Poster A27, Thursday, August 16, 10:15 am – 12:00 pm, Room 2000AB
Perception of Morphologically Complex Words Using Single-Trial EEG
Laurie Lawyer1;1University of Essex
Introduction. Within models of speech perception involving morphologically complex words (eg. Taft (1994), Schreuder & Baayen (1995), Stockall & Marantz (2006), Marslen-Wilson and Tyler (2007)) there is disagreement over whether complex words must be decomposed into their constituent parts during online processing, or whether they may be accessed as whole forms. This study presents a novel approach, using single-trial EEG data to investigate the processing of real words and mispronounced morphologically complex words. Methods. In an ERP experiment, subjects (N=28) were presented 40 complex words and 80 non-words and asked to identify correctly pronounced words. Non-words were created by substituting a single segment in the prefix of an existing complex word with a segment that differed by one to three major phonological features (place, voice, or manner). A linear mixed-effects model was estimated at every time point (-200 to 1400msec) using the single-trial EEG data. The input to the model for each time point contained the amplitude at each electrode site, for each word, for each subject. Predictors were word frequency (from CELEX (Baayen et al., 1995)), and residualised prefix frequency and root frequencies. For modified nonwords, the original frequency values were used. Subject, item, and electrode site were included as random effects. All z-values exceeding 1.96 (estimating p < .05) were considered significant at a given time point. Results. Subject performance showed good discrimination, with 94% of real words correctly labeled, and 87% of mispronounced words correctly identified. In correctly labeled word responses, EEG signal amplitudes were predicted to increase significantly as prefix frequency increased between 260 - 320msec. Word frequency significantly predicted increases EEG amplitude in a later window, from 580 - 1040msec. Root frequency did not achieve statistical significance in any time window. In correctly identified nonwords, EEG signal was predicted to decrease significantly as prefix frequency increased, from 180 - 200msec. Further, root frequency significantly predicted increases in EEG amplitude from 330 - 550msec, and word frequency from 750 - 950msec. Conclusions. The analysis of factors influencing single-trial EEG data show a cascading effect of lexical access. In real words, prefix, root, and word frequencies predict EEG signal amplitude sequentially in separable effects. In cases of prefix mispronunciation, prefix frequency negatively predicted signal amplitude, while root frequency had a similar effect observed in real word processing. This suggests subjects may have used the un-altered roots to bootstrap successful recognition of the mispronounced words, evidenced by the significant word frequency effects. Taken together, this data provides evidence for mutli-staged decompositional parsing of morphologically complex words. The data suggests sub-elements of complex words, such as prefixes, are accessed individually, prior to the fully composed word, which is separately subsequently accessed. This supports the idea that morphologically complex elements are obligatorily decomposed during processing, while also suggesting a different approach to dual-route models where whole-word processing of complex words occurs in a secondary stage of processing.
Topic Area: Grammar: Morphology