Neural correlate of the construction of sentence meaning

Abstract

The neural processes that underlie your ability to read and understand this sentence are unknown. Sentence comprehension occurs very rapidly, and can only be understood at a mechanistic level by discovering the precise sequence of underlying computational and neural events. However, we have no continuous and online neural measure of sentence processing with high spatial and temporal resolution. Here we report just such a measure: intracranial recordings from the surface of the human brain show that neural activity, indexed by γ-power, increases monotonically over the course of a sentence as people read it. This steady increase in activity is absent when people read and remember nonword-lists, despite the higher cognitive demand entailed, ruling out accounts in terms of generic attention, working memory, and cognitive load. Response increases are lower for sentence structure without meaning ("Jabberwocky" sentences) and word meaning without sentence structure (word-lists), showing that this effect is not explained by responses to syntax or word meaning alone. Instead, the full effect is found only for sentences, implicating compositional processes of sentence understanding, a striking and unique feature of human language not shared with animal communication systems. This work opens up new avenues for investigating the sequence of neural events that underlie the construction of linguistic meaning.

Keywords: ECoG; compositionality; language; semantics; syntax.

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Fig. 1.

(A) Cortical models of individual subjects showing all electrodes (black) (see Table 1 for numbers of electrodes), electrodes that show a significant sentences > nonword-lists (S > N) effect in odd-numbered runs (blue), and electrodes that show both a significant S > N effect and a monotonic increase across word positions in the sentence condition in odd-numbered runs, [i.e., our electrodes of interest, EOIs (blue, circled in white)]. (B) The γ-magnitude for sentences and nonword-lists averaged across word positions estimated in even-numbered runs (i.e., data independent from the data used to select the EOIs). (C) The γ-magnitude for sentences and nonword-lists in each of eight word positions in even-numbered runs (see SI Appendix, Part F for additional figures showing data not averaged within each word position). Error bars indicate SEMs over EOIs in both B and C.

Fig. 2.

(A) The γ-magnitude for all conditions (S, sentences; W, word-lists; J, Jabberwocky sentences; N, nonword-lists) averaged across word positions and subjects estimated in even-numbered runs. Darker bars show responses across all trials, lighter bars show responses across the subset of trials for which the memory-probe task was answered correctly. (B) The γ-magnitude for all conditions averaged across subjects in each of eight word positions in even-numbered runs. (C–H) The γ-magnitude for all conditions averaged across word positions (C, E, G) and in each of eight word positions (D, F, H) in even-numbered runs in three groups of EOIs: left frontal (C and D), left lateral temporal (E and F), and left ventral temporal (G and H). All four subjects had some EOIs in left frontal (total: 21 EOIs) and left lateral temporal (total: 20 EOIs) cortex, and two of the four had EOIs in left ventral temporal (total: 9 EOIs) cortex (see SI Appendix, Part G for numbers of EOIs in each group for each subject). Error bars indicate SEMs over subjects in all panels.

Fig. 3.

High γ-magnitude during the early (first two words) and late (last two words) sentence positions estimated across EOIs in even-numbered runs for all conditions. Error bars indicate SEMs over subjects.

Fig. 4.

Comparison of percent signal change (PSC) of correctly and incorrectly answered trials in the word-lists (A), Jabberwocky (B), and nonword-lists (C) conditions, averaged across subjects. For comparison, we include the average of the correctly answered sentence trials (there were too few subjects with incorrectly answered sentence trials to compute averages). Error bars indicate SEMs over subjects.