Priming Sentence Production in Older Adults: Evidence for Preserved Implicit Learning

Abstract

Purpose:

Structural priming- speakers’ unconscious tendency to echo previously encountered message-structure mappings - is thought to reflect the processes of implicit language learning that occur throughout the lifespan. Recently, structural priming has also been used as a means to facilitate language re-learning in age-related language disorders such as aphasia. However, little evidence is available on whether structural priming remains effective in healthy aging, limiting clinical translation of the structural priming paradigm. This study examined the impact of aging on the strength and longevity of abstract structural priming and lexical boost effects.

Method:

Twenty-four young and 24 older adults participated in a collaborative picture-matching task where the participant and experimenter took turns describing picture cards using transitive and dative sentences. In Experiment 1, a target was elicited immediately following a prime (0-lag), whereas two filler items were embedded between a prime and a target sentence in Experiment 2 (2-lag) to examine longer-term priming effects. In both experiments, the verb was repeated for half of the prime-target pairs to examine lexical boost on structural priming.

Results:

At immediate priming, older adults showed both abstract structural priming and lexical boost in transitives and datives, similar to young adults. At longer-term priming, only abstract priming effects were significant in both groups of participants, with no evidence of age-related reduction in priming.

Conclusions:

Both lexically-specific and independent mechanisms of structural priming remain resilient in older adults, supporting the view that structural priming reflects life-long language learning. Further, the findings provide empirical basis for applying structural priming to elderly clinical populations.

Introduction

As adults age, normal changes occur in their cognitive-linguistic skills. In addition, their chances of age-related communication disorders, such as post-stroke aphasia, increase. Therefore, when speech-language pathologists work with the aging population, it is crucial to understand what aspects of cognitive-linguistic processes are preserved or decline as part of normal aging. The present study aims to inform mechanisms of syntactic learning in healthy aging by investigating the extent to which mechanisms of structural priming remain preserved in older adults.

Structural priming refers to a language user’s unconscious tendency to produce or comprehend sentences more easily because they have previously encountered the same sentence structure (Bock, 1986; Pickering & Ferreira, 2008). For example, when an individual hears their conversational partner produce a passive sentence to describe an event, the individual is likely to produce a passive sentence to describe a new event rather than the more frequent, active sentence. Since the seminal study by Bock (1986), structural priming has been observed ubiquitously in healthy speakers in both language production and comprehension tasks, using a variety of syntactic structures as well as across numerous languages (see Mahowald, James, Futrell, & Gibson, 2016; Pickering & Ferreira, 2018 for review). In addition, structural priming has been found in young children 3–4 years of age to adulthood, leading to a proposal that it reflects life-long language learning (Bock & Griffin, 2000; Chang, Dell, & Bock, 2006).

More recently, structural priming has been used with clinical populations such as persons with post-stroke aphasia (PWA) as a means to facilitate their immediate and lasting language production (Cho-Reyes, Mack, & Thompson, 2016; Hartsuiker & Kolk, 1998; Lee & Man, 2017; Lee, Man, Ferreira & Gruberg, 2019; Man, Meehan, Martin, Branigan, & Lee, 2019; Saffran & Martin, 1997; Yan, Martin, & Slevc, 2018; see also Lee, Hosokawa, Meehan, Martin, & Branigan, 2019 for comprehension priming). For example, after hearing or repeating a prime sentence, PWA show more frequent production of complex sentences such as passive or double-object dative sentences (Hartsuiker & Kolk, 1998; Yan et al., 2018). PWA show persistent priming effects over a range of two to 10 intervening fillers, suggesting that priming-induced learning may be preserved in PWA (Cho-Reyes et al., 2016; Lee, Man et al., 2019; Man et al., 2019). In Lee and Man (2017), a PWA, following 12 sessions of structural priming treatment, showed improved production of prepositional dative sentences at 4-week maintenance testing and generalized improvements in connected speech, as demonstrated by increased production of sentence-level utterances. Although preliminary, these findings indicate that structural priming holds clinical potential to create lasting improvements in the syntactic processing of PWA. However, most available structural priming studies with healthy individuals are from children and young adults, with limited research on if and how normal aging affects structural priming. Several cognitive-linguistic processes are associated with the strength and longevity of structural priming effects (see below for details) and certain cognitive-linguistic processes decline with aging. Thus, a better understanding of what aspects of structural priming remain preserved or impaired in older adults will provide a stronger empirical basis for clinical application of structural priming.

Structural priming involves multiple linguistic representations. Abstract structural priming refers to the priming effect that happens at the level of syntactic constituent orders, independent of lexical-semantic content. For example, speakers are more likely to produce a passive structure to describe a new event (e.g., the mailman was chased by the dog) after hearing a passive structure (e.g., the church was struck by lightning) even though no lexical items are shared between the prime and target (Bock, 1986; Bock & Griffin, 2000; Bock & Loebell, 1990). However, the magnitude of a structural priming effect becomes enlarged when lexical items (e.g., verb) are repeated between prime (e.g., the robber was chased by the cop) and target (e.g., the mailman is chased by the dog), i.e., lexical boost effect (Branigan & McLean, 2016; Branigan & Pickering, 1998; Hartsuiker, Bernolet, Schoonbaert, Speybroeck, & Vanderelst, 2008). Most production studies have reported that the lexical boost effect tends to be short-term, dissipating over intervening fillers between a prime and target item (Branigan & McLean, 2016; Hartsuiker et al., 2008). Abstract structural priming, on the other hand, is shown to be longer-term, persisting over intervening fillers between prime and target and across multiple sessions, although the longer-term effects may not be as large as those seen at immediate priming (Bock, Dell, Chang, & Onishi, 2007; Bock & Griffin, 2000; Branigan & McLean, 2016; Hartsuiker et al., 2008; Kaschak & Borreggine, 2008; Kutta & Kaschak, 2012). In Bock et al. (2000; 2007), for example, the young adults showed comparable priming effects at 2-, 4-, and 10 lags of intervening fillers between a prime and a target sentence. These lasting priming effects have also been found in populations whose explicit learning skills are not yet developed or impaired, including young children who were 3–4 years old (Branigan & McLean, 2016; Rowland, Chang, Ambridge, Pine, & Lieven, 2012; Savage, Lieven, Theakston, & Tomasello, 2006) and patients with amnesia (Ferreira, Bock, Wilson, & Cohen, 2008; Heyselaar, Segaert, Walvoort, Kessels, & Hagoort, 2017) and aphasia (Cho-Reyes et al., 2016; Lee & Man, 2017; Lee et al., 2019; Man et al., 2019).

As such, recent theories of structural priming propose that structural priming reflects processes of language learning and have proposed different cognitive bases to account for the different time courses of lexical boost and abstract structural priming effects. The dual-path model proposed by Chang and colleagues (2006; 2012) suggests that abstract structural priming is driven by experience-based implicit learning of syntactic rules that happens throughout the lifetime (see also Jaeger & Snider, 2013). As a language user encounters prime sentences, they make predictions on word order. When their predictions do not match with actual experiences, their syntactic sequencing system learns to adjust connection weights, increasing the probability of generating the primed word order in the future. However, lexical boost is supported by item-specific retrieval in explicit short-term memory, yielding only temporary increase in priming. Reitter, Keller and Moore (2011), on the other hand, explain both short-term and long-term priming within declarative memory-based learning. According to their model, short-lived lexical boost effects are a result of spreading activation. When a linguistic representation is retrieved from declarative memory, there is spreading activation to related representations, resulting in short-term priming effects. However, although the activation decays as a power-law, retrieving a structure from memory repeatedly increases its base-level activation, resulting in longer-lasting priming (learning) effects. Lastly, different from these two models, the most recent proposal by Heyselaar, Wheeldon, and Segaert (2021) suggests that two different types of non-declarative (implicit) memory underlie both short-term and longer-term priming. Specifically, perceptual memory underlies short-term, repetition-based priming. However, conceptual memory supports implicit learning of statistical regularities associated with long-term abstract structural priming effects or cumulative learning of syntax.

Certain linguistic skills are more susceptible to age-related decline. It is well-established that lexical retrieval processes become less efficient in older adults, while their receptive vocabulary remains stable or increases in older adults (Burke & Shafto, 2004; Nicholas, Obler, Alert & Goodglass, 1985; Ramsay, Nicholas, Au, Obler, & Albert, 1999; Zelinski & Burnight, 1997). Mixed findings exist with regards to their syntactic processing. Studies show the ability to process complex syntax diminishes with aging, as indicated by, for example, reduced grammatical complexity in production and longer response times on comprehension tasks (Kemper, Greiner, Marquis, Prenovost, & Mitzner, 2001; Kemper & Sumner, 2001; Kemper, Thompson, & Marquis, 2001; Waters & Caplan, 2001; Wingfield, Peelle, & Grossman, 2003). Age-related differences in syntactic processing become more pronounced when the stimuli or tasks involve increased explicit working memory load, such as encoding of increased numbers of lexical items and embedded clauses (Kemper et al., 2001a, b; Kemper & Summer, 2001; Waters & Caplan, 2001). However, age differences disappear when demands for lexical-semantic processing are controlled for or when real-time measures of sentence processing are examined, indicating that syntactic representations as such may remain preserved in older adults (Caplan & Waters, 2001; Davidson, Zacks, & Ferreira, 2003).

Age-related differences in various domains of cognition may underlie changes in linguistic behaviors seen in older adults. A general decrease in the speed of information processing may result in slower activation of linguistic representations (Salthouse, 1996). The Transmission Deficit Hypothesis holds that aging weakens connections among linguistic representations, resulting in too little excitation to allow a linguistic representation to reach a threshold necessary for activation (Burke & Shafto, 2004). Therefore, within these theories, it can be predicted that syntactic representations might become accessible too slowly or do not reach their full activation in older adults during a structural priming task. Accordingly, this will result in reduced or non-significant structural priming effects in older adults.

More specific to different memory processes, an extensive line of research has reported that explicit, declarative memory abilities requiring conscious recollection are impaired in aging, while there is no or little aging effect on implicit, non-declarative memory (Bennet, Howard, & Howard, 2007; Curran, 1997; Daselaar, Rombouts, Veltman, Raaijmakers, & Jonker, 2003; Fleischman, Wilson, Gabrieli, Bienias, & Bennett, 2004; Howard & Howard, 2001; King, Fogel, Albouy, & Doyon, 2013; Light & Snigh, 1987; Schacter, Cooper, & Valdiserri, 1992; Spaan & Raaijmakers, 2010). For example, Fleischman et al. (2004), in their longitudinal study, found that the ability to explicitly recall visual or verbal materials declined steadily with increasing age at both immediate and delayed time points. However, no consistent age-differences were seen on implicit priming or skill learning tasks. Similarly, Howard and Howard (2001) showed that instructing older adults to find patterns during an alternating serial response time task decreased their accuracy in identifying patterns and slowed their response time rate, compared to when the pattern learning was performed in the context of an implicit task, without involving explicit instructions. However, it should also be noted that there is some evidence suggesting that non-declarative memory associated with skill learning (conceptual memory) can decline with aging, especially in individuals who are in their 70s and 80s (Heyselaar et al., 2017; Maki, Zonderman, & Weingartner, 1999; Schugens, Daum, Spindler, & Birbaumer., 2007). While existing data are not entirely clear cut, given the age-related dissociation between explicit and implicit memory, it can be hypothesized that if structural priming reflects implicit language learning, older adults would show no or minimal reduction in their priming effects.

Only a few studies so far examined syntactic priming in older adults, yielding mixed findings. Some studies have reported intact structural priming in older adults (Hardy, Messenger, & Maylor, 2017; Hardy, Wheeldon, & Segaert, 2020; van Boxtel & Lawyer, 2022). Hardy et al. (2017) found that older adults showed comparable lexical boost and abstract structural priming as young adults in a scripted dialogue-like priming task, wherein participants heard a syntactic (either active or passive) structure from a confederate partner and then immediately (0-lag) described their own transitive action card. Hence, the authors argued that there are no age-related changes in the presence and magnitude of both explicit and implicit memory-based mechanisms of priming. Hardy et al. (2017), however, examined only transitive sentences under the immediate priming condition. Thus, their findings need to be replicated in different structures and at a longer-term priming context. In a recent study, van Boxtel and Lawyer (2022) found that older adults show significant abstract priming and lexical boost effects over 2 intervening fillers during comprehension of reduced relative clauses, using a self-paced reading task. The authors, therefore suggested that both abstract and lexically-mediated syntactic priming may involve implicit memory processes.

Others report non-significant or reduced structural priming effects with aging. In Heyselaar et al (2017), older controls failed to show significant priming effects while patients with amnesia showed significant priming. The authors attributed this lack of priming in older adults to cognitive interference between declarative and nondeclarative memory, unlike in amnesic patients whose explicit memory was impaired (see also Hartsuiker & Kolk, 1998). Recently, Heyselaar, Wheeldon and Segaert (2021) examined how structural priming changes over the lifespan, by recruiting groups of participants from 20 through 85 years of age. The authors found that only those who were in their 70s and 80s showed a decline in long-term, but not short-term priming effects. In addition, they found that while non-declarative perceptual memory (repetition priming) scores were intact, non-declarative conceptual memory (serial reaction time task) scores were reduced in the older adults of 70+ years of age. Thus, these authors concluded that while both short-term and long-term structural priming are supported by non-declarative memory components, long-term structural priming is reduced with aging due to their decreased implicit rule learning processes.

In sum, the existing literature presents with little data regarding if and how aging affects the lexical boost and abstract structural priming effects. Systematic investigation of structural priming in older adults may help to identify the conditions in which syntactic repetition is most facilitative for older adults and have both theoretical and clinical implications. The current study investigated structural priming effects across different structures (transitives and datives) and over different time lags (0- vs. 2-lag) in young and older adults, using a dialogue-like comprehension-to-production syntactic priming task, similar to Hardy et al. (2017). At 0-lag, there were no intervening fillers between prime and target, whereas at 2-lag, two intervening fillers were embedded between prime and target. At both lags, half of the trials were transitive targets and the other half were dative targets (48/structure). Within the target structure, the verb was repeated between prime and target for half of the trials to examine lexical boost effects. The same participants completed both 0- and 2-lag priming tasks, with a minimum of 2 weeks between the experiments and the order of the lags was counterbalanced across participants.

We asked three research questions. First, it was examined whether abstract structural priming remains intact in older adults across different target structures (transitives and datives) and over different temporal lags (zero vs. two intervening fillers). It was predicted that if structural priming truly reflects mechanisms of language learning supported by implicit (or non-declarative) memory, older adults would show significant priming effects at both lags and across the structures (Chang et al., 2006; 2019; Heyslaar et al., 2021). However, if declarative memory-based learning underlies structural priming (Reitter et al., 2011) or if older adults experience slowed or weak spreading activation of linguistic representations (Burke & Shafto, 2004; Salthouse, 2004), they may show reduced or absent priming effects, compared to young adults. Secondly, we asked if lexical boost effects remain preserved in aging. If lexically boosted priming relies on explicit memory, as proposed by the dual-path model of language learning (Chang et al., 2006; 2012) and the declarative memory-based model of structural priming (Reitter et al., 2011), older adults are predicted to show reduced lexical boost effects compared to young adults. However, if non-declarative memory processes support priming effects in general (Heyselaar et al., 2021; see also van Boxtel & Lawyer, 2022), older adults may show comparable lexical boost effects as young adults. Lastly, we conducted an exploratory analysis to compare the performance between the older participants who are in their 60s versus older than 70. This post-hoc analysis was done to ensure that the results of the older adult group were not driven by those in their 60s, given the recent findings that adults who are older than 70 showed reduced structural priming compared to younger older adults (Heyselaar et al., 2021).

Method

Participants.

We tested 24 young adults (YA, 13 females, 11 males; age M (SD) = 21.3 (2.3), range 18–29 years old; education M (SD) = 15.1 (2.1), range 12–22 years) and 24 older adults (OA, 16 females, 8 males; age M (SD) = 70.6 (6.6), range 60–82 years old; education M (SD) = 16.1 (3.0), range 12–23 years). Data from twenty of the older adults were reported in our previous study (Man et al., 2019) as a control group for individuals with aphasia. All participants were monolingual native speakers of English, with no history of neurological or psychiatric disorders that could affect communication. They reported normal or corrected-to-normal vision, and passed a hearing screening at 500, 1000, and 2000 Hz at 40 dB in at least one ear. The participants’ cognitive-linguistic abilities were screened using the Cognitive Linguistic Quick Test (CLQT), as in Table 1 (Helm-Estabrooks, 2001). All participants scored within normal limits on the Severity Rating for their age group (OA: Mean (SD) = 3.98 (.07); YA: Mean (SD) = 3.98 (.06)). On the subdomain scores, YA performed better than OA in the domains of Attention, t (46) = 3.06, p = .004, Executive Function, t (46) = 5.26 p < .001, and Visuospatial Skills, t (46) = 4.12, p < .001. The two groups did not differ statistically on the domains of Memory and Language, t’s < 0.435, p’s > .665. All participants provided informed consent prior to study participation and received monetary compensation for their time.

Table 1.

Participants’ scores on each subdomain (max score) of the Cognitive-Linguistic Quick Test

Group		Attention (215)	Memory (185)	Executive Functions (40)	Language (37)	Visuospatial skills (105)	Composite Severity Rating (4.0)
Young Adults	Mean	204.5	168.9	35.6	33.8	99.4	3.98
	SD	5.3	7.9	2.7	1.6	4.3	0.06
Older Adults	Mean	185.7	168.0	31.3	33.5	91.1	3.98
	SD	29.7	14.8	3.0	1.7	8.9	0.07

Stimuli.

For transitive stimuli, we prepared 48 target and 48 prime sentences and their corresponding black-and-white line drawings. The pictures depicted an animal agent and human patient in one of six actions (bite, chase, kiss, lift, pull, push). For dative stimuli, another set of 96 (48 targets and 48 primes) sentences and associated pictures were prepared. The dative pictures depicted a human agent and goal and an inanimate theme in one of six actions (give, hand, offer, sell, show, throw). The stimuli were taken from previous structural priming studies (Branigan & McLean; 2016; Branigan, Pickering & Cleland, 2000). For both transitive and dative stimuli, each verb was repeated 8 times in the prime pictures, and 8 times in the target pictures with different nouns. The target verb and nouns were written on the picture card. The picture stimuli were printed as 4 1/2 × 3 2/3 inch cards on card stock paper.

As exemplified in Table 2, each target picture was paired once with a ‘preferred’ prime (passive prime for transitives; DO prime for datives) and once with a ‘non-preferred’ prime (active prime for transitives; PO prime for datives). Half of the prime-target pairs had the same verb and the other half had different verbs. Thus, each picture was elicited under the four different priming conditions (e.g., same verb passive, same verb active, different verb passive, and different verb active primes for a transitive target picture). When prime and target stimuli were paired, it was ensured that there was no phonological and semantic overlap between the nouns used in the sentences to minimize additional phonological or semantic influence on priming effects. Additionally, a total of 192 intransitive sentences and corresponding pictures were prepared for fillers. Of these 192 fillers, 31 were used as the ‘Bingo’ items. The bingo items were identical between experimenter and participant, and were included to ensure that the participants actively attended to the task.

Table 2.

A set of example experimental stimuli.

	Example stimuli	Conditions
Transitive
	A frog kissing a queen	Target transitive action
	The sheep is kissing the boy	Same Verb, Active Prime
	The boy is being kissed by the sheep	Same Verb, Passive Prime
	The bear is pulling the witch	Different Verb, Active Prime
	The witch is being pulled by the bear	Different Verb, Passive Prime
Dative
	A ballerina showing a pitcher to a doctor	Target dative action
	The artist is showing the ball to the cowboy	Same Verb, PO Prime
	The artist is showing the cowboy the ball	Same Verb, DO Prime
	The chef is handing the apple to the monk	Different Verb, PO Prime
	The chef is handing the monk the apple	Different Verb, DO Prime

Four lists were created for the experiment for each lag condition. Each list consisted of a total of 96 experimental trials, including 48 transitive prime-target pairs and 48 dative prime-target pairs, and 192 filler trials. Each target picture was presented only once within the list, paired with one of the four different primes across the four lists. The order in which experimental trials were presented within each list was pseudo-randomized such that no same target structure was elicited in two consecutive trials. At 0-lag condition, two filler stimuli followed each prime-target pair (prime-target-filler-filler). At 2-lag condition, two filler stimuli intervened between prime and target items (prime-filler-filler-target), creating a time lag between the prime and target sentence. Each participant received only one list. The order of the list was counterbalanced across the participants.

Task & Procedure.

The experimenter and participant played a collaborative picture description task, where they take turns describing pictures (see Figure 1). The experimenter and participant each had a stack of cards in front of them faced down on the table. The experimenter’s stack contained the prime and filler cards, and the participant’s stack contained the target and filler cards. The participants were told that the goal of the task was to find identical pictures and when they have the same picture as the experimenter, they should say “Bingo”. The experimenter always described their pictures first, using the prime structure indicated by a color cue marked on the bottom right corner of the prime picture card (e.g., green = active, orange = passive). There were no color cues on the participants’ picture cards such that the participant was unaware of the prime manipulation. A set of four practice trials preceded the experimental stimuli.

Figure 1.

An illustration of the priming task for dative targets.

Data coding.

Participants’ responses were transcribed and coded as ‘correct’ if it was produced in an active or passive sentence for transitive targets and a PO or DO sentence for dative targets. Production of synonyms (e.g., girl for woman), variances in verb tense forms (sold, is selling), and disfluencies (e.g., fillers, self-corrected responses) were accepted. For passives responses, a correct response had to contain the correct passive verbal morphology (auxiliary and past participle) and a by-phrase. Only the ‘correct’ responses were then included for the analysis of priming effects. Out of the correct responses, the passive (vs. active) and DO (vs. PO) responses were coded as the ‘preferred’ target structures for the purpose of statistical analyses and were used as the dependent variables.

Statistical analysis.

Data were analyzed using logistic mixed effect regression models (lme 4 package; Bates, Maechler, Bolker, & Walker, 2014). Data were first modeled to include both groups in order to compare priming effects between the groups. These models included prime, verb, group, and their 2- and 3-way interactions as fixed effects. Secondly, a separate model was used for each participant group in order to test for priming and lexical boost effects within the group, by entering the prime type, verb type, and their interaction as fixed effects. For the random effects structure, we included by-participant and by-item intercepts and slopes for all main effects. If the model did not converge, we removed by-item random effects. All models included the full random effect structure, except for the model comparing priming effects between young and older adults for transitive targets at 2-lag. Model comparisons were then performed using the ANOVA test in R, with a threshold of p < .05, to determine whether the interaction or each main effect significantly contributed to model fit. Additionally, effect sizes (Cohen’s d) were computed for each priming condition within participant group to directly compare magnitudes of priming between the groups while factoring out differences in variability (Cohen, 1992). The effect size was determined by calculating the mean difference in the proportion of preferred (e.g. passive) structures between the two (e.g., active and passive) prime conditions and then dividing the result by the pooled standard deviation.

Exploratory analysis.

Heyselaar et al. (2021) recently found that age-related reduction in priming became significant in their participants who were older than 70 years, but not the older adults who were 60–69 years of age. A post-hoc additional analysis was conducted to test if our participants who are older than 70 (OA70+; age range 70–82 years old, n = 13) show reduced priming compared to younger older participants (OA 60–69, n=11) to rule out the possibility that the absence of age-related differences in our results were mainly driven by the younger older participants. To compare their priming effects at 0-lag, the prime, verb, and age and their interactions were entered as predictors. To compare their priming effects at 2-lag, only prime and age group and the interaction were entered as predictors. The verb, and the related 2- and 3-way interactions were not included in this model because priming effects at 2-lag did not vary as an effect of verb type in either YA and OA groups in the main analyses (see Table 5 below). In all models, by-participant and by-item random intercepts and slopes were included for the main effects.

Table 5.

Summary of logistic mixed effects models, Experiment 2 (2-lag).

Predictors	Log-odds Estimate	Std. error	χ2	p-Value
Transitive Targets: YA vs. OA
(Intercept)	−1.786	0.331
Prime	−0.007	0.232	6.836	<.01
Verb	−0.123	0.243	0.104	.746
Age	−0.904	0.469	3.302	.069
Prime x Verb	0.449	0.327	1.135	.286
Prime x Age	0.509	0.367	1.189	.275
Verb x Age	0.070	0.395	0.414	.519
Prime x Verb x Age	−0.448	0.517	0.751	.386
Transitive: YA
(Intercept)	1.938	0.367
Prime	0.051	0.253	2.441	.118
Verb	−0.419	0.254	1.053	.304
Prime x Verb	0.451	0.336	1.802	.179
Transitive: OA
(intercept)	−3.021	0.416
Prime	0.732	0.343	7.692	<.01
Verb	−0.102	0.340	0.080	.778
Prime x Verb	0.061	0.419	0.021	.885
Dative Targets: YA vs. OA
(Intercept)	−2.703	0.569
Prime	1.010	0.352	9.608	<.01
Verb	−0.032	0.312	1.142	.285
Age	−0.442	0.715	3.340	.067
Prime x Verb	−0.089	0.390	0.065	.797
Prime x Age	−0.659	0.526	2.192	.138
Verb x Age	−0.253	0.471	0.573	.449
Prime x Verb x Age	0.029	0.616	0.002	.962
Dative: YA
(Intercept)	−2.987	0.694
Prime	1.317	0.445	10.183	<.01
Verb	0.047	0.312	0.007	.934
Prime x Verb	−0.102	0.394	0.0676	0.795
Dative: OA
(intercept)	−3.159	0.426
Prime	0.481	0.335	3.327	0.068
Verb	−0.308	0.373	1.707	.191
Prime x Verb	−0.049	0.476	0.011	.918

Results

0-lag, Immediate Priming.

For 0-lag priming, 23 and 10 responses were excluded due to experimental errors for YA and OA group, resulting in a total of 2,281/2,304 and 2,295/2,304 scorable responses, respectively. Both groups produced high proportions of correct responses for both transitive (YA: 97% vs. OA: 99%) and dative targets (YA: 97% vs. OA: 99%). The ‘incorrect’ responses mostly consisted of production of non-target structures (e.g., the waitress is trying to show the ballerina how to hold the pitcher for the target the waitress is showing the ballerina a pitcher). The priming effects and the results of the mixed-effects models are summarized in Tables 3 and 4.

Table 3.

Priming results in young adults (YA) and older adults (OA), indicated by proportions of passive and DO responses under different priming conditions. Numbers in parentheses indicate standard errors.

Group	Verb type	Passive prime	Active prime	DO prime	PO prime
Experiment 1 (0-lag)
YA	Same verb	34 (5)	13 (3)	26 (7)	10 (3)
	Different verb	24 (5)	15 (4)	21 (6)	14 (5)
OA	Same verb	50 (6)	3 (9)	23 (5)	5 (2)
	Different verb	28 (4)	7 (2)	18 (4)	7 (2)
Experiment 2 (2-lag)
YA	Same verb	22 (4)	16 (3)	21 (5)	17 (5)
	Different verb	20 (4)	20 (4)	24 (5)	17 (6)
OA	Same verb	14 (4)	10 (3)	11 (4)	8 (3)
	Different verb	14 (3)	9 (2)	12 (3)	8 (3)

Table 4.

Summary of logistic mixed effects models, Experiment 1 (0-lag).

Predictors	Log-odds Estimate	Std. error	χ2	p-Value
Transitive Targets: YA vs. OA
(Intercept)	−2.340	0.418`
Prime	0.807	0.378	51.84	<.0001
Verb	−0.018	0.283	16.23	<.0001
Age	−0592	0.538	0.002	0.965
Prime x Verb	0.688	0.347	17.65	<.001
Prime x Age	0.942	0.551	9.904	0.001
Verb x Age	−1.031	0.530	0.133	0.714
Prime x Verb x Age	1.516	0.601	6.348	0.011
Transitive: YA
(Intercept)	−2.319	0.416
Prime	0.815	0.375	13.056	<.001
Verb	−0.053	0.284	4.095	<.05
Prime Type x Verb	0.697	0.347	4.028	<.05
Transitive: OA
(intercept)	−2.961	0.374
Prime	1.775	0.401	48.252	<.001
Verb	−0.957	0.450	14.187	<.001
Prime Type x Verb	2.115	0.490	18.645	<.001
Dative Targets: YA vs. OA
(Intercept)	−2.749	0.496
Prime	0.367	0.357	43.12	<.0001
Verb	−0.491	0.328	0.352	0.552
Age	−0.820	0.706	0.057	0.810
Prime x Verb	1.043	0.427	10.07	<.01
Prime x Age	1.357	0.519	9.17	<.01
Verb x Age	−0.101	0.509	0.299	0.583
Prime x Verb x Age	−0.102	0.627	0.026	0.870
Dative: YA
(Intercept)	−2.679	0.493
Prime	0.304	0.407	5.288	<.05
Verb	−0.594	0.339	0.003	0.957
Prime x Verb	1.095	0.429	6.511	<.05
Dative: OA
(intercept)	−3.531	0.515
Prime	1.700	0.388	40.831	<.001
Verb	−0.517	0.404	0.438	0.508
Prime x Verb	0.887	0.464	3.652	0.056

For the transitive targets, the model comparing the groups revealed a significant effect of prime type. Both YA and OA produced significantly more passive responses following passive versus active primes, as confirmed by within-group models. A significant effect of verb type indicated more passive responses in the same- versus the different-verb prime condition. However, no age effect was significant. Among 2-way interactions, the prime by verb interaction was significant, indicating increased priming for the same- vs. different-verb primes. In addition, as indicated by the interaction between prime and group, the priming effect was greater for OA than YA. Importantly, the 3-way (prime x verb x group) interaction was significant: the magnitude of lexical boost was greater for OA compared to YA. As seen in Table 2, YA produced a mean of 21% more passives, d =1.0, following passive vs. active primes when there was verb overlap, while the priming effect was reduced to 9%, d = .42, when there was no verb overlap. For OA, the average priming effect was 47% for the same verb, d = 2.06, and 21%, d = 1.21, for the different verb condition.

For the dative targets, there was a significant effect of prime type, indicating both YA and OA produced significantly greater DO responses following DO vs. PO primes. Interestingly, there was a significant prime by group interaction, indicating that OA showed greater priming effects compared to YA. In addition, a significant prime by verb type interaction indicated a lexical boost effect. Within group models indicated that the priming effects were larger for the same verb condition in both YA and OA, although the interaction term did not reach significance for OA. YA showed a mean 16% priming effect, d =.61, in the same verb condition, whereas their mean priming effect was 7% in the different verb condition, d = .29. OA showed a mean of 18% priming effect in the same verb condition, d = .98, and 11% priming effect in the different verb condition, d = .64. No other main effects or 2- or 3-way interactions were significantly contributed to the model fit in the between-group comparisons.

2-lag, lasting priming.

For 2-lag priming, in YA group, we excluded 23 responses from a total of 2,304 responses due to experimental errors, resulting in a total of 2,281 scorable responses. For OA, 6 responses were excluded from a total of 2,304 due to experimental errors, resulting in a total of 2,998 scorable responses. Both groups produced high proportions of correct responses for both transitive (YA: 98% vs. OA: 99%) and dative targets (YA: 97% vs. OA: 99%). The results of the mixed-effects logistic models for the priming effects over the intervening 2 fillers are presented in Table 5.

For transitive targets, the model comparing groups revealed that only the effect of prime type was significant. No other main effects or 2- and 3-way interactions were significant, indicating that the priming effect did not vary as an effect of participant group or verb type. The models conducted within group revealed that the priming effect for YA did not reach significance. This is likely due to the lack of priming effect in the different verb condition (0% difference), although they showed a reliable priming effect in the same verb condition (mean 6% difference), d = .33. OA showed a significant priming effect, demonstrating comparable priming between the same verb (mean 4% difference), d = .27, and different verb (mean 5% difference), d = .32, conditions.

For dative targets, only the effect of prime type was significant, similar to the results in the transitive targets. No other main effect or interaction was significant: the two groups did not differ from each other in the magnitudes of priming or there was no lexical boost effect. Within-group models confirmed that YA showed comparable priming in the same-verb (4% difference), d = .17, and different-verb conditions (7% difference), d = .24. Likewise, OA showed small effect sizes that were comparable between the same-verb (3% difference), d = .20 and different-verb prime conditions (4% difference), d = .30, although the main effect of prime type in the mixed effect model was not significant.

Summary.

The analysis of priming effects revealed that older adults showed intact abstract priming effects at both 0-lag and 2-lag in transitive as well as dative target sentences, with their effects being comparable to or larger than young adults. In addition, older adults showed a larger lexical boost effect than young adults in transitive sentences and comparable lexical boost effect to young adults in dative sentences at 0-lag. Both groups did not show a lexical boost effect at 2-lag, but only significant abstract priming effects.

Exploratory analyses results

The priming effects for these subgroups of OA are presented in Table 6 and the results of statistical analyses are reported in Tables 7 and 8.

Table 6.

Priming results in older adults who were 60–69 years old (OA 60) and older adults who were 70–82 years old (OA70+), indicated by proportions of passive and DO responses under different priming conditions. Numbers in parentheses indicate standard errors.

		Transitive		Dative
Group	Verb type	Passive prime	Active prime	DO prime	PO prime
Experiment 1 (0-lag)
OA60	Same verb	42 (9)	5 (2)	23 (7)	7 (4)
	Different verb	29 (2)	5 (7)	19 (5)	5 (3)
OA70+	Same verb	57 (9)	13 (1)	22 (6)	3 (2)
	Different verb	27 (5)	9 (5)	18 (6)	9 (4)
Experiment 2 (2-lag)
OA60	Same verb	20 (7)	12 (4)	14 (7)	11 (5)
	Different verb	29 (9)	16 (6)	14 (6)	9 (5)
OA70+	Same verb	8 (1)	7 (3)	8 (2)	4 (2)
	Different verb	17 (5)	12 (5)	9 (3)	6 (3)

Table 7.

Summary of logistic mixed effects models comparing OA60 vs. OA70+ groups, Experiment 1 (0-lag).

Predictors	Log-odds Estimate	Std. error	χ2	p-Value
Transitive Targets: OA60 vs. OA70+
(Intercept)	−3.866	0.944
Prime	1.959	0.636	63.048	<.001
Verb	−1.738	0.840	16.025	<.001
Age	0.861	1.049	11.398	<.001
Prime x Verb	2.482	0.900	16.930	<.001
Prime x Age	0.468	0.835	0.137	.711
Verb x Age	1.297	1.002	3.840	.050
Prime x Verb x Age	−0.608	1.088	0.312	.576
Dative Targets: OA60 vs. OA70+
(Intercept)	−3.903	1.035
Prime	1.749	0.630	39.791	<.001
Verb	−0.468	0.626	0.948	.330
Age	0.196	1.217	0.488	.485
Prime x Verb	0.891	0.744	3.188	.074
Prime x Age	0.182	0.836	0.044	.834
Verb x Age	0.097	0.816	0.011	.915
Prime x Verb x Age	−0.071	0.956	0.006	.941

Table 8.

Summary of logistic mixed effects models comparing OA60 vs. OA70+ groups, Experiment 2 (2-lag).

Predictors	Log-odds Estimate	Std. error	χ2	p-Value
Transitive Targets: OA60 vs. OA70+
(Intercept)	−3.097	0.506
Prime	0.302	0.427	11.487	<.001
Age	−0.073	0.776	3.911	0.048
Prime x Age	1.046	0.534	3.837	.050
Dative Targets: OA60 vs. OA70+
(Intercept)	−3.741	0.758
Prime	1.071	0.482	2.223	0.136
Age	0.652	0.874	0.002	.959
Prime x Age	−1.018	0.612	2.768	.096

0-lag immediate priming.

For transitive targets, the model revealed a significant effect of prime type. There were also significant effects of verb type and age group. In addition, the interaction between prime and verb type was significant, indicating that both subgroups of OA showed a significant lexical boost effect. However, importantly, there were no significant age-related 2- and 3-way interactions, indicating that the OA70+ group did not show reduced priming or lexical boost effect compared to the OA60 group. For dative targets, the model comparing the two OA groups revealed only the main effect of prime, indicating that the OA70+ group showed comparable priming effects as the OA60 group. No other main or interaction effects were significant.

2-lag lasting priming.

For transitive targets, there was a main effect of prime, indicating both groups in general produced more passives after passive vs. active primes. There was an effect of age, indicating that the OA70+ group produced fewer passives in general than the OA60 group. The interaction between prime and age was not significant. For dative targets, none of the main or interaction effects were significant. However, the two groups showed comparably small priming effects for dative targets at 2-lag.

In summary, the results of this exploratory analysis indicated that there was no statistically significant reduction in structural priming effects in the OA70+ group, compared to the OA60 group.

DISCUSSION

There has been prolific evidence suggesting that structural priming - humans’ tendency to re-use previously experienced message-syntax associations - is pervasive, creating small but enduring changes in the language production system throughout the lifespan. A growing number of studies has begun to apply structural priming to clinical populations, including persons with aphasia. Specifically, due to its implicit nature, structural priming holds a potential to be used as a remediation strategy to facilitate language re-learning in PWA without heavy reliance on metacognitive rule learning. However, existing studies of structural priming are mostly focused on children and young adults, limiting evidence available for older adults. This study examined the impact of aging on the lexically independent and dependent mechanisms of structural priming in a collaborative picture description task.

The first question examined whether abstract structural priming remains intact in older adults in transitive and dative targets and over different temporal lags (zero vs. two intervening fillers). The older adults showed preserved priming effects in general. At the immediate priming condition, older adults showed increased production of passive sentences after they heard the examiner produce passive compared to active sentences. Similarly, for dative targets, they produced more DO structures after they heard the examiner produce DO compared to PO structures. The older adults indeed showed greater priming effects compared to the young adults in both transitive and dative targets, as indicated by the prime by group interactions. Increased priming in the older adults can be explained by the inverse preference effect, in which the magnitude of priming increases with greater prediction error in less efficient speakers (Anderson & Conture, 2004; Hartsuiker & Kolk, 1998) or more uncommon structures (Bernolet & Hartsuiker, 2010; Jaeger & Snider, 2013; Scheepers, 2003). Because passives and DO structures are used less frequently in older adults, priming experiences might have benefited the older adult group to a greater extent than the young adult group. Over a 2-lag of intervening fillers, priming effects became smaller in both young and older adults compared to the immediate priming condition, consistent with previous findings (Bock et al., 2007; Bock & Griffin, 2000; Branigan & McLean, 2016; Hartsuiker et al., 2008). However, no age-related differences in the magnitude of priming effects were found in both transitive and dative target structures.

These findings add to the growing evidence suggesting that the mechanisms of structural priming are intact in healthy older adults (Hardy et al., 2017; 2020; van Boxtel & Lawyer, 2022). Hardy et al. (2017; 2021) found strong priming effects in transitive sentences for older adults at immediate priming using both a dialogue-like task, similar to ours, (Hardy et al., 2017) and a monologue production-to-production task (Hardy et al., 2020). In the current study, we replicated their findings in transitive targets, but also extended the same findings to dative targets and at a longer-term priming condition. van Boxtel and Lawyer (2022) found no age-differences in priming effects during comprehension of sentences with a reduced relative clause over intervening fillers. Thus, collectively, these findings indicate that structural priming creates adaptation in the abstract syntactic processing system of older adults, independent of tasks of comprehension or production modality (Bock et al., 2007; Chang et al., 2006; Tooley & Bock, 2014).

The lack of age-related differences in the priming effects are not in line with the slowed processing speed theory of aging (Salthouse, 1996) or the Transmission Deficit Hypothesis (Burke & Shafto, 2004). These theories would predict reduced structural priming effects in older adults because their ability to activate the prime structures would not be fast enough or the spreading activation from the prime to the target would not be strong enough to influence their future production. The current results also do not align with the predictions of the declarative memory-based model of structural priming (Reitter et al., 2011). If the older adults relied on declarative memory during the priming task, they would have shown a smaller priming effect than the young adult group. But this prediction was not born out in the current data.

The older adults’ significant abstract priming effects that persisted over a time lag are more in line with theories proposing that structural priming is supported by non-declarative memory or implicit learning (Bock & Griffin, 2000; Chang et al., 2006; 2012; Heyselaar et al., 2017; 2021). Such implicit cognitive processes are generally preserved in older adults (Bennet, et al., 2007; Fleischman et al., 2004; Howard & Howard, 2001; King et al., 1992; Light & Snigh, 1987; but see Heyselaar et al., 2021). Expanding the existing evidence from children and young adults, our findings underscore that the mechanism of abstract structural priming reflects implicit syntactic learning that spans from early childhood through late adulthood (Bock et al., 2007; Bock & Griffin, 2000; Branigan et al., 2000; Branigan & McLean, 2016; Peter et al., 2015; Rowland et al., 2012; Savage et al., 2006). In addition, our findings corroborate previous studies suggesting that the ability to produce and comprehend syntactic structures remains intact in aging (Davidson et al., 2003; Waters & Caplan, 2001).

The second question of this study investigated whether the overlapping lexical information would boost structural priming effects in aging. Both older adults and young adults showed significant lexical boost effects at 0-lag, although the boost dissipated at 2-lag. The absence of lexical boost effects at 2-lag is consistent with previous findings in young adults and children and is in line with the accounts that lexical boost reflects short-term mechanism of priming (Branigan & McLean, 2016; Hartsuiker et al., 2008; Reitter et al., 2011). However, lexical boost effects were very clear at immediate priming conditions for both groups. For transitive targets, the older adults were about 26% more likely to produce passives when the verb was repeated between prime and target compared to when it was not. This lexical boost was greater than the mean 12% boost seen in the young adults and is comparable to the 25.2% lexical boost seen in the older adults of Hardy et al.’s study (2017). For dative targets, the older adults, on average, were 7% more likely to produce double object structures in the same verb compared to different verb prime condition, comparable to the lexical boost of 9% seen in the young adults.

The significant lexical boost effects found in the older adults are consistent with the claim that there is a strong continuity in not only abstract structural but also in lexically-mediated structural priming throughout the life span (Hardy et al., 2017; van Boxtel & Lawyer, 2022). Since we did not obtain specific explicit and implicit memory measures from the participants, it is hard to draw clear conclusions about the nature of cognitive processes supporting lexical boost effects. However, our findings are in line with the studies suggesting that lexical boost may not be solely dependent on explicit memory and at least some non-declarative memory processes are associated with it (Heyselaar et al., 2021; van Boxtel & Lawyer, 2022). This is, however, inconsistent with the dual-path model of language learning, which would predict lexical boost to be reduced with aging because it is thought to be caused by lexical item-specific traces in explicit memory (Chang et al., 2006; 2012; Rowland et al., 2012).

Lastly, our exploratory analysis examined if adults who are older than 70 (70–82 years old) showed reduced structural priming compared to the older adults in their 60s, given that Heyselaar et al. (2021) found a significant decline in structural priming for adults whose age ranged from 70–85 years. We take caution here given the exploratory nature of this analysis. However, different from Heyselaar et al. (2021), there was no clear difference in priming effects between the two subgroups of older adults. This confirms that the significant priming effects in the older adult group of this study are not driven by the performance of those in their 60s. However, this study used a scripted dialogue-like priming task, which could lead to increased priming due to its interactive nature (Branigan et al., 2000), different from Heyslaar et al. (2021) where a production-to-production priming task was used. In addition, we examined only 2-lag condition. Thus, further research is needed to replicate the findings across different tasks and time intervals.

The current findings also provide empirical basis for clinical application of structural priming to age-related disorders such as aphasia. Despite known cognitive-linguistic deficits in elderly populations, both the mechanisms of abstract and lexically-mediated structural priming remain robust and persistent in older adults. More importantly, simply listening to their interlocutor’s production of sentences was sufficient to modify future sentence production in older adults. Similarly, previous studies, albeit a few, showed that structural priming remains preserved and is effective in facilitating production of sentences in older adults with aphasia in both experimental (Cho-Reyes et al., 2016; Lee et al., 2019; Man et al., 2019) and treatment (Lee & Man, 2017) tasks. Therefore, the implicit nature of structural priming may be useful in improving syntactic processing in older adults with language disorders, specifically for those whose metacognitive skills for explicit learning are impaired.

In conclusion, the current study systematically investigated the impact of healthy aging on the mechanisms of abstract structural priming and lexical boost effects. The abstract priming was found to be significant across transitive and dative structures and persistent over a temporal lag of 2 intervening fillers. In addition, lexically-mediated structural priming was preserved in aging, different from theories that predict it should be impaired in OA due to their impaired explicit memory. Taken together the findings demonstrate that the mechanisms of structural priming remain resilient and continue to operate in healthy aging, in line with the theories that attribute structural priming to lifelong implicit language learning. Further, the findings provide empirical basis for applying structural priming as a means to facilitate sentence production in age-related acquired language disorders.