Interactive structure building in sentence production

A


Introduction
A central issue in language production research concerns how speakers sequence words and phrases, and the nature of representations that drive such processes.To explore this issue, researchers have examined structural priming, that is, the speaker's tendency to repeat an earlier encountered structure.For example, when speakers of English describe a so-called dative or ditransitive event of a girl handing a paint brush to a man, they show an increased tendency to use a prepositional object structure as in The girl handed a paint brush to the man, rather than a double object structure as in The girl handed the man a paint brush, after having encountered a prepositional object structure (e.g., Bock, 1986;1989).Structural priming has been found with many other constructions including the active-passive alternation for transitive events (e.g., Bock, 1986) and the use of relative clause structures in noun phrase production (e.g., Cleland & Pickering, 2003;Fukumura & Zhang, 2023).It occurs within and across different languages and in both spoken and written modalities (see Branigan & Pickering, 2017;V. Ferreira & Bock, 2006;Mahowald et al., 2016; for a review and meta-analyses).However, the locus of the representations that drive structural priming and how speakers form such representations remain controversial.The goal of the current study is to address this.
Earlier research on structural priming was guided by the two-stage grammatical encoding account (Bock & Levelt, 1994;Garrett, 1975Garrett, , 1980)), which proposes that grammatical encoding comprises two distinct stages.The first stage involves function assignment, where syntactic functions (e.g., subject, object) are assigned to event participants based on their thematic roles, with the argument structure of the verb playing a key role (e.g., Grimshaw, 1990).For example, the verb hand assigns the agent, theme, and recipient roles to its three arguments, and the selection of an argument structure determines which thematic roles are mapped onto which syntactic function.If speakers choose the argument structure that assigns the direct object to the theme and the oblique object to the recipient, this will result in a functional structure of a prepositional dative structure, such as The girl handed a paint brush to the man.Functional structures are assumed not to specify the linear order of constituents.Therefore, the linear order must be determined in a subsequent stage of constituent assembly.This stage is assumed to be guided by syntactic information, such as syntactic functions (e.g., the subject, the direct object) and categories (e.g., noun phrase, verb phrase, prepositional phrase), without involving thematic role information relevant to function assignment.According to the two-stage account, structural priming occurs during constituent assembly, driven by syntactic information abstracted from the thematic role information involved in function assignment.
Consistent with this two-stage account, earlier evidence appeared to suggest the isolation of structural priming from thematic role information.For example, Bock (1989) reported that the rates of prepositional objects involving a recipient role (relative to double object alternatives) were unaffected by whether the prime involved a recipient role (e.g., The student brought a cake to the teacher) or a beneficiary role (The student brought a cake for the teacher).Similarly, Bock and Loebell (1990) found that prepositional object datives could be elicited by both prepositional locative primes (e.g., The wealthy widow drove her Mercedes to the church) and prepositional dative primes (e. g., The wealthy widow gave the Mercedes to the church), despite the prepositional objects denoting different thematic roles (location vs. recipient).In the same study, it was also found that intransitive primes, such as The airplane is landing by the control power, can elicit passive structures as effectively as passive primes, such as The airplane was alerted by the control tower.Subjects of passive and intransitive sentences represent distinct thematic roles; patients and agents, respectively.Hence, the fact that both sentences equally primed passive structures indicates that structural priming is not driven by thematic roles relevant to function assignment, aligning with the two-stage account.
However, different syntactic structures typically involve distinct function assignments.For example, a prepositional object dative differs from a double object dative not only in abstract phrasal structure (NP VP[V NP PP] vs. NP VP[V NP NP])1 , but also in functional structure.If structural priming can only arise during constituent assembly, function assignment in the target proceeds irrespective of the functional structure of the prime.Therefore, in response to a prepositional object dative prime, speakers may adopt on a functional structure compatible with a double object dative, rather than one compatible with a primed prepositional dative, when planning a target.Thus, for structural priming to occur, the incompatible functional structure needs to be backtracked during constituent assembly.However, backtracking is incompatible with the feed-forward and incremental production processes assumed by the twostage account; that is, each level of processing is guided by the representation processed directly above it, and the order in which the information is passed on to the next stage determines word order (Bock, 1982;Bock & Levelt, 1994;Kempen & Hoenkamp, 1987;Levelt, 1989).Other research suggests that speakers tend not to backtrack on a choice made at an earlier stage (V.Ferreira et al., 2005;Fukumura et al., 2022).
An alternative possibility is that structural priming occurs earlier in the production process, involving thematic role information.Benefactive for-datives and transfer to-datives may prime each other (Bock, 1989), because for-datives can imply the transfer of possession of a theme to a beneficiary, and they can alternate with double object structures only in such cases.For example, I built his bed for him can alternate with I built him his bed, but I drove his car for him cannot alternate with *I drove him his car (Pinker, 1989).Hence, the absence of an interaction may be attributed to the fact that beneficiaries represent similar thematic roles as recipients in the dative alternation.Moreover, contrary to Bock and Loebell (1990), Potter and Lombardi (1998) showed, using a sentence recall task, that prepositional object dative primes such as The tycoon willed that mansion to his young nephew very grudgingly elicit prepositional object datives such as The prompt secretary wrote a message to her boss every week significantly more than prepositional locative primes such as Lenore drove her new convertible to the beach early this afternoon did.In Potter and Lombardi's study, different conditions involved different lexical items.Using better controlled materials, Ziegler and Snedeker (2018) replicated the findings of Potter and Lombardi, further demonstrating that the degree of thematic role overlap between prime and target modulates structural priming.
Indeed, a growing body of evidence is consistent with such a possibility (e.g., Cai et al., 2012;Chang et al., 2003;Cho-Reyes et al., 2016;Fleischer et al., 2012;Hare & Goldberg, 1999;Heydel & Murray, 2000;Köhne et al., 2014;Pappert & Pechmann, 2014;Salamoura & Williams, 2007;Ziegler & Snedeker, 2018;Ziegler et al., 2019).Structural priming has been found between sentences sharing the same constituent structure (Chang et al., 2003;Ziegler & Snedeker, 2018).Sentences with so-called "spray-load" verbs (Levin, 1993), such as The farmer heaped the wagon with straw and The farmer heaped straw onto the wagon, share the abstract syntactic structure (NP VP[V NP PP]), but differ in argument structure underpinning thematic role-function mappings.In the former (heaped the wagon with straw), the location precedes the theme, with the noun phrase (NP) denoting the location (wagon), which is assigned the direct object, and the prepositional phrase (PP) denoting the theme (straw), which is assigned the prepositional object.The thematic role order is reversed in the latter (heaped straw onto the wagon), where the theme is assigned the direct object, and the location is assigned the prepositional object.In a study using a sentence recall task, Chang et al. (2003) showed that participants recalled target sentences using the location-theme order more frequently after having recalled a prime sentence using a location-theme order as opposed to a theme-location order (see also Ziegler & Snedeker, 2018, who replicated the findings using a picture description task).Chang et al. (2006) argued that when abstract constituent structure fails to discriminate structural variants, semantic representations can drive structural alternations.In their Dual-Path model, the meaning system processes the message based on concepts and event roles, while the sequencing system generates sentence structures based on abstract syntactic categories.Although the sequencing system normally operates independently of the meaning system, structural variants can be learned based on thematic role-syntactic function mappings when abstract constituent structure fails to discriminate, as in the case of the theme-location alternation.Indeed, some evidence for thematic role order priming comes from structural variants having the same constituent structure, like the theme-location alternation in English.For example, Köhne et al. (2014) found more theme-recipient orders following theme-recipient order German double object primes (Der Mann verspricht die Putzhilfe der Ehefrau, The man promises [to get] the cleaning woman the wife) than following recipient-theme order German double object primes (Der Mann verspricht der Ehefrau die Putzhilfe, The man promises the wife [to get] the cleaning woman) (see also Pappert & Pechmann, 2014, for related findings).However, findings suggest that thematic role orders can persist across different constituent structures (Bernolet et al., 2009;Cai et al., 2012;Cho-Reyes et al., 2016;Fleischer et al., 2012;Hare & Goldberg, 1999;Salamoura & Williams, 2007), where abstract constituent structure discriminates structural alternatives.
For instance, Hare and Goldberg (1999) found that provide-with sentences in English, such as The officers provided the soldiers with guns prime double-object datives such as The man handed a woman a box of sweets rather than prepositional datives such as The man handed a box of sweets to a woman.Provide-with sentences have the same thematic role order as double-object datives (Agent-Recipient-Theme), whereas they have the same phrasal structure as prepositional datives (NP VP[V NP PP]).Hence, this finding suggests that priming can persist via a different route other than the persistence of function assignment and constituent assembly.Consistent with this possibility, Cai et al. (2012) found that thematic role-linear position mappings persist in dative alternations in Mandarin Chinese, leading to more prepositional object (PO) targets following topicalized double object (DO) primes that had the same theme-recipient order (Naben shu niuzai song-gei le shuishou, The book, the cowboy gave the sailor) than following standard DO primes that had a recipient-theme order (Niuzai song-gei le shuishou naben shu, The cowboy gave the sailor the book).However, topicalized DO primes elicited more DO responses than PO primes did, suggesting that thematic role-syntactic function mappings can also be primed.These findings led Cai et al. to argue that speakers construct a constituent structure by simultaneously tracking thematic role-position mappings and thematic role-syntactic function mappings.However, the critical questions pertinent to the present study remain, concerning what type of representations support such parallel mappings and whether the mappings between thematic roles and positions, as well as between thematic roles and syntactic functions, exert only additive effects on structural choice or if they interact with each other.Bernolet et al. (2009) proposed that both thematic role-position mappings and thematic role-function mappings are influenced by thematic emphasis.In their cross-linguistic priming study, they found that agent-initial Dutch passives (e.g., Door de non wordt de bokser achtervolgd, By the nun is the boxer chased) elicited more agent-final English passives (e.g., The skier is chased by the pirate) than agent-initial Dutch actives (e.g., De non achtervolgt de bokser, The nun chases the boxer) did, demonstrating priming at the functional level.However, agent-initial Dutch passives elicited fewer agent-final English passives than agent-medial Dutch passives (e.g., De bokser wordt door de non achtervolgd, The boxer is by the nun chased) did.Bernolet et al. attributed these findings to the persistence of thematic emphasis; in their rating studies, the agent role in agent-initial Dutch active sentences was perceived as more emphasized than the agent role in agent-initial Dutch passive sentences, which, in turn, was perceived as more emphasized than the agent role in agent-medial and agent-final Dutch passive sentences.Based on these findings, Bernolet et al. argued that structural priming with the theme-location alternation (Chang et al., 2006) is driven by the persistence of thematic emphasis on the theme or location role, rather than by the persistence of thematic role order per se.
Similarly, Vernice et al. (2012) used cleft constructions in Dutch to show that emphasizing the patient role in a sentence increases the likelihood of passive responses in the declarative target sentences.However, in Vernice et al.'s study, thematic emphasis co-varied with thematic role ordering.In experiments using it-clefts (e.g., Het is de cowboy die hij slaat, It is the cowboy who he is hitting) or inverted wh-clefts (e.g., De cowboy is degene die hij slaat, The cowboy is the one who he is hitting), patient-focused primes always involved patient-agent orders, which corresponded to the thematic role order of passive target sentences (De jongen wordt geraakt door de bal, The boy is hit by the ball).One experiment aimed to address this issue, using patient-emphasised wh-cleft (e.g., Degene die hij slaat is de cowboy, The one who he is hitting is the cowboy); however, in this construction, while the identity of the patient (cowboy) is revealed at the end of the sentence, the thematic role order arguably is still patient-agent.Hence, it is not entirely clear whether the priming effect was driven by thematic emphasis or the persistence of thematic role order or both.In Cai et al.'s (2012) study, perceived thematic emphasis was found to be unrelated to the persistence of thematic role-position mappings.

Current study
The current study thus examined the representations underlying structural priming.Specifically, we investigated the locus of the lexical boost effect (Branigan et al., 2000;Carminati et al., 2019;Chang et al., 2015;Fukumura & Zhang, 2023;Hartsuiker et al., 2008;Pickering & Branigan, 1998;Segaert et al., 2013;Scheepers et al., 2017;Van Gompel et al., 2023).While structural priming can occur without any lexical repetition, the magnitude of structural priming typically increases when a lexical item is repeated from the prime to the target.This phenomenon is referred to as lexical repetition 'boosting' structural priming.Pickering and Branigan (1998) first demonstrated this effect in a written sentence completion task.They found that participants were more likely to adopt a prepositional object structure in their completions after having completed a prime that induced a prepositional object structure (e.g., The racing driver showed the torn overall…) compared to a prime that induced a double object structure (e.g., The racing driver showed the helpful mechanic…).Importantly, the priming effect was larger when the verb in the prime was repeated in the target (e.g., show-show) compared to when it was not (e.g., give-show).
K. Fukumura and F. Yang According to Pickering and Branigan (1998), verb repetition boosts the persistence of combinatorial nodes, which are thought to underpin structural priming.The diagram in Fig. 1 illustrates their residual activation model, where a prepositional object (PO) prime activates the NP,PP node, while a double object (DO) prime activates the NP,NP node.The residual activation of these nodes increases the likelihood of corresponding PO or DO responses in the target sentence.Importantly, these combinatorial nodes are associated with the lexical properties of the verb's lemmas, as illustrated in Fig. 1, where the two combinatorial nodes are linked to the verbs show and give.Thus, if a PO prime occurs with the verb show, the processing of the PO prime activates not only the NP,PP node, but also its link to the verb show.As a result, when the verb show is repeated from the prime to the target, the activation of the NP,PP node is boosted by the residual activation of that link.This leads to an even higher likelihood of PO selection in the target sentence compared to cases when a different verb give is chosen in the target sentence.
The critical question is what "combinatorial nodes" entail in the residual activation account.The account followed the architecture of the two-stage model of grammatical encoding account, according to which structural priming arises during constituent assembly, not during function assignment.Consistent with this, Pickering et al. (2002) found that shifted PO structures, in which the prepositional phrase preceded the direct object, as in The racing driver showed to the helpful mechanic the torn overall, do not prime canonical POs any more than the baseline (cf.Hartsuiker & Kolk, 1998).The finding led to the proposal that constituent structure is constructed in a single stage, driven by the selection of a combinatorial node that specifies not only the hierarchical relations of the constituents but also their linear order.Shifted PO structures and canonical POs do not prime each other because, despite having the same function assignment, they differ in the linear order of the constituents and hence they are represented by different combinatorial nodes.Although verb repetition was not manipulated, it was suggested that verb selection is linked to the 'selection of a combinatorial node that mandates the construction of a fully specified constituent structure' (Pickering et al., 2002, p.603).
However, as we have just reviewed, structural priming may occur at multiple levels.This raises the question of precisely what representations are enhanced by lexical repetition.Investigating noun phrase production, Fukumura and Zhang (2023) found that lexical repetition enhances not only the persistence of syntactic structure but also the persistence of conceptual structure.When the noun in the prime was repeated in the target, it resulted in stronger syntactic priming, with an increased tendency to use a relative clause structure following a relative clause structure prime like green bow that's spotted than following a prime with a simpler alternative structure like green spotted bow.However, repeating any of the adjectives in the prime did not enhance syntactic priming.In contrast, when the noun or one of the adjectives in the prime was repeated, it enhanced 'conceptual order priming', with an increased tendency to repeat a color-before-pattern order (green bow that's spotted) following a color-before-pattern order prime (blue sock that's striped) than following a pattern-before-color order prime (striped sock that's blue).The fact that conceptual order priming was affected by adjective repetition as well as noun repetition indicated that the syntactic structure and the conceptual order are represented separately.Interestingly, however, the two priming effects appear to interact.Participants tended to repeat the relative clause of the prime more often when they repeated the adjective order from the prime (e.g., both having a color-before-pattern order) than when they adopted a different adjective order (e.g., a pattern-before-color order after a color-before-pattern order prime).Likewise, participants tended to repeat the adjective conceptual order of the prime more often when they repeated the syntactic structure of the prime than when they did not.These results indicated that different types of information interact during the process of structural priming.
The overarching aim of the current study was thus to investigate the representations underlying structural priming, focusing on the locus of the lexical boost effect.To this end, we began by asking whether and how lexical repetition might enhance structural priming in the theme-location alternation (Chang et al., 2003;Ziegler & Snedeker, 2018) such as sentences (1).The residual activation account posits that the verb's lemma is associated with abstract syntactic information involved in generating constituent structures (Pickering et al., 2002;Pickering & Branigan, 1998), in accord with the assumption that structural priming arises during constituent assembly rather than during function assignment (e.g., Bock & Levelt, 1994).We call this the verb-constituent structure account.In this view, we would not expect to observe verb boost effects in the theme-location alternation, as its structural variants (1a vs. 1b) share the abstract constituent structure (NP VP [V NP PP]).However, as we have just reviewed, structural priming could be driven by representations beyond abstract syntax.Hence, the question is which other representations can be enhanced through verb repetition.K. Fukumura and F. Yang underpinning function assignment.That is, sentences (1a) and (1b) involve different argument structures, specifying different function assignments.Specifically, sentence (1a) has a 'with-structure', where the subject is assigned to the agent, the direct object to the location, and the prepositional object to the theme.In contrast, sentence (1b) has a 'locative structure', where the direct object is assigned to the theme and the prepositional object to the location.Hence, the structural alternation in (1) may be primed due to the persistence of the argument structure from the prime sentence, influencing function assignment in the target sentence.The locative structure and the with-structure can be formalized as x < y, P loc z > and x < y, P with z>, respectively (Rappaport & Levin, 1988).In these formalizations, x represents an external argument (i.e., the subject), y represents the direct argument, and z represents the indirect argument.P loc and P with represent a locative preposition and the preposition with respectively.Thus, when the prime sentence (1a) is encountered, it biases the selection of a with-structure during function assignment in the target, increasing the likelihood of assigning the direct object to the location and the prepositional object to the theme.
Critically, the verb-argument structure account predicts that repeating the verb from primes increases the likelihood of argument structure persistence.The assumption is that verb and argument structure selection are associated because the syntactic features of verbs, such as subcategorization frames (e.g., Chomsky, 1965), can constrain possible argument structures (Bock & Levelt, 1994).Such associations may also arise because the meaning of the verb specifies the thematic role of the direct argument, with the preposition assigning the thematic role to the indirect argument (cf.Marantz, 1984;Rappaport & Levin, 1988).Hence, in the diagram in Fig. 2A illustrating the verb-argument structure hypothesis, the verb load is linked to the two argument structures involved in the theme-location alternation such as in (1).The with-structure node ('with') represents the processes underpinning the selection of the argument structure of the with variant (1a), and the locative structure node ('locative') represents the processes underpinning argument structure selection for the locative variant (1b).The bold lines indicate an increased activation of the with-structure, the verb load, and the link between them, as well as the abstract constituent structure NP VP[V NP PP], following a prime sentence such as (1a).If the activation of the link between the verb load and the with-structure node persists, the repetition of the verb should increase the activation of the with-structure in the target sentence.
Thus, the verb-argument structure account predicts a verb boost effect on the theme-location alternation, such as (1).This contrasts with the earlier-mentioned verb-constituent structure account, which is illustrated in Fig. 2B.According to this account, verb boost effects arise through the verb-constituent structure link, where the verb load is connected to the constituent structure node (NP VP[V NP PP]) rather than to the argument structures.On this view, due to the shared constituent structure between the structural variants of the theme-location alternation (1a vs. 1b), verb repetition should not enhance structural persistence.Crucially, neither account takes into consideration that the theme-location alternation, such as (1), involves distinct thematic role orders.In sentence (1a), the location follows the agent and is subsequently followed by the theme.In contrast, in sentence (1b), the theme, following the agent, precedes the location.As a result, a prime featuring an agent-location-theme order, like (1a), might elicit agent-location-theme orders due to the persistence of the thematic role order from the prime.As discussed earlier, there is indeed evidence suggesting that the thematic role K. Fukumura and F. Yang order of the prime can persist in the target (e.g., Hare & Goldberg, 1999;Ziegler & Snedeker, 2018).Furthermore, lexical repetition has been shown to increase not only syntactic persistence (e.g., the repeated use of a relative clause) but also conceptual order persistence (e.g., the repeated use of a color-before-pattern order such as blue striped bow), and these two priming effects can reinforce each other and lead to stronger overall structural persistence (Fukumura & Zhang, 2023).
Hence, as the third account, the interactive structure-building account proposes that speakers activate abstract thematic role orders separately from argument structures, and verb selection is associated not only with argument structure selection but also with thematic role ordering, as these two processes occur interactively in a single stage.Fig. 2C illustrates the processes and representations proposed by this account, depicting the two thematic role orders as well as the argument structures.The A-L-T node corresponds to the order of agent (A)-location (L)-theme (T) in sentences such as (1a), while the A-T-L node corresponds to the order of agent (A)-theme (T)-location (L) in sentences such as (1b).The with-structure node and the locative structure node denote the representations and processes underlying the selection of the with argument structure (1a), where the direct object is assigned to the location and the prepositional object is assigned to the theme, and the locative argument structure (1b), where the direct object is assigned to the theme and the prepositional object to the location, respectively.These argument structure nodes might also include information relating to semantic representations associated with the argument structures (Anderson, 1971;Goldberg, 1995;Pinker, 1989;Rappaport & Levin, 1988;Levin & Rappaport Hovav, 2005).For example, the locative variant, as exemplified by load the boxes into the van, has been proposed to convey change-of-location meanings (x causes [y to come to be at z]), while the with variant, such as load the van with boxes, is associated with change-of-state meanings (x causes [z to come to be in STATE] by means of x causes [y to come to be at z]) (Rappaport & Levin, 1988;cf. Gropen et al., 1991;Pinker, 1989;Jackendoff, 1983).
In Fig. 2C, the thematic role orders and the argument structure nodes are linked to the verb load, representing a cluster of features associated with the verb's lemma.This cluster includes the verb's meaning (or lexical concept, Levelt et al., 1999) and its syntactic properties.Moreover, the diagram includes a node labelled 'declarative', which signifies the representations and processes relating to the selection of a declarative construction.This includes the conceptual feature or illocutionary force associated with declarative sentences, such as making an assertion (Searle, 1979).Construction nodes are shared among various argument structures and thematic role orders and distinct from abstract constituent structures that define syntactic categories and relationships between words and phrases.Inclusion of a construction node captures the relationship between thematic role ordering and construction selection.Specifically, if speakers intend to emphasize a specific thematic role in English, they may employ cleft constructions.Selection of a particular cleft construction, in turn, influences thematic role ordering.In a what-cleft construction, the emphasized role appears in the sentence's final position (e.g., What the assistant loaded into the van was the boxes).Conversely, in an it-cleft construction, the emphasized role is positioned at the beginning of the sentence (e.g., It was the boxes that the assistant loaded into the van).Furthermore, the co-occurrence of various structural features depends on the specific construction used.While both the with argument structure and locative structure can take different thematic role orders in cleft constructions (see Tables 5 & 10), in declarative constructions, the agent-location-theme order occurs with the with argument structure and the agent-theme-location order occurs with the locative argument structure (see Table 1).Therefore, in the diagram, the links between the A-L-T node and the with-structure node, as well as

Table 1
Example prime and target sentences in Experiments 1-3.

Experiment 1 Verb Repeated Prime
Not-Repeated Prime 2a.The assistant loaded the van with (the boxes).
2c.The assistant packed the van with (the boxes).2b.The assistant loaded the boxes into (the van).
2d.The assistant packed the boxes into (the van).Target Completion 3a.The man loaded (the trolley with the luggage).3b.The man loaded (the luggage onto the trolley).

Experiment 2 Theme Repeated Prime
Not-Repeated Prime 4a.The traveller packed the car with (the suitcases).
4c.The traveller packed the car with (the skis).4b.The traveller packed the suitcases into (the car).
4d.The traveller packed the skis into (the car).Target Completion 5a.The man loaded (the trolley with the suitcases).5b.The man loaded (the suitcases onto the trolley).

Experiment 3 Location Repeated Prime
Not-Repeated Prime 6a.The tour operator packed the trolley with (the rucksacks).
6c.The tour operator packed the sledge with (the rucksacks).6b.The tour operator packed rucksacks onto (the trolley).
6d.The tour operator packed rucksacks onto (the sledge).Target Completion 7a.The man loaded (the trolley with the suitcases).7b.The man loaded (the suitcases onto the trolley).

Note. Example completions in brackets.
K. Fukumura and F. Yang between the A-T-L node and the locative structure node, reflect the typical co-occurrence of these structural features in declarative constructions. 2he bold lines in Fig. 2C illustrate an increased activation of all nodes and connections following a prime sentence like (1a).In line with many theories of human memory and information processing, when a node becomes activated, its activation spreads to other nodes linked to it, increasing their levels of activation.Importantly, when multiple nodes are active, the activation of the connecting links increases due to the intensified transmission of activation among them.This can influence the future co-occurrence of the connected nodes.For instance, a location-theme order (with-structure) prime such as (1a) will activate not only the A-L-T node and the withstructure node, thereby increasing the probability of selecting these structural features in subsequent utterances, but also the verb load, the declarative construction node, and the links connecting these nodes.This forms the basis of the lexical boost effect, enhancing the probabilities of repeating these structural features alongside the verb and the primed construction.Consider a scenario where, following the prime (1a), the verb load is repeated in another declarative construction, resulting in enhanced activation of both the load and declarative nodes.This enhanced activation then spreads to all the interconnected nodes, during which the A-L-T node and the withstructure node receives particularly strong activation.This is because the links from these structural features to load and to the declarative node have been preactivated by the prime; as a result, the residual activation of these links feeds into the A-L-T node and the with-structure node.Hence, repeating the verb in a declarative sentence increases the probability of selecting these structural features.
All three models we have discussed assume that verb repetition leads to a lexical boost effect, enhancing structural priming.However, some researchers have suggested that not only verb repetition but also noun repetition can enhance structural priming.For instance, Chang et al. (2006) proposed that the lexical boost effect can occur with both verbs and nouns because they are content words, and their repetition can serve as a retrieval cue for the explicit short-term memory of the prime (see also Chang et al., 2012;Reitter et al., 2011).On this view, we cannot rely on verb boost effects to examine underlying structural representations, given that any content word repetition can potentially enhance priming effects.Consistent with this idea, Scheepers et al. (2017) found that repeating the agent or recipient role led to larger structural priming effects in dative alternations.Scheepers et al. manipulated the repetition of three nouns with distinct roles (agent, theme, and recipient) in addition to the verb in the same experiments.However, Carminati et al. (2019) failed to replicate these findings when they manipulated noun repetition and verb repetition in separate experiments.Carminati et al. thus proposed that the enhancement of syntactic priming occurs solely with the repetition of syntactic heads, such as verbs.This proposal has been supported by subsequent studies investigating verb boost effects in dative alternations (Huang et al., 2023;Kantola et al., 2023;Van Gompel et al., 2023).However, adjective repetition has been shown to boost adjective order priming (Fukumura & Zhang, 2023), and the role of noun repetition in enhancing structural priming might vary depending on the specific alternations being investigated.For example, Chang et al. (2006) suggested that language users rely more on semantic representations for the theme-location alternation, where constituent structure fails to discriminate between structural alternatives, unlike in dative alternations.On this view, the repetition of role-bearing filler nouns could potentially induce a lexical boost effect, particularly if language users associate the thematic roles with the concepts of the filler nouns in their semantic representations.
Thus, Experiments 1-3 examined if the priming of the theme-location alternation could be enhanced by repeating the verb (Experiment 1), the noun fulfilling the theme role (Experiment 2), and/or the noun fulfilling the location role (Experiment 3) from the prime to the target.As a preview, the findings of these experiments revealed that only verb repetition, not noun repetition, enhances theme-location alternation priming, ruling out the verb-constituent structure account.Experiments 4-6 then contrasted the interactive structure-building account and the verb-argument structure account by independently manipulating the argument structure and thematic role order of the prime using cleft constructions.Below we report each experiment in turn.

Experiment 1: Verb repetition
In all experiments, a web-based written sentence completion task was employed.Participants were tasked with completing a prime prompt in writing, selecting one of the provided words (as shown in the top panel of Fig. 3).Participants then described a given target picture by responding to target prompts (e.g., The man loaded …) using words present in the image (as depicted in the bottom panel of Fig. 3).Examples of sentences from Experiments 1-3 can be found in Table 1.In Experiment 1, each prime prompt featured either the same verb (2a & 2b, load) or a different verb (2c & 2d, pack) compared to the targets (load).The primes induced either a location-theme order completion (with-structure) (2a & 2c) or a theme-location order completion (locative structure) (2b & 2d).If structural priming plays a role, there should be more location-theme order responses (3a) following location-theme order primes (2a & 2c) than following theme-location primes (2b & 2d).Of particular interest is whether this priming effect interacts with verb repetition.According to the verb-constituent structure account, the effect should remain unaffected by verb repetition, because the structural variants of the theme-location alternation share the constituent structure, and verb selection is associated with the selection of an abstract constituent structure.However, both the verb-argument structure account and the interactive structure-building account predict a larger priming effect when the verb is repeated compared to when it is not.These hypotheses propose that verb selection is associated with argument structure selection (verb-argument structure account) or with both the argument structure and thematic role order selected in a particular construction of the prime (interactive structure-building account).priming, Ivanova et al., 2012Ivanova et al., , 2017)), such connections can be formed and can influence structural choice.The diagram omits these ungrammatical connections because they are typically of extremely low activation unless they are specifically primed, and we did not use ungrammatical prime sentences in the current study.

Participants
Eighty native speakers of British English participated in the study, recruited from Prolific (www.prolific.com) in exchange for cash.All participants reported being a student, aged between 18 and 30, raised in an English-speaking monolingual household, and having no reading difficulties.The sample size was based on Experiment 1 of Chang et al.'s (2003) study, which had 83 participants and 16 experimental items.To increase statistical power for investigating interactions, we doubled the number of experimental items, and employed a picture description task, which was expected to result in much fewer exclusions compared to Chang et al.'s sentence recall task, whose exclusion rates were approximately 60 %.

Materials
We constructed 32 experimental items, each comprising sentence fragments for prime and target trials, and a target picture.Both prime and target sentence fragments began with a subject noun, followed by a verb.In the location-theme order prime condition (2a & 2c) (load/pack the van with the boxes), the postverbal noun phrase denoted a location (van) followed by the preposition "with", prompting the mention of a noun phrase denoting a theme (boxes).In the theme-location prime condition (2b & 2d) (load/pack the boxes into the van), the postverbal noun phrase denoted a theme, followed by a locative preposition, prompting the mention of a noun phrase denoting a location.We used 16 verbs from Levin (1993), and verb repetition was manipulated by pairing up verbs with similar meanings, each occurring across four items: load/pack, rub/dab, smear/slather, splash/splatter, drizzle/spray, stuff/cram, pile/stack, and dust/sprinkle (see Appendices 13.1 & 13.2).Each prime prompt appeared across two different items, once as a repeated verb prime and once as a not-repeated verb prime, thereby counterbalancing the prime properties across conditions.We rotated the conditions such that each participant saw each prime only once.
The prime fragments had to be completed by choosing one of the two words provided below the prime fragments (e.g., boxes or castle) (see Fig. 3, top panel).The positions of these words on the screen were counterbalanced.The target fragments always started with the subject (agent) followed by the verb (e.g., The man loaded, as in 3), which had to be completed using the words embedded in the target picture (see Fig. 3, bottom panel).In the repeated verb condition, the prime and target fragments had the same verb, whereas in the not-repeated verb condition, they had different verbs.The prime and the target always had different agents, themes, and locations.The target picture (500 pixels × 500 pixels, 96 pixels per inch) depicted an action involving an agent, a theme, and a location.The  agent and action were named in the target prompt (3), and the theme and location were named (in a font size of 36) in the image (e.g., luggage and trolley in Fig. 3 bottom panel).In addition, we constructed 72 filler items using various sentence structures, including transitive, intransitive, passive, and other forms.Similar to experimental items, each filler item also contained a sentence fragment that could be completed by choosing one of two words given, followed by a target image with another sentence fragment.

Design
We used a 2 (prime structure: theme-location [locative structure] vs. location-theme [with-structure]) × 2 (verb repetition: repeated vs. not-repeated) within-subjects/items repeated measures design, resulting in the creation of four lists.Each list had 32 experimental items, with only one version of each item and 8 items per condition, and 72 filler items.The conditions and items were distributed across each list in a fixed quasi-random order, subject to the constraints that each verb should not occur more than twice and should appear in different prime orders and in different halves of the experiment, and there should be at least 2 filler trials between experimental trials.Twenty participants were randomly assigned to each list.

Procedure
The experiment was conducted online using the Gorilla experiment builder (www.gorilla.sc).Following informed consent, participants confirmed their eligibility by answering a short demographic questionnaire.They then read instructions and went through the procedures and practice trials.Each trial began with the prime trial (Fig. 3 top panel), where participants completed a prime sentence fragment in a blank box by choosing one of the two possible words shown below the box.They then pressed a key, which triggered the presentation of a target sentence fragment.Participants completed the target sentence by describing a picture (Fig. 3 bottom panel).They had to use the words embedded in the picture, which helped to ensure that all the thematic roles would be mentioned.Once completed, they pressed a key to proceed to the next trial.Halfway through the experiment, there was a short break, and the experiment lasted about 30 min on average.During the practice trials, participants were asked to complete the sentence meaningfully and grammatically.They were told that they were free to add any words necessary and choose different word orders, and also told not to worry about spelling.The practice trials involved transitive sentences, passives, and dative alternations but never locative alternations.

Scoring
We scored if participants produced a location-theme response (n = 644) (3a) or theme-location response (3b) (n = 1889).Themelocation responses had a post-verbal theme, followed by a locational preposition (e.g., on, into, onto, in, over, inside, across) and a location.Location-theme responses had a post-verbal location, followed by a preposition and a theme.Almost all location-theme responses had the preposition with (e.g., The boy rubbed hands with soap) (n = 629), though we included responses that used in for the verbs slather/rub/splatter/splash (e.g., the boy slathered his body in sunscreen) (n = 15) (the exclusion of these responses did not alter the results we report below).All other responses were coded as Other (1.1 % of total 2560 responses, n = 27) and excluded from analyses.These included cases where responses included only one thematic role (n = 15; e.g., The woman stuffed the turkey) or omitted prepositions (n = 3; e.g., The boy slathered sunscreen his body) or mentioned non-target prepositions including from (n = 2, e.g., The girl splattered her trousers from the mud) or through (n = 1, The girl splashed liquid through the window); the responses contained an error (n = 1; The woman splashed face onto her face); or there were technical errors (n = 5).These were excluded from analyses.

Results and discussion
Fig. 4 displays the percentages and standard errors of location-theme order (with-structure) responses relative to theme-location order (locative structure) responses (see Appendix 13.10 for count data).Throughout this article, we report results based on Wald z-scores from logit mixed effects models (Baayen et al., 2008;Barr et al., 2013), implemented in the lme4 package (Bates et al., 2015b) in R (version 4.0.3., R Core Team, 2021).The use of Wald z-scores for hypothesis testing is well-established and offers advantages over analyses such as Likelihood Ratio Test (LRT) models.One key advantage is that it allows for determining the significance of each parameter in a single analysis in R, eliminating the need for multiple comparisons required by LRT tests.This reduces computational complexity and ensures computational efficiency.Conducting multiple model comparisons can be challenging, especially when dealing with complex models or binomial distributions that often encounter convergence issues.To check the robustness of our findings, however, we carried out supplementary LRT tests, and the results are completely consistent with those obtained using Wald zscores (see Appendix 13.20).Furthermore, we used Bayes Factors (e.g., Lee & Wagenmakers, 2013) to evaluate predictions regarding null effects and interpreted the results following the schemes proposed by Jeffreys (1961).The Bayes Factors were computed using the bayestestR package (Makowski et al., 2019).Data and R scripts are available at https://osf.io/m5whg/?view_ only=86f83d06ab1a444581cea1b3aa57e4f.
The fixed effects were mean-centered and standardized, unless stated otherwise.This approach allowed us to interpret the results in terms of main effects and interactions by minimizing collinearity between variables (Baayen et al., 2008) and facilitated convergence in R (Gelman & Hill, 2007).All analyses included initial models with random slopes and intercepts by subjects and by items (Barr et al., 2013).To avoid overfitting (Bates et al., 2015a;Matuschek et al., 2017), we reduced the model complexity by suppressing correlations between random effects (Bates et al., 2015a(Bates et al., , 2015b;;Kliegl, 2014;Singmann & Kellen, 2020).Additionally, we simplified the model further by eliminating random effects with close-to-zero variances when encountering indications of failed model convergence (Bates et al., 2015a).This procedure had either no or negligible numerical impact on the results, indicating that including or excluding random effects with zero variances did not influence the outcomes.
In Experiment 1, we analyzed whether the choice of location-theme (with-structure) and theme-location (locative structure) orders (coded as 1 and 0 respectively) was affected by prime structure (location-theme vs. theme-location, coded as 1 and 0 respectively) and verb repetition (repeated vs. not-repeated, coded as 1 and 0 respectively).Table 2 summarizes the results.As expected, participants adopted more location-theme orders (with-structures) following location-theme (with-structures) primes (31.1 %, SE = 1.3 %) than following theme-location (locative structures) primes (19.7 %, SE = 1.1 %), showing an effect of structural priming.Importantly, the effect was larger when the verb was repeated (17.6 %) than when the verb was not repeated (5.1 %), though simple effects were reliable in both cases.The Bayes Factor for this interaction (BF 01 ) was 0.064, providing no support for the null hypothesis.In fact, the small number supports the existence of the verb boost effect.Cruciallly, this verb boost effect cannot be attributed to enhanced constituent structure priming because the structural variants share the constituent structure.Experiment 1 thus provided evidence against the verb-constituent structure account.This account suggests that verb repetition enhances structural priming by boosting the activation of the primed abstract constituent structure.Therefore, according to this view, the magnitude of the structural priming effect should have remained unaffected by verb repetition, because the location-theme orders (with-structures) and theme-location orders (locative structures) share the same abstract constituent structure.

Experiment 2: Theme Repetition
While both the verb-argument structure account and the interactive structure-building account propose that verb repetition leads to increased structural priming, neither account predicts the same effect for the repetition of nouns representing thematic roles.However, as discussed earlier, it has been proposed that both verb and noun repetition can enhance structural priming (e.g., Chang et al., 2006;Reitter et al., 2011;Scheepers et al., 2017), challenging the validity of the structural representations assumed under these two accounts.Although recent findings in the dative alternation have failed to support such a hypothesis, the theme-location alternation might work differently from the dative alternation, as suggested by Chang et al. (2006), possibly relying more upon associations between thematic roles and noun concepts.Experiments 2 and 3 examined this possibility, investigating whether the repetition of nouns can lead to a larger priming effect.Specifically, Experiment 2 focused on assessing the impact of theme repetition.As shown in Table 1, the primes had either the location-theme (with-structure) (4a & 4c) or theme-location (locative structure) (4b & 4d) order, and they featured either the same theme (suitcases, 4a & 4b) or different themes (skis, 4c & 4d) from the target (5).If language users associate the prime structure with the noun fulfilling the theme role, we should observe a stronger tendency to produce sentences with the primed structure when the theme noun is repeated in the target than when it is not repeated.

Participants
A further 80 participants were recruited from the same population via Prolific as before.An additional participant was recruited but removed from analyses as they failed to follow instructions in one third of the experimental trials.

Materials and procedure
These were the same as in Experiment 1, except that we adapted the materials in Experiment 1 such that the prime and target prompts always had different verbs and locations (see Appendices 13.3 & 13.4), and that we manipulated the repetition of theme: Primes had either the same themes (4a & 4b) or different themes (4c & 4d) from targets (5).

Design
We used a 2 (prime structure: theme-location [locative structure] vs. location-theme [with-structure]) × 2 (theme repetition: repeated vs. not-repeated) within-subjects and within-items repeated measures design.The 32 experimental items and 72 filler items were distributed across four lists as in Experiment 1.

Scoring
We scored if participants produced a theme-location response (n = 1995) or a location-theme response (n = 537) using the same coding scheme as in Experiment 1.As in Experiment 1, we included location-theme orders with the preposition in for the verb dab (1.7 %, n = 9; e.g., The man dabbed his hand in some moisturiser).We excluded cases (28 out of 2560, 1.1 %) when only one thematic role was mentioned (n = 8); the preposition was replaced by full of for the verbs pack/cram/stuff/pile (n = 10) or omitted (n = 1) or an unnecessary preposition was erroneously included (e.g., The man dabbed at his hand with moisturiser) (n = 1); the response was structurally ambiguous (n = 1; The man crammed an armchair in the truck with other furniture); the response contained a non-target proposition (n = 1, The girl drizzled vinegar of her dish), a non-target interpretation (n = 1, The boy splashed in the mud next to a bench) or an exchange error (n = 3, e.g., The man dabbed his hand on moisturiser); or there were technical errors (n = 2).

Results and discussion
Fig. 5 reports the percentages and standard errors of location-theme responses (see Appendix 13.10 for count data).The analyses included prime structure (location-theme vs. theme-location) and theme repetition (repeated vs. not-repeated) as fixed effects.Table 3 provides a summary of the results.As in Experiment 1, a main effect of prime order revealed more location-theme orders (withstructures) after location-theme (with-structure) primes (25.5 %, SE = 1.2 %) than after theme-location (locative structure) primes (16.9 %, SE = 1.1 %).However, there was no significant prime order × theme repetition interaction; in fact, the interaction went in the opposite direction from a lexical boost effect.The Bayes Factor (BF 01 ) for the interaction was 38.95, providing strong support for the null hypothesis.This contrasts with Experiment 1, where verb repetition from the prime to the target led to larger structural priming.Simple effects of prime order were reliable in both the theme-repeated condition (7.4 %) and the not-repeated condition (9.8 %).Experiment 2 thus showed that the repetition of the noun bearing the theme role does not enhance the priming of the theme-location alternation.

Experiment 3: Location Repetition
Experiment 3 examined the impact of location repetition on the theme-location alternation priming.As shown in Table 1,  K. Fukumura and F. Yang primes ( 6) had either the location-theme (with-structure) (6a & 6c) or theme-location (locative structure) (6b & 6d) order and either the same location (trolley, 6a & 6b) or different locations (sledge, 6c & 6d) from the target (7).If language users associate the noun fulfilling the location role with the structure of the prime sentence, the effect of prime structure on structural choice in the target (7a vs. 7b) should be larger when the location noun is repeated than when it is not.

Participants
A further eighty participants were recruited from the same population as before.

Materials and procedure
These were the same as before, except that the prime and target always had different verbs and themes, and we manipulated location repetition: Primes and targets had either the same location (6a & 6b) or different locations (6c & 6d) from the targets (7) (see Appendices 13.5 & 13.6).

Design
We used a 2 (prime structure: theme-location [locative structure] vs. location-theme [with-structure]) × 2 (location repetition: repeated vs. not-repeated) within-subjects and within-items repeated measures design.The 32 experimental items and 72 filler items were distributed across four lists as before.

Scoring
We scored if participants produced a theme-location response (n = 2110) or a location-theme response (n = 408) using the same coding scheme as before.We included location-theme responses with the preposition, in, for the verb, slather (n = 3), but excluded cases (1.6 % of total 2560 responses, n = 42), where participants mentioned only one thematic role (n = 5); they used full of instead of a preposition (for the verb stuff) (n = 3) or omitted the preposition (n = 1); the thematic roles were mentioned in separate clauses (n = 1, e.g. the man packed up the boxes and put them in the van); a different location was mentioned (n = 2, e.g., desk rather than shelf); the responses contained exchange errors or conveyed implausible meanings (n = 3); there were technical errors (n = 4).Cases where participants chose the implausible location in the prime trials (n = 23) were also excluded.

Results and discussion
Fig. 6 reports the means and standard errors (see Appendix 13.10 for count data), and Table 4 summarizes the results.As before, a significant main effect of prime structure indicated a higher proportion of location-theme orders following location-theme primes (19.9 %, SE = 1.1 %) compared to theme-location primes (12.5 %, SE = 0.9 %).This main effect did not interact with the repetition of the location noun, and the simple effects of prime order were reliable in both the location-repeated condition (7.5 %) and the notrepeated condition (7.2 %).The results are in line with Experiment 2, where the repetition of the noun representing the location role did not interact with the priming effect.The BF 01 for this interaction was 49.98, again, providing very strong evidence in support of the null hypothesis.
Hence, the results from Experiments 1-3 indicate that verb repetition, but not the repetition of the nouns fulfilling the theme or location roles, reliably enhances the priming of the theme-location alternation.However, the absence of noun boost effects raises the question of whether our experiments had sufficient statistical power to detect significant interactions.To address this concern, we conducted combined analyses for Experiments 2 and 3, doubling the sample size, to investigate whether noun repetition interacts with the prime in the larger dataset.The results showed no significant noun repetition × prime structure interaction, Estimate = -0.02,SE = 0.04, z = -0.42,p = .672,with the Bayes Factor for this interaction (BF 01 = 65.26)providing very strong support for the null hypothesis (see Appendix 13.11 for full results). 4In contrast, another combined analysis contrasting Experiment 1 with the combination of Experiments 2 and 3 revealed a three-way interaction (prime structure × lexical repetition × experiment), confirming that the lexical boost effect depended on the repeated lexical item, Estimate = 0.24, SE = 0.07, z = 3.42, p = .001(see Appendix 13.11 for full results).Overall, these additional analyses confirm that while the repetition of the verb enhances structural priming, the repetition of the theme or location does not have the same magnitude of effect, if any, on enhancing structural priming.

Experiment 4: From What-Cleft Prime To What-Cleft Target
Experiment 1 ruled out the verb-constituent structure account, according to which there should be no verb boost effect when the structural variants share the constituent structure.In this experiment, the thematic role orders and argument structures co-varied in declarative constructions.To examine the representations underlying structural priming further, Experiments 4-6 used cleft constructions.This allowed us to independently manipulate thematic role order and argument structure, as described below.The key question was whether verb repetition exclusively enhances the persistence of argument structure, as predicted by the verb-argument structure account.An alternative possibility was that verb repetition also reinforces the persistence of thematic role order, as posited by the interactive structure-building account, according to which the persistence of thematic role order should interact with that of argument structure when the construction of the prime persists in the target.Thus, in Experiment 4, both primes and targets involved what-cleft constructions.Each prime had either a location-theme order (8a, 8b, 8e, 8f), emphasizing the theme role, or a theme-location order (8c, 8d, 8g, 8h), emphasizing the location role.Within each thematic role order, argument structure and constituent structure were also varied.Sentences (8a, 8c, 8e, 8g) involve a with-structure, whereas sentences (8b, 8d, 8f, 8h) involve a locative structure.In NP-first sentences (8a, 8d, 8e, 8h), the noun phrase representing the direct object preceded the prepositional object.In PP-first sentences (8b, 8c, 8f, 8g), the prepositional phrase containing the prepositional object preceded the direct object.To illustrate the predictions of the two accounts, the diagrams in Fig. 7A and Fig. 7B represent the interactive structure-building account and the verb argument structure account, respectively.The bold lines in the diagrams indicate an increased activation of the relevant nodes and links activated when a what-cleft sentence such as (8a) has been presented as a prime.
According to the interactive structure-building account (Fig. 7A), a prime sentence such as (8a) activates the A-L-T node (representing an Agent-Location-Theme order), the with-structure node (representing a with-argument structure), the verb load, and the Whcleft node (representing the what-cleft construction), along with the links connecting these nodes.The increased activation of the A-L-T node and the with-structure node subsequently enhances the likelihood of repeating a location-theme order and a with-structure in the target, respectively.Now, imagine that the verb load is to be repeated in another what-cleft sentence.This repetition further enhances the activation of the verb load and the Wh-cleft node, leading to a spread of activation throughout the network.In this process, the A-L-T node and the with-structure node receive robust activation from their links to load and the Wh-cleft node.This is due to the pre-activation of these links by the prime (8a), which results in their residual activation contributing to the A-L-T and with-structure nodes when the activation spreads from load and the Wh-cleft node.Consequently, repeating load and a what-cleft construction in the target sentence boosts the activation of the A-L-T node and the with-structure node.This, in turn, increases the probability of repeating a location-theme order and a with-structure, alongside load and the what-cleft construction, in the target sentence.
In this account, verb repetition is not the sole factor capable of enhancing priming effects.Suppose that, following a prime such as (8a), speakers produce another what-cleft sentence with the same thematic role order in the target.This further enhances the A-L-T node and the Wh-cleft construction node.As this enhanced activation spreads from these nodes, the with-structure node, which is linked to these nodes, receives a substantial activation boost.This is because the links from the A-L-T node and the Wh-cleft node to the with-structure have been pre-activated by the prime sentence (8a), resulting in residual activation feeding into the with-structure node as well as the spreading activation from the A-L-T node and the Wh-    Note.NP = Noun Phrase, PP = Prepositional Phrase.Example completions in brackets.
K. Fukumura and F. Yang two nodes spreads through the network, the A-L-T node gains strong activation due to its pre-activated links to the with-structure node and the what-cleft structure node.Thus, further activating the with-structure node and the Wh-cleft construction node in the target increases the likelihood of a persisting A-L-T order alongside a with-structure and a what-cleft construction, given their co-occurrence in the prime.Note that the diagram in Fig. 7A omits constituent structures (NP-first vs. PP-first), reflecting the model's assumption that the primary driver of structural priming is the persistence of argument structure and thematic role order.In this account, while the argument structure and thematic role order of the prime can interact, they can also persist independently.For example, the withstructure of the prime (8a) can elicit a with-structure in the target even if the A-L-T order of the prime (8a) does not persist.Likewise, the A-L-T order of the prime can persist in the target even when a locative structure, rather than a with-structure, is selected in the target.However, constituent structure cannot persist on its own; instead, its persistence hinges on the simultaneous persistence of both argument structure and thematic role order.If the thematic role order of the prime persists in the target (e.g., both employing a location-theme order, 9a & 9b following 8a), NP-first primes (8a) can elicit NP-first targets (9a) by sharing both argument structure (with-structure) and thematic role order (location-theme order).However, if a different thematic role order is adopted in the target (e.g., 9c & 9d following 8a), NP-first primes (8a) will not elicit NP-first targets (9d) due to disparities in argument structure and thematic role order.Likewise, when the prime's argument structure persists in the target (e.g., both having a with-structure, 9a & 9c following 8a), NP-first primes (8a) can elicit NP-first targets (9a) through the shared thematic role order and argument structure.However, if the argument structure does not persist in the target (e.g., 9b & 9d following 8a), NP-first primes (8a) will not induce NPfirst targets (9d) due to the divergence in both argument structure and thematic role order.
In contrast, according to the verb-argument structure account (Fig. 7B), the constituent structure of the prime should be separately primed from its argument structure, because constituent structures are represented distinctly from argument structures.Thematic role orders are not represented in the diagram in Fig. 7B, as they are considered irrelevant for structure building in this model.Hence, accorting to this model, processing the prime (8a) should increase the activation of the constituent structure What-cleft: NP first, as well as the with-structure and the verb load along its link to the with-structure.The increased activation of the with-structure node and the constituent structure What-cleft: NP first should enhance the probability of subsequently selecting a with-structure and an NP-first constituent structure, respectively, in the target.The verb load is linked to the with-structure node but neither to the constituent structure nor to the thematic role order, which is absent in the model.Hence, verb repetition should enhance the persistence of the primed argument structure only.Furthermore, constituent structure persistence should happen not only when the argument structure and the thematic role order persist in the target, but also when different argument structures or thematic role orders are chosen in the target.Moreover, due to the sequential nature of argument structure selection preceding constituent structure selection, the persistence of the argument structure should remain unaffected by whether the linear order of constituents (and thus the thematic role order) also persists in the target or not.

Participants
A further eighty participants were recruited from the same population as before via Prolific and the University of Stirling student population.An additional 15 participants took part but were excluded from analyses because more than 12 out of 32 responses were invalid, mostly failing to include a copula verb after the what-clause (e.g., What the boy packed stationary into the bag).

Materials, design and procedure
We had 32 experimental items.The prime sentences used in Experiments 1-3 were adapted to form what-cleft constructions, and the target prompts began with what-clauses which involved the subject and the verb, and we used identical or similar target events as those used in Experiments 1-3 (see Appendices 13.7 & 13.9).As shown in Table 5, the prime sentences had two thematic role orders: Agent-Location-Theme (A-L-T) order primes (8a, 8b, 8e, 8f) and Agent-Theme-Location order (A-T-L) primes (8c, 8d, 8g, 8h).In the A-L-T order primes, the location (e.g., lift) was mentioned after the verb in the what-cleft construction, with the theme mentioned after the copular verb, emphasizing the theme in the sentence.In the A-T-L order primes, the theme (e.g., equipment) immediately followed the verb in the what-cleft construction, and the location, mentioned after the copular verb, was emphasized.Within each thematic role order condition, what-clefts are derived from either with variants (with-structure, 8a, 8c, 8e, 8g), having the location role as the direct object of the verb, or locative variants (locative structure, 8b, 8d, 8f, 8h), having the theme role as the direct object of the verb.In addition, in NP-first sentences (8a, 8d, 8e, 8h), the verb within the what-clause preceded a noun phrase denoting the direct object, whereas in PP-first sentences (8b, 8c, 8f, 8g), the verb in the what-clause preceded a prepositional phrase containing the prepositional object.The 32 experimental items and 70 filler items were distributed across four lists as before.Because the experimental items always had what-clefts, some filler items had what-clefts (n = 6) or it-clefts, either in primes (n = 19) or targets (n = 9).The rest of the filler items were the same as those used in Experiments 1-3.

Scoring
We scored if participants produced a with-structure (n = 596) and a locative structure (n = 1647) and if they produced locationtheme order (theme-focused) responses (n = 1968) or a theme-location order (location-focused) response (n = 275) when producing a what-cleft target.In addition, we scored whether participants produced NP-first orders (n = 751) or PP-first orders (n = 1492).We included cases where the preposition with was replaced with another preposition for slather and smear (e.g., What the boy slathered his body in was sunscreen) (n = 8).Cases were excluded where a copula verb was omitted (e.g., What the girl splattered the wall with paint; n = K.Fukumura and F. Yang 200); one of the roles was used as a modifier for another role (e.g., What the man rubbed was wax onto the shoe; n = 61); only one argument was mentioned (n = 3); the preposition was omitted (n = 16) or placed in a non-target position (e.g.What the man drizzled the dish was with olive oil; n = 12) or mentioned twice (e.g., What the man stacked the shelves with was with tins; n = 2) and the response contained a word change error (n = 2).Additionally, we excluded cases where participants did not complete the prime fragments as we expected (e.g., by choosing an implausible noun or including a redundant preposition; n = 21).In total, 317 out of 2560 trials (12.4 %) were excluded from analyses.

Results and discussion
Fig. 8 reports the percentages of target responses by prime thematic role order, prime argument structure and verb repetition (see Appendix 13.12 for count data and standard errors).In the following, we present three sets of analyses, each examining the rates of argument structure, thematic role order, or constituent structure in the target, respectively.These analyses were conducted to test specific hypotheses regarding different aspects of the two accounts contrasted in the experiment.

Argument structure persistence
Here we examined the binary choice of argument structure (with-structure vs. locative structure), collapsing across thematic role order in the target.The first analysis focused on determining whether the argument structure of the prime persisted in the target (argument structure priming), and if this priming effect was stronger when the verb was repeated compared to when it was not, as predicted by both the interactive structure-building account and the verb-argument structure account.The model thus included fixed effects such as the thematic role order of the prime (location-theme vs. theme-location, coded as 1 and 0, respectively), verb repetition (repeated vs. not-repeated, coded as 1 and 0, respectively), and prime argument structure (with vs. locative, coded as 1 and 0, respectively) as fixed effects.For target responses, the with-structure was coded 1 and the locative structure as 0. Fig. 9 presents the means and standard errors for this analysis (see Appendix 13.13 for count data) and the results are summarized in Table 6 (1).
A main effect of prime argument structure indicated more with-structure responses after with-structure primes (37.4 %, SE = 1.4 %) than after locative structure primes (15.8 %, SE = 1.1 %).This argument structure priming effect interacted with verb repetition, with a larger effect of prime argument structure with repeated verbs (29.1 %) than with not-repeated verbs (14.2 %), though the effect was reliable regardless of verb repetition.There was no main effect of thematic role order, which indicated that the rates of with-structure and locative structure did not differ reliably depending on whether the prime had a location-theme order or a theme-location order.Neither the effect of prime argument structure nor the verb boost effect interacted with prime thematic role order, either, indicating no evidence that either effect was dependent on a particular thematic role order of the prime.
The second analysis aimed to determine whether the argument structure of the prime exhibited greater persistence when the thematic role order of the prime persisted in the target, as predicted by the interactive structure-building account, but not by the verb argument structure account.Hence, the analysis included a new variable, thematic role order congruence, which was coded as 1 when participants adopted the same thematic role order as in the prime and 0 when they used a different thematic role order (cf.Fukumura & Zhang, 2023).Fig. 10 reports the means for this analysis (see Appendix 13.13 for count data) and Table 6 (2) summarises the results.The analysis revealed an interaction between argument structure and thematic role order congruence: The effect of prime argument structure was larger when the thematic role order of the prime was repeated in the target (29.8%) than when it was not (10.4%).However, the effect of prime argument structure was reliable regardless of whether the same thematic role order was repeated in the target or not.The prime argument structure × thematic role order congruence interaction was not modulated by verb repetition, showing no evidence of a verb boost effect on argument structure persistence being dependent on thematic role order congruence.

Thematic role order persistence
Here we examined the binary choice of thematic role order or emphasis.The first analysis investigated whether the thematic role order of the prime persisted in the target (thematic role order priming) and whether this effect was larger following verb repetition, as predicted by the interactive structure building account, but not by the verb-argument structure account.The analysis thus included prime thematic role order, prime argument structure, and verb repetition as fixed effects.Location-theme order responses were coded as 1, while theme-location order responses were coded as 0. Fig. 11 reports the means (see Appendix 13.14 for count data), and Table 7 (1) summarizes the results.A main effect of prime thematic role order showed a higher rate of location-theme orders after locationtheme primes (95.0 %, SE = 0.6 %) than after theme-location primes (80.3 %, SE = 1.2 %), demonstrating the persistence of prime thematic role order (or emphasis).A prime thematic role order × verb repetition interaction revealed a larger thematic role order effect when the verb was repeated (17.8 %) than when it was not (11.5 %).The absence of a thematic role order × argument structure interaction and a three-way interaction showed no evidence that thematic role order priming and the enhanced priming effect following verb repetition were affected by the argument structure of the prime.
The second analysis examined whether the thematic role order of the prime tended to persist more when the argument structure from the prime persisted in the target, and whether the thematic role order from the prime persisted with a different argument structure from the prime, as predicted by the interactive structure building account.For this analysis, we created a new variable called argument structure congruence, which was coded as 1 when participants adopted the same argument structure as the prime, and 0 when they used a different argument structure.Fig. 12 reports the means for this analysis (see Appendix 13.14 for count data), and Table 7 (2) summarizes the results.The analysis revealed a thematic role order × argument structure congruence interaction, with a larger effect of prime thematic role order when the target adopted the same argument structure as in the prime (22.3 %) than a different argument structure from the prime (3.2 %), though the effect of prime thematic role order was reliable in both cases.This interaction was not modulated by verb repetition, indicating that the relationship between thematic role order priming and argument structure  congruence was not dependent on verb repetition.In this analysis, there was no reliable prime thematic role order × verb repetition interaction, which we discuss in 6.2.4.

Constituent structure persistence
Thus far, our analyses have demonstrated the persistence of both the argument structure and thematic role order of the prime from what-cleft primes to what-cleft targets.While these two effects appeared to interact, both the argument structure and thematic role order of the prime were able to persist separately, even in the absence of their counterpart persisting.Here, we examine the impact of prime constituent structure, specifically, the order of syntactic categories (functions) in the prime (NP [direct object]-first vs. PP [prepositional object]-first), on structural choice in the target.Our primary focus was to determine whether the constituent structure of the prime persists in the target, even when a different argument structure or thematic role order is adopted in the targeta prediction consistent with the verb-argument structure account.An alternative possibility is that constituent structure persistence only occurs  when primes and targets share both thematic role orders and argument structures, as predicted by the interactive structure building account.
The first analysis thus examined whether constituent structure persistence interacted with thematic role order congruence, and if prime constituent structure persisted in the target even when a different thematic role order was selected in the target as predicted by the verb-argument structure account.We analyzed the rates of direct-object first (NP-first) responses (coded as 1) and prepositional object first (PP-first) responses (coded as 0) as a function of prime constituent structure (NP-first vs. PP-first, coded as 1 and 0), thematic role order congruence, and verb repetition.Fig. 13 reports the means for the first analysis (see Appendix 13.15 for count  data), and Table 8 (1) summarizes the results.A main effect of prime constituent structure indicated a higher rate of NP-first responses after NP-first primes (40.8 %, SE = 1.5 %) compared to PP-first primes (26.1 %, SE = 1.3 %).However, there was a significant interaction between prime constituent structure and thematic role order congruence, as well as a significant three-way interaction between prime constituent structure, thematic role order congruence, and verb repetition.When participants adopted the same thematic role order as in the prime, there were more NP-first responses following NP-first primes than following PP-first primes, and this tendency was enhanced by verb repetition.However, when participants adopted a different thematic role order from the prime, there were fewer NP-first responses following NP-first primes than PP-first primes, and this tendency was marginally amplified by verb repetition.
A second analysis examined whether and how the constituent structure of the prime interacted with argument structure congruence in constituent structure selectin in the target, and whether constituent structure priming occurred when a different argument structure was selected in the target.The verb-argument structure account assumes that the constituent structure of the prime should persist in the target even with a different argument structure from the prime is adopted in the target.The analyses thus included prime constituent structure, argument structure congruence and verb repetition as fixed effects.Fig. 14 reports the means for this analysis (see Appendix 13.15 for count data), and Table 8 (2) summarizes the results.The analyses revealed an interaction between prime constituent structure and argument structure congruence.Simple effects revealed that when the prime and target shared the argument structure, there were more NP-first responses following NP-first primes than following PP-first primes.However, there was no effect of prime constituent structure when the prime and target differed in argument structures: NP-first primes did not reliably elicit NP-first targets relative to PP-first primes.

Summary and discussion of Experiment 4
In summary, Experiment 4 yielded several important findings.First, both the argument structure and thematic role order of the prime persisted from what-cleft primes to what-cleft targets.Whilst both effects persisted in the absence of one another, these effects were related.Participants were more likely to repeat the argument structure of the prime when they adopted the same thematic role order as in the prime, and they tended to repeat the thematic role order of the prime more often with the same argument structure as the prime.Second, both effects were enhanced by verb repetition, which appears to have had a larger boost effect on argument structure persistence (14.9 %) compared to thematic role order persistence (6.3 %), suggesting that verbs may be more strongly associated with argument structures than with thematic role orders.We should also note that when the primed argument structure was reused in the target, it boosted thematic role order persistence by 19.1 %, which is substantially larger than the verb boost effect on thematic role ordering, indicating that the thematic role order of the prime may be more strongly associated with the argument structure of the prime than with the verb.This might explain why the model that included the interaction between thematic role order and argument structure congruence showed no verb repetition × thematic role order interaction.
Third, while the argument structure and thematic role order of the prime could persist separately from each other, the prime constituent structure (NP-first vs. PP-first) did not exert an independent influence on structural choice; constituent structure persistence was completely dependent on both argument structure persistence and thematic role order persistence.When participants adopted the same thematic role order as in the prime, more NP-first targets were observed following NP-first primes than following PPfirst primes.In these cases, NP-first primes and targets shared the argument structure, whereas PP-first primes featured different argument structures.When participants adopted a different thematic role order from the prime, we found no evidence for constituent structure persistence; there were fewer NP-first targets and more PP-first targets following NP-first primes than following PP-first primes.This was because PP-first primes, not NP-first primes, shared their argument structure with NP-first targets.Moreover, when participants repeated the argument structure of the prime, NP-first primes led to more NP-first targets; NP-first primes then shared the thematic role order with NP-first targets.When participants selected a different argument structure, however, NP-first primes did not elicit NP-first targets; NP-first primes and targets differed in thematic role order as well as argument structure.Hence, the results indicate that constituent structure can persist only when both the argument structure and thematic role order of the prime simultaneously persist in the target.
All in all, these results are compatible with the interactive structure building account, which proposes that (1) argument structure priming and thematic role order priming interact; (2) verb repetition can enhance both thematic role order priming and argument structure priming by increasing the probability of the primed thematic role order or argument structure being repeated for the same construction as in the prime; (3) the persistence of constituent structure arises from the combination of thematic role order persistence and argument structure persistence; (4) the constituent structure of the prime will not persist when the prime and target differ in argument structure or thematic role order, whereas argument structure and thematic role order can persist in the absence of its counterpart persisting.In contrast, the results were incompatible with the verb-argument structure account, which posits that (1) verb repetition should enhance argument structure priming but not thematic role order priming; (2) the likelihood of argument structure priming should be independent of the persistence of thematic role orders; (3) the constituent structure of the prime should persist separately from the argument structure of the prime.

Experiment 5: From What-Cleft Prime To Non-Cleft Target
The aim of Experiment 5 was to examine the representations underpinning verb boost effects further.According to the interactive structure building account, these effects stem from a general inclination to select structural features based on their previous cooccurrence.When the verb is repeated alongside its associated construction node, the co-activation of the repeated verb and construction nodes in the target boosts the activation of the associated thematic role order and argument structure in the prime.This leads to a higher likelihood of these structural features persisting together with the verb and construction in the target.However, in cases where the prime and the target involve distinct constructions, the likelihood of a verb boost effect is expected to decrease.This is because the links between the structural features (i.e., argument structure and thematic role order) and the target construction have not been primed.Consequently, even though the primed argument structure and thematic role order might gain some activation from their connections to the repeated verb, they cannot receive enhancements from the target construction node.Therefore, verb repetition is less likely to enhance structural persistence across different constructions compared to cases where the construction is repeated from the prime.However, there is an alternative possibility: Verb boost effects may instead solely stem from the links connecting the verb and the argument structure or thematic role order.If this is the case, verb repetition would enhance structural persistence regardless of whether the construction is repeated from the prime to the target.To test these hypotheses, Experiment 5 examined the structural choice in non-cleft targets following what-cleft primes.

Participants
An additional 80 participants were recruited from the same population as before.

Materials, design and procedure
These were the same as Experiment 4, except that the target prompts began with the subject noun (agent) followed by the verb, inducing declarative completions, as in (10) in Table 5 (see Appendices 13.7 & 13.9).

Scoring
We scored if participants produced a theme-location response (n = 2013) or a location-theme response (n = 500) when producing a non-cleft (declarative) target using the same coding scheme as before.We included cases where the preposition in was used for the verb slather (n = 8) or the verb drizzle (n = 2) in the location-theme order responses.Cases were excluded where participants had chosen an implausible alternative word in the prime (n = 13) or included an additional preposition in the prime trials (e.g., What the lawyer stuffed with the banknotes were into briefcases) (n = 23); only one argument was mentioned (n = 1); participants used full of instead of a preposition for the verb stuff or pack (n = 2) or omitted the preposition (n = 3); they conveyed a non-target meaning (n = 1) or used a non-target structure (n = 1) or used a non-target verb (n = 2); or there was a technical error (n = 1).In total, 47 trials (1.8 %) were excluded from analyses.

Results and discussion
Fig. 15 reports the percentages and standard errors of location-theme (with-structure) responses relative to theme-location (locative structure) responses (see Appendix 13.16 for count data).As in Experiments 1-3, participants generally produced theme-location orders more frequently than location-theme orders (as confirmed by the negative intercept in Table 9).This contrasts with Experiment 4, where location-theme orders (theme-focused) were generally chosen more frequently in what-cleft targets, demonstrating an impact of the use of what-cleft construction in the target on thematic role ordering.Importantly, as summarized in Table 9, the analyses, including prime thematic role order (location-theme order vs. theme-location order), prime argument structure (withstructure vs. locative structure) and verb repetition (repeated vs. not-repeated) as fixed effects, revealed a main effect of thematic role order as well as a main effect of argument structure.Participants produced more location-theme order simple declarative sentences K. Fukumura and F. Yang (with-structures) when what-cleft primes had location-theme orders (21.5 %, SE = 1.2 %) rather than when they had theme-location orders (18.3 %, SE = 1.1 %), demonstrating thematic role order priming from what-cleft primes to non-cleft targets.Moreover, participants produced more location-theme orders (with-structures) following with-structure primes (22.7 %, SE = 1.2 %) than following locative structure primes (17.2 %, SE = 1.1 %), indicating the persistence of argument structure from what-cleft primes.Critically, verb repetition did not enhance the effect of prime thematic role order nor prime argument structure.The Bayes Factors (BF 01 ) for the prime argument structure × verb repetition and the prime order × verb repetition interactions were 43.51 and 28.87, respectively, showing strong support for the null hypotheses.
The results of Experiment 5 revealed that argument structure and thematic role order in what-cleft primes influence function assignment and thematic role ordering in non-cleft sentences.Importantly, neither effect was reliably enhanced by verb repetition, indicating that verb boost effects do not simply arise from the co-occurrence of the verb and the argument structure or from the cooccurrence of the verb and the thematic role order in the prime.If this were the case, the repetition of the verb should have enhanced argument structure persistence and thematic role order persistence from what-cleft primes to non-cleft targets.In contrast, the results align with the interactive structure building account, which suggests that verb repetition boosts structural priming due to the cooccurrence of the verb and either an argument structure or thematic role order selected in the particular construction of the prime.When primes and targets involve different constructions, the link between the structural feature and the target construction is not primed.This reduces the likelihood of verb repetition reliably enhancing the probability of the primed structural feature persisting with the verb, compared to cases where the construction node is also repeated alongside the verb; in such cases, the primed structural feature can receive boosts not only from the verb but also from the construction node.
Additionally, the impact of the argument structure of what-cleft primes on argument structure selection in non-cleft targets (5.5 %) (Experiment 5) was smaller compared to that in what-cleft targets (21.6 %) (Experiment 4), Estimate = 0.28, SE = 0.04, z = 6.39,p < .001(see Appendix 13.18 for full results).Similarly, the effect of the thematic role order of what-cleft primes on thematic role ordering in non-cleft targets (3.2 %) (Experiment 5) was reduced in contrast to what-cleft targets (14.7 %) (Experiment 4), Estimate = 0.40, SE = 0.05, z = 7.43, p < .001(see Appendix 13.19 for full results).This cannot be attributed solely to the absence of verb boost effects in Experiment 5.The effects of prime argument structure and thematic role order were smaller in Experiment 5 than in Experiment 4 in both the verb-repeated condition, Estimate = 0.43, SE = 0.07, z = 6.59, p < .001(argument structure); Estimate = 0.47, SE = 0.09, z = 5.06, p < .001(thematic role order), and the not-repeated condition, Estimate = 0.17, SE = 0.06, z = 2.97, p = .003(argument structure); Estimate = 0.32, SE = 0.07, z = 4.42, p < .001(thematic role order).When what-cleft primes were followed by non-cleft targets, interactive priming between thematic role ordering and argument structure selection was not reliable, because the thematic role order and the argument structure of the prime were not primed in association with the non-cleft target construction.Hence, the persistence of a thematic role order in a non-cleft target did not reliably enhance the likelihood of argument structure persistence in the non-cleft target, and vice versa.

Experiment 6: From It-Cleft Prime To Non-Cleft Target
Experiment 5 demonstrated the persistence of thematic role order from what-cleft primes to non-cleft targets.One possible interpretation of this finding is that the persistence from the what-cleft prime to the non-cleft target is driven by thematic emphasis rather than by thematic role order persistence.This interpretation assumes that the alternation between theme-location and locationtheme orders in declarative sentences is largely influenced by the persistence of thematic emphasis (Bernolet et al., 2009).Specifically, a location-theme order what-cleft prime such as What the assistant loaded into the lift was the equipment can elicit location-theme orders in non-cleft declarative sentences such as The man loaded the bus with the luggage because in English, the thematic role of a newly introduced or emphasized entity typically follows that of a previously introduced or de-emphasized entity (e.g., Haviland & Clark, 1974;Givón, 1984;Halliday, 1970;Quirk et al., 1985).Hence, in both what-cleft primes and declarative targets, the last-mentioned thematic role (i.e., the theme role in the above examples) is emphasized.
Experiment 6 thus aimed to determine the impact of thematic emphasis within it-cleft primes on thematic role ordering in declarative targets.If thematic emphasis in the prime sentence influences the ordering of thematic roles in declarative constructions, we should observe an opposite effect of thematic role ordering from it-cleft constructions on non-cleft declarative targets.In it-cleft constructions, it is the first-mentioned role that is emphasized.As shown in Table 10, theme-focused it-cleft sentences like (11a, 11b, 11e, 11f) place the theme role in the first position, while location-focused it-cleft sentences like (11c, 11d, 11g, 11h) place the location role first.Suppose that these it-cleft sentences were presented as primes, followed by non-cleft declarative target sentences, such as (12a and 12b).If thematic emphasis indeed drives the alternation of theme-location in declarative sentences, theme-focused it-cleft primes (11a, 11b, 11e, 11f) should elicit more location-theme order responses, such as The man loaded the bus with the luggage (12a), than location-focused it-cleft primes (11c, 11d, 11g, 11h) should.
As shown in Table 10, all the it-cleft primes mentioned either the theme role or the location role first, whereas in the non-cleft declarative targets, the agent role was mentioned first.Hence, it was unlikely that these it-cleft primes would influence thematic role ordering in the non-cleft target sentences.However, the argument structures of it-cleft primes should persist into non-cleft targets because argument structures do not carry ordering information.Specifically, prime sentences (11a, 11c, 11e, 11g) involve withstructures, whereas sentences (11b, 11d, 11f, 11h) involve locative structures.Therefore, with-structure primes should elicit more location-theme orders (12a, with-structures) as opposed to theme-location orders (12b, locative structures) than locative structure primes should.Critically, it-cleft primes and non-cleft targets involve different constructions.Hence, according to the interactive structure-building account, similar to Experiment 5, verb repetition is unlikely to boost argument structure persistence from it-cleft primes to non-cleft targets.

Participants
A further 80 participants were recruited from the same population as before.An additional participant was recruited but excluded from analyses due to their failure to follow the instructions in the prime trials.

Materials, design, and procedure
These were the same as in Experiment 5, except that prime sentences involved it-clefting, shown in Table 10 (11a-11h) (see Appendix 13.8).We manipulated prime argument structure (with vs. locative structure), prime thematic role order (emphasis) (locationagent-theme/location-focused vs. theme-agent-location/theme-focused) and verb repetition (repeated vs. not-repeated) as within-

Table 10
Example prime sentences and target sentences in Experiment 6. Note.Example completions in brackets.
K. Fukumura and F. Yang subjects and within-items variables, leading to a 2 (argument structure) × 2 (verb repetition) × 2 (thematic role order) repeated measure design.The target events and prompts were the same as in Experiments 4 & 5 (Appendix 13.9).The 32 experimental items and 70 filler items were distributed across four lists as before.

Scoring
We scored if participants produced a theme-location response (n = 1922) or a location-theme response (n = 600) when producing a non-cleft target as before.We excluded cases (n = 17) where the response include only one thematic role (n = 4); they used full of instead of a preposition (n = 3) or in or for in location-theme responses (n = 5) or omitted the preposition (n = 1); the responses conveyed non-target meanings (n = 1), contained an exchange error (n = 1) or an additional unnecessary preposition (n = 1) or verb (n = 1).We also excluded cases (0.8% of total 2560 responses, n = 21) where participants had chosen an implausible alternative word in the prime (n = 15) or included a redundant preposition (e.g., It was on the arm that the tourist rubbed with lotion; It was in the boxes that the babysitter crammed with the toys) (n = 3), or additional words that altered the event meanings (e.g., It was a spoon on the table that the baby splashed with the soup; n = 1) or there were technical errors in the prime trials (n = 2).In total, 38 trials (1.5 %) were excluded from analyses.

Results and discussion
Fig. 16 presents the percentages and standard errors of location-theme (with-structure) responses relative to theme-location (locative structure) responses by prime thematic role order (emphasis), prime argument structure and verb repetition (see Appendix 13.17 for count data).The analyses included prime thematic role order (emphasis) (theme-agent-location/theme-focused vs. location-agent-theme/location-focused, coded as 1 and 0) respectively), prime argument structure (with vs. locative structure, coded as 1 and 0 respectively) and verb repetition (repeated vs. not-repeated, coded as 1 and 0 respectively), which were all mean-centred and standardized.The results are summarized in Table 11.The analyses showed no main effect of thematic role order (emphasis) × verb repetition interaction, with the BF 01 for this interaction being 50.22,showing very strong support for the null hypothesis.In contrast, as in Experiment 5, there was a main effect of argument structure, with more location-theme order responses (with-structures) following a with-structure prime (27.0 %, SE = 1.3 %) than following a locative structure prime (20.6 %, SE = 1.1 %).This effect of prime argument structure was numerically larger when the verb was repeated (8.7 %) than when it was not repeated (4.1 %).However, the effect was only marginal and the BF 01 for this interaction was 11.33, providing support for the null hypothesis.
Thus, Experiment 6 provided no evidence that the thematic role order of it-cleft sentences primes theme-locative alternations in non-cleft sentences.This suggests that the persistence of thematic role order observed in Experiment 5 is unlikely to have resulted from the persistence of thematic emphasis.As Experiment 5, Experiment 6 did find an effect of argument structure priming, where the argument structure of the it-cleft prime influenced argument structure selection in non-cleft targets.This effect was marginally larger when the verb was repeated.While this might imply a certain degree of overlap between it-cleft constructions and declarative constructions, the Bayes Factor for this effect (BF 01 ) favored the null hypothesis.The findings are also in line with those of Experiment 5, where the persistence of both argument structure and thematic role order from what-cleft primes to non-cleft targets was not significantly enhanced by verb repetition.
According to the interactive structure building account, the magnitudes of thematic role order priming and argument structure priming should be larger when the construction is repeated from the prime to the target, regardless of verb repetition.In Experiment 1, both primes and targets employed declarative constructions.Therefore, we might anticipate a larger priming effect in Experiment 1 than in Experiments 5 and 6, where primes and targets featured different constructions.However, unlike in Experiments 5/6, argument structure and thematic role order were confounded in Experiment 1.This prevents us from directly comparing Experiment 1 with K. Fukumura and F. Yang Experiments 5/6, as we cannot discern to what extent the priming effect in Experiment 1 was influenced by the thematic role order, argument structure of the prime, or both.To enable valid comparisons between it-cleft primes and non-cleft primes, investigating a language that permits the independent manipulation of argument structure and thematic role order in declarative sentences would be necessary.Additionally, it is worth noting that Experiment 1 showed a verb boost effect, which was not present in Experiments 5-6.When verb repetition strongly cues the primed structure, the priming effect might decrease in conditions without verb repetition (cf., MacWhinney, 1987).Consistent with this possibility, in the not-repeated condition, the effect of prime was marginally smaller in Experiment 1 (5.1 %), where there was a verb boost effect, than in Experiments 2 and 3 (8.5 %), where there was no verb boost effect (see Appendix 13.11).
In summary, the results from Experiments 4-6 collectively indicate that verb repetition enhances structural priming by increasing the probabilities of repeating the primed argument structure and thematic role order within the same construction as in the prime; when the constructions in the prime and target differ, verb repetition is unlikely to enhance the priming effects.As discussed earlier, the interactive structure-building account proposes that a prime sentence like What the assistant loaded the lift with was the equipment not only activates links from the with-structure and the A-L-T order to the verb but also to the node representing the what-cleft construction (Wh-cleft node) in Fig. 7A.If the verb and the Wh-cleft construction node is subsequently repeated in the target and their activation then spreads through the network, the with-structure node and the A-L-T node receive strong activation due to the prior activation of their links to both the verb and the Wh-cleft node.This robustly increases the chances of the with-structure and the A-L-T order persisting when the verb and the what-cleft construction are repeated in the target.However, if the verb is repeated in a declarative target sentence, the with-structure node only gains a boost from the verb-argument structure link, insufficient to enhance the likelihood of the with-structure appearing in the declarative sentence.In Experiments 5 and 6, cleft primes were followed by non-cleft targets, whereas in Experiment 4, both primes and targets featured the same Wh-cleft constructions.This explains the absence of significant verb boost effects in Experiments 5 and 6, in contrast with their presence in Experiment 4.

Summary
The current study examined the representations underpinning structural priming, focusing on understanding the locus of the lexical boost effect.Three accounts were evaluated.The first two hypotheses were derived from the two-stage grammatical encoding account (e.g., Bock & Levelt, 1994), which posits a strict sequential process of grammatical encoding, with function assignment (involving the assignment of syntactic functions to thematic roles) preceding constituent assembly (determining the linear order based solely on abstract syntactic information rather than on thematic information).The first account, the verb-constituent structure account, assumes that verb selection is associated with the syntactic representations underpinning constituent assembly.On this view, verb repetition is not expected to enhance structural priming with spray-load verbs in declarative sentences, where the structural variants share same constituent structure.On the other hand, the second account, the verb-argument structure account, proposes that verb selection is associated with the selection of argument structures, underpinning thematic role-syntactic function mappings during function assignment.In this view, verb repetition can enhance structural priming with spray-load verbs, increasing the probability of repeating the primed argument structure in the target.However, because function assignment precedes constituent assembly, the persistence of argument structure should be independent of whether the order of constituents (and the thematic role order) is also maintained in the target or not, and the constituent structure of the prime can persist separately from its argument structure and thematic role order.
The third account is the interactive structure-building account.According to this account, structural building proceeds in a single stage, where the selection of a thematic role order, which drives linear ordering, is selected along with an argument structure and other structurally relevant information, such as the verb's semantic and syntactic properties and a sentence construction.This is interactive, because combinations of various structural features are affected by their co-occurrence in prior utterances.Specifically, when the verb and the construction are repeated from the prime in the target, it increases the probability of repeating other associated features, namely, the thematic role order and the argument structure of the prime.Likewise, when the argument structure of the prime is repeated in the same construction in the target, it increases the likelihood of the associated thematic role order being repeated, and vice versa.
Experiments 1-3 revealed that the repetition of verbs, but not nouns fulfilling the theme or location roles, produced a larger structural priming effect on the theme-location alternation in declarative sentences, providing evidence against the verb-constituent structure account.Using what-cleft constructions for both primes and targets, Experiment 4 extended this finding by demonstrating that verb repetition leads to larger likelihoods of argument structure persistence and thematic role order persistence.In addition, the persistence of argument structure was found to interact with the persistence of thematic role order.When participants repeated the same thematic role order from what-cleft primes to what-cleft targets, they were more likely to repeat the argument structure of the prime, and vice versa.These findings were in line with the interactive structure-building account, according to which verb selection is associated with both thematic role ordering and argument structure selection, which also interact with each other.In contrast, the results were inconsistent with the verb-argument structure account, according to which verb repetition should only enhance the persistence of argument structure, and the persistence of argument structure should be unrelated to the persistence of thematic role order persistence.Further evidence against the verb-argument structure account came from the finding that the constituent structure of the prime does not persist independently from the thematic role order nor the argument structure of the prime, whereas both the thematic role order and argument structure of the prime perseverate in the absence of its counterpart persisting.These findings suggest that abstract constituent structure is not a determining factor in structural choices beyond what is determined by the argument structure or thematic role order of the prime.Experiments 5 and 6 further examined whether the verb boost effects observed in Experiments 1 and 4 would be diminished when primes and targets involve different constructions, as suggested by the interactive structure-building account.Experiment 5 found that both the thematic role order and the argument structure of what-cleft primes can influence structural alternations in non-cleft declarative targets, indicating that thematic role orders and argument structures are shared across different constructions.However, these priming effects were not enhanced by verb repetition.Experiment 6 corroborated this finding: While the argument structure of an it-cleft prime persisted into a non-cleft target, verb repetition did not reliably boost this effect, despite a numerical tendency pointing in this direction.Moreover, we observed no evidence that thematic emphasis in it-cleft primes can influence structural choice in non-cleft targets, indicating that thematic role order persistence from what-cleft primes to non-cleft targets is unlikely to be due to the persistence of thematic emphasis.Instead, the effect is likely to be driven by the persistence of abstract thematic role order shared across different constructions.

Implications for theories of structural priming
All in all, the current findings have significant implications for theories of structural priming.We showed that structural priming is primarily driven by the persistence of argument structures that underpin thematic role-function mappings as well as the persistence of thematic role orders that underpin thematic role-position mappings.Importantly, both priming effects are enhanced by verb repetition and appear to interact.In contrast, the persistence of constituent structure represents a combined effect of these two forces.We found no evidence that abstract syntax, more specifically, the orders of syntactic categories and functions, plays an independent role in driving structural alternations.While abstract syntax (e.g., whether a verb is followed by a noun phrase representing a direct object or a prepositional object phrase) might be represented, such information does not appear to drive structural alterations.This contrasts with the residual activation model (Pickering & Branigan, 1998;Pickering et al., 2002), which focused on the persistence of abstract constituent structure as the driving force behind structural priming and verb boost effects.
Moreover, the results from our experiments suggest that verb boost effects reflect general interactive structure-building processes, wherein various structural features, including thematic role orders and argument structures, are selected interactively, guided by their prior co-occurrence and associations.Specifically, when two structural features from a prime are repeated in a target, this increases the likelihood of a third feature that co-occurred with them in the prime being repeated in the target.For instance, when both the verb and a what-cleft construction are repeated from the prime, the probabilities of the argument structure and the thematic role order also persisting along with them increase.Similarly, when the primed thematic role order is repeated in the same what-cleft construction, it enhances the probability of the primed argument structure being repeated in that construction, thus reinforcing the overall structural persistence.However, it is important to note that interactive priming does not necessarily require the repetition of all the structural features that occurred in the prime.In Experiment 4, a verb boost effect on argument structure persistence was observed independently of whether the thematic role order was also repeated or not.This finding suggests that tracking the co-occurrence of all four features (verb, argument structure, construction, and thematic role order) is not a prerequisite for boosting structural persistence.
The current findings, indicating that verb repetition, but not noun repetition, enhances theme-location alternations, provide evidence against hypotheses suggesting that structural priming can be enhanced not only by verb repetition but also by noun repetition, given that both are content words (Chang et al., 2006;Reitter et al., 2011;Scheepers et al., 2017), or that the theme-location alternation functions differently from the dative alternation (Chang et al., 2006).In contrast, the current findings align with recent studies (Carminati et al., 2019;Kantola et al., 2023;Huang et al., 2023;Van Gompel et al., 2023) which found that only verb repetition enhances structural priming involving the dative alternation.However, these results were interpreted based on the traditional residual activation model (Pickering & Branigan, 1998;Pickering et al., 2002), which assumes that verb repetition enhances the persistence of abstract syntax, without involving thematic role information.The current study's demonstration of verb boost effects on both thematic role ordering and argument structure selection provides evidence against this assumption.
Although our results are compatible with Fukumura and Zhang (2023), who showed that syntactic priming of noun phrases is specifically enhanced by head noun repetition, in their study, conceptual category order priming was enhanced not only by noun repetition but also by the repetition of adjectives, which are non-heads.This leads to the question of why noun repetition does not enhance thematic role order priming.It is important to recognize that thematic roles are not inherent features of the concepts bearing those roles.For instance, the role of an object, such as car, can be a theme (e.g., being loaded into a truck) or a location (e.g., being loaded with boxes), depending on its function within the action or event, which is expressed by the verb's meaning.Therefore, the representation of thematic role orders is more closely tied to the verb's meaning than to the noun concepts.This contrasts with adjective conceptual category orders.Conceptual categories like color and pattern are the attributes of individual adjective concepts; for instance, BLUE is a color and STRIPED is a pattern.Hence, adjective category orders are associated with individual adjective concepts because the category orders must correspond to the order of the individual category tokens, the activation of which enables the selection of those categories.
The current findings have significant implications for the ongoing debates surrounding the temporal dynamics of verb boost effects.Researchers have proposed that the lexical boost effect is contingent on explicit short-term memory, yielding transient effects, whereas abstract priming-structural priming devoid of lexical repetition-stems from implicit learning and thus exhibits longer-lasting effects (e.g., Bock & Griffin, 2000;Chang et al., 2006;2012).However, although lexical repetition can enhance the rate of recall of the primed structure (Zhang et al., 2020), explicit memory may also contribute to abstract priming (Bernolet et al., 2016;Zhang et al., 2020; but see Bock et al., 1992;V. Ferreira et al., 2008).Evidence from aphasia studies is somewhat inconsistent.Man et al. (2019) observed no lexical boost effect in aphasic patients with lexical-semantic short-term memory deficits.On the other hand, Yan et al. (2018) found a verb boost effect in aphasic individuals with explicit short-term memory deficits, without a correlation between the magnitude of the lexical boost effect and short-term memory deficits.The temporal durations of different priming effects are difficult to estimate, as they might be influenced by various factors such as experimental tasks, the type of stimuli, the strength of the association, the level of exposure, and others.However, there is evidence suggesting that abstract priming is longer-lasting compared to the lexical boost effect (Bock & Griffin, 2000;Hartsuiker et al., 2008;Branigan & McLean, 2016; see also Bernolet et al., 2016).The difference in duration may not necessarily stem from the lexical boost effect's reliance on explicit memory, as posited by certain theories (e.g., Bock & Griffin, 2000;Chang et al., 2006;2012).Instead, the lexical boost effect may rely on short-term associative memory traces, which are not necessarily explicit but decay faster than abstract priming, which relies on adjustments in the base-level activation of the primed structure (Malhotra et al., 2008).
Another reason why the lexical boost effect may decay faster than abstract priming could be attributed to the complexity of the representations underpinning the lexical boost effect.The current findings suggest that whereas abstract priming relies on the persistence of a single structural feature, either the primed thematic role order or argument structure, the lexical boost effect involves more features and interconnected links.Specifically, verb boost effects rely not only on the verb, the primed structural feature (thematic role order or argument structure) and the construction, but also the links connecting them.If the memory traces for any of these nodes and links become disrupted, the occurrence of a verb boost effect is unlikely.The same principle applies to interactive priming between argument structure and thematic role order, which also relies on multiple features and interconnected links.Consequently, both interactive priming and the lexical boost effect may have shorter-lasting durations compared to abstract priming, owing to a higher likelihood of at least one feature or link being disrupted during intervening trials, as compared to the persistence of an abstract priming effect.Thus, while associations between multiple features can indeed form through exposure to a single prime sentence, one-shot exposures might not be sufficient to establish lasting connections between structural features.A promising avenue for future research lies in the testing of these hypotheses, investigating the temporal dynamics of the priming effects observed in the current study, with a specific focus on examining the relative durations of the persistence of argument structure, thematic role order, their interaction, and the verb boost effect.
The current study highlights the role of 'construction' as an additional structural element in moderating the verb boost effect.In this study, prime sentences appeared in three distinct constructions: What-cleft constructions, it-cleft constructions, and declarative constructions.We have assumed that construction nodes encompass information such as illocutionary force (e.g., making a request, asking a question, giving a command, making a statement) (cf.Searle, 1979), along with global syntactic information (e.g., forms of different cleft constructions) associated with different constructions.What other information might be encoded in these construction nodes?The findings from Chang et al. (2015) suggest that these construction nodes might include information about verb position.They reported that the likelihood of German ditransitive priming increased not only due to verb repetition but also with the repetition of verb position from the prime to the target.In contrast to English, in German, verb position in declarative sentences can vary based on tense and aspect.In their study, prime sentences featured either a simple past tense (e.g., Die Geliebte sendete dem Kavalier die Rechnung, The mistress sent the gentleman the bill), where a finite verb (sendete) occupied the second position, or a present perfect tense (Die Geliebte hat dem Kavalier die Rechnung gesandt, The mistress has sent the gentleman the bill), where an auxiliary (hat) appeared in the second position while the participle (gesandt) was positioned at the end of the sentence.The targets were always in the simple past tense, so they had the finite verb in the second position.Chang et al. found that the prime structure (double object dative vs. prepositional dative) persisted in the target, regardless of whether the verb position in the prime was the same (i.e., second position) or different (i.e., final position), or whether the verb was repeated or not repeated.However, the verb boost effect was reliable only when the verb occurred in the second position in the prime, as was the case in the target.Hence, these findings indicate that the declarative construction node may be represented differently depending on verb position in German.This suggests that the construction nodes examined in the current study, representing declarative sentences, what-cleft sentences, and it-cleft sentences, might also include information relating to the relative position of the verb in the sentence.
Until recently, closed-class words have been assumed to play no role in structural priming (e.g., Branigan & Pickering, 2017).This is because studies have shown, using dative alternations, that structural priming is unaffected by preposition overlap between prime and target; a prepositional object sentence can be elicited by another prepositional object sentence regardless of whether the preposition is the same (as in The student brought a cake to the teacher) or different (as in The student brought a cake for the teacher) (Bock, 1989;Chang et al., 2003;Ziegler & Snedeker, 2018).However, Ziegler et al. (2019) recently reported that the airplane is landing by the control tower primed passive sentences, as reported in Bock and Loebell (1990), but the airplane is landing next to the control tower did not prime passives.
One possibility is that lexical priming of the preposition by influenced the rates of passive responses.On this view, structural priming in the current study could be explained by lexical repetition of prepositions: While location-theme orders almost always involve with, theme-location orders can take different prepositions.Chang et al. (2003) argued against this possibility, pointing out that in their study, locative variants (theme-location orders) primed locative variants just as effectively as with-structure (location-theme) primes elicited with-structure responses.If the lexical priming of with is the driving force behind the structural alternations, structural priming should be more pronounced following with-structure primes than following locative structure primes.In the current study, with variants (31 %) were repeated less frequently than locative variants (80 %), making it unlikely for the lexical priming of with to be the source of the priming effects.
An alternative possibility is that the prepositions are part of the argument structures, as assumed under the interactive structurebuilding account (cf.Pinker, 1989;Rappaport & Levin, 1988).Specifically, the argument structure for with variants involves the preposition with, whereas the argument structure for locative variants involves locative prepositions.Likewise, the preposition by is represented as part of the argument structure underlying a passive sentence, with by specifying the agent role of the prepositional object.Hence, the selection of an argument structure for a passive form was influenced by the prior encounter of by in The airplane is landing by the control tower in Ziegler et al.'s study.This does not conflict with the finding mentioned above, where for-prepositional object dative primes elicit to-prepositional object datives to the same extent as to-prepositional object dative primes do.Both prepositions, for and to, can specify the role that implies receiving or possessing the theme due to the action denoted by the verb (Pinker, 1989).Therefore, both to-datives and for-datives may be construed to have equivalent thematic role orders and argument structures, whereby both to and for specify a similar role, as in the case of the locative in the theme-location alternation, which can take different location prepositions (e.g., on, in, onto) to specify the location role.
Although the current study investigated thematic role ordering, an important question is whether the current findings generalize across different types of word order priming reported in the literature.For instance, Hartsuiker and Westenberg (2000) demonstrated that in Dutch, language users produce auxiliary-final responses more often following an auxiliary-final prime such as omdat de weg geblokkeerd was (because the road was blocked) than following a participle-final prime such as omdat de weg was geblokkeerd.The findings from Konopka and Bock (2009) suggest that this type of word order priming effect can be enhanced by verb repetition.In their study, the position of the particle in phrasal verbs such as broke down the door or broke the door down persisted from the prime to the target, and the effect was larger when the verb and particle were both repeated from the prime to the target.Moreover, the probability of this verb boost effect might also be contingent on the repetition of the construction.Momma (2022) recently showed that the priming of an optional English complementizer that, as in The mechanic mentioned (that) the car could use a tune-up (V.Ferreira, 2003), can be boosted by verb repetition.Momma found that participants were more likely to mention that in their responses after encountering prime sentences containing the complementizer.Interestingly, the verb boost effect was found only when the prime and target either both involved or neither involved a filler-gap dependency (e.g., Who did the director announce (that) he nominated?),while the presence or absence of that was primed across different constructions.Momma proposed that the priming of optional that occurs due to the persistence of the mapping between the use of that and its associated conceptual feature, whereas the verb boost effect is driven by the residual activation of the tree structure in which that is embedded.This is consistent with our proposal that verb selection is associated with the argument structure and thematic role order implemented in a specific construction.

Implications for theories of sentence production
The findings from the present study challenge the fundamental assumptions of the two-stage grammatical encoding account, which posits a strictly serial process of grammatical encoding and the independence of constituent structure from thematic information.However, it may be argued that the two-stage account has received support not only from the structural priming literature but also from other domains of language production research.For instance, the two-stage account originates from Garrett's (1975Garrett's ( , 1990) ) analysis of speech errors, suggesting that word exchange errors (such as I left the briefcase in my cigar, where cigar in my briefcase was intended) occur at the functional stage, while sound exchange errors (such as easy enoughly when easily enough was intended) are associated with a positional stage, with no interaction between these stages.Similarly, some studies on verb agreement errors have provided evidence supporting the two-stage model.For instance, Vigliocco and Nicol (1998) reported that verb agreement errors occur regardless of linear proximity to the attractor noun (*The helicopter for the flights are safe vs. *Are the helicopter for the flights safe?).This finding was regarded as evidence in favor of the two-stage grammatical encoding account, which posits that hierarchical relations relevant to verb agreement are computed prior to constituent assembly (Bock & Miller, 1991;Bock & Cutting, 1992;Garrett, 1980).
However, these interpretations have been controversial.Dell and colleagues (Dell, 1986;Dell & Reich, 1981) examined other error patterns that indicate interactions between different types of representations, which led to the proposal of a word production model that allowed for interactive parallel processing.As for verb agreement errors, other studies have shown that linear proximity can influence verb agreement processes (Franck et al., 2006;Haskell & MacDonald, 2005;Santesteban et al., 2013).For instance, Haskell and MacDonald (2005) found that when producing sentences with disjunctive subject noun phrases in English, participants tend to make the verb agree with the number of the noun nearest to the verb both when the verb follows (e.g., The clock or the horses is/are red) and when it precedes the noun (e.g., Is/Are the horses or the clock red?).Similarly, Santesteban et al. (2013) found that speakers of Basque, a language in which the verb agrees in number with both the subject and object, exhibited a higher frequency of verb agreement errors when the noun was positioned further away from the verb (i.e., the subject in subject-object-verb word order sentences, where the subject is distant from the verb, and the object in object-subject-verb word order sentences, where the object is distant from the verb).It was suggested that syntactic agreement is made over a linearly specified representation, possibly during the mapping process between semantic and syntactic representations.Such a proposal aligns with the interactive structure-building account, according to which the selection of an argument structure underpinning function assignment operates in parallel with the ordering of thematic roles, which allows for the possibility that linear order can interfere with agreement processes.
In addition to speech errors and verb agreement errors, some findings on accessibility effects on sentence production have been interpreted as supporting the two-stage account.For instance, some studies (Bock & Warren, 1985;McDonald et al., 1993) indicated that conceptual accessibility due to animacy or imageability influences word order via function assignment, rather than influencing linear order directly.Using a sentence recall task, McDonald et al. (1993) reported that participants were more likely to recall sentences by assigning the subject role to an animate entity rather than to an inanimate entity (e.g., The child was soothed by the music rather than The music soothed the child).This animacy effect was not observed when the subject comprised a conjoined noun phrase where an animate and an inanimate noun shared the subject function (e.g., The key and the manager were nowhere to be found vs.The manager and the key were nowhere to be found).These findings led to the suggestion that conceptual accessibility impacts structural choice through function assignment rather than through constituent assembly, aligning with the two-stage grammatical encoding account (e.g., Bock et al., 1992;Bock & Warren, 1985).
However, those interpretations have been called into question following findings that the higher accessibility of an entity due to givenness (V.Ferreira & Yoshita, 2003;Prat-Sala & Branigan, 2000) or animacy (Branigan et al., 2008;Kempen & Harbusch, 2004;Prat-Sala & Branigan, 2000;Tanaka et al., 2011;cf. Christianson & F. Ferreira, 2005;Van Nice & Dietrich, 2003) can influence linear ordering as well as function assignment in freer word order languages.For example, Prat-Sala and Branigan (2000) demonstrated in Spanish that when the referent of the noun with the patient role is more accessible due to its prior mention or animacy status, the likelihoods of producing patient-initial actives (A la mujer la atropelló el tren, to the woman she ran over the train, meaning that the train ran over the woman) as well as patient-initial passives (La mujer fué atropellada por el tren, The woman was run over by the train) increase, relative to all alternatives, including agent-initial actives (El tren atropelló a la mujer, The train ran over the woman).Patientinitial actives and agent-initial actives differ in their linear order, but they share the function assignment.The finding thus challenges the assumption of the two-stage models, which suggests that conceptual accessibility affects structural choice exclusively through function assignment.Moreover, in a series of experiments, McDonald et al. (1993) found that when conjoined noun phrases were presented in isolation (i.e., they were not a part of a sentence), participants did tend to recall animate nouns before inanimate nouns (e. g., bachelor and spoon rather than spoon and bachelor), suggesting that animacy can influence linear ordering for conjoined noun phrases.Similarly, conceptual accessibility has been shown to impact the order of both conjoined and non-conjoined adjectives (Fukumura, 2018), with a general preference for color-before-pattern orders (e.g., green and spotted bow rather than spotted and green bow).Therefore, it is not the case that conceptual accessibility solely influences linear order through function assignment, contrary to the two-stage grammatical encoding account.
Therefore, we propose that prior interpretations based on the two-stage model should be re-evaluated in light of the current findings and other available evidence.Specifically, the interactive structure-building account can offer an alternative interpretation of accessibility effects.According to this view, accessibility influences linear ordering at the conceptual level, impacting the selection of an abstract conceptual order, which interfaces with syntactic selection.Fukumura and Zhang (2023) first demonstrated the presence of such interactions, showing speakers repeat a primed relative clause (such as green bow that's spotted as opposed to a simpler alternative structure such as green spotted bow) more often when they also repeat the primed adjective order (i.e., a color-before-pattern order) than when they use a different adjective order (e.g., a pattern-before-color order as in spotted bow that's green) from the prime.In the present study, we observed that in sentence production, thematic role ordering can interact with the selection of argument structure.This leads us to propose that accessibility-based structure building occurs because speakers choose a syntactic structure in conjunction with an abstract conceptual order, such as a thematic role order and an adjective category order, so that a most accessible concept can be placed early, thereby promoting incremental production processes.Thus, by positing that speakers formulate a syntactic structure in tandem with a conceptual order, the interactive structure-building account can explain both accessibility-based grammatical encoding and structural priming in a unified production process.

Conclusions
The current research provides novel insights into interactions between thematic and syntactic representations in sentence production.Overall, our findings demonstrate that the argument structure underpinning thematic role-syntax mappings as well as the order of thematic roles of the preceding sentence can independently influence structural choice across different constructions, while the interaction between them strengthens the magnitude of persistence: The likelihood of argument structure persistence in the target is higher with the same thematic role order than with a different thematic role order from the prime, reflecting a tendency to repeat the same argument structure for the same thematic role order selected in the same construction as in the prime.Verb boost effects arise from the same interactive mechanism, wherein the co-occurrence of different structural features in the preceding sentence influences structural combinations in the upcoming utterance.Thus, verb repetition increases the probabilities of the primed thematic role order as well as the primed argument structure being repeated for the same type of construction.Importantly, while both the argument structure and thematic role order of the preceding sentence can independently influence the structural choice across different constructions, the abstract constituent structure of the preceding sentence cannot be primed separately from its thematic role nor argument structure.
These results are incompatible with traditional two-stage grammatical encoding accounts, and we propose the interactive structure-building account as an alternative.While two-stage accounts have been adopted to explain findings in other areas of language production research, we propose that those interpretations should be re-evaluated in light of the new evidence.We must also acknowledge that the current proposed account and hypotheses require further refinement and revision through future investigations, encompassing not only the domain of structural priming but also other aspects of human language production processes.By stimulating such future investigations, we believe that our work can make a significant contribution to advancing the understanding of structural priming and human sentence production processes.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

3
Fig. 3. Procedure and example prime and target display.3

Fig. 4 .
Fig. 4. Percentages and standard errors of Location-Theme (relative to Theme-Location) responses in Experiment 1.

Fig. 6 .
Fig. 6.Percentages and standard errors of Location-Theme (relative to Theme-Location) responses in Experiment 3.
cleft node.Consequently, when the A-L-T order and the what-cleft construction are repeated in the target, the probability of repeating a with-structure alongside the A-L-T order and the what-cleft construction increases.Similarly, if speakers intend to repeat the with-structure in another what-cleft sentence node in the target, it leads to an enhanced activation of the with-structure node alongside the Wh-cleft node in the target.When the activation from these

Fig. 7 .
Fig. 7. Representations of the interactive structure-building account (A) and the verb-argument structrue account (B).Bold lines indicate an increased activation of relevant nodes and links following the prime (8a).ATL = Agent-Theme-Location, ALT = Agent-Location-Theme.NP = Noun Phrase, PP = Prepositional Phrase.
the assistant packed the lift with was (the equipment).Locative Structure PP-first 8f.What the assistant packed into the lift was (the equipment).Agent-Theme-Location (Location-focused) With-Structure PP-first 8g.What the assistant packed with the equipment was (the lift).Locative Structure NP-first 8h.What the assistant packed the equipment into was (the lift).the man loaded (the bus with was the luggage).Locative Structure PP-first 9b.What the man loaded (onto the bus was the luggage).Agent-Theme-Location (Location-focused) With-Structure PP-first 9c.What the man loaded (with the luggage was the bus).Locative Structure NP-first 9d.What the man loaded (the luggage onto was the bus).Experiment 5 Target Completion Agent-Location-Theme With-Structure NP-first 10a.The man loaded (the bus with the luggage) Agent-Theme-Location Locative Structure NP-first 10b.The man loaded (the luggage onto the bus).

Fig. 9 .
Fig. 9. Percentages and standard errors of With-Structure (relative to Locative Structure) responses by prime argument structure, prime thematic role order, and verb repetition in Experiment 4.

Fig. 10 .
Fig. 10.Percentages and standard errors of With-Structure (relative to Locative Structure) responses by prime argument structure, thematic role order congruence, and verb repetition in Experiment 4.

Fig. 11 .
Fig. 11.Percentages and standard errors of Location-Theme (relative to Theme-Location) orders by prime thematic role order, prime argument structure and verb repetition in Experiment 4.

Fig. 12 .
Fig. 12. Percentages and standard errors of Location-Theme (relative to Theme-Location) orders by prime thematic role order, argument structure congruence and verb repetition in Experiment 4.

Fig. 13 .
Fig. 13.Percentages and standard errors of NP-first (relative to PP-first) responses by prime constituent structure, thematic role order congruence, and verb repetition in Experiment 4. NP = Noun Phrase, PP = Prepositional Phrase.

Fig. 14 .
Fig. 14.Percentages and standard errors of NP-first (relative to PP-first) responses by prime constituent structure, argument structure congruence, and verb repetition in Experiment 4. NP = Noun Phrase, PP = Prepositional Phrase.

Fig. 15 .
Fig. 15.Percentages and standard errors of Location-Theme (With-structure) responses relative to Theme-Location (Locative structure) responses in Experiment 5.
was (the equipment) that the assistant loaded the lift with.Locative Structure 11b.It was (the equipment) that the assistant loaded into the lift.Location-Agent-Theme (Location-focused) With-Structure 11c.It was (the lift) that the assistant loaded with the equipment.Locative-Structure 11d.It was (the lift) that the assistant loaded the equipment into.Not-Repeated Theme-Agent-Location (Theme-focused) With-Structure 11e.It was (the equipment) that the assistant packed the lift with.Locative Structure 11f.It was (the equipment) that the assistant packed into the lift.Location-Agent-Theme (Location-Focused) With-Structure 11g.It was (the lift) that the assistant packed with the equipment.Locative-Structure 11h.It was (the lift) that the assistant packed the equipment into.Experiment 6 Target Completion Agent-Location-Theme With-Structure 12a.The man loaded (the bus with the luggage) Agent-Theme-Location Locative Structure 12b.The man loaded (the luggage onto the bus).

Fig. 16 .
Fig. 16.Percentages and standard errors of Location-Theme (With-structure) responses relative to Theme-Location (Locative structure) responses in Experiment 6.

Table 2
Analyses of structural choice in Experiment 1.

Table 3
Analyses of structural choice in Experiment 2.

Table 4
Analyses of structural choice in Experiment 3.

Table 5
Example prime and target sentences in Experiments 4-5.

Table 6
Analyses of argument structure choice in Experiment 4.

Table 7
Analyses of thematic role order in Experiment 4.

Table 8
Analyses of constituent structure in Experiment 4.

Table 9
Analyses of structural choice in Experiment 5.

Table 11
Analyses of structural choice in Experiment 6.
13.17.Frequencies of target responses inExperiment 6   13.18.Experiment 4 vs.Experiment 5 comparison on argument structure priming (with supplementary likelihood ratio test analyses) Experiment 4 vs.Experiment 5 comparison on thematic role order priming (with supplementary likelihood ratio test analyses)