How Do We Segment Text? Two-Stage Chunking Operation in Reading

The dynamics of ERP responses and clustering results. A, Averaged ERPs waveform responses of all conditions from 32 electrodes (black lines), and RMS waveform response across all electrodes (red line). B, Temporal clustering results of topographies for four conditions (GwLw, GwLn, GnLw, and GnLn) and a baseline symbol condition (symbol). Different colors represent distinct clusters. Samples in the same color but at different time points indicate that they are grouped into the same cluster, sharing similar features but occurring at different times. The temperature of colors represents the rank of the cluster distance relative to the cluster baseline (cluster defined by the baseline period). Approximately 80 ms after stimulus onset, a novel cluster (the second cluster) appears at the same time across five conditions, followed by another new cluster. However, in the symbol condition, the third cluster (∼250 ms) appears earlier with a much shorter duration than four-character string conditions. C, The topographies of each cluster.

The effects of lexicalized chunks revealed in the paired comparisons between GnLn condition and the other three lexical conditions (GwLw, GwLn, and GnLw). A, ERP waveform responses in the representative channels P3 and P4. One-way ANOVA did not reveal any response amplitude differences among conditions. B, RMS waveform responses of all channels. No amplitude difference was found. C, Topographical comparisons of response amplitude. Each row shows a comparison across time. The color scheme depicts the differences in response amplitude between conditions. No significant difference was found on the electrodes after the multiple comparison correction (FDR). D, The temporal dynamics of TANOVA on paired comparisons (uncorrected). The red boxes highlight the earliest latency when the significant differences were observed. All three conditions show evidence of early lexical detection. E, The temporal dynamics of TANOVA on the comparison between GnLn and the average of three lexical conditions, corrected by temporal clustering analysis with a corrected threshold of 0.05 (Maris and Oostenveld, 2007). The lexicality effects emerge around 100 ms.

The processing dynamics of chunks at global and local levels. A, ERP waveform responses of the global lexicality effect in the representative channels P3 and P4; t tests revealed the effects occurred around 160 ms. The shaded area indicates p < 0.05. B, ERP waveform responses of local lexicality effect in the representative channels P3 and P4. The effects occurred around 250 ms, later than the global lexicality effect in A. The shaded area indicates p < 0.05. C, RMS waveform responses of all channels on global lexicality and local lexicality comparisons. No amplitude difference was found. D, Analysis of response amplitude in topographical comparisons between different lexical status at the global level (upper row) and at the local level (lower row) across time. The color scheme depicts the differences in response amplitude between conditions, and the white points superimposed on the topographies indicate the electrodes that showed significant differences after multiple comparison correction (FDR). E, The temporal dynamics of TANOVA results. The results showed a distinct starting time of significant response pattern differences between different lexical status at the global and local levels. The red arrows in all plots indicate the earliest latency of difference in the global level comparison, and the green arrows indicate the earliest latency of difference in the local level comparison. The results were corrected by a temporal clustering analysis with a corrected threshold of 0.05 (Maris and Oostenveld, 2007).