Slide Slam S6 Sandbox Series
Tracking hierarchical processes in minimal linguistic phrases
Piermatteo Morucci1, Jordi Martorell1, Nicola Molinaro1,2; 1Basque Center on Cognition Brain and Language, 2Ikerbasque, Basque Foundation for Science
Does meaning composition during language processing rely on hierarchical or linear computations? Using the frequency-tagging paradigm, Ding et al., (2016) reported that neural oscillations track the incremental combination of words into phrases, and phrases into sentences – that is, they track the unfolding of hierarchical linguistic structures. Yet, computational studies have shown that non-hierarchical models purely based on word-level lexical features could also explain earlier results, suggesting that linear computations may underlie meaning composition (Frank & Yang, 2018). Here we present the experimental design for a magnetoencephalography (MEG) study aimed at testing for the presence of hierarchical processes during minimal linguistic phrase reading. In this experiment, we will use the frequency-tagging paradigm with periodic presentation of written words. We will compare oscillatory responses to 3-word Noun Phrases (NPs) in Spanish containing either linear or hierarchical structures. In both the linear and hierarchical conditions, the first two words are identical (Noun and Colour Adjective; árbol rojo, tree red). The critical manipulation relies on the properties of the third word: in the linear condition, a Size Adjective (e.g., grande, big) modifies the Noun, while in the hierarchical condition a Degree Adjective (e.g., oscuro, dark) modifies the Colour Adjective – giving rise to an Adjective Phrase (AP). Crucially, this resulting AP is embedded within the NP, thus forming a hierarchical structure. By using NPs with identical syntactic categories in both conditions (Noun-Adjective-Adjective), our manipulation specifically targets differences in hierarchical structure, while ruling out potential confounds driven by lexical-category features. Spanish speakers will be presented with trials composed of sequences of 3-word NPs (eight NPs per trial) which instantiate either hierarchical or linear structures. Each word will be presented periodically for 0.5 s. The frequencies of interest will be 2 Hz for words, 1.33 Hz for AP, and 0.67 Hz for NP. After each trial, a picture will appear and participants will be asked to indicate whether or not it matches any of the NPs presented in the trial. This task represents a good proxy to assess whether participants deploy linear/hierarchical processing, as they have to access the compositional meaning of each NP in order to determine the match between its referent and the picture. MEG data will be analyzed through frequency-domain power analysis. We expect to observe power peaks corresponding to both the word and whole NP presentation frequencies in both conditions. Critically, if participants deploy hierarchical processing, we should find an additional peak at the AP frequency in the hierarchical condition, but not in the linear condition. This peak would reflect an extra computation required to access the overall compositional meaning of a hierarchical NP. We will present the experimental design of our study paired with simulations of expected results, as well as preliminary results from a pilot study designed to assess the feasibility of our task at the behavioral level.