You are viewing the SNL 2017 Archive Website. For the latest information, see the Current Website.

 
Poster A1, Wednesday, November 8, 10:30 – 11:45 am, Harborview and Loch Raven Ballrooms

Alpha and theta power are sensitive to semantic but not syntactic retrieval interference

Ashley Lewis1, Julie Van Dyke1;1Haskins Laboratories

According to cue-based retrieval parsing (Lewis, Vasishth, & Van Dyke, 2006) grammatical heads give rise to a retrieval probe, which incorporates retrieval cues associated with critical features (e.g., semantic properties, thematic role, grammatical case) of a constituent that needs to be retrieved from memory for argument integration. Retrieval interference can occur when alternative constituents in the intervening sentence region between a to-be-retrieved constituent and a head share features with the retrieval probe. We used EEG and neural oscillations to investigate the timing of neural events related to syntactic and semantic retrieval interference. Memory retrieval has been linked to alpha, theta, and gamma oscillations (Spitzer et al., 2009). In non-interfering contexts, beta and gamma oscillations have been implicated in semantic and syntactic integration (Lewis, Wang, & Bastiaansen, 2015). Participants' (n=23) EEG was recorded while they read relative clause sentences (RSVP) in one of four conditions, varying the level of syntactic and semantic interference in the region intervening between the animate matrix clause subject and the matrix clause verb. Semantic (animate vs inanimate) and syntactic (potential grammatical subject vs not) feature overlap between an intervening referent and the desired antecedent was used to create high compared to low semantic and syntactic interference respectively. A time-frequency analysis of power in both a low (2-30 Hz) and a high (32-100 Hz) frequency range (multitaper approach; Mitra & Pesaran, 1999) was conducted at the matrix clause verb (target word). Statistical significance was assessed for comparisons of high and low interference using cluster-based permutation statistics (Maris & Oostenveld, 2007). For semantic interference, right centroparietal alpha power (6-13 Hz) at the target word decreased when interference was low (p = 0.01) but not when interference was high. Frontal theta power (5-8 Hz) at the target word increased for high compared to low semantic interference (p = 0.004). There were no statistically significant differences in alpha or theta power for comparisons between high and low syntactic interference. No other frequency bands showed sensitivity to the level or type of interference. We argue that the alpha power effect is related to the allocation of attention (more pronounced alpha power desynchronization = more attention) to the linguistic input stream (Jensen & Mazaheri, 2010). Low retrieval interference results in successful integration of the target word, after which the parser can shift its full attention to the next word in the input, resulting in a pronounced alpha power desynchronization. High retrieval interference on the other hand results in a greater reliance on more deliberate processing (less attention allocated to the linguistic input stream) in order to resolve the interference, resulting in a temporary reduction in alpha desynchronization. Frontal theta power has been linked to increased working memory load (Jensen & Tesche, 2002), which is consistent with the need to select between multiple potential antecedents when interference is high. Behavioral work has suggested that syntactic interference is easier to identify and recover from (Van Dyke, 2007), which may explain why we only observe oscillatory effects when comparing high and low semantic interference.

Topic Area: Computational Approaches

Back to Poster Schedule