Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks

The neural dynamics of sentence production: ECoG reveals sentence-specific networks and temporal patterns

There is a Poster PDF for this presentation, but you must be a current member or registered to attend SNL 2023 to view it. Please go to your Account Home page to register.

Poster B13 in Poster Session B, Tuesday, October 24, 3:30 - 5:15 pm CEST, Espace Vieux-Port

Adam Morgan1, Werner Doyle1, Orrin Devinsky1, Adeen Flinker1; 1NYU School of Medicine

Understanding the neural underpinnings of sentence production is a central goal in the neurobiology of language. While significant progress has been made on single word production, a similarly nuanced understanding of sentence production remains elusive. Here, we collect electrocorticographic (ECoG) recordings from nine awake neurosurgical patients during a controlled sentence production experiment, the first such study we are aware of. Our experiment involved 3 blocks: (1) picture naming, where patients repeatedly named six characters; (2) controlled sentence production, where patients described cartoon scenes involving two of the six characters (e.g., “Frankenstein poked the chicken”), and (3) list production (“chicken Frankenstein”), which, similar to sentence production involves sequencing, but unlike sentence production does not engage event semantics or syntax. Employing an unsupervised machine learning method, non-negative matrix factorization, we clustered electrodes into six functional networks. This revealed four networks that pattern with previously-characterized processes that are common to word, sentence, and list production: stimulus processing, motor planning, articulation, and auditory feedback. However, two additional networks, previously undescribed, were active for sentences but not lists or words. Subsequent analysis of these networks, primarily distributed across middle and inferior frontal gyri, revealed sensitivity to two sentence-specific processes: event semantics (i.e., whether the event involved a physical action) and syntax (whether the upcoming sentence was active/passive). Next, we assessed the commonplace characterization of sentence production as a sequence of single word production processes. If accurate, then by decoding what word a patient is saying throughout the course of sentence production, we should expect each word to come online in the order in which it was said. More precisely, because words consist of multiple representational stages (conceptual, phonological, etc.), we should expect the stages to come online in the same order, too. To test this, we trained multiple classifiers (specific to patients and representational stages) to predict word identity using data from the picture naming task (10-fold cross validation; accuracy ~30%, p < .01 relative to permutation distribution). We then used these classifiers to predict what word the patient was saying throughout the course of sentence production. The majority of classifiers (89%) successfully generalized from single word production to sentence production data, accurately predicting what words the patient said during sentences, with different stages of word representation largely following the same temporal order observed in single word production. Intriguingly, however, we also observed a systematic exception: Objects — the last word of the sentence — were often predicted far earlier relative to their articulation in sentences than in single word production. In fact, 14% of classifiers predicted objects even before the onset of the first word in the sentence. This result aligns with emerging behavioral evidence pointing toward early object planning. Our findings emphasize the importance of expanding beyond single word production studies to achieve a more comprehensive understanding of language production. Moreover, these findings hold potentially important implications for clinical practice, which still relies heavily on single word production paradigms.

Topic Areas: Language Production, Syntax and Combinatorial Semantics

SNL Account Login

Forgot Password?
Create an Account