predict the speed of naming picture objects, word reading, and lexical decision [see, e.g., 16]. It can be assumed that these variables also play a role in the performance of the semantic fluency task. They may be a factor in the variability of the probability of naming certain category members and the variability of categories in productivity. We will evaluate their role by considering, first, their relationships with the probability of mentioning category members in categorical norms and, second, their effectiveness as predictors of category productivity in addition to the degree of category clustering. The hypotheses of this study are that (a) the probability of naming category members, estimated from their frequency of naming, will correlate with psycholinguistic variables and (b) psycholinguistic variables make a contribution to the prediction of category productivity that is independent of the degree of clustering.

 

Method

Ten lexical-semantic categories were analyzed: "four-legged animal," "fruit," "vegetable," "musical instrument," "piece of furniture," "piece of clothing," "carpenter's tool," "type of vehicle," "bird," and "kitchenware." The number of category members included in the analysis varied from 6 ("vegetable") to 30 ("four-legged animal"). The selection of categories and their members was dictated by the following: (1) the category must have data on the degree of clustering, and (2) the category members must have values for one of the psycholinguistic variables—the objective age of word acquisition.[2].

Four psycholinguistic variables were considered: the objective age of word acquisition, word frequency, word imagery, and familiarity of the corresponding objects. The values of these variables are taken from [8]. The objective age of acquisition was determined by presenting images of objects to children of different ages; the children were asked to name the objects depicted. Data on word frequency from the frequency dictionary of modern Russian by O.N. Lyashevskaya and S.A. Sharov were used. Imagery and familiarity of words were assessed by interviewing respondents. Imagery of words was assessed by respondents using a seven-point scale (7 – the word evokes an image very easily, 1 – it is very difficult to create an image), and familiarity was assessed using a five-point scale (5 – the object is very familiar, 1 – completely unfamiliar). Data on the frequency of mentioning category members and the productivity of categories are taken from [2]. Obviously, one can expect that the productivity of categories and the frequency of mentioning their members will negatively correlate with the age of acquisition and positively with the frequency and imagery of words, as well as with the familiarity of the corresponding objects.

 

Results

To assess the relationship between psycholinguistic variables and the probability of mentioning category members in categorical norms, the correlation coefficients of the frequency of mentioning category members with four psycholinguistic variables were calculated for all 10 categories. Correlations with age of acquisition ranged from –0.825 to 0.605 (mean –0.362, 95% confidence interval –0.663, –0.060), with frequency from –0.312 to 0.833 (mean 0.304, 95% confidence interval 0.048, 0.560), with imageability from –0.293 to 0.816 (mean 0.383, 95% confidence interval 0.147, 0.619) and with familiarity from 0.224 to 0.873 (mean 0.551, 95% confidence interval 0.405, 0.697). Only two correlations with age of acquisition were positive and only two correlations with frequency were negative. Thus, the obtained results are consistent with what was expected: previously learned, more frequent, figurative and familiar category members are mentioned more often in the semantic fluency task.

To assess the role of psycholinguistic variables as additional predictors of productivity, in addition to the degree of category clustering, we calculated the mean values of category members for four psycholinguistic variables for each of the 10 categories. These means, along with the productivity and degree of category clustering, are presented in Table 1.

[1] For more details on the data collection procedure for categorical norms, see, for example, [2].

[2] A distinction is made between «objective» and «subjective» age of word acquisition: the first is determined by showing children pictures and asking them to name the objects depicted; the second is determined by asking adult subjects about the age at which, in their opinion, they or their children acquire certain words.

Table 1

Productivity and degree of clustering of 10 categories, average values of age of acquisition, frequency, imagery and familiarity of their members

Category	П	SK	Higher School of Economics	Ч	О	З
Quadruped	8,26	11	54,22	27,20	6,62	1,91
Fruit	7,16	7	60,05	12,37	6,76	3,29
Vegetable	6,99	6	60,50	22,15	6,70	3,15
Musical instrument	7,37	5	108,56	12,83	6,36	1,83
A piece of furniture	8,05	9	61,21	143,88	6,67	3,99
Garment	8,66	11	77,28	35,04	6,60	3,22
Carpenter's tool	5,09	0	92,67	9,88	6,25	1,59
Type of vehicle	7,08	7	54,36	87,41	6,61	2,65
Bird	8,45	11	72,18	13,61	6,57	1,63
Cookware	7,53	10	69,38	37,37	6,63	3,43

 

P – category productivity; CC – category clustering degree; VA – average age of category member acquisition; CH – average frequency of category members; A – average imagery of category members; F – average familiarity of category members.

Table 1

Potency and degree of clustering for 10 categories, mean ages of acquisition, frequencies, imageabilities and familiarities of their members

Category	P	DC	AoA	Fr	I	Fam
A four-footed animal	8.26	11	54.22	27.20	6.62	1.91
A fruit	7.16	7	60.05	12.37	6.76	3.29
A vegetable	6.99	6	60.50	22.15	6.70	3.15
A musical instrument	7.37	5	108.56	12.83	6.36	1.83
An article about furniture	8.05	9	61.21	143.88	6.67	3.99
An article of clothing	8.66	11	77.28	35.04	6.60	3.22
A carpenter's tool	5.09	0	92.67	9.88	6.25	1.59
A type of vehicle	7.08	7	54.36	87.41	6.61	2.65
A bird	8.45	11	72.18	13.61	6.57	1.63
A kitchen utensil	7.53	10	69.38	37.37	6.63	3.43

 

P – category potency; DC – degree of category clustering; AoA – mean ages of acquisition of category members; Fr – mean frequencies of category members; I – mean imageabilities of category members; Fam – mean familiarities of category members.

 

As can be seen in Table 1, the categories exhibit significant variation across all psycholinguistic variables. For example, the average age of acquisition of the names of four-legged animals is 54.22 months, while the names of musical instruments are acquired at 108.56 months. Thus, the names of musical instruments are acquired, on average, approximately four and a half years later than the names of four-legged animals. The average familiarity of furniture items is 3.99, while that of carpenter's tools is only 1.59.

The intercorrelations of the indicators from Table 1 are presented in Table 2.

 

Table 2

Intercorrelations of indicators from Table 1

	SK	Higher School of Economics	Ч	О	З
П	0,945	–0,311	0,230	0,545	0,260
SK		-0,500	0,225	0,653	0,318
Higher School of Economics			–0,414	–0,842	–0,476
Ч				0,313	0,591
О					0,700

 

P – category productivity; CC – category clustering degree; VA – average age of category member acquisition; CH – average frequency of category members; A – average imagery of category members; F – average familiarity of category members.

 

Table 2

Intercorrelations of variables from Table 1

	DC	AoA	Fr	I	Fam
P	0.945	–0.311	0.230	0.545	0.260
DC		-0.500	0.225	0.653	0.318
AoA			–0.414	–0.842	–0.476
Fr				0.313	0.591
I					0.700

 

P – category potency; DC – degree of category clustering; AoA – mean ages of acquisition of category members; Fr – mean frequencies of category members; I – mean imageabilities of category members; Fam – mean familiarities of category members.

 

As can be seen in Table 2, the degree of category clustering correlates most highly with category productivity (0.945). This is higher than the correlation obtained for a wider set of categories (22 categories) (0.838). The remaining correlations with category productivity are significantly lower: the next highest correlation is with the average imageability of category members (0.545). The average age of acquisition and the average imageability of category members correlate quite highly (negatively) (–0.842).

To assess whether the psycholinguistic variables under consideration contribute to the prediction of category productivity independent of the degree of clustering, all possible combinations of these variables were entered into the regression equation for its prediction. None of them significantly increased the prediction accuracy. This result, however, cannot be considered as refuting the hypothesis of the influence of psycholinguistic variables on category productivity. With such a small sample of categories (10) and such a high correlation of productivity with the degree of clustering (0.945), identifying a contribution to the prediction of productivity independent of the degree of clustering is extremely unlikely. Unfortunately, the data on the age of word acquisition that we used determined this particular category sample.

Thus, the conducted analysis of the role of psycholinguistic variables showed that psycholinguistic variables correlate with the probability of word naming in the semantic fluency task. However, it left open the question of the relationship between psycholinguistic variables and category productivity.

 

Evaluation of the generality of the two-stage search model

 

Method

To test the generality of the two-stage search model, we used data obtained from processing material collected for the purpose of compiling categorical norms for the Russian language [2]. Specifically, we used individual data on the number of listed members of each category. Data were analyzed for 22 categories: "precious stone," "quadruped," "type of fabric," "color," "type of sport," "science," "fish," "relative," "part of a building," "flower," "tree," "profession," "fruit," "vegetable," "musical instrument," "piece of furniture," "piece of clothing," "carpenter's tool," "type of vehicle," "bird," "kitchenware," and "metal.".

 

Results

The purpose of the analysis was to assess whether a single or multiple information retrieval mechanisms are used when listing members of structurally distinct categories. Results indicating the use of a single mechanism will indicate the possibility of using a unified model to describe information retrieval, in particular, a two-stage retrieval model. Conversely, results indicating multiple mechanisms will indicate the need to develop separate models for different types of category structure.

To assess the extent to which the available data supported one or the other of these possibilities, a factor analysis was conducted. The results of the factor analysis, which revealed a principal factor—analogous to the general factor in studies of the structure of intelligence—and no specific factors, were interpreted as evidence in favor of the unity of the information retrieval mechanism when listing members of structurally distinct categories and, accordingly, the possibility of using a common model. Principal component analysis was used, examining the results obtained without factor rotation—following the practice of identifying a general factor in studies of the structure of intelligence [see 6, p. 52].

Four factors with eigenvalues greater than 1 were identified. The eigenvalues were 9.893, 1.380, 1.176, and 1.018. It can be seen that a large "difference" exists only between the eigenvalues of the first and second factors. These factors explained 44.966, 6.275, 5.343, and 4.628 percent of the total variance, respectively. The fact that the first factor explains more than 40% of the variance is one of the criteria by which it can be considered general [Ibid.].

We also examined the factor loadings of the categories on these factors. The loadings of all categories on the first factor were positive and quite large (the minimum, 0.469, was for the "metal" category). The presence of positive loadings for all or the overwhelming majority of test items on the first factor is the second criterion for qualifying a test as a general one [Ibid.].

These results indicate that a single information retrieval mechanism operates when listing members of different categories.

Only two categories, "carpenter's tool" and "metal," had significant loadings (0.605 and 0.557) on the second factor. Only one category, "type of fabric," had a significant loading (0.652) on the third factor. On the fourth factor, the highest absolute loading was 0.456, and it belonged to the category "science." Based on all of this, one can only surmise the presence of weak specific factors. The most likely of these points to a commonality of specific characteristics between the categories "carpenter's tool" and "metal.".

Thus, the results of the factor analysis clearly indicate that a single information retrieval mechanism operates across categories when performing a semantic fluency task, and provide only weak hints about the operation of other mechanisms. Using a general model to describe information retrieval when listing members of structurally distinct categories thus appears feasible.

 

 

Literature

1. Alfimova М.В. Semantic verbal fluency: normative data and characteristics of task performance by patients with schizophrenia // Social and Clinical Psychiatry. 2010. Vol. 20. No. 3. P. 20-25.
2. Grigoriev, А.А. Representation of lexical categories in the consciousness of a native speaker. Moscow: Institute of Linguistics, Russian Academy of Sciences, 2004. 179 p.
3. Drozdova К.А., Rupchev Г.Е., Semenova Н.Д. Impaired verbal fluency in patients with schizophrenia // Social and Clinical Psychiatry. 2015. Vol. 25. No. 4. Pp. 9-19.
4. Myskin С.В. From the editor // Organizational psycholinguistics. 2018. No. 1 (1). P. 8-10.
5. Tarasov Е.Ф. Towards the construction of a theory of organizational psycholinguistics // Organizational psycholinguistics. 2018. No. 1(1). P. 11–20.
6. Ushakov Д.В. Psychology of intelligence and giftedness. Moscow: Publishing house "Institute of Psychology of the Russian Academy of Sciences". 2011. 463 p.
7. Battig WF, Montague WE.Category norms for verbal items in 56 categories: A replication and extension of the Connecticut norms //Journal of Experimental Psychology. 1969. V. 80.P. 1-46.
8. Grigoriev A., Oshhepkov I.Objective age of acquisition norms for a set of 286 words in Russian: Relationships with other psycholinguistic variables //Behavior Research Methods. 2013.V. 45. No. 4. P. 1208-1217.
9. Gruenwald PJ, Lockhead GR.The free recall of category examples // Journal of Experimental Psychology: Human Learning and Memory, 1980. V. 6. No. 3. P. 225-240.
10. Kavé G., Shalmon M., Knafo A. Environmental contribution to preschoolers' semantic fluency // Developmental Science. 2013. V. 16. No. 1. P. 124-135.
11. 11. Ryu S.-H., Kim RW, Kim S., Park JH, Kim T-H, Jeong H.-G., Kim JL, Moon SW, Bae JN, Yoon JC, Choo IY, Lee DW, Chang SM, Jhoo JH, Kim S.-K., Cho MJ. Normative study of the category fluency test (CFT) from national data on community-dwelling elderly in Korea // Archives of Gerontology and Geriatrics. 2012. V. 54. P. 305-309.
12. Sherman AM, Massman PJ. Prevalence and correlates of category versus letter fluency discrepancies in Alzheimer's disease // Archives of Clinical Neuropsychology. 1999. V. 14. No. 5. P. 411-418.
13. Sumiyoshi C., Ertugrul A., Yagcioglu AEA, Sumiyoshi T.Semantic memory deficits based on category fluency performance in schizophrenia: Similar impairment patterns of semantic organization across Turkish and Japanese patients // Psychiatry Research. 2009. V. 167. P. 47-57.
14. Sumiyoshi C., Matsui M., Sumiyoshi T., Yamashita I., Sumiyoshi S., Kurachi M. Semantic structure in schizophrenia as assessed by the category fluency test: Effect of verbal intelligence and age of onset // Psychiatry Research. 2001. V. 105. P. 187-199.
15. Tombaugh TN, Kozak J., Rees L. Normative data stratified by age and education for two measures of verbal fluency: FAS and animal naming // Archives of Clinical Neuropsychology. 1999. V. 14. No. 2. P. 167-177.
16. Turner JE, Valentine T., Ellis AW. Contrasting effects of age of acquisition and word frequency on auditory and visual lexical decision // Memory & Cognition. 1998. V. 26. No. 6. P. 1282-1291.
17. Van Overschelde JP, Rawson KA, Dunlosky J. Category norms: An updated and expanded version of the Battig and Montague (1969) norms // Journal of Memory and Language. 2004. V. 50. P. 289-335.

 

References

1. Alfimova MV. Semantic verbal fluency: normative data and task performance in a schizophrenic sample // Social and clinical psychiatry. 2010. Vol. 20. No. 3. P. 20-25.
2. Grigor'ev, AA. The representation of lexical categories in the consciousness of a native speaker. Moscow: Institute of Linguistics, Russian Academy of Sciences, 2004, 179 p.
3. Drozdova KA., Rupchev GE, Semenova ND Verbal fluency deficit in schizophrenia patients // Social and clinical psychiatry. 2015. Vol. 25. No. 4. P. 9-19.
4. Myskin SV. From redaktora [Editorial] // Organizacionnaya psiholingvistika. [Organizational psycholinguistics]. 2018. No. 1 (1). S. 8-10.
5. Tarasov EF. To the construction of a theory of organizational psycholinguistics // Organizational psycholinguistics. 2018. No. 1 (1). P. 11-20.
6. Ushakov DV. Psychology of intelligence and giftedness. Moscow: Institute of Psychology of the Russian Academy of Sciences, 2011. 463 p.
7. Battig WF, Montague WE.Category norms for verbal items in 56 categories: A replication and extension of the Connecticut norms //Journal of Experimental Psychology. 1969. V. 80.P. 1-46.
8. Grigoriev A., Oshhepkov I.Objective age of acquisition norms for a set of 286 words in Russian: Relationships with other psycholinguistic variables //Behavior Research Methods. 2013.V. 45. No. 4. P. 1208-1217.
9. Gruenwald PJ, Lockhead GR.The free recall of category examples // Journal of Experimental Psychology: Human Learning and Memory, 1980. V. 6. No. 3. P. 225-240.
10. Kavé G., Shalmon M., Knafo A. Environmental contribution to preschoolers' semantic fluency // Developmental Science. 2013. V. 16. No. 1. P. 124-135.
11. 11. Ryu S.-H., Kim RW, Kim S., Park JH, Kim T-H, Jeong H.-G., Kim JL, Moon SW, Bae JN, Yoon JC, Choo IY, Lee DW, Chang SM, Jhoo JH, Kim S.-K., Cho MJ. Normative study of the category fluency test (CFT) from national data on community-dwelling elderly in Korea // Archives of Gerontology and Geriatrics. 2012. V. 54. P. 305-309.
12. Sherman AM, Massman PJ. Prevalence and correlates of category versus letter fluency discrepancies in Alzheimer's disease // Archives of Clinical Neuropsychology. 1999. V. 14. No. 5. P. 411-418.
13. Sumiyoshi C., Ertugrul A., Yagcioglu AEA, Sumiyoshi T.Semantic memory deficits based on category fluency performance in schizophrenia: Similar impairment patterns of semantic organization across Turkish and Japanese patients // Psychiatry Research. 2009. V. 167. P. 47-57.
14. Sumiyoshi C., Matsui M., Sumiyoshi T., Yamashita I., Sumiyoshi S., Kurachi M. Semantic structure in schizophrenia as assessed by the category fluency test: Effect of verbal intelligence and age of onset // Psychiatry Research. 2001. V. 105. P. 187-199.
15. Tombaugh TN, Kozak J., Rees L. Normative data stratified by age and education for two measures of verbal fluency: FAS and animal naming // Archives of Clinical Neuropsychology. 1999. V. 14. No. 2. P. 167-177.
16. Turner JE, Valentine T., Ellis AW. Contrasting effects of age of acquisition and word frequency on auditory and visual lexical decision // Memory & Cognition. 1998. V. 26. No. 6. P. 1282-1291.
17. Van Overschelde JP, Rawson KA, Dunlosky J. Category norms: An updated and expanded version of the Battig and Montague (1969) norms // Journal of Memory and Language. 2004. V. 50. P. 289-335.