Structural priming from production to comprehension in aphasia

Abstract

Background:

Many people with aphasia (PWA) show deficits in sentence production and comprehension, in part, due to an inefficient mapping between messages and syntactic structures. Structural priming—the tendency to repeat previously encountered sentence structures—has been shown to support implicit syntactic learning within and across production and comprehension modalities in healthy adults. Structural priming is also effective in facilitating sentence production and comprehension in PWA. However, less is known about whether priming in one modality changes PWA’s performance in the other modality, crucial evidence needed for applying structural priming as a cost-effective intervention strategy for PWA.

Aims:

This study examined (a) whether production to comprehension cross-modality priming is effective in PWA, (b) whether priming-induced changes in syntactic comprehension lasted in the absence of an immediate prime, and (c) whether there is a significant correlation between individuals’ priming effects and the change in their comprehension following priming.

Methods & Procedures:

Thirteen PWA and 13 age-matched control participants completed a pre-test, a production-to-comprehension priming block, and a post-test. In the pre- and post-tests, participants completed a sentence-picture matching task with sentences involving interpretations of an ambiguous prepositional phrase (e.g., The teacher is poking the monk with a bat). Participants were free to choose a picture corresponding to a high attachment (HA; e.g., the teacher is using the bat to poke the monk) or a low attachment (LA; e.g., the monk is holding the bat) interpretation. In the priming block, participants produced LA sentences as primes and then completed a sentence-picture matching task for comprehension targets, similar to the pre-test.

Results:

Age-matched controls and PWA showed a significant priming effect when comparing the priming block to the pre-test. In both groups, the priming effect persisted when comparing picture selections in the pre- and post-tests. At the individual level, age-matched controls who showed larger priming effects also selected more LA pictures in the post-test compared to the pre-test, indicating that the priming effect accounted for the magnitude of change from the pre- to post-test. This correlation was also found in PWA.

Conclusions:

These findings suggest that production-to-comprehension structural priming is effective and persistent in PWA and controls, in line with the view that structural priming is a form of implicit learning. Further, the findings suggest that syntactic representations are shared between modalities, and therefore, production influences future comprehension. Cross-modality structural priming may have clinical potential to improve sentence processing in PWA.

Introduction

Many people with aphasia (PWA) show deficits in sentence processing. These deficits are in part attributed to an inefficient mapping of a message onto a syntactic structure during production or vice versa during comprehension (Berndt & Caramazza, 1980; Mitchum & Berndt, 2008; Rochon et al., 2005; Thompson et al., 2015). Mapping deficits affect both production and comprehension modalities in many PWA, although the degree of impairments may differ between the modalities (Caramazza & Hillis, 1989; Caramazza & Miceli, 1991). However, it remains unclear to what extent the syntactic processes between the two modalities interact in PWA and how the syntactic deficits can be ameliorated effectively.

Most previous experimental studies with PWA have focused on either sentence production or comprehension separately, often assuming modality-specific syntactic processes (see Schröder et al., 2015 for review). Similarly, very few treatment studies have tested cross-modality generalization, yielding no clear evidence for bi-directional cross-modality generalization (Adelt et al., 2016; Jacobs & Thompson, 2000; Schröder et al., 2015). For example, a meta-analysis of sentence treatment studies by Adelt et al. (2016) reported that 10/13 PWA showed generalized improvement to production following treatment of sentence comprehension, whereas only 4/26 PWA showed generalization to comprehension following treatment of sentence production. However, more recent theories of language processing and increasing evidence from structural priming studies suggest that syntactic representations are shared between comprehension and production and that the modalities interact (MacDonald, 2013; Pickering & Garrod, 2004, 2013). In addition, structural priming—the tendency to repeat a recently encountered sentence structure—is viewed to reflect syntactic learning processes (Bock & Griffin, 2000; Chang et al., 2006, 2012). The current study examined cross-modality priming from production to comprehension to investigate the interaction between the two modalities and shed light on processes of implicit language learning in PWA.

Cross-modality structural priming in healthy adults

While earlier models of sentence processing have focused separately on the modalities of comprehension or production at the syntactic level, more recent theories propose that syntactic representations are shared between comprehension and production, predicting interactive bi-directional processes between the modalities (MacDonald, 2013; Pickering & Garrod, 2004, 2013). For instance, the Production-Distribution-Comprehension (PDC) model proposed that comprehension is the result of distributional regularities that come from production processes that shape utterance plans (MacDonald, 2013), indicating that production may influence future comprehension. Although the PDC model does not explicitly describe the specific process of how information may be shared between the modalities, it proposes that cognitive demands during language production influence how sentences are formed, and therefore, these demands may shape language comprehension as well.

With respect to cross-modality structural priming, two different proposals can be found in the literature. One model by Chang and colleagues proposes that during comprehension, speakers predict upcoming linguistic elements and then utilize this knowledge to produce speech (Chang et al., 2000, 2006, 2012; Dell & Chang, 2014). Prediction is a top-down process where listeners use past experience to inform what they predict a speaker might say, and Dell and Chang (2014) state that like prediction, production is a top-down process in that an overall message is used to select linguistic elements at a lower level. Therefore, both comprehension and production have similarities and interact. However, they state that prediction-based changes in the syntactic system primarily arise during comprehension. For example, when the actual utterance that a listener comprehends does not match their prediction, a prediction error is created that changes the weighting of connections in the syntactic system in order to reduce the possibility that the error will reoccur. These changes to the syntactic system result in structural priming during future comprehension or production of sentences. Therefore, cross-modality structural priming would be expected to occur primarily from comprehension to production rather than vice versa.

On the other hand, the Interactive Alignment Model (Pickering & Garrod, 2004, 2013) proposes that priming is a core mechanism that supports how comprehension and production processes are aligned between interlocutors in dialogue contexts. During a dialogue, interlocutors actively adapt their linguistic (syntactic in this case) expressions to their conversation partner in order to ease information processing. This interactive alignment is possible because the same syntactic representations are shared bidirectionally between interlocutors as they listen and speak. Since the alignment between interlocutors is essentially alignment between comprehension and production, this bidirectional interaction between modalities is also applied within the individual, who is engaged in both comprehension and production processes while interacting with an interlocutor. Therefore, within the Interactive Alignment Model, structural priming is expected to occur not only from comprehension to production but also from production to comprehension.

Many cross-modality priming studies have focused on comprehension-to-production priming (e.g., Bock et al., 2007; Branigan & McLean, 2016; Branigan et al., 2000 and others). There is less evidence for priming from production to comprehension. One of the first studies to test production-to-comprehension priming was conducted by Branigan et al. (2005). The authors used sentences with an ambiguous prepositional phrase (PP; e.g., The policeman prodding the doctor with the gun) to test if prior production of one syntactic structure influences subsequent comprehension of a similar ambiguous sentence in young adults. For example, the sentence, “The policeman prodding the doctor with the gun” (taken from Branigan et al., 2005) allows two interpretations. One interpretation is that the PP with the gun modifies or “attaches” to the verb prodding. This is called “high attachment” (henceforth, “HA”) because the verb phrase occurs in a high position in the phrase structure tree. In the other interpretation, the PP attaches to the object noun phrase the doctor. Since the object noun phrase occurs in a low position in the phrase structure tree, it is called “low attachment” (henceforth, “LA”). Because these ambiguous sentences flexibly allow both interpretations, they can be useful in determining which syntactic interpretation is preferred by a language user and whether their “preference” can be altered following priming. They also reduce strategies during structural priming as the surface form is the same for both interpretations whereas in alternating (e.g., active/passive) structures, other factors may influence priming (e.g., difficulty with passive morphology). In Experiment 3 of Branigan et al. (2005), participants completed a production-to-comprehension priming task. For the production prime, they were given a verb and then asked to describe a picture using a sentence. In order to ensure that participants produced the prime sentence using only one attachment type (either HA or LA), the participants were presented with a picture that depicted either a HA or LA action (e.g., The policemen prodding a doctor who is holding a gun for a LA production prime). For the comprehension targets, participants read a new sentence with an ambiguous PP and were presented with two pictures that corresponded to the HA and LA interpretations of the sentence. Results showed successful production-to-comprehension priming. Their young adults were more likely to choose the picture that depicted the same interpretation (e.g., HA) as the prime sentence that they produced previously (e.g., HA).

In another study, Litcofsky and van Hell (2019) tested passive sentences in two cross-modality priming tasks. In the comprehension-to-production priming task, participants listened to a passive or active sentence and then described a picture. In the production-to-comprehension priming task, participants first described a prime picture using either a passive or active sentence, starting from the character on the left of the picture and using the color-cued character as the agent in their response. Then, for comprehension targets, they heard an active or passive sentence while their comprehension was assessed using electroencephalography (EEG). The authors found significant priming effects in passive sentences in the comprehension-to-production task, which was evidenced by more passive sentences produced after a passive rather than an active comprehension prime. Additionally, using event-related brain potentials (ERP), they found that there was increased efficiency in processing passive target sentences following production of passive versus active primes. These findings suggest that comprehension and production might share similar processing systems and that information not only flows from comprehension to production but also from production to comprehension.

Another significant implication of structural priming is that it involves implicit language learning, which is evidenced by a persisting priming effect over time (e.g., across sessions) or over linguistic materials (e.g., fillers intervening between primes and targets) (Ferreira & Bock, 2006; see Pickering & Ferreira, 2008 for review). Bock and Griffin (2000) used a sentence production priming task where participants repeated a prime sentence that they heard and then described the picture on their screen. The authors found that priming persisted even when 10 filler items intervened between the prime and target items, suggesting that priming reflects language learning as memory alone would not be able to explain how priming persisted across longer time lags. Bock et al. (2007) further examined lasting priming by testing participants on a comprehension-to-production structural priming task where participants simply heard a prime sentence and then described a picture using a sentence, with various lag conditions (ranging from 0 to 10 intervening fillers). They found that priming effects persisted even across 10 intervening fillers, indicating not only that priming is long-lasting but that this persistence also occurs across modalities from comprehension to production. Lasting priming effects have also been found across experimental sessions, even when separated by a week (Kaschak, Kutta, & Schatschneider, 2011). In Experiments 1 and 2 of Kaschak, Kutta, and Coyle (2014), participants were first biased to produce one structure depending on whether they were assigned to the DO or PO condition by reading sentences aloud (bias phase). Then, participants were asked to describe pictures using a single sentence (prime phase) either immediately after the bias phase (Exp 1) or at least a week after the bias phase (Exp 2). In both experiments, the authors found that priming effects persisted immediately and at 1-week after the bias phase. They found no significant differences in the priming between the experiments, suggesting that priming persisted up to at least a week and did not significantly decay over time.

The long-term effects of structural priming have also been demonstrated across the lifespan in young children (Branigan & McLean, 2016; Peter et al., 2015; Savage et al., 2006) and healthy older adults (Heyselaar et al., 2020; Lee et al., accepted). One way to test for a lasting effect of structural priming is to compare performance in a participant’s independent comprehension or production of target structures before and after a priming task. For example, to examine if structural priming improves independent production of passive sentences in children, Kidd (2012) used a priming task with three phases: an unprimed baseline (picture description), a priming block, and an unprimed post-test. If priming is truly indicative of implicit learning, the children should have shown an improved ability to produce passive sentences in the post-test, even in the absence of a prime. Indeed, the children maintained the priming effect: they produced more passive sentences at post-testing compared to baseline, indicating a priming-induced, lasting change to their syntactic production.

Structural priming studies in aphasia

Compared to the structural priming literature in healthy individuals, the application of structural priming to aphasia has been limited. However, growing findings suggest that structural priming can be effective in facilitating both immediate and lasting changes in sentence production and comprehension in PWA (Hartsuiker & Kolk, 1998; Keen et al., 2021; Lee & Man, 2017; Lee, Hosokawa, et al., 2019; Lee, Man, et al. 2019; Man et al., 2019; Rossi, 2015; Saffran & Martin, 1997; Verreyt et al., 2013; Yan et al., 2018). Earlier work by Saffran and Martin (1997) and Hartsuiker and Kolk (1998) first demonstrated that structural priming can facilitate the production of dative or transitive sentences in PWA and that PWA have preserved ability to reuse the structure of the prime sentence in their own production without explicit instructions to do so. Later work reported that PWA showed structural priming effects in other structures as well, including production of clitic pronouns for gender and number information in Italian (Rossi, 2015) and comprehension of sentences with an ambiguous prepositional phrase (e.g., the waitress is hitting the boxer with the umbrella; Lee, Hosokawa, et al., 2019). Lastly, more recent studies have shown that priming effects persist over intervening fillers (Cho-Reyes et al., 2016; Lee, Man, et al., 2019; Man et al., 2019) and over time, even lasting up to a month following multiple sessions of structural priming training in production (Lee & Man, 2017), supporting the possibility that structural priming reflects language learning processes in PWA, similar to what has been shown in healthy adults.

Although these findings supporting the use of structural priming in aphasia are promising, there remain important limitations that need to be addressed. First, we do not know whether structural priming occurs bi-directionally across production and comprehension modalities. Most structural priming studies in aphasia have focused on priming either within the production modality (Cho-Reyes et al., 2016; Hartsuiker & Kolk, 1998; Saffran & Martin, 1997; Yan et al., 2018) or within the comprehension modality (Lee, Hosokawa, et al., 2019). For cross-modality priming, there is currently only evidence for comprehension-to-production priming (Keen et al., 2021; Man et al., 2019). These studies reported significant priming effects for transitive sentences in a dialogue-like comprehension-to-production task. However, no study so far has examined production-to-comprehension priming in aphasia. Without clear evidence for bidirectional cross-modality priming, it would be difficult to know to what extent syntactic processes interact between the modalities in PWA.

A second limitation is that most priming studies in aphasia have measured changes in target responses in the presence of prime sentences, failing to provide strong evidence that structural priming creates lasting changes that impact independent (without preceding primes) sentence production or comprehension. Only two studies so far have addressed this question in the domain of sentence production in aphasia. Saffran and Martin (1997) examined prime-independent behavior of transitive and dative sentence production by utilizing post-tests administered at various times after the priming task (ranging from one hour to a week) in five PWA. PWA were more likely to produce passive sentences in the sentence elicitation task in the post-test after showing successful priming effects during a priming task, indicating that priming increased the production of passive sentences even in the absence of a prime. In addition, although there was not a significant priming effect for dative sentences in the priming task, PWA showed a significant increase in producing dative sentences in the post-test following the priming task. In a more recent study, Lee and Man (2017) found that structural priming training of dative sentences improved independent production of target sentences on daily probes at immediate post-testing and a 4-week follow-up session in one patient with agrammatic aphasia. Although these studies provide evidence that priming persists even when a prime is not present, both studies utilized a small sample size, and their long-term effects were measured within the production modality only. This study examined whether PWA show changes in their interpretation of syntactically ambiguous sentences before and after a priming task by using a pre- and post-test.

Current Study

This study aimed to investigate the effectiveness of production priming on syntactic ambiguity resolution during comprehension in PWA and their age-matched controls. A three-phase experimental design was used which included a pre-test, a priming block, and a post-test. The priming block was similar to that of Branigan et al. (2005), where production primes elicited descriptions of a picture with a low attachment (LA) and comprehension targets involved participants selecting a picture that “matched” the target sentence. Using this design, participants’ preference for ambiguous PP interpretations was assessed before priming, in the presence of production priming, and after priming in order to observe whether priming effectively influenced their syntactic choices during comprehension.

Three research questions were investigated. The first question was whether PWA and their age-matched controls would show cross-modality structural priming from production to comprehension in the priming block. It was predicted that both groups would show a significant priming effect during the priming block, as measured by an increased number of LA interpretations in the comprehension target responses compared to the pre-test. This prediction was based on the findings of Branigan et al. (2005) and previous findings where older adults with and without aphasia show preserved priming effects, similar to young adults (Heyselaar et al., 2020; Lee, Hosokawa, et al., 2019; Lee, Man, et al., 2019; Man et al., 2019). The second question addressed lasting priming effects, that is, whether PWA and their age-matched controls would show an increase in their LA comprehension target responses in the unprimed post-test compared to the pre-test. A persistent priming effect in the post-test would indicate that priming made lasting changes to their syntactic comprehension, in support of structural priming as a form of implicit learning. It was predicted that in the post-test, PWA and their age-matched controls would select more LA interpretations in the unprimed post-test compared to the pre-test. The third question examined correlations between individuals’ degrees of priming effects and the pre- to post-test changes in their sentence comprehension. It was predicted that if participants’ increased selection of LA interpretations in the post-test is indeed associated with the cross-modality priming effects, participants who showed larger priming effects in the priming block would also show greater increases in their selection of the LA interpretations in the post-test.

Methods

Participants

Thirteen people with aphasia (PWA; 8 women, 5 men) were enrolled in this study. PWA were between 35 to 82 years old (M = 61.4, SD = 15.9) and completed between 12 to 19 years of education (M = 15.4, SD = 2.5). Fourteen age- and education-matched control participants were also enrolled in this study. Participants were recruited from the Greater Lafayette area, through social media, or referred from aphasia programs in other areas. Informed consent was obtained for all participants prior to testing.

All participants were native speakers of American English with at least an 8^th-grade education and reported no history of neurological or psychological disorders. PWA had no history of neurological or psychological disorders prior to their stroke. All participants self-reported that they had normal or corrected-to-normal hearing and passed an informal hearing screening (500, 1000, and 2000 Hz pure tones at 40 dB at 80% computer volume without headphones). They also self-reported having normal or corrected-to-normal vision and passed an informal near vision screening at a distance of 16 inches using a digital version of the LEA near vision screener (adapted from Hyvärinen, 2018). These screenings were used to ensure that participants could hear and see the stimuli that were used in the task. In addition, control participants completed the Mini Mental State Examination (MMSE; Folstein, Folstein, & McHugh, 1975) and scored within normal limits for cognition based on their age and education (M = 29.1, SD = 1.5).

Language Testing for PWA

Each PWA had a diagnosis of aphasia with a post-stroke onset time of between 24 and 177 months (M = 67.1, SD = 54.2), as verified by their medical records. PWA were assessed using a combination of language tests (see Table 1). Given that the experimental task involved sentence-level stimuli, PWA with mild-to-moderate aphasia severity were recruited as indicated by the Aphasia Quotient (AQ) from the Western Aphasia Battery-Revised (WAB-R; Kertesz, 2006). All PWA showed relatively preserved auditory comprehension of single words and different types of simple sentences on the Auditory Comprehension test of the WAB-R. Their comprehension of verbs was also confirmed to be relatively intact, using the Verb Comprehension Test of the Northwestern Assessment of Verbs and Sentences (NAVS; Cho-Reyes & Thompson, 2012). PWA were able to comprehend and produce simple sentences but were impaired when comprehending and producing more complex sentences as shown in their performance on the Argument Structure Production Test and Sentence Comprehension Test of the NAVS. Participants who completed the study remotely were assessed using remote-adapted versions of the WAB-R and NAVS following the same procedures as previous studies (Dekhtyar et al., 2020; Keen et al., 2021; Pearson Assessments, 2021).

Table 1.

Language testing results for people with aphasia (PWA).

	WAB-R					NAVS
PWA	AQ (100)	Fluency (10)	AC (10)	Repetition (10)	Naming (10)	VNT (100)	VCT (100)	ASPT (100)	SPPT (100)	SCT (100)
A1*	90.2	9	9.8	8.6	8.7	100	95.5	96.9	73.3	80
A2	95.2	9	10	8.6	10	100	100	100	100	86.7
A3*	81.6	5	9.5	9.1	9.2	60	100	90	93.3	76.7
A4	68.8	5	10	5.1	7.3	86.4	955	93.8	80	100
A5*	75.9	5	8.7	6.9	8.4	81.8	100	65.6	60	80
A6*	89.2	9	8.7	8.7	9.2	86.4	100	96.9	50	70
A7	88.1	9	8.9	9.3	7.8	86.4	100	100	100	90
A8	76.7	5	8.9	8.1	8.4	59.1	95.5	53.1	40	60
A9	70.1	5	8.8	6.5	6.8	86.4	100	94.4	33.3	63.3
A10	79.3	6	9	10	6.7	95.5	100	75	80	100
A11	69.6	5	8.8	6.9	7.1	77.3	100	62.5	16.7	76.7
A12*	87.7	9	9.4	7.2	9.3	100	100	100	73.3	83.3
A13*	77	9	7.6	4.8	8.1	63.6	90.9	84.4	66.7	50
Mean	80.7	6.9	9.1	7.7	8.2	83.3	98.3	84.8	66.7	78.2
SD	8.7	2	0.7	1.6	1	14.7	3	16	25.9	14.8

Note. WAB-R = Western Aphasia Battery–Revised; AQ = Aphasia Quotient; AC = Auditory Comprehension; NAVS = Northwestern Assessment of Verbs and Sentences; VNT = Verb Naming Test; VCT = Verb Comprehension Test; ASPT = Argument Structure Production Test; SPPT = Sentence Production Priming Test; SCT = Sentence Comprehension Test.

^*Participant was assessed using remote-adapted versions of the assessments.

Experimental Design & Tasks

All participants completed the experimental sequence which consisted of a pre-test, a priming block, and a post-test in a single experimental session. As shown in Figure 1, the pre-test consisted of a sentence-picture matching task where participants heard and saw a target sentence with an ambiguous prepositional phrase presented along with a pair of pictures that allowed for either a low attachment (LA) or high attachment (HA) interpretation. All participants in both groups chose HA pictures more than 80% of the time in the pre-test, so they were considered to show a preference for HA interpretations of the target stimuli.

Figure 1.

Example of a comprehension target trial in the pre- and post-tests. The black left and right arrows on the screen corresponded to the left and right arrow keys on the participant’s keyboard which were used to select pictures.

After the pre-test, participants completed a priming block that contained the production-to-comprehension priming task. The priming block consisted of prime-target pairs, which were used to test whether priming the opposite interpretation from the participant’s initial preference in the pre-test would significantly influence them to adopt the primed interpretation. Since participants showed a HA preference (selected more than 80% HA pictures in the pre-test), they were all trained on the LA priming block (100% LA primes). The rationale was that in order to induce a larger priming effect, using the participant’s less preferred structure should produce more prediction error and therefore increase priming as a result of the inverse frequency effect (Hartsuiker & Westenberg, 2000; Kaschak, Kutta, & Jones, 2011; Scheepers, 2003). After a brief break, participants started the post-test. The post-test consisted of the same sentence-picture matching task as the pre-test. However, the post-test included different sentence and picture stimuli from the pre-test and the priming block to ensure that participants’ performance in the pre-test was not due to repetition effects of the stimuli.

Stimuli

For prime and target experimental stimuli, a total of 120 sentences with ambiguous prepositional phrases and corresponding black-and-white line drawings were prepared (see Figure 1 for a sample picture stimulus). These stimuli were adapted from Branigan et al. (2005) and Lee, Hosokawa, et al. (2019). To create the sentence stimuli, a set of 6 transitive verbs (hit, thump, hurt, injure, poke, and prod), 17 human characters (e.g., dancer, chef), and 10 instruments (e.g., key, umbrella) were selected. A new instrument—broom—was included in these instruments in order to increase the original stimuli to the required number of items for the tasks. All verbs and nouns were less than three syllables in length (verbs: M = 2.2, SD = 0.4; animate nouns: M = 1.9, SD = 0.6; instruments: M = 1.5, SD = 0.8). Human characters were counterbalanced so that they appeared once in the agent role and once in the patient role. Each verb was repeated 20 times and combined with different human characters (one agent, one patient) and one instrument, yielding a total of 120 novel sentences. Twenty of the sentence stimuli were used as comprehension targets in the pre-test, 80 were used in the priming block which were split further into 40 production primes and 40 comprehension targets, and the remaining 20 were used as comprehension targets in the post-test. The same verbs were used between prime and target sentences as previous evidence has shown that overlapping verbs generally elicit greater priming effects in both sentence comprehension (Arai et al., 2007; Tooley & Traxler, 2010; Tooley et al., 2009) and production (Branigan & McLean, 2016; Man et al., 2019). In addition, a total of 120 intransitive filler sentences were prepared. Filler sentences were combined with black-and-white drawings in the same way as the prime and target sentences.

Setup

This study was created and implemented in the cloud-based experiment builder, Gorilla (www.gorilla.sc). Gorilla presented picture and audio stimuli in each phase and recorded participant utterances on prime trials which were saved as web audio files. The study was also hosted in Gorilla, and participants were provided a URL where the experimenter entered the participant’s study ID and started the study. Testing and procedures were delivered either in person or remotely depending on each participant’s preference. Previous research has demonstrated equivalent results between in-person and remote testing with PWA (Hall et al., 2013), and our previous study has established that structural priming using secure, web-based videoconferencing is as effective as in-person structural priming tasks for PWA (Keen et al., 2021). Participants tested in person completed the study using a laptop in the Purdue Aphasia Lab which was connected to Wi-Fi. Participants who completed the study remotely were required to use a laptop with a screen that was at least 13 inches diagonally and had access to a stable internet connection and a quiet environment, similar to what was required for in-person testing. Remote participants met with the experimenter through a secure version of Zoom (www.zoom.us). The participant was then taught how to share their screen and computer audio with the experimenter, so that the experimenter could monitor their progress throughout the session. The audio from each session was recorded in Camtasia (Camtasia 2022, Techsmith). Seven PWA and 7 controls completed the study in person and the other 6 PWA and 6 controls completed the study remotely. All data collected from participants were stored in secure, HIPAA-compliant cloud storage provided by Purdue University.

Task Procedure

Participants were seated in front of a laptop computer to begin the study. Participants were told that they were completing a study on matching pictures and sentences and describing pictures. Instructions for the study were given on the screen bimodally, presented both in writing and audibly. Once the participant confirmed they understood the instructions, they were asked to press the right arrow key on their keyboard to start the study.

Pre-test and Post-test

For comprehension of the target ambiguous sentences, a sentence-picture matching task was used in the pre-test, as shown in Figure 1 above. Before the pre-test began, 6 practice trials were presented. The two pictures allowed for either a HA or LA interpretation for the target sentence. Participants were asked to press the left or right arrow key on their keyboard (left and right black arrows were also shown under each picture, as illustrated in Figure 1) to select which picture best matched the sentence they heard and saw at the top of their screen. The positions of the pictures were counterbalanced so that HA or LA pictures only appeared 50% of the time on one side of the screen. One comprehension filler was placed after each target item. Similar to the experimental target setup, comprehension filler trials displayed two pictures: one picture that matched the filler sentence the participant heard and saw at the top of their screen (e.g., the alarm is ringing; the bird is singing) and one picture that differed from the filler sentence in either subject (e.g., the phone is ringing) or verb (e.g., the bird is flying). The procedure for the post-test was the same as the pre-test, and the post-test was presented after the priming block.

Priming Block

Figure 2 shows an example of a production prime-comprehension target pair in the priming block. For the production of prime sentences, a cued sentence completion task was employed, similar to Branigan et al. (2005). A prime sentence template with blanks for the goal and theme was presented at the top of participants’ screen to ensure that PWA were able to produce the prime sentence in the required LA structure. Participants were instructed that after they saw “Describe the picture” on the screen, they would hear a beep sound and see a picture. Participants described the picture by using the sentence at the top of the screen and filling in the blanks of the sentence using all the labeled words in the picture (e.g., The dancer is poking the [policeman] with a [broom]). Importantly, different from comprehension target items, the prime items were accompanied by only one picture which allowed for a LA interpretation, ensuring that participants were producing LA sentences.

Figure 2.

Example of a production prime-comprehension target trial in the priming block.

For the comprehension of the target ambiguous sentences, the same sentence-picture matching task was used as in the pre-test. Participants would see “Listen” on the screen and then see two pictures and hear an ambiguous PP sentence that also appeared at the top of their screen. Participants were instructed to select the picture that best matched the sentence. Two filler items occurred after prime-target pairs to disguise the purpose of the task. The first filler item involved production using a sentence completion task where participants saw a single picture and an intransitive sentence frame at the top that was missing the verb. The second filler item was identical to the comprehension filler items in the pre- and post-test.

Participants completed a set of 6 prime-target trials as practice, and then participants began the priming block. Immediate feedback on the production accuracy of the prime sentences was given for the practice trials, but only neutral feedback (e.g., “you’re doing fine”) was provided for experimental trials.

Data Analysis

Accuracy of production primes

Production primes produced by each participant were transcribed by the researcher and trained lab assistants. Each transcribed response was scored as correct (1) if the prime sentence was produced with the correct nouns in the blanks of the sentence frame (e.g., “The dancer is poking the policeman with a broom”). Variations in verb tense forms and omission of articles or copular verbs were ignored for PWA, but controls were required to produce prime sentences without errors. When target verbs were substituted, only semantically similar verbs (e.g., “hitting” for “punching”) were accepted as correct.

Analysis of comprehension trial responses

Comprehension of experimental trials was measured by calculating the proportion of LA picture selections in the pre-test, priming block, and post-test. To test whether immediate production-to-comprehension priming was effective, priming effects were measured as the difference in the proportion of LA pictures selected after a correctly produced LA prime sentence in the priming block versus the proportion of LA pictures that were selected in the pre-test. For example, if the proportion of LA pictures selected in the pre-test was 30% and the proportion of LA pictures selected after a correctly produced LA prime was 60%, the priming effect would be the difference between the two, in this case, 30%. To determine whether the priming effect lasted in the post-test, the difference in LA picture selections from the pre- to post-test was calculated.

Statistical Analysis

Statistical analyses were conducted using mixed-effects logistic regression in R (lme4 package; Bates et al., 2015). To answer the first question on whether immediate cross-modality priming (production to comprehension) successfully occurs, a logistic mixed-effects model was used with fixed factors of phase (pre-test vs. priming block), group (controls vs. PWA), and their interaction. To answer the second question on lasting priming effects in the post-test, another logistic mixed-effects model was performed with fixed factors of phase (pre-test vs. post-test), group (controls vs. PWA), and their interaction. In all of these models, by-participant and by-trial intercepts and slopes were included as random effects (see Supplemental Material 1 for model codes). To answer our third question, a binary correlation analysis was conducted between individual participants’ immediate and lasting priming effects to test whether those who showed priming effects in the priming block also showed a larger increase in selecting LA pictures in the post-test. Effect sizes were calculated using Cohen’s d (Cohen, 1992) to indicate the magnitude of change in LA picture selections between phases in PWA and control participants. Lastly, post-hoc analyses were conducted to examine whether participants’ LA responses changed over trials in the priming block and post-test. Within each phase (priming block, post-test), a logistic mixed-effects model was performed with fixed factors of time (first half vs. second half of the trials), group (controls vs. PWA), and their interaction. By-participant and by-trial intercepts and slopes were included as random effects (see Supplemental Material 1 for model codes).

Results

All participants completed the pre-test, priming block, and post-test in a single session with the exception of A12 who completed the post-test within 24 hours after the priming block due to medical concerns. One control participant’s data were excluded from analysis because the participant reported using an explicit strategy during the priming block.

Prime Accuracy

Age-matched controls produced prime sentences at ceiling level (100% correct). PWA overall showed relatively high accuracy in producing the prime sentences, although their scores were lower than control participants (M = 89.6, SD = 11.76), t(24) = 3.27, p < .001.

Pre-test vs. Priming Block

Participants’ immediate priming effects (i.e., performance in the pre-test vs. priming block) are shown in Figure 3 (see Supplemental Material 2 for individual data). The results of the 2 (group) by 2 (phase) mixed-effects model comparing LA selections in the pre-test vs. the priming block indicated that there was a main effect of phase, such that participants selected a greater proportion of LA pictures in the priming block compared to the pre-test, β = 2.16, SE = 1.09, z = 1.98, p = .047. There was not a significant group effect, indicating that overall, the proportions of LA selections did not significantly differ between the two groups, β = 1.20, SE = 1.16, z = 1.04, p = .30. In addition, the interaction between phase and group did not reach statistical significance, indicating that the magnitude of change in LA selections from the pre-test to the priming block was not significantly different between the two groups, β = −0.93, SE = 1.39, z = −0.67, p = 0.50. The effect sizes calculated within each group indicated a medium effect size for PWA (pre-test: 17.3% vs. priming block: 28.7%), d = 0.55, and for control participants (pre-test: 6.5% vs. priming block: 31.3%), d = 0.78.

Figure 3.

Percent of LA picture selections in the pre-test vs. priming block in PWA and control participants.

Pre-test vs. Post-test

Participants’ lasting priming effects (i.e., performance in the pre-test and post-test) are shown in Figure 4. A 2 (group) by 2 (phase) mixed-effects model revealed a significant main effect of phase, indicating that participants chose LA interpretations significantly more frequently in the post-test compared to the pre-test, β = 2.66, SE = 1.21, z = 2.19, p = .028. The group effect did not reach significance, β = 1.76, SE = 0.92, z = 1.91, p = 0.57. The phase by group interaction was not statistically significant such that the two groups did not statistically differ in their pre-to-post changes in LA picture selections, β = −1.94, SE = 1.50, z = −1.29, p = 0.20. The effect sizes calculated within each group, however, indicated a reduced effect size for PWA (pre-test: 17.3% vs. post-test: 26.9%), d = 0.41, compared to control participants (pre-test: 6.5% vs. post-test: 35%), d = 0.88.

Figure 4.

Percent of LA picture selections in the pre-test vs. post-test in PWA and control participants.

Correlation Analysis

The results from the correlation analyses indicated a strong positive correlation for the control participants, suggesting that those who showed larger priming effects in the priming block tended to show a greater increase in LA responses in the post-test (Figure 5a), r(11) = .97, p < .001. Similarly, a positive correlation was found in PWA (Figure 5b), although it was not as strong as in the control group, r(11) = 0.56, p = .046.

Figure 5.

Correlations between immediate priming effects (pre-test vs. priming block) and lasting priming effects (pre-test vs. post-test) for individual control participants (a) and PWA (b).

Post-hoc Analysis

Participants’ performance in the first and second halves of the priming block and post-test are shown in Figure 6. During the priming block, both groups of participants tended to select more LA pictures in the second half than in the first half of the trials, although the effect did not reach significance, β = 1.29, SE = 0.68, z = 1.89, p = .059. There was not a significant group effect, β = 1.54, SE = 2.00, z = 0.77, p = .44, or an interaction between time and group, β = −0.83, SE = 0.84, z = −0.98, p = .33.

Figure 6.

Percent of LA picture selections in the first half and second half of trials in the priming block and post-test in PWA and control participants.

In the post-test, the effect of time was not significant, indicating that the long-term priming effect during the post-test remained stable over trials, β = −0.87, SE = 0.66, z = −1.33, p = .19. In addition, there was not a significant group effect, β = −0.05, SE = 2.12, z = −0.023, p = .98, or time by group interaction in the post-test, β = 0.63, SE = 0.74, z = 0.86, p = .39.

Discussion

Structural priming effects have been shown to facilitate sentence production and comprehension in people with aphasia (PWA); however, little is known about how priming in production affects subsequent comprehension processes in aphasia. The present study examined immediate and lasting effects of production-to-comprehension structural priming in PWA and age-matched controls, using sentences with ambiguous prepositional phrases. Participants’ preference for pictures with a high attachment (HA) vs. low attachment (LA) interpretation was assessed in three phases: pre-test, LA priming block, and post-test. Three research questions investigated (a) whether production to comprehension cross-modality priming is effective in PWA and control participants during the priming block, (b) whether priming induced changes in syntactic comprehension lasted into the post-test, and (c) whether there is a significant correlation between individuals’ priming effects and the changes in their comprehension in the post-test.

The first set of findings revealed that production-to-comprehension structural priming was preserved in PWA and their age-matched controls. Both groups selected pictures corresponding to LA interpretations more frequently during the priming block compared to the pre-test, where they selected very few LA interpretations in response to the target ambiguous sentences. The evidence of preserved production-to-comprehension priming effects found in our control participants replicate and extend the previous findings from young adults, suggesting that structural priming creates changes in the central syntactic system. In Branigan et al. (2005), the young adults selected 21% more target comprehension responses matched to the production prime interpretation, comparable to the 24% priming effect shown in our age-matched controls. Litcofsky and van Hell (2019) found that production of passive prime sentences facilitated real-time comprehension processing of passive sentences, using EEG.

Together with previous findings, the current study adds to the evidence that the comprehension and production modalities share syntactic representations and that the modalities interact such that prior production influences subsequent comprehension processes. More specifically, the findings support models such as the Interactive Alignment Model which claim that interlocutors align their syntactic expressions to their speech partner, using priming within and between the comprehension and production modalities to ease information processing (Pickering & Garrod, 2004; 2013; see also MacDonald, 2013). Thus, production primes are expected to influence subsequent comprehension, as shown in the current results. On the other hand, the findings of this study are more difficult to explain within models of structural priming that predict priming to only occur from comprehension to production (Chang et al., 2000, 2006, 2012). For example, Chang et al. (2006) claim that structural priming is a result of prediction errors created by a conflict between what a listener predicts they will hear and what they actually hear during comprehension. These prediction errors arise during comprehension, resulting in structural priming effects; thus, priming effects are expected to occur from comprehension to production.

This is the first study showing that production-to-comprehension priming is preserved in PWA. Previous studies on structural priming in aphasia have found within-modality priming (Cho-Reyes et al., 2016; Hartsuiker & Kolk, 1998; Lee, Hosokawa, et al., 2019; Saffran & Martin, 1997; Yan et al., 2018) and priming from comprehension to production (Man et al., 2019). The results of this study indicate that syntactic representations are shared between the modalities and that changes in production can affect the comprehension modality, at least in some PWA (Adelt et al., 2016; Weinrich et al., 2011). The results also support neurolinguistic approaches to aphasia which proposed that deficits in central syntactic processes may underlie sentence production and comprehension errors seen in PWA (Berndt & Caramazza, 1980; Caramazza & Miceli, 1991). However, there seems to be variability in how prior production affects subsequent comprehension in PWA. Schröder and colleagues (2015) found that overall, patients’ comprehension of object-relative clauses did not improve following production treatment. Two out of the seven PWA did show improvements in comprehension, but the result was not significant. On the other hand, there is evidence that sentence comprehension does improve after mapping treatment in sentence production (Weinrich et al., 2011). This variability may indicate that the strength of the link from production to comprehension depends on the severity and nature of linguistic impairments in aphasia. Further research is needed to take into account how the specific degree and nature of deficits between modalities might influence learning from production to comprehension.

The findings from the second and third research questions support the view that structural priming reflects language learning and that priming-based learning is preserved in aphasia. Control participants selected more LA pictures in the post-test compared to the pre-test, indicating that structural priming was effective, and the participants’ implicit learning of the LA structure persisted even in the absence of a prime sentence. Similar to controls, PWA showed increased selections of LA interpretations in the post-test compared to the pre-test. If the cross-modality priming seen in our participants was a product of a mere repetition of the preceding prime’s sentence structure or simply due to visual similarity between prime and target LA pictures during the priming block, we would not have seen significant changes in their interpretation of ambiguous PP sentences in the post-test. The correlation analyses showed that individuals who showed larger priming effects in the priming block also showed larger increases in LA interpretations from the pre- to post-test. This pattern held true in both control participants and PWA, although the correlation was smaller for PWA due to greater individual variability. In addition, the post-hoc analysis revealed that both groups showed a slightly increased (although not statistically significant) priming effect in later trials of the priming block and that neither group showed a significant dissipation in their LA responses over the course of the post-test. Although preliminary, this additional evidence further suggests that the participants showed a greater tendency to make LA interpretations as they produced more LA production primes during the priming block, and this priming-induced learning remained fairly stable during the post-test, once established.

Overall, these results suggest that structural priming from production to comprehension made changes to the syntactic representations available to healthy speakers and PWA which lasted even when priming was not present. These findings are consistent with the view that structural priming reflects lifelong implicit language learning, as evidenced by persisting structural priming effects found in children (Kidd, 2012; Savage et al., 2006), young adults (Shin & Christianson, 2012; Kaschak, Kutta, & Coyle, 2014), and older adults (Hardy et al., 2017; Heyselaar et al., 2020). More importantly, the results from the PWA of this study add to increasing evidence that implicit language learning through structural priming is preserved in aphasia. Recent structural priming studies have found that PWA show persistent priming effects over intervening linguistic materials in both sentence production (Cho-Reyes et al., 2016; Lee, Man, et al., 2019; see Man et al., 2019 for comprehension-to-production priming) and sentence comprehension (Lee, Hosokawa, et al., 2019). Further, PWA showed improved independent production of sentences after priming training (Lee & Man, 2017; see also Saffran & Martin, 1997). The current study shows that even a single session of priming resulted in a measurable change in independent interpretation of syntactically ambiguous sentences. Our PWA showed positive production-to-comprehension priming effects which created lasting changes to their independent comprehension in the post-test. Taken together, these findings provide evidence that structural priming paradigms hold clinical potential for treating language deficits in PWA.

One goal of aphasia rehabilitation is to develop cost-effective treatments, whereby maximum generalization effects can be obtained. Current sentence production treatments, such as the Treatment of Underlying Forms (TUF; Thompson & Shapiro, 2005) and mapping therapy (Rochon et al., 2005), have demonstrated that improvements in the targeted sentence structure can generalize to untrained sentence structures within the production modality, although the outcomes may vary across individuals. However, little evidence is available on whether improvements in these production treatments generalize to the untrained comprehension modality (Adelt et al., 2016; Schroder et al., 2015), indicating a need to improve intervention options for PWA. Considering the results of this study, when attempting to target comprehension deficits by using production treatment, structural priming may be a useful strategy to assist PWA in gaining generalized improvements between the production and comprehension modalities.

There remain limitations in the current study, warranting further research. First, our target sentence structure, which contained ambiguous prepositional phrases, is not frequently used in daily communication and might not be as strongly activated in PWA with difficulty accessing syntactic constructions. Therefore, replicating the findings across a variety of sentence structures would improve generalizability of the findings. The current study also limited the “lasting” effect of priming to the post-test which immediately followed the priming block. Thus, it would be prudent for future studies to test priming effects at longer time points such as 1-week or 1-month post-priming, as shown in previous production-based structural priming studies with PWA (Saffran & Martin, 1997; Lee & Man, 2017). Lastly, it was not within the scope of this study to examine the underlying relationship between individual PWA’s language deficits and their responses to cross-modality priming. More systematic research is needed on how deficits in one or both modalities interact with the degree to which PWA benefit from cross-modality structural priming.

In conclusion, this study used a novel paradigm to investigate whether cross-modality priming from production to comprehension made lasting changes to sentence processing in aphasia. The findings indicate that cross-modality structural priming was preserved in PWA and their age-matched controls, and the effects of priming persisted even in the absence of prime sentences, supporting evidence that structural priming creates lasting changes in how PWA access syntactic representations. Therefore, cross-modality structural priming from production to comprehension holds clinical potential to be used to ameliorate comprehension deficits in aphasia.