SYNTACTIC PRIMING AND REPRESENTATION OF SYNTACTIC INFORMATION

Syntactic priming is the tendency to reproduce the same syntactic structure²

A model of the representation of the syntactic information of verbs in the monolingual lexicon has been proposed by Pickering and Branigan (1998) based on findings of syntactic priming with PO and DO structures independently of repetition of verb form, tense, aspect, and mood. Their account relies on Levelt's (1989) and Roelofs' (1992, 1993) models of lexical access and production and their more detailed version as explicated in Levelt, Roelofs, and Meyer (1999). In these models syntactic information associated with words is represented at the lemma level. More specifically, Roelofs' model for verbs (1993) incorporates information about syntactic category, tense, person, number, and mood. Each particular piece of syntactic information (e.g., verb for syntactic category, present for tense) is represented by a single, separate node. These syntactic nodes are shared among different lemmas that belong to the same category or exhibit the same syntactic feature in a particular context, thus avoiding representational redundancy. Whenever a verb is used in context, in addition to the corresponding (verb) lemma node, the appropriate syntactic nodes are also activated.

Pickering and Branigan (1998) extended this verb model by adding nodes that represent “combinatorial” information of verbs (Figure 1). Combinatorial information “specifies the way in which a word can combine with other linguistic units to form possible expressions of the language. A verb (e.g., eat) can combine with arguments (e.g., the men, the food) that correspond to the participants in the action denoted by the verb” (Pickering & Branigan, 1998, p. 634), that is, its subcategorization frame or argument structure. In Pickering and Branigan's model, the combinatorial nodes are linked to verb lemmas and are active when a verb is used with a particular complement. For example, the lemma of a ditransitive verb that dativises is connected to two combinatorial nodes: one for the PO construction (noun phrase, prepositional phrase [NP], [PP]) and another for the DO construction (NP, NP). According to this account, syntactic priming is affected because combinatorial nodes are shared among verb lemmas that opt for the same postverbal syntactic patterns. The activation of a combinatorial node does not dissipate immediately and can therefore facilitate processing of the same structure in subsequent discourse. Finally, Pickering and Branigan do not exclude the possibility that the combinatorial nodes of verbs include not only their subcategorization information (e.g., NP, PP, and NP, NP for ditransitive verbs) but also their overall combinatorial potential, that is, both their argument and nonargument phrases.

Pickering and Branigan's (1998) partial model of shared representations of the syntactic information associated with verbs at the lemma level in the monolingual production lexicon. From “The representation of verbs: Evidence from syntactic priming in language production,” by M. J. Pickering and H. P. Branigan, 1998, Journal of Memory and Language, 39. Copyright 1998 by Elsevier. Reprinted with permission.

Pickering and Branigan's (1998) model can also be used to frame questions about the relationship between the representations of L1 and L2 syntactic information related to verbs in the bilingual lexicon. On the one hand, the combinatorial nodes of one language (L2) may map onto those of the other (L1), resulting in shared representations, at least for languages that offer the same or equivalent combinatorial possibilities for a particular category of verbs. In such a model, the L1–L2 shared syntactic (combinatorial) information is linked to and can be activated by both L1 and L2 verbs. On the other hand, there may be separate representations of the combinatorial possibilities in each language, resulting in duplicate representations. These language-specific representations are linked to and activated by verbs of a single language only. The former model has the advantage of an economic representational architecture where resources and information from either language can be drawn upon to the maximum, whereas the latter offers processing autonomy and minimizes interlanguage interference (cf. also Hartsuiker, Pickering, & Veltkamp, 2004).

SYNTACTIC PRIMING ACROSS LANGUAGES

In contrast to the monolingual literature, syntactic priming has been less researched in the bilingual literature. Although most of the studies that have employed cross-language syntactic priming thus far employed overlapping type of structures, tasks, and languages, their results are contradictory. Heydel and Murray (1997, described in Heydel & Murray, 2000) presented German–English bilinguals with a German sentence and two pictures per trial. Participants had to decide whether the German sentence was related to the events depicted in the pictures—this was never the case for the critical items, that is, prime sentences and target pictures—and then describe each picture in writing using one English sentence. There were three types of prime sentences: (a) actives (Ein PR-Mann_NOMberät den Manager_ACC, The_NOM PR-man_NOM advises the manager_ACC), (b) passives (Der Manager_NOMwird von einem PR-Mann_DATberaten, The manager_NOM is advised by the PR-man_DAT), and (c) topicalizations (Den Manager_ACCberät ein PR-Mann_NOM, The PR-man advises the manager).

The target pictures could be described in English using only an active or passive construction. The results showed priming of the production of active target constructions following an active prime sentence (an increase of 7%) and priming of the production of passive target constructions after a passive prime sentence (an increase of 8%). Along with passive primes, however, topicalizations also primed the production of passive target responses (an increase of 11%), rather than the production of actives. The effect was, thus, interpreted as arising from the conceptual similarities (in this case the order of thematic roles) between primes (German topicalizations) and targets (English passives).

Two methodological issues should be noted here though. First, in the majority of previous studies, the picture description task was disguised as a memory task. In Heydel and Murray's (2000) tasks, participants were presented with prime sentences and target pictures and had to decide whether the prime sentence was related to the event depicted in the target pictures prior to target picture description. That is, they were invited to consciously elaborate on the semantics of primes and targets looking for similarities or associations. Critical primes and targets were always semantically unrelated (e.g., a prime sentence like Ein PR-Mann berät den Manager (A PR-man advises the manager) paired with two target pictures showing a pelican swallowing a fish); nonetheless, they both entailed an event and two participants. Thus, the conceptual organization of the prime (i.e., the arrangement of thematic roles) might have become unusually potent by virtue of the requirements of the task.

Second, Heydel and Murray (2000) elicited written target responses in their tasks. Although structure priming has been obtained in written production as well, Branigan, Pickering, and Cleland (1999) found that the effect decays rapidly in written production when other structures intervene between prime and target. However, this rapid decay does not apply to oral production (Branigan, Pickering, Stewart, & McLean, 2000). They attributed this difference to the specific nature of written production in that it may induce more conceptual processing than oral production. Writing is a slower and more effortful procedure. More importantly, a written production task allows more time to think and plan the resulting outcome. Consequently, different factors than the ones involved in oral production might come into play. Therefore, it is not unlikely that some processing details are different between written and oral sentence production and differentially affect syntactic priming.

Loebell and Bock (2003) used a picture description task that differed methodologically from Heydel and Murray's (2000). In this task, participants had to repeat the auditory prime sentence, describe the target picture, and then decide whether they had heard the sentence or seen the picture before. They tested four types of primes (actives, passives, POs, and DOs) both L1-to-L2 and L2-to-L1 with German–English bilinguals. They obtained a priming effect for DOs only. No difference between L1-to-L2 and L2-to-L1 priming was obtained.

In a sentence recall task, Meijer and Fox-Tree (2003) primed English target sentences with Spanish PO sentences. Unlike Loebell and Bock (2003), they obtained cross-language syntactic priming of PO structures from Spanish to English. They also showed that Spanish PO sentences are primed by English sentences with locative or instrumental PPs that shared syntactic structure but not thematic roles with the targets. However, in neither experiment were the items rotated across conditions, leaving open the possibility that these effects may be due to specific items.

Hartsuiker et al. (2004) employed a dialogue task (Branigan, Pickering, & Cleland, 2000) with Spanish primes and English targets but the same prime structures as Heydel and Murray (1997, quoted in Heydel & Murray, 2000): actives, passives and “dislocated” actives (corresponding to Heydel and Murray's German topicalizations). Unlike Heydel and Murray, they found that only passive primes produced priming relative to a baseline. The fact that dislocated actives did not produce priming supports the view that the effect resulting from the “topicalizations” in Heydel and Murray might be because of the procedure used, although such an explanation cannot account for the absence of priming from actives in the Hartsuiker et al. study.

A possible explanation for the contradictory results among the previous four studies regarding priming of actives versus passives and POs versus DOs might be the different tasks and methodologies employed. Note, however, that the picture description task and the dialogue task, used by Loebell and Bock (2003) and Hartsuiker et al. (2004), respectively, have also been employed in L1 studies (e.g., Bock & Loebell, 1990; Branigan, Pickering, & Cleland, 2000) where robust syntactic priming effects have been reported with all these structures. The lack of priming with PO structures in the Loebell and Bock study specifically might be because of the particular status of this construction in German. The PO structure is grammatically correct and semantically acceptable only with a small number of German ditransitive verbs; for example, it is ungrammatical with such main ditransitive verbs as geben (give), zeigen (show), anbieten (offer), and leihen (lend). This suggests that the DO, which is perfectly grammatical with all German ditransitive verbs, is the main structure used to express the dative meaning in German, whereas the use of the PO is quite restricted and might thus be underspecified in the mental lexicon of German speakers. Even more revealingly, the within-language effects in a follow-up experiment using only the German material did not reach significance.

More robust cross-language results come from two more recent studies. Desmet and Declercq (2006) demonstrated L1-to-L2 priming of preferences for (high or low) relative clause attachment. Dutch–English bilinguals tended to attach a relative clause to the first postverbal NP of an ambiguous English target sentence (e.g., John met the boss of the employees that…) after completing an equivalent Dutch prime sentence with a relative clause unambiguously attached to the first postverbal NP than after completing a Dutch prime sentence with a relative clause unambiguously attached to the second postverbal NP. However, note that this type of syntactic information is not directly related to specific lexical (here verb) entries, unlike, for example, the PO and DO structures examined in the previous studies. Using a dialogue task, Schoonbaert, Hartsuiker, and Pickering (in press) obtained both L1-to-L2 and L2-to-L1 priming with PO and DO structures. Moreover, L1-to-L2, but not L2-to-L1, priming was significantly boosted when using translation-equivalent prime and target verbs.

In summary, despite some contradictory findings and methodological limitations, previous cross-language studies of syntactic priming support the assumption that at least some structural/syntactic information is shared between languages. In particular, Loebell and Bock (2003), Meijer and Fox-Tree (2003), and Hartsuiker et al. (2004) argued in favor of the existence of shared information at a purely syntactic level of representation. Contrary to Heydell and Murray (2000), they do not postulate any contribution to cross-language syntactic priming from conceptual (thematic role) similarities. However, the bilingual data thus far cannot be conclusive about the nature of cross-language syntactic priming or the conditions under which it takes place. For example, apart from syntactic structure, the prime sentences that induced priming in Loebell and Bock (DO; 2003), Meijer and Fox-Tree (PO; 2003), Hartsuiker et al. (passive; 2004), and Schoonbaert et al. (PO and DO; in press) also fully or partially shared number, type, and order of thematic roles with the target sentences. The cross-linguistic experiments in this study thus tested for syntactic priming with PO and DO structures but using a different language combination from the ones employed thus far (i.e., Greek–English in which POs and DOs are acceptable) and additional priming conditions to determine the source of the effect and the nature of the representation of syntactic information in the bilingual lexicon.

THE PRESENT STUDY

This paper examined two theoretical issues. The first issue, explored in Experiment 1, investigates the shared or independent nature of L1 and L2 syntactic representations in the bilingual lexicon by testing two alternative hypotheses. The first one, which will be termed the language-shared hypothesis, claims that equivalent L1 and L2 syntactic structures share representations in the bilingual mental lexicon. The second hypothesis, which will be called the language-specific hypothesis, states that even equivalent L1 and L2 syntactic structures do not have shared representations. The language-shared hypothesis predicts cross-language syntactic priming with these structures. If participants produce an L1 utterance with a PO structure, they will be more likely to produce a subsequent L2 utterance with a PO rather than a DO structure and vice versa. The language-specific hypothesis predicts no such priming. This prediction is also based on the assumption that activation of syntactic representations perseveres after their use.

The language-shared hypothesis also makes an additional prediction. If cross-language syntactic priming exploits translational (lexical) links and/or shared concept nodes between L1 and L2 lemmas (in addition to syntactic information), priming will be significantly stronger when the L1 prime and L2 target verbs are translation equivalents (e.g., δ

νω/'ðin

/[give]) than when they are unrelated (e.g., πρ oσφ

ρω/pr

'sfεr

/[offer, give]). In contrast, if priming is because of the shared L1–L2 subcategorization information only, then its magnitude will not be affected by whether the primes and targets are translation equivalents. The recent Schoonbaert et al. (in press) data support the former possibility because they found a “lexical boost” with translation-equivalent prime and target verbs and adjacent prime and target trials. To test for purely structural priming, independent of discourse, conceptual, and lexical influences from prime to target structures, we introduced in our task a filler trial between primes and targets containing an intransitive structure.

The second issue that the paper investigates concerns the exact nature of syntactic representations in the bilingual lexicon. Do these representations encode information about syntactic structure in general, about constituent order, order of thematic roles, or a combination of these? Note that the cross-linguistic priming between PO or DO structures obtained in previous studies (e.g., Loebell & Bock, 2003; Schoonbaert et al., in press) cannot distinguish between these three possibilities because PO and DO structures differ across syntactic structure, constituent order, and order of thematic roles (PO: NP V NP PP, agent–theme–goal; DO: NP V NP NP, agent–goal–theme). Evidence that the order of thematic roles and constituent order may drive the priming of PO and DO structures derives from monolingual studies. Hare and Goldberg (1999), for instance, showed that utterances such as The editor credited Bob with the hot story primed DO utterances with which they share order of thematic roles rather than PO utterances with which they share syntactic structure and constituent order. In contrast, Pickering, Branigan, and McLean (2002) found that shifted-PO primes (NP V PP NP, The racing driver showed to the helpful mechanic the torn overall) primed neither PO constructions, with which they share syntactic structure, but not constituent order, nor DO constructions, with which they share order of thematic roles. Instead, they behaved like baseline intransitive primes. On this basis they concluded that the representation of postverbal syntactic structure is fully specified for constituent order. However, it might be argued that the lack of priming from a shifted-PO to a PO structure is because the shifted-PO construction is rarely used in English (e.g., when the DO is particularly long) and might not be fully represented or productive in native English speakers' mental lexicon. A relatively free word order language like Greek where the shifted-PO construction is more commonly used relative to the PO (33.9% shifted-PO vs. 66.1% PO; Lascaratou, 1994) provides a better test case for investigating priming from a shifted-PO prime structure and evaluating the role of constituent order and thematic information during syntactic priming.

Thus, Experiment 2 examined the role of syntactic structure, constituent order, and order of thematic roles in cross-language syntactic priming by testing three alternative hypotheses. The first, which will be termed the syntactic structure hypothesis, claims that syntactic structure alone, irrespective of constituent order variations, is the cause of cross-linguistic syntactic priming. The second, the constituent order hypothesis, states that structural repetition during sentence production across languages is because of the surface constituent alignment of sentences. Finally, the third is the order of thematic roles hypothesis, and states that the array of thematic roles in sentences drives cross-language syntactic priming.

Finally, a further possibility about the nature of syntactic representations that will be tested in Experiment 3 is what we will term the pairing of thematic roles and syntactic structure hypothesis. This hypothesis postulates that the crucial factor in cross-language syntactic priming is the identity of the pairing of thematic roles and syntactic structure in the prime and target sentences. It has two versions. The two-argument pairing of thematic roles and syntactic structure hypothesis requires a complete matching of thematic roles and syntactic structure for both postverbal arguments of the prime and target sentence. The first-argument pairing of thematic roles and syntactic structure hypothesis requires identity of thematic roles and syntactic structure between the first postverbal argument of prime and target sentences only, complying with incremental processes of speech production. Speakers embark on articulating their thoughts as soon as some aspects of the utterance's frame become available rather than wait until all the necessary input for the utterance construction has been retrieved and processed (e.g., Kempen & Hoenkamp, 1987). Accordingly, accessing the first postverbal argument of prime ditransitive constructions, rather than processing the entire (prime) structure, might be sufficient to bias speakers' response towards one of alternative target structures in a syntactic priming task.

All three experiments employed a bilingual version of Branigan, Pickering, Stewart, and McLean's (2000) oral sentence completion task with L1 prime and L2 target items to elicit cross-language syntactic priming. The syntactic structure used was the dative alternation and the L1–L2 pair was Greek–English.

EXPERIMENT 1

Experiment 1 contrasted the language-shared and language-specific hypotheses; that is, it investigated whether representations of L1 and L2 syntactic structures are shared between pairs of languages in which these structures serve the same discourse function and have an equivalent surface syntax. Examples of such equivalent structures are the PO and DO constructions in English and Greek. Sentence 2 corresponds to 3 and sentence 4 to 5.³

• 2. The president gave/offered the gold medal to the winner.
• 3. Ο πρόεδρος
  
  δωσε/πρόσφερε το χρυσό μετ
  
  λλιο στο νικητ
  
  .
• 4. The president gave/offered the winner the gold medal.
• 5. Ο πρόεδρος
  
  δωσε/πρόσφερε του νικητ
  
  το χρυσό μετ
  
  λλιο.

The language-shared hypothesis predicts priming from L1 structures to L2, whereas the language-specific hypothesis does not.

Method

Participants

Thirty-six native Greek-speaking advanced learners of English participated in the experiment. Twenty-six were students at the University of Cambridge, studying a variety of subjects, and 8 were students at or graduates (working as English language teachers) of the department of English Studies of the University of Athens. A more detailed profile of the participants is provided in Appendix A.

Materials

The critical experimental material consisted of English and Greek sentence fragments whose main verb was a ditransitive (e.g., give, στ

λνω/'stεln

/[send]). These fragments were incorporated into 32 sets of items. Each set included two intransitive filler fragments, a prime fragment, and a target fragment. An example set of material can be viewed in Table 1. (See Appendix B for more example items.)

The prime fragment was always in Greek (L1) and the target fragment always in English (L2). The intransitive filler fragment intervening between prime and target was always in English so as to avoid having a language switch just before the target fragment completion. The other intransitive filler fragment was in English in half of the sets and in Greek in the other half.

Every target fragment contained a subject NP followed by a ditransitive verb and could be completed either with a DO construction (i.e., NP NP) or a PO construction (i.e., NP PP). The prime fragments included an additional postverbal NP and varied in two dimensions. First, by introducing this additional NP it was possible to manipulate the prime fragment completion so as to induce either a DO or PO construction. For instance, in (a) the postverbal NP is a plausible theme but an implausible recipient/goal for the action denoted by the ditransitive verb. Thus, participants were more likely to complete this fragment with a PO construction. In contrast, in (b) the postverbal NP is a plausible recipient/goal but an implausible theme. Hence, participants were more likely to complete (b) with a DO construction. Second, half of the prime fragments included an L1 translation of the target L2 verb (e.g., a and b) and the other half a different L1 verb (e.g., c and d). Thus, each of the 32 target fragments occurred with four different prime types: (a) translated-verb–PO-eliciting prime, (b) translated-verb–DO-eliciting prime, (c) different-verb–PO-eliciting prime, and (d) different-verb–DO-eliciting prime. All prime verbs had the same tense, aspect, and number as target verbs.

Nine different Greek verbs were included in the prime fragments and seven different English verbs in the target fragments. Most of the English verbs had been used in previous sentence completion tasks and were reported to produce relatively low proportions of other (i.e., non-PO or non-DO) completions (Branigan et al., 1999; Branigan, Pickering, Stewart, et al., 2000; Pickering & Branigan, 1998). In half of the sets the order of the items was intransitive, prime, intransitive, and target fragment; in the other half the order was prime, intransitive, target, and intransitive fragment. Half of the pairs of intransitive fragments employed the same verb and half a different verb. In each pair of intransitive fragments, one fragment was followed by an adverb; in half of the sets this fragment occurred between prime and target and in the other half of the sets either preceded the prime or followed the target. These manipulations had two purposes. First, they were intended to make the pairs of intransitive fragments as similar as possible to, and to an extent indistinguishable from, the pairs of primes and targets. Second, they were designed so that there was always an intransitive fragment between critical items that would preclude any discourse effects from primes to targets.

In addition, 62 filler fragments of various syntactic form were constructed, for example, noun phrase, noun phrase + verb, noun phrase + verb + noun phrase. None of the filler fragments included a ditransitive verb that could be completed with a DO or PO construction. Some of the fillers shared verbs. Finally, 10 practice fragments were constructed. Two of these fragments were used as buffer trials at the beginning of the main experiment. None of the practice items was repeated in the main experiment.

The experimental items were allocated into four presentation lsts. Each lst contained the 32 sets of items (=32 primes, 32 targets, 64 fillers), 62 filler fragments, and 10 practice fragments, for a total of 200 sentence fragments. Each lst contained eight sets of items from each condition so that each set appeared in only one of its four versions in every lst. The order of items was individually randomized for each participant under the constraints that no more than three sets of items from the same experimental condition occurred in succession and that at least three filler fragments intervened between targets and the following primes.

Procedure

Participants were individually tested using a PC screen. Each was randomly allocated to one of the four lsts. Following Branigan, Pickering, Stewart, et al. (2000), the instructions stated that the purpose of the experiment was to study the types of sentences English learners produce and that participants must repeat aloud and complete sentence fragments in English and Greek in any way they like provided that the resulting sentence is grammatically correct and semantically plausible. It was also stressed that they should start repeating the fragment as soon as it appears on screen and subsequently completing it as quickly as possible with the first completion that comes to their mind, as there was a limited time in which they had to respond. A beep sound warned them that the next fragment was to appear soon. The instructions were presented on the PC screen.

The presentation of the experimental material was controlled by SuperLab software. Each experimental trial consisted of the following stimuli: a fixation point (*) for 1000 ms, the target fragment for 5000 ms, a blank screen for 1000 ms, the beep sound that lasted approximately 118 ms, and finally another blank screen for 1000 ms. The location of the fixation point on the screen marked the beginning of each sentence fragment. Each trial lasted about 8000 ms.

Before the experiment proper, participants were given a brief questionnaire to fill in regarding their language history. A short practice section consisting of 10 sentence fragments preceded the experimental task. Each experimental session lasted approximately 45 min including four breaks, the duration of which was individually determined by each participant.

Scoring

Each experimental session was recorded, transcribed and scored. Branigan, Pickering, Stewart, and McLean's (2000) scoring method was followed. Prime and target completions were scored as PO, DO, or other. A completion was scored as PO if it met the following criteria: (a) if the ditransitive verb was followed by a NP with the thematic role of theme or patient and then a PP with the preposition to in English or σε (to) in Greek and the thematic role of recipient/goal (V NP PP(P NP); in Greek, both postverbal NPs also had to be marked as follows: NPacc PP(P NPacc); (b) if the ditransitive verb was not part of a phrasal verb construction (e.g., show off); (c) if it did not occur in a DO-eliciting prime fragment; and (d) in the case of target fragments if its DO alternative was grammatical. Similarly, a completion was scored as DO under the following conditions: (a) if the ditransitive verb was followed by a NP with the thematic role of recipient/goal and then by a second NP with the thematic role of theme or patient (V NP NP; in Greek, both postverbal NPs also had to be marked as follows: NPgen NPacc); (b) if the ditransitive verb was not part of a phrasal verb construction; (c) if it did not occur in a PO-eliciting prime fragment; and (d) in the case of target fragments, if its PO alternative was grammatical. All other completions were scored as other.

Data analysis

The data analysis was based on the method outlined by Pickering et al. (2002). That is, the dependent variable was the “PO target ratio,” a measure designed to calculate the proportion of PO versus DO target completions excluding other target completions. Target fragments following prime completions that were scored as other were also discarded from the analysis. The analysis was carried out on proportions rather than absolute numbers, because, as participants were free to complete the fragments in any way they liked, the total number of primes completed in the intended way, and hence the number of valid target completions might have varied in each condition (Pickering & Branigan, 1998). Proportions provide a better basis for comparison between and across conditions.

The PO target ratio was calculated as the number of PO target completions divided by the sum of the number of PO and DO target completions in each condition. The use of the PO rather than the DO proportions in the analyses is arbitrary, as the proportions of PO and DO target ratio are complementary. We also calculated an “overall PO target ratio,” which was the number of PO target completions divided by the sum of the number of PO and DO target completions across all four conditions, excluding again other target completions.

In addition to the analysis of the PO target ratio, an analysis of the proportion of other target completions in each of the four conditions was performed. The relevant proportions were derived by dividing the number of other target completions in each condition by the total number of prime completions in the same condition that were completed with the structure by which that condition was meant to elicit. For example, in the same-verb PO condition, the number of other target completions was divided by the total number of PO prime completions in that condition. In all sets of analyses, the proportions were calculated per participant and per item and all experimental factors were within participants and within items.

Results

Of all prime fragments (1,015), 88% (1,152) were completed with a PO or DO; of these, 24% were translated-verb PO completions, 26% translated-verb DO, 23% different-verb PO, and 27% different-verb DO. The percentage of the target fragment completions on these trials was 48% PO (492), 24% DO (242), and 28% other (281). One cell in the translated-verb DO condition and three cells in the different-verb PO condition in the participants analysis had missing values and were replaced by the overall mean. Table 2 displays the PO target ratio and the proportion of other target completions in the four prime conditions.

An analysis of variance (ANOVA) on the other target completions showed no significant difference among prime conditions (all Fs < 2.63, all ps > .1). In the PO target ratio analyses, a two-way ANOVA revealed a significant main effect of prime type, F₁ (1, 35) = 36.29, mean square error (MSE) = .015, p < .01; F₂ (1, 31) = 15.11, MSE = .025, p < .01. The mean PO target ratios in Table 2 show that participants produced more PO target completions following a PO prime completion than following a DO prime (and vice versa for the DO target completions). Neither the main effect of verb type nor their interaction approached significance (all Fs < .79) meaning that the syntactic priming effect was unaffected by the type of verb (translated vs. different) used in the prime fragments.

Finally, we tested for any significant difference in the overall number of PO and DO target completions across all conditions. Because the PO and DO target ratios are complementary (PO + DO = 1), a one-sample t test was employed on the overall PO target ratio across all conditions per participant and per item with a test value of .50. The assumption is that if there is no difference between the PO and DO target ratio, both of them will equal .50. If, therefore, the PO target ratio is significantly different from .50, then it will also be significantly different from the DO target ratio. The test revealed a significant difference in participants' overall production of PO target completions (.69) versus DO target completions (.31), t₁ (35) = 4.21, p < .01; t₂ (31) = 6.58, p < .01.

EXPERIMENT 2

Having established the existence of a syntactic priming effect across languages with PO and DO structures, the goal of Experiment 2 was to evaluate the role of constituent order and thematic roles in cross-language syntactic priming, and hence, in syntactic processing and representation in the bilingual lexicon. This is necessary, as in Experiment 1 the pairs of prime and target constructions that induced priming shared not only syntactic structure but also the order of thematic roles and constituent order.

Using the same bilingual sentence completion paradigm as in Experiment 1, we also introduced, after Pickering et al. (2002), a shifted-PO structure (Ο πρόεδρος

δωσε στο νικητ

…το βραβε

ο = The president gave to the winner… the prize) and an intransitive baseline (Ο πρόεδρος χρεωκόπησε… = The president went bankrupt…) in the set of L1 primes. As in Experiment 1, an L2 intransitive filler fragment always intervened between L1 primes and L2 targets. The three hypotheses tested in this experiment (syntactic structure, constituent order, order of thematic roles) predict the main syntactic priming effect, that is, PO prime completions will prime PO rather than DO target completions and vice versa. This priming is predicted because PO and DO structures differ along their syntactic structure, constituent order, and order of thematic roles. Table 3 provides a summary of the similarities and differences of the prime types with respect to these three variables. All three hypotheses, however, make different predictions concerning the priming that will result from the shifted-PO structure. The syntactic structure hypothesis predicts no difference in priming between PO and shifted-PO primes, that is, both PO and shifted-PO structures will prime PO target completions to the same extent in comparison to the baseline. The constituent order hypothesis predicts no priming of either PO or DO target completions following shifted-PO primes, that is, the proportion of PO target completions after shifted-PO primes will be similar to the proportion of POs after baseline (intransitive) primes. Last, the order of thematic roles hypothesis predicts no difference in priming between DO and shifted-PO primes, that is, both DO and shifted-PO structures will negatively prime PO target completions to the same degree relative to the baseline.

These predictions are valid under the assumption that (a) cross-language syntactic priming, like within-languages priming, is balanced, that is, priming of PO target completions is higher after PO primes and lower after DO primes relative to baseline primes, and (b) the production of shifted-PO target completions is negligible. If priming is not balanced and, for example, PO primes behave like baseline primes, although it is only DO primes that influence target production, then one cannot distinguish between the syntactic structure and the constituent order hypotheses as both make the same prediction: no difference in priming among the PO, shifted-PO, and baseline conditions. Moreover, both in natural speech (Wasow, 1997) and in experimental conditions (e.g., Pickering et al., 2002), native English speakers rarely produce shifted-POs in free production, unless the DO NP is particularly long.⁷

EXPERIMENT 3

Apart from the constituent order hypothesis though, the cross-language data of Experiment 2 can also be explained by an alternative account that we have termed the pairing of thematic roles and syntactic structure hypothesis, and claims that cross-language syntactic priming is driven by the combination of thematic roles and syntactic structure in the prime and target sentences. This pairing could extend to the whole structure (i.e., to both postverbal arguments; the two-argument pairing hypothesis) or may be restricted only to the first postverbal argument (the first-argument pairing hypothesis). Accordingly, accessing the first postverbal argument of prime ditransitive constructions, rather than processing the entire (prime) structure, might be sufficient to bias speakers' response toward one of alternative target structures in a syntactic priming task. Experiment 3 aimed at disentangling these hypotheses (two-argument and first-argument pairing) from the constituent order hypothesis in the context of cross-language syntactic priming.

To do so, it was necessary to employ, in the same task, constructions whose constituent order/syntactic structure and order of thematic roles coincides or contrasts with those of POs and DOs. Thus, Experiment 3 used four types of L1 prime structures: PO, DO, locative (Ο πρóεδρ oς φ

λαξε τ o χρυσó μετ

λλι o μ

σα στ o συρτ

ρι = The president kept the gold medal in the drawer) and provide-with (Ο κ oσμηματ oπ

λησ πρ oμ

θενsigmaε την πελ

τισσ

τ oυ με τ o ακριβó δαχτυλ

δι = The jeweller provided his customer with the expensive ring). Table 4 provides a summary of the similarities and differences of the prime types with respect to syntactic structure, constituent order, order of thematic roles, and pairing of syntactic structure and thematic roles. Locative constructions such as The wealthy widow drove an old Mercedes to the church have been used in the past by Bock and Loebell (1990), who showed that they primed POs such as The girl is handing a paint brush to the boy, despite the difference in thematic roles between the two utterances (locative: agent, theme, location; PO: agent, theme, recipient/goal). However, Hare and Goldberg (1999) argued that the priming demonstrated in Bock and Loebell (1990) between PO and locative structures could also be accounted for by thematic role similarities. Under some linguistic accounts (e.g., Jackendoff, 1983, 1987) the thematic role of recipient/beneficiary in The wealthy widow gave an old Mercedes

and the thematic role of locative in The wealthy widow drove an old Mercedes

can both be subsumed under a broader thematic role, that of goal. To circumvent such a possibility, we used locatives whose second postverbal argument can only assume the thematic role of location (for a detailed description see the Materials section).

Again, all hypotheses predict a main syntactic priming effect for the PO and DO structures. However, they make different predictions about priming from the locative and provide-with primes. The constituent order hypothesis predicts that provide-with structures will prime PO but not DO target completions. In other words, the proportion of PO target completions after provide-with primes will be similar to the proportion of POs after PO primes rather than DO primes. Locative primes will behave differently from PO primes. The two-argument pairing hypothesis predicts that neither the locative will prime POs nor the provide-with will prime DOs. Finally, the first-argument pairing hypothesis predicts that the locative will prime POs although the provide-with will prime DOs, and in addition, provide-with and locative should differ from each other.

Method

Participants

Thirty-six native Greek-speaking advanced learners of English took part. Thirty were graduate students at the University of Cambridge and 6 were students at or graduates of the department of English Studies of the University of Athens. None of them had taken part in Experiments 1 or 2. (See Appendix A for a more detailed profile.)

Materials

The design of the experimental material was again identical to that in Experiment 1 except for the design of the prime items. There were four different types of prime fragments. An example of these prime fragments and a target fragment is given in Table 1. (See Appendix B for more example items.)

All prime fragments consisted of three arguments: a subject NP, a verb, and a postverbal NP. Prime fragments of Type (c) included a locative verb that can take a direct object NP and a locative adverb phrase (AdvP), that is, an AdvP that expresses the thematic role of location. These verbs were the Greek equivalents of Levin's (1993) “keep verbs” (φυλ

ω/fl'la

/[keep], αφ

νω/a'fin

/[leave], απ oθηκε

ω/ap

θi'ķεv

/[store], στ oιβ

ζω/sti'vaz

/[stack]) and “verbs of concealment” (κρ

βω/'kriv

/[hide]). The verb ξεχνacuteαω/ksε'xna

/(forget) was also added, used in the sense of “leave something behind,” “forget to take something.” Unlike the motion verbs (e.g., drive, return, move) used in the prepositional locatives by Bock and Loebell (1990), none of the locative verbs in this task express motion of any sort, and thus the locative PP can be interpreted as pure locative rather than the goal of the verb's action. Keep verbs “relate to maintaining something at some location. They do not describe the actual putting of any entity at this location” (Levin, 1993, p. 146, my emphasis). Similarly, verbs of concealment relate “to keeping something out of view” (Levin, 1993, p. 143, my emphasis). In these primes the postverbal NP was always a plausible theme for the action denoted by the verb and was followed by a locative adverb (μ

σα/'mεsa/[in]/[inside], π

νω/'pan

/[on], κ

τω/'kat

/[under], π

σω/'pis

/[behind]). This locative adverb required a locative preposition; in this case, either σε/sε/(in, at, on, to) or από/a'p

/[from], followed by an NP as a complement V NP AdvP(Adv PP[P NP]). The locative adverb and the locative preposition form a periphrastic or complex preposition. Finally, prime fragments of Type (d), which we term provide-with primes, employed Greek verbs equivalent to Levin's (1993) “verbs of fulfilling” (πρ oμηθε

ω/pr

mi'θεv

/[provide], εφ oδι

ζω/εf

'ðjaz

/[supply]), “equip verbs” (εξoπλ

ζω/εks

'pliz

/[equip], απ oζημι

νω/ap

zi'mj

n

/[compensate], ανtauαμε

βω/a

da'miv

/[reward], φ oρτ

νω/f

'rt

n

/[burden], επιβαρ

νω/εpiva'rin

/[charge sb [with money/a chore]], επιφ oρτ

ζω/εpif

'rtiz

/[entrust sb [with a task]]), and “pelt verbs” (β oμβαρδ

ζω/v

mva'rðiz

/[bombard]).

⁸

All verbs in the provide-with category express their arguments in the NP V NP PP(with NP) pattern, where the postverbal NP expresses the recipient/goal argument and the PP expresses the theme argument. Only πρ oμηθε

ω/par

mi'θεv

/(provide) and φ oρτωνω/f

'rt

nc/(burden) can also occur in the NP V NP PP(to NP) pattern, where the postverbal NP expresses the theme argument and the PP expresses the recipient/goal argument.

In these primes the postverbal NP always signified the recipient/goal of the action of the verb and was followed by the preposition με/mε/(with). The preposition με was meant to initiate a PP that would be the theme for the action conveyed by the verb. In total, the Greek prime fragments included nine ditransitive verbs, six locative verbs and nine provide-with verbs. The English target fragments used six ditransitive verbs. Each of the 32 target fragments occurred with four different prime types: (a) PO-eliciting prime, (b) DO-eliciting prime, (c) locative-eliciting prime, and (d) provide-with eliciting prime.

Procedure and data analysis

The procedure and data analysis were the same as in Experiment 1.

Scoring

It was identical to that of Experiment 1 apart from the following modifications. Prime completions were scored as PO, DO, locative, or provide-with. A prime completion was scored as locative if the locative adverb at the end of the sentence fragment was followed by a preposition and an accusative noun phrase that had the thematic role of location V NPacc AdvP(Adv PP[P NPacc]). It was scored as provide-with if the preposition με/mε/(with) at the end of the sentence fragment was followed by an accusative noun phrase that had the thematic role of recipient/goal (V NPacc PP[P NPacc]).

Discussion

As in Experiments 1 and 2, there was cross-linguistic syntactic priming (10%) between PO and DO. Moreover, L1 provide-with structures primed L2 DOs rather than L2 POs. L1 locatives behaved similarly to L1 PO primes in both participant and item analysis and differed significantly from L1 provide-with primes concerning priming of L2 POs in the participant analysis. The difference in priming between POs and provide-with structures indicates that the overall surface structure is not the source of priming.⁹

Even if it is assumed that the AdvPs of locative structures in fact act as PPs, the overall priming observed in Experiment 3 cannot be explained by a purely syntactic structure and constituent order account, as such an account could not accommodate the fact that the provide-with structures prime DO targets, although their syntactic structures do not fully match.

The only account that can plausibly accommodate the results of both Experiment 2 and 3 is the first argument pairing hypothesis, that is, it is the combination of thematic roles and syntactic structure in the first postverbal argument of the prime and target sentences that triggers syntactic priming.

The argument holding for Bock and Loebell's (1990) material that to-PPs express the same thematic role (Goal) in both locative structures with motion verbs and POs (e.g., Jackendoff, 1987), cannot apply to the present data. The locative AdvPs in combination with the specific verbs used in this task (e.g., φυλ

ω… μ

σα σε [keep…in]) denote location rather than goal. Moreover, despite the priming between L1 provide-with and L2 DO structures and the tendency of L1 locatives to prime L2 POs, the order of thematic roles up to the first postverbal argument (DO-provide-with: agent, recipient/goal; PO-locative: agent, theme) is not a plausible explanation of cross-language syntactic priming because it cannot accommodate the data of Experiment 2. In that task, an L1 shifted-PO did not prime L2 DO responses, although both structures share the same order of thematic roles (agent, recipient/goal, theme).