If you're not familiar with grammatical gender, let me explain: some languages are wonderfully weird and sort their nouns into gender categories. As a Portuguese speaker, a tree is feminine to me, but as a Spanish speaker, it's masculine!
My topics of interest
Grammatical gender
about my gender.
I want to understand how speakers of gendered languages represent gender in the mind and process it during language production and comprehension, using EEG and behavioral methods across language families. My questions here include how a word's form (its orthography and phonology) cues its gender through dual-route and pseudomorphological mechanisms; whether the masculine and feminine values are processed differently, and if so whether the reasons are linguistic, social, or both; and how bilinguals handle gender when a word's translation carries a different value.
What fascinates me most is how gender is expressed in the form of nouns, and how much this varies across languages. In Dutch you cannot predict a noun's gender from its form; in German you can, but it is quite hard; in Spanish it's remarkably easy, thanks to two magical cues (-a and -o) that end 60–70% of nouns. Due to this, Spanish is considered a “gender transparent” language just like Italian and Portuguese. This variation has led me to question whether gender is represented and processed the same way across languages, and to develop a model that accommodates both transparent and opaque systems. Concretely, I am testing my adaptation of the AUSTRAL model of lexical access for gender retrieval across language families, and refining a computational model (built with colleagues) that quantifies just how transparent a language's gender system is.
Social bias in grammar
Grammatical gender is a linguistic category, often defined as abstract and arbitrary. There is no reason for a table to be masculine in German but feminine in Spanish (grammatical gender is not related to the meaning of words). But does the mind keep it purely linguistic?
Recent research suggests that grammatical gender processing is influenced not only by linguistic factors but also by social and extra-linguistic factors. Nouns whose grammatical gender aligns with gender stereotypes are processed more efficiently (e.g., in Spanish, falda [skirt] is feminine and stereotypically associated with women, whereas corbata [tie] is also feminine but stereotypically associated with men. This coincidence or mismatch seems to influence how quickly these words are processed). Additionally, individual differences might matter: women tend to be more sensitive to gender agreement and feminine forms, while traditional gender roles, opposition to inclusive language, and hostile sexism are associated with a preference for masculine forms and reduced sensitivity to the feminine gender value (e.g., Casado, Sá-Leite et al., 2023; Casado et al., 2018; Pesciarelli et al., 2019).
Beliefs and attitudes related to our sex-based reality somehow seem to shape the cognitive representation of grammar, challenging modular views of human cognition and the Fodorian conceptualization of the language system.
Based on this, I am testing whether gender roles, social attitudes, and sex-based stereotypes are tied to the abstract and/or semantic gender of nouns during real-time processing, asking where our mental representation of grammar meets social cognition. I am doing so in languages with different types of gender systems and across populations from different cultures, extending these questions to a total of twelve languages across three continents.
Emotion in language
Language does not only convey meaning; it also carries emotion. I study how the affective value of words shapes language comprehension across different speaker populations, from German bilinguals to Spanish mono-, bi-, and trilinguals. Rather than treating emotion as an isolated psychological variable, I am interested in how affective information interacts with ongoing grammatical computations during real-time language processing, and whether emotional engagement with language differs depending on linguistic experience, proficiency, or multilingualism.
As part of Isabel Fraga’s project funded by the Spanish Government, we are testing the relationship between emotion and grammatical processing, asking how these two fundamental aspects of human cognition interact during real-time language comprehension. Following previous work on the neural correlates of grammatical and emotional processing, this project moves beyond traditional approaches by examining individual and contextual differences and extending the investigation beyond native language processing. Focusing on number agreement across Spanish, Galician, and English, we aim to understand whether and when emotional information modulates morphosyntactic processing, combining behavioral and electrophysiological measures to capture both the time-course and consequences of this interaction. In another smaller project, I am testing whether emotional valence modulates lexical access during language switching in German-English bilinguals. We examine whether emotional information influences word recognition even when it is irrelevant to the task itself.
Reading in an AI world
Today's readers build meaning from text alongside a machine that offers to do the reading for them. Most teenagers already use chatbots for schoolwork, many of them to summarize the very things they were assigned to read. AI has come into our lives not silently, but abruptly and sometimes even overwhelmingly. Yet we know almost nothing about what generative AI does to deep reading. Comprehension is not one thing: we can retain the surface wording of a text, its explicit content, or the deeper situation model that integrates what we read with what we already know (Kintsch, 1998). My question is whether AI-assisted summarization leaves this deepest, most effortful level intact, or quietly erodes it while sparing the shallower ones.
I am beginning with adults, with an already pre-registered study where summarization of texts can be done with or without AI, and comprehension questions follow with and without a delay. Comprehension is measured at all three representational levels, so that a selective cost at the situation-model level would become visible where a coarser measure would miss it. Because the very act of AI-assisted summarizing may offload the effortful, self-generated work that makes learning durable (the kind of work implicated in the generation effect, levels of processing, and desirable difficulties) I expect any cost to concentrate in deep comprehension and in delayed memory, rather than in immediate recall of wording.
Still, reading is not only cognitive. I also ask what AI does to the feeling of reading: whether summarizing a story with AI hampers narrative engagement and transportation, i.e., the emotional experience through which fiction is thought to nurture empathy. And because readers differ, I test who is most affected and who is protected, bringing individual differences to the table. I consider how deeply participants rework the AI's output into their own words, how their reading style affects the results, as well as their prior experience with AI, and their personality traits.
Adults are the starting point. Adolescence (when deep reading is still consolidating and these tools are already in every classroom) is where the question matters the most, and extending the work there is the project I am now looking to fund. That is the future this research is directed at: understanding how a new generation of human minds learns to read and think, alongside AI.