Parallelism Between Sentence Structure and Nominal Phrases in Japanese: Evidence from Scrambled Instrumental and Locative Adverbial Phrases.

The present study investigated the canonical position of instrumental and locative adverbial phrases in both Japanese sentences and noun phrases to determine whether the canonical positions are parallel. A series of sentence/phrase decision tasks were used to compare sentences with different word-orders, including sentences with SAdvOV (S is subject phrase, Adv adverb, O object phrase and V verb), AdvSOV, SAdvOV and SOAdvV word orders. SAdvOV word order was found to be the most quickly processed, for both instrumental adverbial (Experiment 1) and locative adverbial phrases (Experiment 2). Thus, the canonical position for these adverbial phrases is identified as the position immediately preceding the object (Theme argument). This finding was replicated when the same experimental methods were applied to event-denoting noun phrases. Adverbial adjuncts in the initial position (AdvON, N is noun phrase) were processed more quickly and accurately than noun phrases with adverbial phrases in the second position (OAdvN), for both instrumental adverbial (Experiment 3) and locative adverbial phrases (Experiment 4). Therefore, the position immediately preceding the object is the canonical position for both instrumental and locative adverbial phrases in sentences and in noun phrases. In conclusion, this indicates that the base structure of a sentence is shared by its related noun phrase.

Introduction

Some sentences have corresponding noun phrases with no change of fundamental meaning. Taking an example from Chomsky (1970), the active sentence the enemy destroyed the city corresponds to the noun phrase the enemy’s destruction of the city. Here, the subject of the noun phrase is marked by the possessive –s and the object of the noun phrase is introduced with the preposition of. This is referred to as ‘nominalization’. Although the past tense is lost in the noun phrase, both the sentence and the noun phrase carry essentially the same meaning and have similar elements. In this example, the noun destruction is a nominal derived from the verb destroy and acts as the head of the whole noun phrase. Using the genitive case marker –no, nominalization in Japanese is easily accomplished with relatively free word order. Thus, the word order in Japanese, where scrambling is allowed in both sentences and noun phrases, is much more flexible than in English. A study using eye-tracking (Tamaoka & Mansbridge, 2019; Tamaoka et al., 2014) showed longer eye-fixation times for the nominative-marked (-ga) noun phrase in scrambled sentences of OSgapV (gap refers to the original object position) than in its counterpart of the accusative-marked (-o) noun phrase in canonical sentences of SOV. Assuming faster processing times for canonically ordered sentences and possibly noun phrases compared to scrambled orders, the present study investigated whether the derived noun phrase (i.e., nominalization) has fundamentally the same syntactic structure as its corresponding sentence. More specifically, the present study addressed this question from two perspectives; (1) by attempting to determine the canonical position of instrumental or locative adverbial phrases in both sentences and noun phrases, and when this was clarified, (2) by determining whether the canonical order of a sentence with an instrumental or a locative adverbial phrase is the same as the canonical order of its corresponding noun phrase.

A Japanese Sentence and Its Corresponding Noun Phrase and Scrambling

Japanese nominals that can have sentential correlates are formed by adding the genitive case-marker –no to the postpositional phrase (PP) and noun phrase (NP) within the whole NP. For instance, a sentence Gakusei-ga yuubin-de zyuken-o moosikon-da, N(student)-NOM(-ga) N(post)-P(-de) N(exam)-ACC(-o) V(apply.for)-PAST, ‘The student applied for the exam by post’ has a corresponding noun phrase Gakusei-no yuubin-de-no zyuken-no moosikomi, N(student)-GEN(-no) N(post)-P(-de)-GEN(-no) N(exam)-GEN(-no) N(application), ‘the student’s application for the exam by post’. In this noun phrase, the nouns gakusei ‘the student’ and zyuken ‘exam’, and the PP consisting of yuubin ‘post’ and the instrumental postposition –de are marked with the genitive case marker –no. The nominative marker –ga and accusative case marker –o are obligatorily dropped when the genitive case marker –no is added.

Furthermore, Japanese allows flexible word orders in both sentences and noun phrases. In other words, the aforementioned example sentence can be scrambled to yuubin-de gakusei-ga zyuken-o moosikon-da by moving [PP N(post)-P(-de)] to a higher position than [NP N(student)-NOM(-ga)]. This sentence has the corresponding noun phrase yuubin-de-no gakusei-no zyuken-no moosikomi. Like the sentence, this noun phrase allows the order of gakusei-no [NP N(student)-GEN(-no)] and yuubin-de-no [PP N(post)-INS(-de)-GEN(-no)] to be shifted, but both those constituents must be marked with the genitive case marker -no. The meaning of the scrambled sentence and its corresponding noun phrase remains the same.

In psycholinguistic experimental studies on scrambling (e.g., Imamura et al., 2016; Koizumi & Tamaoka, 2004, 2006, 2010; Mazuka et al., 2002; Miyamoto & Takahashi, 2004; Tamaoka et al., 2005; Tamaoka et al., 2014; Tamaoka & Mansbridge, 2019; Ueno & Kluender, 2003; Witzel & Witzel, 2016), the canonical order of SOV (S is subject phrase, O object phrase and V verb) was found to be faster to process than the scrambled (i.e., different) order of OSV. The delay in processing time between scrambled OSV-ordered sentences and their SOV canonical counterparts is known as the scrambling effect. The sentence processing model of gap-filling parsing (Frazier & Clifton, 1989; Frazier & Flores D’Arcais, 1989; Frazier, 1987; Stowe, 1986) provides one possible explanation for the delay with the scrambled OSV order. This scrambling can be explained as a syntactic operation of phrasal movement from the original locus of the object (NP-o) in the canonical position to the sentence initial position as in [CP NP-o1 [IP NP-ga [I’ [VP gap1 V] I]]] where IP refers to inflectional phrase and CP to complement phrase. The gap1 indicates the original position in the canonical order from which the NP-o1 was moved to the sentence initial position. To process the scrambled sentence, native Japanese speakers must recognize the initial NP-o as the filler, and then find its original position in the specifier of VP (gap1) to establish the filler-gap dependency. Here, due to the degree of syntactic complexity, a canonical SOV-ordered sentence is expected to be processed more quickly than its OSV-ordered scrambled counterpart.

Some of the difficulties thought to be associated with scrambling can be explained in part by limitations in working memory or computational resources (Gibson, 1998), the increased syntactic complexity of the scrambling operation (Hawkins, 2004), expectation of constituents (Levy, 2008) and discourse contexts (Kaiser & Trueswell, 2004; Yano & Koizumi, 2018). Discourse and frequency effects have been found to modulate the overall scrambling effect such that they can attenuate the processing difficulty (Imamura et al., 2016). The eye-tracking studies (Tamaoka & Mansbridge, 2019; Tamaoka et al., 2014) clearly indicated longer eye fixation times in the area (noun phrase) assumed to have a gap and in the area of a head verb under the scrambled condition. In these eye-tacking studies, there were no discourse contexts presented prior to the scrambled sentences. In such conditions, scrambled sentences seem to result in a heavier cognitive load for processing; scrambling should be the main source of difficulty in processing a single sentence. Using this nature of the scrambling effect, the present study examines the canonical position of instrumental and locative adverbial phrases.

Possible Canonical Positions of Multiple Adverbs

Adverbs are optional elements of a sentence, categorized as adjuncts that can be placed in various positions. Koizumi (1993) classified Japanese adverbs into three types based on their canonical positions, (1) adverbs in a modal phrase (MP adverbs) such as saiwainimo ‘fortunately’, tabun ‘perhaps’, (2) adverbs in an inflection phrase (IP adverbs) such as kinoo ‘yesterday’, mukasi ‘once upon a time’, and (3) adverbs in a verb phrase (VP adverbs) such as konagonani ‘into pieces’, kossori ‘secretly’. Each type of adverb is argued to have a canonical position—i.e., a position it must occupy before scrambling.

Canonical positions of various types of adverbs are identified as follows (Adv refers to adverbs):

[MP (MP ) [IP (IP ) Subject (IP ) [VP (VP ) Object (VP ) Verb] I] M].

The canonical position of the modal phrase adverb (MP adv) saiwainimo ‘fortunately’ is located within the modal phrase as in saiwainimo Kenzi-ga han’nin-o tukamae-ta ‘Fortunately Kenzi caught a criminal’. Temporal adverbs like kinoo ‘yesterday’, kyoo ‘today’, or asita ‘tomorrow’ are classified as the inflectional phrase adverb (IP adv), taking their canonical positions before or after the subject. For example, the canonical positions of the time adverb kinoo can be placed either before the subject as in Kinoo Tomoko-ga kami-o kit-ta ‘Yesterday Tomoko cut her hair’ or after the subject as in Tomoko-ga kinoo kami-o kit-ta.

According to Koizumi (1993), both manner and resultative adverbs are classified as verb phrase adverbs (VP adv). There are two possible canonical positions within VP adverbs (i.e., canonically neutral), either before the object or after the object (before the verb), or SAdvOV and SOAdvV. For example, the canonical position of the resultative adverb konagonani ‘into pieces’ can be either before the object as in Kenzi-ga konagonani kabin-o kowasi-ta ‘Kenzi broke the vase into pieces’ or after the object (before the verb) as in Kenzi-ga kabin-o konagonani kowasi-ta. This claim is supported by an experimental sentence-processing study by Koizumi and Tamaoka (2006), whose reading-time measurements revealed that the canonical position of manner and resultative adverbs is either SAdvOV or SOAdvV. Furthermore, a corpus study by Namba and Tamaoka (2016) counted frequencies of adverb occurrences, indicating that manner adverbs showed near equal distribution for 50.0% of SAdvOV and 48.4% of SOAdvV positions. There were only 1.7% in AdvSOV. As with the psycholinguistic study by Koizumi and Tamaoka (2006), manner adverbs appear to be canonically neutral between those two positions. In sentences with a resultative adverb, the corpus frequency of word order is 80.7% for SOAdvV, 18.0% for SAdvOV, and 1.3% for AdvSOV. Although manner adverbs appear randomly in either the SAdvOV position or the SOAdvV position, resultative adverbs appear most frequently in the SOAdvV position. Thus, manner and resultative adverbs may have different canonical positions within the VP adverbial position.

Possible Canonical Positions of Instrumental and Locative Adverbial Phrases

The canonical positions of instrumental and locative adverbial phrases have yet to be clearly defined. In fact, Koizumi and Tamaoka (2006) did not investigate instrumental and locative adverbial phrases ending with -de (i.e., [PP NP + Postposition(-de)]) in their investigation. These can be placed in three different positions in Japanese sentences: in the sentence-initial position as in Tomoko-ga terebi-o kesi-ta ‘Tomoko turned off the television with remote control’, with the order AdvSOV; in the post-subject position (SAdvOV) as in Tomoko-ga terebi-o kesi-ta; or in the post-object position (SOAdvV) as in Tomoko-ga terebi-o kesi-ta. It is possible, however, to consider the instrumental and locative adverbial PP phrase with –de to have one or two basic positions. As the first step of such an investigation, the present study examines where the canonical position of these might be located.

Because these two types of adverbial phrases, instrumental and locative, are generally taken to modify the whole VP (i.e., the verb and its object without tense) as opposed to IP including tense, it is assumed that they are at the VP level. The potential canonical position (SAdvOV) is depicted in Fig. 1. There are two examples in Fig. 1. An example of the instrumental adverbial phrase remokon-de ‘by remote control’ is Tomoko-ga terebi-o kesi-ta ‘Tomoko turned off the television with remote control’. In this sentence, the adverbial phrase is positioned in the specifier of V’ as the postposition (PP) adverbial phrase. Likewise, the locative adverb heya-de ‘in the room’ is also found in the same position as Tomoko-ga mimotu-o ake-ta ‘Tomoko opened the package in the room’.

Fig. 1

The canonical position of instrumental and locative adverbial phrases at the sentence level Note: The PP refers to the assumed canonical position of the adverb. INST refers to instrumental adverbs and LOC to locative adverbs

This SAdvOV canonical order is proposed by Vinka (2009) who claims locatives and instrumentals in Japanese are similar to the extent that they both take VPs as arguments. Takezawa (2000; as cited by Ogawa & Niinuma, 2012) claims that locatives and resultatives in Japanese are likely to differ in their ordering in relation to a theme determiner phrase (Theme DP) or the object phrase (O), where the former is likely to precede the Theme (SAdvOV), and the latter to follow it (SOAdvV) at the VP level (VP adverbs). Prior to the present experiments, it was tentatively assumed that SAdvOV is the canonical order for both instrumental and locative adverbial phrases. Alternatively, the order SOAdvV is another possibility for the canonical order. However, we hypothesize that SAdvOV is canonical, and that the order SOAdvV is derived by scrambling.

In both examples in Fig. 2, the final nouns of moosikomi ‘application’ and hanbai ‘sale’ can be considered to have a verb-like argument structure (i.e., have a subject and object), as in their verb forms of moosikomu ‘to apply’ and hanbaisuru ‘to sell’. Although these noun phrases including instrumental and locative adverbial phrases do not have a subject, they can be considered equivalent to the (S)AdvOV-ordered sentence shown in Fig. 1. Thus, it is assumed that the noun phrase with adverb-initial position AdvON (N is noun phrase) would be the canonical order. The same holds for the attested word-order within noun phrases whose heads have an argument structure. Within the noun phrase in Fig. 2, the instrumental and locative adverbial phrases can be re-ordered as OAdvV as zyuken-no moosikomi and yasai-no hanbai. In this study, we hypothesize that AdvON is canonical, and that the order OAdvV is derived by scrambling.

Fig. 2

The canonical position of instrumental and locative adverbial phrases at the noun phrase level Note: The PP refers to the assumed canonical position of the adverb. INST refers to instrumental adverbs and LOC to locative adverbs

It is also assumed that both instrumental and locative adverbial phrases in noun phrases (PP + the genitive maker -no) are placed in the initial position AdvON as depicted in Fig. 2. The instrumental adverbial phrase yuusoo-de-no ‘by postal mail’ is located at the specifier of NP or in the beginning of the noun phrase in the AdvON order as zyuken-no moosikomi ‘applying for examination by postal mail’. Likewise, the locative adverbial phrase rozyoo-de-no ‘on the street’ is also located at the same position of NP specifier as in yasai-no hanbai ‘the sale of vegetables on the street’.

Nominalization is a prevalent cross-linguistic feature, and it is heavily featured in Japanese as well (see Kishimoto, 2006; Miyagawa, 2012; Sugioka, 1992). The nominalized structure may impose the same restrictions on the ordering of elements as a sentential vP structure. However, it is not yet known whether order (i.e., syntactic position) restrictions will also be prevalent within the hierarchal structure of an NP or DP in Japanese.

If native Japanese speakers process the noun phrase in the same manner as they process its corresponding sentence, the scrambling effect should be observed in the processing of an OSV-type of noun phrase as well. This would suggest that the psycholinguistic nature of sentences is shared by their corresponding noun phrases. Instrumental and locative adverbial phrases are excellent candidates for investigating the parallelism of sentences and their noun phrase counterparts because these adverbial phrases allow us to directly compare the syntactic structures of sentences and noun phrases. Subject (S) and object (O) cannot be scrambled in a noun phrase and replaced by the genitive. For instance, the noun phrase, Nihonzin-no siboo-no torisugi, N(Japanese)-GEN N(fat)-GEN N(taking-too-much), ‘Too-much eating of fat by Japanese’ does not make sense once it is scrambled as in *Siboo-no nihonzin-no torisugi, N(fat)-GEN N(Japanese)-GEN N(taking-too-much). Hence, the subject phrases must be omitted in investigating scrambling in NPs. In addition, a sentence with a resultative adverb cannot be nominalized: the sentence Konagona-ni hakai-si-ta, Adv(into-pieces) V(destroy)-PAST, ‘(I) destroyed (it) into pieces’ cannot be expressed as an NP *Konagona-e-no hakai, ‘destruction into-pieces’. Likewise, a manner adverb used in a sentence has to change its lexical category, from adverb to adjective, in order to be nominalized: the sentence Subayaku ootoo-si-ta, Adv(quickly) V(respond)-PAST, ‘I quickly responded (to it)’ becomes a noun phrase where ‘response’ is modified by the adjective ‘quick’, as subayai ootoo, Adj(quick) N(response), ‘a quick response’.

To verify the hypothesis that noun phrases are processed in the same way as sentences, this study investigated the existence of the scrambling effect in both sentences and noun phrases with different instrumental and locative adverb orders. In the present study, we conducted four experiments, each relying on the scrambling effect to identify the canonical word order of the structure containing an adverb. These experiments provide empirical evidence supporting (1) the canonical positions for the instrumental and locative adverbial phrases as SAdvOV depicted in Fig. 1 at the sentence level and as AdvON in Fig. 2 in the noun phrase and (2) sentences and noun phrases sharing the parallel basic structure (i.e., canonical order).

Experiment 1: Sentences with Instrumental Adverbial Phrases

Experiment 1 tested whether instrumental adverbial phrases have a canonical position of SAdvOV in sentences with transitive verbs.

Method

Participants

Twenty-four graduate and undergraduate students (19 females and 5 males) at a university in Japan, all native speakers of Japanese, participated in Experiment 1. The participants were from various academic backgrounds, and their ages ranged from 18 years and 2 months to 27 years and 8 months. The average age was 21 years and 4 months with a standard deviation of 2 years and 8 months on the day of testing.

Materials

Stimuli consisted of sentences that included instrumental adverbial phrases in one of three possible word orders: AdvSOV, SAdvOV and SOAdvV. These stimuli were divided into two classes – semantically coherent “Yes” stimuli, and semantically anomalous “No” stimuli. To create the “Yes” stimulus sentences, twenty-four sentences with an AdvSOV structure, e.g., Ziro-ga repooto-o kai-ta ‘Ziro wrote a report with a personal computer’, had their adverbial phrase positions altered into SAdvOV, e.g., Ziro-ga repooto-o kai-ta, and SOAdvV, e.g., Ziro-ga repooto-o kai-ta. The only difference among the three types of sentences was the position of adverbial phrase with –de. In the same manner, in the case of the “No” stimuli, 24 sets of semantically infelicitous sentences with AdvSOV order, e.g., * Ziro-ga Kazuko-ni soodansi-ta ‘Ziro consulted Kazuko with the vegetables’, had the position of their instrumental adverbial PP phrase changed into SAdvOV, e.g., Ziro-ga Kazuko-ni soodansi-ta, and SOAdvV, e.g., Ziro-ga Kazuko-ni soodansi-ta. Again, the only differences among the three were the positions of the adverbial phrases.

Reading times are likely to become shorter when participants read sentences in sequence containing the same words. In order to prevent this effect of repeated encountering, we counterbalanced the list of sentences with a Latin square design. Three lists of sentences were given to three groups of participants (8 each). Each list consisted of 24 sentences (8 each for AdvSOV, SAdvOV and SOAdvV) for “Yes” responses and 24 sentences (8 each of AdvSOV, SAdvOV and SOAdvV) for “No” responses. In addition, 20 dummy sentences (10 semantically coherent and 10 semantically anomalous) were put in each list as filler sentences, such as Watasi-no tokuina ryoori-wa kareeraisu da ‘My favorite thing to cook is curry and rice’. A total of 68 stimuli were used, consisting of 24 semantically coherent, 24 semantically anomalous, and 20 dummy sentences. Experiments 1 and 2 were conducted together in a single experiment.

Procedure

The present study utilized a reaction time paradigm that measures the elapsed time between the presentation of a sensory stimulus and the participant’s subsequent behavior. This interval is called reaction (or processing) time. The experiment employed a whole sentence correctness decision task, by showing one sentence at a time on a computer screen. The presentation was controlled by Microsoft’s Visual Basic 6.0 + Microsoft DirectX8 computer program. Stimuli with both semantically coherent and semantically anomalous responses were presented to participants in a random order, in the center of a computer screen 600 ms after the appearance of asterisks ‘********’ indicating an eye-fixation point. Participants were asked to decide whether the sentences were semantically acceptable, by pressing a “Yes” or “No” button. They were also asked to answer as quickly as possible, while maintaining accuracy. Twenty-four practice trials were given to the participants prior to actual testing. Participants were expected to process simple sentences with a high accuracy. Sentences with a scrambled order were expected to require longer processing times than the corresponding ones with canonical order, due to syntactic manipulations.

Results

There was one extreme “Yes” response and one extreme “No” response in reaction times among sentence correctness decision times (i.e., less than 500 ms or longer than 5000 ms). These two extreme responses were excluded from the reaction time data. Furthermore, only stimulus items with correct responses (i.e., the participants answered “Yes” when the sentence made sense and “No” when it did not) were used in the analyses of reaction times. Before performing the analysis, reaction times outside of 2.5 standard deviations plus and minus the mean reaction time at both the high and low ranges were replaced by boundaries indicated by 2.5 standard deviations from the individual means of participants in each condition. Only one reaction time among correctly-responded “Yes” items fell into this category. The means of “Yes” and “No” reaction times and error rates for sentence correctness decisions are presented in Table 1.

Table 1

Reaction times and error rates for sentences with instrumental adverbs in experiment 1

Response type	Position of adverb (Adv)	Reaction time (ms)		Error rates (%)
		M	SD	M (%)	SD (%)
"Yes" responses	Adv SOV	1807	456	6.25	11.66
	SAdv OV	1527	361	7.81	12.12
	SOAdvV	1663	479	2.60	6.36
"No" responses	AdvSOV	1667	409	16.17	20.21
	SAdvOV	1674	436	14.33	17.96
	SOAdvV	1671	466	15.17	17.73

A series of one-way analyses of variance (ANOVAs) with repeated measures for three sentence types (AdvSOV, SAdvOV and SOAdvV) of “Yes” responses was conducted on reaction times (milliseconds) and error rates (percent), using participant (F1) and item (F2) variabilities. There were significant main effects in both participant analysis [F1(2, 46) = 8.262, p < 0.001] and item analysis [F2(2, 46) = 6.008, p < 0.01] for reaction times, but no main effect for error rates [F1(2, 46) = 2.120, p = 0.132, ns; F2(2, 46) = 1.523, p = 0.229, ns]. Simple contrast comparisons were conducted among the three AdvSOV, SAdvOV and SOAdvV conditions, indicating that SAdvOV was the most quickly recognized word order. As for “No” responses (incorrect sentences), neither reaction times [F1(2, 46) = 0.007, p = 0.993, ns; F2(2, 46) = 0.072, p = 0.930, n.] nor error rates [F1(2, 46) = 0.365, p = 0.696, ns; F2(2, 46) = 1.181, p = 0.322, ns] showed significant main effects.

Discussion

Sentences with canonical word order are predicted to be processed more quickly than sentences with scrambled word orders. Experiment 1 clearly indicated that the sentences in which the instrumental adverbial phrases were placed between the subject phrase and the object phrase were processed the most quickly among the three types of word order. Unlike the canonical position of manner and resultative adverbs being either SAdvOV or SOAdvV (Koizumi & Tamaoka, 2006), the canonical position of instrumental adverbial phrases was only SAdvOV. On the other hand, reaction times and error rates for correctly perceived “No” responses are observed to be null main effects, showing a minute difference in reaction times and a similar trend in error rates. This null effect may be a result of participants having multiple ways to reject these stimuli for “No” responses. For example, a participant may have been able to correctly reject the target sentence after comprehending only the semantically implausible part of the sentence (e.g., an adverbial phrase of ‘consult with vegetable’). Importantly, the “No” responses serve as a reference, and are not a major concern to the main argument for the canonical position of adverbial phrases (hereafter the same holds true for the other three experiments).

Experiment 2: Sentences with Locative Adverbial Phrases

Experiment 2 tested whether locative adverbial phrases have a canonical position (i.e., SAdvOV) in sentences with transitive verbs.

Experiment 2 was carried out at the same time as Experiment 1, so participants were the same in both experiments.

As in Experiment 1, “Yes” responses consisted of sentences with adverbial phrases in the phrase orders of AdvSOV, SAdvOV and SOAdvV. Twenty-four sentences with an AdvSOV structure had their adverbial phrase positions altered into SAdvOV and SOAdvV. In the same way, for the case of “No” responses, 24 sets of semantically anomalous sentences with AdvSOV had the position of their locative adverbial phrase changed into SAdvOV and SOAdvV. A counterbalance with a Latin square design was used to assign participants to different sentences. Since Experiment 2 was conducted with Experiment 1, the rest of the stimulus conditions were the same as Experiment 1.

The procedure for Experiment 2 was the same as the procedure for Experiment 1.

There was only one extreme “Yes” response and no extreme “No” responses in reaction times among sentence correctness decision times (i.e., less than 500 ms or longer than 5000 ms). This item was removed from the reaction time data. In addition, only stimulus items of correct responses were used in the analyses of reaction times. The data trimming process was the same as Experiment 1. No reaction times among correctly-responded “Yes” and “No” items fell outside the range of the mean plus or minus 2.5 standard deviations. The means of “Yes” and “No” reaction times and error rates for sentence correctness decisions are presented in Table 2.

Table 2

Reaction times and error rates for sentences with locative adverbs in experiment 2

Response type	Position of adverb (Adv)	Reaction time (ms)		Error rates (%)
		M	SD	M (%)	SD (%)
"Yes" responses	Adv SOV	1740	440	5.88	8.35
	SAdv OV	1626	365	4.83	9.08
	SOAdvV	1866	530	7.46	10.51
"No" responses	AdvSOV	1719	410	15.17	17.73
	SAdvOV	1635	423	15.25	22.91
	SOAdvV	1667	434	16.67	20.06

A series of ANOVAs with repeated measures for three sentence types (AdvSOV, SAdvOV and SOAdvV) of “Yes” responses was conducted on reaction times and error rates, using participant (F1) and item (F2) variabilities. Significant main effects were found in both participant analysis [F1(2, 46) = 9.255, p < 0.001] and item analysis [F2(2, 46) = 6.337, p < 0.01] for reaction times, but no main effect for error rates [F(2, 46) = 1.028, p = 0.366, ns; F(2, 46) = 0.603, p = 0.552, ns]. Simple contrast comparisons were conducted among the three conditions, AdvSOV, SAdvOV and SOAdvV, again indicating that SAdvOV was the most quickly processed phrase order. As for “No” responses (incorrect sentences), neither reaction times [F1(2, 46) = 1.007, p = 0.373, ns; F2(2, 46) = 0.905, p = 0.416, ns] nor error rates [F1(2, 46) = 0.315, p = 0.731, ns; F2(2, 46) = 0.063, p = 0.940, ns] showed significant main effects.

As in Experiment 1, Experiment 2 indicated that sentences with SAdvOV phrase order were the most quickly processed among sentences of three different phrase orders. Thus, as with instrumental adverbial phrases, SAdvOV must be the canonical position of locative adverbial phrases. Once again, this result differed from the canonical position of manner and resultative adverbial phrases identified as either SAdvOV or SOAdvV (Koizumi & Tamaoka, 2006).

Experiment 3: Noun Phrases with Instrumental Adverbial Phrases

Experiment 3 examined whether instrumental adverbial phrases have the canonical position of AdvON in noun phrases.

Twenty-four graduate and undergraduate students (8 females and 16 males) at a university in Japan, all native speakers of Japanese, participated in Experiment 3. These participants did not participate in Experiments 1 and 2. Ages ranged from 19 years and 2 months to 26 years and 9 months. The average age was 21 years and 3 months with a standard deviation of 1 years and 7 months on the day of testing.

“Yes” responses consisted of noun phrases with instrumental adverbial phrases with the orders AdvON and OAdvN. Twenty-four AdvON noun phrases, e.g., zyuken-no moosikomi ‘application for entrance examination by mail’, were prepared and had their adverbial phrase positions altered into OAdvN, e.g., zyuken-no moosikomi. The only difference between the two types of noun phrases was the position of the instrumental adverbial phrase with -de. For the case of “No” responses, a set of the same number of semantically anomalous noun phrases with AdvON, e.g., densirenzi-de-no obentoo-no kiritori ‘cutting a lunch box by microwave’, were prepared and the adverbial phrase position shifted into OAdvN as in obentoo-no densirenzi-de-no kiritori. Again, the only difference between noun phrases was the adverbial phrase position. These noun phrases are semantically unacceptable, so the participants were expected to reject them as incorrect phrases by pressing the "No" key.

In order to prevent the problem of repeatedly encountering noun phrases with the same words, a counterbalanced (or Latin square) design was used to assign participants to different noun phrases. Two lists of noun phrases were given to two groups of participants (12 each). Each list consisted of 24 coherent noun phrases (12 each for AdvON and OAdvN) for the “Yes” responses and 24 anomalous noun phrases (12 each for AdvON and OAdvN) for the “No” responses. In addition, 20 dummy noun phrases (10 correct and 10 incorrect) were added in each list as filler noun phrases, e.g., hanagara-no kawaii sukaato ‘a cute skirt with pictures of flowers’. The total of 68 noun phrases were used for Experiment 3, consisting of 24 semantically coherent, 24 semantically anomalous, and 20 dummy noun phrases. Experiments 3 and 4 were conducted together in a single experimental trial.

The experiment employed a whole-phrase correctness decision task, by showing one noun phrase at a time on a computer screen. The presentation was controlled by Microsoft’s Visual Basic 6.0 + Microsoft DirectX8 computer program. Stimuli with both semantically coherent and semantically anomalous responses were presented to participants in a random order, in the center of a computer screen 600 ms after the appearance of asterisks ‘********’ indicating an eye-fixation point. Participants were asked to decide whether the noun phrases made sense, by pressing a “Yes” or “No” button. They were additionally asked to answer as quickly as possible, while maintaining accuracy. Twenty-four practice trials were given to the participants prior to actual testing.

There were two extreme “Yes” responses but no extreme “No” responses in reaction times among noun phrase correctness decision times (i.e., less than 500 ms or longer than 5000 ms). These extreme responses were removed from the reaction time data. In addition, only stimulus items with correct responses were used in the analyses of reaction times. The data trimming process was the same as Experiments 1 and 2. Four reaction times for “Yes” items and 11 reaction times for “No” items fell outside the range of the mean plus or minus 2.5 standard deviations. The means of “Yes” and “No” reaction times and error rates for noun phrase correctness decisions are presented in Table 3.

Table 3

Reaction times and error rates for noun phrases with instrumental adverbs in experiment 3

Response type	Position of adverb (Adv)	Reaction time (ms)		Error rates (%)
		M	SD	M (%)	SD (%)
"Yes" responses	Adv ON	1909	572	8.33	9.83
	OAdv N	2263	720	16.67	13.90
"No" responses	AdvON	1937	490	2.08	5.63
	OAdvN	2000	632	2.08	5.63

A series of ANOVAs with repeated measures for two noun phrase types (AdvON and OAdvN) of “Yes” responses was conducted on reaction times and error rates, using participant (F1) and item (F2) variabilities: Two adverbial phrase positions, the adverbial phrase initial position and the adverbial phrase second position, were examined in the processing of noun phrases with instrumental adverbial phrases. The analyses indicated that noun phrases of the adverbial phrase initial position (i.e., AdvON) were processed more quickly [F(1, 23) = 18.846, p < 0.001; F(1, 23) = 11.158, p < 0.01] and more accurately [F(1, 23) = 5.750, p < 0.05; F(1, 23) = 9.095, p < 0.01] than noun phrases with the adverbial phrase second position (i.e., OAdvN). For “No” responses (incorrect noun phrases), neither reaction times [F1(1, 23) = 0.354, p = 0.558, ns; F2(1, 23) = 0.186, p = 0.671, ns] nor error rates [F1(1, 23) = 0.000, p = 1.000, ns; F2(1, 23) = 0.000, p = 0.984, ns] showed significant main effects.

Experiment 3 indicated that the canonical position of instrumental adverbial phrases in noun phrases is AdvON (adverb initial position). This result replicates the general findings of Experiment 1 for the sentence structure. Thus, this provides evidence that nominal phrases are likely to adhere to similar restrictions of canonical order as the sentence structure.

Experiment 4: Noun Phrases with Locative Adverbial Phrases

Experiment 4 further investigated whether the canonical position of locative adverbial phrases is also AdvON in noun phrases.

Experiment 3 was carried out with Experiment 4, so participants were the same in both experiments.

As in Experiment 3, “Yes” responses were noun phrases with locative adverbial phrases in the phrase structure of AdvON and OAdvN. Twenty-four semantically coherent noun phrases with AdvON were altered so that their adverbial phrase position was OAdvN. The only difference between two types of noun phrases was the position of the locative adverbial phrase with –de. In the case of “No” responses, 24 sets of semantically anomalous noun phrases AdvON were prepared with the adverbial phrase position shifted to OAdvN. As in Experiment 3, a counterbalanced (or Latin square) design was used to assign participants to different noun phrases. Because Experiment 4 was conducted with Experiment 3, the rest of the stimulus conditions were the same.

The procedure for Experiment 4 was the same as the procedure for Experiments 1, 2 and 3.

There were two extreme “Yes” responses and no extreme “No” responses in reaction times among noun phrase correctness decision times (i.e., less than 500 ms or longer than 5000 ms). Once again, these extreme outliers were removed from the reaction time data. Additionally, only stimulus items of correct responses were used in the analyses of reaction times. The data trimming process was the same as Experiments 1, 2 and 3. Two reaction times for “Yes” items and six reaction times for “No” items fell outside the cutoff boundary of the mean plus or minus 2.5 standard deviations. The means of “Yes” and “No” reaction times and error rates for noun phrase correctness decisions are presented in Table 4.

Table 4

Reaction times and error rates for noun phrases with locative adverbs in experiment 4

Response type	Position of adverb (Adv)	Reaction time (ms)		Error rates (%)
		M	SD	M (%)	SD (%)
"Yes" responses	Adv ON	1728	436	6.94	10.90
	OAdvN	2031	633	15.28	16.24
"No" responses	AdvON	1764	460	4.17	8.86
	OAdvN	1906	470	3.47	6.91

A series of ANOVAs with repeated measures for the two noun phrase types (AdvON and OAdvN) of “Yes” responses was conducted on reaction times and error rates. Two adverbial phrase positions, the adverbial phrase initial position and the adverbial phrase second position, were examined in the processing of noun phrases with locative adverbial phrases. The analyses indicated that noun phrases with adverbial phrases in initial position (i.e., AdvON) were processed more quickly [F(1, 23) = 17.751, p < 0.001; F(1, 23) = 10.456, p < 0.01] and more accurately [F(1, 23) = 4.312, p < 0.05; F(1, 23) = 4.545, p < 0.05] than noun phrases with adverbial phrase in second position (i.e., OAdvN). As for “No” responses (incorrect noun phrases), reaction times for noun phrases with adverbial phrases in initial position (i.e., AdvON) were shorter than for noun phrases with adverbial phrases in second position (i.e., OAdvN) for the participant analysis [F1(1, 23) = 4.749, p < 0.05], but this result did not hold for the item analysis [F2(1, 23) = 0.977, p = 0.333, ns]. Thus, we interpret this main effect result as not significant. There was no significant main effect on error rates [F1(1, 23) = 0.074 p = 0.788, ns; F2(1, 23) = 0.097, p = 0.758, ns].

Experiment 4 indicated that, as with instrumental adverbial phrases in Experiment 3, the adverbial initial position of AdvON was the canonical position of locative adverbial phrases in noun phrases.

General Discussion

The present study conducted four experiments to investigate whether the base structure of a sentence is shared by its related noun phrase. As depicted in Figs. 1 and 2, the present study hypothesized that instrumental and locative adverbial phrases with –de have the same canonical position, SAdvOV for sentences, and AdvON for noun phrases. Extending on prior studies of the scrambling effect in Japanese (e.g., Koizumi & Tamaoka, 2004, 2006, 2010; Mazuka et al., 2002; Tamaoka et al., 2005; Tamaoka et al., 2014; Tamaoka & Mansbridge, 2019), which showed that canonical word orders are processed more quickly than scrambled orders, the present study investigated the canonical position of instrumental and locative adverbial phrases.

Experiments 1 and 2 were conducted to find the canonical position of instrumental and locative adverbial phrase marked by –de within sentences. Among the three possible phrasal orders of AdvSOV, SAdvOV and SOAdvV, the order of SAdvOV was found to be the most quickly processed, both in the case of instrumental in Experiment 1 and locative in Experiment 2. Thus, the canonical position for these adverbial phrases can be identified as the position between the subject and the object. Experiments 3 and 4 examined the canonical position of instrumental and locative adverbial phrases in nominal phrases. The result showed for both the instrumental adverbial phase in Experiment 3 and the locative adverbial phrase in Experiment 4, the adverbial phrase initial position (i.e., AdvON order) was processed more quickly and more accurately than the adverbial phrase second position (i.e., OAdvN order). These results suggest that the canonical adverbial phrase position for sentences, SAdvOV or PP position in Fig. 1, remains the same in the case of noun phrases, AdvON or PP position in Fig. 2.

Takezawa (2000; as cited by Ogawa & Niinuma, 2012) claims that locatives and resultatives in Japanese are likely to differ in their ordering in relation to a theme determiner phrase (Theme DP) or an object phrase (O), where the former is likely to precede the Theme (SAdvOV), and the latter to follow it (SOAdvV) at the VP level (VP adverbs). Furthermore, although adverbial phrases are adjuncts and thought to be optional pieces of a sentence, the present study showed that adverbial phrases have canonical base positions in both a sentence and a noun phrase.

According to Koizumi’s (1993; see also Kishimoto, 2006; Miyagawa, 2012; Sugioka, 1992) categorization, the instrumental and locative adverbial phrases tested in this study should be classified as VP adverbs. Under this assumption, their canonical position is defined either SAdvOV or SOAdvV. In fact, the psycholinguistic study by Koizumi and Tamaoka (2006) indicated that the canonical position of manner and resultative adverbs is as either SAdvOV or SOAdvV. However, instrumental and locative adverbial phrases with –de, those tested in Experiments 1 and 2 for sentences, and Experiments 3 and 4 for noun phrases, turned out to have only one canonical position: SAdvOV for a sentence (PP position in Fig. 1), and AdvON for noun phrases (PP position in Fig. 2). Additional support for this order is found in the self-paced reading study by Nambu and Nakatani (2014). Using nominative-genitive alternation, they also provide evidence that locative adverbial phrases (i.e., PPs) are likely to be located under the specifier of VP giving the order SAdvOV.

The ultimate purpose of the present experiments was to examine whether the basic word order of a sentence would be shared by its corresponding noun phrases. The four experiments in the present study found that the canonical position for instrumental and locative adverbial phrases remains parallel across both sentences and noun phrases. The adverbial phrase position of SAdvOV in sentence is considered equivalent to AdvON for a noun phrase. This study’s findings support the assumption that the base structure of a sentence is shared by its related noun phrase.