Korean, the native language of both South Korea and North Korea, is estimated to have more than 80 million people in the world who speak Korean as a first, second, or heritage language (Pae, 2024). Korean is a verb-final language with the canonical order of subject, object, and verb (SOV). The order of subject and object is relatively flexible. A simple transitive sentence in Korean can have one of two basic orders: either SOV or OSV. Noun phrases (NPs) are marked by one of three case markers (or particles): the subject or nominative marker-이/가, −i/ka (NP_NOM), the object or accusative marker-을/를, −eul/leul (NP_ACC), or the topic marker-은/는, −eun/neun (NP_TOP). In this study, we represent sentences in Korean using standard romanization. These combinations enable the creation of four types of sentences (Lee and Ramsey, 2000; Pae, 2024). These, with either SOV or OSV, are illustrated in Sentences (1) to (4). All sentences fundamentally carry the same meaning. Sentence (1) is the canonical order, which employs the scrambled order OSV, as shown in Sentence (2). Topicalization in Korean is achieved by adding the marker -eun/neun to a noun phrase, as illustrated in Sentence (3) with subject topicalization and Sentence (4) with object topicalization.

(1) SOV: Canonical order.

Eom-ma -ga sa-gwa -leul meog -eoss-da.

NP(mother) NOM NP(apple) ACC V(eat) PST.

엄마가 사과를 먹었다.

‘(My) mother ate (an) apple.’

(2) OSV: Scrambled order.

Sa-gwa -leul eom-ma -ga meog -eoss-da.

NP(apple) ACC NP(mother) NOM V(eat) PST.

사과를 엄마가 먹었다.

(3) SOV: Subject topicalized order.

Eom-ma -neun sa-gwa -leul meog -eoss-da.

NP(mother) TOP NP(apple) ACC V(eat) PST.

엄마는 사과를 먹었다.

(4) OSV: Object topicalized order.

Sa-gwa -neun eom-ma -ga meog -eoss-da.

NP(apple) TOP NP(mother) NOM V(eat) PST.

사과는 엄마가 먹었다.

In Sentence (1), ‘(My) mother ate (an) apple,’ follows the SOV order with ‘(my) mother’ marked by the nominative case marker -ga (NP_NOM), ‘(an) apple’ marked by the accusative case marker -leul (NP_ACC), and the past tense verb (V-PST) meog-eoss-da ‘ate’ at the end. In Sentence (2), the positions of NP_NOM and NP_ACC are scrambled, as is characteristic of the OSV order. Once again, the final verb ‘ate’ appears at the end of the sentence. The OSV order is formed by moving the object to the beginning of the sentence. This word order was termed scrambled by Ross (1967), who primarily discussed this phenomenon in relation to Germanic languages (Neeleman, 1994; Broekhuis, 2008).

Several psycholinguistic studies conducted in Japanese, which shares similar syntactic features with Korean, have investigated sentence processing. These studies (e.g., Mazuka et al., 2002; Ueno and Kluender, 2003; Koizumi and Tamaoka, 2004, 2010; Miyamoto and Takahashi, 2004; Tamaoka et al., 2005, 2014; Imamura et al., 2016; Witzel and Witzel, 2016; Tamaoka and Mansbridge, 2019) consistently found that the canonical SOV order is processed faster than the scrambled OSV order. For instance, in the methodology employed by Tamaoka et al. (2005), participants were tasked with evaluating the correctness of each sentence, considering both its semantic coherence and grammatical accuracy. This approach is hereafter referred to as a ‘sentence correctness decision task.’ Utilizing this task, the study measured the processing time for both canonical SOV and scrambled OSV sentence orders in Japanese, employing sentences analogous to Korean Sentences (1) and (2). Without any preceding context, the study found that Japanese canonical SOV sentences were processed both more rapidly and accurately compared to their scrambled OSV counterparts. Drawing on these results, the current study predicts a similar processing advantage for canonical sentences in Korean over their scrambled alternatives.

The processing inefficiency observed in both accuracy and speed for scrambled sentences is commonly referred to as the scrambling effect. One potential explanation for the delay in processing the scrambled OSV order in Japanese comes from the gap-filling parsing model (Frazier and Rayner, 1982; Stowe, 1986; Frazier, 1987; Frazier and Clifton, 1989; Frazier and Flores D’Arcais, 1989). According to this model, native Korean speakers likely identify the initial NP_ACC marked by -leul as the filler and subsequently search for its original position in the specifier of the gap to establish the filler-gap dependency. Given that the OSV scrambled order entails a syntactically more complex structure than its corresponding SOV order, the process of gap-filling parsing for the OSV order is expected to be slower than for the SOV order. Therefore, we anticipate observing similar processing dynamics in Korean as those identified in Japanese, where scrambled sentences are processed less efficiently compared to canonical sentences.

In the processing of scrambled word order in Japanese and potentially in Korean, the filler-gap dependency is activated. While the subject typically appears first in a sentence, the object precedes it in a scrambled sentence. Given that Korean and Japanese are null-subject (or pro-drop) languages, the subject of a sentence can be omitted. Consequently, native speakers of Korean and Japanese may initially interpret the sentence as having a null subject. However, in a scrambled sentence, the subject follows the object (OS order). Native speakers of Korean and Japanese will then search for the gap₁, representing the position where the object would occur in the SOV canonical order. They establish the filler-gap dependency between the object and the gap as O₁ S gap₁ before encountering the verb to fully comprehend the scrambled sentence. This additional processing step may prolong the processing time and potentially lead to comprehension errors.

In Korean, canonical and scrambled word orders overlap with topicalization orders. The subject in the SOV canonical order is typically positioned at the beginning of the sentence unless a pro-drop occurs. Therefore, the canonical S_NOMOV and the topicalized S_TOPOV share the same word order. For instance, S_NOMOV for 경찰이 범인을 잡았다 (Gyeong-chal-i beom-in-eul jab-ass-da, ‘The police caught the culprit’) and S_TOPOV for 경찰은 범인을 잡았다 (Gyeong-chal-eun beom-in-eul jab-ass-da) both have the subject in the initial position, which is canonical. It is important to note that Korean lacks definite and indefinite articles, so nouns in a sentence are typically denoted without an article. Similarly, O_ACCSV and O_TOPSV also share the same word order. For example, O_ACCSV for 범인을 경찰이 잡았다 (Beom-in-eul gyeong-chal-i jab-ass-da, ‘The culprit, the police caught’) and O_TOPSV for 범인은 경찰이 잡았다 (Beom-in-eun gyeong-chal-i jab-ass-da) both have the object in the scrambled position. Since the word orders of subject and object topicalization share the same pattern as canonical and scrambling word orders, it is difficult to measure the effect of subject and object topicalization independently from the processing of canonical and scrambling structures.

In Korean, noun phrases can be topicalized using the marker -eun/neun (NP_TOP). As topicalization is a discourse feature, NP_TOP is positioned at the beginning of a sentence to indicate the topic. A subject or an object can be topicalized using the marker -eun/neun. Sentence (3) exemplifies a subject-topicalized (NP_SUB-TOP) sentence in the SOV canonical order, beginning with ‘Speaking of (my) mother.’ Subject topicalization also implies an exclusionary meaning, referring specifically to ‘mother’ and not other family members. Sentence (4) illustrates an object-topicalized (NP_OBJ-TOP) sentence in the OSV scrambled order, commencing with ‘Speaking of (the) apple.’ Object topicalization similarly implies an exclusionary meaning, specifically referring to ‘(the) apple’ and not other fruits.

It is also notable that the distinct grammatical particles (or makers) used for the subject (−이/가, −i/ga) and the object (−을/를, −eul/leul) differ from auxiliary particles, such as -은/는, −eun/neun. In particular, auxiliary particles that serve functions beyond topicalization may interfere with processing due to their multiple functions. For instance, the marker -eun/neun can be employed for emphasis, as in 그는 영어는 잘 한다 (Neun yeong-eo-neun jal han-da), which translates to ‘(He is not so good at other subjects, but) he is good at English.’ Another usage is to indicate contrast, as seen in 인생은 짧고 예술은 길다 (In-saeng-eun jjalb-go ye-sul-eun gil-da), meaning ‘Life is short, art is long.’ Additionally, it can function as a pseudo-subject particle in certain contexts where the subject has already been mentioned. In such cases, its function may not necessarily be as a topic marker. Additionally, the Korean language is also a highly contextual language. Depending on the use of the auxiliary particle, the nuance of a given sentence varies because it carries semantic information rather than the syntactic function. Thus, it is essential to be cautious of the multiple functions of the particle (marker) -eun/neun when interpreting the results of an experiment.

In verb-final languages such as Korean and Japanese, the sentence-ending verb also plays a crucial role in properly understanding a sentence, particularly when there is no animacy contrast between the subject and the object. This processing tendency can be described as a backward argument-verb dependency (Tamaoka and Mansbridge, 2019). In the absence of an animacy contrast, native Japanese speakers rely on information provided by the verb at the end of the sentence to establish the structural properties of scrambled constituents. Previous studies (Hsiao and Gibson, 2003; Pu, 2007; Vasishth et al., 2013; Kwon et al., 2019 for Mandarin Chinese; Mak et al., 2002 for Dutch) have indicated that animacy features affect the processing of relative clauses. Similarly, an animacy contrast may influence the processing of Korean and Japanese sentences, given that native speakers of these languages encounter the verb only at the end of the sentence. In such cases, animacy information might play a crucial role in constructing sentence structure, particularly in verb-final languages such as Korean and Japanese. Therefore, a semantically driven analysis for sentence processing would be considered an important additional factor.

It is widely accepted that both Korean and Japanese belong to the group of Altaic languages, sharing many linguistic features (Lee and Ramsey, 2000; Pae, 2024), although the debate over the designation of language family remains unsettled. In both Korean and Japanese, simple transitive sentences can exhibit either SOV or OSV word orders. In Japanese, noun phrases (NPs) are denoted by one of three case markers: the nominative -ga (NP_NOM), the accusative -o (NP_ACC), or the topicalization -wa (NP_TOP). This system mirrors Korean in that the word orders demonstrate overlap in two primary aspects. First, sentences with a topicalized subject adhere to the same SOV structure as canonical sentences. Second, sentences where the object is topicalized align with the OSV order typical of scrambled sentences.

A study on topicalization in Japanese by Imamura et al. (2016) found the order of processing speed to be as follows: canonical S_NOMO_ACCV (M = 1,410 ms) = subject topicalized S_TOPO_ACCV (M = 1,414 ms) < scrambled O_ACCS_NOMV (M = 1,512 ms) < object topicalized O_TOPS_NOMV (M = 1,626 ms). This study showed that the processing time for the subject topicalized word order of S_TOPO_ACCV was the same as that of the canonical S_NOMO_ACCV. Therefore, as Imamura et al. (2016) suggested, this order seems to be commonly used, resulting in both structures being easily processed within a short time. A simpler explanation may be that when the nominative marker -ga is used, the topic marker -wa in the SOV order appears to identify the subject or its equivalent. Another possible explanation is that a sentence topic placed at the beginning of a sentence may speed up processing by initially providing an overall theme of the sentence. This discourse feature may explain equivalencies in processing speed. Additionally, it may be that a combination of these factors speeds up processing time.

However, the processing speed of object topicalized sentences was slower than their equivalent canonical sentences, and furthermore, even slower than their scrambled sentences. Thus, as Kuroda (1987) suggested, object topicalization could involve the process of both scrambling and topicalization. Additionally, it is possible to add the explanation that object topicalization focuses on ‘(an) apple’. This focus implies an exclusionary meaning of ‘an apple’ and ‘not any other fruits.’ Consequently, this focus may further delay the processing speed compared to corresponding scrambled sentences. Once again, because subject/object topicalization shares the same word order with canonical/scrambling, the effect of subject/object topicalization cannot be measured independently from canonical/scrambling.

In the Austronesian language of Tongan, the canonical word order is VSO (Verb-Subject-Object), commonly used in transitive sentences. However, a VOS (Verb-Object-Subject) order is also grammatically possible (Churchward, 1953; Dixon, 1979, 1994; Otsuka, 2000, 2005a,b; Custis, 2004). Tongan, being an ergative language (Otsuka, 2005a,b, 2010), marks both the subject of an intransitive sentence and the object of a transitive sentence with the same absolutive (ABS) marker ‘a, while the subject of a transitive sentence is marked by the ergative (ERG) marker ‘e (i.e., ERG/ABS case marking pattern). Tongan verbs have limited inflectional morphology (e.g., no inflection for tense). The present study focuses only on transitive sentences.

In Tongan, a noun phrase (NP) that is topicalized is positioned before the verb in either SVO or OVS orders, marked with the topic marker ‘ko’. This results in four potential word orders for transitive sentences: VSO, VOS, SVO, or OVS. Unlike in Japanese and Korean, when topicalization occurs in Tongan, the resulting word order does not overlap with either the VSO canonical or scrambled orders. To observe the topicalization effect as an independent phenomenon, the present study experimentally investigated processing times for these four orders of transitive sentences. A verb-initial language such as Tongan is ideal for investigating this processing function.

In Tongan, a sentence denoting ‘the woman ate the fish’ is assumed to be in the canonical order, as shown in Sentence (5), where NP refers to the noun phrase, PST refers to the past tense, ERG (‘e) refers to the ergative case marker, ABS (‘a) refers to the absolutive case marker, and REF refers to the referential/specific article (Anderson and Otsuka, 2006; Macdonald, 2014).

(5) VSO: Canonical transitive Tongan sentence.

Na’e kai ‘e he fefine ‘a e ika.

PST V(eat) ERG REF NP (woman) ABS REF NP (fish)

‘The woman ate the fish.’

The VOS scrambled order is derived from the VSO canonical order by moving the object (O) between the verb (V) and the subject (S), as shown in Sentence (6), constructing a more complex structure (VO₁Sgap₁) than the canonical (VSO) order. The VOS scrambled order retains the same meaning of ‘The woman ate the fish.’

(6) VO₁Sgap₁: Scrambled transitive Tongan sentence.

Na’e kai ‘a e ika ‘e he fefine.

PST V(eat) ABS REF NP (fish) ERG REF NP (woman)

‘The fish, the woman ate.’

According to the gap-filling parsing model (Frazier and Rayner, 1982; Stowe, 1986; Frazier, 1987; Frazier and Clifton, 1989; Frazier and Flores D’Arcais, 1989), native Tongan speakers possibly process VOS scrambled sentences in the following manner: The noun phrase ‘a e ika ‘the fish’ after the verb is an absolutive case marked (‘a) noun phrase indicating the object (O). Native Tongan speakers perceive that the object should come after the subject (S) to conform to the VSO canonical order. The object is temporarily kept in working memory (Chen, 1986; Carpenter and Just, 1989; King and Just, 1991) while the original object position of gap₁ is identified as following S. Finally, speakers establish a relationship between O₁ and gap₁, and ‘the fish’ is filled into the gap₁ position. The multiple steps required to use the active filler strategy should account for the extra processing time needed for the VOS scrambled order compared to the canonical VSO order.

In Tongan, there is a distinction in the topicalization of the subject and object. In the case of subject topicalization, a resumptive pronoun such as ‘ne’ (third person singular) appears in place of a gap in the relative clause (Otsuka, 2006), positioned after the past tense na’a (phonological change from na’e) and before the verb kai ‘eat’. In the English sentence, ‘This is the girl that I do not know what she said,’ ‘she’ in a relative clause refers to the previously mentioned noun ‘the girl’. Similarly, the resumptive pronoun ne in Tongan in Sentence (7) refers to e fefine ‘the woman’. In fact, the resumptive pronoun ne is necessary in the relativization of ergative NPs. Thus, topicalization behaves similarly to relativization in this aspect, indicating that it involves A-bar movement.

(7) S₁Vgap₁O: Subject topicalized transitive Tongan sentence

Ko e fefinei na’a nei kai ‘a e ika

TOP REF NP (woman) PST PRO V(eat) ABS REF NP (fish)

‘(It is) the woman that (she) ate the fish.’

Custis (2004) posits that topicalization results from a movement marked by ko. Therefore, according to the gap-filling parsing model, a topicalized subject phrase is shifted in front of the verb. Native Tongan speakers identify the initial topicalized phrase as the filler and then seek its original position in the specifier of the gap to establish filler-gap dependency. As this dependency extends beyond the verb, processing an SVO subject topicalized sentence might require even more processing time than a VOS scrambled sentence. It is noteworthy that both subject and object topicalization in Tongan also convey an exclusionary meaning, similar to what is seen in Korean and Japanese.

The object can also be topicalized by moving it in front of the verb. As with subject topicalization, the object is also marked by the topic marker ko, as seen in ko e ika (‘the fish’) in Sentence (8), illustrating object topicalization (O_TOP). However, unlike subject topicalization, object topicalization does not require a resumptive pronoun. Both Sentences (7 and 8) follow the structure ko NP₁ V NP₂. The object topicalization in Sentence (8) does not necessitate a resumptive pronoun.

(8) O₁VSgap₁: Object topicalized transitive Tongan sentence.

Ko e ika na’e kai ‘e he fefine.

TOP REF NP (fish) PST V(eat) ERG REF NP (woman).

‘About the fish, the woman ate.’

Object topicalization in Tongan may involve an even longer distance movement (O₁VSgap₁) than subject topicalization (S₁Vgap₁O) because the object moves ahead of the subject and the verb. Native Tongan speakers may read a topicalized sentence to find gap₁ at the end of the sentence as O₁VSgap₁. Once they establish the relationship between O₁ and gap₁, the topicalized O₁ is filled in gap₁. The distance between filler and gap in O₁VSgap₁ is longer than the distance between filler and gap in either VO₁Sgap₁ or S₁Vgap₁O. Therefore, based on the syntactic structure (moved distance), Tongan speakers keep O₁ in working memory longer for O₁VSgap₁ than VO₁Sgap₁ and S₁Vgap₁O. Thus, the gap-filling parsing model predicts that an OVS topicalized order will take longer to process than either a VOS scrambled or SVO topicalized order. Additionally, if native Tongan speakers can obtain the argument information from the verb at the beginning of the sentence, they could easily construct a whole sentence based on that information. The processing of argument-verb dependency would function well to enable the construction of a whole syntactic structure of VSO and VOS. Since the subject and the object are placed before the verb in topicalization, native Tongan speakers may have to process the backward verb-argument dependency. In such a case, SVO and OVS orders may be disadvantaged in processing compared to VSO and VOS orders: a distinct difference in reaction times and possibly in accuracy may be observed between the verb-initial orders (VSO and VOS) and the verb-second orders (SVO and OVS).

From the perspective of syntactic complexity based on filler-gap dependency, processing difficulties among the four-word orders in Tongan sentences can be predicted as follows: The VS_ERG O_ABS canonical order is processed the fastest. In the V O_ABS1 S_ERG gap₁ structure, the object is moved in front of the subject, while in the topicalized S_TOP1 V gap₁ O_ABS sentence, the subject phrase is moved before the verb. Due to the distance of one phrase movement, the speed of sentence processing would be assumed to be similar between the scrambled VO_ABS1 S_ERG gap₁ and topicalized S₁Vgap₁O orders. However, in object topicalization, the object phrase moves beyond both the verb and the subject phrases. Therefore, the O_TOP1 V S_ERG gap₁ object topicalized order has a much longer distance of filler-gap dependency, making it syntactically more complex than other word orders. This syntactically complex order would require the longest processing time among the four orders.

The present study conducted three experiments to investigate differences in word orders resulting from scrambling and topicalization. For native speakers, the accuracy of sentence correctness decisions tends to be consistently higher. This makes the reaction time required to determine whether a sentence is correct a more sensitive indicator. Therefore, sentence processing for various word orders can be predicted by processing speed.

Experiment 1 focused on (verb-final) Korean to compare the findings with those of Imamura et al. (2016) on Japanese. Given the syntactic similarities between Korean and Japanese, this experiment aimed to determine if a similar processing trend exists between the two languages. We anticipated that sentences in the S_NOM O_ACC V order would be processed most quickly because they are in the canonical order. Next, since topicalized sentences of S_TOP O_ACC V also have the same word order as the canonical order, we expected that they would be processed at the same speed as the canonical order, similar to Japanese sentences. Given that the scrambled order of O_ACC1 S_NOM gap₁V is known to be significantly slower than the canonical order, as shown by various studies (e.g., Mazuka et al., 2002; Ueno and Kluender, 2003; Koizumi and Tamaoka, 2004, 2010; Miyamoto and Takahashi, 2004; Tamaoka et al., 2005, 2014; Imamura et al., 2016; Witzel and Witzel, 2016; Tamaoka and Mansbridge, 2019), we expected O_TOP1 S_NOM gap₁V to be slower than both the canonical S_NOM O_ACC V and topicalized S_TOP O_ACC V orders. Furthermore, according to Imamura et al. (2016), the topicalized order of O_TOP1 S_NOM gap₁V was even slower than the scrambled order of O_ACC1 S_NOM gap₁V in Japanese, suggesting a similar result may be observed in Korean. Therefore, based on the filler-gap dependency (the movement-based anticipation), the speed of Korean sentence processing was predicted as follows:

Prediction 1: Korean Sentence Processing Based on a Filler Gap Dependency. S N O M O A C C V = S TOP O A C C V < O A C C 1 S N O M g a p 1 V < O TOP 1 S N O M g a p 1 V .

Experiment 2 involved a phrase-by-phrase processing experiment using a maze task to verify whether the VSO order is indeed canonical in the verb-initial language of Tongan. Similar to Experiment 1 in Korean, this experiment aimed to ascertain the canonical word order in Tongan sentence structure. Participants were presented with sentences having both VSO and VOS orders, broken down into phrases, and their processing speed for each phrase was measured. The results were then analyzed to determine whether the VSO order demonstrated superior processing efficiency compared to the VOS scrambled order, confirming its canonical status in Tongan sentence construction.

Experiment 3 in Tongan was conducted in a similar manner to Experiment 1 in Korean, enabling a comparative analysis of sentence processing strategies across the two languages. This experiment was designed to delve deeper into the effects of both scrambling and topicalization in Tongan sentence processing using a sentence correctness decision task. Specifically, it aimed to separate the influence of topicalization from the impact of canonical and scrambled orders. By comparing sentence processing across four different orders—VSO, VOS, SVO, and OVS—participants’ reaction times and accuracy rates were recorded and analyzed. This comprehensive investigation allowed for an understanding of how subject (SVO) and object topicalizations (OVS), in conjunction with canonical (VSO) and scrambled (VOS) orders, affect sentence processing in Tongan. Based on the filler gap dependency (movement-based anticipation), the speed of Tongan sentence processing was predicted as follows:

Prediction 2: Tongan Sentence Processing Based on a Filler Gap Dependency. V S ERG O ABS < V O ABS 1 S ERG g a p 1 = S TOP 1 V g a p 1 O ABS < O TOP 1 V S ERG g a p 1 .

The findings from these three experiments would contribute to our understanding of sentence processing mechanisms in both verb-final and verb-initial languages, shedding light on the universality and language-specific aspects of sentence structure and comprehension.

The results of the Korean sentence correctness decision task from Experiment 1 demonstrated a pattern similar to that observed in the study by Imamura et al. (2016) in Japanese, largely confirming Prediction 1, with the exception of no difference observed between O_ACC S_NOMV and O_TOP S_NOMV. First, akin to Japanese, the scrambled order in Korean was processed more slowly than the canonical order. Second, consistent with the findings of the previous Japanese experiment (Imamura et al., 2016), the subject topicalized order was processed at a comparable speed to the canonical order. This observation suggests that similar to Japanese, the S_TOPO_ACCV subject topicalized order in Korean might have been construed simply as the subject, akin to the canonical ordered sentence S_NOMO_ACCV. Third, unlike Japanese, there was no discernible difference in processing time between object topicalized and scrambled sentences. This discrepancy between Japanese and Korean could be attributed to the animacy effect, as Korean stimulus sentences featured animacy contrast, unlike the study by Imamura et al. (2016). Further elaboration on this aspect will be provided in the General Discussion section. Additionally, similar to Japanese, the SOV canonical order in Korean overlapped with the subject topicalized order, while the OSV scrambled order in Korean overlapped with the object topicalized order. Consequently, neither Japanese nor Korean could distinguish between the effects of scrambling and topicalization. Hence, a more comprehensive investigation is warranted in Tongan, which exhibits a distinct word order for canonical/scrambled and subject/object topicalization.

Experiment 2 utilized a maze task to assess the processing of Tongan AdvVSO canonical and AdvVOS scrambled orders phrase-by-phrase. The results concerning the third and fourth crucial phrases occurring after the verb indicated that native Tongan speakers process the VSO order faster than the VOS order. Consistent with previous linguistic studies (Churchward, 1953; Dixon, 1979, 1994; Otsuka, 2000, 2005a,b, 2010), the present psycholinguistic investigation confirmed that the VSO order is canonical in Tongan. Experiment 2 provided support for the scrambling effect (V S_ERG O_ABS < V O_ABS1 S_ERG gap₁) as depicted in Prediction 2 based on filler-gap dependency.

Experiment 3 addressed the primary objective of this study: examining the effects of scrambling and topicalization in Tongan sentences. The efficiency of processing for the four word orders (VSO, VOS, SVO, and OVS) was measured in terms of speed and accuracy. Similar to Experiment 2, the results indicated that VSO canonical sentences were processed more rapidly and accurately than VOS scrambled sentences, supporting the existence of a scrambling effect. Interestingly, OVS object topicalized sentences were processed as swiftly and accurately as VSO canonical sentences, while SVO subject topicalized sentences were processed more slowly than VSO canonical sentences but faster than VOS scrambled sentences. Both SVO subject and OVS object topicalized sentences were processed more efficiently than VOS scrambled sentences. The object topicalization sentences exhibited a positive (efficient) processing effect, whereas the subject topicalization sentences were processed less efficiently than object topicalization sentences but more efficiently than VOS scrambled sentences. These findings did not entirely align with Prediction 2 based on the filler-gap dependency. Further details were discussed in the subsequent General Discussion section.

Various syntactic similarities are found between Korean and Japanese. In both languages, a scrambled order is created by moving the object in front of the subject, as in O₁SVgap₁. The result of Experiment 1 indicated that the scrambled O₁SVgap₁ order in Korean was processed more slowly and less accurately than the SOV canonical order. This result confirmed the existence of the scrambling effect, as also found in previous studies on Japanese sentence processing (e.g., Mazuka et al., 2002; Ueno and Kluender, 2003; Koizumi and Tamaoka, 2004, 2010; Miyamoto and Takahashi, 2004; Tamaoka et al., 2005, 2014; Imamura et al., 2016; Witzel and Witzel, 2016; Tamaoka and Mansbridge, 2019). The processing delay for the syntactic structure of the O₁SVgap₁ scrambled order in both Japanese and Korean could be explained by the gap-filling parsing model (Frazier and Rayner, 1982; Stowe, 1986; Frazier, 1987; Frazier and Clifton, 1989; Frazier and Flores D’Arcais, 1989) which was presented in the introductory section of this article.

Regarding topicalization, in agreement with Imamura et al. (2016), the subject topicalized order in Korean was processed equally as quickly as the canonical order. However, it is still uncertain whether the processing speed of subject topicalization is accelerated by the influence of the SOV canonical order. Additionally, there were no differences in processing speed between the object topicalized order and the scrambled order in Korean. Once again, since the OSV scrambled order overlaps with the object topicalized order, the null difference result may have been caused by the influence of the OSV scrambled order and bears no relation to the object topicalization effect. It is quite possible that the scrambling effect may override the topicalization effect. It is also conceivable that the particle -eun/neun can be a pseudo-subject. This might be a partial reason for the fact that no difference has been found in this study between S_NOMOV and S_TOPOV in accuracy and reaction time.

One difference was found between Japanese and Korean. There was no difference in processing speed between the object topicalized and the scrambled order in Korean. By contrast, the object topicalized order was slower than the scrambled order in Japanese. This different result may be created by the animacy effect (Mak et al., 2002; Hsiao and Gibson, 2003; Pu, 2007; Vasishth et al., 2013; Kwon et al., 2019). In the Japanese study (Imamura et al., 2016), proper nouns such as Satoo and Suzuki were used for both subjects and objects. For example, サトウがスズキを褒めた (Satoo-ga Suzuki-o home-ta) meaning ‘Sato praised Suzuki’ contains two animate nouns. These two animate proper nouns can function as either subject or object. Unlike the Japanese study, the stimulus sentences used in the current Experiment 1 using Korean were mostly constructed by pairing an animate noun with an inanimate noun, such as in 여성이 물을 삼켰다 (Yeoseong-i mul-ul sam-keossda) meaning ‘The woman drank water.’ This animacy contrast between the subject and the object in Korean may have facilitated the processing of OSV object topicalized order, which is a plausible explanation for why there was no difference in the speed of sentence processing between OVS object topicalized and OSV scrambled orders. However, the animacy of noun phrases could be closely related to thematic assignments and argument structures (Comrie, 1981; Mak et al., 2002; Traxler et al., 2002, 2005; Bornkessel-Schlesewsky and Schlesewsky, 2006; Gennari and MacDonald, 2008; Wagers and Phillips, 2014; Ness and Meltzer-Asscher, 2017). This issue should be investigated in future studies.

Previous studies (e.g., Churchward, 1953; Dixon, 1979, 1994; Otsuka, 2000, 2005a,b, 2010; Custis, 2004) have suggested that the canonical order in Tongan is VSO, while the scrambled order is VOS. However, no large-scale corpus study has been conducted in Tongan to verify that the VSO order is dominant. Thus, before investigating the processing of subject/object topicalized sentences, a maze task was conducted in Experiment 2 to determine whether VSO is truly the canonical order of Tongan transitive sentences. The lexical maze task is a unique task in which participants make a series of continuous lexical decisions that reveal how they are processing sentences without the participants’ awareness of this background operation. This task makes it possible to compare phrase-by-phrase sentence processing between VSO and VOS orders.

Because the first phrase consisting of the adverb mahalo (‘maybe’) and the second phrase containing a verb is the same in AdvVSO canonical and AdvVOS scrambled sentences in the lexical maze task, no differences in processing speed for the first and second phrases were expected. The phrases of interest appear after the verb. In the third subject/object phrase, the noun with the ergative ‘e marker (S) was processed faster than the noun with the absolutive ‘a marker (O). Since VSO is considered the most used order, the subject would be expected to appear after the verb. In the following fourth object/subject phrase, the noun with the absolutive ‘a marker (O) was processed faster than the noun with the ergative ‘e marker (S). This result was surprising because after the third phrase was identified, the following final fourth phrase would automatically have been an object for the canonical or a subject for the scrambled order. Even after the processing of the third phrase, scrambling continues to affect the processing of the sentence-final subject. This could be due to the search for the gap after the subject in the AdvVO₁Sgap₁. As anticipated by Prediction 2, the processing difference between VSO and VOS in the third and fourth phrases in Experiment 2 supports the proposal that VSO is the canonical order.

The effect of scrambling in Experiment 2 was further confirmed in Experiment 3. The VSO canonical order was processed more quickly and accurately than the VOS scrambled order. As Otsuka (2005b, 2010) proposed, VOS scrambled order in Tongan is understood as movement (VO₁Sgap₁) motivated by a new information focus. Otsuka argues that the position immediately following the verb is reserved for new information. In other words, given a context in which the object is new information and the subject is old information (e.g., answering an object wh-question), the object is placed before the subject. According to this proposal, as in Japanese and Korean, gap-filling parsing (Stowe, 1986; Frazier, 1987; Frazier and Clifton, 1989; Frazier and Flores D’Arcais, 1989; Traxler and Pickering, 1996) may have been used for processing the Tongan VOS scrambled order. Consequently, Experiments 2 and 3 provided clear evidence for the linguistic claim that the VSO order is canonical and the VOS is scrambled in Tongan (Churchward, 1953; Dixon, 1979, 1994; Otsuka, 2000, 2005a,b, 2010).

Furthermore, some Tongan verbs can be used both as transitive and intransitive verbs (i.e., ambitransitivity). Since Tongan is an ergative language and the absolutive case marker ‘a also marks the subject of intransitive sentences, the VO_ABS order can be interpreted as being an intransitive canonical sentence in VS_ABS order. When a noun phrase marked by the ergative case e’ follows, the sentence is considered a transitive VO_ABSS_ERG scrambled sentence. In that case, it may be that the delay in processing a VO_ABSS_ERG ordered sentence is caused by verb transitivity confusion. Nine of 32 verbs in Experiment 3 fall into this ambitransitivity category; these are kai ‘eat’, talitali ‘wait’, ui ‘call’, tuli ‘chase’, huo ‘hoe’, fo ‘wash clothes’, inu ‘drink’, tui ‘wear’, and lau ‘read.’ However, the nouns used in Experiment 3 are basically a pair of animate nouns for the subject (e.g., ‘woman’, ‘teacher’, ‘children’) and inanimate nouns (e.g., ‘clothes’, ‘bicycle’, ‘money’) for the object. Transitivity confusion could be avoided by using animacy contrast in the stimulus sentences in Experiment 3. Thus, we do not consider the possibility of the influence of ambitransitivity to be very likely. Hence, it would be parsimonious to conclude that the differences in accuracy and speed were primarily driven by the effect of scrambling order: specifically, VOS scrambled order exhibited a 4.07% lower accuracy rate and was 110 ms slower in speed compared to the VSO canonical order.

The SVO and OVS sentence-fronting topicalized orders are always ko-marked (Custis, 2004), as shown in Sentences (7) and (8) above. Custis (2004) considers that both SVO and OVS orders involve A-bar movement similar to relativization. If that is the case, then both orders would be predicted to be recognized by native Tongan speakers as a more complex syntactic structure than the VSO canonical order.

According to the gap-filling parsing model (Stowe, 1986; Frazier, 1987; Frazier and Clifton, 1989; Frazier and Flores D’Arcais, 1989; Traxler and Pickering, 1996), the processing of Tongan topicalized sentences is described as follows: a topicalized subject phrase is moved in front of the verb; Tongan speakers recognize the initial topicalized phrase as the filler, and then look for its original position in the specifier of gap to establish the filler-gap dependency. In this framework, the OVS order should have a longer processing time than the SVO order because the movement involved in the object topicalization O₁VSgap₁ is greater in distance than the movement involved in the subject topicalization S₁Vgap₁O. In addition, the object relative clause (O₁VSgap₁ in Tongan) for many of the world languages has been found to be more difficult to process and comprehend than the subject-extracted (S₁Vgap₁ in Tongan) relative clause (e.g., Staub, 2010; Jäger et al., 2015).

However, an unexpected result was found in Experiment 3. There was no significant difference in processing speed and accuracy between the OVS object tropicalized order and the VSO canonical order (OVS=VSO). By contrast, the processing of the SVO subject topicalized order was slower than for the VSO canonical order (VSO < SVO) but faster than for the VOS scrambled order (SVO < OVS). The contrasting result between SVO and OVS may be related to the prominence of the topic. Custis (2004) considers Tongan to be a topic-marking language. A topic can be placed in one of two positions in Tongan—either immediately before the verb (SVO/OVS) or immediately after the verb (VOS). The placement of a topic is pragmatically motivated and syntactically results in one of the three-word orders of SVO, OVS, or VOS, while the VSO canonical order is pragmatically neutral. Custis (2004) observed (using the corpus data) that SVO and OVS orders are used to mark less salient topics and VOS to mark more salient topics in discourse. By contrast, the VOS scrambled order is used for presenting a salient topic. The activation levels infer cognitive status reflecting the frequency of a discourse topic: namely, how often would it be talked about? When the NP is frequently mentioned in discourse, it is more accessible and hence less salient in the discourse. This is the case for the fronted subject of SVO or the fronted object of OVS orders. On the other hand, a low activation level indicates that the topic is less frequently talked about so that it would be more salient and less accessible. This is the case for the object of VOS scrambled order.

The idea of cognitive status by Custis (2004) may account for the result of the processing speed in VSO < SVO < VOS. The pragmatically-neutral VSO is the fastest, while the VOS scrambled order is the slowest. The SVO subject topicalized order falls in between the VSO and VOS orders. Then, what causes the difference in processing speed between SVO and OVS orders? One possibility is that, as shown in Sentence (7), the resumptive pronoun ne appears after the tense and before the verb in the case of subject topicalization, so this visible syntactic feature may require extra processing time than the OVS object topicalized order with no resumptive pronoun. Prediction 2 (i.e., VSO < VOS = SVO < OVS) was constructed according to the syntactic complexity constructed by the filler-gap dependency (Stowe, 1986; Frazier, 1987; Frazier and Clifton, 1989; Frazier and Flores D’Arcais, 1989; Traxler and Pickering, 1996), regarding the word order in Tongan scrambling and topicalization. However, the results of Experiment 3 (i.e., VSO = OVS < SVO < VOS) indicated that word orders of both subject and object topicalization facilitated in sentence processing, deviating from the movement-based anticipation of Prediction 2.

In conclusion, the present study sheds light on the processing dynamics of Tongan sentence structures, offering insights into the cognitive mechanisms underlying language comprehension. The observed processing speeds, with VSO being the fastest, followed by SVO and then VOS, suggest a nuanced interplay between syntactic structure and cognitive processing. The unexpected processing efficiency of OVS orders compared to the predicted pattern raises intriguing questions about the underlying mechanisms at play. The discrepancy between the observed results and Prediction 2 highlights the limitations of relying solely on syntactic complexity metrics such as the filler-gap dependency. Instead, the findings suggest that the facilitation effect of both subject and object topicalization in Tongan may be better explained by a broader consideration of cognitive status and pragmatic factors. Further investigations into the processing dynamics of topicalized word orders in Tongan, as well as comparative studies across different languages, are warranted to deepen our understanding of how linguistic structure interacts with cognitive processing mechanisms. This study opens avenues for future research into the intricate interplay between syntax, cognition, and language comprehension.