Syntactic processing of Subject-Verb Number
Agreement: The effects of distance and
complexity on Agreement Processing.
Vicente Soto de Amesti
Student Number: 10391290
Supervisor: Judith Rispens
Co-Supervisor: Sible Andringa
October 20, 2014
AbstractThe aim of the current study was to investigate how the brain interprets subject-verb number agreement during language comprehension. Event re-lated potentials (ERP’s) where measured in order to investigate the neural correlates of syntactic disagreement processing in native language process-ing. High density EEG measurements were recorded while 18 native Spanish speaking participants read grammatical sentences as well as sentences con-taining four types of different subject-verb number disagreements. ERP’s were elicited to examine the temporal brain dynamics during on-line sen-tence processing of subject-verb number agreement and disagreement with varying levels of distance and complexity. Subject-verb agreement violations elicited widespread negativities in the 350ms - 500ms time window, which is interpreted as strong negativities responding to morphosyntactic ungram-maticality. A typical P600 effect is also observed in response to agreement violations. Modulation of the P600 amplitude are examined in relation to
the effect of distance. Longer sentences elicit a larger P600 component
then short sentences. The effect of morphosyntactic ungrammaticality is investigated and the possibility that subject-number agreement processing mechanisms are sensitive to aggregated distance between noun and verb is discussed. The conclusions of this study is an effect of aggregating distance between a subject and its corresponding verb on the P600 component.
1
Introduction
For producing and comprehending/reading language there are multiples di-mensions of analysis that must be resolved before language can be under-stood. A basic well established idea in psycholinguistics is the fact that during language processing, properties regarding rules and information re-garding content represent two distinct levels of processing (Ferreira and Clifton Jr, 1986) associated with specific time-locked electrophysiological
responses in the brain. In order for humans to efficiently communicate
language both these levels of processing must be quickly resolved and inte-grated. In native speakers event related potentials (ERP’s) can be stably elicited by presenting subjects with sentences containing a different types of experimental manipulations on spoken or visually presented sentences (Kaan, 2007). If errors in semantic features are present, for instance, sen-tences tend to loose their meaning and become nonsense (Sanders et al., 2002; Kutas and Hillyard, 1980). On the other hand, errors in syntactic properties though ungrammatical, in the strictest sense, maintain meaning and allow on-line error detection and repair (Friederici et al., 2002). Subject-verb agreement, for instance, rely on purely morphosyntactic, rule-based, relationships between the subject and the verb. The purpose of this study is to investigate brain responses to subject-verb disagreement processing in native speakers of Spanish.
When processing subject-verb number agreement, the syntactic number features of the noun must match the number features of the succeeding verb. An example of a deviation to this type of agreement is typically produced in second language acquisition of English in sentences like ”Can we *uses the car?”. In this example the plural subject ”we” does not share features with its corresponding verb ”*uses”. Though ungrammatical, this sentence is still semantically interpretable and maintains meaning. Different syntactic features are present in order to sustain agreement relationships and main-tain their syntactic structure in a sentence. During comprehension, number agreement relationships require storing the features of the noun and main-taining them until the corresponding verb appears. Agreement relationships, as such, are crucial mechanisms for structurally organizing language. They allow for co-referencing, maintaining anamorphic relationships as well as processing some long distance relationships in languages with inflectional morphology (Silva-Pereyra and Carreiras, 2007). Specific brain processes have been related to agreement processing. In this study we aim to investi-gate how the brain resolves agreement relationships in different cases were it has been violated. specifically, it investigates specific cortical brain activ-ity involved in detecting and processing variations of subject-verb number disagreements in Spanish.
1.1
Morphosyntactic agreement processing in Spanish
In Spanish, grammaticality correct sentences require correspondence in per-son, number and gender syntactic features between subject and verb. Num-ber morphological representation of any concept is variable because it can adopt at least two different forms, singular and plural. Never the less, num-ber features are always considered a conceptual feature identifying the quan-tity of the referent and are, there for, unambiguous in this regard. Gender, on the other hand, bares relevant conceptual characteristics when regarding animate ones, but is arbitrarily assigned to inanimate entities, for instance ”la[fem] mujer[fem]” / ”the woman” opposed to ”la[fem] guitarra[fem]” / ”the guitar”. In a correct sentence then, the noun and it’s corresponding
verb must coincide in number, like this simple sentence: ”Los perros est´an
en la casa” (”The dogs are in the house”). In this sentence both the subject
(Perros / ”dogs”) and verb (est´an / ”are”) are plural generating agreement
between features. In longer, more complex sentences intervening material can be present between the noun and the verb ”The dog of the boys is in the house”, but the singular features of the head noun are maintained. Given a linguistic architecture where verb features must agree with those of the noun, some authors argue in favour of a hierarchical account for processing syntactic features (Carminati, 2005; Greenberg, 2005; Harley and Ritter, 2002). According to this idea, agreement features have a particular estab-lished hierarchy (P erson > N umber > Gender). These studies propose that hierarchical organization of agreements features represents a direct re-flection of the feature’s cognitive significance. In this way ’person’ would be more salient than ’number’, and ’number’ would be more salient that ’gender’. Sylva-Pereira & Carreiras (2007) found evidence against the hier-archical ordering of features in Spanish. They show no difference in syntactic cognitive processing (salience/hierarchy) between features number and per-son. In this investigation number agreement were investigated in different sentences containing one or two nouns, as well as sentences including and adverb to append additional distance between linguistic constituents. Neu-ral processes behind subject-verb number agreement are examined in native Spanish speakers.
1.2
Left Anterior Negativity (LAN)
In investigating linguistic relationships, EEG has proven a pioneer research tool by establishing time referenced events to evoked responses in the brain. Language specific ERP’s have been documented in response to multiple
types of linguistic processes (Molinaro et al., 2011). Syntactic language
properties, for instance, are typically associated with three major ERP com-ponents in the brain, early left anterior negativity (ELAN), left anterior negativity (LAN) and the syntactic positive shift (P600).
ELAN, occurring typically between 100ms - 300ms post stimulus on-set time, is commonly related to first pass syntactic parsing (Hahne and Friederici, 1999), but the functional interpretation of this component, never the less, has been controversial. Friederici et al. (2002) . Negativities elicited by critical words in sentences containing violations to word category
viola-tions like ”The scientist criticized Max’s of proof the theorem” (Neville
et al., 1991) differ from the negativities elicited by to morphosyntactic vio-lations (occurring between 300 and 500 ms post stimulus onset time) elicited by sentences like All turtles have four *leg and a tail (Kutas and Hillyard, 1983). Nevertheless, due to the very quick timing of this earlier response some authors claim that baseline problems and contextual factors may
con-tribute to ELAN effects in word category violation studies. This raises
serious issues regarding the validity of these results and the interpretation of this early response (Steinhauer and Drury, 2012).
The LAN component is a negative-going deflection typically appearing as a response to morphosyntactic violations during online sentence process-ing. It reaches its maximum amplitude between 300ms and 500ms after the onset of an anomalous word. Due to that it shares a time window with an-other language related ERP (N400), laterality is a critical for determining the identity of the LAN. The LAN is functionally related to morphosyntactic violations and is normally interpreted as reflecting an initial stage in which morphosyntactic information is processed (Friederici, 1995). In this early stage, word form violations such as number violations, are detected as well. The laterality and location of this component show less consistency across studies. In two studies (Hinojosa et al., 2003; Kaan, 2002) results show evi-dence of a bilateral anterior negativity response between 200ms and 500ms from the apparition of a deviant word, but not specific to the left hemisphere (Hagoort et al., 2003; Hagoort, 2003; Kaan, 2007). Syntactic processing in sentence comprehension proves to be among the first mechanisms to be en-gaged in language. Hinojosa et al. (2003) study two types of violation in Spanish, phrase structure and morphosyntactic violations. They found no differences in neural activity in response to both of these violations for the LAN component, and significantly different waveforms resulting in the later P600 component.
1.3
The P600
The P600 is a positive-going deflection for a syntactic event related potential (EPR) that peaks around 600ms stimulus onset time and has a posterior distribution in the brain (Osterhout and Holcomb, 1992). Also referred to as the Syntactic Positive Shift (SPS), this component is a well studied highly reliable ERP responding consistently to syntactic violations. The P600 has a centro-posterior distribution across scalp locations (Hagoort et al., 1993)
and has been typically interpreted as representing syntactic repair processes in sentence comprehension (Friederici et al., 2002). Multiple variations of morphosyntactic deviations can elicit a P600 response. There is consistent evidence for two main type of deviations generating this neural component; agreement mismatches between the subject and the verb (to number, gen-der or person) (Mancini et al., 2011) and ambiguously constructed sentences where the correct attribution of features is unclear (i.e. garden path sen-tences). In both these cases the syntactic content in the sentence requires identification, revision and retrieval of the correct expression (Gouvea et al., 2010). Molinaro et al. (2011) conducted a survey of investigations into agree-ment processing. They report 74% of subject-verb number agreeagree-ment stud-ies, including studies in Spanish, find evidence for a LAN / P600 pattern to syntactic deviations, as such, we expect to find a similar evoked pattern for syntactic deviations in our results. In interpreting the functional relevance of the P600, Osterhout et al. (1994) propose that the amplitude of the P600 waveform is in direct relation to the difficulty to generate a grammatical representation. Kaan et. al., (2000) suggest a slightly different functional interpretation of the P600. They provide evidence to support the idea that the P600 is better understood as an index for general syntactic integration difficulty. This investigation examined how experimental manipulations on distance and complexity to sentences subject-verb number disagreements modulate resulting ERP.
1.4
Distance and complexity in agreement processing
Linear distance (i.e. the amount of words separating a noun from the verb) (Kaan, 2002; Hammer et al., 2008) and complexity (i.e. the presence of an intervening prepositional phrase in a sentence) (Pearlmutter, 2000) can have important effect on agreement processing. In this investigation linear distance has been distinguished from structural distance in order to closely examine differences in elicited neural responses. Sentences in which an ad-verb is located between the noun and ad-verb are understood as linearly more distant than sentences in where the noun is immediately next to the verb. Sentences containing an additional prepositional noun phrase necessarily in-crease the amount of distance between noun and verb as well. In these ”com-plex” sentences nevertheless, an additional active noun becomes present and the features of the intervening material must be monitored accordingly. Re-search shows that distance is not the only confining element to establish the relative difficulty of a sentence, but any element that requires higher pro-cessing loads would increase propro-cessing demands. The effect of complexity on agreement processing has also been researched in few other studies (Kaan and Swaab, 2003; Kolk et al., 2003). In the present study, complexity is ma-nipulated by adding a second, intervening, prepositional phrase in between
the subject and the verb. Pearlmuter (2000) concluded that complex inter-vening material, like noun phrases, would decrease stored subject-number information necessary to recognize subject-verb number agreement errors. Kaan (2002) tested the effects of altering the number properties in the head noun and the object. The results show a delayed P600 component was when sentences contained an intervening plural noun, supporting the idea that processing requirements limit recognition of subject-verb agreements. Kaan (2002) also found that more offline judgment errors were made when the stimuli sentences contained a subject that was singular and a plural object, compared to any other number combinations between subject and object. A following study by Kaan and Swaab (2003) compared simple and complex sentences (containing two noun phrases) and found a smaller P600 amplitude for the complex condition compared to the simple ones in their ERP components. Finally there is a know illusory effect within subject-verb agreement for complex sentence processing. Language production studies evidence that speakers sometimes generate sentences where the verb inap-propriately agrees with a proximal noun phrase. Bock and Miller (1991) show that in production sentences like (1), where the singular subject ”key” does not agree with the plural verb ”are” are often regarded as correct.
(1) The key to the cabinets are on the table.
In comprehension studies sentences containing subject-verb disagree-ments overall reading times increase compared to their grammatical coun-terpart. In sentences like (1), nevertheless, the error is commonly overlooked and no delays in reading times are observed (Pearlmutter et al., 1999; Wagers et al., 2009). The number features of the interceding plural prepositional phrase are transferred to the verb and trigger a illusion of grammaticality. The presence of a illusory agreement effect will be investigated in this study given that neural responses could be different than in normal subject-verb disagreement processing.
Gibson (2000) presents the Dependancy Locality Theory (DLT) for re-source use in sentence comprehension. At the base of the DLT is the idea that the mechanisms behind sentence parsing (i.e. online consecutive im-puting of each word given preceding information) require at least two com-putational resources. Structural integrations (i.e. incorporating a new word into the existing information) and keeping the sentence structure, including features, in memory. Given this proposal, ”greater the distance between an incoming word and the most local head or dependent to which it attaches, the greater the integration cost” (Gibson, 1998). Though wildly assumed that propinquity between the subject and verb is the determining factor in how difficult processing a particular sentence is, there is evidence to sup-port the idea that added distance between the noun and the verb can also facilitate processing (Konieczny, 2000).
1.5
Research questions
This investigation looks into the differences to the ERP components elicited by sentences containing one, and two nouns at different linear distances. We include manipulations to simple and complex sentences via the addition of an interceding adverb (positioned in all cases immediately before the verb) to study the effect of additional linear distance on LAN and the P600 as well. The underlying assumption here is that keeping track of noun’s fea-tures requires additional processing loads and that this will be reflected in the resulting ERP’s. We expect to see modulations of the LAN and P600 amplitude to ungrammatical sentences containing two nouns opposed to one, as well as significantly larger mean amplitudes in high distance, compared to low distance, ungrammatical conditions. The aim of the current study was to analyse differences in evoked brain responses for agreement violations in simple sentences containing one noun (*NP - V - PP) and complex sen-tences containing two nouns (*NP - PP - V - PP). Distance manipulation were constructed via the addition of a interceding adverb between the noun and verb. regarding distance manipulation we also expect significantly larger ERP modulations given the additional processing requirements of maintain-ing the features of the head noun. We incorporate both variables into our analysis in four types of sentences; Low distance Simple [NP-VP-PP], Low distance Complex [NP -NP-VP-PP], High distance Simple [NP-Adv-VP-PP], and High distance Complex [NP-NP-Adv-VP-PP] (details in Materials subsection). We view modulations to LAN and P600 component waveforms in response to deviations in each condition. We aim to answer the question if syntactic ERP’s are sensitive to experimental manipulation on complexity and linear distance. Specifically, does the addition of linear distance gener-ate a difference in the average amplitude of the LAN and P600 component? Do increments in complexity generate a significantly different neural activ-ity in agreement processing? And will we see a typical LAN/P600 ERP biphasic pattern in response to subject-verb disagreements? We expect to see significantly larger LAN and P600 amplitude in response to morphosyn-tactic ungrammatical complex and long sentences than in simple and short ones.
2
Methods
2.1
Participants
Twenty native speakers of Spanish participated in the experiment. Two par-ticipants were removed from the data, one due to a low score in the judgment task and the second given irregularities in their EEG measurements. The data of 18 subjects was used in the final analysis (n=18, 11 female). Ages
ranged from 20 to 33 years old ( 25.9 mean age SD= 4.05). All partici-pants had normal or corrected to normal vision and three of them were left handed. Participants where recruited online using native Spanish Facebook groups. Native Spanish speaking participants originated from four differ-ent countries, Spain(n=10), Chile (n=5), Colombia (n=2) and Argdiffer-entina (n=3). One Spanish and one Chilean participants’ data was excluded from the sample.
All participants were measured at the research lab of the ACLC in Am-sterdam. Before testing, participants were sent an e-mail containing an ini-tial questionnaire regarding their language history (Annex Section: ACLA sheet). All participants had Spanish family backgrounds and coursed ele-mentary and high school at least in Spanish. No participant reported any language impediments or reading problems of any kind. All participants, but two free volunteers, were paid 20.- euros for their participation.
2.2
Stimuli
6 initial practise trials commenced the experiment. Two versions of 225 test sentences (half grammatical) and 80 (half semantically implausible) fillers were used in the experiment. Each individual test sentence was constructed containing at the least one head noun phrase (HN), one verb (V) and one prepositional phrase (PP). Both experimental manipulations - number and complexity - where crossed yielding four experimental sentence types.
Table 1: Test Sentence Structures for Conditions
Low distance Simple Low distance Complex High distance Simple High distance Complex
LdS LdC HdS HdC
Total N = 48 Total N = 64 Total N = 48 Total N = 64
24 Grammatical 32 Grammatical 24 Grammatical 32 Grammatical
12 [N(sg.)-V(sg.)] 16 [HN(sg.)-N(pl./sg.)-V(sg.)] 12 [N(sn.)-V(sg.)] 16 [HN(sg.)-N(pl./sg.)-V(sg.)] 24 Ungrammatical 32 Ungrammatical 24 Ungrammatical 32 Ungrammatical
12 [N(sg.)-V(pl.)] 16 [HN(sg.)-N(sg./pl.)-V(pl.)] 12 [N(sn.)-V(pl.)] 16 [HN(sg.)-N(sn./pl.)-V(pl.)] 12 Singular Nouns 16 Singular Head Noun 12 Singular Nouns 16 Singular Head Noun 12 Animated Nouns 16 Animated Head Nouns 12 Animated Nouns 12 Animated Head Nouns
Table 1: Structures for all conditions and amount of stimuli per condition displayed. N = amount of stimuli per condition. Complex sentences have bigger number of stimuli sentences per condition given that the additional NP also required extra contra balancing to control for number and animacy. In Complex conditions the head noun was always kept singular. HN (head noun) - N (noun) - V (verb) - sg. (singular) - pl. (plural). In Low distance Complex conditions ungrammatical sentences with a plural interceding PP are examined for an illusory agreement effect.
(Adv.) was inserted between the HN and the V so as to introduce linear distance between the HN and the corresponding V. In complex conditions and additional NP interceded between the HN and the V. Nouns where controlled for animacy (half aniamate), gender (half male), number (half singular) and frequency. Nouns with a frequency under 5 per million where
not used to create stimuli, we used Corpus del Espa˜nol Online and only
consulted 20th century texts. Adverbs where not controlled for frequency for our high distance conditions. Ungrammatical filler sentences where con-structed in the same ratio and manner to the experimental conditions but containing a semantic violation in the verb position. All sentences were constructed in present indicative grammatical tense. To check the stimuli, full lists of stimuli were sent out to ten native Spanish speakers who were asked to judge each sentence as correct or incorrect. Items with low ratings (< .85) were corrected or excluded the test material altogether.
In doing so we created four sentence types for our experiment. Low distance simple sentences (2), low distance complex sentences (3), high dis-tance simple sentences (4), high disdis-tance complex sentences (5) and filler sentences (6). (2) El The mueble cabinet guarda contains los the platos. dishes. ‘The cabinet contains the dishes’.
(3) La The secretaria secretary intencionalmente intentionally saluda greet a [to la the] gente. people. ‘The secteraty intentionally greets people’.
(4) El The perro dog del [of/the] ni˜no boy camina walk en in la the calle street ‘The dog of the boy walks in the street’
(5) Las The amigas friends de of la the ni˜na girl tipicamente typically comen eat en at su her casa. house. ‘The friends of the girl typically eat at home’.
(6) Filler El The perro dog de of la the ni˜na girl naturalmente naturally llueve rain en at su her casa. house. ‘The dog of the girl naturally rains at home’.
Ungrammatical versions of each item were created by changing the verb to plural or singular as in (6). Two versions of the items were used in the experiment. If a sentence was ungrammatical in version 1 of the stimuli the same item would be grammatical in version 2. Full list of the stimuli in Annex 1.
(7) *El *The perro dog del [of/the] ni˜no boy caminan walk en in la the calle street ‘*The dog of the boy walk in the street’
2.3
Procedure
All participants were measured in a testing booth, seated comfortably in a chair at approximately 90cms away from a LCD display monitor. Instruc-tions where given verbally and presented on screen in the booth whilst the EEG cap and electrodes were being set up. Six practice trials proceeded in-struction, three of these correct. All sentences were presented in white over a black background. Punctuation and use of upper and lower case letters was normal. The presentation sequence was as follows: Before the first and each succeeding item a green circle appeared onscreen (3s) followed by a red fixation cross (1s) indicating the beginning of a sentence. Sentences were presented in a chunk by chunk manner. Each chunk was composed of 1 to 3 words and remained on the screen for 350 ms followed by a blank screen of 150 ms between chunks. The critical verb always constituted its own chunk and always presented in second or third position. Subjects were instructed to blink during the presence of the green circle on screen in order to reduce the possibility of blinks at critical moments. Participants were asked to read the sentences carefully for comprehension. After each sentence participants answered a judgment task: Was the sentence incorrect or correct? (¿Es
la oraci´on incorrecta (bot´on rojo) o correcta (bot´on blanco)? ). After a
re-sponse the following item would commence. To avoid end word (wrap up) effects, critical verb was never in the end position, but always followed by a PP (min 2 words max 4 words). The critical verb agreed with the subject in the grammatical versions, but not in the ungrammatical. Participants were asked to judge the correctness of the sentence they just read and to press the corresponding button on the response box to answer the question. After the response the following trial started. The complete experiment consisted of 4 blocks of around 76 sentences each. Subjects controlled amount of rest time in between blocks. Complete experiment lasted between 1,5h - 2h per participant.
2.3.1 EEG Measurements
Electroencephalografic (EEG) measurements where recorded using 64 scalp electrodes were placed in a standard BioSemi 64-channel elastic cap. Seven additional flat electrodes were placed above and below the left eye, at the outer canthi, at both mastoids and on the nose. ECOG recorded horizon-tal and vertical eye movements including eye blinks. The EEG signal was amplified with a direct current amplifier build into each electrode. EEG
data was on-line referenced to the linked mastoid electrodes and the nose electrode. Data was high-pass filtered at 0.5 Hz and low-pass filtered at 35 Hz with the sample rate always kept under 10khz (8kHz). ERP’s where time-locked to the critical word.
Region of Interest Layout
Figure 1: Approximate location of 64 electrodes on scalp and ROI division.
2.3.2 EEG analysis
The EEG data was pre-processed using BrainVisionAnalyzer2.0 (Brain Prod-ucts, Munich, Germany). Only grammatical and ungrammatical trials judged correctly were used in the analysis. All channels were referenced to the av-erage of the nose and mastoids activity, data was high pass filtered at 0.5Hz and low pass filtered at 35Hz. ERP’s where computed offline using 1700ms epochs (-200 ms to 1500ms), time locked to stimulus onset time being the
presentation of the critical verb. All trials judged incorrectly in the post sentence judgment task were removed, leaving us only with correctly per-ceived items. A semi-automatic artefact rejection was conducted and data was visually inspected for epoch removal. Our criteria for removal included segments with electrical activity exceeding +-100uV and/or epochs that contained blinks in critical segments. (8.9% trials removed). ERP baseline correction was preformed using -200ms to 0ms pre-stimulus time window.
Six regions of interest (ROI) where created over anterior and posterior electrode sites as showed in Figure 1. Regions of Interest were distributed as follows: Anterior Left (Fp1, AF7, AF3, F5, F3, F1, FC1, FC3, FC5), Midline Anterior (Fpz, AFz, Fz, FCz), Right Anterior (Fp2, AF8, AF4, F6, F4, F2, FC2, FC4, FC6), Posterior Left (CP5, CP3, CP1, P5, P3, P1, PO3), Midline Posterior (Cz, CPz, Pz, POz) and Posterior Right (CP6, CP4, CP2, P6, P4, P2, PO4).
3
Results
Statistical analysis for each time window consisted in a omnibus 6 (ROI) x 2 (Distance) x 2 (Complexity) x 2 (Grammaticality) repeated measures ANOVA to check for main effects in our data. When main effects are encoun-tered, follow up analysis are performed. Given that in both time-windows epsilon values > 0.75 in the Mauchly’s Test of Sphericity, we report the Huynh-Feldt correction to test for within-subject effects (Huynh and Feldt, 1976).
3.1
Acceptability Judgment Task Response Ratings
Data of one participant was excluded from all analysis due to a low score on judgment task (cutoff < 85%, p = .67). Accordingly, low rating items (< 85%) where discarded from the analysis a well. Overall, the partici-pants were equally accurate in judging grammatical from ungrammatical
sentences. According to the acceptability judgment after each sentence,
on average participants rated 91,5% of the all correct sentences presented as acceptable and 89,2% of the ungrammatical sentences were judged as unacceptable. This rating demonstrates that participants were adequately sensitivity to syntactic ungrammaticality in our stimuli sentences. Table 2 shows the average accuracy ratings for all participants for each condition.
Due to a non normal distribution of the response ratings for the post sentence judgment tests, non-parametric test were preformed on this set of data. A Wilcox signed ranks test was performed on all ungrammatical conditions against each other. Results indicates that subjects had signifi-cantly better accuracy at answering LdS (M=.925) sentences than any other condition, LdC (M = .888), Z = −2.949, p < .005; HdS (M = .871), Z =
Table 2: Average Judgement Scores per Condition. LdS LdC HdS HdC Grammatical (%) 91.1 91.4 94.4 89.1 Ungrammatical(%) 92.5 88.8 87.1 88.5 Mean .919 .884 .906 .888 Std. Dev. 5.15 6.15 6.32 5.01 −3.043, p < .005; HdC (M = .885), Z = −2.443, p < .005. This results is not unexpected as LdS conditions are the simplest type of sentence and very high accuracies were expected. A follow-up Wilcox Signed-ranks test was performed to checked the response rates for one particular subgroup of stimuli for an illusory agreement effect. We compared average accuracy to ungrammatical complex sentences with a ”singular head noun - plural noun - and plural verb” structure, to the accuracy rating of ungrammatical ”singular head noun - singular noun - plural verb” complex sentences. If an illusory agreement effect were present in our data more errors would be expected in identifying ungrammatical ”singular head noun plural noun -and plural verb” sentences. Ratings between these conditions are not dif-ferent from each other, Z = −1.22, p < .05 and, as such, no evidence of a illusory agreement effect was found in the judgment response data.
3.2
ERP analysis
LAN: 350ms to 500ms Time Window
For the 350ms-500ms time window, an initial 2(Gram) x 2(Complexity) x 2(Distance) x 6(ROI) Omnibus repeated ANOVA was conducted. Results show no significant main effects for Distance, Complexity or ROI F < 1 at this time window. Interestingly, no significant main effect was found for Grammaticality F (1, 17) = 2.135, p = .162. A marginally significant inter-action between Distance and ROI F (1, 17) = .271 , p = .060 was observed, but no other interaction appeared significant, F < 1.
To follow up the trend in the interaction between Distance and ROI an additional analysis was conducted with Distance, ROI and Grammaticality as factors. A 3 (Anterior ROI) x 2 (Grammaticality) x 2(Distance) repeated measures ANOVA provides results showing a main effect for Grammatical-ity, F (1, 17) = 9.195, p = .009 but not for ROI or Distance, F < 1 over
the anterior sites. Unlike our expectations, this result indicates a general negativity effect over anterior and posterior sites; these results are displayed in Table 3.
Table 3: Distance x ROI x Grammaticality Mean Amplitude for the 350ms – 500ms Time
window Table 3: Mean amplitudes elicited by high and low distances sentences after simple and complex conditions were collapsed for the 350ms – 500ms time window. Differences in the waveforms elicited from grammatical and ungrammatical sentences are not contained to anterior sites, but are also present in posterior ones as well.
Distance ROI Grammaticality Mean (µV)
Low Distance
Left Anterior Grammatical -.038 Ungrammatical -1.369 Mid Anterior Grammatical -.059 Ungrammatical -1.813 Right Anterior Grammatical .047 Ungrammatical -1.465 Left Posterior Grammatical -.257 Ungrammatical -1.593 Mid Posterior Grammatical -.730 Ungrammatical -1.978 Right Posterior Grammatical -1.742 Ungrammatical -.948
High Distance
Left Anterior Grammatical -1.652 Ungrammatical -1.630 Mid Anterior Grammatical -1.115 Ungrammatical -1.489 Right Anterior Grammatical -.633 Ungrammatical -.455 Left Posterior Grammatical -.596 Ungrammatical -1.605 Mid Posterior Grammatical -.487 Ungrammatical -1.315 Right Posterior Grammatical -.249 Ungrammatical -1.456
Table 3 summarizes the mean amplitudes elicited by grammatical and ungrammatical sentences. Results evidence differences in both anterior and posterior ROI. As such, our results evidence a lack of LAN in response to subject-verb number disagreements. These results present evidence for a general negativity effect to ungrammaticality present over anterior and posterior regions at this time. This effect is illustrated in Figure 2.
Topo-graphical maps show a medial-posterior negativity that becomes increasingly posterior over time. These widespread negativities are followed by strong positivities in posterior ROI in the following time window, between 500ms and 1000ms. These results are presented in the next subsection.
Grand Average Topographical Plots: Ungrammatical Sentences between 350ms-500ms
Figure 2: Topographical maps display the mean activity, relative to baseline, in response to Ungrammatical sentences between 350ms and 500ms from the apparition of the critical verb.
3.3
P600: 500ms to 1000ms Time Window
As in the earlier time window, a 2(Grammaticality) x 2(Complexity) x 2(Distance) x 6(ROI) omnibus repeated measures ANOVA was performed. Mauchly’s Test of Sphericity shows that epsilon > .75 in all effects and conse-quently a Huynh-Feldt statistic is reported for within subject effects. A main effect for Grammaticality was observed, F (1, 17) = 12.086, p = 0.019 as well as a marginally significant effect for Distance, F (1, 17) = 12.086, p = 0.072. No effect was observed for Complexity, F (1, 17) = 0.303, p = 0.589 or ROI F (1, 17) = 0.337, p = 0.881. A marginally significant interaction between ROI and Grammaticality, F (1, 17) = 2.748, p = 0.066 was observed but all other interactions proved not significant, F < 1. The main Grammaticality effect reflects a significant difference in the waveform amplitude elicited by
grammatical and ungrammatical sentences.
Table 4: Distance x ROI x Grammaticality Mean Amplitude for the 500ms – 1000ms Time window
Distance ROI Grammaticality Mean (µV)
Low Distance
Left Anterior Grammatical -.241 Ungrammatical 1.577 Mid Anterior Grammatical 1.221 Ungrammatical .553 Right Anterior Grammatical -.298 Ungrammatical 1.325 Left Posterior Grammatical -1.414 Ungrammatical 1.355 Mid Posterior Grammatical -1.313 Ungrammatical 1.023 Right Posterior Grammatical -1.164 Ungrammatical 1.515
High Distance
Left Anterior Grammatical .467 Ungrammatical 2.325 Mid Anterior Grammatical .371 Ungrammatical 1.827 Right Anterior Grammatical .236 Ungrammatical 2.791 Left Posterior Grammatical .195 Ungrammatical 3.456 Mid Posterior Grammatical -.032 Ungrammatical 3.302 Right Posterior Grammatical .176 Ungrammatical 3.506
Table 4: Mean amplitudes elicited by high and low distances sentences after simple and complex conditions were collapsed for the 500ms to 1000ms time window. Significant differences in the mean amplitudes are considerably greater over posterior sites than anterior ones. The greatest differences in elicited amplitude between grammatical and ungrammatical sentences are found at posterior sites in high distance conditions.
To investigate the marginally significant effect of Distance found in the omnibus ANOVA, second 6 (ROI) x 2 (Grammaticality) x 2 (Distance) was performed. Results show a significant effect of Grammaticality, F (1, 17) = 12.086, p = 0.003; ROI F (1, 17) = 4.167, p = 0.018 and a marginally sig-nificant effect of Distance, F (1, 17) = 3.745, p = 0.069. These main re-sults where further described by a marginally significant interaction be-tween Distance and ROI, F (1, 17) = 2.589, p = 0.078; but no other
in-teractions were observed Distance/Grammaticality, ROI/Grammaticality, Distance/ROI/Grammaticality, F < 1. Table 4 shows the mean amplitudes by ROI for high and low distance, grammatical and ungrammatical sen-tences. Ungrammatical sentences elicit larger positivities than grammatical ones in most ROI (with the exception of the anterior medial region), but larger differences between the waveforms are observed over the posterior sites. Following analysis were performed over posterior ROI.
A follow-up 3 (ROI) x 2 (Distance) x 2 (Grammaticality) repeated mea-sures Anova was performed to unpack this marginal interaction. Effects for Grammaticality, F (1, 17) = 15.101, p = 0.001;ROI, F (1, 17) = 6.951, p = 0.007; and for Distance, F (1, 17) = 7.315, p = 0.015 were found. These effects are qualified by a marginally significant interaction between tance and ROI, F (1, 17) = 3.501, p = 0.055. No other interactions, Dis-tance/Gram, ROI/Gram, Distance/ROI/Gram were relevant, F < 1.
Posterior electrodes Grammatical and Ungrammatical conditions:
Figure 3: Grand Average ERPs elicited by Ungrammatical (full line —) and Grammatical (dotted line - - -) sentences.
The mean amplitudes elicited by ungrammatical sentences were com-pared at high and low distance conditions in all posterior ROI. Results il-lustrate that the high distance sentences elicit a greater positivity than low
distance sentences in Left t(17)= -2.696; p = .015, and Right t(17)=-2.403; p = 0.028 anterior ROI. Grand average ERP’s are illustrated in Figure 4.
Posterior electrodes low v/s high distance ungrammatical sentences
Figure 4: Average ERP waveforms for High Distance ungrammatical sentences (dotted line - -) and Low Distance ungrammatical sentences (full line —) for posterior electrodes. The critical verb is presented at time 0. This figure depicts differences in ungrammatical sentences between 500ms and 1000ms.
Grand Average Topographical Plots: Ungrammatical Sentences between 500ms-1000ms
Figure 5: Topographical maps display the mean activity, relative to baseline, in response to Ungrammatical sentences between 500ms and 1000ms from stimulus onset time (this being the apparition of the critical verb).
Overall, gran average ERPs showed that ungrammatical sentences elicit a significantly larger positivity than grammatical ones as illustrated in Figure 3. As expected, results depict a P600 component in response to ungrammat-icality elicited by subject-verb agreement violations. Figure 5 show topo-graphical maps of the average electrical activity over the scalp for the 500ms - 1000ms time window. These plots show a typical P600 posterior positivity that reaches its maximum at around 800ms post critical word onset.
4
Discussion
The aim of this investigation was to view the effect of different levels of complexity and distance on subject-verb agreement processing. To do this, Spanish stimuli sentences were created in which the head noun was separated from the verb by either an adverb (distance) or an additional prepositional noun phrase (complexity). Behavioural data and ERP analysis results are discussed next. The main results are as follows, behavioural responses indi-cate subjects understood sentences correctly. An absence of illusory agree-ments effects was found. Electrophysiological waveforms display a
Nega-tivity / P600 pattern in response to subject-verb number disagreements in which only the later component seems differences in distance. Because subject-verb distance effects are only present in the P600 component, the conclusion is drawn that early ERPs related to syntactic processing provide error detection while later components represent repair and integration of the syntactic agreement features.
4.1
Behavioural Responses
Behavioural responses, to the post stimuli judgment task, suggest that sub-jects detect deviations to subject-verb number agreement in simple short sentences significantly better than in complex and/or long sentences. In addition, the existence of an illusory agreement effect was also investigated via the manipulation of the noun number features in complex conditions. We expected to find more judgment errors made to complex ungrammatical sentences with a ”singular head noun - plural noun - plural verb” struc-ture due to a known effect of grammatical asymmetry in which the feastruc-tures of the intervening noun are transferred to the verb (Wagers et al., 2009), but surprisingly, no effect of illusory agreement was found. No difference was found in the accuracy ratings between ”singular head noun - plural noun - plural verb” complex sentences and ”singular head noun - singular noun - plural verb” complex sentences in grammatical or ungrammatical conditions. One possible explanation is that in the current investigation the intervening prepositional phrases used in creating complex conditions were constructed with a preposition (determiner) and a noun like in (8).
(8) El The gato cat de of las the amigas girls camina walk en in la the calle. street . ‘The cat of the girls walks in the street’.
It is possible that the determiner of the second noun effectively narrows the morphological possibilities of the following word in the sentence. Instead of incrementing the distance, and therefore increase processing difficulty, the determiners may effectively predict the possible words it will encounter and immediately detect deviations to this expectation. Ultimately this would reduce, instead of increasing, computational demands. Konieczny (2000) shows evidence that in perception, at least, verb features can be anticipated through incremental integration of their arguments. Konieczny argues that locality is relevant for production rather than comprehension. During pro-cessing of a sentence, the type of the verb (i.e, its idiosyncratic feature) becomes continuously constrained to allow only certain arguments to be imputed, thus, facilitating syntactic processing, decreasing response times and increasing accuracy. In doing so he tests the locality-based preferences issue in sentence processing finding behavioural evidence against the DLT
(Gibson, 2000). Following this idea, in sentence (8), for instance, the de-terminer ’de las’ could actually be effectively facilitating processing rather than complicating it. By reducing the possible features of the following verb and generating a prediction, syntactic deviation processing could be assisted by the presence of the determiner ’of the’.
Another possibility an effect of word and structure frequency. Though
this particular sentence the construction ”el gato de las ni˜nas” is quite
common, other sentences like (9), containing an animate head noun followed by an inanimate intervening noun the sentence ensembles are much less frequent. (9) La The se˜nora woman de of los the pantanos swamps profundamente deeply duerme sleeps por in la the ma˜nana. morning.
‘The woman of the swaps deeply sleeps in the morning.’
It is possible that subjects were particularly careful in answering sen-tences they considered odd and where less attend to frequent sensen-tences of the same condition. If so, then the generation of expectancy to these ”odd-reading” sentences could also promote the use of strategies to answer the judgement tasks correctly, rather than rely on normal reading processes.
There are know effects of frequency tampering the overall difficulty of sentence processing (Van Petten and Kutas, 1990). In the current inves-tigation, Nouns were individually controlled for frequency. The complete sentence structure, however, was not. Carns and Foss (1971) propose that sentence structure plays a role in determining the relevance of a words
fre-quency. They found that reading times were prolonged following a low
frequency adjective. Carns and Foss (1971) conclude that adjectives are less predictable than verbs in a sentence, and there for more likely to be subject to a frequency effect. In our experiment adverbs were used to create linear distance between noun and verb. Adverbs, in this context, seem equally unpredictable as adjectives and there for as likely to be susceptible to infre-quency or expectancy of violation effects. This could also help explain the existence of a P600 and lack of a LAN component in our EEG data. ERP components are discussed next.
4.2
ERP Components: Negativity and the P600
In multiple studies (Wicha et al 2004, Barber et al 2004, Barber and Car-reiras 2005) a two-stage pattern, characterised by a left anterior negativity followed by posterior positivities is described in response to subject-verb disagreements to syntactic agreement violation in Spanish. In this inves-tigation, results point to a widespread negativity effect in response to
un-grammaticality rather than the expected LAN effect between 350ms - 500ms. These negativities are followed by positivities reflecting a P600 effect.
Sentences containing a subject-verb number disagreement clearly evoked a different brain response to critical verbs than correct sentences, but no specific location was found in the resulting ERPs. As such our predictions regarding the modulation of the LAN response as result of experimental manipulations on the test sentences cannot be clearly resolved. Never the less we provide a alternative explanation of these findings. Similar results were found by Hinojosa et al. (2003) in response to verb inflection violations like in sentence (10); taken from their stimuli.
(10) *La The prueba evidence ocultada hidden por by el the fiscal prosecuter aparec´ı. (’I’)appeared. ‘ The evidence (that was) hidden by the prosecuter (I) appeared.’ They a found non-lateralized central-anterior negativities to inflection violations. As expected negativities are found in response to agreement vi-olations. Unlike our results, Hinojosa et al. (2003) find that the negativities are restricted to anterior regions of the brain. Though it is not addressed in the study, sentence structure frequency and word context could be influ-encing the distribution of the ERP component. In a similar manner to our investigation, the structure of the stimuli sentences used in Hinojosa et al. (2003) is quite uncommon, particularly, the deviation to subject-verb person agreement is scarcely produced in natural speech. Unnaturally constructed sentences could be an issue for our investigation as well due to the fact that frequency differences, rather than purely morpho-syntactic deviations, could be generating a neural response.
In our experiment ungrammatical conditions do elicit a significantly greater negativity than grammatical sentences in the 350ms-500ms time win-dow, but this effect is not limited to anterior locations. Three other studies in Spanish (Hinojosa et al., 2003; Kaan, 2002; Leinonen et al., 2008) have found similar effects to these according to a recent review (Molinaro et al., 2011). But in all cases negativities are segregated to anterior locations. Of these three Kaan (2002) was the only to use multi-word intervening material and no verb inflection violation. Her results also evidence a early (300ms -500ms) bilateral anterior negativity effect in response to ungrammaticality and, as in our results, this component seems insensitive to distance manip-ulations. A anterior negativity effect has been reported to detect responses lexical-syntactic violations and is typically considered to reflect first stage parsing of agreement errors. This understanding of the anterior negativity component response explains, to some extent, the occurrence of a negativity effect in the present study as a response to disagreements. Coulson et al. (1998) found results more closely related to the ones presented in this study. They report different distributions in negativities responding to
ungrammat-ical sentences between 300ms and 500ms after the onset of a critungrammat-ical verb. Though no main effects of Complexity, Distance or ROI seem to influence the amplitude of the initial negativities, Distance does generate differences in the later P600 waveforms. As expected, between 500ms -1000ms from the appearance of a critical verb, ungrammatical sentences elicit a significantly greater negativity in posterior ROI and in anterior ones. Furthermore, once the overall data was collapsed over complexity, results show a relevant effect of distance on the P600. In the ERP analysis for our later time window, we view a differences in evoked responses to ungrammatical low distance sentences like (11) when compared to similar ungrammatical long distance sentences like (12). (11) *El *The perro dog caminan walk por on la the vereda. sidewalk. ‘*The dog on the sidewalk.’
(12) *La *The madre mother de of las the ni˜nas girls rapidamente quickly caminan walk a to casa. house. ‘*The mother of the girls quickly walk home.’
A clear P600 response was elicited in posterior brain regions between 500ms - 1000ms after the onset of the critical verb. Results show that the P600 is not sensitive to processing complexity increments within a sentence. Complex and simple conditions were collapsed after this, and a significant effect for distance on the P600 was found. The average positivity elicited by ungrammatical sentences was significantly larger in high distance sentences, compared to low distance sentences. Kaan (2002) used tailored stimuli with fairly complex intervening material to view the effects of distance on agree-ment processing. She used ”long conditions” that had three of four word phrase intervening material between the subject and the verb. Kaan re-ported a P600 at two time windows, an early P600 and a late P600. They collapsed their deviant-number condition in the analysis, and found that intervening material between a subject and the mismatch verb could trigger additional working memory processing difficulties reflected in a temporally delayed P600 waveform peak. In the current investigation, high distance conditions only used a one word adverb to add distance. It is likely that one word, independent of how long, is not sufficient material to raise the difficulty of the sentence enough to increase processing demands sufficiently. Complex sentence conditions, on the other hand, present multi word inter-vening material, but show no individual effect on sentence processing. As such, when collapsed over complexity, the distance that was factored into the analysis was augmented. Following this, we report an effect of distance on the P600s amplitude. We interpret this finding as a result of increasing processing demands via the addition of distance between the noun and the verb by 3 or 4 words. Since in complex conditions contain a higher amount
of intervening material due to the additional noun phrase (3 or 4 words), we propose that distance can indeed affect agreement processing in Spanish. These results provide further evidence to claims made by Kaan (2002) in regards to the effects of distance on feature tracking in sentence processing. Given that three or four words generate an important increase to the time and difficulty required to maintain the noun’s features in working memory and posteriorly reactivate them to match with the verb, larger amplitudes of the P600 are encountered. Kaan and Swaab (2003) provide evidence to support the idea that a posterior distribution of the P600 represents an in-dex of syntactic processing difficulty, including repair and syntactic revision. They posit a frontally distributed positivity would represent ambiguity res-olution responding to increase in the level of complexity of a sentence. This would explain the lack of differences in amplitudes for complex and simple sentences on the P600. Also it can provide an account for a increased pos-itivity in our high distance conditions. In longer sentences like (12), linear distance is sufficient to require significantly greater syntactic revision mecha-nisms. In this way the dimension of the syntactic repair process is indexed in the P600 amplitude. Given that in the present study short subject-verb dis-agreements elicit smaller P600 amplitudes than long disagreeing sentences, but only when collapsed across both simple and complex conditions, we postulate the idea that there is a threshold amount of intervening necessary to elicit a significantly different response. Kaan and Swaab (2003) present evidence for a ceiling effect in the P600 amplitude due to maximum resource use in sentence processing. To this theory we add the idea of a minimal re-source demand necessary for differential morphosyntactic parsing processes in native speakers. Also, due to the experimental design of the investiga-tion high and low distance complex sentences only had 16 items per subject and some of these where incorrectly answered, and there for not included in the ERP analysis. Low amount of stimuli could account for marginally significant effect in our results, this is further discussed in the Concluding Remarks Section of this report.
Though we do not find evidence of a typical biphasic LAN/P600 pattern in response to subject-verb disagreements. We report a negativity/positivity biphasic pattern in response to ungrammatical sentences containing subject-verb agreement deviations. Results evidence a general negativity, followed by a strong positivity for ungrammatical sentences. We interpret this neg-ativity response in our results as a detection of subject-verb agreement vi-olations. The component is followed by a strongly positive P600 compo-nent signalling syntactic repair or integration (Kaan and Swaab, 2003). As such subject-verb agreement processing in Spanish demonstrates a two stage process in which each epoch is identified by specific polarities in the ERP waveforms of each time window.
results indicate that initial syntactic parsing processes (before 500ms from critical verb onset time) relate to widespread cortical negativities, whilst secondary parsing mechanisms (between 500ms and 1s) relate to an ongoing positivity with posterior-cortical neural generators. Furthermore, additional distance between the noun and the verb (more than three intervening words) generate differences on the max amplitude of the P600.
5
Concluding Remarks
To summarise, behavioural judgment task response show adequate sensi-tivity to stimuli sentences and evidence no illusory agreement effect. The simplest sentences are judged more accurately that complex and long sen-tences, as expected. The ERP results presented show no evidence of a clear LAN response in the EEG data. A general negativity was, however, found in the early time window as a response to subject-verb disagreement. Pos-teriorly, a clear P600 response was detected in the second time window. The later component seems to respond to the addition of distance between subject and verb, but only in very high distance conditions.
As a final remark we wish to include some methodological considerations with regards to our work. The experimental design for this investigation
proved unnecessarily complex. This was due to this investigation being
modified from its original form where native speakers of Spanish would be compared to highly proficient second language learners. Given the absence of participants in the second language group during the pilot phase of the investigation, the experimental design and stimuli were altered resulting in a very complex nested design. Future research in this field should consider generating specific, non overlapping, experimental conditions to further test the effect of distance on subject-verb agreement in Spanish. In the same manner, due to contra balancing for plurality, animacy and gender in our experiment, the smallest experimental conditions in the ERP analysis had relatively few items (between 12 and 16 sentences). This could be causing marginally significant effects in our results due to lack of power. Future studies should account for amount of stimuli per condition used and try to maximize the expected effect. Hopefully this investigation will provide an important stepping stone for following researchers and establish general guidelines regarding investigations into agreement processing.
5.1
Acknowledgments
I would like to firstly acknowledge my supervisor Dr. Judith Rispens for the countless hours of supervision and teaching. This project would have in no way have been possible without her assistance and guidance. Secondly a
spatial thanks to Dirk-Jan Vet for all the technical support and to Tamara Plankman for all her help in setting up and conducting the pilot of this study. Thirdly, we acknowledge Dr. Sible Andringa for co-assessing this investigations and finally, special thanks to the ACLC for allowing open access to laboratories and equipment in the centre.
Bibliography
Carminati, M. N. (2005). Processing reflexes of the feature hierarchy
(per-son¿ number¿ gender) and implications for linguistic theory. Lingua,
115(3):259–285.
Carns, H. S. and Foss, D. J. (1971). Falsification of the hypothesis that word frequency is a unified variable in sentence processing. Journal of Verbal Learning and Verbal Behavior, 10(1):41–43.
Coulson, S., King, J. W., and Kutas, M. (1998). Expect the unexpected: Event-related brain response to morphosyntactic violations. Language and cognitive processes, 13(1):21–58.
Ferreira, F. and Clifton Jr, C. (1986). The independence of syntactic pro-cessing. Journal of memory and language, 25(3):348–368.
Friederici, A. D. (1995). The time course of syntactic activation during language processing: A model based on neuropsychological and neuro-physiological data. Brain and language, 50(3):259–281.
Friederici, A. D., Hahne, A., and Saddy, D. (2002). Distinct neurophysio-logical patterns reflecting aspects of syntactic complexity and syntactic repair. Journal of psycholinguistic research, 31(1):45–63.
Gibson, E. (1998). Linguistic complexity: Locality of syntactic dependen-cies. Cognition, 68(1):1–76.
Gibson, E. (2000). The dependency locality theory: A distance-based theory of linguistic complexity. Image, language, brain, pages 95–126.
Gouvea, A. C., Phillips, C., Kazanina, N., and Poeppel, D. (2010). The lin-guistic processes underlying the p600. Language and Cognitive Processes, 25(2):149–188.
Greenberg, J. H. (2005). Language universals: With special reference to feature hierarchies. Walter de Gruyter.
Hagoort, P. (2003). How the brain solves the binding problem for language: a neurocomputational model of syntactic processing. Neuroimage, 20:S18– S29.
Hagoort, P., Brown, C., and Groothusen, J. (1993). The syntactic posi-tive shift (sps) as an erp measure of syntactic processing. Language and cognitive processes, 8(4):439–483.
Hagoort, P., Wassenaar, M., and Brown, C. M. (2003). Syntax-related erp-effects in dutch. Cognitive Brain Research, 16(1):38–50.
Hahne, A. and Friederici, A. (1999). Electrophysiological evidence for two steps in syntactic analysis: Early automatic and late controlled processes. Cognitive Neuroscience, Journal of, 11(2):194–205.
Hammer, A., Jansma, B. M., Lamers, M., and M¨unte, T. F. (2008).
Inter-play of meaning, syntax and working memory during pronoun resolution investigated by erps. Brain research, 1230:177–191.
Harley, H. and Ritter, E. (2002). Person and number in pronouns: A feature-geometric analysis. Language, 78(3):482–526.
Hinojosa, J., Mart´ın-Loeches, M., Casado, P., Mu˜noz, F., and Rubia, F.
(2003). Similarities and differences between phrase structure and mor-phosyntactic violations in spanish: An event-related potentials study. Language and Cognitive Processes, 18(2):113–142.
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 and Behavioral Statistics, 1(1):69–82. Kaan, E. (2002). Investigating the effects of distance and number
interfer-ence in processing subject-verb dependencies: An erp study. Journal of Psycholinguistic Research, 31(2):165–193.
Kaan, E. (2007). Event-related potentials and language processing: A brief overview. Language and Linguistics Compass, 1(6):571–591.
Kaan, E. and Swaab, T. (2003). Repair, revision, and complexity in syntactic analysis: An electrophysiological differentiation. Cognitive Neuroscience, Journal of, 15(1):98–110.
Kolk, H. H., Chwilla, D. J., van Herten, M., and Oor, P. J. (2003). Structure and limited capacity in verbal working memory: A study with event-related potentials. Brain and language, 85(1):1–36.
Konieczny, L. (2000). Locality and parsing complexity. Journal of Psy-cholinguistic Research, 29(6):627–645.
Kutas, M. and Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity. Science, 207(4427):203–205.
Kutas, M. and Hillyard, S. A. (1983). Event-related brain potentials
to grammatical errors and semantic anomalies. Memory & Cognition, 11(5):539–550.
Leinonen, A., Brattico, P., J¨arvenp¨a¨a, M., and Krause, C. M. (2008).
Event-related potential (erp) responses to violations of inflectional and deriva-tional rules of finnish. Brain research, 1218:181–193.
Mancini, S., Molinaro, N., Rizzi, L., and Carreiras, M. (2011). When persons
disagree: an erp study of unagreement in spanish. Psychophysiology,
48(10):1361–1371.
Molinaro, N., Barber, H. A., and Carreiras, M. (2011). Grammatical agree-ment processing in reading: Erp findings and future directions. cortex, 47(8):908–930.
Neville, H., Nicol, J., Barss, A., Forster, K., and Garrett, M. (1991). Syn-tactically based sentence processing classes: Evidence from event-related brain potentials. Cognitive Neuroscience, Journal of, 3(2):151–165.
Osterhout, L. and Holcomb, P. J. (1992). Event-related brain
poten-tials elicited by syntactic anomaly. Journal of memory and language, 31(6):785–806.
Osterhout, L., Holcomb, P. J., and Swinney, D. A. (1994). Brain potentials elicited by garden-path sentences: evidence of the application of verb in-formation during parsing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(4):786.
Pearlmutter, N. J. (2000). Linear versus hierarchical agreement feature pro-cessing in comprehension. Journal of Psycholinguistic Research, 29(1):89– 98.
Pearlmutter, N. J., Garnsey, S. M., and Bock, K. (1999). Agreement
pro-cesses in sentence comprehension. Journal of Memory and language,
41(3):427–456.
Sanders, L. D., Newport, E. L., and Neville, H. J. (2002). Segmenting non-sense: an event-related potential index of perceived onsets in continuous speech. Nature neuroscience, 5(7):700–703.
Silva-Pereyra, J. F. and Carreiras, M. (2007). An erp study of agreement features in spanish. Brain Research, 1185:201–211.
Steinhauer, K. and Drury, J. E. (2012). On the early left-anterior negativity (elan) in syntax studies. Brain and language, 120(2):135–162.
Van Petten, C. and Kutas, M. (1990). Interactions between sentence context and word frequencyinevent-related brainpotentials. Memory & Cognition, 18(4):380–393.
Wagers, M. W., Lau, E. F., and Phillips, C. (2009). Agreement attraction in comprehension: Representations and processes. Journal of Memory and Language, 61(2):206–237.
ANNEX SENTENCES BY CONDITION Item Con
ditio n
NP NPⁿ Adverb VP PP
1 LdS Los gatos camina en la calle 2 LdS El músico huyen en la estación 3 LdS Las perras corren en la calle 3 LdS La amiga observa el suceso 5 LdS Los edificios cae al suelo 6 LdS El fuerte aguantan el ataque 7 LdS Las lámparas iluminan en la noche 4 LdS La comida se enfría en la mesa 9 LdS Los hombres roba en el mercado 10 LdS El motor hierven en el coche 11 LdS Las mujeres practican su instrumento 12 LdS La hormiga construye su colonia 13 LdS Los negocios vende los zapallos 13 LdS El mueble guardan los platos 15 LdS Las estufas calientan el departamento 16 LdS La cárcel permite las visitas 17 LdS Los ciudadanos vota en la elección 14 LdS El ratón comen mucho queso 19 LdS Las suegras regaña en la mesa 20 LdS La reina detiene la ejecución 21 LdS Los teléfonos suena en la mañana 22 LdS El vaso contienen líquido transparente 23 LdS Las ventanas tiemblan en la tormenta 23 LdS La llave cruje en la cerradura 25 LdS Los jefes despide al empleado 26 LdS El coche viajan a la playa 27 LdS Las tortugas nadan en el mar 24 LdS La mariposa vuela en el jardín 29 LdS Los puentes cruza la bahía 30 LdS El hotel cierran en el verano 31 LdS Las cosechas alimentan al pueblo 32 LdS La habitación aloja al invitado 33 LdS Los policías frena al vehículo 33 LdS El bombero extinguen el incendio 35 LdS Las viudas lloran en el funeral 36 LdS La enfermera trabaja en el hospital 37 LdS Los libreros afirma los estantes 34 LdS El filtro limpian el aire 39 LdS Las represas previenen la inundación 30 LdS La mañana comienza muy temprano 31 LdS Los pájaros planea en el aire 32 LdS El músico tocan el instrumento 33 LdS Las vecinas cocinan en la tarde 33 LdS La mesera lava en la pieza
35 LdS Los abrigos mantiene el calor 36 LdS El alimento engordan a las gallinas 37 LdS Las velas arden en la oscuridad 34 LdS La pipa quema el tabaco 39 LdC El amigo de los directores presentan un reclamo 50 LdC Los coches del actor corre mucho por la vía 51 LdC El ratón de los almacenes espanta a los clientes 52 LdC Los guardias del bar disculpan a los intrusos 53 LdC El edificio de los ancianos acogen a visitantes 53 LdC Las prendas del vendedor está en el canasto 55 LdC El suelo de los árboles sufre por la contaminación 56 LdC Los anillos del barril mantienen su integridad 57 LdC La profesora de las modelos sonríen al entrar 54 LdC Las abogadas de la sobrina llora por lo que sucede 59 LdC La dueña de las granjas cultiva tomates orgánicos 60 LdC Las empleadas de la empresa abren la gran puerta 61 LdC La escuela de las enfermeras cierran temprano hoy 62 LdC Las autoridades de la zona establece las tarifas 63 LdC La compuerta de las represas se sostiene por ahora 63 LdC Las luces de la pieza se queman con este voltaje 65 LdC La recepcionista de los barcos viajan en la noche 66 LdC Las monturas del caballo se resbala al correr 67 LdC La secretaria de los hoteles huele muy mal 64 LdC Las brujas del pueblo condenan a los insolentes 69 LdC La herencia de los porteros pierden hoy su valor 70 LdC Las chimeneas del departamento arde con llamas 71 LdC La llave de los gabinetes está en la calle 72 LdC Las pulseras del cantante golpean contra el suelo 73 LdC El gato de las niñas pasean en la calle 73 LdC Los caballeros de la dama conduce el desfile 75 LdC El rey del las haciendas dispone de los frutos 76 LdC Los herederos de la mansión reciben turistas de verano 77 LdC El bolso de las mujeres desaparecen en la estación 74 LdC Los huesos de la pierna fortalece el cuerpo humano 79 LdC El autor de las pinturas honra al maestro 40 LdC Los tapetes de la escalera decoran la casa fea 41 LdC El perro de los niños caminan por la calle 42 LdC Los peces del océano cambia de color
43 LdC El músico de los funerales finaliza lentamente la canción 43 LdC Los hombres del experimento fracasan en su labor 45 LdC El dolor de los veteranos dependen del medicamento 46 LdC Las estructuras del juego entretiene a los niños 47 LdC El cabello de los maniquíes cae al suelo 44 LdC Las aguas del temporal superan lo previsto 49 LdC La esclava de las princesas reposan al sol 90 LdC Las camareras de la nave salta por la turbulencia 91 LdC La jirafa de las praderas llena de mosquitos 92 LdC Las meseras de la taberna juntan bastante dinero
93 LdC La pianola de las alumnas resuenan mucho cuando tocan 93 LdC Las ideas de la vecina sorprende a su amigo 95 LdC La música de las trompetas desafina en la canción 96 LdC Las teteras de la cocina están hirviendo fuertemente 97 LdC La visita de los enfermos se aburren en el hogar 94 LdC Las ruedas del automóvil estalla por accidente 99 LdC La mujer de los negocios cancela el acuerdo 100 LdC Las señoras del fiordo encantan a los viajantes 101 LdC La mesa de los hermanos detienen el derrumbe 102 LdC Las redes del pescador rinde frutos maravillosos 103 LdC La mochila de los exploradores se extravía en el bosque 103 LdC Las cerraduras del cajón están fuera de uso 105 LdC El amante de las vecinas manejan por la ciudad 106 LdC Los soldados de la reina muere en la revuelta 107 LdC El campeón de las peleas sube de categoría hoy 104 LdC Los jefes de la bailarina permiten atrasos justificados 109 LdC El jardinero de las locas huyen hacia el campo 110 LdC Los ordenadores de la secretaria sufre de un virus 111 LdC El palacio de las suegras alberga a muchas personas 112 LdC Los puentes de la frontera evitan el paso de inmigrantes 113 HdS Los barcos casualmente navega a la costa 113 HdS El vendedor ágilmente negocian el precio 115 HdS Las empleadas básicamente pierden el empleo 116 HdS La modelo tranquilamente estrena los zapatos feos 117 HdS Los bares ocasionalmente abre sus puertas 114 HdS El bolso fácilmente engañan a los compradores 119 HdS Las haciendas desafortunadamente arden en el incendio 120 HdS La cerradura lógicamente abre la puerta 121 HdS Los pilares simplemente sostiene la estructura 122 HdS El ciego lamentablemente dependen de su familia 123 HdS Las enfermeras afortunadamente atienden en el hospital 123 HdS La profesora inicialmente castiga al alumno 125 HdS Los sombreros típicamente tapa las orejas 126 HdS El experimento raramente explotan en la sala 127 HdS Las pulseras a veces dañan su mano 124 HdS La idea afortunadamente incita el cambio 129 HdS Los presidente diariamente come en la mesa 130 HdS El juez extrañamente liberan al reo 131 HdS Las meseras siempre ganan más dinero 132 HdS La bicicleta claramente mejora el rendimiento 133 HdS Las cucharas discretamente derrama la comida 133 HdS El fuerte fácilmente defienden al pelotón 135 HdS Las granjas anualmente cosechan patatas premiadas 136 HdS La propiedad tradicionalmente pertenece a la ciudad 137 HdS Los osos imponentemente camina en el bosque 134 HdS El gato relajadamente miran el atardecer 139 HdS Las hijas tranquilamente juegan en la sala 130 HdS La perra indiscutidamente come en el jardín
