No looking back: the effects of visual cues on the lexical boost in structural priming

ABSTRACT

Four structural priming experiments investigated the lexical boost effect in structural priming. In two experiments, we tested whether repeating the subject in prepositional object or double object ditransitive structures boosted structural priming. In two other experiments, we manipulated the repetition of the verb. Repetition of the subject noun affected structural priming, but only when the prime remained visible while participants produced the target sentence. In contrast, repetition of the verb boosted priming regardless of whether participants could see the prime and target simultaneously. We conclude that the subject noun repetition effect is more strategic in nature than the verb boost effect. Structures are automatically associated with the verb, their syntactic head, whereas repetition of the subject noun only affects priming if the presentation method makes the repetition highly explicit.

KEYWORDS:

• Language production
• syntactic representation
• structural priming
• lexical boost
• syntactic head

Introduction

Research has shown that language producers tend to repeat syntactic structures across utterances. This finding has been referred to as syntactic or structural priming. It has been observed in naturally occurring text and conversation as well as in psycholinguistic experiments (see Pickering & Ferreira, 2008 for an overview). One critical finding has been the lexical boost effect, the finding that structural priming is larger when particular words are repeated between prime and target. In a seminal study, Pickering and Branigan (1998) demonstrated this effect in a written sentence completion task. Participants received booklets containing a list of sentence fragments. They first wrote a completion to a prime fragment that encouraged the production of either a prepositional object (PO) ditransitive structure (1a) or a double object (DO) structure (1b). Next, they provided a written completion to a target fragment that allowed either a PO or DO structure (2).

1a. The racing driver showed the torn overall … 

1b. The racing driver showed the helpful mechanic … 

2. The patient gave … 

They found that their participants produced more PO target structures after PO than DO primes in cases such as (1-2), but critically, the priming effect was larger when the verb was the same in prime and target (e.g. “showed” in both) than when it was not (e.g. “showed” and “gave”). Based on their findings, Pickering and Branigan proposed a residual activation model in which combinatorial information (i.e. the structures with which a word can co-occur) is associated with the verb. Structural priming in the absence of verb repetition is due to residual activation of the combinatorial information (e.g. activation of the PO or DO structure), whereas the lexical boost effect is due to activation of the associative link between verb lemmas and combinatorial information that specifies the structures with which the verb can occur (e.g. a link between the PO structure and “show”).

Many studies have since replicated the lexical boost with verb repetition using different structural priming methods (e.g. Branigan et al., 2000; Chang et al., 2015; Corley & Scheepers, 2002; Hartsuiker et al., 2008; Schoonbaert et al., 2007; Segaert et al., 2016; Van Gompel et al., 2012). However, the question remains whether the repetition of other words from the prime causes a similar boost. Pickering and Branigan’s (1998) residual activation model suggests that the lexical boost only occurs when the verb is repeated, because it is the syntactic head that licenses the PO or DO. The boost should not occur when other words in the sentence are repeated because they do not license these structures.

Pickering and Branigan’s (1998) account of the lexical boost contrasts with that of Chang et al. (2006) and Reitter et al. (2011), which predict that the repetition of any content word, regardless of whether it is the head of the primed structure, should cause a boost. Chang et al. assumed that the lexical boost occurs because a repeated word acts as an explicit memory cue for the structure in the prime sentence and therefore any repeated content word, whether a verb or a noun, should cause a priming boost. In their model, the memory mechanisms underlying the lexical boost are very different from the memory mechanisms when no content words are repeated. The repetition of a word activates an explicit short-term memory trace of the sentence, whereas abstract priming when no words are repeated is due to implicit learning. Similarly, Reitter et al.’s ACT-R model assumes that the two effects are different in nature. Abstract structural priming is due to base-level learning and spreading activation, whereas the lexical boost is due to spreading activation in combination with associative learning, during which structures become associated with word meanings.

Recently, Scheepers et al. (2017) and Carminati et al. (2019) investigated whether a lexical boost also occurs with the repetition of the nouns in ditransitive PO/DO sentences, which are not the head of the PO or DO. Their experiments yielded conflicting findings. Scheepers et al. conducted three experiments using a “random word array” target task. Participants read either a PO or DO prime sentence aloud and then had to produce a sentence from words presented in a random order (e.g. “bible”, “evangelist”, “youngster”, “sold”). In all three experiments, Scheepers et al. found a lexical boost with noun repetition (in particular, the subject and the recipient), whereas the abstract structural priming effect was small and only significant when they combined the data from all three experiments. In contrast, Carminati et al. conducted five experiments using different tasks (including the random word array task) and observed abstract structural priming with PO/DO structures but no lexical boost with noun repetition. The difference in results may be partly due to the fact that Carminati et al. repeated a single noun in their experimental trials, whereas Scheepers et al.’s Experiments 2 and 3 contained many trials where several nouns were repeated; repetition of several nouns may have made participants aware that the prime and target structures were related. However, it does not explain the results of Scheepers et al.’s Experiment 1, where they only repeated a single noun. The discrepancy in results between Scheepers et al. and Carminati et al. suggests that the non-head noun boost may come and go depending on quite subtle differences in methodology. This contrasts with the verb head boost, which has been observed in many experiments using a variety of methods. This suggests that the head and non-head boosts may be due to different underlying cognitive mechanisms.

In the current experiments, we therefore investigated whether the head and non-head boosts are differently affected by the method. Based on Scheepers et al.’s (2017) and Carminati et al.’s (2019) research, we expected that the head boost effect is more stable across methods than the non-head boost. In order to investigate this, we compared a method in which the word repetition was very explicit with one where it was less so. We made the word repetition highly explicit by giving participants booklets with prime-target pairs that allowed them to see the prime fragment and its completion while they completed the target; thus, they did not need to rely on their memory to realise that a word was repeated between prime and target. We contrasted it with a method in which participants could not see the prime and target simultaneously, so the word repetition was less explicit; participants wrote completions on a computer screen and the completion disappeared after they had completed it, so the prime was no longer visible when they completed the target. Both methods have been used previously (e.g. Corley & Scheepers, 2002; Desmet & Declercq, 2006; Kantola & Van Gompel, 2011; Pickering et al., 2002; Pickering & Branigan, 1998; Scheepers, 2003; Van Gompel et al., 2012), but the current study is the first to directly contrast them to investigate task-specific effects on the lexical boost.

Models of structural priming and the lexical boost such as the residual activation model by Pickering and Branigan (1998) and the ACT-R model by Reitter et al. (2011) currently do not make predictions for how the visibility of the prime sentence affects the lexical boost. This is also true for Chang et al.’s (2006) explicit memory account of the lexical boost. However, if the boost is due to the explicit memory for the words in the prime, that raises the possibility that the boost is larger when participants can look back at the prime, because the simultaneous visibility of the repeated word in the prime and target may act as an overt, explicit cue for the prime structure. In contrast, when participants cannot look back, the repeated word may act as a cue only if it is still represented in explicit memory. If the repeated prime word is no longer in explicit memory, no boost may occur. This account predicts that we should observe a stronger lexical boost if participants can look back than if they cannot, both when the noun and when the verb is repeated. However, given that the head boost appears more robust across studies than the non-head boost, an alternative account is that different cognitive processes may underlie the verb and noun boost effects. That is, the non-head boost may be due to explicit memory for the prime, whereas the head boost may be unaffected by explicit memory processes (and rely on more automatically engaged implicit memory processes), and as a result, the non-head boost may depend on how explicit the word repetition is, whereas the head boost may not.

To test this, we conducted four structural priming experiments on PO/DO ditransitive structures: In Experiments 1 and 2, we manipulated the repetition of the subject noun (a non-head), whereas in Experiments 3 and 4, we manipulated the repetition of the verb (the head). In Experiments 1 and 3, participants could see the prime when they completed the target, but in Experiments 2 and 4, they could not.

Experiment 1

Experiment 1 tested whether a lexical boost occurred when the subject noun in PO/DO ditransitives was repeated between prime and target. We made the repetition highly explicit by using a written completion task in which participants could see the prime while they completed the target.

Method

Participants

Fifty-six native speakers of English took part. All were undergraduate students at the University of Dundee who received course credits for their participation. None of the participants reported any previous reading difficulties. All experiments in this article were approved by the University of Dundee Ethics Committee.

Materials

We constructed 40 item sets such as (3-4). Participants had to write a completion to a prime fragment that either encouraged a PO (3a, c) or a DO completion (3b, d) and then had to write a completion to a target fragment (4) that allowed either a PO or DO. The PO eliciting prime fragments consisted of a definite subject noun phrase, a ditransitive verb and a theme noun phrase modified by an adjective. The DO primes were the same except that the noun phrase after the verb was a recipient noun phrase modified by an adjective. The target fragments consisted of an indefinite subject noun and a ditransitive verb. The subject noun in the prime was either the same as in the target (3a, b) or different (3c, d). The verb was always different in the prime and target. We used an indefinite noun phrase in the target to indicate that the subject in the prime and target were not coreferent. In addition to the conditions in (3-4), we had four subject noun counterbalancing conditions where the nouns (e.g. “farmer” and “seller”) were swapped in both the primes and targets.

3a. The farmer gave the new potatoes . . . (PO prime, noun repeated)

3b. The farmer gave the potential buyer . . . (DO prime, noun repeated)

3c. The seller gave the new potatoes . . . (PO prime, noun not repeated)

3d. The seller gave the potential buyer . . . (DO prime, noun not repeated)

4. A farmer showed . . . (target)

The experiment also included 115 filler sentence fragments of a variety of structures that did not specifically encourage a PO or a DO completion. These fragments elicited complete clauses (e.g. “The answering machine was broken because … ”), verb phrases (e.g. “The shopping centre … ”) or a subject plus verb phrase (e.g. “After last night’s party … ”).

Procedure

Participants provided hand-written completions to sentence fragments in a booklet. They were asked to complete the sentences so that they would be meaningful and grammatical. The target fragments appeared immediately underneath the primes so that participants could see both at the same time. The experiment took 45–75 minutes to complete. Participants gave informed consent to take part in the experiment.

Design

The experiment contained 40 experimental items in eight experimental conditions defined by the variables prime structure (PO vs. DO), subject noun repetition (repeated vs. not repeated) and subject noun counterbalancing (e.g. target noun was “farmer” vs. “seller”). The design was a 2×2×2 within-participants and within-items factorial design with complete Latin square counterbalancing of all three variables. The experimental items and 155 fillers occurred in a random order that was the same across lists, with at least two fillers between experimental items. Seven participants were randomly assigned to each list.

Results

We scored whether participants produced a PO, DO, or other completion to the prime and target fragments. A completion was scored as a PO if the verb was followed by a theme noun phrase and a recipient prepositional object (in that order). It was scored as a DO if the verb was followed by a recipient indirect object noun phrase and a theme direct object noun phrase (again in that order). Responses where participants added a particle (e.g. as in “showed off”) were considered a PO or DO if they had the appropriate structure. Completions were only scored as a PO or DO if they were reversible into the alternative structure.

First, we excluded target trials on which the prime completion was not the intended PO or DO structure (i.e. we excluded other completions, but also cases where a PO eliciting prime was completed as a DO, or vice versa). This resulted in 17.2% exclusions. Next, to examine structural priming, we analysed the proportions of PO target completions out of all PO or DO completions of the target. We excluded other responses from the targets, so the proportion of DOs equals 1 - p(PO). Table 1 presents the mean proportions of POs in all experiments in this paper. We carried out logit mixed effect analyses using the glmer function from the lme Package in R (version 3.5.1; CRAN project; The R Core Team, 2018). The fixed variables prime structure (PO vs. DO prime) and subject noun repetition (repeated vs. not repeated) were centred. We included by-participants and by-items random intercepts and random slopes for all fixed variables (Barr et al., 2013), by enforcing zero correlations between random effects in order to avoid overparameterization or false convergence (Matuschek et al., 2017).

Table 1. Mean proportions of target completions by Experiment and condition.

	Mean proportions of PO target completions out of all PO and DO completions			Mean proportions of other target completions
	PO prime	DO prime	Priming effect	PO prime	DO prime
Experiment 1 (prime visible)
Subject noun repeated	59.9 (3.1)	45.7 (3.1)	14.2	43.4 (2.4)	45.9 (2.3)
Subject noun not repeated	54.8 (3.3)	53.1 (3.1)	1.7	48.2 (2.4)	48.1 (2.3)
Experiment 2 (prime not visible)
Subject noun repeated	61.6 (2.8)	51.0 (2.9)	10.6	38.6 (2.2)	41.2 (2.2)
Subject noun not repeated	64.3 (2.9)	56.5 (3.0)	7.8	44.1 (2.3)	44.6 (2.2)
Experiment 3 (prime visible)
Verb repeated	77.4 (2.5)	51.3 (3.0)	26.1	34.0 (2.3)	34.7 (2.3)
Verb not repeated	67.1 (3.2)	55.1 (3.3)	12.0	48.6 (2.4)	45.2 (2.4)
Experiment 4 (prime not visible)
Verb repeated	76.0 (2.5)	51.6 (2.8)	24.4	38.8 (2.2)	34.7 (2.2)
Verb not repeated	68.9 (2.9)	61.4 (3.0)	7.5	45.3 (2.3)	47.4 (2.2)

Notes. Standard errors of the means in parentheses. Conditions where neither the subject nor the verb was repeated were identical between experiments. The means were calculated including only trials where participants produced either a PO or DO structure in the prime.

There was a main effect of prime structure (β = 0.208, z = 2.58, p = .010), indicating that participants produced more PO target completions after PO than DO primes. There was no main effect of subject noun repetition (β = 0.080, z = 0.89, p = .376), but critically, there was an interaction between prime structure and noun repetition: β = −0.200, z = −2.50, p = .013, indicating that structural priming was stronger when the subject was repeated. Follow-up analyses showed that priming was significant when the subject noun was the same in prime and target (β = 0.387, z = 3.40, p < .001), but not when it was different (β = 0.030, z = 0.26, p = .800).

For completeness, we also analysed the proportions of other target completions out of all trials where participants produced a valid (i.e. PO or DO) prime completion. The means are shown in Table 1. Most of these other completions were monotransitive structure completions. Logit mixed effect modelling, conducted in the same way as before, showed no effect of prime structure, verb repetition or interaction between the two variables (ps > .19).

Discussion

The experiment showed an interaction between subject noun repetition and prime structure, but no structural priming in the absence of subject repetition. Both results contrast with previous findings. The subject noun repetition effect on priming in the current experiment contrasts with the absence of such an effect in Carminati et al. (2019). In four of their experiments, they used a (spoken) sentence completion task, but critically, participants could not look back at the prime when they completed the target. Our current results raise the possibility that when participants could see the prime and target simultaneously and the subject noun was repeated, the highly explicit word repetition may have functioned as a cue that reactivated the prime structure, resulting in priming in the repeated noun conditions. We directly tested this possibility in Experiment 2.

The absence of a structural priming effect in the absence of noun repetition contrasts with previous studies that have found evidence for abstract structural priming with PO/DO ditransitives in English (e.g. Bock, 1986, 1989; Corley & Scheepers, 2002; Pickering & Branigan, 1998). One possible explanation is that because participants could see the prime and target simultaneously in our experiment, they may have realised that the first content word in the sentence was often repeated, and they may have started checking for this word repetition. In trials where the first word was repeated, this strategy would have highlighted the noun repetition (resulting in priming), but in trials where it was not repeated, the additional processes involved in checking for word repetition may have interfered with the structural priming, resulting in no abstract structural priming. This predicts that we should find evidence for abstract structural priming when participants cannot look back at the prime to check for word repetition, a prediction that we also tested in Experiment 2.

Experiment 2

The results from Experiment 1 contrast with those from Carminati et al.’s (2019) study, in which participants could not look back to the prime and no interaction between noun repetition and prime structure was detected. One possibility is that the interaction in Experiment 1 was due to the fact that participants could look back at the prime. However, Scheepers et al. (2017) found a noun boost effect with a method in which participants could not look back (using the random word array target task). In order to establish whether the lexical boost in Experiment 1 was indeed due to looking back or some other, unknown reason, we conducted Experiment 2, in which the prime disappeared after participants completed it, so they could no longer see it when they produced the target. In addition, this also allowed us to test whether the absence of abstract structural priming in Experiment 1 was due to the method.

Method

Participants

Forty-eight participants from the same population as in Experiment 1 took part.

Materials

We used the same experimental item sets as in Experiment 1, but added 8 further sets in order to increase the statistical power to observe a lexical boost. We also added 23 filler fragments.

Procedure

We used the Gorilla experimental software (Anwyl-Irvine et al., 2020) to present the sentence fragments on a computer screen. Participants wrote completions using a keyboard. Each sentence disappeared after it had been completed so that participants could not see the prime and target at the same time. As in Experiment 1, participants were asked to complete the sentences so that they would be grammatical and meaningful. The experiment lasted between 45 and 75 minutes.

Design

The design was the same as in Experiment 1, except that we had 48 experimental items and six participants per list.

Results

The scoring was conducted in the same way as in Experiment 1. In 14.7% of trials, participants did not complete the prime with the intended PO or DO structure; responses to the subsequent targets were excluded from the analyses. We conducted logit mixed effect modelling in the same way as in Experiment 1. The mean proportions of PO target completions out of all PO and DO completions are presented in Table 1. We observed a main effect of prime structure (β = 0.356, z = 4.47, p < .001), indicating that participants produced more PO target completions after PO than DO primes. There was no main effect of subject noun repetition structure (β = 0.075, z = 0.97, p = .331), and critically, no interaction between prime structure and subject noun repetition (β = −0.011, z = −0.139, p = .890), indicating that there was no subject noun boost. Follow-up analyses showed that there was a prime structure effect both when the subject noun was repeated (β = 0.325, z = 3.01, p = .003) and when it was not repeated (β = 0.360, z = 3.07, p = .002).¹

As in Experiment 1, we also analysed the proportions of other completions out of all targets where participants produced a PO or DO prime completion, using the same analysis method as before. The means are shown in Table 1. There was a main effect of subject noun repetition (β = −0.120, z = −2.20, p = .028), showing that participants produced more other responses when the subject noun was different in prime and target than when it was the same. There was no effect of prime structure nor an interaction between prime structure and verb repetition on the proportions of other target completions (ps > .700).

Discussion

The results from Experiment 2 differ from Experiment 1. In contrast to Experiment 1, we observed an abstract structural priming effect but no interaction between prime structure and subject noun repetition. This suggests that the interaction we obtained in Experiment 1 was task-specific. We assume that when the prime is not visible during target completion, subject noun repetition is a much less explicit cue for the structure in the prime sentence, and therefore, it does not affect priming. In addition, because the prime does not remain visible, unlike in Experiment 1 there are no checking-back processes that might interfere with the priming effect when the subject is not repeated. This explains why, in contrast to Experiment 1, Experiment 2 did show abstract structural priming.

Experiment 3

Experiments 1 and 2 showed that the effect of subject noun repetition is strongly dependent on the task, but the question remains whether the head boost effect is more stable across tasks. Therefore, in Experiments 3 and 4 we investigated whether verb repetition would yield a boost using the same methods as in Experiments 1 and 2. In Experiment 3, we manipulated the verb repetition between prime and target, using the method in which the prime remained visible when participants completed the target.

Method

Participants

Forty participants from the same populations as in the previous experiments took part.

Materials

The materials were the same as in Experiment 1, except that we manipulated the repetition of the verb instead of the subject noun and instead of counterbalancing the target subject noun, we counterbalanced the target verb (e.g. “showed” and “gave”).

Procedure

The task was the same as in Experiment 1. Participants wrote completions to sentence fragments in a booklet.

Design

The design was the same as in Experiment 1, except that we had five participants per list.

Results

As before, we excluded all target trials on which the prime was not completed with the intended PO or DO structure, resulting in 12.2% exclusions. Table 1 shows the mean proportions of POs out of all PO or DO target completions. The data were analysed in the same way as in the previous experiments. There was a main effect of prime structure (β = 0.526, z = 6.25, p < .001), indicating that participants produced more PO target completions after PO than DO primes. There was no main effect of verb repetition (β = −0.067, z = −0.79, p = .427), but critically, there was an interaction between prime structure and verb repetition: β = −0.225, z = −2.85, p = .004, indicating a verb boost effect. Follow-up analyses showed that priming was significant when the verb was the same in prime and target (β = 0.712, z = 6.48, p < .001; no random slope due to lack of convergence), but also when it was different (β = 0.326, z = 2.73, p = .006).

We also analysed the proportions of other target completions, using the same analysis method as before. Table 1 shows the means. There was a main effect of verb repetition (β = −0.323, z = −4.86, p < .001), showing that participants produced more other completions when the verb was not repeated than when it was. There was no effect of prime structure nor an interaction between prime structure and verb repetition on the proportions of other completions (ps > .300).

Discussion

Experiment 3 showed a lexical boost with verb repetition and also an abstract priming effect in the absence of verb repetition. Both findings are consistent with previous studies (e.g. Branigan et al., 2000; Corley & Scheepers, 2002; Hartsuiker et al., 2008; Pickering & Branigan, 1998; Schoonbaert et al., 2007). However, the abstract priming effect contrasts with the absence of this effect in Experiment 1, in which the same method and materials were used in the no-word repetition conditions. This suggests that the difference between these two experiments comes from different strategies adopted by the participants. Because the subject noun is the first content word in the sentence, participants may be more aware of the repetition of it and may have started checking whether the subject noun was repeated before completing the target in Experiment 1. This may have caused interference with abstract priming. Repetition of the verb, which comes later in the sentence, may not have led to such a strategy in Experiment 3.

Experiment 4

Experiment 4 tested whether the effect of verb repetition on structural priming is similarly sensitive to the method as the subject noun repetition effect, or whether, as we hypothesised in the introduction, it is stable across methodologies. Experiment 4 was the same as Experiment 2, except that we now manipulated the repetition of the verb rather than the subject noun. Participants could not see the prime when they completed the target.

Method

Participants

Forty-eight participants from the same population as before took part.

Materials

The materials were the same as in Experiment 2, except that we manipulated the repetition of the verb.

Procedure

The procedure was the same as in Experiment 2. Participants typed completions on a computer screen and they could not see the prime when they completed the target.

Design

The design was the same as in Experiment 2, except that we had six participants per list.

Results

Across conditions, we excluded 15.1% of data because participants did not complete the prime with the intended PO or DO structure. See Table 1 for the mean proportions of POs out of all PO and DO target completions. The analyses showed a main effect of prime structure (β = 0.495, z = 5.35, p < .001), indicating that participants produced more PO target completions after PO than DO primes. There was no main effect of verb repetition (β = 0.028, z = 0.36, p = .720), but there was an interaction between prime structure and verb repetition: β = −0.296, z = −3.81, p < .001. Follow-up analyses revealed that priming was significant when the verb was the same in prime and target (β = 0.728, z = 6.58, p < .001), but the 7.5% difference between the prime conditions was not significant when the verb was different (β = 0.170, z = 1.42, p = .157).

As in Experiment 3, analysis of the proportions of other target completions (Table 1) showed a main effect of verb repetition (β = −0.247, z = −3.93, p < .001). There was no main effect of prime structure nor interaction between prime structure and verb repetition (ps > .095).

Discussion

Experiment 4 showed that structural priming was stronger when the verb was repeated than when it was not, even though participants could not see the prime when they completed the target. This interaction effect (16.9%) was comparable to Experiment 3 (14.1%). Thus, the head verb boost appears to be unaffected by the change in task. When the verb was not repeated, the priming effect was not significant. However, the 7.5% difference between the prime conditions was very similar to the 7.8% abstract priming effect in Experiment 2 (using the same method), which was significant.

General discussion

We investigated the effects of different methods on structural priming with PO/DO structures when either the verb or the subject noun was repeated. The main finding was that the effect of subject noun repetition on structural priming was dependent on the priming methodology we used, whereas the head verb boost was not. Structural priming only interacted with subject noun repetition if participants could see the prime when they completed the target, whereas an interaction with head verb repetition occurred regardless of whether the prime was present or not during target completion. This suggests that the effects of head and non-head repetition are due to different processes.

Our results suggest that the prime structure is automatically associated with the head verb, regardless of the participants’ task, but this is not the case with the subject noun. Instead, the subject noun repetition effect only occurs if the repetition is highly explicit; only when participants were allowed to see the prime when they completed the target, did repetition of the subject affect structural priming. When participants did not see the prime and they only had their memory of the prime sentence (either explicit or implicit memory) to rely on, the repetition of the subject had no effect on priming.

The experiments also showed that abstract structural priming (that is, in the absence of word repetition) can be affected by experiment-specific strategies. In particular, Experiments 1 and 3 used the same task and materials in the non-repeated conditions, but Experiment 3 showed abstract priming, whereas Experiment 1 did not. Presumably, this difference is due to the particular words that were repeated in the other conditions in the experiments. In Experiment 1, the subject noun was repeated in the other conditions, whereas in Experiment 3, the verb was. Because the subject noun was the first content word in the sentence, it is plausible that its repetition was more obvious than verb repetition, and that as a result, participants started checking for subject noun repetition. This checking for word repetition between prime and target may have resulted in interference to the abstract priming. In fact, the effect of noun repetition on priming in Experiment 1 may primarily have been due to the absence of priming in the no-repetition conditions, rather than due to enhanced priming in the repeated subject conditions. In Experiment 1, the priming effect in the repeated subject conditions was 14.2%, which was only slightly larger than the priming in the no repetition conditions in Experiment 3 (12.0%). This is consistent with our explanation that checking for word repetition in Experiment 1 resulted in interference to priming and therefore, no priming occurred when the subject noun was not repeated. In addition, priming when the noun was repeated was fairly weak because, although the subject noun repetition functioned as a cue for the structure in the prime sentence, checking for repetition may have caused interference that reduced the structural priming effect.

Our finding that the noun boost was affected by the explicitness of the prime sentence, but the verb boost was not, does not fit well with models that assume that these effects are due to the same underlying memory mechanisms. The finding that the noun boost depended on whether participants could look back at the prime may be compatible with Chang et al.’s (2006) account, because the simultaneous visibility of the prime and target makes the repetition more explicit and participants do not need to only rely on their explicit memory to reactivate the prime structure. However, because the noun and verb boosts should be due to the same explicit memory mechanisms, it does not explain why the verb boost occurred regardless of whether the prime and target were simultaneously visible. Similarly, Reitter et al.’s (2011) ACT-R model does not explain why the lexical boost with the subject noun and the verb is different, because in both cases, the additional priming with word repetition should be due to the same mechanism (associative learning of word-structure combinations). Instead, we argue that this difference suggests that participants rely on different memory representations. Structural priming when the noun is repeated may indeed rely on the presence of an explicit cue, but the verb boost does not, suggesting that the latter is due to more automatically engaged memory processes.

The residual activation model (Pickering & Branigan, 1998) does not make specific predictions for how the simultaneous presence of the prime and target affects the lexical boost, but it can more easily deal with the distinction between the subject noun and the verb. Because the verb is the syntactic head of the PO/DO structure, this structure is automatically associated with the verb during the processing of the prime. As a result, a verb repetition boost should occur and it should occur regardless of the method. In contrast, the prime structure is not normally associated with the subject noun, because it is not its head. As we have proposed above, the effect of subject repetition on priming may be due to additional processes that participants adopt when its repetition is highly explicit due to the simultaneous visibility of the prime and target. When the subject noun is repeated, these processes result in reactivation of the prime structure, whereas they interfere with its activation when there is no subject repetition.

The explanation that we propose for our data is a dual-process account, which assumes that the head-verb boost and abstract structural priming are due to automatic processes (either residual activation or implicit learning), whereas the subject noun boost is the result of more strategically engaged processes that participants use when they can look back at the prime sentence. It is worth noting though that our dual-process explanation is different from Chang et al.’s dual-process account, which postulates a distinction between abstract structural priming, which is due to implicit learning of the prime structure, and the lexical boost, which occurs because a repeated content word acts as an explicit memory cue for the prime structure. In contrast to our account, their account assumes that the subject noun and head verb boosts rely on the same memory processes.

Although our current findings do not support Chang et al.’s dual-process account, one advantage of this account is that it straightforwardly explains why abstract structural priming and the verb boost differ in their time-course, with abstract priming persisting across several intervening sentences between prime and target, while the verb boost decays more rapidly (Branigan & McLean, 2016; Hartsuiker et al., 2008): Chang et al. argued that the implicit learning of abstract structures is long lasting, whereas the lexical boost is due to short-term explicit memory. However, using computational modelling, Malhotra et al. (2008) showed that the difference in time course can also be accounted for in a residual activation model, by assuming that the syntactic nodes are mutually inhibitory, whereas the bindings between syntactic and lexical nodes are excitatory. Carminati et al. (2019) therefore argued that the time-course findings do not constitute evidence against a residual activation model of abstract structural priming and the head-verb boost.

Our findings may also shed light on the results of Scheepers et al. (2017). In their experiments, they observed structural priming when a noun was repeated between prime and target, but structural priming in the absence of content word repetition was very small (only significant when they combined the data from all their experiments), similar to what we found in Experiment 1. It is possible that because nouns were frequently repeated in their experiments and all experiments had trials on which the first noun was repeated, participants used the repetition as a prompt when constructing a sentence from the random word array. The use of the first noun as a prompt may have been similar to that in our Experiment 1, even though participants could not simultaneously see the prime and target in their experiments. As in Experiment 1, checking for word repetition may have led to interference to the priming when no word was repeated (explaining why there was only a very small abstract priming effect), whereas the presence of a word repetition cue may have led to reactivation of the prime structure (explaining why priming occurred with noun repetition).

Our findings also have important methodological implications. The difference in findings between the two tasks in our experiments suggests that both the lexical boost and abstract priming can be affected by the method. When participants can see the prime while completing the target, this can result in a lexical boost with non-head repetition that is not otherwise observed. Several previous experiments have used sentence completion tasks in which participants could see the prime while they completed the target. Because these experiments either did not repeat any words (e.g. Desmet & Declercq, 2006; Kantola & Van Gompel, 2011; Pickering et al., 2002; Scheepers, 2003) or repeated only the verb (Branigan et al., 1999; Pickering & Branigan, 1998), it is unlikely that the results obtained in these studies were task-specific. However, given the potentially strategic nature of this task, it may be better avoided in future studies. More generally, researchers using the structural priming method need to be aware of possible task effects. Structural priming research has been conducted using a wide range of tasks, including picture description under the guise of a memory task (e.g. Bock, 1986, 1989) or without a memory task (e.g. Vernice et al., 2012), target sentence recall (e.g. Potter & Lombardi, 1998), sentence completion (e.g. Pickering & Branigan, 1998), production of the target sentence from words presented in a random order (e.g. Scheepers et al., 2017) and dialogue with a confederate listener (e.g. Branigan et al., 2000), but it remains largely unclear how the different methods might affect structural priming. Future research will need to establish to what extent these methods tap into the same structural representations.

Acknowledgement

We would like to thank Grant Hamilton and Grene Jagna-An for assistance with the experiments. The experimental data can be downloaded via the Open Science Framework at http://dx.doi.org/10.17605/OSF.IO/2MK4V

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research was supported by the Economic and Social Research Council [grant no ES/P001866/1], awarded to Roger van Gompel and Leila Kantola

Notes

1 Due to an error, the eight items we added to the items from Experiment 1 had a definite subject noun phrase in the target. We therefore also conducted an analysis without these items. This analysis showed the same effects as the analysis with these items.