Electrophysiological correlates of morphosyntactic integration in German phrasal context

Abstract

The morphosyntactic paradigm of an inflected word can influence isolated word recognition, but its role in multiple-word phrasal integration is less clear. We examined the electrophysiological response to adjectives in short German prepositional phrases to evaluate whether strong and weak forms of the adjective show a differential response, and whether paradigm variables are related to this response. Twenty native German speakers classified serially presented phrases as grammatically correct or not while the electroencephalogram (EEG) was recorded. A functional mixed effects model of the response to grammatically correct trials revealed a differential response to strong and weak forms of the adjectives. This response difference depended on whether the preceding preposition imposed accusative or dative case. The lexically conditioned information content of the adjectives modulated a later interval of the response. The results indicate that grammatical context modulates the response to morphosyntactic information content, and lends support to the role of paradigm structure in integrative phrasal processing.

Full text

This article was downloaded by: [Max Planck Institut Fur Psycholinguistik]
On: 07 February 2012, At: 06:43
Publisher: Psychology Press
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Language and Cognitive Processes
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/plcp20
Electrophysiological correlates of
morphosyntactic integration in
German phrasal context
D. J. Davidson a , A. Hanulíková a b & P. Indefrey c d
a Basque Centre on Cognition, Brain and Language, Donostia,
Spain
b Max Planck Institute for Psycholinguistics, Nijmegen, The
Netherlands
c Abt. für Allgemeine Sprachwissenschaft, Institut für Sprache
und Information, Heinrich-Heine-Universität, Düsseldorf,
Germany
d Donders Centre for Cognitive Neuroimaging, Donders Institute
for Brain, Cognition, and Behavior, Nijmegen, The Netherlands
Available online: 03 Feb 2012
To cite this article: D. J. Davidson, A. Hanulíková & P. Indefrey (2012): Electrophysiological
correlates of morphosyntactic integration in German phrasal context, Language and Cognitive
Processes, 27:2, 288-311
To link to this article: http://dx.doi.org/10.1080/01690965.2011.616448
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-
conditions
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any
representation that the contents will be complete or accurate or up to date. The
accuracy of any instructions, formulae, and drug doses should be independently
verified with primary sources. The publisher shall not be liable for any loss, actions,
claims, proceedings, demand, or costs or damages whatsoever or howsoever caused

arising directly or indirectly in connection with or arising out of the use of this
material.
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

Electrophysiological correlates of morphosyntactic
integration in German phrasal context
D. J. Davidson
1
, A. Hanulı
´kova
´
1,2
, and P. Indefrey
3,4
1
Basque Centre on Cognition, Brain and Language, Donostia, Spain
2
Max Planck Institute for Psycholinguistics, Nijmegen, The Netherlands
3
Abt. fu
¨r Allgemeine Sprachwissenschaft, Institut fu
¨r Sprache und Information,
Heinrich-Heine-Universita
¨t, Du
¨sseldorf, Germany
4
Donders Centre for Cognitive Neuroimaging, Donders Institute for Brain,
Cognition, and Behavior, Nijmegen, The Netherlands
The morphosyntactic paradigm of an inflected word can influence isolated word
recognition, but its role in multiple-word phrasal integration is less clear. We examined
the electrophysiological response to adjectives in short German prepositional phrases to
evaluate whether strong and weak forms of the adjective show a differential response, and
whether paradigm variables are related to this response. Twenty native German speakers
classified serially presented phrases as grammatically correct or not while the electro-
encephalogram (EEG) was recorded. A functional mixed effects model of the response to
grammatically correct trials revealed a differential response to strong and weak forms of
the adjectives. This response difference depended on whether the preceding preposition
imposed accusative or dative case. The lexically conditioned information content of the
adjectives modulated a later interval of the response. The results indicate that
grammatical context modulates the response to morphosyntactic information content,
and lends support to the role of paradigm structure in integrative phrasal processing.
Keywords: Paradigm; Syntactic integration; Parsing; Information theory; Inflectional
morphology.
INTRODUCTION
Recent approaches to the recognition of inflected words have employed a statistical
framework based on information theory and word-and-paradigm morphology (Milin,
Kuperman, Kostic´, & Baayen, 2008). In this approach, the speed of a classification
Correspondence should be addressed to D. J. Davidson, Basque Centre on Cognition, Brain and
Language, Mikeletegi Pasealekua 69, Solairua 220009 Donostia Spain. E-mail: d.davidson@bcbl.eu
This work was supported by the Dutch and German science foundations Nederlandse Organisatie voor
Wetenschappelijk Onderzoek (NWO) and the Max-Planck-Gesellschaft (MPG), respectively. Daniel von
Rhein recruited subjects, helped to construct materials, and assisted in data collection. Christiane Fritz
constructed the set of German materials, and assisted in data collection. Victor Kuperman provided
valuable feedback on an earlier version of this article. Portions of this work were presented at the Third
International Conference of Cognitive Science 2008, Moscow; as well as the 6th International Morphology
Processing Conference (MOPROC) 2009 in Turku.
LANGUAGE AND COGNITIVE PROCESSES
2012, 27 (2), 288311
#2012 Psychology Press, an imprint of the Taylor & Francis Group, an Informa business
http://www.psypress.com/lcp http://dx.doi.org/10.1080/01690965.2011.616448
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

response to a word in a lexical decision task is inversely related to the relative amount
of information provided by that word, estimated from a probability model. This
probability is taken from a frequency estimate of the word relative to a corpus sample
of a given size, or by more complex measures of the probability distribution associated
with a word. The experiments based on this approach have shown that paradigm-
specific information content influences the response to words that have been presented
in isolation. However, much less is known about how paradigm-related information
content influences morphosyntactic integration in phrasal (syntagmatic) context. This
issue is relevant for paradigm-based models because the choice of morphological
suffix in an inflected word is a reflection of the grammatical context in which it occurs
(see section below entitled ‘‘Information Content of Morphosyntactic Paradigms’’). If
the paradigm-specific information content is used during phrasal processing, this
would add further support for the proposal that paradigmatic organisation is relevant
for comprehension. More generally, there has been increased interest in the application
of information theory to sentence processing (Levy, 2008), but the specific interaction
of the paradigm distribution and grammatical integration has yet to be addressed.
Here, we use electrophysiology to examine how the distribution of an inflected form
within a paradigm affects the response to the strong and weak adjectival declension in
German. In the remainder of the introduction, we first review the existing literature on
variables related to lexical paradigms. Then we describe how electroencephalography
(EEG) has been used to examine morphosyntactic recognition in phrasal contexts, and
finally we summarise and propose several hypotheses.
INFORMATION CONTENT OF MORPHOSYNTACTIC PARADIGMS
Several studies have now shown that morphological paradigms can influence isolated
word recognition in visual lexical decision tasks. Kostic´(1991, 1995) showed that a
statistical measure that incorporated the frequency of an inflected form within a
paradigm, as well as the number of grammatical functions or meanings of a word, was
highly positively correlated with lexical decision times during the recognition of
Serbian nouns. This measure was based on information theory, indicating the relative
amount of information that an inflection suffix provides, relative to its paradigm.
Moscoso del Prado Martı´n, Kostic´, and Baayen (2004) found that lexical decision
times for Dutch nouns were positively correlated with inflectional entropy. Inflectional
entropy increases in a paradigm when there are more inflectional variants possible,
and/or when the variants have similar probabilities. Baayen, McQueen, Dijkstra, and
Schreuder (2003) showed that the relative frequency of inflected forms of Dutch nouns
influenced visual and auditory lexical decision times. Milin et al. (2008) reviewed more
recent evidence supporting the role of information content and entropy in lexical
decision. Also note that the existing literature suggests important differences between
grammatical categories. Kostic´and Katz (1987) compared lexical decision times for
isolated Serbian nouns, adjectives, and verbs and found that, unlike nouns, responses
to adjectives were not fastest to most common default nominative case forms, and
secondly, that surface frequency predicted response times. The results of these studies
indicate that paradigm distribution can influence isolated word recognition, but it is
not clear that the effects are the same across grammatical categories.
Less is known about whether there is an influence of morphological paradigms in
phrasal constructions, although there is good evidence that phrasal context has a
significant impact on morphosyntactic processing. For example, an extensive series of
MORPHOSYNTACTIC INTEGRATION 289
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

experiments with Serbian has shown that inflected words are recognised more quickly
(as revealed by lexical decision times) in correct grammatical contexts as compared to
grammatical-violation contexts (Gurjanov, Lukatela, Lukatela, Savic´, & Turvey, 1985;
Gurjanov, Lukatela, Moskovljevic´, Savic´, & Turvey, 1985; Lukatela, Kostic´, Feldman,
& Turvey, 1983; Lukatela, Kostic´, Todorovic´, Carello, & Turvey, 1987). Also, related to
the present study, Bo
¨lte and Connine (2004) have shown that correct inflection
facilitates the recognition of German nouns when the noun is preceded by a determiner,
also using a lexical decision task. However, recently Hyo
¨na
¨, Vainio, and Laine (2002)
have shown that lexical factors which operate in isolated lexical decision tasks do not
necessarily have the same effect in sentence context (in Finnish), and Vainio, Hyo
¨na
¨,
and Pajunen (2008) have shown relatively delayed effects of grammatical agreement
during sentence reading, in that the facilitatory effects of agreement emerged in total
reading time measures rather than first-pass fixation measures. Thus, it is an open
question whether the lexical factors identified in earlier work using isolated-word
lexical decision also operate during phrasal processing, and if so, when.
There is some evidence that paradigm organisation can determine word-level
processing in German, but the question remains whether it affects phrase-level processing.
Clahsen, Eisenbeiss, Hadler, and Sonnenstuhl (2001) have shown that paradigm specificity
influences repetition priming patterns in German adjectives and verbs. However, it is not
yet clear how repetition priming maps onto phrasal composition, or whether similar
specificity results would be obtained in phrasal contexts (also see Barber & Carreiras,
2005). Penke, Janssen, and Eisenbeiss (2004) found evidence consistent with an under-
specified morphological paradigm for German adjectival paradigms in a sentence
matching task, but they did not investigate the influence of the item-specific probability
distribution for the paradigm. Unlike the behavioral results for lexical decision, the tasks
that have been used for sentence contexts so far do not allow for an estimate of when
(within the sentence) the effect of paradigm distribution would be expected to operate.
PROBABILITY MODEL FOR THE GERMAN STRONG AND WEAK
ADJECTIVAL PARADIGMS
In the experiment reported below, the influence of the paradigm probability distribution
is investigated in German adjectives using a probability model derived from Milin et al.
(2008). German determiners and adjectives show agreement with their head nouns for
the features of case, gender, and number. The degree to which these features are
expressed by an adjective suffix differs between three classes termed the strong, weak,
and mixed declension. An adjective may take suffixes of all three classes, depending on
the preceding elements of the noun phrase (NP). This dependency is considered to be a
syntactic dependency rather than a semantic or phonological dependency (Zwicky,
1986) and, following Schlenker (1999), it can be described as a rule according to which
syntactic features are to be expressed onlyon the first inflectable element of an NP. If the
adjective is the first inflectable element, it takes on a strong suffix. The suffix em in [1],
for example, specifies dative case and nonfeminine gender. Please note that the examples
are indexed with square brackets and equations are indexed with parentheses. By
contrast, if the adjective is preceded by a definite determiner that expresses the feature
information as in [2], the adjective has a weak suffix en that is compatible with the
feature specification of the determiner but does not express the features itself.
[1] mit kleinem Boot STRONG
‘‘with a small boat’’
290 DAVIDSON, HANULI
´KOVA
´, INDEFREY
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

[2] mit dem kleinen Boot WEAK
‘‘with the small boat’’
As a consequence of this, a greater variety of suffixes (e, en, em, es, er) are
used with the strong declension than the weak (e, en), as shown in Table 1.
The five German adjective suffixes show a high degree of syncretism expressing 72
possible combinations of case, gender, number, and declension class. Linguistic analyses
have dealt with the syncretism by using a hierarchical feature specification, such that
some suffixes are specified for case, number, and gender features and others are treated
as default or ‘‘elsewhere’’ forms with reduced or no specification (Bierwisch, 1967;
Blevins, 1995; Cahill & Gazdar, 1997; Clahsen, Eisenbeiss, Hadler, & Sonnenstuhl,
2001; Penke et al., 2004; Schlenker, 1999; Wunderlich, 1997; Zwicky, 1986). Commonly
assumed features in analyses of German determiner and adjective morphology are
9OBLIQUE and 9GOVERNED for case (nominative
OBL GOV
, accusative
OBL
GOV
,dative
OBL GOV
, and genitive
OBL GOV
), 9MASCULINE and
9FEMININE for gender (masculine
MF
, neuter
MF
, and feminine
MF
), and
9PLURAL for number (singular
PL
, plural
PL
). All analyses assume to some degree
homophonic suffixes and in most cases the suffixes are analysed separately for strong
and weak declension with the five suffixes of the strong declension carrying more feature
specifications than the two suffixes of the weak declension. Following Schlenker’s (1999)
general idea of a left-to-right accumulation of features in the processing of a German
noun phrase, the adjective suffix em in example [1] specifies the features OBL,
GOV, F, and PL. In example [2] the same features are specified by the determiner
dem and the adjective suffix en is a default suffix adding no feature information to the
NP. Note that due to the syncretism of the strong suffix em, gender remains
underspecified as nonfeminine in the dative case. In accusative case NPs such as [3],
which is governed by the preposition ohne, the strong suffix es underspecifies case but
fully specifies neuter gender with the features OBL, M, F, and PL. In the
corresponding definite NP [4] the determiner again specifies the same features. Even
though in this case the weak adjective suffix especifies number as singular, from a
processing view this is redundant and adds no further feature information to the
NP. Note that this is different for NPs with the definite determiner die indicating
feminine singular or plural of all genders. To keep the weak conditions homogeneous, we
therefore only used masculine and neuter nouns in the present experiment.
[3] ohne kleines Boot STRONG
‘‘without a small boat’’
[4] ohne das kleine Boot WEAK
‘‘without the small boat’’
The pattern of weak and strong endings can be quantified by the respective entropy
values of the strong and the weak paradigms. To illustrate this, Table 2 shows the
TABLE 1
Strong and weak adjective declension in German according to case (nominative, accusative,
dative, genitive), gender (masculine, feminine, neuter), and number (singular, plural)
Masc-sg Neu-sg Fem-sg -pl
Nom r/e s/e e/e e/n
Acc n/ns/ee/e e/n
Dat m/nm/nr/n n/n
Gen n/n n/n r/n r/n
Note: Suffixes used in the current experiment shown in italics.
MORPHOSYNTACTIC INTEGRATION 291
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

relative frequency F, estimated proportion Prp, and estimated information Iwxfor the
stemsuffix combinations of an example adjective stem wklein-, as well as for both
the strong and weak adjective suffixes as such. The frequencies and proportions were
obtained from a tagged corpus sample of German text (the Digitales Wo
¨rterbuch der
deutschen Sprache-Kerncorpus, 2008; DWDS core corpus). The DWDS corpus
consists of a text corpus of over 100 million words spanning the time period 1900
2000. The corpus contains text from five different genres (approximate percentage in
parenthesis, see Geyken, 2007): journalism (27%), literary text (26%), science literature
(22%), and other nonfiction (20%). The estimates were obtained by using a regular
expression search over sequences of grammatical category tags corresponding to
‘‘Prep Adj Noun’’ for the strong or ‘‘Prep Det Adj Noun’’ for the weak. Please see
www.dwds.de for more information about the composition of the corpus.The amount
of information conveyed by the suffix x
Ix¼log2PrpðxÞ(1)
is calculated from the probability of the suffix Pr
p
(x), estimated from the frequency of
the suffixes F(x) relative to the frequency of the suffixes in the paradigms p:
Prp¼FðxÞ
PxFðxÞ(2)
The distribution in Table 2 shows that the amount of information is more evenly
distributed over the strong paradigm (the probabilities are more similar to each other)
compared to the weak. In addition, the two suffixes that are used only in the strong
paradigm, em (F(x)80,398) and es (F(x)9,272), are less common than the
TABLE 2
Probability distribution for the strong and weak inflected forms of the adjective
klein-
(
F
(
w
x
),
Pr
p
(
w
x
), Iwx) in the prepositional phrase
Prep
(
Det
)
Adj Noun
, as well as the distribution of
suffixes for the same phrases over all adjectives (
F
(
x
),
Pr
p
(
x
),
I
x
)
Class w
x
F(w
x
)Pr
p
(w
x
)IwxF(x)Pr
p
(x)I
x
Strong klein-en 2,216 0.6149 0.7016 184,273 0.3523 1.50495
klein-e821 0.2278 2.1341 100,691 0.1925 2.37686
klein-em 265 0.0735 3.7655 80,398 0.1537 2.70156
klein-er 262 0.0727 3.7820 148,360 0.2837 1.81769
klein-es 40 0.0111 6.4935 9,272 0.0177 5.81777
Weak klein-en 2,888 0.8273 0.2736 371,216 0.7864 0.34661
klein-e600 0.1719 2.5406 97,548 0.2067 2.27468
klein-em* 3 0.0009 10.1845 93 0.0002 12.30935
klein-er* 0 0.0000 3,066 0.0065 7.26636
klein-es* 0 0.0000 103 0.0002 12.16201
Note: The first three numerical columns refer to the properties of the specific item klein-, while the last
three numerical columns refer to the properties of the suffixes over all adjectives. Estimated frequencies taken
from the tagged DWDS-Kerncorpus (www.dwds.de) comprising data over the years 19002000.
*Observed in the corpus for either the klein- or other stems, but normatively incorrect.
292 DAVIDSON, HANULI
´KOVA
´, INDEFREY
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

others. The entropy of the strong declension, 2.0218, is therefore higher than the weak,
0.7949, calculated with (3) over the suffixes in Table 2.
H¼XxPrpðwxÞlog2PrpðwxÞ(3)
The entropy indices capture the intuition expressed by the descriptive terms
‘‘strong’’ and ‘‘weak’’ because the weak form is thought to convey less information,
because it has already been expressed on the definite determiner in the phrase. Here,
we wish to extend the previous work by Kostic´and colleagues using lexical decision
tasks to phrasal context. If language comprehenders are sensitive to the amount of
information conveyed by a term in phrasal context, then there should be a larger
response time and error cost associated with recognising the strong form relative to the
weak.
In addition to the global probability model, lexeme-specific measures of informa-
tion can be derived. The amount of information conveyed by a suffix combined with a
particular stem
Ixwj¼log2PrpðwxÞþlog2PrpðwÞ(4)
is calculated from the probability of the inflected form Pr
p
(w
x
) and Pr
p
(w). The first
probability can be estimated from the frequency of the inflected form wxrelative to the
sum of the frequencies of all forms of the stem’s paradigm,
PrpðwxÞ¼ FðwxÞ
PxFðwxÞ(5)
while the second, Pr
p
(w), is estimated from the frequency of the adjectival stem win
the corpus (this is also termed lemma frequency by some authors).
PrpðwÞ¼FðwÞ
N(6)
A related quantity is the relative amount of information conveyed by a particular
stemsuffix combination relative to the number of word forms Nin the corpus
Iwx¼log2PrpðwxÞ(7)
which is derived from PrNðwxÞ(also sometimes termed the form frequency).
PrNðwxÞ¼FðwxÞ
N(8)
The difference between I
xjw
and Iwxis that I
xjw
models the amount of information
relative to the paradigm, while Iwxmodels information relative to the entire corpus.
Finally, we can calculate the relative entropy (KullbackLieber divergence, D)for
each adjective as in (9), as an estimate for how much the distribution of suffixes in the
strong and weak paradigms of a particular stem deviates from the distribution of
suffixes in the strong and weak declension paradigms in general. Here, Pis the
probability distribution of the paradigm for the stem, and Qis the distribution of the
MORPHOSYNTACTIC INTEGRATION 293
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

declension class (here strong and weak). The value of Dfor klein- for the strong
declension would be 0.3207, and for the weak, 0.0145, using the values in Table 2
(replacing values of 0 in the table with 1), indicating greater divergence of the suffixes
of the adjective klein- from the strong general paradigm than the weak general
paradigm, for which Dis close to 0 (no divergence).
DPQ
k
ðÞ¼
XxPrpðwxÞlog2
PrNðwxÞ
PrNðxÞ(9)
ELECTROPHYSIOLOGICAL ACTIVITY RELATED TO MORPHOSYNTAX
It is possible that electrophysiological brain responses recorded with EEG may also be
related to the variation in the different probability measures in a paradigm. This could
provide additional information about when these variables have an influence on
contextualised recognition, because EEG reflects some of the brain activity involved in
recognition as it unfolds over time. However, it is not clear from the previous
behavioral work using either single word lexical decision or sentence processing, which
cortical systems are responsible for processing morphosyntax, or what the expected
timing of the evoked response should be. It can be assumed based on previous lexical
decision tasks that the morphological information is available in under 1 s, as most
average behavioral responses (response times, fixation durations) in these tasks are
well within this interval. However, because these behavioral measures capture only a
limited number of data points related to activity per trial, the full profile of activity is
not known. Also, behavioral measurements are limited to motor responses, which
reflect the timing of motor system activity, and not directly the system of interest. In
contrast, the EEG yields a more densely sampled time series, which reflects (some
fraction of) the activity that is directly linked to the system of interest.
Electrophysiological studies employing contrasts between grammatical violations
and controls (Barber & Carreiras, 2005; Gunter, Friederici, & Schriefers, 2000; Mu
¨nte
& Heinze, 1994) have shown several event-related potential (ERP) components
responsive to morphosyntactic agreement. These include the left anterior negative
component (LAN), as well as a late positive (P600) component. If, for instance, the
syntactic agreement of gender between a noun and its preceding determiner is violated
(den
acc,masc,sing
*Land
neut,sing
), the LAN can be observed as early as 0.3safter the onset
of the word that creates a mismatch with the preceding context, and usually shows a
left frontal maximum (but see Hagoort, Wassenaar, & Brown, 2003; Mu
¨nte, Matzke, &
Johannes, 1997). Early negativities have been observed to morphosyntactic violations
in German (e.g., Gunter et al., 2000; Koester, Gunter, Wagner, & Friederici, 2004),
Spanish (Barber & Carreiras, 2005), English (Mu
¨nte, Heinze, & Mangun, 1993), and
Hebrew (Deutsch & Bentin, 2001). In addition to a LAN, these studies report the P600
starting about 0.5 safter the onset of the violation with a maximum at posterior
electrode locations. These two ERP components have been reported for violations that
occur in both isolated phrases and sentences. The P600 is sometimes described as a
‘‘late’’ component, in that it follows the early (N100 or P200) or mid-latency (N400 or
LAN) components, but it is understood to be elicited by the same critical word. There
are studies suggesting that the late positivity can be found for processing syntactically
incorrect structures (Hagoort, Brown, & Groothusen, 1993) as well as infrequent
structures (Osterhout & Holcomb, 1992). The anterior negativities are often observed
294 DAVIDSON, HANULI
´KOVA
´, INDEFREY
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

for incorrect structures (Friederici, Hahne, & Mecklinger, 1996; Mu
¨nte et al. 1993,
1997; Osterhout & Mobley, 1995) but have been observed for noncanonical structures
as well (e.g., Ro
¨sler, Pechmann, Streb, Ro
¨der, & Hennighausen, 1998).
EEG findings concerning the violation effects have motivated a cognitive model of
processing for both noun phrase and verbal agreement that involves several stages of
processing (Barber & Carreiras, 2005; Faussart, Jakubowitz, & Costes, 1999;
Molinaro, Vespignani, & Job, 2008). In this view, there is an early stage of lexical
access during which a word is selected based on the perceptual input. This is followed
by a stage of lexical recognition, during which morphosyntactic or semantic features
are accessed. Finally, depending on the previous phrasal context, context integration
and possibly re-analysis occurs (following a violation). In the previous literature, the
ERP violation effects described above have been used to determine whether some
features such as person, gender, or number are accessed before others, or whether the
violations of different features are re-analysed with different time courses. For the
most part, the relative insensitivity of ERP components to different types of
agreement violations has been taken as evidence against feature hierarchies during
lexical recognition or integration (Barber & Carreiras, 2005; Silva-Pereyra &
Carreiras, 2007), an issue which the present study does not address. However, the
influence of different types of features on relatively late components (e.g., P600 or
later, see also Vainio et al., 2008) has been taken as evidence for the stages of
processing in this model (Molinaro et al., 2008). Interestingly, the late influence of
features is taken to be a by-product of the differential access to the stem and suffix
during re-analysis. This was taken to be consistent with the hypothesis proposed by
Barber and Carreiras (2005), elaborated by Molinaro et al. (2008), that the timing of
P600 responses with different latencies might reflect different stages of a re-analysis
process. However, to date the probability distribution of the combined stem and suffix
during grammatical phrasal processing has not figured prominently in this research.
The present study can be seen as providing more information about the influence of
these statistical factors on the lexical recognition and integration phases of the parsing
of grammatical phrases during agreement processing.
Another line of comprehension research has examined so-called attraction effects
in number violation sentences, mainly in English. Similar to the well-known attraction
effect seen during production (Bock & Miller, 1991), subjects show an attenuated
reading time disruption for violations of subjectverb number agreement when there is
an intervening noun that agrees with the verb (Pearlmutter, Garnsey, & Bock, 1999),
and this is also reflected in a reduced P600 amplitude (Kaan, 2002). Theoretical
explanations of this data pattern have emphasised either mechanisms of feature
transmission during the construction of the subject noun phrase, or alternatively,
errors in retrieving the correct subject representation when processing the verb (see
Wagers, Lau, & Phillips, 2009 for a recent review). Although the processes that are
hypothesised to explain the attraction effects seen in the comprehension literature are
similar to those used to explain grammatical violation effects in the electrophysiolo-
gical literature (see e.g., Nevins, Dillon, Malhotra, & Phillips, 2007), it is important to
note that the attraction effect occurs over a dependency spanning a longer distance
than the local agreement effects studied in the present paper. Nevertheless, some of the
processes might be common to different types of agreement. For example, Wagers
et al. (2009) suggest that a plurality effect observed in several reading studies might
have basis in morphological complexity (see their Table 12). Statistical measures of
morphosyntactic complexity, such as those described here, might be informative in
this regard.
MORPHOSYNTACTIC INTEGRATION 295
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

More recently, lexical variables such as morphological family size have been
considered in neurophysiological studies. Using an English visual lexical decision task,
Solomyak and Marantz (2010) have recently shown with magnetoencephalography
(MEG) that activity in a time window around 0.17 sis correlated with the conditional
probability of a word given its suffix. This approach used a region-of-interest analysis
with a source reconstruction to examine the time course of the brain response to the
presentation of words presented as text in English. This result provides support for the
approach of relating continuous lexical variables to single-trial electrophysiological
responses. Effects of morphological family size have also been observed with MEG
(Pylkka
¨nen, Feintuch, Hopkins, & Marantz, 2004), but as with the behaviorally
oriented work reviewed above, not in phrasal context. The present study compares the
responses to strong and weak forms in phrasal contexts, and will focus mainly on
grammatically correct phrases.
Other work with EEG supports the role of the amount of (semantic) information in
phrasal processing. The original report of the N400 semantic violation effect in
sentence processing (Kutas & Hillyard, 1980), as well as the work that followed, shows
that words with high information content (e.g., low probability), as determined by the
preceding semantic or discourse context, elicit a larger amplitude potential. In another
example, Gunter et al. (2000) found the LANP600 complex to gender violations
only when the closure probability of the noun is high. Low closure probability nouns
elicited the LAN, but the P600 was strongly reduced. Other studies have argued that
the P600 violation effect, seen in response to grammatically incorrect sentences,
reflects the lower probability of the ungrammatical sentences as compared to the
grammatical sentences (Coulson, King, & Kutus, 1998; but see Osterhout & Hagoort,
1999).
SUMMARY AND PREDICTIONS
The research reviewed above suggests that the statistical distribution of the inflectional
suffixes within a paradigm can modulate isolated word recognition, so that words with
higher information content are judged more slowly and lead to a larger amplitude
brain response. Second, EEG research has shown that violations of a morphosyntactic
constraint in sentence or phrasal context are registered as early as 0.3safter the onset
of a violation. The general aim of the present research is to determine whether the
statistical distribution of inflectional suffixes in the German adjectival paradigm will
modulate the cortical response to the adjective in phrasal context.
In phrasal context, information content might influence the process of recognising
a word or its integration into the previous grammatical and semantic context.
Previously, Vainio et al. (2008) have proposed that effects as late as the P600 might be
taken as relatively late integrative effects, while ERP effects that are much earlier (e.g.,
B0.2 s) might be taken as an influence on the word recognition process itself. On the
other hand, it is clear from the grammatical violation literature that morphosyntactic
violations can elicit LAN effects that begin approximately 0.3 safter the onset of a
critical word or morpheme. Thus, we take 0.3sas the lower bound for the beginning of
integrative processing. To the extent that information content effects (including
interactions) begin before 0.3 sthis would be taken as evidence that it has likely
influenced word recognition rather than integration. Effects that begin later are more
likely to reflect phrasal integration.
296 DAVIDSON, HANULI
´KOVA
´, INDEFREY
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

In the experiment described below, these issues are investigated by comparing the
responses of nonviolation prepositional phrases in strong and weak configurations.
The phrases are presented using prepositions with different grammatical case
requirements in order to test whether any observed differences between the strong
and weak configurations depend on the specific grammatical case, and whether
lexically-specific variables influence the response. In the experimental task, partici-
pants were presented with short prepositional phrases; half the phrases were
grammatical, and half were ungrammatical. The task for participants was to judge
whether the phrases were acceptable or not.
Several linking assumptions about the electrophysiological response should be
made explicit (Barber & Kutas, 2007; Schall, 2004; Tanenhaus & Trueswell, 2004). The
experiment reported here attempts to correlate electrophysiological activity recorded
with EEG to a probability model for German morphosyntax. Although EEG has
good temporal resolution relative to other noninvasive neurophysiological recording
techniques, it is nevertheless only able to record volume currents that are aggregated
over relatively large spatial and temporal scales relative to the activity of neurons or
cortical columns (Nunez & Srinivasan, 2006). Our main assumption is that some of
the electrical potentials present in patches of connected cortical areas responsible for
morphosyntactic or lexical processing will have slow enough time dynamics in order to
summate so that they can be observed with EEG.
Given these assumptions, we hypothesise that the amount of information conveyed
by a word determines the integration of the word form into its context (e.g., forming a
linked short-term memory representation with the previously established grammatical
and semantic context). If the relative amount of morphosyntactic information
conveyed by a word is higher, then the word will not be integrated as quickly because
it will take longer to establish this linked memory representation. Our additional
assumption is that the late ERP component magnitude will be greater when it takes
longer to form this linked memory. If this assumption holds, the magnitude of the
ERP response to the adjective should be larger in the strong form, because it carries
more information. Our assumption is that creating a linked memory representation
involves neural synchronisation within a given patch of cortex, and that this gives rise
to amplitude differences observed in the average ERP. It should also lead to longer
judgment times for the strong form to the extent that this delay is propagated to the
motor system.
METHOD
Participants
Native right-handed German participants (N20) were recruited with posted
advertisements from Radboud University in Nijmegen, a city near the east border
of The Netherlands with Germany. The advertisements described a generic EEG
experiment, and asked for native German-speaking participants. No participant
reported problems with hearing, vision, or prior neurological injury.
Design, materials, and procedure
The design was intended to contrast the ERP response to the adjective in the strong
versus the weak form of the phrases following the German inflectional paradigm (see
Tables 1 and 3, and the earlier examples [1] and [2]. The main experimental variables
MORPHOSYNTACTIC INTEGRATION 297
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

included the contrast between the two conditions strong and weak (Dcln), as well as
the contrast between dative and accusative preposition (Prep), both variables within-
subjects. The strong phrases were constructed without a determiner, while the weak
phrases included a determiner. We used two different prepositions (mit,ohne)to
provide two different cases (dative and accusative, respectively). To complete the
phrases, we chose nouns corresponding to two genders (masculine, neuter) in singular
form (see Table 3 for examples). In addition to the Dcln and Prep factors, the
numerical covariates Iwx,I
xjw
, and Dwere calculated according the equations (7), (4),
and (9), respectively. Multiple linear regression was used to orthogonalise these
variables with respect to each other prior to modeling.
Along with the grammatically correct phrases analyzed here, an equal number of
violation phrases were presented, consisting of a violation of declension (e.g., ‘‘mit
dem *kleinem Boot’’). These violation stimuli were similar to those presented to native
speakers and second language learners in Davidson and Indefrey (2009). The original
goal of the present design was to follow up on Davidson and Indefrey (2009) using a
wider variety of lexical materials with native speaker participants. In this paper, we
concentrate not on the violation-control contrast, but rather the control conditions
primarily. This is done in order to test an assumption, namely whether the EEG
sources underlying the violation response are also sensitive to nonviolation
TABLE 3
Examples of the experimental stimuli. The column
N
refers to the number of items presented in
each experimental list
Declension Preposition Gender Example N
Weak Dative Masc mit dem großen
Mann
40
Neu mit dem großen Haus 40
Accusative Neu ohne das große Haus 160
Strong Dative Masc mit großem Mann 20
Neu mit großem Haus 60
Accusative Masc ohne großen Mann 40
Neu ohne großes Haus 80
Filler category Example N
Violation-declension mit dem großem
Mann
40
ohne das großes Haus 40
Control-other mit dem Haus 40
ohne den Mann 40
Violation-other mit großes Haus 40
ohne großem Mann 40
mit dem großes Haus 40
ohne das großem
Haus
40
mit große Haus 20
mit große Mann 20
mit das Haus 40
ohne dem Mann 40
ohne das große Mann 40
mit großen Mann 80
Note: Haus (house), Mann (man).
298 DAVIDSON, HANULI
´KOVA
´, INDEFREY
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

morphological information, and if so, in what way. The violation stimuli are treated as
fillers except for the purpose of selecting electrodes of interest (the violation response
to the declension violation is used to define groups of electrodes for the later wavelet-
regression analysis). It should be noted that for the nonviolation stimuli, the design is
not completely crossed (or balanced) for the three factors of declension, preposition,
and gender (see Table 3) because of our choice of the violation-control contrasts.
Also, note that the materials consisted of prepositional phrases, rather than full
sentences. Although it would be preferable to employ full sentential stimuli, the
reduced length of the word series allows for a greater number of trials.
The stimuli consisted of 587 adjective and noun pairs, chosen to capture most of
the surface form frequency range, and were arranged into the phrases by a native
German speaker. The adjectives and nouns were arranged into eight different lists. The
stimuli were constructed by choosing adjectivenoun pairs (for the masculine, and the
neuter nouns, separately), and arbitrarily assigning groups of items to the conditions
shown in Table 3. After creating the specific combinations of preposition, determiner
(if necessary), adjective, and noun for each condition, the lists were randomised
without any constraints separately for each participant. Table 3 shows the number of
items for each combination of declension, preposition, and gender.
In the experimental task, the participants were asked to classify the prepositional
phrases as grammatically acceptable or unacceptable following serial visual presenta-
tion at 0.6 sISI/word (words were shown for duration of 0.3 s, with a blank period of
0.3 s between word onsets). The entire experiment consisted of one session in which a
series of the phrases were presented as trials, with breaks approximately every 30 trials,
or when requested by the participant. Participants were asked to relax or rest but were
otherwise not provided a task during the breaks. Each trial began with a white fixation
cross for 1 s, followed by the serial presentation of the phrase. At 0.5 safter the onset
of the final word, another fixation cross was presented in yellow to indicate that a
classification response was requested. The next trial began 2 safter the classification
response.
Recordings and data analysis
EEG was recorded from 64 electrodes referenced to the left mastoid using battery-
powered BrainVision BRAIN AMP Series amplifiers (Brain Products GmbH,
Mu
¨nchen, Germany), later re-referenced to an average reference. Signals were sampled
at 500 Hz, with a low-pass filter at 200 Hz and a high-pass filter with a time constant
of 159 sduring acquisition. Electrodes were applied to an equivalent inter-electrode
distance Easy-Cap (Brain Products). Impedance levels were kept below 10 kVat the
electrode-skin interface, with input impedance at the amplifiers at 10 MV(see Ferree,
Luu, Russell, & Tucker, 2001). An additional electrode was placed below the left eye,
referenced to an electrode above the eye, to record activity related to blinks or vertical
eye movements. Lateral eye movement activity was recorded as the difference between
channels at the left and right outer canthus.
The recorded EEG data were screened for eye movement, muscle, and other noise
artifacts (artifact trials were excluded), filtered with a low-pass filter at 50 Hz (two-
pass 6th-order Butterworth finite impulse response filter), and segmented into 1 s
epochs consisting of 0.1 sbefore the onset of the critical word (CW) and 0.9 sfollowing
the CW. The resulting epochs were baselined with respect to the 0.1sbaseline interval
and averaged according to experimental condition. Only trials with correct responses
MORPHOSYNTACTIC INTEGRATION 299
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

were included in the analyses, and for the wavelet-regression analysis described below
only grammatically correct phrases were analysed.
Two electrode groups were chosen to summarise the ERP response, based on the
contrast between the declension violation and control conditions described earlier.
There were two time windows for this contrast at 0.3 to 0.5 sand at 0.5 to 0.9 s, in line
with the time windows typically chosen for the analysis of the LAN and P600
violation effects, respectively (see also Davidson & Indefrey, 2009). The analysis was
conducted using a clustering and randomisation test (Maris & Oostenveld, 2007) over
electrodes in these time windows. See Supplementary Section 1 for discussion of an
alternative analysis based on analysis of variance (ANOVA). The goal of this analysis
was to determine whether there was a larger amplitude response for the violation
condition, and if so, to identify which electrodes were sensitive to this contrast. These
electrodes were then used for the later wavelet-regression analysis of the control
conditions. For the LAN-contrast, there was a significantly more-negative amplitude
response for the violation condition compared to the control condition over eight left
frontal electrodes (Supplementary Figure 1 ), summedT20.12, p.025. For the P600
contrast, there was a significantly more positive amplitude response on 14 central-
posterior channels for the violation condition compared to the control condition
(Supplementary Figure 2 ), summedT48.18, pB.001. Based on these results, the
left anterior group was comprised of the average of the electrodes near F7 (electrodes
8:9, 18:20, 33:34, and 39 on the Easy-Cap montage, see Supplementary Figure 3), and
a posterior group comprised of the average of the electrodes surrounding Pz
(electrodes 1, 4:7, 12:16, 26:27, and 29:30). Note that these sets of electrodes also
correspond to the approximate locations at which the negative spatial mode of the
LAN effect, and the positive spatial mode of the P600 violation effect have been
observed in previous studies (e.g., Gunter et al., 2000; Koester et al., 2004).
A functional mixed effects regression model (Morris & Carroll, 2006) was used to
estimate the effect of the regressors described in the Materials section. The functional
regression is similar to mixed effect regression except that bcoefficients are modeled as
functions of time (see Davidson, 2009 for an overview and application to EEG data).
Briefly, the single trials from an EEG experiment are transformed into wavelet
coefficients using a discrete wavelet transform. Posterior coefficients for the regression
are obtained using Markov chain Monte Carlo simulation, and the resulting
coefficients are transformed back into the time domain. This approach also provides
a Bayesian false discovery rate, calculated according to Morris, Brown, Herrick,
Baggerly, and Coombes (2008). Posterior credibility intervals around the functional
response are also derived. These can be interpreted as a confidence interval for the
functional response that takes into account the dependencies between data points that
arise in time series analyses. The credibility interval width is constrained so that
a0.1, and the minimum effect size was set at 0.1 mV, the recording amplifier
resolution. In order to place the regression parameters on a common scale, factor
regressors were coded as 0.5/0.5 and other predictors were standardised to a zero-
mean, two standard deviation norm. For a similar, nonfunctional regression model
approach, see Hauk, Davis, Ford, Pulvermuller, and Marslen-Wilson (2006). Note that
the wavelet-based approach reported here includes only participants as a random
effect. Although in principle both participants and items could be specified as
(crossed) random effects with this approach, numerical instability prevented this in the
present case.
300 DAVIDSON, HANULI
´KOVA
´, INDEFREY
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

RESULTS
The behavioral responses suggested a slower response time for the strong form than
the weak (see Figure 1). Classification times for the weak form were faster than the
strong form by 87 ms (highest posterior density interval 30, 136 ms). Subjects also
made slightly fewer errors with the weak form (4%) versus the strong form (7%;
p.001). There were no main effects of preposition, nor an interaction between
declension and preposition, for either response times or error rates.
In contrast to the behavioral results, the ERP results suggested a more complex
pattern of results. The grand average ERPs for the strong and weak responses, for left
frontal and posterior electrode groups, are shown in Figure 2, for the dative (mit) and
accusative (ohne) prepositions.
The posterior electrode group shows a main effect of the strongweak contrast for
the accusative (beginning at approximately 0.4 s, see Figure 2d), but no effect is
apparent for the dative. While there is some indication of a difference between strong
and weak conditions for the frontal electrode group for the dative preposition at
approximately 0.3 s(Figure 2a), the evidence is not strong.
1
The functional regression analysis supported the above description for the main
grammatical factors, and in addition showed statistical relationships with several of
the information quantity regressors in several time windows (see Figure 3 for both the
left frontal and the posterior electrode groups). These will be described below. In all of
the plots, the highlighted time points are the mean values that exceed the minimum
effect size and whose estimated probability exceeds the false discovery rate threshold.
Additional analyses (conducted separately) including word length, simple form
frequency, simple lemma frequency, and trial block showed no evidence of main
effects or interactions with declension or preposition, so these factors will not be
considered further. See Supplementary Section 2 for the analysis of the factor gender.
For the frontal group, the first plot (Ave) of Figure 3a shows the average response
at the left frontal interest region to the weak condition, at the average level of the other
1
Note that with a more restricted set of electrodes centered around the approximate location of F7, there
is stronger evidence for this Dcln*Prep interaction, with the same functional form as shown in Figure 2a.
Figure 1. Average response times (left) and errors (right) for grammatical classification. Error bars indicate
standard error of the mean over participants.
MORPHOSYNTACTIC INTEGRATION 301
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

Figure 2. Grand average event-related potential response to the strong and weak forms of the adjectives for the left frontal and posterior electrode groups, for the dative and
accusative prepositions.
302 DAVIDSON, HANULI
´KOVA
´, INDEFREY
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

Figure 3. Coefficients (functional beta weights) for the left frontal (aj) and posterior (kt) electrode groups for the effects of the experimental factors declension (Dcln) and
preposition (Prep), as well as interactions with several information quantities (see main text for description).
MORPHOSYNTACTIC INTEGRATION 303
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

variables. The highlighted points are those that are likely to be greater or less than zero
(i.e., the grand average response). There were no main effects for Iwx,I
xjw
,orD. There
is a brief main effect of Prep at approximately 0.8 s, and of declension (Dcln) also at a
late interval (0.75 s). There is no strong evidence of an interaction (Dcln*Prep)
between declension and preposition on the frontal electrodes. Finally, the information
measure I
xjw
modulated the interaction between Dcln and Prep in a late time window
(Figure 3i). The interaction was more negative at approximately 0.6 to 0.8 swhen the
paradigm-specific information I
xjw
was greater. This modulates the nonsignificant late
negative difference that is apparent in Figure 3g. Figure 4 shows the average ERP
amplitudes for strong and weak declension in this late time window for the two cases,
for high and low values of I
xjw
(based on a median split). The difference between
strong and weak is similar for the dative and accusative case for high values of I
xjw
.In
contrast, for low values of I
xjw
, the dative case shows the opposite pattern as the
accusative case. For the dative case, there was a greater magnitude response for the
weak declension compared to the strong but for the accusative, the weak declension
led to a more negative response than the strong. In sum, for the frontal electrode
group the interaction indicates that I
xjw
modulates the response to the declension
contrast, depending on the type of preposition, relatively late within the time series.
On the posterior electrodes, there were significant main effects of preposition
(Figure 3o) and declension (Figure 3p), as well as a similar interaction with the
information quantity regressors (see Figure 3s) as was the case with the frontal
response. In addition, there was a two-way interaction between declension and
preposition (Figure 3q), indicating that for the accusative preposition, the strong
declension had a more negative potential than the weak, starting at approximately
0.4 sand remaining until approximately the end of the observation interval (see also
Figure 2). The brief main effects of preposition and declension should be seen in light
of this interaction. The paradigm-specific information content I
xjw
modulated the
Figure 4. Average potentials for the interaction of declension (Dcln) and preposition (Prep), at high and
low values of I
x|w
, for the left frontal electrode group.
304 DAVIDSON, HANULI
´KOVA
´, INDEFREY
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

interaction between declension and preposition (Figure 3s) in a sustained late time
window, starting at approximately 0.8 s.
Figure 5 shows the average ERP amplitudes for strong and weak declension in this
late time window for the two cases, for high and low values of I
xjw
. The difference
between strong and weak is similar for the dative case for both high and low values of
I
xjw
(again, based on a median split). In contrast, for the accusative case there was no
difference between strong and weak for high values of I
xjw
, but a greater magnitude
response for the weak declension compared to the strong for low values of I
xjw
.As
with the frontal response, this interaction indicates that I
xjw
modulates the response to
the declension contrast, depending on the type of preposition, but only relatively late
within the time series.
There were also several other brief higher-order interactions (e.g., Figure 3g,h,r),
but because they were not sustained they will not be considered here further.
DISCUSSION
The experiment reported here has shown a differential ERP response to the strong and
weak declension of the German adjective, presented in a short prepositional phrase via
rapid serial presentation. Also, participants were slower and (slightly) less accurate to
give judgments for phrases presented in the strong form, compared to the weak. In
addition, the adjective-specific information content of the inflection (e.g., I
xjw
) affected
the ERP response to the adjective, but at a relatively late time interval (i.e., at
approximately 0.8 s). This provides positive evidence that lexically-specific variables
can modulate the ERP response in phrasal context.
The behavioral results are largely congruent with previous studies of the influence
of sentential or phrasal context on lexical decision times. Previous experiments have
shown that readers and listeners can use phrasal context to aid lexical decision-making
Figure 5. Average potentials for the interaction of declension (Dcln) and preposition (Prep), at high and
low values of I
x|w
, for the posterior electrode group.
MORPHOSYNTACTIC INTEGRATION 305
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

(Bo
¨lte & Connine, 2004; Gurjanov, Lukatela, Lukatela, Savic´, & Turvey, 1985;
Gurjanov, Lukatela, Moskovljevic´, Savic´, & Turvey, 1985). The present results could
be interpreted to reflect a similar process. When participants encounter a German
adjective within an attributive adjectival phrase that also includes a determiner, the
determiner provides additional grammatical context for the processing of the adjective
and the noun to appear after the adjective. In addition, the grammatical information is
presented over a longer time period with the weak form of the phrases, allowing
comprehenders more time to integrate the information from each additional word
before encountering the end of the phrase. These factors would explain why
participants were faster to judge the weak forms of the phrases than the strong.
In the behavioral task, participants responded with the meta-linguistic judgment at
the end of the phrase, so it is not clear when the differential effect of declension might
have arisen. The judgment was cued at 0.5 safter the end of the phrase, meaning that
short-lived effect might have been missed using the judgment procedure employed
here. Future studies using acceptability judgments during the presentation of the
phrases themselves or reading time studies (e.g., eye-tracking) might be informative
about the timing of the effects of declension.
Nevertheless, the ERP results do offer some evidence about when the effects might
arise. The ERP contrasts revealed that with the accusative (but not dative)
prepositions for the posterior electrode group, there was a differential response in a
time window beginning at approximately 0.4 s. The timing of this response difference
would suggest that this activity could reflect grammatical or semantic integration, as
previous ERP studies have shown that the effect of grammatical or semantic context
most often is apparent at this time (for review see Kutas, Van Petten, & Kluender,
2006). Beyond this, it is not clear why the two different prepositions would lead to a
differential response. Although the accusative preposition ohne (‘‘without’’) also
conveys a difference in semantic meaning (negation) compared to the dative mit
(‘‘with’’), this does not straightforwardly lead to the prediction of the difference that
was observed.
Another potential explanation for the differential response might be a difference in
featural information. If we follow Schlenker’s (1999) general idea of featural
information accumulating from left to right then we can ask what new information
the adjective form provides in the series of words. In the task the prepositions were
always present, so case information was present before the determiner or adjective. In
the weak conditions, the determiners provide the following new information:
[5] mit (dative) dem (F, Pl)
That is, gender is incompletely specified (masc or neuter), whereas number is
specified.
[6] ohne (acc) den (M, Pl)
[7] ohne (acc) das (MF, Pl)
In the accusative case, gender is fully specified, as well as number. In the three weak
conditions, the following adjective with suffix en provides no new information.
In the strong conditions, the adjectives provide the following new information:
[8] mit (dative) m(F, Pl)
That is, in the dative case, gender is incompletely specified (masc or neuter), and
number is specified.
[9] ohne (acc) n(M, Pl)
[10] ohne (acc) s(MF, Pl)
In the accusative case, gender is fully specified, as well as number.
306 DAVIDSON, HANULI
´KOVA
´, INDEFREY
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

Under this feature account, the weak adjective form provides no new information
for dative or accusative. The strong form provides new information: plural for both
dative and accusative, but fully specified gender information only for accusative (for
dative, only underspecified gender information). It may be the case that the posterior
electrode group picks up this difference between the amount of gender information
provided by strong forms in dative and accusative contexts from 0.4sonwards. This
difference may then be further modulated by adjective-specific information content of
this suffix relative to all forms of the particular adjective stem at later stages (after
0.8 s). In future work, it would be advisable to use a greater variety of prepositions to
determine whether the interaction is due to a global property of case, or lexically-
specific properties, or both. It would also be useful to obtain a source reconstruction
for this response difference to determine whether it results from activity within distinct
cortical areas (e.g., frontal and temporal areas), or alternatively, activity within a
similar region but with a difference in orientation or geometry. This would help
constrain the interpretation of the response difference to the extent that semantic or
grammatical effects are differentiated by anatomical location.
The adjective-specific information content modulated the interaction of case and
declension. This result supports the contention of earlier work, conducted using
lexical decision tasks, that the statistical organisation of words within a morphological
paradigm has an influence on word processing (e.g., Milin et al., 2008). In addition the
results show that these effects can be seen in adjectives, as some previous experiments
have observed that adjectives are not always sensitive to these statistical variables
(Kostic´& Katz, 1987). Perhaps most importantly, however, the results presented here
extend this finding to phrasal contexts, and provide some evidence about the timing of
an electrophysiological correlate of this factor. The relatively late time interval for the
interaction is consistent with earlier reports from studies employing eye movement
recordings (Hyo
¨na
¨et al., 2002; Vainio et al., 2008). These studies have shown that
morphological complexity variables seem to exert an influence relatively late within
the parsing and interpretation process during the processing of sentences. In the
present study, although the grammatical context factors influenced the ERP response
magnitude beginning at approximately 0.4 s, the lexically-specific information
modulated the response only late within the response interval.
Although these ERP effects were present in grammatical phrases, it is worthwhile
considering in what ways they might be similar to previous ERP results using either
ungrammatical or noncanonical phrases. In earlier studies of violation effects (see
Introduction), two or three different stages of processing have been proposed. In the
earliest stage, lexical access occurs, during which morphosyntactic or semantic
features as accessed. During later stages, context integration, and in the case of input
with a grammatical error, re-analysis takes place. If one assumes that these stages
correctly characterise the parsing process, then the current results suggest that the
influence of statistical factors related to the inflectional paradigm operate primarily on
the later integration stage. Because the interaction was observed relatively late, it
seems more likely that integrative rather than lexical access processes were involved.
Also, because the interaction was observed in grammatically correct phrases, it seems
unlikely that a process of re-analysis would be involved. Earlier reports have shown
LAN and P600 effects for noncanonical compared to canonical phrasal order in
German (e.g., Ro
¨sler et al., 1998), and it appears that this type of effect would be more
similar to the present study. In studies of this response in German, the first
manifestation is observed at the determiner of the first noun phrase within the
sentence as a greater amplitude response to noncanonical phrasal order. In this sense,
MORPHOSYNTACTIC INTEGRATION 307
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

the effect can be seen as an interaction of inflectional morphology of the determiner
with the grammatical structure of the construction. There are important differences,
however. First, the canonicity effect is first observed within the time window of the
LAN, while the interaction here was first seen within the time window of the P600.
Second, the canonicity effect is seen in response to a determiner, while the interaction
here was present for inflected adjectives. Perhaps one reason for the difference in
timing could be due to this difference in word class, or possibly greater complexity
associated with identifying the suffix given the adjective, compared to the determiner.
In any case, the present results may imply that modulation of late components may
reflect the statistical information content of the inflected forms in grammatical
context. Future work might explore how probabilistic relationships at the phrasal
level, in addition to the morphosyntactic level, influence the electrophysiological
response.
One important caveat for this interpretation is that EEG, like all noninvasive
physiological measures, is selective for certain types of brain activity (as described in
the Introduction). It cannot be concluded, for example, that an earlier modulation of
the response is not present in cortex. Rather it was not apparent for the present
recording setup and design. In the absence of any evidence that the modulation is
earlier, however, the most parsimonious conclusion is that the influence only occurs
relatively late in the response interval. Future work might examine both EEG and eye-
tracking performance using the same materials, to determine whether there is item-
covariance between the effects seen in EEG and eye-tracking. Also, it is important to
consider that the responses reported here are restricted to the left frontal and posterior
electrode groups. It may be possible in future studies to extend the wavelet-regression
technique employed here to include spatial topography as well as time, or ideally, to
source waveforms (e.g., see Solomyak & Marantz, 2010). A final caveat is that only
certain parts of the German adjective declension were examined here. Future studies
might obtain a better picture of the response to the adjectival declension contrasts by
sampling adjectives that occur in a greater variety of cases and genders. The present
results serve as an initial estimate of the response for the paradigm.
In common with some of the lexical decision experiments reviewed in the
Introduction, the present study employed a task in which participants provided a
meta-linguistic judgment about the acceptability of the phrases. It might be the case
that the ERP patterns we observed were influenced by the judgment task, and it is
worth considering whether the need for a judgment may have influenced the late
responses that we observed. It seems possible that the late interaction effects seen in
the present study are analogous to a late verification phase of the P600 violation
effects seen previously in the literature (see Introduction). This might have been the
case if, as suggested by an anonymous reviewer, the presence of violation stimuli in the
experiment, and/or the need to make acceptability judgments, induced participants to
employ the same verification mechanisms on the control trials as on the violation
trials. This cannot be excluded in the present design. Future studies might employ a
task that does not involve meta-linguistic judgments, and which does not include
violation stimuli, such as passive reading or listening. This would take advantage of
the property of EEG measurements, which do not require that a specific behavioral
task is used in order to measure ERP components.
At a very general level, the results demonstrate that ERP amplitude differences can
be observed without (necessarily) contrasting violation and control stimuli, which
remains one of the most popular methods for using electrophysiology to study
sentence comprehension. The contrast between strong and weak declension led to a
308 DAVIDSON, HANULI
´KOVA
´, INDEFREY
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

differential response within grammatically licensed phrases, and this result was
apparent in the presence of several other regressors for lexical factors. These
functional regression results also add further support to the general approach of
using linguistically conditioned statistical variation as a means of investigating the
neural correlates of language processing, and ultimately the function of the networks
that are responsible for linguistic comprehension. Although both approaches have
been taken (use of nonviolation paradigms, the influence of lexical factors) previously
in the EEG sentence processing literature, the intended contribution of the present
experiment was to illustrate how linguistically conditioned morphosyntactic variation
might be reflected in the ERP response to words in phrasal context, where the
variation is conditioned by the phrasal context. While the results presented here are
preliminary in many respects due to design and task limitations, the general approach
may prove useful in further investigations of declension and other morphosyntactic
processes, especially those that attempt to link the results from behavioral and
electrophysiological methods.
Manuscript received 13 October 2009
Revised manuscript received 17 August 2011
REFERENCES
Baayen, R. H., McQueen, J., Dijkstra, T., & Schreuder, R. (2003). Frequency effects in regular inflectional
morphology: Revisiting Dutch plurals. In R. H. Baayen & R. Schreuder (Eds.), Morphological structure in
language processing (pp. 355390). Berlin: Mouton de Gruyter.
Barber, H. A., & Carreiras, M. (2005). Grammatical gender and number agreement in Spanish: An ERP
comparison. Journal of Cognitive Neuroscience,17, 137153.
Barber, H. A., & Kutas, M. (2007). Interplay between computational models and cognitive electrophysiology in
visual word recognition. Brain Research Reviews,53,98123. doi:10.1016/j.brainresrev.2006.07.002.
Bierwisch, M. (1967). Syntactic features in morphology: General problems of so-called pronominal inflection in
German. In To honor Roman Jakobson (pp. 239 270). The Hague: Mouton.
Blevins, J. P. (1995). Syncretism and paradigmatic opposition. Linguistics and Philosophy,18, 113152.
Bock, K., & Miller, C. A. (1991). Broken agreement. Cognitive Psychology,23,4593.
Bo
¨lte, J., & Connine, C. M. (2004). Grammatical gender in spoken word recognition in German. Perception &
Psychophysics,66, 10181032.
Cahill, L., & Gazdar, G. (1997). The inflectional phonology of German adjectives, determiners, and pronouns.
Linguistics,35(2), 211245.
Clahsen, H., Eisenbeiss, S., Hadler, M., & Sonnenstuhl, I. (2001). The mental representation of inflected words:
An experimental study of adjectives and verbs in German. Language,77, 510543.
Coulson, S., King, J., & Kutas, M. (1998). Expect the unexpected: Event-related brain response to
morphosyntactic violations. Language and Cognitive Processes,13,2158.
Davidson, D. J. (2009). Functional mixed-effect models for electrophysiological responses. Neirofiziologiya/
Neurophysiology,41,7987.
Davidson, D. J., & Indefrey, P. (2009). An event-related potential study on changes of violation and error
responses during morphosyntactic learning. Journal of Cognitive Neuroscience,21, 433446.
Deutsch, A., & Bentin, S. (2001). Syntactic and semantic factors in processing gender agreement in Hebrew:
Evidence from ERPs and eye movements. Journal of Memory and Language,45, 200224.
DWDS-Kerncorpus. (2008). Berlin-Brandenburgische Akademie der Wissen-schaften [Berlin Brandenburg
Academy of Science]. Berlin. Available from: www.dwds.de.
Faussart, C., Jakubowitz, C., & Costes, M. (1999). Gender and number processing in spoken French and
Spanish. Revista di Linguistica,11, 357374.
Ferree, T. C., Luu, P., Russell, G. S., & Tucker, D. M. (2001). Scalp electrode impedance, infection risk, and
EEG data quality. Clinical Neurophysiology,112, 536544.
Friederici, A. D., Hahne, A., & Mecklinger, A. (1996). Temporal structure of syntactic parsing: Early and late
event-related brain potential effects. Journal of Experimental Psychology: Learning, Memory.Cognition,22,
12191248.
MORPHOSYNTACTIC INTEGRATION 309
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

Geyken, A. (2007). The DWDS corpus: A reference corpus for the German language of the 20th century. In
C. Fellbaum (Ed.), Collocations and idioms: Linguistic, lexicographic, and computational aspects. (pp. 23
41). London: Continuum Press.
Gunter, T. C., Friederici, A. D., & Schriefers, H. (2000). Syntactic gender and semantic expectancy: ERPs reveal
early autonomy and late interaction. Journal of Cognitive Neuroscience,12, 556568.
Gurjanov, M., Lukatela, G., Lukatela, K., Savic´, M., & Turvey, M. T. (1985). Grammatical priming of inflected
nouns by the gender of possessive adjectives. Journal of Experimental Psychology: Learning, Memory and
Cognition,11, 692701.
Gurjanov, M., Lukatela, G., Moskovljevic´, J., Savic´, M., & Turvey, M. T. (1985). Grammatical priming of
inflected nouns by inflected adjectives. Cognition,19,5571.
Hagoort, P., Brown, C. M., & Groothusen, J. (1993). The syntactic positive shift (SPS) as an ERP measure of
syntactic processing. Language and Cognitive Processes,8, 439483.
Hagoort, P., Wassenaar, M., & Brown, C. M. (2003). Syntaxrelated ERPeffects in Dutch. Cognitive Brain
Research,16,3850.
Hauk, O., Davis, M. H., Ford, M., Pulvermuller, F., & Marslen-Wilson, W. D. (2006). The time course of visual
word recognition as revealed by linear regression analysis of ERP data. Neuroimage,30, 13831400.
Hyo
¨na
¨, J., Vainio, S., & Laine, M. (2002). A morphological effect obtains for isolated words but not for words
in sentence context. European Journal of Cognitive Psychology,14, 417 433.
Kaan, E. (2002). Investigating the effects of distance and number interference in processing subject-verb
dependencies: An ERP study. Journal of Psycholinguistic Research,31, 165193.
Koester, D., Gunter, T. C., Wagner, S., & Friederici, A. D. (2004). Morphosyntax, prosody, and linking
elements: The auditory processing of German nominal compounds. Journal of Cognitive Neuroscience,16,
16471668.
Kostic´, A. (1991). Informational approach to processing inflected morphology: Standard data reconsidered.
Psychological Research,53,6270.
Kostic´, A. (1995). Informational load constraints on processing inflected morphology. In L. B. Feldman (Ed.),
Morphological aspects of language processing. Hillsdale, NJ: Lawrence Erlbaum.
Kostic´, A., & Katz, L. (1987). Processing differences between nouns, adjectives, and verbs. Psychological
Research,49, 229236.
Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity.
Science,207, 203.
Kutas, M., Van Petten, C. K., & Kluender, R. (2006). Psycholinguistics electrified II (19942005). In M. A.
Gernsbacher & M. Traxler (Eds.), Handbook of psycholinguistics (2nd ed, pp. 83143). New York: Elsevier.
Levy, R. (2008). Expectation-based syntactic comprehension. Cognition,106, 11261177.
Lukatela, G., Kostic´, A., Feldman, L., & Turvey, M. T. (1983). Grammatical priming of inflected nouns.
Memory & Cognition,11,5963.
Lukatela, G., Kostic´, A., Todorovi, D., Carello, C., & Turvey, M. T. (1987). Type and number of violations and
the grammatical congruency effect in lexical decision. Psychological Research,49,3743.
Maris, E., & Oostenveld, R. (2007). Nonparametric statistical testing of EEG- and MEG-data. Journal of
Neuroscience Methods,164, 177190.
Milin, P., Kuperman, V., Kostic´, A., & Baayen, R. H. (2008). Words and paradigms bit by bit: An information-
theoretic approach to the processing of inflection and derivation. In J. P. Blevins & J. Blevins (Eds.), Analogy
in grammar: Form and acquisition (pp. 214278). Oxford: Oxford University Press.
Molinaro, N., Vespignani, F., & Job, R. (2008). A deeper reanalysis of a superficial feature: An ERP study on
agreement violations. Brain Research,1228, 161176.
Morris, J. S., Brown, P. J., Herrick, R. C., Baggerly, K. A., & Coombes, K. R. (2008). Bayesian analysis of mass
spectrometry proteomics data using wavelet based functional mixed models. Biometrics,64, 479489.
Morris, J. S., & Carroll, R. J. (2006). Wavelet-based functional mixed models. Journal of the Royal Statistical
Society, B,68(2), 179199.
Moscoso del Prado Martı´n, F., Kosti, A., & Baayen, R. H. (2004). Putting the bits together: An information-
theoretical perspective on morphological processing. Cognition,94,118.
Mu
¨nte, T. F., & Heinze, H. J. (1994). ERP negativities during syntactic processing of written words. In H. J.
Heinze, T. F. Mu
¨nte, & G. R. Mangun (Eds.), Cognitive electrophysiology: Basic and clinical applications
(pp. 211238). Boston: Birkhu
¨ser.
Mu
¨nte, T. F., Heinze, H. J., & Mangun, G. R. (1993). Dissociation of brain activity related to syntactic and
semantic aspects of language. Journal of Cognitive Neuroscience,5, 335344.
Mu
¨nte, T. F., Matzke, M., & Johannes, S. (1997). Brain activity associated with syntactic incongruities in words
and pseudo-words. Journal of Cognitive Neuroscience,9, 300311.
Nevins, A., Dillon, B., Malhotra, S., & Phillips, C. (2007). The role of feature-number and feature-type in
processing Hindi verb agreement violations. Brain Research,1164,8194.
310 DAVIDSON, HANULI
´KOVA
´, INDEFREY
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012

Nunez, P. L., & Srinivasan, R. (2006). Electric fields of the brain: The neurophysiology of EEG (2nd ed). Oxford:
Oxford University Press.
Osterhout, L., & Hagoort, P. (1999). A superficial resemblance does not necessarily mean you are part of the
family: Counterarguments to Coulson, King and Kutas (1998) in the P600/SPS-P300 debate. Language and
Cognitive Processes,14,114.
Osterhout, L., & Holcomb, P. J. (1992). Event-related brain potentials elicited by syntactic anomaly. Journal of
Memory and Language,31, 785806.
Osterhout, L., & Mobley, L. A. (1995). Event-related brain potentials elicited by failure to agree. Journal of
Memory and Language,34, 739773.
Pearlmutter, N. J., Garnsey, S. M., & Bock, K. (1999). Agreement processes in sentence comprehension. Journal
of Memory and Language,41, 427456.
Penke, M., Janssen, U., & Eisenbeiss, S. (2004). Psycholinguistic evidence for the underspecification of
morphosyntactic features. Brain and Language,90, 423433.
Pylkka
¨nen, L., Feintuch, S., Hopkins, E., & Marantz, A. (2004). Neural correlates of the effects of
morphological family frequency and family size: An MEG study. Cognition,91, B35B45.
Ro
¨sler, F., Pechmann, T., Streb, J., Ro
¨der, B., & Hennighausen, E. (1998). Parsing of sentences in a language
with varying word order: Word-by-word variations of processing demands are revealed by event-related
brain potentials. Journal of Memory and Language,38, 150176.
Schall, J. D. (2004). On building a bridge between brain and behavior. Annual Review of Psychology,55,2350.
Schlenker, P. (1999). La flexion de l’adjectif en allemand: la morphologie de haut en bas [The inflection of
adjectives in German: Morphology from top to bottom]. Recherches linguistiques de Vincennes,28, 115132.
Silva-Pereyra, J., & Carreiras, M. (2007). An ERP study of agreement features in Spanish. Brain Research,1185,
201211.
Solomyak, O., & Marantz, A. (2010). Evidence for early morphological decomposition in visual word
recognition. Journal of Cognitive Neuroscience,22, 20422057.
Tanenhaus, M. K., & Trueswell, M. K. (2004). Eye movements as a tool for bridging the language-as-product
and language-as-action traditions. In J. C. Trueswell & M. K. Tanenhaus (Eds.), Approaches to processing
world-situated language: Bridging the product and action traditions (pp. 338). Cambridge, MA: MIT Press.
Vainio, S., Hyo
¨na
¨, J., & Pajunen, A. (2008). Processing modifier-head agreement in reading: Evidence for a
delayed effect of agreement. Memory & Cognition,36, 329340.
Wagers, M. W., Lau, E. F., & Phillips, C. (2009). Agreement attraction in comprehension: Representations and
processes. Journal of Memory and Language,61, 206237.
Wunderlich, D. (1997). Der unterspezifizierte Artikel [The underspecified article]. In M. S. U. & K.-H. R.
Christa Drscheid (Eds.), Sprache im Fokus (pp. 4755). Tu
¨bingen: Niemeyer.
Zwicky, A. M. (1986). German adjective agreement in GPSG. Linguistics,24, 957990.
MORPHOSYNTACTIC INTEGRATION 311
Downloaded by [Max Planck Institut Fur Psycholinguistik] at 06:43 07 February 2012