Aging and the word frequency effect: A lexical decision investigation

.Yruropsycholoyro. Vol.27.No.9,pp.1197-1703, ,989 Prmtedin Great Bnlam

002X-3932/89 s?.OO+O.OO tr 1989PergamonPress plc

NOTE AGING AND THE WORD FREQUENCY EFFECT: INVESTIGATION MARIE-J• SZPHETAINTURIER,*Q

MARTINTKEMBLAY*S;

A LEXICAL

DECISION

and ANDRE ROCH LECOURS*&~

*Centre de recherche du Centre hospitalier C&e-des-Neiges, Montrtal, Canada; TDtpartement de psychologie, Universitb de Mont&al, Canada; and $Facult& de mtdecine, Universite de Montreal, Canada (Receitzd

16 January

1989; accepted 27 April 1989)

Abstract-It is known that speed and accuracy in recognizing words are constrained by their frequency of occurrence (“frequency esect”). This study bears on the diachrony of the word frequency effect. Our postulates in this respect were (1) that a significant frequency effect should be present throughout adulthood, irrespective of age, and (2) that the magnitude of this effect should be greater among the elderly. Twenty young and 20 older healthy adults were submitted to a lexical decision experiment. Results confirmed our first postulate but invalidated the second one, that is, significant frequency effects were found in both groups but these effects were documented to be of identical magnitude. An attempt is made at explaining the latter result from a theoretical standpoint. The former is interpreted as further evidence that senescence (normal aging) does not interfere with passive. automatic and unconscious mental processes. Moreover, it is suggested that-if observed among otherwise apparently healthy elderly adults-modifications of the frequency effect might be taken as a cognitive marker of disease.

INTRODUCTION IT HAS become

obvious that “classical” psychometric batteries are disqualified as a potential source of information yielding an early diagnosis of the degenerative pathologies of cognition. Consequently, aging research has recently tended to rely on theoretical constructs and methods which n priori have little to do with the notions of senescence and senility. The constructs and methods have arisen from the field of cognitive psychology. They have led to the gathering of an ever-increasing body of pin-pointed data with respect to various aspects of human cogmtion. One might insist that this trend was most welcome since understanding normal cognition in its diachrony is a critical prerequisite to understanding its dysfunctions as they occur at various ages. Moreover, beside providing temporary or enduring answers to theoretically justified questions concerning the structures and functions of the human mind, several of the methods of cognitive neuropsychology might also provide sets of exact data on the diachrony of welldefined aspects of cognitive senescence and there is no obvious reason to predict that they might not also yield early and conceivably preclinical cognitive markers of disease. The subject matter of this note has to do with the senescence of language. It has long been considered, mostly on the basis of WAIS scores, that linguistic abilities are not sensitive to aging 1131. However, a number of studies have recently led their authors to suggest that certain linguistic abilities change with senescence [3]. Thus, GO~DGLASS [9] and others [4, 20, 21, 22, 251 have reported data suggesting that normal elderly adults do show evidence of change in the processes of accessing and retrieving units from the mental lexicon, a suggestion others have challenged [7]. The ontogenesis of one’s mental lexicon no doubt begins when one begins to “learn” words and probably lasts, across one’s lifespan, as long as one keeps making use of and remains exposed to words, occasionally new ones. It

$Correspondence to be addressed to : Marie-Jostphe Tainturier, Laboratoire Tht-ophile-Alajouanine, Centre de recherche du Centre hospitalier Chte-de+Neiges, 4565 Chemin Queen-Mary, Montrbal. Qukbec, Canada H3W-1 WS. 1197

seems reasonable. therefore, to conceive of the mental lexicon as ofa dynamicentity with a diachrony ofits own and to suppose that modifications might occur, with passing time, in its structure and:‘or subserving procedures. Given appropriate testing conditions, such modifications might emerge as differences in the behavior of subjects of various age groups. Indeed, and as mentioned above, such changes have been observed in tasks such as picture naming and verbal fluency 122, 251, which involve an active use of the lexicon. Conversely. no such change-and more precisely no decline-has so far been observed in tasks as lexical decision, which imply a comparatively more passive, automatic and unconscious resort to the mental lexicon [2,5, 14, 151. In brief, it is now believed that normal aging may alter “controlled” although not automatic cognitive processing [12]. More precisely, the subject matter of this note has to do with senescence of the word frequency effect as measurable through lexical decision experiments. The notion of word frequency effect refers to the fact that the speed and accuracy of word recognition processes are constrained by their frequency of use and,‘or exposure 1161. In the present context, we predict that the frequency effect should not be disrupted in normal aging since it is constdered to reflect, specifically, those automatic processes which are believed not to decline with senescence 1121. and might therefore constitute a reliable indication of pathology in cases of perturbation in the elderly. It could however be that the strength of this effect increases with age, i.e. given that growing old in a standard human environment means, almost by definition, progressive accumulation of exposures to words and uses of them. we hypothesize that the word frequency effect might in some manner get “stronger” with aging. Indeed, it is not unreasonable to assume that the rate ofexposure to words across time will be greater for high than for low frequency items. many of which are only encountered occasionally and in very specific contexts. In this perspective, the mental correlate of the word frequency function of older individuals might become steeper than that of younger ones, a phenomenon which should manifest itself in a greater amplitude of the frequency effect among elderly adults. In the current study, we have attempted to explore the above issues by submitting two groups of neurologically healthly subjects (“young” vs “old”) to a lexical decision task bearing on words of various frequencies. Although it has already been shown that older individuals do, like younger ones, take longer to recognize low frequency words 161, this study is to our knowledge the first to focus on the diachrony of the word frequency effect. Apart from the theoretical interest they may convey, such data are of potential usefulness-if only as a source of normative material&to anyone involved in studying lexical access among brain-damaged subjects of various age groups.

METHOD A total of 40 adult right-banders served as subjects in this experiment. They were divided into a subpopulation comprised of 20 “young” subjects (YG Group) and another of 2O”old” subjects (OD Group). Age varied between 62 and 77 years (mean = 67) within the latter. and between 20 and 33 years (mean = 28) within the former. All subjects were nemologically healthy (as assessed through a routine neurological questionnaire in the case of YG subjects. and through routine questionnaire and examination for the OD ones) and none was hospitalized or otherwise Institutionalized at the time of testing. Level of education was comparable within the two groups. All subjects were right-handed native speakers and fluent readers of French: most also had some knowledge of English.

The experimental list that we used in pursuing this research was comprised of 1X4 stirnull (92 words and 92 nonwords) (Appendix I). Forty-six of the word items belonged to the open class and 46 to the closed class. The reason for including an equal number of “function” and “content” words was that we are currently pursuing our study with pathological populations in which this variable is of potential interest. The open- and closed-class stimuli arc paired with one another according to their frequency, literal and syllabic length. as well as graphemic structure. All stimuli are unambiguous as to class membership. Open-class ones are of the non-derived type, i.e. they are not derived from another unit by the application of a productive word formation rule 1261: the unmarked form was selected whenever the notion of inflection was pertinent. The frequency values that were attributed to word-stimuli are those listed for written French by JUIL.I_AIUI) PI rli. [IX]. Items were selected with the purpose of covering as wide a frequency range as possible given our pairing the frcqucncies thus rcpreacnted in our experimental list vary from 0 to 3.600 criteria. Asexpressed in logarithms,,,. (Mcan=2.017, S.D.=0.672). Nonwords were created to match the verbal set with regard to length and structure. This was done by substituting each grapheme ofcvery word item by a grapheme of the same class [e.g. “friend: trini”( ami )= “ona” ( ‘ona )]. The resulting nonwords are all orthographically and phonologically legal. None of them is the homophonc of a I’rcnch or English word.

Stimuli were composed in black-on-white uppercase characters and individually projected at the center of a Macintosh computer screen. Each word remained on the screen until the subject responded. A constant interval of I .5 xc separated each response from the next stimulus presentation. Subjects were orally instructed (a) IO

1199

NOTE

indicate-by pressing either a “yes” or a “no” button-whether or not each stimulus corresponded to a word they knew, (b) to respond with their right index finger, and (c) to respond as quickly and accurately as possible. Testing began with a practice list of 100 items with characteristics similar to those ofthe experimental list. Subjects were aware that this was meant to familiarize them with the procedure and were eventually given verbal feedback as to their performance. They were thereafter informed that formal testing was beginning but were not told that the first 20 stimuli did not belong to the experimental list and would not be taken into account in data analysis. The order ofstimuli presentation varied semi-randomly from one subject to the other (a maximum of three words or nonwords, and of two words of the same class could appear in succession). Stimulus presentation was controlled by a Macintosh computer which also recorded reaction times (in milliseconds) and errors.

RESULTS The mean reaction times (RTs) for the correct responses to each of the words were computed for each group of subjects (Appendix 1). The results ofone of the younger subjects were discarded from the corpus because his RTs and error rates exceeded by more than three standard deviations those of the other individuals in his group. Word outliers were excluded from the analysis using the following cut-off criterion: the RT of a particular subject for a given item was excluded if it exceeded by two standard deviations this subject’s mean RT for all of the other items of the same class. This trimming procedure led to the exclusion of4.5% of the data. Moreover, two items were excluded from this computation because they exceeded the mean error rate for the other items by three standard deviations. In order to keep the list balanced, we also excluded the two words to which they were paired. All missing values were replaced by the subject’s mean RT for a given (open- or closed-class) sublist. Table 1 summarizes the data. Table 1. Mean RTs (in msec), SDS and error rates for young and old subjects

Mean RTs SD Error rates

YG group

OD group

551 25 0.76%

681 26 0.43%

The effect of word frequency on reaction times was evaluated for the YG vs the OD group. This was done by analyzing the degree of correlation (r of Pearson) between the logarithms,,, of word frequencies and the mean RTs expressed in msec. These data and the results of simple regression analysis are summarized in Table 2 and plotted in Fig. 1. The results of these analyses indicate that both groups of subjects showed a significant frequency effect; moreover, as illustrated in Fig. 1, this effect appears to be of similar magnitude in both. Table 2. Effects of frequency

Correlation Regression Intercept Slope

coefficient

on RTs for young

and old subjects

YG group

OD group

-0.46 F= 22.98 P < 0.0001 589 - 18.49

-0.48 F=25.57 P<0.0001 723 ~ 20.27

Another analysis of the frequency effect was done by computing the mean RT’s of the 30 most frequent vs those of the 30 least frequent items (Table 3). A repeated measures ANOVA opposing the two-modality factor “Frequency” and the repeated factor “Subjects” (YG vs OD) showed both a main effect of age [F(l, 58)=2003.71. P
DISCUSSION In summary, our study indicates that, although the lexical decision RTs were found to be longer among older subjects, they and the younger subjects showed a frequency effect of similar amplitude. It should parenthetically be stressed that the longer latencies observed among the elderly adults cannot be

1200

Norr l a.D

8OC

750

700 2 G 5

650

4 5

600

Y 550

0

.5

Fig, I. Magnitude

1

L.oGoFFREau~Y 1.5

of the frequency

2.5

3

3.5

4

effect in the YG and OD subgroups

Table 3. Mean RTs (in msec) of the young :rnd old groups for the 30 most and 30 least frequent items YG group

OD group

s41 564

670 699

__~ Most frequent Least frequent

presumed to retlect lengthelring of access time per se (estimated to be of :he order of I SO mscc in undergraduate students [I]). Lexical decision obviously implies several steps beside acccs~ proper. sumc of which cg. visual analysis, motor programmation and exccutionP~arc known to be age sensitive 1271. Whether or not access time per .$c increases with age thus remains an open question. As predicted and in line with current beliefs regarding the ac.ltom;ltic,controlled dichotomy (cf. .supr~). the word frequency effect was found to remain unimpatred wi!h normal aging. Contrary to our prediction. the amplitude of the frequency effect was not found to increase with aging, which might be interpreted in several uays. One might suggest that our initial hypothesis was correct bu: that an age-linked increase in !he amplitude 01 the word frequency effect would be masked by a RT floor effect in older subjects. In our opinion, :his interpretation is excluded since vve observed a ve:y similar pattern of results when the 20 most frequent items of our list and or the 20 items with the fastest RTs among older adults were excluded from the analysis. Taking our results at face value then, one might argue th,tt the mental Icxicon is not such a dynsmicen!ity after all, and that once individual lexical entries have been ascribed a certain mental frequency value, however and whenever this is supposed to occur, such values remain stable for the rest of one’s lifespan. As a matter of fact, our resu!ta are consistent with the tennets of certain models of lexical access like. for instance. the one claboratcd by FOKSTPK [ %1. This model accounts for the frequency effect by postulating that visual word recognition supposes the serial scanntng of an orthographic access tile in which entries are listed from the moat to the lcast frequent. In this perspective, the vvord-frequency elfects amounts tc a ranking etfcxt and further exposure to words with an ascribed rank within the file should have no impact on recognition time Beyond the ground of our personal preconceptions as to the ontogenesis of the mental lexicon (cf. stcpru), we also exclude thts intcrpretatton because we have recently documented the existence of an interactron between schooling and the word frequency cKect 1241. which we see as clearly incompatible with Forster’s views on the internal corrclatcs of word frcqucncy.

1201

NOTE

As a consequence, it seems to us that the following alternative interpretation of our results can be entertained. In our opinion, it remains reasonable to assume that the rate of exposure to words across one’s lifespan will be proportionally greater for higher than for lower frequency items. Nonetheless, counterbalancing phenomena exerting their influence on low and/or high frequency entries would explain why this does not lead to an increase in the amplitude of the frequency effect with aging. On the one hand, it has been shown that repetition priming-that is, the benefit in decision time associated, for a given stimulus, to prior presentation of the same item-is greater for low than for high frequency words [17], including among normal elderly adults 161. Now, if one attempts to link this finding to the dynamics of the mental lexicon across time, it might be suggested that the subjective frequency of lower frequency entries might increase proportionally faster than that of higher frequency ones, which should in turn, rather than flattening the frequencyeffect slope, contribute to stabilizing the consequence of greater exposure to frequent words with aging. On the other hand. high frequency words might reach an asymptote after which further exposure will have very small if any impact on recognition thresholds, a phenomenon which might also contribute to counterbalance the consequence of greater exposure to high frequency words with time passing. GORDON [IO] has documented that above a certain level offamiliarity, RTs are no longer a linear function of word frequency as attested by the existence ofan asymptote of the reaction times for very high frequency words. This was interpreted as the behavioral correlate of a saturation in the mechanisms subserving sensibility to word frequency. Furthermore, GORDON and CARAMAZZA [l I] showed that, given appropriate testing procedures. this effect can be distinguished from an absolute floor in reaction times. Along this line of argumentation, one might wish to take into account a connectionist simulation in which the difference between recognition threshold for high and low frequency tended to diminish with “training” 1231 which, iflinkable to the natural increment in exposure to words with aging, might entail the prediction that the amplitude of the frequency effect should diminish rather than increase with aging, a prediction to which our results do not conform. The results we obtain can be accounted for by arguing that the asymptote effect and the one of higher rate oi exposure to high frequency words wili tend to neutralize each other. In summary, one of our predictions in the present study was that the word frequency effect would change across time because exposure rates to words are not the same for low vs high frequency items. We argued that this factor aione would entail greater amplitude of the frequency effect among older subjects. This was not verified. For the time being and by reference to our postulate regarding greater diachronic exposure to high-frequency words, we attribute our results to the counterbalancing effects of other lexical phenomena resulting in stabilization of the frequency effect across time (a) by a!lowing the recognition thresholds of lower frequency words to decrease more markedly and!or (b) by curbing those of high frequency words. As underlined in our introduction to this note, it is our impression that theoretically-founded exact cognitive measurements-such as the frequency efiect observed in lexical decision-are more likely to yield early reliable markers of senility than are the data gathered through use of classical psychometric batteries. One of us 1191 has recently attempted to define the notion of cognitive marker as it pertains to the domain of gerontology and geriatrics. and to list, in the ideal. the characteristics of a good cognitive marker. According to his definition, a coynit& mark would be a precise behavioral fact which lends itself to objective measures and can be considered either as characteristic of a given phase in the diachrony of cognitive senescence, or else as a pathognomonic indicator of a disease that interferes with cognition, or again as a reliable witness to the effects of a neuropsychotropic drug. In our opinion, and we are currently verifying this point, it is conceivable that anomalies of the frequency effect and/or of less automatized phenomena, whether generalized to the mental lexicon as a whole or restricted to particular word subsets or to particular modalities. might turn out to be markers ofdisease. hopefully early ones. ALknr)wleciUemenrs~~This research was supported by Grant PG-28 of the Conseil de la recherche Canada and by a Fends de la recherche en santt du Qukbec studentship to the first author.

mkdicale

du

REFERENCES 1. BALOTA, D. A. and CHUMBLEY. J. I. The locus of the word frequency effect in the pronunciation task: lexical access and or production? J Mrm. Larly. 24, 89 -106, 1985. 2. BALOTA, D. A. and DU(.HEK, J. M. Age-related differences in lexical access, spreading activation, and simple pronunciation. Ps~~chol. Aginy 3, 84 93, 1988. 3. BAYI.ES,K. A. and KASNIAK, A. W. Communication and Cognition in Normal Aging and Dumenria. College Hill Press, Boston 1987. 4. BOWLES, N. L.. OBLER, L. K. and ALBERT. M. L. Naming errors in healthy aging and dementia. Corrr.~ 23, 519-524. 1987. 5. BURKE, D. M. and YEE, P. L. Semantic priming during sentence processing by young and older adults. Drrl Psycho/. 20, 903 910, 1984. 6. CLARKE, E. 0. Semantic and episodic memory impairment in normal and cognitively impaired elderly adults. In Language and Communication in the Elderly, L. K. OBLER and M. L. ALBERT (Editors). Lexington Books, 1980.

1202 7

x. 9. IO. II. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 21.

NOTE

FLICKER, C., FERRIS, S. H., CROOK, T. and BARTUS, R. Implications of memory and language dysfunction in the naming deficit of senile dementia. Bruin Lang. 31, 187-200, 1987. FOKSTER, K. I. Accessing the mental lexicon. In /Vew Approuches to Languuyr ~Merhanism.s, R. J. WALES and E. WALKER (Editors), pp. 257 286, New Holland. Amsterdam, 1976. GOOIXLASS, H. Naming disorders in aphasia and aging. In Languayr and Communciation in (he Elderl~~. L. K. OBLER and M. L. ALBERT (Editors). Lexington Books. Massachusetts, 1980. GORDON. B. Subjective frequency and the lexical decision latency function: implications for mechanisms of lexical access. J. Mem. Lany. 24, 631 645, 1985. GORDON, B. and CARAMAZZA, A. Lexical access and frequency sensitivity: frequency saturation and open, closed class equivalence. Coynition 21, 95 115, 1985. HASHER. L. and ZACHS. R. T. Automatic and effortful processes in memorv.. J. L’.x~.Ps,chol.: Gen. 108,356- 388. 1979. HOCHANAIXL., G. and KAPLAN, E. Neuropsychology of normal aging. In C‘limcul Neuro/o~/~ of A