INFORME

The Reading Signatures of Agreement Attraction

Sol Lago1

, Carlos Acuña Fariña2, and Enrique Meseguer3

1Goethe University Frankfurt
2University of Santiago de Compostela
3University of La Laguna

Palabras clave: agreement attraction, comprensión, cue-based retrieval, similarity-based interference,
Español

un acceso abierto

diario

ABSTRACTO

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

/

.

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

The comprehension of subject-verb agreement shows “attraction effects,” which reveal that
number computations can be derailed by nouns that are grammatically unlicensed to control
agreement with a verb. Sin embargo, previous results are mixed regarding whether attraction
affects the processing of grammatical and ungrammatical sentences alike. In a large-sample
eye-tracking replication of Lago et al. (2015), we support this “grammaticality asymmetry”
by showing that the reading profiles associated with attraction depend on sentence
grammaticality. In ungrammatical sentences, attraction affected both fixation durations and
regressive eye-movements at the critical disagreeing verb. Mientras tanto, both grammatical and
ungrammatical sentences showed effects of the attractor noun number prior to the verb, en el
primero- and second-pass reading of the subject phrase. This contrast suggests that attraction
effects in comprehension have at least two different sources: the first reflects verb-triggered
processes that operate mainly in ungrammatical sentences. The second source reflects
difficulties in the encoding of the subject phrase, which disturb comprehension in both
grammatical and ungrammatical sentences.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

INTRODUCCIÓN

Over the past 30 years agreement has been a popular topic in the literature on syntactic process-
En g. This research started with a focus on the agreement errors made in sentence production
(Bock & Molinero, 1991) and later addressed how these errors are processed in comprehension
(Nicol et al., 1997). The interest in agreement is well-founded because it displays many features
that make language processing an exciting research area. Primero, agreement often capitalizes on
the existence of a morphological component, whose productivity is known to differ crosslinguis-
tically. Por ejemplo, whereas the English phrase “all the white broken dishes” contains one mor-
phological plural marker, there are five in its Spanish translation (“todos los platos blancos
rotos”). Más, agreement can occur between nonadjacent elements, thus implicating the
use of working memory (Luis & Vasishth, 2005). Finalmente, agreement can be studied in produc-
ción (where processing goes from meaning to form) or comprehension (where form precedes
significado), and it can be evaluated with different methodologies: judgment tasks, self-paced
lectura, event-related potentials. Here we use the eye-tracking method to establish the reading
signatures of the comprehension of agreement in Spanish, a language with previous conflicting
findings.

Citación: Lago, S., Acuña Fariña, C., &
Meseguer, mi. (2021). The Reading
Signatures of Agreement Attraction.
Mente abierta: Discoveries in Cognitive
Ciencia, 5, 132–153. https://doi.org/10
.1162/opmi_a_00047

DOI:
https://doi.org/10.1162/opmi_a_00047

Materiales suplementarios:
https://osf.io/f8bt7/

Recibió: 14 Enero 2021
Aceptado: 5 Agosto 2021

Conflicto de intereses:
The authors declare no conflict
de interés.

Autor correspondiente:
Sol Lago
sollago@em.uni-frankfurt.de

Derechos de autor: © 2021
Instituto de Tecnología de Massachusetts
Publicado bajo Creative Commons
Atribución 4.0 Internacional
(CC POR 4.0) licencia

La prensa del MIT

The Reading of Agreement Attraction

Lago et al.

In production, subject-verb agreement errors sometimes occur when a noun phrase con-
tains a plural modifier: speakers are more likely to accidentally produce a plural verb in sen-
tences like (1b) than (1a), consistent with the idea that the plural modifier “attracts” the verb,
misleading agreement computations. These so-called attraction errors are crosslinguistically
robust in production, having been attested in languages like English, Español ( Vigliocco
et al., 1996), italiano ( Vigliocco et al., 1995), Francés (Franck et al., 2002), Dutch (Hartsuiker
et al., 2003), and Russian (Lorimor et al., 2008), among others.

(1)

a. *The key to the cabinet are rusty from many years of disuse.
b. *The key to the cabinets are rusty from many years of disuse.

A dominant model of these errors, known as “marking and morphing,” was originally de-
veloped as a production model (Eberhard et al., 2005). Marking and morphing maintains that
attraction arises during the computation of the number of the subject phrase. The plural feature
of the attractor noun cabinets affects this computation and renders it more likely to be plural in
(1b) than in (1a). On this account, the subject phrase number is misconstrued, and the verb
merely receives the result of the previous miscalculation (Bock & Cutting, 1992; Bock &
Eberhard, 1993; Bock & Molinero, 1991; Eberhard, 1997; Vigliocco et al., 1996; Vigliocco &
Nicol, 1998). One strength of the marking and morphing model is that it can capture a com-
monly observed pattern known as the number markedness effect: number mismatch effects
tend to be stronger when the head noun is singular and the attractor plural (as in 1b) than
in the reversed arrangement (p.ej., “The keys to the cabinet”). According to the model, the lack
of a morphological singular marker reduces the contribution of the attractor to the subject
number computation.

Mientras tanto, comprehension studies are more recent and originally sought to evaluate
whether the configurations responsible for attraction in production elicited processing differ-
ences in comprehension. But an important advantage of comprehension studies is that they
make it possible to explicitly manipulate the number of the verb triggering an agreement com-
putation: Whereas verb number is decided by participants in production, in comprehension
researchers can manipulate whether a plural or a singular verb appears after the subject
phrase. Por lo tanto, comprehension designs include not only ungrammatical sentences such
como (1a–b) but also their grammatical counterparts (2a–b), in which the auxiliary verb is singular
and agrees in number with the subject head:

(2)

a. The key to the cabinet is rusty from many years of disuse.
b. The key to the cabinets is rusty from many years of disuse.

A clear prediction of the marking and morphing model is that if attraction occurs in com-
prehension, it should affect both grammatical and ungrammatical sentences. This is because
both versions—and thus subject number computations—are identical prior to the (y)gram-
matical verb. Fundamentalmente, the marking and morphing model predicts that attraction should yield
opposite effects depending on grammaticality. A plural attractor should facilitate the process-
ing of the ungrammatical verb: If participants are more likely to misconstrue the subject phrase
as plural in (1b), they should sometimes fail to notice the agreement violation at the plural
auxiliary in (1b) compared to (1a), which has an unambiguously singular subject.

Por el contrario, a marking and morphing account predicts that a plural attractor should disrupt
the comprehension of a grammatical verb. If participants are more likely to misconstrue the
subject phrase as plural in (2b), they should mistake the singular auxiliary for an agreement
violation. This should result in more processing difficulty in (2b) than in (2a), which contains

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

133

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

.

/

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

/

.

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

an unambiguously singular subject. En breve, the marking and morphing model predicts that a
plural attractor should disrupt processing in a grammatical sentence like (1b) but facilitate it in
an ungrammatical sentence like (2b).

An alternative account, cue-based retrieval, makes different predictions (Engelmann et al.,
2019; Jäger et al., 2017; Jäger et al., 2020; Luis & Vasishth, 2005; for a recent overview, ver
Vasishth et al., 2019). While there are several computational implementations of cue-based re-
trieval, they all propose that attraction does not result from a faulty encoding of the subject
phrase, but rather from similarity-based interference during the retrieval process initiated by a
verb. When a verb is encountered, speakers use the syntactic, semantic, and morphological fea-
tures of the verb as cues to retrieve an appropriate number controller from memory. Memoria
chunks corresponding to preceding items in the sentence are queried in parallel, and the chunk
with the most features matching the cues of the verb is the most likely to be retrieved.

In an ungrammatical sentence like (1b), interference at retrieval is high if there is a plural
attractor, because the subject head matches the syntactic cues of the verb but the plural at-
tractor matches the number cue. Due to this partial match, the plural attractor is sometimes
misretrieved, allowing comprehenders to license the verb in number and resulting in process-
ing facilitation: This effect is known as facilitatory interference. Por el contrario, interference in
grammatical sentences is low with a plural attractor, because the subject head fully matches
the verb in syntactic and number cues while the attractor does not. Por el contrario, retrieval in-
terference in grammatical sentences is higher with a singular attractor, as the number overlap
between both nouns increases competition at retrieval: This effect is known as inhibitory in-
terference. En breve, retrieval-based accounts predict that plural attractors should facilitate pro-
cessing in both grammatical and ungrammatical sentences like (1b) y (2b).

To summarize, both marking and morphing and retrieval-based accounts predict that plural
attractors should ease the processing of ungrammatical verbs. Sin embargo, their predictions differ
for grammatical sentences. Marking and morphing predicts that plural attractors should cor-
rupt the representation of the subject phrase and disrupt the processing of grammatical verbs.
Por el contrario, retrieval-based accounts predict that plural attractors should reduce interference
at retrieval and facilitate the processing of grammatical verbs. De este modo, the study of attraction
effects in grammatical and ungrammatical sentences is critical to arbitrate between different
accounts of attraction. Which of them is more consistent with previous results?

Attraction in Comprehension

Desafortunadamente, attraction effects in comprehension have a mixed empirical record. The text
below summarizes four contested empirical patterns related to potential differences between
grammatical and ungrammatical sentences. Our study was an attempt to clarify these patterns
through the use of a large participant sample and a paradigm—reading eye-tracking—with
good temporal resolution.

First Issue: The Grammaticality Asymmetry. Early work supported the predictions of the marking
and morphing account (Nicol et al., 1997; Pearlmutter et al., 1999). Por ejemplo, in two self-
paced reading and one eye-tracking study, Pearlmutter and colleagues (1999) showed that a
plural attractor created difficulty in grammatical sentences but eased difficulty in ungrammatical
unos. These effects appeared sometimes at the verb but more often at the word following it.
When ungrammatical verbs were preceded by plural attractors, participants showed shorter total
reading times and fewer first-pass regressions than when verbs were preceded by singular attrac-
tores. But after grammatical verbs, the opposite was found, with longer total reading times and

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

134

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

/

.

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

/

.

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

more regressions with plural attractors. Por lo tanto, the authors concluded that the plural attractor
occasionally caused the subject to be misrepresented as plural in both production and compre-
hension. In the literature since then, accounts placing the origin of attraction in the equivocal
representation of the subject phrase are known as representational accounts (Eberhard et al.,
2005; Franck et al., 2002; Hammerly et al., 2019; Hartsuiker et al., 2001; Staub, 2009).

Sin embargo, while later studies largely replicated attraction effects in ungrammatical sentences,
they raised questions about their existence in grammatical sentences. Wagers and colleagues
(2009) were the first to reason that in grammatical sentences such as “The key to the cabinets
is …” there is a danger of misattributing attraction to an unrelated factor: the difficulty of
processing a plural noun per se. Plurals are morphologically and semantically more complex
than singulars, and noun number influences processing times (Lau et al., 2007; Lehtonen &
Laine, 2003; New et al., 2004). Por lo tanto, processing difficulty in grammatical sentences may
reflect the downstream effects of having previously encountered the plural noun “cabinets,"
rather than attraction. This confound is mitigated in ungrammatical sentences because attrac-
tion there emerges as faster reading times at the verb after a plural attractor, which is the
opposite than expected from a spillover effects of noun plurality.

Wagers and colleagues (2009) addressed the potential confound in two ways. Primero, ellos
incorporated another word after the plural noun, in order to decrease the cost of attractor plu-
rality prior to the verb, Por ejemplo, “The key to the cabinets unsurprisingly was/were … .”
Segundo, they tested for attraction using object relative clauses (RC), in which the attractor did
not linearly intervene between the RC subject and the verb, Por ejemplo, “The cabinets that
the key opens/open … .” In several self-paced reading studies, they observed that attraction in
grammatical sentences was absent when either an adverb was added or a non-intervening
attractor was used. En tono rimbombante, attraction effects still occurred in ungrammatical sentences.
Por lo tanto, it was proposed that in comprehension attraction occurred only in ungrammatical
oraciones, with little or no effect in grammatical ones. This pattern is known as the “grammat-
icality asymmetry.” In the factorial 2 × 2 designs often used in comprehension (resulting in the
four conditions 1a–b and 2a–b above), the asymmetry is operationalized as an interaction be-
tween the factors grammaticality and attractor number, which demonstrates quantitatively dif-
ferent attraction effects in grammatical vs. ungrammatical sentences.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

.

/

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

Actualmente, the empirical status of the grammaticality asymmetry is unclear, as shown by a
recent review by Hammerly et al. (2019). Por un lado, studies that have measured num-
ber attraction effects in both grammatical and ungrammatical sentences have indeed found
significant effects more often in ungrammatical sentences: 91% vs. 38%, respectivamente.
Sin embargo, solo 60% del 45 studies that tested the grammaticality asymmetry found a sig-
nificant interaction between grammaticality and attractor number, which is the statistical dem-
onstration of the asymmetry (for discussion about the need of interactions to support claims
about asymmetric effects, see Gelman & Stern, 2006; Nieuwenhuis et al., 2011).

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Second Issue: Attraction in Grammatical Sentences. A second issue is that even when attraction is
observed in grammatical sentences, its direction is unclear. Consistent with Pearlmutter et al.
(1999), some studies have reported processing difficulty, consistent with representational ac-
cuenta (Acuña-Fariña et al., 2014; Nicol et al., 1997; Patson & Husband, 2016). But many
studies have found the opposite: processing facilitation (Franck et al., 2015; Nicenboim
et al., 2018; Smith et al., 2021: experimento 1; Villata et al., 2018).

Por ejemplo, Villata and colleagues (2018) conducted a self-paced reading study with 130
English participants. Their main analysis failed to show attraction effects online, pero un

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

135

The Reading of Agreement Attraction

Lago et al.

supplementary analysis that included more trials (containing reading times up to 8,000 EM)
found a marginal facilitatory attraction effect after the verb in grammatical sentences with
plural attractors. These findings align with those of Nicenboim et al. (2018), who observed
processing facilitation in grammatical sentences with plural attractors: in a sample of 184
German participants, the mean size of the effect at the verb was 9 ms and the range of effect
sizes deemed likely with 95% probability was [0, 18] ms—this is known as a 95% Bayesian
credible interval. This modest effect size is consistent with the estimate from a meta-analysis by
Jäger et al. (2017): 13 [2, 28] EM. De este modo, large participant samples seem necessary to detect
attraction in grammatical sentences, although this is still uncommon in most studies on agree-
mento. Our large-sample study seeks to address this limitation.

En años recientes, several researchers have suggested that read-
Third Issue: Encoding Interference.
ing time effects previously attributed to the retrieval triggered by a verb may instead be due to
a different cognitive process: encoding interference (Ness & Meltzer-Asscher, 2017; Herrero
et al., 2021; Villata et al., 2018; Villata & Franck, 2020). Encoding interference arises when
two elements share similar features, which degrades their distinctiveness in memory (Oberauer
& Kliegl, 2006). Encoding interference reflects difficulty in the initial memory encoding of
elementos, and thus it occurs prior to their retrieval by elements appearing later in a sentence,
Por ejemplo, verbos. Por lo tanto, encoding interference effects should be observable prior to
the appearance of a verb (in number attraction studies) or in configurations in which the
manipulated feature(s) is not a retrieval cue.

The evidence for these two predictions is scarce and inconsistent in the attraction literature
( Jäger et al., 2015; parker & Konrad, 2020), but it is widespread in other grammatical
domains.1 For instance, Gordon and colleagues (2002) reported that the comprehension of
object relative clauses was facilitated when the subject and the object were of different types,
Por ejemplo, a definite description and a pronoun vs. two definite descriptions: “The barber
that you / the lawyer admired.” The facilitation due to the pronoun was observed both at the
second noun phrase and at the relative clause verb. While the effect at the pronoun itself could
be due to lexical variables (pronouns are shorter and more frequent than definite descriptions),
the reading times at the verb more strongly suggest an encoding effect, as the distinction
between noun types is unlikely to be a retrieval cue (for similar findings see Barker et al.,
2001; Fedorenko et al., 2006; Gordon et al., 2002; Hofmeister & Vasishth, 2014; Jäger
et al., 2015; Kush et al., 2015).

In agreement attraction studies, some evidence of encoding interference has been provided
by Villata et al. (2018). In addition to the experiment reported above, the authors conducted a
second self-paced reading experiment that manipulated the gender (instead of the number) de
attractor and target nouns in Italian object RCs, Por ejemplo, “The ballerinaFEM [that the
waiterMASC / the waitressFEM has surprised] … .” Results from 167 Italian participants showed
that the participle verb was read more quickly when the attractor and target noun bore differ-
ent gender, as opposed to when they matched. The surprising aspect of this result is that past
participles in Italian object RCs do not instantiate gender agreement. Por lo tanto, gender could
not have possibly been used as a retrieval cue at the participle, disqualifying a retrieval-based
explanation and favoring an encoding interference-based one.

1 But note that number attraction effects that are consistent with encoding interference in grammatical sen-
tences have been attested in question comprehension measures in adult and developmental studies in German,
Swedish, hebreo, and Italian (Adani et al., 2010, 2014; Belletti et al., 2012; Jäger et al., 2015; Villata & Franck,
2020). Sin embargo, since attraction effects were only probed with offline measures, we do not discuss them in the
main text because we focus on processing effects.

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

136

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

/

.

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

Encoding interference effects in attraction cannot be explained by marking and morphing
or by retrieval-based accounts. But they can be explained by a different framework, self-
organized sentence processing (SOSP), which is the third account relevant for the present
estudiar (Kempen & Vosse, 1989; Smith et al., 2018). As the dynamics of SOSP are quite com-
complejo, here we offer only a brief summary focused on the two issues under discussion: attraction
effects in grammatical sentences and the grammaticality asymmetry (for a detailed account,
see Smith et al., 2021). In SOSP, encountering each word activates a treelet with semantic
and syntactic features, similar to memory chunks in cue-based retrieval (p.ej., +notario público, −plural).
Activated treelets attempt to combine in all possible ways with other activated treelets, y el
strength of these connections grows over time. Fundamentalmente, and in contrast with cue-based
retrieval, these interactions happen continuously, rather than only at retrieval. This allows
SOSP to capture encoding interference effects.

Por ejemplo, in the preamble “the key to the cabinets,” after the singular head noun “key”
is encountered, the subject node of the sentence initially gravitates toward [+notario público, −plural]
(note that gravitation is a gradual process in which features increase their values from 0 toward
a maximum of 1; for a useful visualization see Villata et al., 2018, cifra 3). When the attractor
is encountered later, its treelet competes with the “key” treelet to attach to the subject node. Si
the attractor is plural, competition is weak, because the attractor solely matches the subject
node in the +NP feature. But if the attractor is singular, it will match both features of the subject
nodo [+notario público, −plural], increasing its competition with the head noun for attachment and delay-
ing processing both prior and during the processing of the verb. The same competition mech-
anism operates in ungrammatical sentences, with the only difference that in this case the
subject node is never a perfect match for the verb, regardless of which treelet attaches to it.
De este modo, SOSP predicts facilitation due to a plural attractor in both grammatical and ungrammat-
ical sentences and it explains encoding and retrieval interference through the same
mechanism.

The feature-competition property of SOSP also allows it to explain the grammaticality
asymmetry. In grammatical sentences, all constraints align to form the correct parse, such that
the attractor only has a weak influence. Por el contrario, in ungrammatical sentences, it is impos-
sible to satisfy all the constraints and the system remains in a blended, intermediate state, con
a benefit for the plural attractor condition due to reduced competition in the encoding of the
subject phrase. En breve, SOSP predicts processing facilitation in both grammatical and
ungrammatical sentences as well as the grammaticality asymmetry, similarly to retrieval-based
accounts and in contrast with the marking and morphing model. But in contrast to cue-based
retrieval, SOSP can model encoding interference and explain attraction effects prior to the verb.
Sin embargo, studies to date have failed to find preverbal encoding interference effects, cual es
unexpected for SOSP. One potential explanation is that regions prior to the verb are seldom
analyzed in reading studies (and when they are, the results are confounded by the lexical effect
of attractor plurality, as shown by Wagers et al., 2009).

Fourth Issue: Differences Between Languages and Reading Paradigms. A final problem in evaluating
previous results lies in the scarcity and marked differences between studies. The review by
Hammerly et al. (2019) shows that only six languages have been tested to date, con
English comprising almost 90% of the evidence. Previous studies also range from self-paced
reading and acceptability judgments to eye-tracking and event-related potentials. Considering
that an important issue is the timing of verb-driven attraction effects vs. attractor-driven plu-
rality effects, it is unfortunate that few studies have used the eye-tracking method (Dillon et al.,
2013; Jäger et al., 2020; parker & Phillips, 2017; Pearlmutter et al., 1999). De estos, the only

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

137

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

.

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

non-English study was by Acuña-Fariña et al. (2014). Due to the speed and highly automatic
nature of eye-movements, eye-tracking is a useful tool for examining temporal profiles and it
also helps alleviate some concerns about response bias and task strategies that apply to judg-
ment and self-paced reading tasks (Hammerly et al., 2019; Villata & Franck, 2020).

The Current Study

Our work seeks to clarify the empirical status of attraction effects in comprehension. We focus
on number attraction in Spanish, a language in which previous results are mixed and based on
only two studies. Acuña-Fariña and colleagues (2014) used eye-tracking to measure the pro-
cessing of prepositional phrase constructions in grammatical sentences such as “The name of
the boy(s) was German” (attractor bolded). The results were similar to those of Pearlmutter
et al. (1999), but attraction appeared immediately at the verb, which elicited more regressions
and longer regression-path and total reading times when following a plural attractor.
“Regression-path time” describes the reading times to a word prior to it being exited to the
bien, including the rereading of prior words.

In a later study, Lago and colleagues (2015) conducted four self-paced reading experiments
that examined attraction effects in grammatical and ungrammatical sentences. They used
object relative clauses such as “The note(s) [RC that the girl writes/*write in the class] … .”
As mentioned, object RCs avoid the confounding effects of plural morphology, because the
plural attractor doesn’t immediately precede the verb and thus it is less likely to influence its
lectura. All experiments showed processing facilitation after ungrammatical verbs. Sin embargo,
the results in grammatical sentences were inconsistent: while no attraction was seen in one
experimento, which used thematic verbs such as “write,” processing difficulty was observed
with auxiliary verbs (“is going to write”): grammatical sentences with a plural attractor elicited
longer reading times than those with singular attractors, consistent with representational
accounts and with Acuña-Fariña et al. (2014). An additional experiment was performed to
follow up on this effect, but it failed to replicate the attraction effect in grammatical sentences.

Tomados juntos, these findings offer conflicting answers about the existence and direction-
ality of attraction effects in Spanish grammatical sentences. Por un lado, the attraction
effect in the study of Acuña-Fariña et al. (2014) may be attributed to the spillover effect of the
attractor plurality, which may have confounded reading times at the verb. Por otro lado,
Lago et al. (2015) may have failed to observe this effect due to the limited temporal resolution
of the self-paced reading method, which does not allow participants to reread prior material.
This precludes the measurement of regressive eye movements and rereading patterns, two key
markers of attraction in eye-tracking studies.

To address these issues, we revisited the grammaticality asymmetry in Spanish and sought
to equalize testing conditions by using the same language, the same methodology, y el
same design and linguistic constructions. We also recruited the largest number of participants
that was feasible given our resources, motivated by previous findings that attraction effects in
grammatical sentences are small and may require large samples to be detectable. We used the
non-intervening configurations of Lago et al. (2015) to minimize spillover effects due to attrac-
tor plurality, but we employed an eye-tracking paradigm like Acuña-Fariña et al. (2014) a
maximize the temporal resolution of our measures. We reasoned that eye-tracking should
allow us to more comprehensively test for attraction in grammatical sentences, while repli-
cating facilitation effects in ungrammatical ones. Más, the use of non-intervening config-
urations allowed us to compare reading profiles at the region before the critical verb, the RC
subject region.

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

138

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

.

/

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

/

.

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

EYE-TRACKING EXPERIMENT

Our study was a large-sample eye-tracking replication of the first experiment of Lago et al.
(2015). The materials featured grammatical and ungrammatical sentences with a non-
intervening attractor noun (p.ej., “the notes”). An example is shown in (3) with the regions
of analysis marked by vertical bars. These regions comprised the relative clause verb and a
following 2–3-word adverbial (p.ej., “wrote in the class,” henceforth, verb/spillover region).
The region preceding the verb, consisting of the RC subject, was also analyzed to probe for
encoding interference effects (p.ej., “la chica,” RC subject region). This region was identical
across conditions, thus avoiding lexical differences between conditions due to noun plurality
( Wagers et al., 2009).

(3)

a. GRAMMATICAL, SINGULAR ATTRACTOR

La nota que | la chica | escribió | en la clase | alegró a su amiga.

b. GRAMMATICAL, PLURAL ATTRACTOR

Las notas que | la chica | escribió | en la clase | alegraron a su amiga.

C. UNGRAMMATICAL, SINGULAR ATTRACTOR

*La nota que | la chica | escribieron | en la clase | alegró a su amiga.

d. UNGRAMMATICAL, PLURAL ATTRACTOR

*Las notas que | la chica | escribieron | en la clase | alegraron a su amiga.

Translation: The note(s) that the girl wrote.SG / *wrote.PL in the class cheered up her
amigo.

Based on the attraction literature, we expected plural attractors to ease the processing of un-
grammatical verbs, as reported by Lago et al. (2015) in Spanish and as predicted by SOSP, y
representational and retrieval-based accounts. Our goal was to properly assess the magnitude
of attraction effects in reading using a large participant sample and restricting our analysis to
those eye-tracking measures independently supported by previous studies (see Analysis).

Además, we sought to clarify the empirical record regarding the grammaticality asym-
metry and the direction of attraction effects in grammatical sentences. If the grammaticality
asymmetry holds in reading comprehension, we should obtain a statistical interaction between
the number of the attractor noun and the grammaticality of the sentence ( Wagers et al., 2009).
Mientras tanto, attraction in grammatical sentences is critical to arbitrate between different ac-
cuenta. According to representational accounts, the plural attractor should disrupt the reading
of the grammatical verb in (3b). According to retrieval-based accounts and SOSP, attraction
should facilitate it. We note that due to the different perspectives adopted by these accounts,
attraction effects in grammatical sentences are often reported inconsistently in the literature.
Studies within representational and SOSP approaches treat the plural attractor condition as the
condition of interest and compare it against the singular condition (which functions as a base-
line condition). Por el contrario, retrieval-based studies treat the plural attractor condition as a
baseline because their focus is on the singular attractor condition, in which similarity-based
interference is expected in grammatical sentences. To avoid ambiguity, here we consistently
take the plural attractor condition as the condition of interest in both grammatical and ungram-
matical sentences, always using the singular attractor conditions as baseline (Dillon et al.,
2013; Hammerly et al., 2019; Villata et al., 2018).

Finalmente, we assessed whether encoding interference contribute to attraction effects. El
ability to explain encoding interference separates SOSP from representational and cue-based
retrieval accounts. According to encoding interference one should be able to measure

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

139

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

/

.

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

processing differences due to the attractor number prior to the verb. Sin embargo, such effects
have not been consistently found in online attraction studies, despite being attested in
offline/untimed comprehension measures ( Jäger et al., 2015; Villata et al., 2018) and in the
comprehension of other linguistic phenomena (p.ej., Gordon et al., 2004; Kush et al., 2015).
Our study was optimized to detect encoding interference because the preverbal region was
identical across conditions (thus avoiding a confound with noun plurality) and because the
eye-tracking paradigm allowed measuring not only fixations but also regressive eye move-
mentos, a key diagnostic of attraction in previous studies.

Métodos

Participantes. Eye-movement data from 160 native speakers of Spanish were collected at the
University of La Laguna in Tenerife, España. All participants had normal or corrected-to-normal
visión. Due to experimenter error, their chronological age was not stored, but all participants
were university students. Participants provided informed consent and received course credit or
payment for their participation. All procedures were in accordance with the Declaration of
Helsinki.

Materials. Materials were taken from Lago et al. (2015). Some of them were modified in order
to avoid uncommon expressions in Castilian Spanish like “lapicera” (‘pencil’), “aplicante”
(‘applicant’), “reporte” (‘report’), “manejó” (‘drove’), and “fallas” (‘faults’), as well as Latin
American geographical or cultural references, such as ‘the war of Paraguay,’ ‘the ancient
Mayas,’ or ‘the Chilean border.’ The sentences were shortened to 80 characters when neces-
sary to make them fit in one line of the display monitor.

The experimental sentences consisted of 48 sentence sets arranged in a 2 × 2 dentro-
subjects design, with grammaticality (grammatical/ungrammatical) and attractor number (sin-
gular/plural) as factors. In the grammatical conditions, the relative clause (RC) subject and verb
were both singular (es decir., they agreed in number), while in the ungrammatical conditions the RC
subject was singular but the verb was plural. The RC verb was always in the simple past tense
and perfective aspect. The singular suffix for this tense-aspect combination is one character
largo (p.ej., “escribi-ó,” ‘write.3sg’), while the plural suffix is 4 characters long (p.ej., “escribi-
eron,” ‘write.3pl’). The attractor noun was always inanimate while the RC subject head was
always animate.

El 48 sentence sets were distributed across four lists in a Latin square design, and were
combined with 104 sentences from a different experiment (not reported here). These filler
sentences were all grammatical and consisted of simple subject-verb-object-adverbial sen-
tences like “The priest found the bishop in the morning.” This resulted in 15.8% del
items being ungrammatical. Materials, datos, and analysis code are available at https://osf
.io/f8bt7/.

Procedimiento. Eye movements were recorded by a video-based Eyelink 1000 Plus sampling at
1000 Hz (SR Research). Sentences were presented in lowercase on a monitor that displayed
hasta 80 characters per line. When necessary, line breaks occurred after the critical region.
Participants were seated 73 cm away from the monitor, and three characters equaled 1 degree
of visual angle. Viewing was binocular but only the right eye was recorded.

Experimental and filler items were followed by yes/no comprehension questions to ensure
that participants were attentive. In the experimental items, the questions never referred to the
agreement dependency to avoid focusing participants’ attention on it. De este modo, response accuracy
was not a measure of interest in the experimental items. Each session began with a calibration

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

140

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

.

/

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

on a 9-point grid and recalibration between trials was conducted if necessary. Participantes
were instructed to read at a natural pace and answer the comprehension questions. They were
not informed that some sentences would contain grammatical errors. Three practice items
were presented. Before each trial, participants fixated on a marker above the first word of
the upcoming sentence. Upon fixation on this marker, the text appeared. The order of exper-
imental and filler items was randomized for each list. An entire experimental session lasted
approximately 30–45 min.

Análisis. Eye-tracking data were processed as follows. Fixations equal or shorter than 80 EM
within one character of another fixation were merged. We also merged fixations shorter than
40 ms falling within three characters of another fixation. All remaining fixations below 80 EM
o superior 800 ms were removed. Trials in which a region was skipped in initial reading were
treated as missing data for that region.

Our analyses focused only on measures consistently found to show attraction in previous
estudios, to minimize the number of multiple comparisons, which can increase Type I error
rates in eye-tracking (Godfroid & Hui, 2020; von der Malsburg & Angele, 2017). Within
frequentists frameworks, this issue can be addressed by applying corrections, como el
Bonferroni correction (Bonferroni, 1936). This is not possible in a Bayesian framework—
the framework used here—because the Bayesian approach is concerned with quantifying
the strength of evidence for a hypothesis given its prior probability, regardless of the number
of hypotheses under investigation (for discussion see Dienes, 2011). From a frequentist per-
perspectiva, this may render the results anticonservative and, de este modo, they should be interpreted
with caution.

Based on previous studies, we analyzed the probability of first-pass regressions (Dillon
et al., 2013; Jäger et al., 2020; Pearlmutter et al., 1999), regression-path (Acuña-Fariña
et al., 2014; parker & Phillips, 2017), and total reading time (Acuña-Fariña et al., 2014;
Dillon et al., 2013; Jäger et al., 2020; parker & Phillips, 2017; Pearlmutter et al., 1999). El
probability of first-pass regressions (also known as “regressions out”) is an early processing
measure denoting the probability of initiating a regression when first encountering a region.
Por el contrario, total time is a global measure that denotes the whole amount of time spent in a
región, including rereading. Finalmente, regression-path time (also called “cumulative reading
time” or “go-past time”) describes the amount of time spent since first entering a region from
the left until leaving it to the right (including regressions to preceding regions).

Our critical region comprised the verb and spillover region, as shown in example (3).
Following Cunnings and Sturt (2018), we analyzed these regions jointly and included
REGION as a fixed effect. This procedure helps minimize the number of statistical tests per re-
gion, it allows evaluating potential timecourse differences between regions and it increases
power to observe small effects that may be nonsignificant at individual regions but neverthe-
less consistent across them. Sin embargo, one drawback of the procedure is that pooling the
regions in the same model assumes equal variances in both regions. This is unlikely to be true
in our materials, in which the verb region comprised a single word, but the spillover region
comprised a 2–3-word adverbial. De este modo, we conducted an additional analysis of both regions
separately. The results of this analysis were largely similar to those of the joint analysis, con
the main difference concerning the first-pass regression measure (see the Supplemental
Materials).

In addition to the verb/spillover region, we examined the reading of the relative clause
sujeto, which appeared immediately before the verb: the RC subject region. If encoding

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

141

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

.

/

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

interference occurs, the effect of the attractor number may already show up in the reading of
the RC subject, prior to the appearance of the verb. Respectivamente, previous eye-tracking studies
have shown that the difficulty elicited by object RCs (in comparison to subject RCs) arises at
the noun phrase following the relative pronoun (Staub, 2010b; Staub et al., 2017). Para
instancia, Staub (2010b) showed that this noun phrase elicits more regressions and longer
regression-path times than the same phrase in a subject RC or in a complement clause.
Por lo tanto, we examined first-pass regressions, regression-path, and total times in the RC
subject region. We report this analysis in a separate section, Exploratory Analysis, porque
our decision to perform it was taken during the analysis stage, after data collection took place.

Reading data were analyzed with mixed-effects logistic regression (first-pass regression
measure) and mixed-effects linear regression (regression-path and total time measures). Todo
models were fit in a Bayesian framework using the brms package in R (Bürkner, 2017; R
Development Core Team, 2020). Bayesian models are valuable because they combine prior
information with evidence from the data in order to obtain a probability distribution over the
plausible values of a parameter—the parameter’s posterior distribution. De este modo, an experimental
effect can be quantified in terms of the likelihood of its possible values, which is more infor-
mative than a binary statement about whether the effect is significant, because it puts the focus
on determining an effect size and direction, along with its uncertainty (Cumming, 2014;
Kruschke & Liddell, 2018).

The procedure for fitting Bayesian models and assessing their convergence followed Jäger
et al. (2020). We used a hierarchical lognormal likelihood function to model the raw reading
times in milliseconds. This is equivalent to log-transforming the values and fitting a hierarchi-
cal linear model with a normal likelihood. All models assumed correlated varying intercepts
and slopes for items and participants for all predictors of interest and their interactions
(GRAMMATICALITY, ATTRACTOR NUMBER, and REGION; see Barr et al., 2013). Más, since there were
two non-independent datapoints from each trial (p.ej., one observation from the verb and one
from the spillover region), we included a random intercept for trial, defined as each unique
participant-item pairing.2

Two models were computed to address our research questions. The first model used fixed
effects of GRAMMATICALITY, ATTRACTOR NUMBER, and REGION, as well as their interaction. Its goal
was to examine the existence of attraction and of the grammaticality asymmetry, which should
appear as an interaction between grammaticality and attractor number. Since ungrammatical
verbs were systematically longer than grammatical verbs, the factor REGION LENGTH was added
to the models of the verb/spillover region (centered and operationalized as the number of char-
acters per region).

The second model assessed the nested effects of attraction. Its goal was to estimate the
effect of attraction in grammatical and ungrammatical sentences separately. Nested models
were run for the three reading measures in the combined verb/spillover region, and for total
times in the RC subject region. Separate nested models for grammatical and ungrammatical
sentences were not run for first-pass regressions and regression-path times at the RC subject

2 We also tried to include a by-region slope for the trial random effect. But this led to several problems, en-
cluding divergent transitions, an R-hat statistic higher than 1.01 and chains with a low estimated Bayesian
Fraction of Missing Information. Because the problem persisted after changing the optimization parameters
and doubling the number of iterations (Stan Development Team, 2020; https://mc-stan.org/misc/warnings
.html), we removed the by-region slopes for the models reported in this article.

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

142

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

/

.

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

/

.

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

because these measures index processing prior to the appearance of the verb, when all
conditions were grammatical. Por el contrario, total reading times include the second-pass reading
of the RC subject region after it was exited to the right, and thus they may be affected by the
grammaticality of the verb. In the nested model of the verb/spillover, REGION was always used
as a factor. In the nested model of the RC subject, which comprised a single identical region
across condition, neither REGION nor REGION LENGTH were used.

To avoid making strong a priori assumptions about possible effect sizes, we used mildly
informative priors (Gelman et al., 2014). Específicamente, we used a standard normal distribution
norte (0, 1) for all fixed effects except the intercept, which had an N (0, 10) previo. The prior for the
random effects and the residual variance used a half-normal distribution N+ (0, 1), porque
random effects and residual variance cannot be negative. Within the variance-covariance
matrices of the by-participant and by-item random effects, priors were defined for the corre-
lation matrices using a so-called Lewandowski-Kurowicka-Joe (LKJ) previo (Lewandowski et al.,
2009). This prior has a parameter η, cual, when set to 2, has the regularizing effect of dis-
favoring extreme correlations.

All contrasts were sum-coded as ± 0.5, such that the model parameters reflected differences
between condition means. For the factor grammaticality (−0.5 grammatical / +0.5 ungrammat-
ical), a positive coefficient reflects a slowdown in ungrammatical sentences, eso es, Procesando
ruptura. For attractor number (−0.5 singular / +0.5 plural), a negative coefficient reflects a
speedup for sentences with a plural attractor. For the grammaticality × attractor number inter-
acción, a negative coefficient shows stronger attraction effects in ungrammatical than gram-
matical sentences. For the factor region (−0.5 verb / +0.5 spillover), a positive coefficient
reflects longer reading times (or more regressions) in the spillover than in the verb region.

Resultados

Mean accuracy in the comprehension questions was 91.4% (DE = 28%). For the reading data,
we focus on the effects of interest but include the full output of the main models in the
Materiales suplementarios. The Supplemental Materials also contain the mean reading times
in all sentence regions, to provide a visual summary of how readers navigated the sentences
(Figure S1). We report the mean of the posterior probability distribution of each effect together
con un 95% credible interval (CrI), which represents the interval where it is 95% certain that
the true effect lies given the data and the model. For easier interpretability, reading measures
are back-transformed to their original scale in the text, but statistical analyses and inferences
are based on the untransformed model coefficients.

Verb and Spillover Region. The verb region was skipped on 2.7% of trials and the spillover
region on 0.56% of trials. The reading patterns were consistent across measures (Cifra 1).
All measures showed main effects of grammaticality and attraction (Cifra 2). Específicamente, y-
grammatical verbs disrupted processing by eliciting more regressions (posterior mean 6.1%,
95% CrI [4.5, 7.7] %) and longer regression-path (55 [44, 67] EM) and total times (56 [43,
69] EM). Mientras tanto, plural attractors facilitated processing by reducing regressions (−1.2
[−2.2, −0.3] %), regression-path (−13 [−18, −7] EM), and total times (−10 [−17, −4] EM).
Finalmente, the mean estimates of the grammaticality × attractor interaction were negative across
measures, consistent with larger attraction effects in ungrammatical than grammatical sen-
tenencias. Sin embargo, although the credible intervals were clustered on negative values, they also
included zero and a few positive values: −1.3 [−3.2, 0.5] % regressions, −9 [−21, 1] EM
regression-path, and −10 [−23, 4] ms total time. De este modo, these results are consistent with the

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

143

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

/

.

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

/

.

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

Cifra 1.
error bars 95% confidence intervals calculated for each condition across participants and items.

Empirical reading measures for the verb (“wrote.SG/PL

") and spillover region (“in the class”). Points show condition means and

claim of larger attraction effects in ungrammatical sentences—the grammaticality asymmetry—
but they do not conclusively support it.

The nested model assessed the size of attraction effects separately for ungrammatical and
grammatical verbs. Ungrammatical verbs showed facilitation due to attraction: verbs preceded
by plural attractors elicited fewer regressions (−1.9 [−3.1, −0.7] %) and faster regression-path
(−17 [−25, −9] EM) and total times (−15 [−25, −6] EM) than verbs preceded by singular attrac-
tores. Grammatical verbs showed similar facilitatory patterns, but attraction effects were incon-
clusive because they spanned negative and positive values: −0.5 [−2, 0.9] % regressions, −8
[−16, 0] ms regression-path and −5 [−14, 4] ms total time.

Exploratory Analysis: RC Subject Region. The RC subject was skipped on average on 2.27% de
ensayos. The earlier processing measures, first-pass regression and regression-path times, no lo hizo
show a grammaticality effect, as expected since all conditions were grammatical prior to the
verb. But the number of the attractor noun modulated regression-path times: when the attractor
was plural, the RC subject—which was always singular—elicited shorter regression-path
veces: −17 [−29, −5] EM. De este modo, reading was facilitated when the attractor and the RC subject
bore different numbers, consistent with encoding interference (Cifra 3). A similar pattern was
observed in first-pass regressions, with fewer regressions after a plural attractor: −2.7 [−4.8,
−0.8] %. Sin embargo, this effect should be taken with caution, because visual inspection of
Figure S1 (Materiales suplementarios) revealed a similar pattern already in the preceding region.
Por lo tanto, we can’t rule out that the encoding effect at the RC subject in first-pass regressions
was due to spillover from the previous region.

The RC subject showed a clear effect of grammaticality in total reading times (24 [11, 37] EM)
and also of the number of the attractor noun, with shorter total reading times when the attractor
and the target noun bore different numbers: −22 [−36, −8] EM. The nested comparisons showed
that the number mismatch between the attractor and target noun facilitated processing in un-
grammatical and grammatical sentences similarly: −19 [−36, −2] ms and −26 [−43, −8] EM,
respectivamente (Cifra 4).

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

144

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

/

.

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

/

.

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

.

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

Cifra 2. Posterior estimates of the statistical model for first-pass regressions, regression-path, and total time measures in the
verb/spillover region. Red circles show posterior means and thicker horizontal lines show 95% credible intervals (thinner lines show 99%
credible intervals). Dashed vertical lines correspond to an effect size of zero. For the factor grammaticality, positive estimates reflect longer
reading times (or more regressions) in ungrammatical sentences. For the attractor factor, negative estimates reflect faster reading times (or fewer
regressions) in plural attractor sentences. For the interaction, negative estimates suggest stronger attraction effects in ungrammatical than gram-
matical sentences.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

GENERAL DISCUSSION

Our study examined the reading profiles of subject-verb agreement attraction in Spanish. Nosotros
were interested in whether attraction effects would differentially disrupt the comprehension of
grammatical and ungrammatical sentences, giving rise to the so-called grammaticality asym-
metry. In Spanish, a nuestro conocimiento, only two studies have examined the reading profiles of
number attraction, and only one of them probed for the grammaticality asymmetry. But their
direct comparison is challenging because they used different designs, grammatical construc-
ciones, and methods. Our study sought to equalize testing conditions by deploying the same
design and linguistic constructions as Lago et al. (2015) but using eye-tracking and a large
participant sample to increase the sensitivity of the measurements.

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

145

The Reading of Agreement Attraction

Lago et al.

Cifra 3. Empirical reading measures for the relative clause subject (“the girl”). Points show condition means and error bars 95% confi-
dence intervals calculated for each condition across participants and items.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

.

/

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Cifra 4. Posterior estimates of the statistical model for first-pass regressions, regression-path, and total time measures in the
verb/spillover region. Red circles show posterior means and thicker horizontal lines show 95% credible intervals (thinner lines show 99%
credible intervals). Dashed vertical lines correspond to an effect size of zero. For the factor grammaticality, positive estimates reflect longer
reading times (or more regressions) in ungrammatical sentences. For the attractor factor, negative estimates reflect faster reading times (or fewer
regressions) in plural attractor sentences. For the interaction, negative estimates suggest stronger attraction effects in ungrammatical than gram-
matical sentences.

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

146

The Reading of Agreement Attraction

Lago et al.

Our results can be summarized as follows. Primero, we replicated the finding that attraction
eases the reading of ungrammatical verbs, as predicted by all the accounts of attraction con-
sidered here: SOSP, representational, and retrieval-based accounts. Facilitatory effects were
observed in all reading measures: first-pass regressions, reading-path, and total times. Estos
attraction effects were small. For comparison, the disruption due to an ungrammatical verb (como
opposed to a grammatical one) increased regression-path and total reading times 50 ms on
average and the regression probability about 6%. Mientras tanto, attraction-induced facilitation
in ungrammatical sentences was on average 13 ms in reading-path times, 10 ms in total times,
and less than 2% in first-pass regressions. These modest sizes are not unexpected: Large
sample studies can provide more realistic estimates of an effect size, as significant effects
in low-powered studies are often overestimates ( Vasishth et al., 2018).

Our remaining results concern the empirical effects with a mixed record in previous studies:
the grammaticality asymmetry and attraction effects in grammatical sentences. The direction of
attraction effects in grammatical sentences is crucial to arbitrate between processing models,
because both SOSP and retrieval-based predict facilitation with grammatical verbs, mientras
representational accounts predict processing difficulty. Desafortunadamente, our analyses failed to
detect conclusive effects of attraction in grammatical sentences in the verb/spillover region.
Sin embargo, the mean estimates of the effects, as well as their credible intervals, consistently
suggested facilitation, with faster reading times and fewer regressions in the plural attractor
condición.

Based on these reading profiles, our view is that our results align with those of large-sample
studies and of a meta-analysis ( Jäger et al., 2017; Nicenboim et al., 2018; Villata et al., 2018)
and suggest that attraction in grammatical sentences is unlikely to emerge as processing diffi-
culty during reading. Under this view, the unexpected results in previous studies are those that
showed processing difficulty in grammatical sentences (Acuña-Fariña et al., 2014; Nicol et al.,
1997; Pearlmutter et al., 1999; Patson & Husband, 2016). Since all these studies used prepo-
sitional phrase constructions like “The key to the cabinet(s) is …,” a likely explanation is that
the spillover effects of noun plurality created the processing difficulty measured at the gram-
matical verb ( Wagers et al., 2009). We mitigated this confound by testing object RCs, en el cual
the plural attractor did not immediately precede the verb. We believe that object RCs provide
a better way to diagnose attraction in grammatical sentences, but we acknowledge an alter-
native proposal that attraction effects in relative clauses may obey a different mechanism,
so-called “predication confusion” (Staub, 2009, 2010a). While we cannot rule out this possi-
habilidad, under this proposal it is surprising that prepositional phrases and object RCs often show
the same attraction effects. Experiments using other constructions (or languages in which
modifiers linearly precede their heads) may help resolve this issue.

Encoding Interference in Agreement Attraction

The novel finding from our study is that participants read the RC subject more easily when it
differed in number with the preceding attractor noun, and that this effect emerged prior to the
critical verb. In the plural attractor condition, regression-path times for the singular RC subject
were reduced on average 17 ms and regressions decreased about 3%. Total reading times,
which incorporate refixations to the RC subject after encountering the verb, showed equiva-
lent reductions in ungrammatical and grammatical sentences: on average 20 y 25 EM,
respectivamente. Fundamentalmente, facilitation effects were similarly sized in grammatical and ungrammat-
ical sentences at the RC subject, but they were numerically larger for ungrammatical sentences
at the verb/spillover region. This suggests that the two effects are unlikely to reflect the same

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

147

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

.

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

proceso. While the grammaticality asymmetry at the verb/spillover region is expected under
retrieval-based accounts, the symmetric effect at the RC subject requires an additional
explicación.

Regressions and regression-path time measures at the RC subject reflect processing prior to
the verb, and so the effect at the RC subject cannot be explained by retrieval-based accounts.
Sin embargo, one explanation that would still be consistent with a retrieval-based account is that
the effect at the RC subject reflected a parafoveal-on-foveal effect, such that the parafoveal
processing of the RC verb was complete enough to trigger the retrieval of an object and affect
fixations durations at the RC subject. We think that this is unlikely for two reasons. Primero, bajo
such an explanation—without additional assumptions—we should have also found a gram-
maticality effect at the RC subject. Segundo, the existence of parafoveal-on-foveal effects in
experimental reading paradigms is uncertain (Schotter, 2018). Most of the supporting evidence
for these effects comes from corpus analyses, which lack tight experimental controls (Kennedy
& Pynte, 2005; Kliegl et al., 2006). Mientras tanto, experimentally controlled studies have found
little to no evidence of parafoveal-on-foveal effects of the lexical properties of a following
palabra, such as lexical frequency, predictability, and plausibility (Angele et al., 2008; Angele
& Rayner, 2011; Angele et al., 2015; Henderson & Ferreira, 1993; Inhoff et al., 2000; Rayner
et al., 2007; Staub et al., 2007). De este modo, we think that parafoveal-on-foveal effects are an unlikely
explanation for the reading patterns at the RC subject region.

En cambio, the effect at the RC subject is consistent with encoding interference. Within
memory-based frameworks, encoding interference can be conceptualized as a feature over-
writing process (Nairne, 1990; Oberauer & Kliegl, 2006). The proposal is that shared features
increase the competition between two items, reducing their distinctiveness and thus their
memory activation. Encoding interference effects are common in psycholinguistics. Para
instancia, the comprehension of object relative clauses is facilitated when the subject and
object are of different types, Por ejemplo, a definite description and a pronoun vs. two definite
descripciones: The barber that you / the lawyer admired (Gordon et al., 2002; see also Barker
et al., 2001; Fedorenko et al., 2006; Hofmeister & Vasishth, 2014). Similarmente, Ha sido
shown that participants read words within a cleft clause more slowly when they rhyme with
those of a currently maintained memory word list, and that this slowdown occurs prior to the
verb/retrieval site (Kush et al., 2015).

Asombrosamente, encoding interference is seldom found in online attraction studies, a pesar de
being common in offline comprehension measures (Adani et al., 2010, 2014; Belletti et al.,
2012; Jäger et al., 2015; parker & Konrad, 2020; Villata & Franck, 2020). Only two previous
self-paced reading studies reported effects consistent with encoding interference, but these
effects were found only when including very long reading times (typically considered outliers
in self-paced reading; Villata et al., 2018) or they were inconsistent across experiments (Herrero
et al., 2021). One explanation for why we were able to detect encoding interference is that we
used an eye-tracking paradigm, which can capture difficulties in information encoding by
measuring regressive eye movements. Por el contrario, self-paced reading studies don’t allow
participants to reread previous material, y, de este modo, encoding difficulty can only emerge as
excessively long reading times, which are likely to be discarded during analysis.

Of the theoretical accounts discussed here, only SOSP can capture encoding interference
efectos. This is because SOSP describes the building of structure as a competition between
treelets with semantic and syntactic features, such that the treelets corresponding to the attrac-
tor and target noun compete to attach to the treelet corresponding to the RC subject node.
When the attractor and target noun overlap in features, competition for attachment is stronger,
which delays processing prior to the verb in grammatical and ungrammatical sentences alike.

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

148

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

/

.

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

/

.

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

Por el contrario, current retrieval-based accounts cannot explain encoding interference, en
either their activation-based (Luis & Vasishth, 2005) or direct access implementations
(Nicenboim & Vasishth, 2018). This is because these accounts assume that interference is
due to an overlap between the retrieval cues of the verb and the features of previously encoded
items in memory. Por lo tanto, attraction can only emerge after the verb is encountered, como el
verb initiates the retrieval process. Sin embargo, the dynamics of encoding interference are similar
to that of another mechanism that is already part of the cue-based retrieval framework: the fan
efecto (Anderson et al., 2004; Luis & Vasishth, 2005). The fan effect penalizes the retrieval of
an item when the retrieval cue also matches other items, eso es, when a feature is shared by
two or more items. The critical difference is that the fan effect only operates at retrieval, mientras
encoding interference affects the creation of items in memory ( Jäger et al., 2015).

As pointed out by Villata et al. (2018), incorporating a fan-like effect at encoding in a cue-
based retrieval framework creates new challenges. In current implementations, the activation
level of a newly created memory chunk does not depend on its feature overlap with other
chunks. This assumption would need to be changed to penalize items whose features overlap
with those of previously encoded items (p.ej., an “activation leveling” process, as suggested by
Villata et al., 2018). Notwithstanding the specific computational implementation, incorporating
a mechanism to generate encoding interference will be critical for future cue-based retrieval
implementations, if they are to appropriately capture the processing patterns attested in the
estudio actual.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

.

/

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Limitations and Open Questions

Our study has several limitations and it leaves some open questions. The first concerns our
research question about the existence of a grammaticality asymmetry, eso es, la posibilidad
that attraction affects grammatical and ungrammatical Spanish sentences differentially. El
statistical correlate of the grammaticality asymmetry is an interaction between grammaticality
and attraction, but our interaction effects were not conclusive. This is puzzling, as all reading
measures showed quantitatively larger attraction effects in ungrammatical than grammatical
oraciones. Sin embargo, the posterior probability distributions of the interaction in Figure 2 también
suggest a broad distribution for these effects, across a relatively large set of values. This indi-
cates that an even more highly powered design might be needed to increase the precision of
the estimates in order to reliably detect the interaction.3 The estimates obtained here could be
used by future studies in Spanish to run power analyses in order to estimate the sample size
needed to detect such evidence.

Segundo, previous work has proposed that the grammaticality asymmetry can be explained
under a view of retrieval as a repair-based mechanism (Lago et al., 2015; Schlueter et al., 2019;
Wagers et al., 2009). The hypothesis was that participants only used memory retrieval to

3 Finding an interaction may have also been particularly challenging because attraction effects in our study
had the same direction in grammatical and ungrammatical sentences. This may help explain why reports of the
grammaticality asymmetry are inconsistent across studies. As mentioned in the Introduction, estudios previos
have coded comparisons in the grammatical sentences differently, some taking the plural attractor condition as a
baseline and others doing the reverse. With the former coding, attraction effects go in opposite directions in
grammatical and ungrammatical sentences, which maximizes the likelihood of finding a significant interaction.
We do not have a solution to this problem, as the alternative coding schemes are motivated by different theo-
retical accounts. Here we decided to use a coding scheme that reflected the linguistic manipulation in our ma-
terials because it seemed like a more theory-neutral way to describe the data.

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

149

The Reading of Agreement Attraction

Lago et al.

“repair” a sentence after a number violation had already been encountered. This repair-based
hypothesis predicts that reading disruptions due to ungrammaticality should temporally pre-
cede attraction effects, as readers should first detect an agreement violation and only afterward
experience attraction due to misretrieval. In our study, we did not find evidence of an earlier
onset of grammaticality vs. attraction effects. First-pass regressions, which index early process-
En g, already showed evidence of both effects. Por lo tanto, our results fail to support a repair-
based view of retrieval. While such evidence might have emerged if other measures of early
processing had been examined, we limited ourselves to measures that have shown attraction in
estudios previos, to reduce the number of multiple comparisons. Más, the use of other early
processing measures would be unadvisable, since these are particularly sensitive to lexical
properties like word length and frequency (Rayner, 1998). In our study, ungrammatical verbs
were always plural and thus longer and less frequent than grammatical verbs. Este problema es
common in attraction studies (Avetisyan et al., 2020; Dillon et al., 2013; Lago et al., 2015;
Schlueter et al., 2019). While we tried to minimize it by using region length as a predictor in
the analysis, future studies might consider a different manipulation of grammaticality to better
compare it with attraction effects.

Finalmente, our finding of encoding interference at the RC subject was only obtained in an
exploratory analysis. This analysis was labeled “exploratory” because, even though encoding
effects are consistent with the SOSP account of attraction, we only thought of performing it
during the analysis stage, as we became acquainted with recent work on attraction in online
comprensión ( Villata et al., 2018; Smith et al., 2021). Por lo tanto, it will be important for
future studies to replicate our finding, especially due to its implication for theoretical accounts
of attraction, either in terms of supporting an SOSP framework or in showing the need to
incorporate a feature overwriting process in cue-based retrieval models. Fundamentalmente, our findings
suggest that to detect encoding interference effects in comprehension, the use of large partic-
ipant samples is advisable.

EXPRESIONES DE GRATITUD

We thank Anna Laurinavichyute, Garrett Smith, and two anonymous reviewers for useful
feedback and discussions.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

.

/

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

CONTRIBUCIONES DE AUTOR

SL: Conceptualización: Equal; Curación de datos: Lead; Análisis formal: Secundario; Writing –
original draft: Lead; Escritura – revisión & edición: Lead. CAF: Conceptualización: Equal;
Escritura – borrador original: Equal; Escritura – revisión & edición: Equal. EM: Conceptualización:
Equal; Curación de datos: Equal; Análisis formal: Equal; Investigación: Lead; Recursos: Lead;
Escritura – borrador original: Equal; Escritura – revisión & edición: Equal.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

REFERENCIAS

Acuña-Fariña, j. C., Meseguer, MI., & Carreiras, METRO. (2014). Gender
and number agreement in comprehension in Spanish. Lingua,
143, 108–128. https://doi.org/10.1016/j.lingua.2014.01.013
Adani, F., Forgiarini, METRO., Guasti, METRO. T., & Lely, h. k. j. V. D. (2014).
Number dissimilarities facilitate the comprehension of relative
clauses in children with (Grammatical) Specific Language
Impairment. Journal of Child Language, 41(4), 811–841. https://
doi.org/10.1017/S0305000913000184, PubMed: 23806292
Adani, F., van der Lely, h. k. J., Forgiarini, METRO., & Guasti, METRO. t.
(2010). Grammatical feature dissimilarities make relative clauses

easier: A comprehension study with Italian children. Lingua, 120(9),
2148–2166. https://doi.org/10.1016/j.lingua.2010.03.018,
PubMed: 21151323

anderson, j. r., Bothell, D., Byrne, METRO. D., Douglass, S., Lebiere, C.,
& Qin, Y. (2004). An integrated theory of the mind. Psicológico
Revisar, 111(4), 1036–1060. https://doi.org/10.1037/0033-295X
.111.4.1036, PubMed: 15482072

Angele, B., & Rayner, k. (2011). Parafoveal processing of word n + 2
during reading: Do the preceding words matter? Diario de
Psicología experimental: Percepción y desempeño humanos,

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

150

The Reading of Agreement Attraction

Lago et al.

37(4), 1210–1220. https://doi.org/10.1037/a0023096, PubMed:
21553996

Angele, B., Schotter, mi. r., Slattery, t. J., Tenenbaum, t. l.,
Bicknell, K., & Rayner, k. (2015). Do successor effects in reading
reflect lexical parafoveal processing? Evidence from corpus-
based and experimental eye movement data. Diario de
Memory and Language, 79–80, 76–96. https://doi.org/10.1016/j
.jml.2014.11.003

Angele, B., Slattery, t. J., Cual, J., Kliegl, r., & Rayner, k. (2008).
Parafoveal processing in reading: Manipulating n + 1 y N + 2
previews simultaneously. Visual Cognition, 16(6), 697–707.
https://doi.org/10.1080/13506280802009704, PubMed:
19424452

Avetisyan, S., Lago, S., & Vasishth, S. (2020). Does case marking
affect agreement attraction in comprehension? Diario de
Memory and Language, 112, Article 104087. https://doi.org/10
.1016/j.jml.2020.104087

Pregonero, J., Nicol, J., & Garrett, METRO. (2001). Semantic factors in the pro-
duction of number agreement. Journal of Psycholinguistic Research,
30(1), 91–114. https://doi.org/10.1023/a:1005208308278,
PubMed: 11291185

barr, D. J., Exacción, r., Transportistas, C., & Teja, h. j. (2013). Aleatorio
effects structure for confirmatory hypothesis testing: Keep it
maximal. Journal of Memory and Language, 68(3), 255–278.
https://doi.org/10.1016/j.jml.2012.11.001, PubMed: 24403724
Belletti, A., Friedmann, NORTE., Brunato, D., & Rizzi, l. (2012). Hace
gender make a difference? Comparing the effect of gender on
children’s comprehension of relative clauses in Hebrew and
italiano. Lingua, 122(10), 1053–1069. https://doi.org/10.1016/j
.lingua.2012.02.007

Bock, K., & Cutting, j. C. (1992). Regulating mental energy:
Performance units in language production. Journal of Memory
and Language, 31(1), 99–127. https://doi.org/10.1016/0749
-596X(92)90007-k

Bock, K., & Eberhard, k. METRO. (1993). Significado, sound and syntax in
English number agreement. Language and Cognitive Processes,
8(1), 57–99. https://doi.org/10.1080/01690969308406949

Bock, K., & Molinero, C. A. (1991). Broken agreement. Cognitivo
Psicología, 23(1), 45–93. https://doi.org/10.1016/0010-0285
(91)90003-7, PubMed: 2001615

Bonferroni, C. mi. (1936). Teoria statistica delle classi e calcolo delle
probabilità. Pubblicazioni Del R Istituto Superiore Di Scienze
Economiche e Commerciali Di Firenze, 8, 3–62.

Bürkner, P.-C. (2017). brms: An R Package for Bayesian multilevel
models Using Stan. Journal of Statistical Software, 80(1). https://
doi.org/10.18637/jss.v080.i01

Cumming, GRAMO. (2014). The new statistics: Why and how.
ciencia psicológica, 25(1), 7–29. https://doi.org/10.1177
/0956797613504966, PubMed: 24220629

Cunnings, I., & Sturt, PAG. (2018). Retrieval interference and semantic
interpretación. Journal of Memory and Language, 102, 16–27.
https://doi.org/10.1016/j.jml.2018.05.001

Dienes, z. (2011). Bayesian versus orthodox statistics: Which
side are you on? Perspectives on Psychological Science, 6(3),
274–290. https://doi.org/10.1177/1745691611406920,
PubMed: 26168518

Dillon, B., Mishler, A., Sloggett, S., & Phillips, C. (2013). Contrasting
intrusion profiles for agreement and anaphora: Experimental
and modeling evidence. Journal of Memory and Language, 69(2),
85–103. https://doi.org/10.1016/j.jml.2013.04.003

Eberhard, k. METRO. (1997). The marked effect of number on subject–verb
agreement. Journal of Memory and Language, 36(2), 147–164.
https://doi.org/10.1006/jmla.1996.2484

Eberhard, k. METRO., Cutting, j. C., & Bock, k. (2005). Making syntax of
sense: Number agreement in sentence production. Psicológico
Revisar, 112(3), 531–559. https://doi.org/10.1037/0033-295X
.112.3.531, PubMed: 16060750

Engelmann, F., Jäger, l. A., & Vasishth, S. (2019). The effect of
prominence and cue association on retrieval processes: A com-
putational account. Ciencia cognitiva, 43(12), Article e12800.
https://doi.org/10.1111/cogs.12800, PubMed: 31858626

Fedorenko, MI., Gibson, MI., & Rohde, D. (2006). The nature of work-
ing memory capacity in sentence comprehension: Evidencia
against domain-specific working memory resources. Diario de
Memory and Language, 54(4), 541–553. https://doi.org/10.1016
/j.jml.2005.12.006

Franck, J., Colonna, S., & Rizzi, l. (2015). Task-dependency and
structure-dependency in number interference effects in sentence
comprensión. Fronteras en psicología, 6. https://doi.org/10
.3389/fpsyg.2015.00349, PubMed: 25914652

Franck, J., Vigliocco, GRAMO., & Nicol, j. (2002). Subject-verb agreement
errors in French and English: The role of syntactic hierarchy.
Language and Cognitive Processes, 17(4), 371–404. https://doi
.org/10.1080/01690960143000254

Gelman, A., Carlin, j. B., Stern, h. S., & Frotar, D. B. (2014).
Bayesian data analysis (3tercera ed.). Chapman and Hall/CRC Press.
https://doi.org/10.1201/b16018

Gelman, A., & Stern, h. (2006). The difference between “signifi-
cant” and “not significant” is not itself statistically significant.
The American Statistician, 60(4), 328–331. https://doi.org/10
.1198/000313006X152649

Godfroid, A., & Hui, B. (2020). Five common pitfalls in eye-
tracking research. Second Language Research, 36(3), 277–305.
https://doi.org/10.1177/0267658320921218

gordon, PAG. C., Hendrick, r., & Johnson, METRO. (2004). Effects of noun
phrase type on sentence complexity. Journal of Memory and
Idioma, 51(1), 97–114. https://doi.org/10.1016/j.jml.2004.02.003
gordon, PAG. C., Hendrick, r., & Levin, W.. h. (2002). Memory-load
interference in syntactic processing. ciencia psicológica, 13(5),
425–430. https://journals.sagepub.com/doi/10.1111/1467-9280
.00475, PubMed: 12219808

Hammerly, C., Staub, A., & Dillon, B. (2019). The grammaticality
asymmetry in agreement attraction reflects response bias:
Experimental and modeling evidence. Psicología cognitiva,
110, 70–104. https://doi.org/10.1016/j.cogpsych.2019.01.001,
PubMed: 30798061

Hartsuiker, R. J., Antón-Méndez, I., & van Zee, METRO. (2001). Object
attraction in subject-verb agreement construction. Diario de
Memory and Language, 45(4), 546–572. https://doi.org/10.1006
/jmla.2000.2787

Hartsuiker, R. J., Schriefers, h. J., Bock, K., & Kikstra, GRAMO. METRO. (2003).
Morphophonological influences on the construction of subject-
verb agreement. Memoria & Cognición, 31(8), 1316–1326.
https://doi.org/10.3758/BF03195814, PubMed: 15058692

Henderson, j. METRO., & Ferreira, F. (1993). Eye movement control
during reading: Fixation measures reflect foveal but not parafo-
veal processing difficulty. Canadian Journal of Experimental
Psicología, 47(2), 201–221. https://doi.org/10.1037/h0078814,
PubMed: 8364530

Hofmeister, PAG., & Vasishth, S. (2014). Distinctiveness and encoding
effects in online sentence comprehension. Fronteras en psicología,
5. https://doi.org/10.3389/fpsyg.2014.01237, PubMed: 25566105
Inhoff, A. w., Starr, METRO., & Shindler, k. l. (2000). Is the processing of
words during eye fixations in reading strictly serial? Percepción &
Psicofísica, 62(7), 1474–1484. https://doi.org/10.3758
/bf03212147, PubMed: 11143457

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

151

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

.

/

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

Jäger, l. A., Benz, l., Roeser, J., Dillon, B. w., & Vasishth, S.
(2015). Teasing apart retrieval and encoding interference in the
processing of anaphors. Fronteras en psicología, 6. https://doi.org
/10.3389/fpsyg.2015.00506, PubMed: 26106337

Jäger, l. A., Engelmann, F., & Vasishth, S. (2017). Similarity-based
interference in sentence comprehension: Literature review and
Bayesian meta-analysis. Journal of Memory and Language, 94,
316–339. https://doi.org/10.1016/j.jml.2017.01.004

Jäger, l. A., Mertzen, D., Van Dyke, j. A., & Vasishth, S. (2020).
Interference patterns in subject-verb agreement and reflexives re-
visited: A large-sample study. Journal of Memory and Language,
111, 104063. https://doi.org/10.1016/j.jml.2019.104063,
PubMed: 33100507

Kempen, GRAMO., & Vosse, t. (1989). Incremental syntactic tree for-
mation in human sentence processing: A cognitive architecture
based on activation decay and simulated annealing. Connection
S c i e n c e , 1 ( 3 ) , 2 7 3 – 2 9 0 . h t t p s : / / d o i . o r g / 1 0 . 1 0 8 0
/09540098908915642

Kennedy, A., & Pynte, j. (2005). Parafoveal-on-foveal effects in nor-
mal reading. Vision Research, 45(2), 153–168. https://doi.org/10
.1016/j.visres.2004.07.037, PubMed: 15581917

Kliegl, r., Nuthmann, A., & Engbert, R. (2006). Tracking the mind
during reading: The influence of past, present, and future words
on fixation durations. Revista de Psicología Experimental:
General, 135(1), 12–35. https://doi.org/10.1037/0096-3445.135
.1.12, PubMed: 16478314

Kruschke, j. K., & Liddell, t. METRO. (2018). The Bayesian New
Estadísticas: Hypothesis testing, estimation, meta-analysis, y
power analysis from a Bayesian perspective. psiconomico
Boletín & Revisar, 25(1), 178–206. https://doi.org/10.3758
/s13423-016-1221-4, PubMed: 28176294

Kush, D., Johns, C. l., & Van Dyke, j. A. (2015). Identifying the role
of phonology in sentence-level reading. Journal of Memory and
Idioma, 79–80, 18–29. https://doi.org/10.1016/j.jml.2014.11
.001, PubMed: 26893535

Lago, S., Shalom, D. MI., Sigman, METRO., Lau, mi. F., & Phillips, C. (2015).
Agreement attraction in Spanish comprehension. Diario de
Memory and Language, 82, 133–149. https://doi.org/10.1016/j
.jml.2015.02.002

Lau, mi. F., Rozanova, K., & Phillips, C. (2007). Syntactic prediction and
lexical surface frequency effects in sentence processing. Universidad
of Maryland Working Papers in Linguistics, 16, 163–200.

Lehtonen, METRO., & Laine, METRO. (2003). How word frequency affects
morphological processing in monolinguals and bilinguals.
Bilingualism: Language and Cognition, 6(3), 213–225. https://doi
.org/10.1017/S1366728903001147

Lewandowski, D., Kurowicka, D., & Joe, h. (2009). Generating ran-
dom correlation matrices based on vines and extended onion
método. Journal of Multivariate Analysis, 100(9), 1989–2001.
https://doi.org/10.1016/j.jmva.2009.04.008

Luis, R. l., & Vasishth, S. (2005). An activation-based model of sen-
tence processing as skilled memory retrieval. Ciencia cognitiva,
29(3), 375–419. https://doi.org/10.1207/s15516709cog0000_25,
PubMed: 21702779

Lorimor, h., Bock, K., Zalkind, MI., Sheyman, A., & Beard, R.
(2008). Agreement and attraction in Russian. Language and
Cognitive Processes, 23(6), 769–799. https://doi.org/10.1080
/01690960701774182

Nairne, j. S. (1990). A feature model of immediate memory.
Memoria & Cognición, 18(3), 251–269. https://doi.org/10.3758
/BF03213879, PubMed: 2192233

Ness, T., & Meltzer-Asscher, A. (2017). Working memory in the
Interference and
long-distance dependencies:

processing of

filler maintenance. Journal of Psycholinguistic Research, 46(6),
1353–1365. https://doi.org/10.1007/s10936-017-9499-6,
PubMed: 28528512

Nuevo, B., Brysbaert, METRO., Segui, J., Ferrand, l., & Rastle, k. (2004).
The processing of singular and plural nouns in French and
Inglés. Journal of Memory and Language, 51(4), 568–585.
https://doi.org/10.1016/j.jml.2004.06.010

Nicenboim, B., & Vasishth, S. (2018). Models of retrieval in
sentence comprehension: A computational evaluation using
Bayesian hierarchical modeling. Journal of Memory and
Idioma, 99, 1–34. https://doi.org/10.1016/j.jml.2017.08.004
Nicenboim, B., Vasishth, S., Engelmann, F., & Suckow, k. (2018).
Exploratory and confirmatory analyses in sentence processing: A
case study of number interference in German. Ciencia cognitiva,
42(S4), 1075–1100. https://doi.org/10.1111/cogs.12589,
PubMed: 29411408

Nicol, j. l., Forster, k. I., & Veres, C. (1997). Subject–verb agreement
processes in comprehension. Journal of Memory and Language,
36(4), 569–587. https://doi.org/10.1006/jmla.1996.2497

Nieuwenhuis, S., Forstman, B. Ud., & carpinteros, E.-J. (2011).
Erroneous analyses of interactions in neuroscience: A problem of
significance. Neurociencia de la naturaleza, 14(9), 1105–1107. https://doi
.org/10.1038/nn.2886, PubMed: 21878926

Oberauer, K., & Kliegl, R. (2006). A formal model of capacity limits
in working memory. Journal of Memory and Language, 55(4),
601–626. https://doi.org/10.1016/j.jml.2006.08.009

parker, D., & Konrad, k. (2020). Teasing apart encoding and
retrieval interference in sentence comprehension: Evidencia de
agreement attraction. In S. Denison, METRO. Mack, Y. Xu, & B. C.
Armstrong (Editores.), Proceedings of the 42nd Annual Conference
of the Cognitive Science Society (páginas. 1446–1452). Cognitivo
Science Society.

parker, D., & Phillips, C. (2017). Reflexive attraction in comprehen-
sion is selective. Journal of Memory and Language, 94, 272–290.
https://doi.org/10.1016/j.jml.2017.01.002

Patson, norte. D., & Husband, mi. METRO. (2016). Misinterpretations in
agreement and agreement attraction. Quarterly Journal of
Psicología experimental, 69(5), 950–971. https://doi.org/10
.1080/17470218.2014.992445, PubMed: 25624171

Pearlmutter, norte. J., Garnsey, S. METRO., & Bock, k. (1999). Agreement
processes in sentence comprehension. Journal of Memory and
Idioma, 41(3), 427–456. https://doi.org/10.1006/jmla.1999.2653
(2020). R: A Language and
for Statistical Computing. R Foundation for

R Development Core Team.

Environment
Statistical Computing.

Rayner, k. (1998). Eye movements in reading and information
Procesando: 20 years of research. Boletín Psicológico, 124(3),
372–422. https://doi.org/10.1037/0033-2909.124.3.372,
PubMed: 9849112

Rayner, K., Juhasz, B. J., & Marrón, S. j. (2007). Do readers obtain
preview benefit from word n + 2? A test of serial attention shift
versus distributed lexical processing models of eye movement
control in reading. Revista de Psicología Experimental. Humano
Perception and Performance, 33(1), 230–245. https://doi.org/10
.1037/0096-1523.33.1.230, PubMed: 17311490

Schlueter, Z., parker, D., & Lau, mi. (2019). Error-driven retrieval in
agreement attraction rarely leads to misinterpretation. Fronteras en
Psicología, 10. https://doi.org/10.3389/fpsyg.2019.01002,
PubMed: 31133936

Schotter, mi. R. (2018). Reading ahead by hedging our bets on seeing
the future: Eye tracking and electrophysiology evidence for par-
afoveal lexical processing and saccadic control by partial word
recognition. In The psychology of learning and motivation:

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

152

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

/

.

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

The Reading of Agreement Attraction

Lago et al.

Current topics in language (páginas. 263–298). Elsevier Academic
Prensa. https://doi.org/10.1016/bs.plm.2018.08.011

Herrero, GRAMO., Franck, J., & Tabor, W.. (2018). A self-organizing approach
to subject–verb number agreement. Ciencia cognitiva, 42(S4),
1043–1074. https://doi.org/10.1111/cogs.12591, PubMed:
29388256

Herrero, GRAMO., Franck, J., & Tabor, W.. (2021). Encoding interference
effects support self-organized sentence processing. Cognitivo
Psicología, 124, Article 101356. https://doi.org/10.1016/j
.cogpsych.2020.101356, PubMed: 33285355

Stan Development Team. (2020). RStan: The R interface to Stan. R

package version 2.16.2. https://mc-stan.org

Staub, A. (2009). On the interpretation of the number attraction
efecto: Response time evidence. Journal of Memory and
Idioma, 60(2), 308–327. https://doi.org/10.1016/j.jml.2008.11
.002, PubMed: 20126291

Staub, A.

(2010a). Response time distributional evidence for
distinct varieties of number attraction. Cognición, 114(3),
447–454. https://doi.org/10.1016/j.cognition.2009.11.003,
PubMed: 20003964

Staub, A. (2010b). Eye movements and processing difficulty in
object relative clauses. Cognición, 116(1), 71–86. https://doi.org
/10.1016/j.cognition.2010.04.002, PubMed: 20427040

Staub, A., Dillon, B., & Clifton, C. (2017). The matrix verb as a
source of comprehension difficulty in object relative sentences.
Ciencia cognitiva, 41(S6), 1353–1376. https://doi.org/10.1111
/cogs.12448, PubMed: 27813146

Staub, A., Rayner, K., Pollatsek, A., Hyönä, J., & Majewski, h.
(2007). The time course of plausibility effects on eye movements
in reading: Evidence from noun-noun compounds. Diario de
Psicología experimental. Aprendiendo, Memoria, and Cognition,
33(6), 1162–1169. https://doi.org/10.1037/0278-7393.33.6
.1162, PubMed: 17983320

Vasishth, S., Mertzen, D., Jäger, l. A., & Gelman, A. (2018). El
statistical significance filter leads to overoptimistic expectations of

replicability. Journal of Memory and Language, 103, 151–175.
https://doi.org/10.1016/j.jml.2018.07.004

Vasishth, S., Nicenboim, B., Engelmann, F., & Burchert, F. (2019).
Computational models of retrieval processes in sentence pro-
cesando. Tendencias en Ciencias Cognitivas, 23(11), 968–982. https://
doi.org/10.1016/j.tics.2019.09.003, PubMed: 31668586

Vigliocco, GRAMO., Butterworth, B., & Garrett, METRO. F. (1996). Subject-verb
agreement in Spanish and English: Differences in the role of con-
ceptual constraints. Cognición, 61(3), 261–298. https://doi.org/10
.1016/S0010-0277(96)00713-5, PubMed: 8990974

Vigliocco, GRAMO., Butterworth, B., & Semenza, C. (1995). Constructing
subject-verb agreement in speech: The role of semantic and
morphological factors. Journal of Memory and Language, 34(2),
186–215. https://doi.org/10.1006/jmla.1995.1009

Vigliocco, GRAMO., & Nicol, j. (1998). Separating hierarchical relations
and word order in language production: Is proximity concord
syntactic or linear? Cognición, 68(1), B13–B29. https://doi.org
/10.1016/S0010-0277(98)00041-9, PubMed: 9775519

Villata, S., & Franck, j. (2020). Similarity-based interference in
agreement comprehension and production: Evidence from object
agreement. Revista de Psicología Experimental: Aprendiendo,
Memoria, and Cognition, 46(1), 170–188. https://doi.org/10
.1037/xlm0000718, PubMed: 31033310

Villata, S., Tabor, w., & Franck, j. (2018). Encoding and retrieval
interference in sentence comprehension: Evidence from agree-
mento. Fronteras en psicología, 9. https://doi.org/10.3389/fpsyg
.2018.00002, PubMed: 29403414

von der Malsburg, T., & Angele, B. (2017). False positives and other
statistical errors in standard analyses of eye movements in read-
En g. Journal of Memory and Language, 94, 119–133. https://doi
.org/10.1016/j.jml.2016.10.003, PubMed: 28603341

Wagers, METRO. w., Lau, mi. F., & Phillips, C. (2009). Agreement attrac-
tion in comprehension: Representations and processes. Diario de
Memory and Language, 61(2), 206–237. https://doi.org/10.1016/j
.jml.2009.04.002

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
oh
pag
metro

i
/

yo

a
r
t
i
C
mi
–
pag
d

F
/

d
oh

i
/

i

/

/

.

1
0
1
1
6
2
oh
pag
metro
_
a
_
0
0
0
4
7
1
9
6
9
1
6
3
oh
pag
metro
_
a
_
0
0
0
4
7
pag
d

.

/

i

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
7
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

MENTE ABIERTA: Descubrimientos en ciencia cognitiva

153

Descargar PDF