Parafoveal Processing of Grammatical Gender in Spanish

Do readers process articles and nouns together?

Marina Serrano-Carot & Bernhard Angele

2026-05-06

Short version of this talk:

  • Core question: How much syntactic information is processed parafoveally during reading?
  • Two complementary paradigms in Spanish:
    • Word flanker (grammatical categorization/lexical decision/semantic categorization)
    • Gaze-contingent boundary (sentence reading with eye tracking and a preview manipulation)
  • Shared phenomenon: article–noun agreement (gender/number)
  • Overall finding: Whether discordant articles affect ongoing processing depends on the task. There does not seem to be an immediate effect on the reading process, but the impact of a discordant article preview can be observed in subsequent fixations.

Main research questions

  1. Do readers process articles and nouns together?
  2. Do they compute agreement before fixation?
  3. What happens when agreement is violated?
  4. Are effects task-dependent (decision task vs. natural reading)?

Overview

  1. Theoretical background and predictions
  2. Flanker study: Articles as flankers
  3. Boundary study: Boundary preview of articles
  4. Cross-study synthesis and implications

Background: Why articles?

  • Extremely frequent, short, and often skipped in reading
  • In Spanish, articles are morphosyntactically rich:
    • Gender: el/la
    • Number: el/la vs. los/las
  • Ideal test case for parafoveal syntactic processing (Angele & Rayner, 2013).
English
the
Spanish
ellaloslas
German
derdiedasdesdemden

Background: Reading and parafoveal processing

  • Reading understood as the act of visually scanning written language in order to decode the symbols and comprehend their meaning.
  • A complex cognitive process that involves both, foveal and parafoveal processing.
    • Foveal Processing: Words directly fixated by the eyes. Highest visual acuity and detail perception.
    • Parafoveal Processing: The ability of the eyes to take in information from words that have not yet been directly fixated.
    • With eye-tracking, we can study how readers use this parafoveal information to facilitate reading comprehension and speed. These studies can measure, for example, how the context of words that are perceived parafoveally influences the anticipation of upcoming words.

Background: Skipping behavior

  • Word skipping is the act of moving one’s gaze past a word without fixating it.
  • This is essential for efficient reading.
  • When readers decide whether to skip a word, they need to use parafoveal information about the next word.
  • But which information do they use?
    • Do they use information about the preceding context when they skip words?
    • Do they use information about the subsequent context?

Parafovea vs. preceding context

  • For the preceding context, previous studies in English (Angele et al., 2014; Angele & Rayner, 2013) have shown that readers tend to skip short and high-frequency words, such as the definite article “the”.
  • Some of these studies have shown that readers skip articles even when they are incompatible with the preceding context.
    • Angele & Rayner (2013):
She was sure she would
the ace
all the tests.
  • Even when the context is extremely constrained, readers skip the article preview:
    • Abbott et al. (2015): “If you are shot in the heart, you will surely die/the immediately”.

What about the subsequent context?

  • It is unclear if this is also true when both the article and the incompatible context are in the parafovea.
  • In a language like Spanish, this could happen when there is a mismatch between the article and the noun in terms of grammatical gender.
  • La mesa vs. el* mesa

Background: Article-noun agreement

  • In languages with grammatical gender, like Spanish (or German), articles must agree with the noun in gender and number (Corbett, 1991).
  • In Spanish, gender is often transparent (e.g., -a for feminine mesa), but can be opaque (e.g., raíz is feminine).
  • Agreement violations (el mesa) lead to the gender agreement effect:
  • Research Question: Is this agreement processed parafoveally?

Are articles and nouns processed together?

  • Word Grouping Hypothesis (Radach, 1996):
    • Articles and nouns may be targeted by saccades as a single functional unit.
    • Fixations often aim for the center of the article-noun group, leading to high skipping rates for articles.
  • Multi-Constituent Unit (MCU) Hypothesis (Zang, 2019):
    • Frequently co-occurring words (like la mesa) may be processed as a single lexicalized unit.
  • Parallel Processing: Does the parafoveal article influence the foveal noun’s identification?

Early detection of syntactic violations

  • Evidence from RSVP and ERP studies suggests syntactic violations are detected very quickly and often unconsciously.
  • Early mismatch signals:
    • Batterink & Neville (2013): Early negativity for violations even when not consciously reported.
    • Lucchese et al. (2017): Mismatch negativities in auditory processing occur before full word recognition.
    • Pyatigorskaya et al. (2023): Masked function word primes trigger agreement costs without conscious perception.
  • Question: Does this rapid detection translate to parafoveal processing?

Parafoveal processing and shallow processing

  • The Missing Letter Effect shows readers frequently miss target letters in high-frequency function words (like “the”) compared to content words (Healy, 1994; Koriat & Greenberg, 1993).
  • Similarly, repeated articles (e.g., “the the”) are frequently missed during natural reading (Staub et al., 2019).
  • Does this mean articles are processed with a shallow, late integration mechanism where only basic definiteness is extracted quickly, and syntactic anomalies are ignored until later?

Theoretical implications

  • Early integration view: Readers combine parafoveal article + noun features rapidly to guide skipping and eye movements.
  • Late integration / shallow parafoveal view: Early eye guidance (skipping) is driven by visual factors (length, frequency); syntactic mismatch effects emerge only later during integration.
  • Modeling the process:
    • Serial attention shift: e.g., E-Z Reader; Reichle et al. (1998); Reichle et al. (2003): Attention shifts sequentially; nouns are not processed until the article is identified.
    • Parallel/Gradient processing: e.g., SWIFT; Engbert et al. (2005): Multiple words processed at once; allows (theoretically) for joint article-noun effects.
    • Noisy Channel / Perceptual Inference: Readers may “correct” noisy parafoveal input to maintain fluency.

The Flanking Letters Paradigm

  • Used to investigate whether parafoveally presented information obligatorily influences foveal word processing.
  • A central target word is flanked by context words/letters:
  • Can be used with any forced choice task such as:
    • semantic categorization
    • lexical decision
    • grammatical gender decision
  • Allows us to ask: Is gender information in the parafovea processed obligatorily, and does it vary by task?

Flanker study overview

  • Serrano-Carot et al. (2026)
  • Three experiments:
    • Exp 1: grammatical gender decision
    • Exp 2: lexical decision
    • Exp 3: semantic categorization
  • Central noun with article flankers (or no flankers in Exp 1)

Flanker study manipulation (Exp 1: Gender categorization)

Concordant (AGR)

+
la  mesa  la
mesa

Target noun and flankers agree in gender.

Discordant (DIS)

+
el  mesa  el
mesa

Flankers violate noun gender agreement.

Flanker study manipulation (Exp 2: Lexical decision)

Word Concordant

+
la  mesa  la
mesa

Decide if target is a word.

Pseudoword Discordant

+
el  mepa  el
mepa

Decide if target is a word.

Flanker study manipulation (Exp 3: Semantic categorization)

Non-animal Concordant

+
la  mesa  la
mesa

Decide if target is an animal.

Animal Discordant

+
el  vaca  el
vaca

Decide if target is an animal.

Hypotheses from the AUSTRAL+ model

Architecture

  • Syntactic Level: Assumes nodes for grammatical gender.
  • Dual Routes:
    • Sublexical: Input from morpheme endings (e.g., -a).
    • Lexical: Input from the complete word lemmas.

Task Predictions

  • Explicit (Gender categorization):
    • Queries gender nodes \(\rightarrow\) Strong.
  • Implicit (Lexical decision):
    • Sublexical route is secondary \(\rightarrow\) Small.
  • Semantic (Categorization):
    • No direct link \(\rightarrow\) Minimal/None.

Method: Flanker study samples and tasks

  • Exp 1: 109 participants.
  • Exp 2: 54 participants
  • Exp 3: 60 participants
  • RT models with Bayesian mixed effects; accuracy also modeled.

Raincloud plot: Flanker study Exp 1 RT distributions

Raincloud plot: Flanker study Exp 2 (Words) RT distributions

Raincloud plot: Flanker study Exp 2 (Pseudowords) RT distributions

Raincloud plot: Flanker study Exp 3 RT distributions

Comparison Plot: Difference in RT (Discordant vs. Concordant/Control)

Flanker study: Interim conclusions

  • Evidence for article–noun joint processing is strongest when grammar is task-relevant
  • With less grammar-relevant tasks, agreement effects weaken or vanish
  • Suggests parafoveal/foveal agreement signals can influence behavior, but task demands modulate the impact
  • How does this translate to natural reading conditions?
    • Sentence reading with a gaze-contingent display change

Boundary study

  • Source: Serrano-Carot & Angele (2026), Journal of Eye Movement Research (Serrano-Carot & Angele, 2026)
  • Two experiments that manipulate the preview readers receive of an article before fixatiing it.
    • Exp 1: Gender agreement manipulation
    • Exp 2: Number, gender, and definiteness manipulation

Boundary study (Experiment 1)

Dissimilar gender preview condition

He olvidado limpiar el la sala del fondo.

Animation: Gaze (red dot) moves across the sentence. When it crosses the invisible boundary (dashed line), the mismatched parafoveal preview (“el”) changes to the correct target (“la”).

Boundary study (Experiment 2)

Number/Gender mismatch condition

Ayer encontré los una mesa nueva en la tienda.

Animation: Preview of a plural/masculine article (“los”) changes to the singular/feminine target (“una”) during the saccade to the noun.

Method: Experiment 1

  • EyeLink Portable Duo, 1000 Hz recording frequency.
  • LCD screen, 240 Hz refresh rate.
  • 24 native Spanish speakers, aged 18-35.
  • 210 sentences in Spanish with a noun phrase.
    • 105 with masculine nouns (e.g., “coche/s”).
    • 105 with feminine nouns (e.g., “mesa/s”).
      • Phrase components: definite article (“el”/ “la”/ “los”/ “las”) + noun.
      • High frequency words from EsPal.
      • Word length of 4-6 (singular), 5-7 (plural).

Experimental Design: Examples

Gender Number Identical preview Dissimilar preview
Fem Sing Necesito que me ayudes a poner la mesa para la cena. Necesito que me ayudes a poner el* mesa para la cena.
Fem Plur Necesito que me ayudes a poner las mesas para la cena. Necesito que me ayudes a poner los* mesas para la cena.
Masc Sing Mi abuelo miraba el coche con curiosidad. Mi abuelo miraba la* coche con curiosidad.
Masc Plur Mi abuelo miraba los coches con curiosidad. Mi abuelo miraba las* coches con curiosidad.

Hypotheses

What happens when the article preview disagrees with the noun?

  1. Functional Unit Hypothesis (Early Integration)
    • Readers process the article and noun together before fixation.
    • Prediction: Mismatch \(\rightarrow\) Increased fixation times and Reduced skipping rates.
  2. Shallow Processing Hypothesis (Visual-Only)
    • Skipping is driven by word length and frequency, not syntax.
    • Prediction: No effect on skipping or fixation times from gender/number mismatch.
  3. Late Integration Hypothesis (Spillover)
    • Syntactic agreement is computed only after the word is fixated.
    • Prediction: Increased fixation times (especially on the noun/post-target), but No effect on skipping.

Method: Experiment 2

  • EyeLink Portable Duo, 1000 Hz recording frequency.
  • LCD screen, 240 Hz refresh rate.
  • 24 native Spanish speakers, aged 18-35.
  • 210 sentences in Spanish with a noun phrase.
    • 105 with masculine nouns (e.g., “coche/s”).
    • 105 with feminine nouns (e.g., “mesa/s”).
      • Phrase components: definite masculine plural article (“los”) + noun vs indefinite feminine article (“una”) + noun.
      • “Ayer encontré una mesa nueva en la tienda”.
      • “Ayer encontré los*mesa nueva en la tienda”.

Article skipping

TVT: Experiment 1

Comparison Plot: Difference in TVT (Dissimilar vs. Identical Preview)

Article skipping in Experiment 2

TVT distributions in Experiment 2

Discussion: Skipping and fixation

  • Readers showed no significant change in skipping probability based on the preview condition. This suggests two possibilities:
    1. Readers may not have enough parafoveal information about the upcoming noun to detect the violation.
    2. The lack of article-noun agreement might not influence the decision to skip.
  • Some evidence was found for an effect of the article preview on fixation times on the noun.
  • Indicates that the incorrect article preview is partially processed.
  • This processing leads to disruption during later stages of reading.

A potential modelling account

  • The SEAM model (Rabe et al., 2024) combines the oculomotor and word identification components of the SWIFT model and the syntactic components of the Lewis & Vasishth (2005) model.
  • Regressions are triggered by memory retrieval.
subject object subject
The robber that the policeman in the patrol car chased escaped.

Applied to our study

  • Perhaps reaching a subsequent word or even the end of the sentence triggers a recall of the noun phrase.
  • This recall may be disrupted by memory traces of the preview.
  • May trigger regressions back to “poesías” and increase TVT.
object: F object: M
Este libro contiene los las poesías más bonitas que he leído nunca.

Conclusion (Boundary study)

  • Skipping decisions might rely more on general word characteristics than on specific syntactic agreement.
  • Disruptions in later reading stages suggest that syntactic mismatches do affect processing, but perhaps not at the stage of making the skipping decision.
  • Indefinite articles could be skipped more due to being less specific.

Answers to the core questions

1) Do readers process article+noun together?

  • Yes, at least sometimes: both studies show mismatch costs, especially in later measures.

2) Do they compute agreement early enough for skipping?

  • Usually no in natural sentence reading: skipping is largely unchanged.

3) Consequences of lack of agreement?

  • Downstream cost: longer TVT and spillover/reanalysis.
  • In explicit tasks, mismatch also affects decision RT/accuracy.

Theoretical implications: Word recognition

  • Task-Dependent Mandatory Processing:
    • Interference in gender/lexical tasks but not semantic tasks suggests that syntactic processing is modulated by task relevance.
  • Challenges to “All-or-Nothing” Lexical Access:
    • Contradicts models where “lexical access” is a single moment when all info becomes active.
    • Supports distributed models (e.g., Triangle model) where orthographic, syntactic, and semantic nodes are connected but can be queried independently.
  • Support for the AUSTRAL Model (Sá-Leite & Lago, 2024):
    • Supports the dual-route account: gender categorization queries syntactic nodes directly (strong effect), while lexical decision does not (weaker effect).

Theoretical implications: Reading models

  • Informational Economy & Strategic Deferral:
    • Readers “skip now, check later.” Highly automated processing of short articles prioritizes landing on the noun (semantic load).
  • Perceptual Inference & Noisy Channel (Staub et al., 2019):
    • Readers may “correct” noisy or mismatched parafoveal input to maintain fluent forward movement, deferring reanalysis.
  • Memory Retrieval & The SEAM Model:
    • Delayed disruption (late TVT) suggests that processing the noun triggers a retrieval of the article from memory.
    • Mismatch results in retrieval failure/interference, leading to late reanalysis and regressions.
  • Serial vs. Parallel Processing:
    • Absence of early effects supports serial accounts (E-Z Reader) or parallel models where syntactic integration lags behind oculomotor decisions.

Final conclusions

  • Spanish readers do not reliably use parafoveal article–noun agreement to decide skipping.
  • But disagreement is not ignored: it generates later processing costs.
  • Task demands strongly modulate whether agreement affects immediate behavior.
  • Together, the two studies sharpen our understanding of when grammatical agreement matters during reading.

References

Abbott, M. J., Angele, B., Ahn, Y. D., & Rayner, K. (2015). Skipping syntactically illegal the previews: The role of predictability. Journal of Experimental Psychology: Learning, Memory, and Cognition, 41(6), 1703–1714. https://doi.org/10.1037/xlm0000142
Alemán Bañón, J., & Rothman, J. (2016). The role of morphological markedness in the processing of number and gender agreement in spanish: An event-related potential investigation. Language, Cognition and Neuroscience, 31(10), 1273–1298. https://doi.org/10.1080/23273798.2016.1218032
Angele, B., Laishley, A. E., Rayner, K., & Liversedge, S. P. (2014). The effect of high- and low-frequency previews and sentential fit on word skipping during reading. Journal of Experimental Psychology. Learning, Memory, and Cognition, 40(4), 1181–1203. https://doi.org/10.1037/a0036396
Angele, B., & Rayner, K. (2013). Processing the in the parafovea: Are articles skipped automatically? Journal of Experimental Psychology: Learning, Memory, and Cognition, 39(2), 649–662. https://doi.org/10.1037/a0029294
Barber, H., & Carreiras, M. (2005). Grammatical gender and number agreement in spanish: An ERP comparison. Journal of Cognitive Neuroscience, 17(1), 137–153. https://doi.org/10.1162/0898929052880101
Batterink, L., & Neville, H. J. (2013). The human brain processes syntax in the absence of conscious awareness. The Journal of Neuroscience, 33(19), 8528–8533. https://doi.org/10.1523/jneurosci.0618-13.2013
Corbett, G. G. (1991). Gender. Cambridge University Press.
Dare, N., & Shillcock, R. (2013). Serial and parallel processing in reading: Investigating the effects of parafoveal orthographic information on nonisolated word recognition. Quarterly Journal of Experimental Psychology, 66(3), 487–504. https://doi.org/10.1080/17470218.2012.703212
Engbert, R., Nuthmann, A., Richter, E. M., & Kliegl, R. (2005). SWIFT: A dynamical model of saccade generation during reading. Psychological Review, 112(4), 777–813.
Healy, A. F. (1994). Letter detection: A window to unitization and other cognitive processes in reading text. Psychonomic Bulletin & Review, 1, 333–344. https://doi.org/10.3758/BF03213975
Koriat, A., & Greenberg, S. N. (1993). Prominence of leading functors in function morpheme sequences as evidenced by letter detection. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19(1), 34–50. https://doi.org/10.1037/0278-7393.19.1.34
Lewis, R. L., & Vasishth, S. (2005). An activation-based model of sentence processing as skilled memory retrieval. Cognitive Science, 29(3), 375–419. https://doi.org/10.1207/s15516709cog0000_25
Lucchese, G., Hanna, J., Autenrieb, A., Miller, T. M., & Pulvermüller, F. (2017). Electrophysiological evidence for early and interactive symbol access and rule processing in retrieving and combining language constructions. Journal of Cognitive Neuroscience, 29(2), 254–266. https://doi.org/10.1162/jocn_a_01038
Pyatigorskaya, E., Maran, M., & Zaccarella, E. (2023). Testing the automaticity of syntax using masked visual priming. Language, Cognition and Neuroscience, 38(7), 925–949. https://doi.org/10.1080/23273798.2023.2173790
Rabe, M. M., Paape, D., Mertzen, D., Vasishth, S., & Engbert, R. (2024). SEAM: An integrated activation-coupled model of sentence processing and eye movements in reading. Journal of Memory and Language, 135, 104496. https://doi.org/10.1016/j.jml.2023.104496
Radach, R. (1996). Blickbewegungen beim lesen: Psychologische aspekte der determination von fixationspositionen [eye movements in reading: Psychological factors that determine fixation locations]. Münster/New York: Waxmann.
Reichle, E. D., Pollatsek, A., Fisher, D. L., & Rayner, K. (1998). Toward a model of eye movement control in reading. Psychological Review, 105(1), 125–157.
Reichle, E. D., Rayner, K., & Pollatsek, A. (2003). The e-z reader model of eye-movement control in reading: Comparisons to other models. Behavioral and Brain Sciences, 26(4), 445–476.
Sá-Leite, A. R., & Lago, S. (2024). The role of word form in gender processing during lexical access: A theoretical review and novel proposal in language comprehension. Psychonomic Bulletin & Review, 31(5), 1934–1953. https://doi.org/10.3758/s13423-023-02426-8
Serrano-Carot, M., & Angele, B. (2026). Spanish readers skip articles regardless of gender and number agreement. Journal of Eye Movement Research, 19(1), 6. https://doi.org/10.3390/jemr19010006
Serrano-Carot, M., Gomez, P., & Angele, B. (2026). Articles as flankers: The effect of grammatical gender depends on the task. Language, Cognition and Neuroscience, 0(0), 1–23. https://doi.org/10.1080/23273798.2026.2657830
Snell, J., Van Leipsig, S., Grainger, J., & Meeter, M. (2018). OB1-reader: A model of word recognition and eye movements in text reading. Psychological Review, 125(6), 969–984. https://doi.org/10.1037/rev0000119
Staub, A., Dodge, S., & Cohen, A. L. (2019). Failure to detect function word repetitions and omissions in reading: Are eye movements to blame? Psychonomic Bulletin & Review, 26(1), 340–346. https://doi.org/10.3758/s13423-018-1492-z
Taft, M. (2023). Localist lexical representation of polymorphemic words: The AUSTRAL model. Routledge.
Zang, C. (2019). New Perspectives on Serialism and Parallelism in Oculomotor Control During Reading: The Multi-Constituent Unit Hypothesis. Vision (Basel, Switzerland), 3(4), 50. https://doi.org/10.3390/vision3040050