Repetition priming of words and nonwords in Alzheimer's disease and normal aging

Abstract

Objective

This study examines the magnitude and direction of nonword and word lexical decision repetition priming effects in Alzheimer’s disease (AD) and normal aging, focusing specifically on the negative priming effect sometimes observed with repeated nonwords.

Method

Probable Alzheimer's disease (AD) patients (30), elderly normal controls (34), and young normal controls (49) participated in a repetition priming experiment using low-frequency words and word-like nonwords with a letter-level orthographic orienting task at study followed by a lexical decision test phase.

Results

Although participants' reaction times were longer in AD compared to elderly normal, and elderly normal compared to young normal, the repetition priming effect and the degree to which the repetition priming effect was reversed for nonwords compared to words was unaffected by AD or normal aging.

Conclusion

AD patients, like young and elderly normal participants, are able to modify (in the case of words) and create (in the case of nonwords) long-term memory traces for lexical stimuli, based on a single orthographic processing trial. The nonword repetition results are discussed from the perspective of new vocabulary learning commencing with a provisional lexical memory trace created after orthographic encoding of a novel word-like letter string.

There has been considerable interest in repetition priming—defined as faster or more accurate processing of a repeated stimulus due to its prior processing—in elderly normal (EN) and probable Alzheimer’s disease (AD) individuals. In repetition priming paradigms, participants are generally not told about any possible relationship between the initial and later occurrences of stimulus items; thus, these paradigms are implicit tests of memory. This is in contrast to episodic memory paradigms in which specific mention is usually made of previous experience with the stimulus items, and participants are asked to recall previously seen items or to make “yes”/ “no” recognition judgments for previously seen and foil items; these paradigms are explicit tests of memory. Dissociations between repetition priming and episodic/event memory performance have often been found in AD and normal aging, with deficits on episodic memory tasks (to a much greater extent for AD than for EN groups), in the face of normal performance on repetition priming tasks (Craik & Rose, 2012; Zacks & Hasher, 2006). A particularly interesting aspect of lexical decision repetition is that nonwords can be repeated; in this case, a performance decrement is sometimes found for word-like nonwords, presumably because they become more difficult to distinguish from real words in some way (e.g., McKoon & Ratcliff, 1979, Exp. 2).

Prior to describing the design features and specific rationale for the current study of repetition priming in AD, we will provide a brief overview of repetition findings in AD and normal aging, discuss the advantages of utilizing lexical decision as a repetition priming task, and review models of lexical decision and repetition priming.

Brief Overview of the Repetition Priming Literature

Repetition priming has been assessed extensively in AD and normal aging with riddle-like tasks involving word generation or retrieval such as word-stem completion or word-fragment completion, and to a lesser degree with simpler, more passive tasks, usually involving reaction time (RT) measures and relatively more shallow access to lexical-semantic memory. Such tasks (e.g., word identification, word pronunciation, and lexical decision) produce smaller, less consistent deficits in repetition priming due to AD or aging, and sometimes no deficit at all. EN individuals showed repetition priming in RT-based pronunciation or lexical decision tasks equivalent to that of young normal (YN) individuals in, for example, Balota and Duchek (1991), Balota and Ferraro (1996), Light and Kennison (1996), and Ober, Shenaut, Jagust, and Stillman (1991). Contrasting studies in which EN participants showed reduced repetition priming relative to YNs in relatively more conceptually-based tasks are plentiful; for reviews, see Light (2012), Fleischman and Gabrieli (1998) and Fleischman (2007). Examples of AD participants showing repetition priming not differing from that of EN in RT-based pronunciation and/or lexical decision tasks include Balota and Ferraro (1996), Carlesimo, Mauri, Fadda, Turriziani, and Caltagirone (2001) and Ober et al. (1991); however, a study utilizing very long delays between repeated stimuli showed significantly less repetition priming of lexical decision for AD compared to EN (Schnyer, Allen, Kaszniak, & Forster, 1999). Examples of AD participants showing reduced repetition priming compared to EN in relatively more conceptual-level and/or word-retrieval-based tasks, include Fleischman et al. (2005) and Mitchell and Schmitt (2006); for reviews, see Fleischman and Gabrieli (1998) and Fleischman (2007). Several within-subject studies provide evidence for a dissociation such that AD patients show less-than-normal repetition priming effects when the test task requires word production, but normal priming when it does not (Fleischman & Gabrieli, 1998; Fleischman et al., 2001; but see Lazzara, Yonelinas, & Ober, 2001).

Lexical Decision: Advantages as a Repetition Priming Task

RT-based lexical decision, in contrast to word-stem or word-fragment completion, is an ideal task for the assessment of lexical repetition priming in neuropsychological populations with word-finding problems (such as AD and EN) because no word retrieval is required. There is ample evidence that AD patients perform at comparable-to-normal levels of accuracy on two-choice (“word” vs. “nonword”) lexical decision, even though their RTs are relatively long (e.g., Dunabeitia, Marin, & Carreiras, 2009; Perri et al., 2003). In fact, AD patients produced comparable-to-normal RT and accuracy in a “go-no-go” lexical-decision task in which a button press was made only to real words (Ober, Shenaut, & Reed, 1995). Moreover, the lexical decision task requires the use of nonword targets as foils, so priming of both words and nonwords can be evaluated if desired, with the standard two-choice task.

In spite of these advantages, there have been relatively few studies of repetition priming of lexical decision in normal aging, and fewer still that included AD patients. The available evidence suggests that both AD and EN groups have intact repetition priming for both nonwords and words, although the evidence is much more limited for nonwords than for words. This is because the majority of experiments that have used the lexical decision task with AD patients were actually semantic priming experiments focused on priming due to associative relations between words, each critical target word being used once in a semantically related pair and once in a semantically unrelated pair. This systematic repetition of target words provided the opportunity to assess repetition priming for words but not nonwords, the latter not having been repeated systematically. When (word) repetition priming has been assessed within this type of semantic priming paradigm, it has generally been found to be equivalent for AD, EN, and YN groups (e.g., Ober & Shenaut, 1988; Ober et al., 1991).

Models of Lexical Decision and Repetition Priming

The most pervasive element in models of lexical decision is some notion of familiarity. For example, in the two-stage model of Balota and Chumbley (1984), the initial stage involved rapid computation of familiarity, with relatively high familiarity biasing toward a fast "yes" (word) response and relatively low familiarity biasing toward a fast "no" (nonword) response; a second stage occurred whenever the familiarity fell between the participant’s criteria for a word compared to nonword response, and this stage involved more analytic processing of the lexical/semantic properties of the stimulus before a (necessarily slower) response was made. Familiarity has long been thought to be incorporated into the representation of lexemes in some way that lowers their activation threshold (Forster, 1978; Morton, 1969), either absolutely as part of the permanent lexical structure due to the frequency of access over a long period of time, or perhaps also relatively as the result of recent access (Coane, Balota, Dolan, & Jacoby, 2011). Models of lexical decision that are focused more directly on repetition priming of lexical decision, however, have emphasized the levels at which the stimuli are processed during the study and test phases (Duchek & Neely, 1989), or the degree to which identical cognitive processes are used for the study and test tasks (e.g., Franks, Bilbrey, Lien, & McNamara, 2000). For example, in experiments focusing directly on nonword priming, Zeelenberg, Wagenmakers, and Shiffrin (2004) demonstrated that lexical-decision priming for nonwords was inhibitory when a letter-height task was used at study, but facilitatory when a lexical-decision task was used at study. Their explanation was based on a two-process model, with a relatively fast familiarity assessment followed by a slower episodic retrieval of the memory of making either a “word” or “nonword” response during the study phase. Their study also included a second experiment using lexical decision during both study and test, which found negative nonword priming when participants were asked to respond very quickly, and positive nonword priming when they were asked to respond more slowly. The quick-response condition was interpreted as having given preference to the familiarity process, whereas the slow-response condition was said to allow the retrieval of the prior response (to the stimulus during the study phase). It is important to note the contrast between the non-familiarity component of Balota and Chumbley's two-stage model and the non-familiarity process of Zeelenberg et al.’s two-process model: in the former, access to permanent lexical/semantic memory is assumed, while in the latter, it is access to a recent, episodic memory that is used. In any case, all models of repetition priming of lexical decision involve the modification or creation of long-term memory traces representing studied items that can affect access to those items at the time of test.

Specific Rationale for the Current Study

A study by Balota and Ferraro (1996) assessed repetition priming of lexical decision in AD and normal aging without embedding lexical decisions into a semantic priming task. Their study—which focused on word-frequency effects—employed a rhyme decision task in the study phase, in which participants indicated whether pairs of stimuli rhymed or not (e.g., cause/laws versus cause/raise for words, spart/cart versus spart/hurt for nonwords), with both word and nonword rhyme items being potential targets in a later lexical decision task. The repetition priming effect, as assessed by lexical decision RT, was positive for words and negative for nonwords (slower lexical decision RTs for repeated nonwords is more likely when the study task does not involve lexical decision, as will be discussed later). Most importantly, the study found that the magnitudes of (negative) nonword and (positive) word repetition priming did not differ across AD, EN, and YN participants, providing evidence that neither normal aging nor AD impairs the ability to create the necessary memory traces for nonwords.

Given that episodic memory is significantly impaired in AD, whereas semantic memory and implicit memory are relatively preserved, we can assume that the memory processes underlying preserved repetition priming of lexical decision are unlikely to depend on episodic memory; on the other hand, the relative preservation of lexical/semantic knowledge would allow it to be used normally. The previously published study most similar to ours is that of Balota and Ferraro (1996). However, our study differed from theirs in two principal areas: the lexicality of the study task and the overall similarity of word versus nonword stimuli. Instead of rhyme decision, which is relatively likely to involve lexical access, our study task involved matching substrings within word and nonword strings in order to minimize word-level processing of the stimuli by focusing the participants on letter-by-letter (orthographic) processing. Our nonword stimuli were obsolete English words, and our words were all of low familiarity/low frequency, our intention being to maximize the similarity of words and nonwords in form and in familiarity. Moreover, unfamiliar, low-frequency words show significantly longer lexical decision RTs—and significantly greater repetition priming effects—than high-frequency words in YNs (e.g., Balota & Spieler, 1999; Scarborough, Cortese, & Scarborough, 1977), ENs (Balota & Ferraro, 1996), and very mild AD patients (but not mild-to-moderate AD, per Balota & Ferraro, 1996). These experimental design features were intended to provide a strong test of whether AD and EN individuals differ from YN individuals in their ability to modify (in the case of words) or create (in the case of nonwords) long-term memory traces, on the basis of only relatively shallow orthographic processing of lexical stimuli, such that these traces would facilitate later processing of these stimuli.

Study task

In contrast to the phonological, word-level study task was used by Balota and Ferraro, we were interested in using a sub-word level orthographic study task comparable to the letter-height study task used by Zeelenberg et al. (2004) with YN participants, in which the letter string was simply scanned with no requirement of any processing at the lexical level. Eliminating the requirement for word-level processing at study would provide a test of whether participants form new lexical/semantic memory traces for nonwords when there is unlikely to have been any prior conscious lexical processing of them. Unfortunately, the letter-height task, which requires participants to ascertain whether a mentally-generated, lower-case version of an upper-case (presented) stimulus has more ascending letters (e.g., h, t) or descending letters (e.g., j, q), could be difficult for AD patients due to their deficiencies in working memory. The study task we decided to use here, substring matching, had the same orthographic-scanning characteristic as the letter-height task, but did not require participants to hold a different version of the presented stimulus in working memory.

Nonword characteristics

Nonwords that closely resemble words tend to show repetition priming effects of greater magnitude than those that do not (Stone & Orden, 1993; Wagenmakers, Zeelenberg, Steyvers, Shiffrin, & Raaijmakers, 2004), keeping in mind that, depending on experimental parameters, the nonword repetition priming effect may sometimes manifest as facilitation and sometimes as inhibition (Zeelenberg et al., 2004). The usual method of generating word-like nonwords has been to modify real words by replacing one or two letters, as was done in Balota and Ferraro (1996), but this approach produces nonwords that may violate standard English letter-sequence frequencies, and presentation of such nonwords has been shown to activate, under some circumstances, the source words from which they were derived (Deacon, Dynowska, Ritter, & Grose-Fifer, 2004). To avoid these problems in the current study, we used obsolete English words, that is, words that once were active in the language, but which have been considered obsolete for 100 years or more; this results in words that “feel” much more like low frequency words than many nonwords produced by altering current words. They are generally more likely to contain familiar morphology and perhaps a few participants may have seen one or more of them in a historical text or dictionary, or as a cognate word in a foreign language. In any case, previous testing has shown that participants have little or no trouble classifying them correctly as nonwords, at least on first presentation (Ober et al., 1991).

Word frequency

Low frequency words benefit more from repetition than high frequency words (Balota & Ferraro, 1996; Duchek & Neely, 1989). This effect has usually been interpreted as indicating that high frequency words are already quite familiar, and therefore do not become more familiar from recent presentation. In order to maximize the potential amount of word priming in the current study, we used only low frequency words, whereas Balota & Ferraro used half high frequency and half low frequency words, because word frequency was a critical variable in their study. Also, the use of only unfamiliar, low-frequency words further reduced the difference in familiarity between words and nonwords in our study compared to theirs.

Methods

Participants

A total of 113 native speakers of English participated in this study: 30 who had been diagnosed with probable AD, 34 EN, and 49 YN. Demographic and other information about the participant groups is presented in Table 1. The AD individuals were recruited from the University of California, Davis, Alzheimer’s Disease Clinical Centers (Martinez and Sacramento sites). All met the NINCDS-ADRDA (McKhann et al., 1984) criteria for probable AD. None of the AD participants had a history of psychiatric treatment, drug or alcohol abuse, heart disease, uncontrolled high blood pressure, stroke, or head injury; none were taking any medications known to affect cognition (including cholinesterase inhibitors). The AD participants were mildly-to-moderately demented, with scores on the Mini-Mental State Examination (Folstein, Folstein, & McHugh, 1975) ranging from 12–28.

Table 1.

Mean (SD in parentheses) Age, Education, MMSE, and NAART for each Participant Group

	N	Agea	Educ.	MMSE	NAARTb
YN	49	19.3 (1.6)	13.1 (1.2)	---	26.41 (8.31)
EN	34	71.5 (6.5)	15.4 (2.4)	29.5 (0.7)	16.41 (8.66)
AD	30	78.4 (6.1)	13.3 (2.4)	22.2 (3.9)	27.31 (10.28)

Note. YN = Young Normal; EN = Elderly Normal; AD = Alzheimer’s Disease; MMSE = Mini-Mental Status Exam (number correct out of 30); NAART = North American Adult Reading Test (number of errors, i.e., incorrectly pronounced words, out of 61).

^aA smaller sample of EN participants that was more closely age-matched to the AD group was utilized in a re-analyses of the EN and AD experimental data; details are included in the results section.

^bThe NAART was unavailable for one AD participant, due to experimenter error.

The EN participants were recruited from the community, via advertisements posted in senior centers and placed in community newspapers. All of the EN individuals met the same exclusionary criteria as the AD participants and none had any complaints about memory or other cognitive functions; they all had scores on the MMSE in the range of 27–30. The YN participants were recruited from among research subject pools in Psychology and Human Development at the University of California, Davis. These younger participants met the same exclusionary criteria as the EN participants. All participants were given the North American Adult Reading Test (NAART; Spreen & Strauss, 1991), which involves the pronunciation of irregularly spelled words, and which is strongly correlated with verbal IQ. The AD and EN participants received a nominal fee, whereas the YN participants received course credit for their participation. The study was approved by the Institutional Review Boards of the VA Northern California Health Care System and the University of California, Davis.

Stimulus Materials

A set of 50 low-familiarity, low frequency word stimuli were collected from the MRC Psycholinguistics Data Base: Machine Usable Dictionary, Version 2 (Wilson, 1988). The words ranged from 217 to 448 (median 411) in the MRC composite familiarity norm, and ranged from 4 to 15 (median 5) in the MRC Francis and Kucera (1982) frequency norm. Examples of word stimuli are ardent, kerosene, sonata, and tweed. In addition, 50 pronounceable nonwords were collected from a list of obsolete words (Ober et al., 1991). No nonword appeared either in the familiarity norm or in the Francis and Kucera norm. They were marked as obsolete in the Oxford English Dictionary (1971) and were found neither in Webster's Modern (1902) or New Collegiate (1981) dictionaries. Examples of the nonwords are: chalon, famble, jocant, and sanglier. Both word and nonword stimuli had from five to eight letters and from one to four syllables.

Items were divided into two counterbalance sets such that each stimulus was used equally often as a studied and an unstudied item (see Table 2). Six words and nonwords were randomly selected to be used as fillers, two words and two nonwords in each of the three phases, the same filler items being used in both counterbalancing sets. The remaining 44 words and 44 nonwords were divided equally and randomly between the two sets for the study phase, resulting in 22 words and 22 nonwords for each set. Three-letter matching substrings of each item were selected randomly (e.g., sonata: nat), then modified by random letter substitution to produce three-letter substrings that did not match the item by only one letter (e.g., nat: nbt). Of the 22 studied words and 22 studied nonwords in Set 1, 10 words and 10 nonwords were randomly assigned to become studied items in the practice phase of Set 1 and the unstudied items in the practice phase of Set 2; 12 words and 12 nonwords were randomly assigned to become the studied items in test phase of Set 1 and the unstudied items in the test phase of Set 2. The same procedure was used in the opposite direction for the 22 words and 22 nonwords in the study phase of Set 2. Subjects were assigned sequentially to Set 1 or Set 2, and items within the lists were shuffled randomly for each participant, except for filler items, which always came first.

Table 2.

Counterbalancing scheme for study, practice, and test phases

	Set 1	Set 2
Study Phase	Fillers for study phase (2 words, 2 nonwords)
	P1: Studied items for practice phase, Set 1 (10 words, 10 nonwords)	P2: Studied items for practice phase, Set 2 (10 words, 10 nonwords)
	T1: Studied items for test phase, Set 1 (12 words, 12 nonwords)	T2: Studied items for test phase, Set 2 (12 words, 12 nonwords)
Practice Phase	Fillers for practice phase (2 words, 2 nonwords)
studied:	P1 (10 words, 10 nonwords)	P2 (10 words, 10 nonwords)
unstudied:	P2 (10 words, 10 nonwords)	P1 (10 words, 10 nonwords)
Test Phase	Fillers for test phase (2 words, 2 nonwords)
studied:	T1 (12 words, 12 nonwords)	T2 (12 words, 12 nonwords)
unstudied:	T2 (12 words, 12 nonwords)	T1 (12 words, 12 nonwords)

Note: The same filler items are used in both sets; all other items are used in one set as a studied item and in the other as an unstudied item, as indicated by the labels P1, P2, T1, and T2. Studied and unstudied items within a phase were shuffled together before each run.

Design and Procedure

A 3 Group (AD, EN, and YN) × 2 Lexicality (word vs. nonword) × 2 Repetition (studied vs. unstudied) design was employed, with lexicality and repetition being manipulated within subjects. The experimental procedure involved three phases, with a brief pause (about 1 min) between them (see above and Table 2 for details on stimulus selection and counterbalancing). In the first (study) phase the participants performed a substring matching task. Each of the study phase stimuli was presented on a computer screen with a three letter string beneath it. Participants were instructed to press a large green button (yes) if the lower string was contained in the longer string above it, and to press a large red button (no) if the lower string was not contained in the longer string. For half of these trials “yes” was the correct answer (e.g., flyer - yer, brike - rik); for the other half, “no” was the correct answer (e.g., mutton - dut, drazel - drx). The second and third phases consisted of two-choice lexical decision trials, in which subjects were instructed to respond “yes” or “no” depending on whether a stimulus was a word. The second session was a practice session for the third, which was the critical test phase. One half of the studied words and one half of the studied nonwords had been presented with a matching substring, while the other half had been presented with a nonmatching substring during the study phase. None of the test phase stimuli had been utilized in the practice phase. Participants were not given any information about the relationship between the substring and lexical decision tasks.

Results

The 113 participants that were included in the data analyses had all achieved accuracy levels of 75% or more on the critical word and nonword trials. Additional participants were tested who did not meet the predetermined accuracy criteria; these included five YN, one EN, and seven AD participants. In order to reduce RT variance due to outliers in the data for the final 113 participants, a data trimming procedure was employed in which the trials with the highest and the lowest RT in each cell of the experimental design were eliminated for each participant, as suggested by Bush, Hess, and Wolford (1993).

The mean RT priming effects (unstudied - studied) for the three groups across all experimental conditions are displayed in Figure 1. As can be seen in the figure, the pattern of results for RT priming is that all three groups showed evidence of a numerically positive priming effect for words (i.e., shorter RT for studied words than for unstudied words) and a negative priming effect for nonwords (i.e., longer RT for studied nonwords compared to unstudied nonwords). In order to address the primary question as to whether word and nonword repetition priming of lexical decision RTs is modulated either by age or by dementia, the AD and EN groups were included in a first set of ANOVAs (dementia-related effects), whereas the EN and YN groups were included in a second set of ANOVAs (age-related effects). Secondary analyses included: an analysis of lexical decision errors, a test of the effect of the seven-year mean age difference between our EN and AD groups, a test of the effect of general slowing due to aging and dementia, and a test to determine the extent to which individual groups manifested RT priming effects for words and nonwords. All effect sizes are expressed in terms of Cohen’s d statistic; Cohen has suggested that d ≥ .20, .50, and .80, be interpreted as small, medium, and large effects, respectively (Cohen, 1988; pp. 24–27).

Figure 1.

Word and nonword reaction time priming effect means (±SE) for AD (Alzheimer's Disease, N = 30), EN (Elderly Normal, N = 34), and YN (Young Normal, N = 49) participants.

Analysis of Differences Due to Age

The RT means are presented in Table 3. A 2 Group (YN, EN) × 2 Lexicality (word, nonword) × 2 Repetition (studied, unstudied) ANOVA revealed the expected overall longer RT for the EN compared to YN groups, F(1, 81) = 12.2, p = .001, d = .78, and the expected longer RTs for nonwords versus words, F(1, 81) = 68.7, p < .001, d = 1.8. The overall effect of repetition was negligible, F(1, 81) < 1, due to the positive repetition effect for words canceling out the negative repetition effect for nonwords. The Lexicality × Repetition interaction effect was quite robust, F(1, 81) = 20.1, p < .001, d = 1.0, and reflects the opposite effects of repetition on words versus nonwords. All of the remaining interactions were insignificant, including Group × Lexicality, and Group × Lexicality × Repetition, both with F(1, 81) < 1. In sum, the increase in RT for nonwords versus words was proportionate for the YN and EN groups, and the degree to which the repetition priming effect was reversed for nonwords (negative priming) versus words (positive priming) was not modulated by age group.

Table 3.

Mean RT and Error Rate (SD) Across Groups and Repetition Conditions

		Words		Nonwords
		Repeated	Not Repeated	Repeated	Not Repeated
RT (ms)	YN	680 (114)	719 (109)	852 (239)	821 (201)
	EN	824 (183)	867 (216)	1006 (295)	971 (319)
	AD	1140 (363)	1205 (395)	1797 (1091)	1686 (870)
Errors (%)	YN	6 (7)	11 (10)	10 (9)	8 (9)
	EN	3 (5)	3 (5)	3 (5)	6 (9)
	AD	4 (7)	6 (7)	11 (11)	9 (12)

Note. YN = Young Normal; EN = Elderly Normal; AD = Alzheimer’s Disease

Of secondary interest were the error data, which are also presented in Table 3. The error rate was low, ranging from 3 to 11 percent across experimental conditions. A Group × Lexicality × Repetition ANOVA showed that EN were more accurate than YNs, F(1, 81) = 28.3, p < .001, d = 1.2; a marginal two-way interaction of Lexicality × Repetition, due to repeated words showing overall fewer errors than nonrepeated words, in contrast to equivalent levels of accuracy for repeated versus nonrepeated nonwords, F(1, 81) = 3.79, p = .052, d = .43; and a robust three-way interaction of Group × Lexicality × Repetition (absent in the analysis of the RT data), F(1, 81) = 10.3, p = .002, d = .71. The three-way interaction was due to YN participants showing the same pattern of positive repetition priming for words versus negative priming for nonwords as they showed for the RT data, while EN participants showed no priming for words and positive priming for nonwords, a pattern different from that exhibited in the RT data. There was a marginal indication of a repetition effect, with repeated items showing greater accuracy than nonrepeated items, F(1, 81) = 2.80, p = .094, d = .37; all remaining Fs < 1.

Analysis of Differences Due to Dementia

We now turn to the RT data for the EN versus AD groups (see the relevant rows of Table 3). As expected, the AD group responded more slowly than the EN group across all experimental conditions, F(1, 62) = 22.0, p < .001, d = 1.2, and there were longer RTs for nonwords compared to words, F(1, 62) = 26.6, p < .001, d = 1.3. The overall effect of repetition was negligible, F(1, 62) < 1, due to the positive effect of repetition for words canceling out the negative effect of repetition for nonwords. As was the case for the YN and EN groups’ combined data, the analysis on the EN and AD groups’ combined data showed a significant Lexicality × Repetition effect, F(1, 62) = 5.75, p = .018, d = .61, due to the opposite effects of repetition on words versus nonwords. The AD group showed a disproportionate overall increase in RT for nonwords versus words (563 ms, a 45% increase) as compared to the EN group (142 ms, a 17% increase), which was reflected in a significant Group × Lexicality interaction effect, F(1, 62) = 10.3, p = .002, d = .81. The remaining interaction effects, including the important Group × Lexicality × Repetition effect, produced Fs < 1. In summary, the AD group exhibited greater slowing for nonwords compared to words than did the EN group, and most importantly, the degree to which the repetition priming effect was reversed for nonwords (negative priming) versus words (positive priming) did not differ for the AD compared to EN participants.

The error rate ranged from 3 to 12 percent across experimental conditions. The Group × Lexicality × Repetition ANOVA indicated that the AD group showed lower accuracy than the EN group, F(1, 62) = 9.09, p = .004, d = .77; words showed higher accuracy than nonwords, F(1, 62) = 10.9, p = .002, d = .84; and the AD group made disproportionately more errors on nonwords than words, compared to the EN group, F(1, 62) = 3.69, p = .056, d = .49, for Group × Lexicality. The three-way interaction was insignificant, F(1, 62) = 2.61, p = .107, d = .41. but at a level that did not completely exclude the possibility that the EN group's greater accuracy for studied than for unstudied nonwords differed from the AD group's pattern in the opposite direction. This would be parallel to the same interaction found for EN compared to YN. All remaining main effects and interactions were insignificant (Fs < 1).

Possible Effects of an Age Difference Between the EN and AD Groups

Given the almost seven-year age difference between our AD and EN samples (see Table 1), we reanalyzed the RT data with a smaller sample of EN participants more closely matched on years of age to the AD group. The youngest of the EN participants were dropped until we ended up with a group of 21 ENs with a mean age of 75.7 (SD = 4.1), which resulted in t (49) = 1.73, p = .086, for the age difference (2.7 years) between AD and ENs, in contrast to t (62) = 4.34, p = .001, for the age difference (6.9 years) between the full sample of 34 ENs and the AD group. The pattern of results was nearly identical for this as compared to the full-sample analyses. The only differences were that the F values were somewhat decreased (and corresponding p values were somewhat increased), presumably due to a decrease in power with the smaller sample.

A parallel ANOVA was performed on the error rate date with the smaller sample of 21 EN participants who were matched more closely in age with the AD; the results were virtually identical to those using the full sample of EN participants, including the direction of the EN group's accuracy difference for studied versus unstudied nonwords, except for a minor reduction of F values and slightly increased p values.

Possible Effects of Age- and Dementia-Related General Slowing

To examine possible effects of general slowing on the between-group RT priming results, we applied a method suggested by Madden, Nebes, and Allen (1992). First, in order to get the best possible estimate of the task-complexity function of the YN and AD groups relative to the EN group using the data available in this experiment, we computed the group means of eight different types of correct trials: for the study phase task (substring matching, not semantic), we computed the mean RT for word/match, word/no-match, nonword/match, and nonword/no-match; for lexical decision (semantic), we included both the practice block and the experimental block, and computed the means for studied/word, studied/nonword, unstudied/word, and unstudied/nonword. Then, using those eight variables, we computed linear regression equations for EN and AD, AD = −69.3 + 1.68 (EN), r² = .974, and for EN and YN, YN = 287 + 0.543 (EN), r² = .987. We then transformed the YN and AD lexical decision data by reversing the equations (i.e., subtracting the mean and dividing by the coefficient), and repeated the analyses described earlier on aging and dementia. For overall RT, neither the AD group nor the YN group differed from the EN group, Fs < 1. For all other factors and interactions, the pattern of results was identical to that of the original analysis, suggesting that general slowing had little if any impact on the results.

Priming in All Groups Combined and Within Individual Groups

In spite of the pattern of effects reported above, it is still of interest to determine whether individual groups produced positive RT priming for words and negative priming for nonwords. As a preliminary, we looked at word and nonword RT priming for all three groups combined in a pair of 3 Group (AD, EN, YN) × 2 Repetition (studied, unstudied) ANOVAs, using just word or just nonword data. There was a significant overall positive priming effect for words, F (1,110) = 15.9, p < .001, d = 0.76, and also a significant (but less robust) overall negative priming effect for nonwords, F(1, 110) = 4.75, p = .029, d = 0.42.

Next, we performed a series of t-tests on priming within each individual group (means and standard errors are displayed in Figure 1). YN participants showed significant priming for words, t(48) = 2.86, p = .006, but insignificant priming for nonwords, t(48) = 1.49, p = .138; EN participants showed significant priming for words, t(33) = 3.24, p = .003, and marginal priming for nonwords, t(33) = 1.81, p = .076; and AD patients showed a marginal priming effect for words, t(29) = 1.83, p = .074, and insignificant priming for nonwords, t(29) = 1.33, p = .191. While all of these results are in the predicted direction, and all three groups showed priming for words (if only marginally for AD), only the EN group showed a (marginal) priming effect for nonwords.

Discussion

Summary of Findings

The RT data from this lexical decision, repetition priming experiment showed that the pattern of priming was not modulated by age (YN vs. EN) or by dementia (AD vs. EN), and that there was overall significant positive priming for words as well as overall significant negative priming for nonwords. A caveat is that although each of the three individual groups showed at least marginal priming for words, only the EN group showed marginal priming for nonwords, indicating that there may have been insufficient power relative to within-group variability to test for these priming effects in the individual groups. The accuracy data generally resembled the RT data, with the following caveats: the overall accuracy was high, with rather small differences among conditions; and, unlike the AD and YN groups, the EN group did not show the expected increase in accuracy with repeated words, or the expected decrease in accuracy with repeated nonwords. There did not appear to be any effect of a slight age difference between the AD and EN groups, nor did there appear to be any effect of general slowing, due either to age or to dementia.

The current study differed from a earlier one by Balota and Ferraro (1996) in several ways. Our study task, substring matching, was intended to maximize the role of orthographic, letter-by-letter processing, in contrast to the test task (lexical decision), which involved lexical, word-level processing, whereas their study task, rhyme decision, was at the lexical level. Moreover, the use of low frequency word targets and obsolete English nonword targets resulted in the a priori familiarity of the targets requiring a “yes” decision being much closer to that of targets requiring a “no” decision, than was the case for Balota and Ferraro, who changed one or two letters in real words to produce their nonwords, and used both high and low frequency words. The combination of the orthography-based study task and the lexicality-based test task minimized the role of strategies based on explicit memory for the encoding context, because different levels of processing as well as different responses were required during the study versus test phase. In spite of these changes, the results of our study are quite similar to those of Balota and Ferraro as pertains specifically to the pattern of word and nonword repetition priming in aging and dementia; note however that Balota and Ferraro did not report on priming within individual groups nor provide per-group standard errors. Moreover, it is of interest that the absolute magnitude of priming in our study was greater than in theirs for every group and for both words and nonwords (our EN group being compared to both of theirs, and our AD group to both of theirs). This suggests that the methodological differences in our study—a much shallower study task and much lower frequency words and more word-like nonwords—may have enhanced both positive priming for words and negative priming for nonwords in every group, at least to some degree.

Word Repetition Priming

The invariance of the positive word-repetition priming effect with AD and EN, in spite of the slower overall RTs obtained with these participant groups on the lexical decision task relative to YN, and in spite of the well-documented deficits in explicit memory tests associated with AD and normal aging (Fleischman & Gabrieli, 1999; Old & Naveh-Benjamin, 2008), is another indication of its robustness. The present findings for words are consistent with findings of preserved repetition priming in normal aging and AD across several types verbal tasks, when the tasks are relatively more perceptual than conceptual in nature and when word retrieval is not required (as reviewed by Fleischman, 2007). Nonetheless, the present AD study is one of very few that have focused on repetition priming of lexical decision; the majority of experiments on lexically-based repetition priming in AD have utilized tasks such as threshold word-identification, and word-stem or word-fragment completion (e.g., Arroyo-Anllo, Ingrand, Neau, Aireault, & Gil, 2004; Karlsson, Borjesson, Adolfson, & Nilsson, 2002; La Voie & Faulkner, 2008).

Nonword Repetition Priming

The use of a two-choice lexical decision paradigm to assess repetition priming has the benefit of allowing the researcher to evaluate nonword repetition priming effects. We were particularly interested in the nonword priming effect because it provides a test of the intactness of the memory processes which underlie repetition priming for novel (i.e., never before encountered) stimuli. The design of the present experiment, with different study- versus test-phase tasks, very low frequency words (i.e., very unfamiliar words), and very word-like nonwords (i.e., obsolete English words) was meant to ensure a robust nonword familiarity effect (i.e., negative priming for nonwords), which we could then compare across participant groups. We did, indeed, obtain a negative priming effect for nonwords, which was equivalent between the AD and EN groups, as well as between the EN and YN groups. Negative priming effects for nonwords have been obtained with YNs (Bowers, 1994; McKoon & Ratcliff, 1979; Zeelenberg et al., 2004), and with AD and EN participants (Balota & Ferraro, 1996). We have provided a replication and extension of equal-to-normal, negative, nonword priming for AD patients, utilizing a different type of study task (requiring a focus on sub-lexical orthography) and a more challenging lexical decision task (in terms of the similarity of the nonwords to the words) than employed in previous research. Our findings provide increased support for the ability of mild-to-moderate AD patients (like EN and YNs) to create long-term memory traces for novel, word-like letter strings.

There is evidence for two opposing processes in making a lexical decision to a repeated nonword: a relatively fast, familiarity-based process, which inhibits the correct response to a repeated nonword (i.e., it is biased toward a “word” response) and results in a negative nonword priming effect; and a relatively slow, instance-based processes, which facilitates the correct response to a repeated nonwords (i.e., it is biased toward a “nonword” response) and results in a positive nonword priming effect. This opposing-process evidence was obtained by the manipulation of speed-stress (Wagenmakers, Zeelenberg, et al., 2004) and similarity of the study and test tasks (Zeelenberg et al., 2004), in lexical-decision experiments with YN participants. In their discussions of the nature of nonword memory representations, Wagenmakers, Zeelenberg, et al. (2004), Zeelenberg et al. (2004), and Wagenmakers, Steyvers, et al. (2004) assume that the initial presentation of a nonword does not result in a lexical-semantic memory trace for that nonword. Instead, their assumption is that in addition to an instance (context-based) trace being created, the presentation of a nonword triggers activation of (actual) words in lexical-semantic memory that are orthographically and/or phonologically similar to the presented nonwords. These types of “neighborhood” activation effects have, in fact, been demonstrated for nonwords (Siakaluk, Sears, & Lupker, 2002) as well as for words (Sears, Hino, & Lupker, 1995; Yates, Locker, & Simpson, 2004). Repeated presentation of a nonword results in at least some of these “neighborhood” words being reactivated, thus increasing the feeling of familiarity to the nonword stimulus. In any case, it is assumed by Wagenmakers, Zeelenberg, and colleagues (e.g., Zeelenberg et al., 2004) that the familiarity-based process underlying negative nonword priming does not entail lexical-semantic representations for the nonwords per se. In the next section, we argue that such representations may indeed be involved in the processing of repeated word-like nonwords.

A Theoretical Conjecture: Nonword Repetition Priming and Novel Word Learning

It is not unusual to encounter word-like nonwords when reading. A well-known example is in Chapter 1 of J. K. Rowling's (1997) Harry Potter and the Philosopher's Stone:

“Even Muggles like yourself should be celebrating, this happy, happy day!” And the old man hugged Mr. Dursley around the middle and walked off. Mr. Dursley stood rooted to the spot. He had been hugged by a complete stranger. He also thought he had been called a Muggle, whatever that was. [p. 10]

This “word-like nonword” is used a few more times later in the same chapter, but is not defined explicitly until Chapter 4.

There is considerable similarity between encountering an unknown word while reading a text, and encountering a word-like nonword in an experimental context. In a textual context, it is reasonable to assume that the string is processed as a word that was previously unknown, with fragments of meaning, grammatical function, and pronunciation stored in a rudimentary, provisional lexical entry. When the same string is encountered later, additional meaning is added to the previously created entry from the later context; this is repeated as part of the normal process of word acquisition. Obviously, the textual context provides considerable information that can be stored in the new lexical entry, even if there is not enough to provide a complete understanding of the word. In an experimental context, particularly one such as in the present experiment where the initial encoding of the items was quite shallow, there is much less information available for storage in a provisional lexical entry. However, even in such a case there may be at least some information available, including when and where the item was encountered, what the reader's mental state was at the time, what the response was, and hypotheses as to possible morphological, phonological, and semantic attributes based on the item's orthography. If such provisional lexical entries resulted from single study-phase exposures to word-like nonword letter strings, then during the test phase of a repetition-priming experiment the entry would be activated when the previously viewed letter string was presented again, along with any attributes stored there, including item familiarity. Incidentally, if such provisional lexical entries are created as the result of an exposure as shallow as that involved in our experiment's study task, it is most likely that the process is an automatic one, as opposed to one requiring specific effort or explicit processing of the item.

The concept of a provisional lexical entry is compatible with both the Tenpenny (1995) "weakly episodic" and the Bowers (2000) “weakly abstractionist” view of repetition priming. Both views allow for the contribution of episodic and abstract memory traces to repetition priming, albeit with one type of trace playing a more important role in each view. Moreover, this type of approach is compatible either with a multiple-systems view in which lexical and semantic processing are supported by separate memory systems (Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001), or with a single-system view in which lexical and semantic processing are served by the same memory system (e.g., Dilkina, McClelland, & Plaut, 2010).

Coane et al.’s (2011) results, which differentiated a long-lasting “absolute” familiarity (resulting from experience with the item across the lifespan) from an episode-specific “relative” familiarity (a change in familiarity caused by recent access), provides an interesting nuance to the idea of a provisional lexical entry for nonwords. Their dual-familiarity model suggests that words with high absolute familiarity have little potential to accrue relative familiarity, whereas words with low absolute familiarity have high potential; this duality can explain certain aspects of the word frequency effect. Since a newly created “nonword” lexeme would have near-zero absolute familiarity, then it would have maximum potential to attain high relative familiarity, making such lexemes subjectively more similar to unfamiliar words in terms of overall familiarity, and perhaps accounting for the negative nonword repetition effect such as we report here.

In order to account for positive nonword priming based on a memory of making a previous “no” response in experiments using lexical decision at both study and test, Wagenmakers, Zeelenberg, et al. (2004) posited an episodic, instance-based component in their model of lexical-decision repetition priming, motivated in part by their explicit assumption that nonwords have no lexical or semantic memory entries. It is unclear what the participant's decision to classify a word-like letter string as “not a word” would have on the formation of provisional lexical entries. The existence of such entries would presumably facilitate retrieval of memories of previous responses by acting as a verbal index into episodic memory; alternatively, one could suppose that the previous lexical decision is stored in the provisional lexeme itself, being activated along with its other attributes when the letter string is repeated. Ironically, in this case part of the information stored in the lexical entry would be “this is not a word”. However, this is not unprecedented: consider Lewis Caroll's “slythy toves” and other more or less familiar “words” whose meaning is essentially that they are meaningless, plus some bits of source information. The process whereby an initially sparse provisional lexeme can become increasingly word-like as it accumulates meaning through subsequent experience seems similar to the Wagenmakers, Steyvers, et al. (2004) account of how new information is added to an existing lexical/semantic trace; the addition of the idea of provisional lexemes coming into existence upon encoding of word-like letter strings would provide a theoretical starting point for that process.

Interestingly, there is evidence for implicit learning of novel words by way of reports of successful acquisition of new words by amnesics (e.g., Westmacott & Moscovitch, 2001) and partial acquisition (across incidental-learning exposures) and retention of a novel verb by AD patients (Grossman et al., 2007). Both AD and amnesic patient groups have significant deficits on explicit memory tasks; thus, the novel-word learning must be solely or mainly implicit in nature. In the case of AD, these patients’ preserved novel-word learning is consistent with the evidence for preserved lexical-decision priming, and with a general preservation of implicit learning and memory for lexical stimuli. It should be noted, however, that there was evidence that the AD patients were deficient compared to controls in the acquisition of the semantic attributes of newly-learned words in the Grossman et al. study. An interesting demonstration of novel word learning in YNs is provided by Tamminen and Gaskell (2013) who were able to obtain semantic priming by preceding previously known words by newly learned words (after only a brief period of training); semantic priming is, of course, indicative of the new words having been incorporated into the lexical/semantic network.

In sum, our conjecture is that an initial presentation of a nonword, whether in an experimental task or in a text, may automatically cause a provisional lexical entry to be created, storing mainly orthographic or phonemic features (but relatively few semantic features). Consistent with this proposition are research findings in the domain of novel-word learning that are suggestive of changes in the lexical/semantic processing for novel words occurring after just a single exposure (e.g., Nora et al., 2012; Tamminen & Gaskell, 2013). These and related findings have led to the “complementary systems account” of word learning that distinguishes short- from long-term (after being repeatedly linked to semantic information) mental representations of novel words, with the long-term representations becoming less distinguishable from those of existing words with regard to behavioral and neural response patterns (Lindsay & Gaskell, 2010).

EN Error-Rate Priming

Finally, it was unexpected that the EN group's error data did not display the pattern seen across all groups for RT priming and in the YN and AD group for errors, namely positive priming for words and negative priming for nonwords. Instead, the EN error rate showed near-zero priming for words and positive priming for nonwords. While this could be simply a fluke, there is another possibility: the well-known cautiousness of elderly participants (e.g., Starns & Ratcliff, 2010). EN participants may have noticed during the study phase that the items seemed to be very infrequent words, and so implicitly tried to find a meaning for each one (e.g., in case they were asked about them later) while they performed the substring matching task, with failure being more salient than success. In this case, an “I found no meaning for X” memory trace could have facilitated their error performance for studied nonwords while having little effect on words. Note that although in our experiment this contrary effect was visible only in EN error performance, the explanation is somewhat compatible with the finding of Zeelenberg et al. (2004) of positive RT priming for nonwords when lexical decisions are made at study. Perhaps if processing time were constrained during the study task, gratuitous lexical processing could be reduced, parallel to what was also seen by Zeelenberg et al. for explicit lexical decisions with constrained study time.

Conclusions and Future Directions

The results of this study extend the stimulus and task parameters under which repetition priming of lexical decision, based mainly (if not exclusively) on implicit memory processes, is preserved in AD as well as normal aging. Specifically, the repetition priming effect and the degree to which the repetition priming effect was reversed for nonwords compared to words was unaffected by AD or normal aging. Thus, AD and EN individuals, like YN individuals, are able to form new long term memory traces based on purely orthographic encoding of novel word-like letter strings.

One obvious question for future research is whether significant differences in priming effects would be obtained for AD compared to EN groups, or EN compared to YN groups with a lexical decision test task, if the time between the study and test phases had been many hours (or several days), as opposed to the approximately 10-minutes of the current study. A related question for future research is whether there would be differences in the longevity of the lexical-decision priming effects based on the nature of the study task (e.g., orthographic, phonemic, or semantic) and/or the type of word versus nonword stimuli (including the similarity between the two types of stimuli), in addition to differences due to age or disease.

Another direction for future research is toward clarifying the link between nonword repetition priming and the acquisition of new vocabulary. To what extent do the abstract structures posited by Tenpenny (1995) and the Bowers (2000) act like lexical entries? Complicating this task is the lack of an accepted way to determine objectively when a newly encountered nonword such as “Muggle” enters the lexicon and what information is stored with it during the ongoing process of lexicalization. Given the Grossman et al. (2007) findings regarding novel-word learning abilities in AD (which are consistent with preserved nonword repetition priming in AD), another interesting research avenue would be to extend that work via paradigms that make greater use of implicit memory tasks, on which AD patients perform relatively well (e.g., tasks that do not require word retrieval). Such implicit tasks could involve pairings of novel words with pictures, sounds, tactile stimuli, etc., in addition to verbal material such as sentence frames.