Is the subject-before-object preference universal? An event-related potential study in the Kaqchikel Mayan language

Transcription

1 Language, Cognition and Neuroscience ISSN: (Print) (Online) Journal homepage: Is the subject-before-object preference universal? An event-related potential study in the Kaqchikel Mayan language Daichi Yasunaga, Masataka Yano, Yoshiho Yasugi & Masatoshi Koizumi To cite this article: Daichi Yasunaga, Masataka Yano, Yoshiho Yasugi & Masatoshi Koizumi (2015) Is the subject-before-object preference universal? An event-related potential study in the Kaqchikel Mayan language, Language, Cognition and Neuroscience, 30:9, , DOI: / To link to this article: The Author(s). Published by Taylor & Francis Published online: 23 Sep Submit your article to this journal Article views: 254 View related articles View Crossmark data Citing articles: 1 View citing articles Full Terms & Conditions of access and use can be found at Download by: [Tohoku University] Date: 18 October 2015, At: 18:41

2 Language, Cognition and Neuroscience, 2015 Vol. 30, No. 9, , Is the subject-before-object preference universal? An event-related potential study in the Kaqchikel Mayan language Daichi Yasunaga a, Masataka Yano b, Yoshiho Yasugi c and Masatoshi Koizumi d* a Faculty of Letters, Institute of Human and Social Sciences, Kanazawa University, Kakuma-machi, Kanazawa, Ishikawa , Japan; b Graduate School of Humanities, Kyushu University, , Hakozaki, Higashi-ku, Fukuoka , Japan; c Department of Cultural Research, National Museum of Ethnology, 10-1 Senri Expo Park, Suita, Osaka , Japan; d Department of Linguistics, Graduate School of Arts and Letters, Tohoku University, 27-1, Kawauchi, Aoba-ku, Sendai, Miyagi , Japan (Received 30 June 2013; accepted 30 July 2015) The processing load of sentences with different word orders in the Kaqchikel Mayan language was investigated using eventrelated potentials. We observed a P600 for subject-verb-object and verb-subject-object sentences as compared to verb-objectsubject (VOS) sentences, suggesting that VOS order is easier to process than the other orders. This is consistent with the traditional interpretation in Mayan linguistics that the syntactically determined basic word order is VOS in Kaqchikel, as in many other Mayan languages. More importantly, the results revealed that the preference for subject-object word order in sentence comprehension observed in previous studies may not be universal; rather, processing load in sentence comprehension is greatly affected by the syntactic nature of individual languages. Keywords: Basic word order; field-based psycholinguistics; Guatemala; processing load; syntactic complexity 1. Introduction In many flexible word order languages, sentences with a transitive verb (V) in which the subject (S) precedes the object (O) (SO Word Order = SOV, SVO, VSO) are reported to induce a lower processing load in comprehension than those in which the opposite occurs (OS Word Order = OSV, OVS, VOS) (e.g. Bader & Meng, 1999; Kaiser & Trueswell, 2004; Mazuka, Itoh, & Kondo, 2002; Sekerina, 1997; Tamaoka et al., 2005). The question naturally arises as to why this should be the case, together with the related question of whether this difference in processing loads is universal, holding of all human languages. In the psycholinguistic literature, a number of suggestions have been made regarding the factors affecting word order preference in sentence comprehension and production. These may be divided into two broad theoretical positions (Koizumi et al., 2014). One view, which we refer to as Individual Grammar Theory, posits that grammatical factors of individual languages such as syntactic complexity primarily determine the word order preference in each language, that is, a language s syntactically determined basic word order is preferred to other possible word orders (see Gibson, 2000; Hawkins, 2004; Marantz, 2005; O Grady, 1997; Pritchett & Whitman, 1995, among others). In contrast, what may be called Universal Cognition Theory hypothesises that word order preferences are largely attributable to human cognitive features that are universal (e.g. conceptual accessibility) so that SO word order is preferred regardless of the basic word order of any individual language (Bornkessel-Schlesewsky & Schlesewsky, 2009a, 2009b; Kemmerer 2012; Tanaka, Branigan, McLean, & Pickering, 2011, to mention just a few.). Both these theories correctly predict that SO word order is preferred in languages in which the subject precedes the object in the syntactically basic word order (SO languages). Sentence-processing studies conducted so far have all targeted SO languages, except for a few recent ones such as Clemens et al. (2015) and Norcliffe, Konopka, Brown, and Levinson (2015). Hence, it remains unclear whether the preference for the SO word order is a reflection of word order in individual languages or a reflection of more universal human cognitive features. To determine which of these two theories is correct, it is necessary to examine languages for which the two theories offer different predictions, namely, languages in which the object precedes the subject in the syntactically basic word order (OS languages). Here, we report the results of an event-related potential (ERP) study of Kaqchikel, a Mayan language spoken in Guatemala. Although, in general, the word order of Kaqchikel is flexible, the syntactically determined basic word order of Kaqchikel is VOS *Corresponding author. masatoshi.koizumi.a4@tohoku.ac.jp 2015 The Author(s). Published by Taylor & Francis This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License ( by-nc-nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited, and is not altered, transformed, or built upon in any way.

3 1210 D. Yasunaga et al. (García Matzar & Rodríguez Guaján, 1997, p. 333). We found that the late positive ERP components (P600 or syntactic ERP) were elicited by SVO and VSO sentences as compared to VOS sentences, which indicates that the processing load of VOS is lower than that of the two other commonly used word orders. This suggests that the preference for SO word order in sentence comprehension may not be universal, but rather that syntactic features of individual languages significantly influence sentence-processing load. 2. Previous cross-linguistic work: SO word order preference Evidence that SO word orders are easier to process than OS word orders in flexible word order languages is abundant in the psycholinguistic and neurolinguistic literature. For instance, a study has found that in terms of behavioural indices, Japanese readers take less time to judge whether a sentence makes sense when it has an SOV word order than when it has an OSV word order (Tamaoka et al., 2005). Other studies, using self-paced reading and eye-tracking methodologies, have found longer reading times for OSV as compared to SOV sentences in Japanese (Imamura & Koizumi, 2008; Mazuka et al., 2002). Parallel results from the processing of orthographically and phonologically presented sentences have been reported for many other languages (see, e.g. Kaiser & Trueswell, 2004, for Finnish; Kim, 2012, forkorean; Sekerina, 1997, for Russian; Tamaoka, Kanduboda, & Sakai, 2011, for Sinhalese). In terms of neurophysiological indices, studies using functional magnetic resonance imaging have found that there is a greater activation of the left inferior frontal gyrus during the processing of OS word orders compared to SO word orders (Grewe et al., 2007, for German; Kim et al., 2009; and Kinno, Kawamura, Shioda, & Sakai, 2008, for Japanese). ERP research also supports the claim that SO word orders are easier to process. For instance, compared with SO orders, OS orders elicit a P600 and/or (sustained) anterior negativity components, suggesting that processing OS word orders place a larger load on working memory (Erdocia, Laka, Mestres-Missé, & Rodriguez-Fornells, 2009, for Basque; Roesler, Pechmann, Streb, Roeder, & Hennighausen, 1998, for German; Ueno & Kluender, 2003; and Hagiwara, Soshi, Ishihara, & Imanaka, 2007, for Japanese). Thus, solid evidence exists to support the notion that SO word orders are preferred to OS word orders in many languages of the world. The question then arises as to the sources of this preference in sentence comprehension. A possibility that immediately comes to mind is that it is primarily due to syntactic canonicity (i.e. individual grammar theory). According to many sentence-processing theories, a language s syntactically determined basic word order is easier to process than other grammatically possible but non-canonical derived word orders in the language (e.g. Gibson, 2000; Hawkins, 2004; O Grady, 1997; Pritchett & Whitman, 1995). Thus, according to the individual grammar theory, SO word orders were preferred in previous studies because they were the syntactically basic word orders in the target languages. An alternative explanation is that the SO word order preference in sentence comprehension may be largely attributable to human cognitive features that are more universal (i.e. universal cognition theory). That there may be such features is strongly suggested by the fact that a vast majority of the world s languages have one of the SO word orders as the basic word order (SOV 48%, SVO 41%, VSO 8%, VOS 2%, OVS 1%, and OSV 0.5%, according to Dryer, 2005; see also Gell-Mann & Ruhlen, 2011). In particular, several studies have shown that entities that are prominent as a result of properties such as agency, animacy, concreteness, prototypicality, and prior mention in the discourse tend to appear as sentenceinitial subjects (cf. Branigan, Pickering, & Tanaka, 2008; Bock & Warren, 1985; Bornkessel-Schlesewsky & Schlesewsky, 2009a; Hirsh-Pasek & Golinkoff, 1996; Primus, 1999; Slobin & Bever, 1982). Universal cognition theory, therefore, leads to the expectation that SO word order has a low processing load regardless of the basic word order of any individual language; this is consistent with what has been reported in the literature so far. Both individual grammar theory and universal cognition theory correctly predict the SO word order preference in sentence comprehension in SO languages. However, their predictions diverge when it comes to OS languages. According to individual grammar theory, OS word orders should be processed faster than SO word orders in these languages. However, universal cognition theory predicts that the opposite should be the case. It is therefore necessary to study OS languages to test these theories. To this end, Koizumi et al. (2014) conducted an experiment with a sentence plausibility judgement task in Kaqchikel, a VOS Mayan language with a flexible word order. In this experiment, grammatical transitive sentences, either semantically plausible or implausible, in three different word orders (i.e. VOS, VSO, and SVO) as well as filler sentences were aurally presented in a random order to the participants through headsets. The participants were asked to judge whether each sentence was semantically plausible and to push a YES button (natural sentence) or NO button (unnatural sentence) as quickly and accurately as possible according to their judgement. The time from the beginning of each stimulus sentence until the button press was measured as the reaction time. They found that semantically plausible sentences in VOS order were processed faster than those in SVO or VOS order. This suggests that VOS, an OS order, is preferred to SVO and VSO, both of which are SO orders, in Kaqchikel

4 Language, Cognition and Neuroscience 1211 sentence comprehension. Based on these results, Koizumi et al. (2014) concluded that the SO order preference in sentence comprehension may not be universal. The test items used in Koizumi et al. (2014) were nonreversible sentences (i.e. sentences with an animate subject and an inanimate object), but the results were replicated with reversible sentences (i.e. sentences with an animate subject and an animate object) as reported in Kiyama, Tamaoka, Kim, and Koizumi (2013). There are, however, at least two remaining questions. One has to do with contextual effects. As explicitly mentioned and discussed in Koizumi et al. (2014) as well as in Section 3 of the present paper, in Kaqchikel, VOS order can be felicitously used in a wide range of contexts, including the absence of any substantial context, whereas SVO is frequently used in contexts where the subject is a topic (Ajsivinac Sian, García Mátzar, Cutzal, & Alonzo Guaján, 2004, pp ; García Matzar & Rodríguez Guaján, 1997, p. 334; Tichoc Cumes et al., 2000, pp ). It is therefore likely that the higher processing load of SVO in their experiment is at least partially attributable to the lack of felicitous contexts. SVO sentences might be easier to process than VOS sentences if an appropriate context is given. In other words, it is not clear the extent to which the OS preference in Kaqchikel found in the Koizumi et al. (2014) and Kiyama et al. (2013) experiments is due to syntactic rather than contextual factors. Another weakness of these studies has to do with the index of processing load used. They measured reaction times for sentence plausibility judgements. Although reaction times for sentence plausibility judgements provide us with useful data reflecting the processing load of a sentence as a whole, they convey no information about the time course of sentence processing. Accordingly, we cannot determine, based on them, whether the processing load increased in the expected sentence regions at the expected times. In order to overcome these difficulties, in the present study, we recorded ERPs during a picture sentence matching task. In this task, a picture was presented in the centre of a computer screen in front of the participant and, after the picture disappeared, a Kaqchikel sentence was aurally presented through a headset. The task of the participant was to judge whether the meaning of the sentence was congruent with the content of the picture. The measurement of ERPs enabled us to investigate the time course of sentence processing. Moreover, there is good reason to believe that the presentation of a picture prior to the corresponding sentence provided an appropriate non-verbal context, not only for VOS word order, but also for SVO word order. For instance, in a production study that used a picture description task, Kaqchikel speakers used SVO word order more than 70% of the time (Kubo, Ono, Tanaka, Koizumi, & Sakai, 2012; see also Section 4). 3. Kaqchikel Kaqchikel, previously spelled Cakchiquel, is one of the 21 Mayan languages spoken in Guatemala. It is mainly used in the highlands west of Guatemala City; it has over 450,000 speakers (Brown, Maxwell, & Little, 2006, p. 2; Lewis, 2009; Tay Coyoy, 1996, p. 55). Like other Mayan languages, Kaqchikel is headmarking. Subjects and objects are unmarked, and person and number agreement for both subjects and objects is obligatorily expressed on the verb. Kaqchikel is ergative, like other Mayan languages. The order of morphemes in the verb is [Aspect-Absolutive-Ergative-Verb stem]. 1 An example is given in (1) below. (1) Y-e -in-to IC-ABS3pl-ERG1sg-help I help them. Because Kaqchikel is a pro-drop language, (1) functions as an independent sentence. Like many other Mayan languages, Kaqchikel allows different grammatical word orders. However, according to García Matzar and Rodríguez Guaján (1997) and others, its basic word order is VOS, as exemplified in (2a), in which neither the subject nor the object is topicalised or focused (Ajsivinac Sian et al., 2004, p. 162; García Matzar & Rodríguez Guaján, 1997, p. 333; Rodríguez Guaján, 1994, p. 200; Tichoc Cumes et al., 2000, p. 195). VOS is thus typically used in a neutral context. If the sentence is semantically irreversible, VSO is also possible as in (2b), although VOS is favoured. (2) a. X-Ø-u-chöy ri chäj ri ajanel. [VOS] CP-ABS3sg-ERG3sg-cut DET pine.tree DET carpenter b. X-Ø-u-chöy ri ajanel ri chäj. [VSO] CP-ABS3sg-ERG3sg-cut DET carpenter DET pine.tree The carpenter cut the pine tree. In cases like (3a,b), where the sentence is semantically reversible (i.e. it makes sense when the object and subject are reversed), a VOS interpretation is overwhelmingly favoured (even though a VSO interpretation is still possible). (3) a. X-Ø-r-oqotaj ri me s ritz i. CP-ABS3sg-ERG3sg-run.after DET cat DET dog The dog ran after the cat. b. X-Ø-r-oqotaj ri tz i ri me s. CP-ABS3sg-ERG3sg-run.after DET dog DET cat The cat ran after the dog. In the Kaqchikel of the sixteenth century, VSO was employed when the object was complex. In current

5 1212 D. Yasunaga et al. speech, VSO sentences are only used in very restricted contexts (García Matzar & Rodríguez Guaján, 1997, p. 342). SVO is used when the subject is topicalised by moving it before the verb (García Matzar & Rodríguez Guaján, 1997, p. 334). Similarly, in OVS sentences, the object has been fronted as a topic (García Matzar & Rodríguez Guaján, 1997, p. 335). These are illustrated in (4). (4) a. Ri ajanel x-ø-u-chöy ri chäj. [SVO] DET carpenter CP-ABS3sg-ERG3sg-cut DET pine.tree The carpenter cut the pine tree. b. Ri chäj x-ø-u-chöy ri ajanel. [OVS] DET pine.tree CP-ABS3sg-ERG3sg-cut DET carpenter The carpenter cut the pine tree. Thus, all the word orders other than VOS are pragmatically and syntactically marked. 2 Furthermore, not only arguments, but also adverbs and adjuncts can be topicalised to the left of the verb, giving rise to a derived word order. This is exemplified in (5). When certain types of adjuncts (locative, instrumental, etc.) are preposed, the particle wi is inserted in the original position of the fronted constituent, as illustrated in (5b) (García Matzar & Rodríguez Guaján, 1997, p. 349; cf. also Yasugi, 2005). (5) a. Aninäq x-ø-u-b in-isa-j la ak wal la achi rapidly CP-ABS3sg-ERG3sg-walk-CAU-VT DET girl DET man The man made the girl walk rapidly. (Tichoc et al., 2000, p.228) b. Wawe Iximche n-ø-u-b än wi ri r-ochoch here Iximche IC-ABS3sg-ERG3sg-do WI DET POS3sg-house ri a Waqi Kej. DET CL Waqi Kej Here in Iximche, Waqi Kej builds his house. (García Matzar & Rodríguez Guaján, 1997, p. 354) It is thus clear that the preverbal position is not the dedicated subject position in Kaqchikel. For these and other reasons, many Mayan language researchers consider the syntactically determined basic word order of modern Kaqchikel to be VOS (Ajsivinac Sian et al., 2004, p. 162; García Matzar & Rodríguez Guaján, 1997, p. 333; Rodríguez Guaján, 1994, p. 200; Tichoc et al., 2000, p. 195). 3 Although precise syntactic structures of Mayan languages are still under debate, for the purpose of the present study, it is sufficient to assume, for Kaqchikel transitive sentences with different word orders, the schematic syntactic structures shown in (6), in which VOS is structurally simpler than the other orders (cf. England, 1991; Preminger, 2011; Tada, 1993; see also Aissen, 1992 and Coon, 2010, among many others, for other Mayan languages). (6) Order Schematic syntactic structure VOS [VOS] VSO [[V gap i S] O i ] SVO [S i [VO gap i ]] OVS [O i [V gap i S]] It should be noted at this point that even though Kaqchikel has VOS as its syntactically basic word order, it is the SVO order that is most frequently used in this language (England, 1991, p. 472; Kubo et al., 2012; Maxwell & Little, 2006, p. 102; Rodríguez Guaján, 1994, p. 201). According to Kubo et al. (2012), for example, of all the sentences with a transitive verb and nominal subject and object produced in their sentence production experiment with a picture description task, sentences with the SVO, VOS, and VSO word order constituted 74.4%, 24.2%, and 1.4% of the total sentences produced, respectively (see also Kubo, Ono, Tanaka, Koizumi, & Sakai, 2015). SVO, in which the subject is topicalised, is most frequently employed partly because it produces cohesion in discourse. Furthermore, in some dialects/idiolects of Kaqchikel (e.g. Patzún Kaqchikel), VOS sentences are only interpreted as questions when both the subject and object are definite and at an equal level of animacy (Kim, 2011; see also England, 1991). For these reasons, some researchers suggest that at least some dialects of Kaqchikel may be shifting or have shifted from a VOS to an SVO language (Brown et al., 2006; England, 1991; Kim, 2011). However, if indeed a part of the modern Kaqchikel community is currently shifting from using VOS to SVO as the syntactically basic word order, this shift has not yet been reflected in the internal grammar of the majority of native Kaqchikel speakers, as suggested, for example, by the fact that the VOS order was processed faster than the SVO order in several behavioural experiments (e.g. Kiyama et al., 2013; Koizumi et al., 2014) as well as the fact that the participants of the present study accept declarative VOS sentences more than 95% of the time (see Section 5.5.1). Not only in Kaqchikel but also in many Mayan languages, word orders in which subjects are preposed appear more frequently than the syntactically determined basic word order. Therefore, it has been suggested that when examining the basic word order of Mayan languages, syntactically determined word order from the standpoint of syntactic complexity needs to be distinguished from pragmatically determined word order, commonly used for pragmatic purposes (Brody, 1984; England, 1991). The discrepancy between the most frequently used word order and syntactically basic word order, if framed in terms of word order differences, might appear to be unique to OS languages. However, from a

6 Language, Cognition and Neuroscience 1213 more general viewpoint, similar situations are also observed in SO languages such as Korean and Japanese. Take Japanese, an SOV language, for example. In Japanese, the subject is marked with the nominative case marker, and the object with the accusative case marker, in pragmatically neutral contexts. When the referent of the subject is a discourse topic, the subject is preposed and marked with the topic marker (Kitagawa, 1982; Kuroda, 1988; Saito, 1985; Shibatani, 1990; Tateishi, 1990; among many others). This is schematically shown in (7) (Shibatani, 1990, p. 274). (7) a. [S-nom O-acc V] b. [S-top [ O-acc V]] [S-nom O-acc V] vs. [S-top [ O-acc V]] in Japanese seems to be parallel to VOS vs. SVO in Kaqchikel. [S-nom O-acc V] in Japanese and VOS in Kaqchikel are syntactically simple, and used typically in pragmatically neutral contexts. [S-top [ O-acc V]] and SVO are syntactically more complex and used in contexts where the subject is a topic. The production frequencies of [S-top [ O-acc V]] and SVO are several times higher than those of [S-nom O-acc V] and VOS, respectively (see Imamura & Koizumi (2011) for the production frequency in Japanese). It may, therefore, be the norm rather than an exception that sentences with a topicalised subject (which is presumably associated with the most commonly used information structure) are more frequently produced than corresponding sentences with a non-topicalised subject in languages that morpho-syntactically distinguish between the two kinds of subject. Viewed this way, there is nothing surprising about the fact that in Kaqchikel, SVO with a topicalised subject is used more frequently than VOS with a non-topicalised subject (see also Section 6). 4. Why Kaqchikel? As pointed out in Section 1, to test the two theories (i.e. individual grammar theory and universal cognition theory), it is necessary to examine OS languages, for which the two theories make different predictions. Given that Kaqchikel is an OS language with the features described in Section 3, the following predictions were made about processing load in the comprehension of Kaqchikel sentences: if the preference for SO word order shown by speakers of SO languages is closely related with the syntactic canonicity of the individual language, as suggested by individual grammar theory, VOS sentences will have a lower processing load than sentences with the other word orders shown in (6) above. On the other hand, if SO triggers a lower processing load than OS regardless of the basic word order of the individual grammar, as suggested by universal cognition theory, then Kaqchikel VOS sentences should create a greater processing load than SVO and VSO sentences. Another advantage of using Kaqchikel in this study has to do with production frequency. As mentioned in the previous section, the SVO order is used more frequently than the basic VOS order. In the psycholinguistic literature, there are cases in which the frequency with which the words and sentence structures appear affects the sentence-processing load (e.g. Trueswell, Tanenhaus, & Kello, 1993). That is, speakers of the language are more proficient in sentence structures and words that are used frequently, and they are more likely to process these with great speed and a high level of accuracy. Thus, it is interesting to observe how production frequency influences sentence processing in Kaqchikel. We will return to this issue in Section Experiment 5.1. Participants Sixteen native speakers of Kaqchikel participated in the experiment (7 females and 9 males, M = years, SD = 11.24). The place of origin and residence of the participants were distributed evenly throughout a wide range of the central Guatemala highlands, without any concentration on a particular region/dialect. They also used Spanish in daily life. All participants had normal or corrected-to-normal vision and self-reported as being righthanded. Prior to the experiment, written informed consent was obtained from each participant. They were paid for their participation. Approval for the study was obtained from the Ethics Committee of the Graduate School of Arts and Letters, Tohoku University Method and stimuli Electroencephalogram (EEG) was recorded while the participant engaged in the picture sentence matching task. In this task, a picture was presented in the centre of a computer screen for three seconds and, after the picture disappeared, a Kaqchikel sentence was aurally presented through a headset. 4 After the sentence ended, a question mark appeared in the centre of the screen. The participant was instructed to answer, upon seeing a question mark, whether the content of the picture was congruent with the meaning of the sentence by clicking a mouse button to indicate either YES or NO (Figure 1). A practice session was carried out to allow the participants to habituate to the task. As a general rule, the experimental session was started after the participant s accuracy exceeded 80% in the practice session.

7 1214 D. Yasunaga et al. Figure 1. Design of the task used in the experiment. Participants were instructed to judge whether the sentence they heard was congruent with the preceding picture, and to respond by pressing one of two buttons (indicating YES or NO ). EEGs were recorded while participants listened to the sentences. Each picture used in this experiment depicted a transitive action describable with one of the following six verbs commonly used in Kaqchikel: ch äy hit, jïk pull, nïm push, oyoj call, pixab aj bless, and xib ij surprise. Either the agent(s) or the patient(s) consisted of two persons, and the other consisted of just a single person. The agent(s) and patient(s) were painted in different colours; the colours used were the four colours describable with the following familiar colour terms in Kaqchikel: käq red, xar blue, säq white, and q ëq black. Transitive sentences congruent with the contents of the pictures were arranged into each of the four word orders (VOS, VSO, SVO, and OVS) as exemplified in (8). 5 Forty-eight sets of sentences yielding a total of 192 target sentences were created in this way. All the target sentences were so-called reversible sentences, with a definite human subject, definite human object, and action verb in past tense. In order to morpho-syntactically differentiate the two argument roles, half of the sentences contained a singular subject and plural object, whereas the other half contained a plural subject and singular object. As these 192 sentences were congruent with the contents of the corresponding pictures, the correct response is YES for these sentences. (8) a. Xkoyoj/ri xar/ri taq käq. [VOS] CP-ABS3sg-ERG3pl-call/ DET blue/ DET PM red The reds called the blue. b. Xkoyoj/ri taq käq/ri xar. [VSO] c. Ri taq käq/xkoyoj/ri xar. [SVO] d. Ri xar/xkoyoj/ri taq käq. [OVS] Another 96 sentences were created such that they did not match the corresponding pictures (in that the roles of the subject and object were reversed, the colour of either the subject or object was incorrect, or the action depicted by the verb was incorrect); thus, the correct response to these sentences was NO. A total of 288 stimuli sentences were presented to each participant in random order. The sentences were recorded by a male native Kaqchikel speaker. The length of time duration of each sentence was edited in Praat ver to create an equal duration across the four word order conditions by slightly shortening the duration of some pauses between phrases. There was no particular word order condition whose sentences were significantly more heavily edited (in terms of the total duration shortened) than the sentences of any other conditions. After the editing, all the test items were judged as natural in terms of prosody by our native Kaqchikel consultants. The averages and standard deviations of time duration for word order are given in Table 1. A oneway analysis of variance (ANOVA) showed no significant differences among the word orders in terms of time duration between the onset and offset of the sentence [F(3, 141) = 0.986, p =.40, ns.]. The stimulus onset asynchrony (SOA) between the first and second phrases (i.e. the duration of Region 1) and the SOA between the second and third phrases (i.e. the duration of Region 2) are more than 900 ms in all the four conditions. Thus, it can be safely assumed that EEG signals up to around 900 ms after the onset of each phrase were not affected by the input of the subsequent phrase. Phrases directly compared with each other (as stated in Section 5.3) were comparable and not statistically different in duration. In particular, there was no main effect of word order among the four conditions on the time duration of the third region [F(3, 141) = 1.08, p =.36]. The location of trigger pulses in stimuli for EEG recordings/analyses was determined in manner of visual inspection of the speech waveform of each sentence and in an auditory manner, using Praat, by the experimenters. The trigger pulses were synchronised with the onset of each region for all stimulus sentences, as illustrated in Figure 2 with sentence (8a) as an example.

8 Language, Cognition and Neuroscience 1215 Table 1. Mean duration of semantically plausible transitive sentences ( Yes items) of each word order. (M = mean, SD = standard deviation). Whole sentence (ms) Duration of R1 (ms) Duration of R2 (ms) Duration of R3 (ms) Word order M SD M SD M SD M SD VOS VSO SVO OVS Note: n = Predictions Each stimulus sentence consisted of V, O, and S. The sets of words used for S and O were identical to each other across the four word order conditions. Thus, the ERP components elicited by S and O could be directly compared. The set of words used for V were different from those used for S and O in a number of respects, including meaning, grammatical category, production frequency, and the number of phonemes and morphemes. As these factors affect the distribution of ERP components (e.g. Bornkessel-Schlesewsky & Schlesewsky, 2009b), we did not compare the ERP responses elicited by S and O, on the one hand, and V, on the other hand. We used a late positive ERP component, commonly referred to as the P600, as an index of an increased fillergap integration cost. It has been repeatedly observed that the greater syntactic processing difficulty causes a large positive-going shift, broadly distributed over the scalp, after presentation of the difficulty-inducing word. This is referred to as the P600 effect (Hagoort, Brown, & Groothusen, 1993; Osterhout et al., 1992). This effect is rather robust, and is often observed across types of difficulty with processing loads against word order, verb subcategorisation, syntactic reanalysis, dependency formation (Kaan & Swaab, 2003b), types of languages including word order languages (Dutch; Hagoort, & Brown, 2000), casemarked languages (Japanese; Hagiwara et al., 2007), and isolating languages (Chinese; Yang, Charles, & Liu, 2010), and various methodological factors, such as modality of the stimulus (i.e. visual or auditory). A P600 effect is also observed for syntactically ill-formed sentences (e.g. The hungry guests helped himself to the food; Osterhout & Mobley, 1995), thematically reversed Figure 2. Trigger pulses were synchronised to the onset of each region for all sentences. sentences (e.g. The hearty meal was devouring the kids; Kim & Osterhout, 2005), and picture sentence mismatches (e.g. The triangle stands in front of the square, after presenting a picture depicting a triangle behind a square; Vissers, Kolk, Van de Meerendonk, & Chwilla, 2008) (for more general review, see Bornkessel-Schlesewsky & Schlesewsky, 2009b). However, only grammatical sentences congruent with a preceding picture were analysed in the present study. In addition, although grammatically non-preferred continuation elicits a P600, our experimental stimuli did not include such continuation requiring revision of an initial syntactic structure (e.g. the man is painting the house and the garage is already finished) (Kaan & Swaab, 2003a). Therefore, the P600 effect in the experiment is correlated with filler-gap integration difficulty. In particular, it has been observed that the P600 is elicited at a gap position regardless of whether the corresponding filler precedes or follows the gap. For example, Yano, Tateyama, and Sakamoto (2014) examined the processing of a gap-filler (rather than filler-gap) dependency in Japanese cleft constructions, and found P600 effects at the pre-filler gap position, but no P600 effect at the subsequent filler position. Another possible interpretation of the late positive components has to do with the experimental task. In the present experiment, the pictures were presented before the sentences, and a decision was required after sentence offset. Thus, the positivities observed might be a variant of P300, indexing decision-related processes (Coulson, King, & Kutas, 1998; Haupt, Schlesewsky, Roehm, Friederici, & Bornkessel-Schlesewsky, 2008; Sassenhagen, Schlesewsky, & Bornkessel-Schlesewsky, 2014). This, however, is unlikely for the following reason. Only those sentences that were correctly marked YES were included for the analysis of ERPs, and given the design of the experimental sentences described above, decisions for the YES sentences cannot be made until encountering the third regions (e.g. the S of VOS). Thus, there is no reason to believe that decisionrelated processes differed across the four word order conditions, yielding the decision-related P3. Given this, if SVO sentences involve a filler-gap dependency, as schematically shown in (6) above, we

9 1216 D. Yasunaga et al. Table 2. Word order Accuracy (%) in the picture sentence matching task. Accuracy (%) VOS VSO SVO OVS Note: n = 16. should observe a late positive ERP component, that is, the P600, in response to the third region (i.e. O) of SVO sentences as compared to the third region of canonical VOS sentences (i.e. S), as the P600 is elicited when a filler is integrated with a corresponding gap position or, more generally, when processing load increases. In addition, we expected to observe a P600 component in response to the second, rather than the third, region of the VSO sentences (i.e. S) as compared to the second region of the VOS M SD sentences (i.e. O). As for OVS sentences, if they included a gap after the verb, as represented in (6), we would expect P600 effects upon encountering the verb (second region) when compared with SVO sentences, which in turn would elicit syntactic positivity at the third region (i.e. for the O of SVO relative to the S of OVS). We might also observe an N400 for the O of OVS relative to the S of SVO due to the extremely low production frequency of O initial sentences Data analysis The EEG for the congruent sentences was recorded from 17 Ag-AgCl electrodes located at Fz, Cz, Pz, F7, F8, F3, F4, T7, T8, C3, C4, P7, P8, P3, P4, O1, and O2 according to the International system (Jasper, 1958). The linked earlobe served as a reference. All electrode impedances were kept below 5 kω. Additional electrodes were placed on the left side of the left eye and beneath the left eye to monitor eye movements and blinks for later rejection. Band-pass filtering was set from 0.01 to 50 Hz. Recorded Figure 3. Panel A: The grand averaged ERP waveforms for the third region in the SVO (O, blue lines) and VOS (S, black lines) word order conditions for all 17 electrode sites. Negativity is plotted upward. Panel B: Voltage maps for the difference ERPs (SVO-VOS) for every 100 ms from 300 to 800 ms.

10 Table 3. Statistical results for the third region of SVO versus VOS. The third region of SVO versus VOS ms ms ms ms midline parasagittal temporal midline parasagittal temporal midline parasagittal temporal midline Parasagittal temporal WordOrder F-value Significance ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. W*Anteriority F-value Significance ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. W*Hemisphere F-value N/A N/A N/A N/A Significance N/A ns. ns. N/A * ns. N/A ns. + N/A * * W*A*H F-value N/A N/A N/A N/A Significance N/A + ns. N/A ns. ns. N/A ns. + N/A ns ms ms ms ms midline parasagittal temporal midline parasagittal temporal midline parasagittal temporal midline Parasagittal temporal WordOrder F-value Significance * ** ** ns. * + ns. * ns. + * ns. W*Anteriority F-value Significance ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. W*Hemisphere F-value N/A N/A N/A N/A Significance N/A * * N/A * *** N/A ns. ns. N/A + ns. W*A*H F-value N/A N/A N/A N/A Significance N/A ns. * N/A ns. ns. N/A ns. *** N/A ns. * Note: ns.: p >.10, +: p <.10, N/A: not applicable. *p <.05. **p <.01. ***p <.005. ****p <.001. Language, Cognition and Neuroscience 1217

11 1218 D. Yasunaga et al. Figure 4. Panel A: The grand averaged ERP waveforms for the second region in the VSO (S, red lines) and VOS (O, black lines) word order conditions for all 17 electrode sites. Negativity is plotted upward. Panel B: The voltage maps of the difference ERPs (VSO-VOS) for every 100 ms from 300 to 800 ms. Figure 5. The grand averaged ERP waveforms for the second region in the SVO (blue lines) and OVS (green lines) word order conditions for all 17 electrode sites. Negativity is plotted upward.

12 Table 4. Statistical results for the second region of VSO versus VOS. The second region of VSO versus VOS ms ms ms ms midline parasagittal temporal midline parasagittal temporal midline parasagittal temporal midline Parasagittal temporal WordOrder F-value Significance ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. W*Anteriority F-value Significance ns. + + ns. ns. ns. ns. ns. + ns. ns. ns. W*Hemisphere F-value N/A N/A N/A N/A Significance N/A ns. ns. N/A + ns. N/A ns. ns. N/A ns. ns. W*A*H F-value N/A N/A N/A N/A Significance N/A ns. ns. N/A ns. ns. N/A + + N/A ns. ns ms ms ms ms midline parasagittal temporal midline parasagittal temporal midline parasagittal temporal midline Parasagittal temporal WordOrder F-value Significance ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. W*Anteriority F-value Significance ns. ns. ns. ns. ns. ns. ns. ns. *** + ns. * W*Hemisphere F-value N/A N/A N/A N/A Significance N/A ns. ns. N/A ns. ns. N/A * ns. N/A * + W*A*H F-value N/A N/A N/A N/A Significance N/A ns. ns. N/A ns. ns. N/A ns. ns. N/A ns. ns. Note: ns.: p >.10, +: p <.10, N/A: not applicable. *p <.05. **p <.01. ***p <.005. ****p <.001. Language, Cognition and Neuroscience 1219

13 Table 5. Statistical results for the second region of SVO versus OVS. The second region of SVO versus OVS ms ms ms ms midline parasagittal temporal midline parasagittal temporal midline parasagittal temporal midline Parasagittal temporal WordOrder F-value 0.01 > Significance ns. ns. ns. ns. ns. ns. ns. ns. * * ** ** W*Anteriority F-value Significance * ns. ns. ** ns. * *** ** *** *** + ** W*Hemisphere F-value N/A N/A N/A N/A Significance N/A + ns. N/A ns. ns. N/A ns. ns. N/A ns. ns. W*A*H F-value N/A N/A N/A N/A Significance N/A ns. ns. N/A ns. ns. N/A + ** N/A ns. ns ms ms ms ms midline parasagittal temporal midline parasagittal temporal midline parasagittal temporal midline Parasagittal temporal WordOrder F-value > Significance ns. ns. + ns. ns. ns. ns. ns. ns. * + ns. W*Anteriority F-value Significance * ns. * ns. ns. ns. ns. ns. ns. ns. ns. ns. W*Hemisphere F-value N/A N/A N/A N/A Significance N/A ns. ns. N/A ns. ns. N/A ns. ns. N/A ns. ns. W*A*H F-value N/A N/A N/A N/A Significance N/A * ns. N/A * * N/A * + N/A * + Note: ns.: p >.10, +: p <.10, N/A: not applicable. *p <.05. **p <.01. ***p <.005. ****p < D. Yasunaga et al.

14 Language, Cognition and Neuroscience 1221 signals were digitised at a sampling rate of 250 Hz. ERPs were quantified by averaging for each region in the time window of ms for each participant in all conditions. The baseline was set to 100 ms prior to stimulus onset. Trials with ERP artefacts exceeding ±80 μv in these time windows were automatically eliminated. The error trials were also excluded from the averaging Results Behavioural data The accuracy rate for the picture sentence matching task was calculated for each word order for each participant (Table 2). A one-way repeated measures ANOVA with task accuracy as the dependent variable was performed with WordOrder (four levels; VOS, VSO, SVO, OVS) as the within-participants factor. The ANOVA revealed a main effect for WordOrder [F(3, 45) = 6.24, p =.001]. The Bonferroni multiple comparisons revealed that the accuracy for VOS (95.4%, SD = 4.3) was significantly greater than the accuracy for OVS (88%, SD = 8.8) (p =.009). 6 The difference between the SVO (95.4%, SD = 3.8) and OVS word orders was marginally significant (p =.074). No other comparison between word orders was statistically significant Electrophysiological data A repeated measures ANOVA was used to statically analyse the mean voltage within time windows for every 100 ms immediately after the onset of the region for each word order. The statistical analyses were conducted separately at the midline (Fz, Cz, and Pz), parasagittal (F3, F4, C3, C4, P3, P4, O1, and O2), and temporal (F7, F8, T7, T8, P7, and P8) arrays. The midline analysis consisted of repeated measures ANOVAs with two within-group factors: WordOrder (two levels) Anteriority (three levels). The parasagittal and temporal analyses consisted of three within-group factors: WordOrder (two levels) Hemisphere (two levels) Anteriority (four levels in parasagittal arrays and three levels in temporal arrays). The Greenhouse Geisser correction for lack of sphericity was applied whenever applicable (Greenhouse & Geisser, 1959). The original degrees of freedom were reported with the corrected probability level SVO versus VOS. A visual inspection of the ERPs indicated that the O of SVO, as compared to the S of VOS, shifted positively in the time window of ms, being distributed mainly in the left parietal domain (Figure 3). This distribution was confirmed by the results of the ANOVAs (Table 3). In the time window of ms, the interactions between WordOrder and Hemisphere were significant at the parasagittal and temporal domains [parasagittal: F(1, 15) = 6.46, p <.05, temporal: F(1, 15) = 5.23, p <.05]. In both domains, the simple main effects of WordOrder were significant in the left hemisphere [parasagittal: p <.05, temporal: p <.05]. In the time window of ms, the interaction between WordOrder, Hemisphere, and Anteriority was significant at the temporal domain [F(2, 30) = 4.19, p <.05], and the effects of WordOrder were significant at the sites P7 (p <.05). The interactions between WordOrder and Hemisphere were significant at the parasagittal and temporal domains [parasagittal: F(1, 15) = 6.44, p <.05, Figure 6. The grand averaged ERP waveforms for the third region in the SVO (blue lines) and OVS (green lines) word order conditions for all 17 electrode sites. Negativity is plotted upward.

15 Table 6. Statistical results for the third region of SVO versus OVS. The third region of SVO versus OVS ms ms ms ms midline parasagittal temporal midline parasagittal temporal midline parasagittal temporal midline Parasagittal temporal WordOrder F-value Significance ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. W*Anteriority F-value Significance ns. ns. ns. ns. ns. ns. + ns. ns. + ns. ns. W*Hemisphere F-value N/A N/A N/A N/A > Significance N/A ns. ns. N/A ns. ns. N/A ns. ns. N/A ns. ns. W*A*H F-value N/A 0.75 > 0.00 N/A N/A N/A Significance N/A ns. ns. N/A ns. ns. N/A ns. *** N/A ns. ** ms ms ms ms midline parasagittal temporal midline parasagittal temporal midline parasagittal temporal midline Parasagittal temporal WordOrder F-value > > Significance ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. W*Anteriority F-value Significance ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. ns. W*Hemisphere F-value N/A N/A N/A N/A Significance N/A ns. ns. N/A ns. ns. N/A ns. ns. N/A ns. ns. W*A*H F-value N/A N/A N/A N/A Significance N/A ns. *** N/A ns. ns. N/A ns. *** N/A ns. *** Note: ns.: p >.10, +: p <.10, N/A: not applicable. *p <.05. **p <.01. ***p <.005. ****p < D. Yasunaga et al.

16 Language, Cognition and Neuroscience 1223 temporal: F(1, 15) = 6.31, p <.05]. In both domains, the simple main effects of WordOrder were significant in the left hemisphere [parasagittal: p <.001, temporal: p <.001]. The interaction between WordOrder and Hemisphere was significant in the time window of ms in the parasagittal and temporal arrays [parasagittal: F (1, 15) = 7.96, p <.05, temporal: F(1,15) = 10.49, p <.01]. These interactions showed a simple main effect of WordOrder in the left hemisphere [parasagittal: p <.05, temporal: p <.01]. The main effect of WordOrder was observed in the parasagittal array in the time windows of ms and ms [ ms: F(1, 15) = 4.72, p <.05, ms: F(1, 15) = 7.93, p <.05]. Finally, the interactions of WordOrder, Hemisphere, and Anteriority in the time windows of ms and ms were significant in the temporal array [ ms: F(2, 30) = 5.98, p <.01, ms: F(2, 30) = 4.78, p <.05]. These interactions indicate the simple-simple main effect at P7 [ ms: p <.001, ms: p <.001]. These results suggest that the O of SVO elicited a long-lasting positivity relative to the S of VOS VSO versus VOS. Figure 4 and Table 4 show the results of the comparison of the S of VSO and the O of VOS. In the time windows of ms and ms, the interactions between WordOrder and Hemisphere were significant at the parasagittal array [ ms: F (1, 15) = 6.62, p <.05, ms: F(1, 15) = 6.22, p <.05]. The results of the multiple comparisons, however, did not show statistical significance in the time windows of ms and ms. In the time window of ms, the interaction between WordOrder and Anteriority was significant at the temporal array [F(1, 15) = 8.74, p <.05]. The Bonferroni post hoc comparison revealed significance at the parietal domain (p <.005). Overall, we can conclude that the S of VSO elicited positivity at the parietal domain relative to the O of VOS. ANOVAs were also conducted for the third region of the VSO condition (i.e. O) and VOS condition (i.e. S). The effect of WordOrder was not significant for any of the time windows SVO vs. OVS. Other than the comparison reported above, we compared SVO and OVS. In the first region of the SVO condition (i.e. S) and the OVS condition (i.e. O), no effects of WordOrder reached significance for any time windows. That is, effects of relative production frequency were not observed here. This may be because, at the first noun phrase (NP), it is not yet clear whether it is S or O, which will be made explicitly clear at the second region by agreement morphemes on the V. As for the second region, the V of OVS elicited a late positivity relative to the V of SVO (Figure 5 and Table 5), consistent with our expectation that the O would be associated with a gap upon encountering the V in the processing of OVS sentences. Finally, a visual inspection indicated that the O of SVO, as compared to the S of OVS, shifted positively in the time window of ms (Figure 6), as expected. Figure 7. The grand averaged ERP waveforms for the first region in VOS + VSO (black dashed lines) and SVO + OVS (blue dotted lines) for all 17 electrode sites. Negativity is plotted upward.