Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Investigating the Planning of Manner, Path and Deixis in Mandarin Direction Serial Verb Constructions using EEG-MVPA
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Runze Liang1, Laura Giglio; 1University of Zurich
How speakers transform an event into a structured utterance remains a central question in language production. Mandarin directional serial verb constructions (DSVCs) offer a useful test case because a single predicate for a motion action can encode several event components, typically including manner of motion, path, and deictic orientation. A motion event can be described with a three-morpheme sequence such as 跑_run+上_up+来_come or 走_walk+进_in_去_go, where the first element specifies manner, the second path, and the third direction of movement relative to the speaker’s perspective. Analyses have argued that such constructions are internally structured rather than unanalyzable lexical compounds, but it remains unclear whether these components are planned incrementally or as a single element during speech production. The present study tested this question using EEG-based time-resolved multivariate pattern analysis (MVPA) during Mandarin DSVC production. Participants (N=25) described visual motion-event prompts, where one person moves towards or away from another person in different manners and paths, with a three-morpheme DSVC in a fixed manner-path-deixis order. The stimulus set factorially manipulated 4 manners (run, walk, jump and skate), 4 paths (in, out, up and down), and 2 deictic orientations (come vs. go), with 4 visual versions per condition (e.g. mirroring the images). First, we asked whether EEG activity during the planning epoch contains information about each component of the DSVC independently of the others. Analyses were conducted separately for manner, path, and deixis from picture onset using correctly produced trials. Time-resolved classifiers were evaluated against the appropriate chance level (25% for manner and path, 50% for deixis) using cross-validation. Significant above-chance classification was determined using cluster-based permutation. In addition, temporal-generalization analyses tested whether component-specific patterns were transient or stable across the planning interval. Group results showed significant above-chance decoding for all three components, but with different temporal profiles. Path was decoded earliest, peaking at 352 ms after picture onset (accuracy = 0.351; chance = 0.25), with significant intervals from 179-727 ms and 765-844 ms. Deixis showed the strongest and most sustained effect, peaking at 441 ms (accuracy = .634; chance = .50) and remaining significant from 117-924 ms. Manner emerged later, with significant decoding between 678-1000 ms, peaking at 980 ms (accuracy = .348; chance = .25). Temporal-generalization analyses converged with these patterns: deixis showed broad cross-time generalization, path showed mid-epoch effects extending across neighboring train-test times, and manner showed weaker, later, and more restricted generalization. Together, these results suggest a graded planning profile, but not in the order of articulation. The first articulated DSVC component, manner, seems to be encoded last, just before the speech onset. Earlier decoding for path and deixis suggests that these elements are retrieved earlier and perhaps maintained until articulation. Overall, this study demonstrates the feasibility of applying EEG-MVPA to a multi-component predicate, extending previous single word production-decoding work to the formulation of complex Mandarin motion descriptions. The results suggest that these serial verbs are planned incrementally but not in order of mention.
Topic Areas: Language Production, Syntax and Combinatorial Semantics