An ERP study of the processing of subject and object relative

ERPs to Japanese Relative Clauses 1
An ERP study of the processing of subject and object relative clauses in Japanese
Mieko Ueno
Department of Linguistics, University of California, San Diego
Susan M. Garnsey
Department of Psychology, University of Illinois, Urbana-Champaign
Page headline: ERPS to Japanese Relative Clauses
July 2006
Address correspondence to:
Mieko Ueno
Department of Linguistics
University of California, San Diego
9500 Gilman Drive
La Jolla, CA 92093-0108
phone: (217) 721-8227
fax: (858) 534-4789
email: ueno@ling.ucsd.edu
ERPs to Japanese Relative Clauses 2
Abstract
Using reading times and event-related brain potentials (ERPs), we investigated the processing of
Japanese subject/object relative clauses (SRs /ORs). English ORs take longer to read (King &
Just, 1991) and elicit left-lateralized/bilateral anterior negativity (LAN) between fillers and gaps
(King & Kutas, 1995), which is largely attributed to a longer filler-gap distance. In contrast to
English, gaps in Japanese relative clauses precede their fillers, and the linear gap-filler distance is
longer in SRs than in ORs. Nevertheless, Japanese ORs take longer to read (Ishizuka, Nakatani,
& Gibson, 2003; Miyamoto & Nakamura, 2003), perhaps because in both English and Japanese,
ORs involve a longer structural filler-gap/gap-filler distance in their syntactic representations
(O'Grady, 1997). We investigated how gap-filler association in Japanese would compare to
filler-gap association in English, and whether it is linear or structural distance that determines
comprehension difficulty. The results showed higher processing costs from the head noun to the
sentence-end for ORs than SRs in both reading times and ERPs. Our data also suggest that gapfiller association in Japanese relative clauses involve a long-lasting integration process after the
filler, as indexed by the continuous centro-posterior positivity to ORs (cf. Kaan, Harris, Gibson,
& Holcomb, 2000). Finally, both ERP and reading time data seem more consistent with a
structural distance account than a linear distance account, at least for Japanese relative clauses.
ERPs to Japanese Relative Clauses 3
Acknowledgements
This work was supported by the University of Illinois Research Board and by an NIMH
postdoctoral fellowship (T32 MH19554) to the first author. We thank Gary Dell, Kara
Federmeier, Cindy Fisher, Tomoko Ishizuka, Andy Kehler, Robert Kluender, Nayoung Kwon,
Brian Lin, Charles Lin, Edson Miyamoto, and Tamara Swaab for helpful discussion and
suggestions, Yoko Ieuji, Eri Imura, Jinya Kurita, Kyung Sil Lee, Tom Shoemaker, and Michelle
Wang for help with stimulus generation and data collection, and Laura Matzen and Katie McGee
for editorial support.
ERPs to Japanese Relative Clauses 4
An ERP study of the processing of subject and object relative clauses in Japanese
Introduction
The fundamental question addressed in this paper is how syntactically distinct languages
are processed in the brain. By investigating such a question, we hope to find both languageuniversal and language-specific aspects of sentence comprehension and thereby to narrow the
gap between linguistic and cognitive neuroscientific approaches to language. Event-related brain
potentials (ERPs) are useful in this endeavor, as they reveal millisecond-by-millisecond changes
in neural activity during language comprehension.
This study investigates the processing of Japanese subject and object relative clauses
using both self-paced reading times and ERPs, and compares the results to similar studies done
in English and other languages. Consider subject and object relative clauses in English, as
illustrated below in (1). The sentence fragment in (1a) is called a subject relative (SR) because
the head noun the reporter is the subject of the relative clause (RC) who attacked the senator, i.e.,
the reporter is the one who attacked the senator. The sentence fragment in (1b) is called an
object relative (OR) because the head noun the reporter is the object of the RC who the senator
attacked, i.e., the senator attacked the reporter.
(1) a. Subject relative (SR)
head noun
↓
the reporter
relative clause (RC)
[who __ attacked the senator]
FILLER GAP
b. Object relative (OR)
the reporter
[who the senator attacked __ ]
FILLER
GAP
ERPs to Japanese Relative Clauses 5
A large number of studies using various methodologies have shown that ORs are harder to
process than SRs in English (e.g., reading times: King & Just, 1991; eye-tracking: Traxler,
Morris, & Seely, 2002; ERPs: King & Kutas, 1995; fMRI: Caplan et al., 2001). The
explanation for this is often tied to the notion of filler-gap dependencies. In (1), the displaced
wh-element who is called a filler while the canonical position in the sentence for an element with
its grammatical function (e.g., subject and object) is called a gap, and they are said to be
dependent on each other, as the interpretation of a filler involves associating it with its gap (cf.
Fodor, 1989). While the filler and its gap are immediately adjacent to each other in SRs, there
are words intervening between them in ORs, and this distance between the filler and its gap is
said to be the source of difficulty (e.g., Gibson, 1998, 20001). Simple distance between words in
a sentence is conventionally called linear distance. Alternatively, distance between a filler and
its gap can be characterized in another way, in terms of hierarchical syntactic structure, as shown
in Figure 1 (e.g., Chomsky, 1981). The object gap position is more deeply embedded in the
phrase structure than the subject gap position, so there are more syntactic nodes between the
filler and its gap in ORs than in SRs. This kind of distance is conventionally called structural
distance, and could be another source of comprehension difficulty in ORs (O’Grady, 1997)2.
(Figure 1 about here)
RCs in Japanese differ from those in English in several important ways. Some of those
differences arise because of the differences in the two languages’ basic word orders. With
respect to RCs specifically, Japanese RCs are prenominal (the RC precedes its head noun) as
shown in (2). In English, in contrast, relatives are postnominal, as illustrated above in (1).
Another difference between the two languages is that Japanese has no overt relativizers while
English has relative pronouns such as who and that. Thus, instead of a relative pronoun, it is the
ERPs to Japanese Relative Clauses 6
head noun that functions as the filler in word-by-word sentence processing in Japanese. The
consequence of these differences in Japanese and English is that a gap precedes the head noun
that fills it in Japanese, while a gap follows the relative pronoun that fills it in English. As will
be discussed later, this word order difference is likely to lead to important differences in how
RCs are processed in Japanese and English.
(2) a. Subject relative (SR) in Japanese
relative clause (RC)
head noun
↓
[ __
議員を
[ __
[ __
giin-o
hinanshita]
senator-ACC(USATIVE) attacked ]
非難した]
GAP
記者
kisha
reporter
FILLER
‘the reporter [(who) __ attacked the senator]’
b. Object relative (OR) in Japanese
[ 議員が
__
[ giin-ga
[ senator-NOM(INATIVE) __
非難した]
記者
hinanshita]
attacked ]
kisha
reporter
GAP
FILLER
‘the reporter [(who) the senator attacked __ ]’
Another feature of Japanese that is important here is that it allows pro-drop, i.e., the
complete omission of words that would instead be pronominalized in English. Subject pro-drop
(3a) is much more common than object pro-drop (3b) in Japanese (Ueno & Polinsky, submitted),
but both can occur in certain kinds of discourse.
ERPs to Japanese Relative Clauses 7
(3) a. Subject pro-drop sentence
議員を
非難した
pro
pro
giin-o
hinanshita
pro
senator-ACC attacked
‘(I/you/he/she/we/they) attacked the senator’
b. Object pro-drop sentence
議員が
pro
pro
非難した
giin-ga
hinanshita
attacked
senator-NOM
‘The senator attacked (me/you/him/her/us/them)’
Furthermore, Japanese has a clause type called fact-clause, which is a clause that
modifies certain nouns such as ‘fact’, ‘news’, and ‘rumor’ without a gap, as in ‘the fact that the
reporter attacked the senator’. Following the basic word order of Japanese, these fact-clauses are
also prenominal, as shown in (4). Notice that just as there is no overt relativizer in Japanese RCs,
there is also no complementizer in Japanese fact-clauses, i.e., there is no word equivalent to the
English complementizer that.
(4) Fact-clause
[記者が
議員を
[kisha-ga
[reporter- NOM
giin-o
hinanshita]
senator- ACC attacked]
非難した]
事実
jijitsu
fact
‘The fact [(that) the reporter attacked the senator]’
Fact-clauses also differ from their English parallel in that they allow pro-drop, as shown in (5).
ERPs to Japanese Relative Clauses 8
(5) a. Fact-clause with subject pro-drop
[pro
議員を
非難した]
事実
[pro
giin-o
hinanshita]
jijitsu
fact
[pro
senator-ACC attacked]
‘The fact [(that) (I/you/he/she/we/they) attacked the senator]’
b. Fact-clause with object pro-drop
[議員が
pro
pro
非難した]
事実
[giin-ga
hinanshita]
jijitsu
attacked]
fact
[senator-NOM
‘The fact [(that) senator attacked (me/you/him/her/us/them)]’
Taken together, the combination of the prenominal position of noun-modifiers such as
RCs and fact-clauses, the absence of overt relative markers, and the possibility of pro-drop
means that there can be temporary ambiguity about whether a sequence of words constitutes a
RC, a simple mono-clausal sentence, or a fact-clause. In other words, a RC like (2a) or (2b)
could be initially interpreted as a simple mono-clausal sentence as in (3a) or (3b) or as a factclause as in (5a) or (5b). It is only at the point when the head noun appears that it becomes clear
that the word string must constitute a RC modifying the head noun that has appeared. That these
words constitute a noun-modifying clause is signaled by the fact that a noun follows a verb,
while verbs are always clause-final in Japanese. In other words, if a verb is followed by a noun,
that verb must be the end of a prenominal clause modifying that noun. The possibility that the
sentence is a fact-clause like (5a) or (5b) would be ruled out at the head noun position since the
noun is not a ‘fact’-type noun.
These differences between Japanese and English RCs have several consequences for how
they could be comprehended. First, as discussed above, Japanese RCs are temporarily
ambiguous. In English, comprehenders would typically know that a sequence of words is part of
a RC because many cues, such as a relative pronoun immediately following a head noun, are
ERPs to Japanese Relative Clauses 9
available early in the sentence. In contrast, Japanese comprehenders may not know that there is
a RC noun structure until its very last word, the head noun, appears. Another difference is that in
Japanese, it is SRs that involve a longer linear gap-filler distance as shown in (2), while in
English it is ORs that involve a longer linear filler-gap distance, as shown in (1). While there are
many differences between Japanese and English RCs, they are argued to have similar
hierarchical structures, as shown in Figure 2. In the Japanese structures, there is a covert
operator (an entity that identifies the gap) instead of a relative pronoun coindexed with both the
head noun and the gap (e.g., Kaplan & Whitman, 1995)3, but otherwise the structure is the same
in both languages. In spite of the fact that the left-to-right order of the words in the phrase
structure differs between English and Japanese, the hierarchical relationship of the syntactic
configuration is the same, which means that the object gap position is more deeply embedded
than the subject gap position. Thus, despite surface differences between the two languages that
result in a longer linear distance for ORs in English and a longer linear distance for SRs in
Japanese, in both languages ORs involve a longer structural distance than SRs. Therefore, unlike
English, linear and structural distance accounts of processing costs of RCs yield different
predictions for Japanese, in that a linear distance account predicts a higher processing cost for
SRs while a structural distance account predicts a higher processing cost for ORs. This
dissociation between linear and structural distance in Japanese affords the opportunity to
investigate which is the more important factor in determining the difficulty of RC
comprehension.
(Figure 2 about here)
As discussed above, ORs are typically harder to process than SRs in English, possibly
due to either longer linear filler-gap distance, longer structural filler-gap distance, or perhaps
ERPs to Japanese Relative Clauses 10
both. Earlier studies on the processing of filler-gap dependencies in RCs and wh-questions
(questions involving wh-words such as what and who) argued that associating a displaced whfiller with its gap increases working memory load, and that this processing cost is reflected in an
ERP component known as the left anterior negativity (LAN) between the filler and the gap. The
LAN is a negative voltage deflection that is larger at the front of the head than at the back and is
often left-lateralized. It has been observed in both a slow form, with a duration of several
seconds, and a phasic from, with a duration of a few hundred milliseconds (e.g., Kluender &
Kutas 1993; King & Kutas 1995). For instance, King and Kutas (1995) reported a frontal
bilateral slow negative potential in response to ORs in English (6b) when compared to SRs (6a),
starting after the relative pronoun and continuing through Regions 1 and 3. In addition, there
was a phasic LAN effect immediately following the gap in the OR condition (6b), i.e., at the
main verb admitted in Region 3.
Region 1
(6) a. SR
b. OR
Region 2
The reporter [who __ harshly attacked the senator]
The reporter [who
the senator
Region 3
admitted the error.
harshly attacked __] admitted the error.
Other recent studies have reported P600 effects at the gap location instead of or in addition to
LAN effects. The P600 is a positive peak at around 600 ms after stimulus onset. For instance,
Kaan, Harris, Gibson, and Holcomb (2000) compared embedded yes-no and wh-questions as
shown in (7), and found P600 effects at the pre-gap main verb position (imitated in this example)
in the wh-conditions ((7b) and (7c)).
ERPs to Japanese Relative Clauses 11
(7) a. Emily wondered [whether the performer in the concert had imitated a pop star] for the
audience’s amusement.
b.
Emily wondered [who the performer in the concert had imitated __ ] for the audiences
amusement.
c.
Emily wondered [which pop star the performer in the concert had imitated __ ] for the
audience’s amusement.
Kaan et al. (2000) argued that although the P600 had previously been attributed specifically to
syntactic reanalysis (e.g., Osterhout & Holcomb, 1992), their results showed that it can also
indicate syntactic integration difficulty in general, such as the relative difficulty of integrating
different types of wh-fillers with the rest of the sentence. More recent studies have reported the
combination of both LAN and P600 effects in wh-questions in both English (Phillips, Kazanina,
& Abada, 2005) and German (Fiebach, Schlesewsky, & Friederici, 2001, 2002).
In English, both wh-questions and RCs involve wh-movement and the filler-gap
dependencies that necessarily result. Thus, it is not surprising that the same ERP components
have been found to be sensitive to properties of both RCs and wh-questions in English and
similar languages. However, in Japanese wh-words typically stay in situ (i.e., they remain in
their canonical subject/object position in the sentence), and thus wh-questions are syntactically
distinct from RCs. Therefore, the processes underlying the comprehension of these two kinds of
structures in Japanese may be less similar than they are in English, and thus perhaps less likely to
lead to similar ERP effects. Japanese wh-questions have been investigated using ERPs (e.g.,
Ueno & Kluender, 2003), but Japanese RCs have not, to the best of our knowledge. However,
Japanese RCs have been investigated using reading time measures in a small number of studies
(Ishizuka, Nakatani, & Gibson, 20034; Miyamoto & Nakamura, 2003), as have Korean RCs,
which have very similar syntactic properties (Kwon, Polinsky, & Kluender, in press, in prep.).
These studies have shown that ORs take longer to read than SRs at or starting at the head noun
ERPs to Japanese Relative Clauses 12
position. Notice that it is ORs that produce slower reading times, just as in English. Since ORs
involve longer structural distances in both languages, while only in English do they also have
longer linear distance, these results suggest that structural distance may be the more important
kind of distance. Notice also that the difference between Japanese SRs and ORs does not appear
until the head noun position at the end of the RC, where it finally becomes clear that the structure
is in fact a RC, since it is not until then that it is both clear that there is a RC and what kind of
RC it is. This structural ambiguity contrasts with English, where it is clear very early on that
there is a RC, and also which kind it is (by whether is it a verb or a noun that follows the relative
pronoun, see (1) above), and where differences between SRs and ORs emerge in various
processing measures beginning immediately after the relative pronoun.
Given all of the above, we wanted to find out whether the kinds of LAN and/or P600
effects that have been found to be sensitive to filler-gap processing in English would replicate
with gap-filler processing in Japanese, and also whether these effects in Japanese would correlate
with linear or structural distance. If Japanese gap-filler association is like English filler-gap
association, we might expect to see LAN effects between gap and filler, possibly followed by
P600 effects when the gap is filled at the filler. However, since the presence of a gap is not
clearly indicated until the filler is encountered in Japanese, it is possible that there will not be any
differences between SRs and ORs until at or after the head noun position. Additionally, if linear
distance is more important, we should see more processing costs indexed by these ERP effects
for Japanese SRs in comparison to ORs. On the other hand, if structural distance is more
important, as suggested by the reading time studies described above, we would expect to see
ERP effects indexing difficulty in ORs.
ERPs to Japanese Relative Clauses 13
Experiment 1
Experiment 1 used a self-paced reading task to examine reading times for the
experimental stimuli, in order to determine whether our stimuli would show the same pattern of
results seen in previous reading time studies, in which ORs took longer to read than SRs.
Method
Participants
Forty native speakers of Japanese (31 females, 9 males; age 18-41 years, mean 26 years)
participated in the experiment. They were residents of the San Diego or Urbana-Champaign area
and had been outside of Japan for less than 10 years. Participants were reimbursed for their time.
Materials
Stimuli consisted of pairs of sentences with singly-embedded Japanese (a) SRs and (b)
ORs, as shown in Table 1. Note that on the surface, the only difference between SR and OR
conditions was the case marker of the first noun in the RC (‘senator-ACC’ for SRs and ‘senatorNOM’
for ORs in the example). All the experimental items had the same syntactic structures and
the same number of words as the example sentences shown in Table 1.
(Table 1 about here)
Following Ishizuka et al. (2003), dative-topic-marking was used on the head noun in
order to control for potential confounds between RC type and both case-mismatch (Sauerland &
Gibson, 1998) and perspective-shift (Sheldon, 1974; MacWhinney, 1982). The case-marking on
the head noun indicates its grammatical function in the main clause, rather than its function in
the RC. If the head noun is the subject of the main clause, it has a nominative case marker,
which is also the case marker it would have in an SR if it were actually present there, since it is
also the subject of the RC. In contrast, when a main clause subject noun is modified by an OR,
ERPs to Japanese Relative Clauses 14
its case-marking is still nominative, but the case-marking that it would have if it were actually
present in the RC would be accusative, since it is the object in the RC. This match or mismatch
between the case-marking on the head noun of the RC and its grammatical function within the
RC is necessarily confounded with RC type, and has been argued to lead to differences in
processing difficulty between the two kinds of RCs (Sauerland & Gibson, 1998). To avoid this
potential confound, the head nouns in the experimental items were all given dative-topic marking
so that there would be an equivalent discrepancy between the case marking on the head noun and
the case marking it would receive if it were actually present within the RC for both SRs (dativetopic/oblique vs nominative/subject) and ORs (dative-topic/oblique vs accusative/object). This
approach also controlled for a related potential confound between RC type and degree of
perspective shift required between the relative and main clauses, which has also been argued to
influence the processing difficulty of RCs (Sheldon, 1974; MacWhinney, 1982). Since the head
noun is an oblique instead of the subject in the main clause, there should be a similar degree of
perspective shift between the main and relative clauses for both SRs and ORs.
In order to confirm that nouns used at the head noun position (‘reporter’ in the example in
Table 1) were equally plausible as both the subject or the object of the RC verb (‘attacked’), 20
native speakers of Japanese (who did not participate in Experiment 1 or Experiment 2) rated
simple transitive sentences that were created by replacing the gap in each type of RC with the
corresponding head noun, as in ‘The reporter attacked the senator’ vs ‘The senator attacked the
reporter’. One hundred pairs of such sentences were constructed and distributed over two lists in
a Latin square design and intermixed with 50 filler sentences in a pseudo-random order.
Participants rated the sentences on a scale from ‘1’ (strange) to ‘5’ (natural). Twenty pairs of
sentences that yielded the largest within-pair differences were discarded. Ratings for the
ERPs to Japanese Relative Clauses 15
remaining 80 pairs did not differ significantly [F1(1,18) = 1.02, p > .1; F2(1,78) = 2.58, p > .1]
when the head noun was used as the subject (mean rating = 4.4) or the object (4.3).
Another norming study was conducted to test whether the verbs used in the RC were
strongly transitive-biased. This was done to try to ensure that it would be clear at the RC verb in
both SRs and ORs that one of its arguments was missing (i.e., its agent/subject in an SR and its
patient/object in an OR). If the RC verbs were not strongly transitive, then the absence of an
accusative-marked noun preceding them in ORs could simply mean that the verb was being used
intransitively in a mono-clausal sentence or a fact-clause, rather than that an argument was
missing. In contrast, in SRs, where there is no nominative-marked noun preceding the RC verb,
it is clear that an argument is missing, although there are still multiple possibilities with regard to
the structure of the sentence. It could be a mono-clausal sentence with pro-drop, or a fact-clause
with pro-drop, or a RC. Thus, the appearance of the RC verb would provide different kinds of
information in SRs and ORs, which could lead to processing differences that are not strictly due
to RC type. For example, processing might be easier at the RC verb in ORs because it is still
possible at that point to interpret the word string as an intransitive construction in a mono-clausal
sentence or a fact-clause, while in SRs it becomes clear at the point that an argument is missing.
At the next word, the head noun of the RC, relative difficulty could then reverse because the
appearance of a noun following an intransitive-bias verb would be the first cue in ORs that there
is a RC, while in SRs there was already some indication at the RC verb itself To try to equate
the informativeness of the RC verb about the possible presence of a RC in both SRs and ORs, the
bias of the verbs was normed to ensure that only strongly transitive verbs were used, so it would
be clear at the verb that an argument was missing in both types of RCs. It is impossible,
however, to completely avoid some asymmetry in the cues provided by the RC verb and the head
ERPs to Japanese Relative Clauses 16
noun in SRs and ORs, because subjects are much more likely to be dropped than objects (Ueno
& Polinsky, submitted). Thus, when a strongly transitive verb appears after only a nominativemarked noun, it may seem stranger than when the same verb appears after just an accusativemarked noun, since subjects are generally dropped more often than objects are. This asymmetry
in subject/object pro-drop must be kept in mind when examining the pattern of results in our
studies.
Another reason to use only strongly transitive verbs was to avoid another possible reason
for ORs to be more difficult than SRs at the head noun. If the verb were intransitive-biased and
then it became clear at the head noun that it nonetheless has an object in this sentence, that might
lead to greater processing cost that is again not strictly due to RC type.
The 69 different verbs used in the RC of the stimulus sentences were combined with 31
filler verbs that can never take direct objects (e.g., ‘rise’) and given to each participant in a
different pseudo-randomized order. Ten native speakers of Japanese, a subset of those who had
participated earlier in either Experiment 1 or Experiment 2, did the norming study 0-369 days
(mean 146 days) after participating in the original experiment. Participants were presented with
the verbs and asked to type the first sensible sentence that came to mind for each verb.5
Collapsed across participants, 96% of the sentences that were generated with the RC verbs had
overt direct objects, confirming that these verbs were highly transitive-biased. Interestingly,
only 65% of these sentences had overt subjects (= 35% subject-drop), indicating that subjects can
easily be dropped even in the total absence of discourse context.
Eighty pairs of stimulus sentences representing the two experimental conditions were
constructed and placed in a Latin square design to create two parallel lists containing 80
experimental items each, such that no one participant saw more than one sentence from any pair,
ERPs to Japanese Relative Clauses 17
and every participant saw an equal number of items in each condition. Filler items consisted of
80 sentences of five different types of bi-clausal and mono-clausal constructions (16 sentences
per type), including (a) embedded wh-questions, (b) bi-clausal declaratives, (c) ditransitives, (d)
transitives in canonical word order, and (e) transitives in scrambled word order. The 80 filler
sentences were added to each list, and then sentences in these two lists were pseudo-randomized
and divided into 5 blocks of 32 sentences each.
Procedure
Participants were seated in front of an IBM compatible laptop computer running the EPrime software package (Schneider, Eschman, & Zuccolotto, 2002). Participants were timed in a
word-by-word self-paced non-cumulative reading task. Stimuli were presented on the computer
screen in Japanese characters one bunsetsu at a time. A bunsetsu consists of one free morpheme
(lexical word or pronoun) and the bound morpheme/s associated with it (particles modifying the
noun/verb), and will be referred to as a “word” hereafter. Each word was presented at the center
of the screen and participants pressed the spacebar to reveal each subsequent word of the
sentence. Yes/no comprehension questions were presented after each sentence and participants
pressed one of two keys on the keyboard to answer them, after which they received feedback.
Before beginning the experiment, participants were given a practice set of 20 sentences. The
experiment took participants approximately 45 minutes to complete.
Data Analysis
Analyses were conducted on both question-response accuracy and reading times per word.
Reading times were trimmed so that data points beyond 2 standard deviations from the relevant
subject x condition x position cell mean were replaced with the corresponding cutoff value,
affecting 5% of the data. The means and analyses presented below are based on the trimmed
ERPs to Japanese Relative Clauses 18
reading times. For each sentence position as well as for the post-RC region, a repeated measures
analysis of variance was conducted with RC type as a within-group factor, and either subject
(with list as a nested factor) (F1) or item (with item group as a nested factor) (F2) as a random
factor. An alpha level of .05 was used for all statistical tests, with a p-value of .10 considered
marginally significant.
Results
Comprehension Task Accuracy
The mean correct response rate to all the comprehension questions across subjects was
92% (range 81-100%, S.D. 5%). Thus no subject's data were excluded from the reading time
analyses based on poor comprehension. The mean correct response percentage did not differ
significantly between SRs (91%) and ORs (90%) [Fs < 1].
Reading Times
(Figure 3 about here)
Figure 3 shows the trimmed reading times by sentence position. In the RC region (words
1-3), SRs took longer to read than ORs at the first word position (the subject modifier, ‘new’ in
Table 1) [F1(1,38) = 4.74, p < .05; F2(1,78) = 8.10, p < .01], although items were completely
identical across conditions at this point (see Discussion). There was no significant difference at
the second word position (the RC noun, ‘senator-ACC/NOM’) [Fs < 1]. At word 3 (the RC verb,
‘attacked’), however, SRs took longer than ORs, but this was reliable only in the item analysis
[F1(1,38) = 2.27, p > .1; F2(1,78) = 8.25, p < .01].
In the post-RC region (words 4-7), ORs took longer than SRs at the head noun (word 4
‘reporter-DAT-TOP’) [F1(1,38) = 5.60, p < .05; F2(1,78) = 4.79, p < .05], and also at the following
word 5 (‘long-term’), though only marginally so [F1(1,38) = 2.89, p < .1; F2(1,78) = 3.30, p < .1].
ERPs to Japanese Relative Clauses 19
There were no significant differences at word 6 (‘colleague-NOM’) [Fs <1] or word 7 (‘existed’)
[F1(1,38) = 1.19, p > .1; F2(1,78) = 1.01, p > .1]. When reading times for the entire post-RC
region (w4-w7) were collapsed together, ORs were read significantly more slowly than SRs
[F1(1,38) = 8.24, p < .01; F2(1,78) = 6.99, p < .01].
Discussion
Slow reading times in the first word position must have been due to noise, given that the
word in that position was identical in both conditions. The slowdown did not extend to the
following word and so will be ignored in the remainder of the discussion. Spurious effects are
sometimes observed on the first word in the sentence in reading time studies because participants
get ahead of themselves and unintentionally press the button starting the next sentence when they
meant to take a short break between trials. They then take that break, making times at the first
word more variable (cf. Miyamoto & Nakamura, 2003).
At word 2, different case markers (nominative in ORs and accusative in SRs) on the noun
did not lead to reading time differences, even though it is more typical to start a sentence with a
nominative- than accusative-marked noun. However, in the self-paced reading paradigm, effects
are often delayed a word, so we could speculate that the accusative-marked noun in SRs caused a
slowdown one word later at word 3, although the difference was reliable only by items. A
similar trend has been reported in other reading time studies in Japanese and Korean (Miyamoto
& Nakamura, 2003; Kwon et al., in press, in prep.), with SRs taking longer to read than ORs
within the RC region. Miyamoto and Nakamura reported longer reading times for SRs when all
the words within the RC were collapsed together, but they did not provide a plausible
explanation for this effect. Kwon et al. reported longer reading times for SRs at the word
immediately following the accusative-marked noun position in the RC, which is similar to our
ERPs to Japanese Relative Clauses 20
results, though in Kwon et al.’s materials that word was an adverb while in ours it was a verb.
Kwon et al. (in prep.) attributed the effect to the possibility that the parser had to look for the
missing subject after seeing the accusative-marked noun in SRs, while there was no such need in
ORs since they only contained a nominative-marked noun. A similar process might have been
taking place in our study as well.
Upon reaching the head noun position, the relative difficulty of ORs and SRs reversed,
with ORs taking longer than SRs at both the head noun (word 4) and the following word (word
5), and also when the reading times were combined across the entire post-RC region (words 4-7).
Thus there seems to be a reliable processing cost for ORs, starting from the position which
clearly indicates that there is a RC, replicating previous studies (Ishizuka et al., 2003; Miyamoto
& Nakamura, 2003; Kwon et al., in press, in prep.).
Notice that the reading times on the head noun were the longest in the sentence for both
SRs and ORs. One might wonder whether this was due to some lexical property of the head
nouns, rather than to RC processing difficulty. The words at this position actually had slightly
higher character counts (4.6 on average) than other nouns in the sentence (3.6-3.7 characters on
average), probably due to the extra topic-marker in addition to the dative-marker. However,
word length could not be the reason for the long reading times, since we also analyzed lengthcorrected residual reading times (calculated as described in Ferreira and Clifton (1986)) and
found that the pattern of results was identical to the one shown in Figure 3. Long reading times
on the head noun are also not due to the frequency of the head nouns, since the head nouns were
no lower in frequency (mean frequency = 12,848) than the other nouns in the sentence (word 2,
mean frequency = 10,510; word 6 = 11,552; F < 1), based on a corpus count of a popular
Japanese newspaper (Amano & Kondo, 2000)6. Nor was it likely to be due to the visual
ERPs to Japanese Relative Clauses 21
complexity of the head nouns, since the character counts of Chinese characters, which tend to be
more visually complex than syllabic characters, were not significantly higher for the head nouns
(mean Chinese character count = 1.96) than for the other nouns in the sentence (word 2 in RC =
1.89; word 6 in main clause = 1.83; F < 1).
Therefore, the long reading times at the head noun position seem most likely to be due to
the difficulty of processing the RC structure for both SRs and ORs. This is the point in the
sentence where the parser should be constructing the RC structure and filling the gap with the
head noun for both SRs and ORs, and the reading times show that that is harder in ORs. This
provides additional support for the structural distance account of the RC difficulty, since it is
ORs that have greater structural distance in Japanese, just as in English.
Experiment 2
Experiment 2 investigated ERPs in response to Japanese SRs and ORs. To recap, our
major questions were (a) whether there would be LAN and/or P600 effects in Japanese RCs like
those observed previously for English RCs and wh-questions (Kluender & Kutas, 1993; King &
Kutas, 1995; Kaan et al., 2000; Phillips et al., 2005) and (b) whether ERP results would lead to a
similar conclusion as the reading time results showing that Japanese ORs are more difficult, and
thus provide additional support for the structural distance account of the difference.
Methods
Participants
Thirty-three native speakers of Japanese (23 females, 10 males; age 20-34 years, mean 26
years) who did not participate in Experiment 1 were included in the study.7 Participants were
right-handed and had normal or corrected-to-normal vision. They were residents of the Urbana-
ERPs to Japanese Relative Clauses 22
Champaign area and had been outside of Japan for less than 10 years. Participants were
reimbursed for their time.
Materials
The stimuli and design were identical to Experiment 1.
Procedure
Participants were seated facing a computer monitor in a sound-attenuated room. Stimuli
were presented on the center of a computer screen in Japanese characters one word (or bunsetsu)
at a time with 450 ms duration and 650 ms8 stimulus onset asynchrony. Yes/no comprehension
questions were presented after each sentence and participants responded using the mouse, after
which they received feedback. Before beginning the experiment, participants were given a
practice set of 20 sentences. Participants were given as much rest as they wished between blocks
of sentences. The experiment took participants approximately 2.5 hours.
Electrophysiological Recording
The electroencephalogram (EEG) was recorded from 25 positions, using Ag/AgCl
electrodes attached to an elastic cap (Easy-Cap). Electrodes were positioned on the two mastoid
processes, and data were collected using the left mastoid as the reference. The EEG was later
algebraically re-referenced to the mean of the activity at these two mastoid electrodes. To detect
blinks and lateral eye-movements for later correction, additional electrodes were placed beneath
the right eye and at the outer canthi of both eyes. Impedances were kept below 10KΩ. The EEG
was amplified with a bandpass of 0.01 to 30 Hz, digitized at 200 Hz, and stored for off-line
analysis. Data with excessive blinks were corrected using an eye movement correction
procedure (Gratton, Coles, & Donchin, 1983).
ERPs to Japanese Relative Clauses 23
Data Analysis
Analyses were conducted on question-response accuracy and ERP waveforms. ERP
measurements were examined for two-word averages to reveal phasic effects, and four-word
averages for longer-lasting effects. The analysis epochs in the two-word averages were 1750 ms
long, including a 100 ms prestimulus baseline, while four-word averages were 3050 ms long,
again including a 100 ms prestimulus baseline.
The statistical analyses were done separately on midline (Fz, Cz, and Pz), parasagittal
(AF3/4, F3/4, FC3/4, C3/4, CP3/4, P3/4, PO7/8), and temporal (F7/8, FT7/8, T3/4, T5/6)
electrodes, as well as on individual electrodes. Midline analyses consisted of repeated measures
ANOVAs with two within-group factors, including two levels of RC type and three levels of
anterior/posterior sites, and subject (with list as a nested factor) as a random factor. Parasagittal
analyses consisted of repeated measures ANOVAs with three within-group factors, including
two levels of RC type, seven levels of anterior/posterior sites, and two levels of hemisphere.
Temporal analyses consisted of repeated measures ANOVAs with three within-group factors,
including two levels of RC type, four levels of anterior/posterior sites, and two levels of
hemisphere. An alpha level of .05 was used for all statistical tests, with a p-value of .10
considered marginally significant. The Huynh-Feldt correction for lack of sphericity was applied
whenever applicable (Huynh & Feldt, 1976). Original degrees of freedom are reported with the
corrected probability level.
Results
Comprehension Task Accuracy
The mean correct response rate to all the comprehension questions across subjects was
91% (range 84-98%, S.D. 4%). Thus no subject's data were excluded from the ERP analyses
ERPs to Japanese Relative Clauses 24
based on poor comprehension. The mean correct response percentage did not differ significantly
between SRs (89%) and ORs (87%) [F1(1,31) = 1.60, p > .1; F2(1,78) = 1.02, p > .1].
ERPs
Six percent of the data were rejected due to eye movement and other artifacts. Sentences
were compared in the RC region (RC + head noun, ‘new senator-ACC/NOM attacked reporterDAT-TOP’
in Table 1), to examine how gap-filler dependencies in Japanese RCs were processed,
and also at the post-RC region (head noun + remainder of the sentence, ‘reporter-DAT-TOP longterm colleague-NOM existed’), to determine whether there were any effects of RC type after the
gap was filled by the head noun.
Relative clause (RC) region. Visual inspection of the RC region showed bilateral anterior
negativity at the RC verb and head noun position (‘attacked reporter- DAT-TOP’ in Table 1) of
ORs (see Figure 4). To quantify this observation, ANOVAs were performed in the latency
window of 300 to 1250 ms poststimulus onset of ‘attacked’, covering the 300-600 ms (standard
latency for LAN) windows of both ‘attacked’ and ‘reporter-DAT-TOP’. A significant or marginal
main effect of RC type was found in both the midline [F(1, 31) = 7.36, p < .05] and parasagittal
[F(1, 31) = 3.04, p < .1] arrays. In addition, there was a marginal interaction between RC type
and anteriority in the midline array [F(2, 30) = 3.04, p < .1]. ANOVAs run on individual
electrodes in the same latency window revealed a significant main effect of RC type at Fz, AF3,
F4, FC4, and C4, as well as a marginal main effect at Cz, F3, and C3. These effects indicated
that ORs were more negative than SRs, especially at frontal regions.
(Figure 4 about here)
Post-relative clause (RC) region. Figure 5 shows the comparison between RC type at the
post-RC region, from the head noun position to the sentence end (‘reporter-DAT-TOP long-term
ERPs to Japanese Relative Clauses 25
colleague-NOM existed’ in Table 1). Visual inspection of these four-word averages showed a
slow continuous divergence between SRs and ORs, in which ORs were more positive than SRs,
starting at about 500 ms after the onset of ‘reporter-DAT-TOP’ and continuing throughout the
epoch to the sentence end. To capture this observation, ANOVAs were performed in the latency
window of 500 ms to 2950 ms poststimulus onset of ‘reporter-DAT-TOP’. There was a significant
or marginal main effect of RC type in all three arrays [midline: F(1, 31) = 7.77, p < .01;
parasagittal: F(1, 31) = 4.82, p < .05; temporal: F(1, 31) = 3.36, p < .1], as well as a significant
RC type x anteriority interaction in the temporal array [F(3, 29) = 3.38, p < .05]. ANOVAs run
on individual electrodes in the same latency window revealed a significant main effect of RC
type at Fz, AF4, T4, C4, T5, T6, CP3, P3, and P4, as well as a marginal main effect at Cz, C3,
PO7, and PO8. These effects indicated that ORs were more positive than SRs, especially at
centro-posterior regions.
(Figure 5 about here)
Discussion
To summarize, ORs elicited greater bilateral anterior negativity than SRs at the RC verb
and head noun positions in the RC region (Figure 4). In the post-RC region starting from the
head noun and continuing to the end of sentence, ORs elicited greater centro-posterior positivity
than SRs (Figure 5). The anterior effect in the RC region has the typical latency and anterior
scalp distribution for the LAN, but it is bilaterally distributed rather than lateralized. Although
anterior negativities have been found to be left-lateralized to varying degrees in several studies
(hence the usual name LAN), there have also been previous observations of bilaterally
distributed and even right-predominant versions of anterior negativities (e.g., Müller, King, &
Kutas, 1997; Ueno & Kluender, 2003). The functional and neurophysiological significance of
ERPs to Japanese Relative Clauses 26
different scalp distributions for anterior negativities during sentence processing remain to be
determined. We assume here that both bilaterally distributed and lateralized versions signal
similar kinds of processing difficulty, though differences in the nature and/or degree of difficulty
may contribute to differences in scalp distribution. The centro-posterior effect in the post-RC
region is harder to clearly identify as one of the ERP components previously observed in
sentence processing studies. In what follows, we will discuss each of these effects in turn.
Relative Clause (RC) Region
One possible interpretation of the anterior negativity observed in the RC region (as
schematized in (8)) is that it reflects demands placed on working memory by both the storage of
a gap and its subsequent retrieval for gap-filling (cf. Kluender & Kutas, 1993). On this
interpretation, the parser recognizes the gap in ORs upon seeing a transitive-biased verb
(‘attacked’ in (8b)) immediately following a nominative-marked noun, since there is no object in
the usual position. This leads to processes that tax working memory until the object-gap in the
RC is filled by the head noun. The parser would also recognize a gap in SRs either immediately
at the accusative-marked noun or one word later at the RC verb (since at the accusative-marked
noun itself, it is possible that the subject noun has been dropped, or that the accusative-marked
noun is scrambled out of the default subject-object order), but this seems to be easier, as
indicated by the ERP results.
(8) a. SR [ ___ new senator-ACC attacked] reporter-DAT-TOP ...
AN
b. OR [new senator-NOM ___ attacked] reporter-DAT-TOP ...
As pointed out earlier, another possible explanation has to do with differences in the
likelihood of subject- and object-drop in Japanese. It is much more common to drop subjects
ERPs to Japanese Relative Clauses 27
than objects with transitive-biased verbs like those used in this experiment, as shown in our
norming test. If the sentence is parsed as a simple mono-clausal sentence (or a fact-clause with
pro-drop like (5a) or (5b)) up until the head noun, at the RC verb the ORs appear to be missing
an object, while SRs appear to be missing a subject. Since missing objects are less common, the
appearance of a missing object in the ORs may contribute to a greater processing cost. In
addition to reflecting increased working-memory load for the processing of structures involving
filler-gap dependencies, anterior negativities have also been linked to syntactic processing
difficulty in general, such as that triggered by violations of phrase structure (e.g., Neville, Nicol,
Barss, Forster, & Garrett, 1991) or morphosyntactic constraints (e.g., Coulson, King, & Kutas,
1998), by non-preferred disambiguations of temporarily ambiguous sentences (Kaan & Swaab,
2003), and by case-marked determiners (in German) that indicate that argument nouns are not in
their canonical order (Rösler, Pechmann, Streb, Röder, & Hennighausen, 1998)9. Thus it would
be possible to link the anterior negativity found in the present study to the processing of a
syntactically atypical construction with object-drop. Participants might have (incorrectly)
recognized and reacted to such a construction at the RC verb position, and the effect might have
carried over to the following head noun position.
Post-Relative Clause (RC) Region
The slowly developing divergence between SRs and ORs across this sentence region has
a scalp distribution like that of P600, but a timecourse that is not at all like the usual P600
timecourse. Thus, it is not clear whether or not this effect is related to previously observed P600
effects, or to an entirely different phenomenon. Given that, it is not even clear whether to take
the perspective that ORs produce more centro-posterior positivity or that SRs produce more
centro-posterior negativity. However, since previous observations of ERP effects related to
ERPs to Japanese Relative Clauses 28
sentence-processing difficulty due to factors other than word meaning have all found greater
difficulty to be associated with anterior negativity and/or posterior positivity, it seems to make
more sense to describe the posteriorly distributed effect10 observed here as greater positivity in
ORs. In addition, the fact that reading times on the RC head noun were longer for ORs than for
SRs in Experiment 1 suggests that it is ORs that are more difficult, providing support for the idea
that it makes more sense to describe the ERP effect as greater positivity for ORs. However,
before further interpreting the results from that perspective, it is important to rule out a couple of
other alternatives.
A somewhat remote possibility is that the effect observed here is related to N400 effects,
in which words whose meanings are harder to integrate with their contexts lead to greater centroposterior negativity. Such effects have been reported at the end of sentences containing syntactic
and/or semantic anomalies that evoke P600 and/or N400 effects earlier in the sentence
(Osterhout & Holcomb, 1992; Hagoort, Brown, & Groothusen, 1993; Osterhout & Nicol, 1999),
and have been interpreted as showing that the kinds of processing problems that initially evoke
N400/P600 effects also make it difficult to integrate the meanings of subsequent words. If that
were the source of the effect found here, it would be more appropriate to describe it as greater
negativity for SRs. However, unlike these studies, our effect started at the fourth word from the
end of grammatical, rather than ungrammatical, sentences, and it was not preceded by
P600/N400 effects earlier in the sentence. Thus, it seems unlikely that the posteriorly distributed
effect observed here is related to these N400-like “sentence-end wrap-up” effects.
Another possibility concerns artifacts that can be introduced by the waveform baselining
process. The larger anterior negativity elicited during ORs started at the RC verb and extended
over the head noun following it. Re-baselining the 4-word waveforms starting at the head noun
ERPs to Japanese Relative Clauses 29
thus could artifactually introduce a later difference in the opposite direction. That is, if the
greater anterior negativity elicited by ORs slowly returned to a more positive default state over
the words following the head noun, forcing the waveforms together at the start of the head noun,
as the baselining procedure does, would make it appear that the waveforms gradually diverged in
the opposite direction from their earlier divergence, when in fact they were just continuing to
converge from that earlier divergence. However, if that were the source of the gradually
increasing positivity observed in OR sentences baselined at the head noun, its scalp distribution
should be the same as that for the negativity observed earlier in the sentence, i.e., larger
differences at the front of the head than at the back. Since the scalp distribution of the difference
observed here starting at the head nouns was larger over the back of the head than the front, it
cannot be explained in this way11. Therefore, we now return to an interpretation of the
differences as greater positivity in response to ORs and continue our discussion.
The difference has a reasonable scalp distribution for a P600 effect, yet is in the form of a
steady long-lasting shift instead of a local peak. However, LAN effects have been reported in
both phasic and long-lasting versions (e.g., Kluender & Kutas, 1993; King & Kutas, 2005;
Fiebach et al., 2001, 2002), so perhaps the same variations are possible for P600 effects. In fact,
Van Petten and Kutas (1991) reported a slow positive shift in response to syntactically
incoherent sentences, such as Be place prefer the was city it and sure be perfume, and more
recently, Casado, Martín-Loeches, Muñoz, & Fernández-Frías (2005) reported a long-lasting
positive shift for a phrase structure reallocation process in Spanish. These effects are argued to
index syntactic processes typically related to the P600, and the positivity effect in our experiment
may be a similar variant.
ERPs to Japanese Relative Clauses 30
The continuous positivity starting at the head noun is more clearly due specifically to RC
processing than is the anterior negativity evoked by earlier words, since there is no longer any
ambiguity about whether the sentence fragment so far is a part of a mono-clausal sentence or a
fact-clause, or instead contains a RC. Consistent with previous results obtained by Kaan et al.
(2000), this positivity seems to index the greater syntactic integration costs of filling object-gaps.
This fits well with the structural distance account of differences in difficulty between SRs and
ORs, in that the object gap position is more deeply embedded and thus more structurally distant
from the head noun, leading to more retrieval and integration costs.
However, one may wonder why the effect was long-lasting rather than phasic. As shown
in the reading times, the head noun position for both SRs and ORs seems to require substantial
processing resources, since it had the longest reading times in the sentence. This is the position
in the sentence where it becomes clear that there is a RC, so if participants have not begun to do
so earlier, this is where they have to recognize and construct a RC structure. It may have been
especially difficult to simultaneously do both that and search farther down the phrase structure
for the gap for ORs, which may have led to a sustained positivity. A possibly related finding in
Kluender and Kutas (1993) was that LAN effects were observed across several words after a gap
had been filled, suggesting that gap-filling is a resource-demanding process whose effects
continue to be felt for some time over subsequent words.
Another possibility is that the sustained positivity reflects reanalysis costs, as the P600
has been argued to do (e.g., Osterhout & Holcomb, 1992; Osterhout, Holcomb, & Swinney,
1994). However, that is unlikely to be the case here. First, because the verbs used in the RC
were highly transitive-biased, it is unlikely that participants expected OR sentences to continue
as an intransitive construction and then had to revise that analysis at the head noun position.
ERPs to Japanese Relative Clauses 31
Second, as discussed above, subject-drop is much more common than object-drop in Japanese.
Thus SRs involving missing subjects should have been more likely to make participants think
that the sentence would continue as a mono-clausal sentence (or perhaps a fact-clause) than ORs
with missing objects, and this should have resulted in a higher reanalysis cost at the head noun
position in SRs rather than in ORs.
General Discussion
OR > SR
The two experiments reported here both suggest that Japanese ORs are harder to process
than SRs. However, there were some discrepancies between them. In particular, in the RC
region the effects appear to go in opposite directions for reading times and ERPs. At the RC
verb, there was a trend in Experiment 1 for slower reading times in SRs than in ORs, suggesting
that SRs had a higher processing load at that point. In contrast, at that same position in
Experiment 2, there was more anterior negativity elicited by the RC verb in ORs than in SRs,
suggesting a higher processing load for the ORs at that point.
We speculated in the discussion of Experiment 1 that one factor contributing to longer
reading times for the RC verb in SRs may have been the atypicality of a sentence beginning with
an accusative-marked noun, as the SRs did. Since effects in self-paced reading times are
sometimes delayed a word, perhaps reading times were longer at the verb because of spillover
from the previous accusative-marked noun, due to its atypicality as the first noun in the sentence.
There is no reason to expect effects to be similarly delayed a word in the ERPs, so it would be
possible to get different patterns of results at the RC verb in the two kinds of measures.
However, although this account suggests that there should have been some kind of difference in
the ERP waveforms evoked by the accusative- vs nominative-marked nouns themselves, there
ERPs to Japanese Relative Clauses 32
was no such difference.12 Instead, the earliest difference between conditions in the ERP
waveforms was an anterior negativity to the RC verb in ORs, which continued over the following
head noun. In the discussion of Experiment 2, we suggested that this may have been due either
to the greater difficulty of processing of the gap-filler dependency in ORs, or to the noncanonicity of object-drop if a mono-clausal or fact-clausal interpretation was being constructed.
The discrepancy between reading times and ERPs might have occurred because multiple
different kinds of processes were taking place at the RC region, as the parser had to entertain
multiple possibilities because of the degree of temporary ambiguity at that point in the sentences.
Thus it is possible that the two methodologies are sensitive to different aspects of the processing
that was taking place at that point.
Starting at the head noun position, which clearly indicates the RC structure, both reading
times and ERPs showed extra processing costs for ORs that continued to the end of the sentence.
In addition, although the difference was never statistically significant, mean accuracy on
comprehension questions was also slightly lower for ORs than for SRs in both the reading time
study (SR: 91% vs OR: 90%) and the ERP study (SR: 89% vs OR: 87%)13, suggesting extra
processing costs for ORs.
What Kind of ERP?—Linear vs Structural Distance
Our reading time experiment replicated the results of previous studies finding that ORs
took longer to read than SRs starting at the head noun position. Our ERP experiment showed
further that similar ERP components are elicited by English and Japanese ORs: anterior
negativity is observed between filler and gap or gap and filler, and posterior positivity is
observed when the gap is filled, although the posterior positivity is phasic for English and longlasting for Japanese. If we interpret the anterior negativity effect in the RC region as associated
ERPs to Japanese Relative Clauses 33
with gap-filler dependencies, we could further say that the pattern of results involving anterior
negativity and posterior positivity suggests similar parsing operations for both filler-gap
dependencies in English and gap-filler dependencies in Japanese.14 However, in the discussion
of Experiment 2 we raised a possible alternative interpretation of the anterior negativity that is
not specifically about RC processing, i.e., that ORs initially appear to be missing an object,
which is more unusual than a missing subject in Japanese sentences. Therefore, what we can say
with more certainty is that gap-filler association in Japanese RCs seems to involve a long-lasting
integration process after the filler, as indexed by the continuous posterior positivity in ORs. If
we assume that the continuous positivity indicates greater syntactic integration cost for ORs,
both ERP and reading time data are more consistent with a structural rather than linear distance
account of RC processing difficulty in Japanese.
An Alternative Interpretation
We have been focusing on the linear vs structural distance accounts, but we are aware
that this is not the only possible interpretation. It could be that the discourse prominence of
subjects over objects causes the extra processing cost to ORs. Some linguists have proposed that
Japanese RCs do not involve syntactic movement and that gaps are actually pros (null argument
pronouns) instead of traces left by syntactic movement (e.g., Comrie & Horie, 1995; Matsumoto,
1989; Murasugi, 2000). On this interpretation, the gap in the RC is identified in the same way as
a pronoun, either by an entity in a given discourse or by a cataphoric (forward reference,
backwards anaphora) link with the head noun in the absence of plausible discourse referents.
Since there was no discourse context in the present study, gaps in our stimuli would have been
identified in the latter manner. 15 Then rather than gap-filler distance (either linear or structural),
ERPs to Japanese Relative Clauses 34
what would actually matter is that the subject pro is easier to link cataphorically with the head
noun than the object pro is, possibly due to its discourse prominence.
There has been much work showing that antecedents in the subject position are easier to
access than antecedents in the object position in anaphor resolution (e.g., Frederiksen, 1981;
Crawley, Stevenson, & Kleinman, 1990), suggesting that subjects are more prominent than
objects for pronoun reference. To the best of our knowledge, there has not been much work
from the opposite perspective about whether cataphoric/anaphoric pronouns in the subject
position are easier to identify than those in the object position. However, one recent reading time
study in Korean (Kwon et al., in press, in prep.) has reported a higher processing cost when
object pros are cataphorically referenced than when subject pros are, indicating that
subject/object asymmetry coming from a pronoun instead of its antecedent may also influence
processing difficulty.
Possibly related to this line of interpretation, several ERP studies (e.g., Cowles, Kutas, &
Kluender, 2003; Van Berkum, Brown, Hagoort, & Zwitserlood, 2003) have reported anterior
negativity and/or P600 effects in the establishment of anaphoric links. Cowles et al., in
particular, reported a sustained anterior negativity when the parser is carrying referential
ambiguity, as well as a P600 effect when that referential gap was filled. Thus the present results
could be linked with this type of interpretation, and we hope to move the discussion forward by
exploring more types of filler-gap/gap-filler dependencies in our future research.
Conclusions
To conclude, our experiments have shown that Japanese ORs are harder to process than
SRs, revealing higher processing costs from the head noun to the sentence end for ORs than SRs
in both reading times and ERPs. Our data also suggest that gap-filler association in Japanese
ERPs to Japanese Relative Clauses 35
RCs involve a long-lasting integration process after the filler, as indexed by a continuous centroposterior positivity in ORs. Finally, both ERP and reading time data seem more consistent with
a structural distance account than a linear distance account for Japanese RCs.
ERPs to Japanese Relative Clauses 36
References
Amano, N., & Kondo, M. (2000). NTT database series nihongo-no goikokusei: Lexical
properties of Japanese (Vol. 7). Tokyo: Sanseido.
Caplan, D., Vijayan, S., Kuperberg, G., West, C., Waters, G., Greve, D., & Dale, A.M. (2001).
Vascular responses to syntactic processing: Event-related fMRI study of relative clauses.
Human Brain Mapping, 15, 26-38.
Casado, P., Martín-Loeches, M., Muñoz, F., & Fernández-Frías, C. (2005). Are semantic and
syntactic cues inducing the same processes in the identification of word order? Cognitive
Brain Research 24, 526-543.
Chomsky, N. (1981). Lectures on government and binding. Dordrecht, Netherlands: Foris
Publications.
Comrie, B., & Horie, K., (1995). Complement clauses versus relative clauses: Some Khmer
evidence. In W. Abraham, T. Givón, & S. A. Thompson (Eds.), Discourse grammar
typology (pp. 65-75). Amsterdam, Netherlands: John Benjamins Publishing.
Coulson, S., King, J. W., & Kutas, M. (1998). Expect the unexpected: Event-related brain
response to morphosyntactic violations. Language and Cognitive Processes, 13, 21-58.
Cowles, H. W., Kutas, M., & Kluender, R. (2003). Different ERP results from identical
sentences: The importance of prior context in sentence processing. Poster presented at
the 16th annual CUNY conference on human sentence processing, March 27-29,
Massachusetts Institute of Technology, Cambridge, MA.
Crawley, R. A., Stevenson, R. J., & Kleinman, D. (1990). The use of heuristic strategies in the
interpretation of pronouns. Journal of Psycholinguistics Research, 19, 245-264.
ERPs to Japanese Relative Clauses 37
Ferreira, F., & Clifton, C. (1986). The independence of syntactic processing. Journal of Memory
and Language 25, 348-368.
Fiebach, C. J., Schlesewsky, M., & Friederici, A. D. (2001). Syntactic working memory and the
establishment of filler-gap dependencies: Insights from ERPs and fMRI. Journal of
Psycholinguistic Research, 30, 321-338.
Fiebach, C. J., Schlesewsky, M., & Friederici, A. D. (2002). Separating syntactic memory costs
and syntactic integration costs during parsing: The processing of German WH-questions.
Journal of Memory and Language 47, 250-272.
Fodor, J.D. (1989). Empty categories in sentence processing. Language and Cognitive Processes,
4, 155-209.
Frederiksen, J. R. (1981). Understanding anaphora: Rules used by readers in assigning
pronominal referents. Discourse Processes, 4, 323-347.
Fukui, N., & Takano, Y. (2000). Nominal structure: An Extension of the symmetry principle. In
P. Svenonius (Ed.), The derivation of VO and OV (pp. 219-254). Amsterdam,
Netherlands: John Benjamins.
Gibson, E. (1998). Linguistic complexity: Locality of syntactic dependencies. Cognition, 68, 176.
Gibson, E. (2000). The dependency locality theory: A distance-based theory of linguistic
complexity. In A. Marantz, Y. Miyashita, & W. O’Neil (Eds.), Image, language, brain:
Papers from the first mind articulation project symposium (pp. 95-126). Cambridge, MA:
MIT Press.
ERPs to Japanese Relative Clauses 38
Gordon, P. C., Hendrick, R., & Johnson, M. (2001). Memory inference during language
processing. Journal of Experimental Psychology: Learning, Memory and Cognition, 27,
1411-1423.
Gratton, G., Coles, M. G. H., & Donchin, E. (1983). A new method for off-line removal of
ocular artifact. Electroencephalography and Clinical Neurophysiology, 55, 468-484.
Hagoort, P., Brown, C., & Groothusen, J. (1993). The syntactic positive shift (SPS) as an ERP
measure of syntactic processing. Language & Cognitive Processes, 8, 439-483.
Han, C., & Kim, J.B., (2004). Are there “double relative clauses” in Korean? Linguistic Inquiry
35, 315-337.
Hawkins, J. A., (1994). A performance theory of order and constituency. Cambridge, United
Kingdom: Cambridge University Press.
Huynh, H., and Feldt, L.S. (1976). Estimation of the Box correction for degrees of freedom from
sample data in randomized block and split-plot designs. Journal of Educational Statistics,
1, 69-82.
Ishizuka, T., Nakatani, K., & Gibson, E. (2003). Relative clause extraction complexity in
Japanese. Poster presented at the 16th annual CUNY conference on human sentence
processing, March 27-29, Massachusetts Institute of Technology, Cambridge, MA.
Ishizuka, T., Nakatani, K., & Gibson, E. (2006). Processing Japanese relative clauses in context.
Paper presented at the 19th annual CUNY conference on human sentence processing,
March 23-25, CUNY Graduate School & University Center, New York, NY.
Kaan, E., Harris, A., Gibson, E., & Holcomb, P. (2000). The P600 as an index of syntactic
integration difficulty. Language and Cognitive Processes, 15, 159-201.
ERPs to Japanese Relative Clauses 39
Kaan, E., & Swaab, T. Y. (2003). Electrophysiological evidence for serial sentence processing:
a comparison between non-preferred and ungrammatical continuations. Cognitive Brain
Research, 17, 621-635.
Kaplan, T. I., & Whiteman, J. B. (1995). The category of relative clauses in Japanese, with
reference to Korean. Journal of East Asian Linguistics, 4, 29-58.
King, J. W., & Just, M. A. (1991). Individual differences in syntactic processing: The role of
working memory. Journal of Memory and Language, 30, 580-602.
King, J. W., & Kutas, M. (1995). Who did what to when?: Using word- and clause-level ERPs to
monitor working memory usage in reading. Journal of Cognitive Neuroscience, 7, 376-395.
Kluender, R., & Kutas, M. (1993). Bridging the gap: Evidence from ERPs on the processing of
unbounded dependencies. Journal of Cognitive Neuroscience, 5, 196-214.
Kwon, N., Polinsky, M., & Kluender, R. (in press). Subject preference in Korean. In
Proceedings of the 25th West Coast Conference on Formal Linguistics. Somerville, MA:
Cascadilla Press.
Kwon, N., Polinsky, M., & Kluender, R. (in prep.). Processing of long-distance dependencies in
Korean.
MacWhinney, B. (1982). Basic syntactic processes. In S. Kuczaj (Ed.), Language development
(Vol. 1; pp. 73-136). Hillsdale, NJ: Lawrence Erlbaum.
Matsumoto, Y. (1989). Japanese-style noun modification ... in English. In K. Hall, M. Meacham,
& R. Shapiro (Eds.), Proceedings of the Fifteenth Annual Meeting of the Berkeley
Linguistics Society (pp. 226-237). Berkeley, CA: BLS.
Miyamoto, E. & Nakamura, M. (2003). Subject/object asymmetries in the processing of relative
clauses in Japanese. In G. Garding & M. Tsujimura (Eds.), Proceedings of the 22nd West
ERPs to Japanese Relative Clauses 40
Coast Conference on Formal Linguistics (pp. 342-355). Somerville, MA: Cascadilla
Press.
Müller, H., King, J. W., & Kutas, M. (1997). Event related potentials elicited by spoken relative
clauses. Cognitive Brain Research, 5, 193-203.
Murasugi, K. 2000. An antisymmetry analysis of Japanese relative clauses. In A. Alexiadou, P.
Law, A. Meinunger, & C. Wilder (Eds.), The syntax of relative clauses (pp. 231-263).
Amsterdam, Netherlands: John Benjamins Publishing.
Neville, H., Nicol, J., Barss, A., Forster, K., & Garrett, M. (1991). Syntactically based sentence
processing classes: Evidence from event-related brain potentials. Journal of Cognitive
Neuroscience, 3, 151-165.
O’Grady, W. (1997). Syntactic development. Chicago, IL: Chicago University Press.
Osterhout, L., & Holcomb, P. J. (1992). Event-related brain potentials elicited by syntactic
anomaly. Journal of Memory and Language, 31, 785-806.
Osterhout, L., Holcomb, P. J., & Swinney, D. A. (1994). Brain potentials elicited by garden-path
sentences: Evidence of the application of verb information during parsing. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 20(4), 786-803.
Osterhout, L., & Nicol, J. (1999). On the distinctiveness, independence, and time course of the
brain responses to syntactic and semantic anomalies. Language and Cognitive Processes,
14, 283-317.
Phillips, C., Kazanina, N., & Abada, S. (2005). ERP effects of the processing of syntactic longdistance dependencies. Cognitive Brain Research 22, 407-428.
Rösler, F., Pechmann, T., Streb, J., Röder, B., & Hennighausen, E. (1998). Parsing of sentences
in a language with varying word order: Word-by-word variations of processing demands
ERPs to Japanese Relative Clauses 41
are revealed by event-related brain potentials. Journal of Memory and Language, 38,
150-176.
Sauerland, U., & Gibson, E. (1998). How to predict the relative clause attachment preference.
Paper presented at the 11th annual CUNY conference on human sentence processing,
March 19-21, Rutgers University, New Brunswick, NJ.
Schneider, W., Eschman, A., & Zuccolotto, A. (2002). E-Prime reference guide. Pittsburgh, PA:
Psychology Software Tools.
Sheldon, A. (1974). The role of parallel function in the acquisition of relative clauses in English.
Journal of Verbal Learning and Verbal Behavior, 13, 272-81.
Traxler, M.J., Morris, R.K., & Seely, R. E. (2002). Processing subject and object relative clauses:
Evidence from eye movements. Journal of Memory and Language, 47, 69-90.
Ueno, M. & Kluender, R. (2003). On the processing of Japanese wh-questions: Relating
grammar and brain. In G. Garding and M. Tsujimura (Eds.), Proceedings of the 22nd West
Coast Conference on Formal Linguistics (pp. 491-504). Somerville, MA: Cascadilla
Press.
Ueno, M. & Polinsky, M. (submitted). Maximizing processing in an SOV language.
Van Berkum, J., Brown, C. M., Hagoort, P., & Zwitserlood, P. (2003). Event-related brain
potentials reflect discourse-referential ambiguity in spoken language comprehension.
Psychophysiology, 40, 235-248.
Van Petten, C., & Kutas, M. (1991). Influences of semantic and syntactic context in open- and
closed-class words. Memory & Cognition, 19, 95-112.
Warren, T., & Gibson, E. (2002). The influence of referential processing on sentence complexity.
Cognition, 85, 79-112.
ERPs to Japanese Relative Clauses 42
Appendix
Materials
Experimental Conditions:
a. Subject Relative (SR)
新任の/議員を/非難した/記者には/長年の/相棒が/いた。
b. Object Relative (OR)
新任の/議員が/非難した/記者には/長年の/相棒が/いた。
Only the SR versions are given below, since the only difference between them and the OR
versions is the case-marking particle on the first noun, as highlighted in the examples above.
1a.
2a.
3a.
4a.
5a.
6a.
7a.
8a.
9a.
10a.
11a.
12a.
13a.
14a.
15a.
16a.
17a.
18a.
19a.
20a.
21a.
22a.
23a.
24a.
25a.
26a.
27a.
28a.
29a.
30a.
31a.
32a.
新任の/議員を/非難した/記者には/長年の/相棒が/いた。
真面目な/父を/疎んじた/兄には/まともな/友達が/少なかった。
評判の/先輩を/妬んだ/選手には/練習熱心な/がんばり屋が/多かった。
人気者の/先生を/疎んじた/生徒たちには/根暗な/人が/多かった。
高名な/人間国宝を/批判した/人々には/派手な/中年女性が/多かった。
無口な/父を/支えた/母には/熱烈な/信奉者が/多かった。
頑固な/社長を/見慣れてた/部下には/優秀な/企業戦士が/多かった。
年寄りの/母を/褒め称えた/看護士には/臨機応変な/者が/多かった。
中年の/女性社員を/助けた/同僚には/誠実な/人物が/多かった。
頭脳明晰な /いとこを/うらやんだ/学生には/青白い/優等生が/多かった。
保守派の/教頭先生を/嫌がった/母親達には/高学歴の/インテリが/多かった。
几帳面な/書記を/ほめた/委員長には/高校生の/娘が/いた。
冷血漢の/スパイを/調査した/組織には/優秀な/手下が/多かった。
照れ屋な/寮母さんを/ほめた/女子学生には/同郷の/友人が/少なかった。
お天気屋の/社長を/追い出した/株主には/うさん臭い/仲間が/ついていた。
聡明な/国王を/騙した/貴族には/ずる賢い/悪党が/ついていた。
わがままな/ちびっこを/叩いた/保母さんには/適齢期の/長女が/いた。
タバコ屋の/未亡人を/なだめた/紳士には/たくさんの/知人が/いた。
賢明な/店長を/信頼した/コックには/たくさんの/知り合いが/いた。
排他的な/政治家を/排除した/運動家には/熱狂的な/支持者が/いた。
屈託ない/跡継ぎを/見限った/父親には/年若い/愛人が/いた。
取引先の/御曹司を/見初めた/令嬢には/双子の/妹が/いた。
近所の/囲碁仲間を/愛した/曾祖父には/長生きな/連れ合いが/いた。
古株の/会員を/排除した/委員長には/若い/情婦が/いた。
隣人の/亭主を/追い返した/舅には/意中の/女性が/いた。
アルバイトの/若者を/指導した/婦人には/手強い/敵が/いた。
陽気な/嫁を/うるさがった/曾祖母には/ベテランの/家政婦さんが/ついていた。
石頭の/上司を/嫌がった/刑事には/入院中の/相棒が/いた。
偉大な/指導者を/接待した/貴婦人には/年若い/恋人が/いた。
泣き上戸の/常連客を/呼び止めた/ママには/出っ歯の/用心棒が/ついていた。
汚い/浮浪者を/殴り倒した/不良には/はすっぱな/ガールフレンドが/いた。
お屋敷の/坊ちゃんを/敬愛した/婆やには/愛らしい/孫娘が/いた。
ERPs to Japanese Relative Clauses 43
33a.
34a.
35a.
36a.
37a.
38a.
39a.
40a.
41a.
42a.
43a.
44a.
45a.
46a.
47a.
48a.
49a.
50a.
51a.
52a.
53a.
54a.
55a.
56a.
57a.
58a.
59a.
60a.
61a.
62a.
63a.
64a.
65a.
66a.
67a.
68a.
69a.
70a.
71a.
72a.
73a.
74a.
75a.
76a.
77a.
78a.
79a.
80a.
太っ腹な/スポンサーを/探し出した/監督には/有能な/マネージャーが/ついていた。
せっかちな/連れ合いを/なだめた/細君には/しっかり者の/お手伝いさんが/ついていた。
皮肉な/教頭を/脅した/教員には/かんしゃく持ちの/奥さんが/いた。
若い/係長を/批判した/秘書には/酒好きな/女友達が/いた。
雅やかな/王子を/捕らえた/盗賊には/忠実な/手下が/いた。
売れっ子の/芸能人を/使った/カメラマンには/金持ちの/スポンサーが/ついていた。
小太りの/苛めっ子を/逃がした/わんぱく坊主には/のろまな/妹が/いた。
エレガントな/ウグイス嬢を/引き止めた/立候補者には/有力な/政治家が/ついていた。
新人の/役者を/賞賛した/演出家には/元女優の/愛妻が/いた。
のんべいの/父親を/叱った/娘には/病気がちな/母親が/いた。
夜勤明けの/医者を/呼び止めた/入院患者には/献身的な/女房が/ついていた。
アメリカ人の/選手を/選んだ/コーチには/ヨーロッパ人の/奥さんが/いた。
昔の/アイドルを/けなした/マネージャーには/長年の/悪友が/いた。
地元の/名士を/からかった/芸者には/大富豪の/旦那が/ついていた。
おませな/女の子を/追いかけ回した/母親には/小学校以来の/幼なじみが/いた。
男勝りな/姉を/どなり付けた/課長には/社内の/支持者が/少なかった．
野暮な/客を/無視した/バーテンダーには/素敵な/恋人が/いた。
しっかり者の/お手伝いを/手なずけた/祖母には/単身赴任中の/息子が/いた。
ずうずうしい/大家を/どなった/借主には/無責任な/親が/いた。
赤毛の/少女を/ナンパした/高校生には/幼なじみの/彼女が/いた。
気短な/板前を/非難した/板長には/有望な/後継者が/いた。
しぶとい/ボクサーを/倒した/ライバル選手には/忠実な/ファンが/多かった．
野暮ったい/田舎者を/からかった/老人には/頼りない/長男が/いた。
非情な/借金取りを/追いまわした/ライターには/貧乏な/親戚が/いた。
怒りっぽい/夫を/殴った/息子には/同い年の/彼女が/いた。
敬謙な/信者を/訪ねた/神父には/多数の/協力者が/いた。
喧嘩早い/男子を/挑発した/教師には/意地悪な/悪友が/いた。
男勝りな/女課長を/いびった/係長には/強気な/妻が/いた。
グラマーな/タレントを/おちょくった/司会者には/口うるさい/上役が/いた。
無口な/旦那を/追い払った/おかみさんには/うわさ好きな/おしゃべり仲間が /いた。
中堅どころの/小説家を/ひやかした/編集者には/美人の/秘書が/ついていた。
内気な/姉を/はげました/妹には/大好きな/アイドル歌手が/いた。
有名な/スター女優を/好んだ/監督には/青年層の/ファンが/多かった。
やんちゃな/男の子を/見つけた/保母には/大学生の/息子が/いた。
手強い/敵を/殺した/兵士には/老齢の/母親が/いた。
直属の/部下を/裏切った/上司には/親しい/仲間が/少なかった．
暴れん坊の/ならず者を/捕らえた/ちんぴらには/手下の/弟分が/いた。
せこい/小役人を/ののしった/市民には/幼い/娘が/いた。
野党の/議員を/支持した/官僚には/上流階級出身の/妻が/いた。
名家の/奥方を/追いやった/姑には/陰湿な/小姑が/ついていた。
勇敢な/名将を/切り倒した/兵士には/同郷の/戦友が/いた。
元気な/お袋を/追いかけた/配達人には/意地悪な/雇い主が/いた。
同室の/入院患者を/起こした/少女には/憧れの/医師が/いた。
古株の/ピアニストを/無視した/指揮者には/天才肌の/弟が/いた。
新人類の/部下を/避けた/次長には/受験生の/長男が/いた。
高徳の/僧を/招待した/武士には/盲目の/長男が/いた。
おませな/女の子を/ひやかした/腕白坊主には/厳格な/父親が/いた。
女たらしの/プレイボーイを/誘惑した/人妻には/無関心な/夫が/いた。
ERPs to Japanese Relative Clauses 44
Footnotes
1
More strictly speaking, Gibson (1998, 2000) concerns the number of new discourse
referents between a filler and its gap. When a noun between a filler and its gap is reduced to a
pronoun, processing difficulty is said to be reduced (cf. Gordon, Hendrick, & Johnson, 2001;
Warren & Gibson, 2002).
2
Also see Hawkins (1994) for an account based on the combination of linear and
structural distance.
3
Whether the head noun is considered to originate outside (e.g., Han & Kim, 2004) or
inside (e.g., Fukui & Takano, 2000) the RC is controversial, but none of our arguments or
conclusions depend on which of these is the better analysis.
4
More recently, Ishizuka, Nakatani, and Gibson (2006) reported the opposite effect when
Japanese SRs and ORs were presented in a discourse context. See footnote 15 in General
Discussion.
5
The sentences produced by participants in the norming study did not resemble the
stimulus sentences in general. Participants did not use the same agent or patient (or a reversed
agent or patient) that were used in the stimulus sentence with the corresponding verb, except in
six sentences out of the 690 sentences examined (0.9%).
6
We only included the frequency counts for nouns (without any case-marking particles)
whose exact orthography (i.e., the particular combination of Chinese characters (kanji) and
syllabic characters (kana)) was in the entry of Amano and Kondo (2000). This resulted in 70
data points for the nouns in the RC at sentence position 2, 68 data points for the head nouns at
sentence position 4, and 70 data points for the main clause nouns at sentence position 6.
ERPs to Japanese Relative Clauses 45
7
The total number of participants actually run was 51. However, due to a hardware
problem, data from the first 19 participants had to be discarded. Four other participants had
additional recording problems with too much noise or drift, and their data were also discarded.
In addition, another participant’s data were accidentally deleted. Six participants from the 19
participants affected by the hardware problem came back for a second session 52-139 days
(mean 113 days) after their original session. Exclusion of these six participants does not alter the
patterns in the data but weakens the statistical power.
8
The presentation rate of 650 ms per word is longer than the 500 ms per word used in
many English studies, but was deemed optimal after consulting five native speakers of Japanese.
Given both the visual complexity of Chinese characters often used in Japanese and the fact that
many of the Japanese bunsetsu translate as multiple English words, it is not surprising that
readers needed more time per bunsetsu.
9
To what extent all of these processes involve working-memory load is controversial (cf.
Kluender & Kutas, 1993; Coulson et al., 1998).
10
There are two frontal electrodes that showed a significant difference between SRs and
ORs, namely, Fz and AF4. However, recall that there was anterior negativity in response to ORs
at the relative clause verb (in addition to the head noun) in the relative clause region. This
negativity can push ORs down further than actual when waveforms are re-baselined as discussed
in the present section. This could be obscuring the comparison for these electrodes.
11
Perhaps except for the few frontal electrodes discussed in footnote 10 above.
12
Such a difference might be predicted to take either the form of an anterior negativity or
a posterior positivity or both. In German, where case-marking particles precede nouns, Rösler et
al. (1998) found anterior negativities in response to case-marked determiners that indicated that
ERPs to Japanese Relative Clauses 46
phrases were not in canonical order, but posterior positivities to the nouns that followed those
determiners. Since in Japanese the case-marking particle was presented together with its noun,
either effect or a combination of both might have been expected.
13
The reason why mean accuracy for the ERP experiment was 2-3% lower overall than
that for the reading time experiment may be because participants could not control the
presentation rate of the sentences in the ERP study. They may have needed more time to read
some parts of the sentences.
14
If we classify both English wh-questions and relative clauses together as "filler-gap
dependencies", we can easily link both our anterior negativity and centro-posterior positivity
effects to the processes of filler-gap/gap-filler associations in both English and Japanese.
However, if we focus only on the processing of relative clauses in English, neither of the two
ERP studies on English SRs vs ORs that we know of reported any posterior positivity effects in
response to ORs. However, this may be at least partly because these earlier studies paid little
attention to P600 effects. Single word averages of King and Kutas (1995: 383) seem to show a
small posterior positivity effect at the RC verb (the pre-gap position similar to Kaan at al.
(2000)) of ORs. King and Kutas only examined the 300-500 ms time window and reported a RC
type x electrode interaction at this position (p. 385), but the interaction could in part have been
due to a posterior positivity effect, in addition to the LAN they reported. Müller et al. (1997)
conducted an auditory version of King and Kutas, but did not seem to reveal any posterior
positivity effect around the gap of ORs. This may have something to do with the auditory
presentation of their stimuli.
15
Recently, Ishizuka et al. (2006) investigated the processing of Japanese SRs and ORs
in a discourse context. They compared SRs and ORs in the following discourse context and
ERPs to Japanese Relative Clauses 47
reported that SRs took longer to read than ORs at the RC verb (‘interviewed’ in the example),
while there was no difference at the head noun and later positions.
ある番組でレポーターが作家をインタビューした。/
一方その作家は次の作品の材料に別のレポーターをインタビューした。/
A reporter interviewed a writer on a TV program./
Then the writer interviewed another reporter for his new novel./
Taro: /
選挙に/立候補したのは/どちらの/レポーター
‘Which reporter stands as a candidate for the election?’
Hanako: /
SR: 作家を/インタビューした/レポーター/だった/らしいよ
writer-ACC / interviewed / reporter / was / it seems
‘It seems to be the reporter who interviewed the writer’
OR:
作家が/インタビューした/レポーター/だった/らしいよ
writer-NOM / interviewed / reporter / was / it seems
‘It seems to be the reporter who the writer interviewed’
Ishizuka et al.’s (2006) data can be interpreted according to the alternative analysis of
Japanese RCs discussed in this section. It is possible to argue that the gap was filled by (or the
pro was coreferenced with) the corresponding referent in the discourse before the head noun
position. This may be why there was no reading time difference at the head noun position in
their data, unlike other reading time studies.
Ishizuka et al. (2006) argue that their data support the linear distance account over the
structural distance account. They argue that the longer reading times for ORs in previous studies
were due to greater temporary ambiguity of ORs, which are more likely to be interpreted as
mono-clausal with sentence-initial nominative-marked nouns than are SRs with sentence-initial
accusative-marked nouns. They argue further that SRs actually take longer to read than ORs
when participants are led to expect a RC construction by the discourse context. However, the
prevalence of subject-drop in the sentences produced in our norming study suggests that even
ERPs to Japanese Relative Clauses 48
when there is no discourse context, participants could imagine a mono-clausal sentence with a
missing subject upon seeing an accusative-marked noun in SRs. Thus ORs do not necessarily
have to be more temporarily ambiguous than SRs. Then the longer reading times for SRs at the
RC verb may have been either because the accusative-marked noun caused a slowdown one
word later, just as we observed in Experiment 1, or because the part of the discourse related to
the answer to the SR question than the OR question was mentioned earlier (i.e., the event of a
reporter interviewing a writer was mentioned before the event of the writer interviewing another
reporter in the given discourse), and thus may have taken longer to access.
ERPs to Japanese Relative Clauses 49
Table 1
Stimuli
_____________________________________________________________________________
a. Subject Relatives (SRs)
[新任の
議員を
非難した]
記者には
長年の
相棒が
いた。
[shinninno giin-o
hinanshita] kisha-ni-wa
naganenno aibou-ga
ita.
[new
senator-ACC attacked]
reporter-DAT-TOP long-term colleague-NOM existed
‘(For the reporter [(who) attacked the new senator], a long-term colleague existed →)
The reporter [(who) attacked the new senator] had a long-term colleague.’
b. Object Relatives (ORs)
[新任の
議員が
非難した]
記者には
長年の
相棒が
いた。
[shinninno giin-o
hinanshita] kisha-ni-wa
naganenno aibou-ga
ita.
[new
senator- NOM attacked] reporter-DAT-TOP long-term colleague-NOM existed
‘The reporter [(who) the new senator attacked] had a long-term colleague.’
_____________________________________________________________________________
ERPs to Japanese Relative Clauses 50
Figure Captions
Figure 1. Syntactic structures for subject relatives (SRs) and object relatives (ORs) in English.
Figure 2. Syntactic structures for subject relatives (SRs) and object relatives (ORs) in Japanese.
Figure 3. Mean trimmed reading times for each sentence position for subject relatives (SRs) and
object relatives (ORs).
Figure 4. ERPs from frontal (AF3/4, F3/4, FC3/4, Fz) electrodes at the relative clause verb and
head noun positions of SRs vs ORs. Negativity is plotted up.
Figure 5. ERPs from all electrodes at the post-relative clause region (from the head noun to the
sentence-end) of SRs vs ORs. Negativity is plotted up.
ERPs to Japanese Relative Clauses 51
Figure 1
SR
OR
NP
NP
3
3
NP
S’
NP
S’
the reporteri 3
the reporteri 3
NP
S
NP
S
3
whoi
3
whoi
NP
VP
NP
VP
the senator 3
3
___i
V
NP
V
NP
attacked
___i
attacked
the senator
ERPs to Japanese Relative Clauses 52
Figure 2
SR
OR
NP
3
S’
NP
3 reporteri
記者
NP
S
Opi 3
NP
VP
___i 3
NP
V
Senator-A
attacked
議員を
非難した
NP
3
S’
NP
3 reporteri
NP
S 記者
Opi 3
NP
VP
senator-N 3
議員が NP
V
attacked
___i
非難した
ERPs to Japanese Relative Clauses 53
Figure 3
Reading Times (msec)
900
Head Noun
800
700
600
RC Verb
SR
500
OR
400
300
w1
w2
w3
w4
w5
[新任の 議員を/が 非難した] 記者には 長年の
w6
w7
相棒が
いた。
[new senator-A/N attacked] reporter-D-T long-term colleague existed.
ERPs to Japanese Relative Clauses 54
Figure 4
AF3
AF4
µV
-10
F3
F4
Fz
650
1300ms
0
10
FC3
FC4
Subject Relatives (SRs)
[新任の 議員を
非難した]
記者には
new
attacked
reporter-DAT-TOP long-term colleague-NOM existed
senator-ACC
長年の
相棒が
いた。
Object Relatives (ORs)
[新任の 議員が
new
非難した]
senator-NOM attacked
記者には
長年の
相棒が
いた。
reporter-DAT-TOP long-term colleague-NOM existed
ERPs to Japanese Relative Clauses 55
Figure 5
AF4
AF3
F7
F3
FT7
FC3
T3
C3
T5
CP3
Fz
Cz
F4
F8
FC4
FT8
C4
T4
CP4
T6
µV
-10
P3
P4
Pz
-5
0
5
10
PO7
0
1300
2600
ms
PO8
Subject Relatives (SRs)
[新任の 議員を
非難した] 記者には
new
attacked reporter-DAT-TOP long-term colleague-NOM
senator-ACC
長年の
相棒が
いた。
existed
Object Relatives (ORs)
[新任の 議員が
new
非難した] 記者には
長年の
相棒が
senator-NOM attacked reporter-DAT-TOP long-term colleague-NOM
いた。
existed