Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Complex word processing in single trial EEG shows evidence for partial decomposition
Poster Session B, Friday, September 12, 4:30 - 6:00 pm, Field House
Laurel Lawyer1, Ehsan Solaimani2; 1University of California Davis, 2University of York
The processing of complex words has been a long-standing area of interest in psycholinguistics. While recent evidence from MEG studies (e.g., Stockall et al. (2019)) supports the separate processing of prefixes in visually presented words, little work has addressed whether this is also the case in speech perception. The present study uses single-trial EEG data to investigate auditory processing of complex spoken words, providing evidence for partial decomposition. In an ERP experiment, participants (N=28) were presented 40 real and 80 altered complex words and asked to indicate whether each was a well-pronounced word of English. Words were coded for frequency of the full word (e.g., 'unlikely'), the root (e.g., 'like'), and the prefix (e.g., 'un-'). Altered words were derived by substituting a single segment in the prefix of an existing complex word (e.g., 'tisprove' from 'disprove'), and were coded for root frequency ('prove'), as well as whole-word and prefix frequency of the original words ('disprove' and 'dis-'). Data from midline sites (Fz, Cz, Pz) was separately modeled for words and altered words using a generalized additive mixed model (GAMM). Root and prefix frequencies were each residualized against whole-word frequency to reduce colinearity, and all three frequency measures were log normalized. Tensor product smooths were used to estimate the non-linear interaction between time and each frequency measure. Subject, item, and channel within subject were included as random effects in the models. In word responses, we found significant interactions between time and both prefix and whole-word frequency (ps < 0.005), with no significant interaction between time and root frequency (p > 0.1). Visual inspection of the tensor product smooths illustrates that in real words, an greater negativity is observed for low-frequency prefixes onsetting around 450msec, and for whole-word frequency onsetting around 600msec. For altered words, all three measures of frequency were significant (prefix, word, and non-word; all ps > 0.001). Visual inspection of these smooths shows greater negativity for low-frequency roots onsetting around 500msec, and for whole words and prefixes both around 600msec. This data suggests that complex words are subject to partial decomposition, but not full decomposition. In real words, prefix and whole word frequencies predict EEG signal amplitude sequentially in separable effects. However, this decomposition is not total, as participants do not appear to separately access the root. In altered words, where prefixes do not provide a reliable cue to the current complex word, we observe a response that indexes access to the root first, followed by the prefix and whole-word frequency effects thereafter. Taken together, this data provides evidence for partial decompositional parsing of morphologically complex words, with sub-elements of complex words, such as prefixes, accessed individually, prior to the fully composed word. Only in cases where the identity of the prefix is unreliable do we find evidence of separate access to the root.
Topic Areas: Morphology, Speech Perception