Lack of selectivity for syntax relative to word meanings throughout the language network

Abstract

To understand what you are reading now, your mind retrieves the meanings of words and constructions from a linguistic knowledge store (lexico-semantic processing) and identifies the relationships among them to construct a complex meaning (syntactic or combinatorial processing). Do these two sets of processes rely on distinct, specialized mechanisms or, rather, share a common pool of resources? Linguistic theorizing, empirical evidence from language acquisition and processing, and computational modeling have jointly painted a picture whereby lexico-semantic and syntactic processing are deeply inter-connected and perhaps not separable. In contrast, many current proposals of the neural architecture of language continue to endorse a view whereby certain brain regions selectively support syntactic/combinatorial processing, although the locus of such "syntactic hub", and its nature, vary across proposals. Here, we searched for selectivity for syntactic over lexico-semantic processing using a powerful individual-subjects fMRI approach across three sentence comprehension paradigms that have been used in prior work to argue for such selectivity: responses to lexico-semantic vs. morpho-syntactic violations (Experiment 1); recovery from neural suppression across pairs of sentences differing in only lexical items vs. only syntactic structure (Experiment 2); and same/different meaning judgments on such sentence pairs (Experiment 3). Across experiments, both lexico-semantic and syntactic conditions elicited robust responses throughout the left fronto-temporal language network. Critically, however, no regions were more strongly engaged by syntactic than lexico-semantic processing, although some regions showed the opposite pattern. Thus, contra many current proposals of the neural architecture of language, syntactic/combinatorial processing is not separable from lexico-semantic processing at the level of brain regions-or even voxel subsets-within the language network, in line with strong integration between these two processes that has been consistently observed in behavioral and computational language research. The results further suggest that the language network may be generally more strongly concerned with meaning than syntactic form, in line with the primary function of language-to share meanings across minds.

Keywords: Cognitive neuroscience; Composition; Language architecture; Lexical semantics; Syntax.

Figure 1:. A (non-exhaustive) set of theoretically possible architectures of language.

Distinct boxes correspond to distinct brain regions (or sets of brain regions; e.g., in 1a–d, “combinatorial processing” may recruit a single region or multiple regions, but critically, this region or these regions do not support other aspects of language processing, like understanding word meanings). The architectures differ in whether they draw a (region-level) distinction between the lexicon and grammar (a vs. b–f), between storage and access of linguistic representations (1a–b vs. 1c–f), and critically, in whether combinatorial processing is a separable component (1a–d vs. 1e–f).

Figure 2:. Sample stimuli for each condition in Experiments 1–3.

Two examples are provided for each condition in each experiment. For Experiment 1, the top row shows the beginning of a sentence, and the next rows show different possible continuations. For Experiments 2–3, the top row shows one sentence from a pair, and the next rows show different possibilities for the other sentence in that pair. Red: Lexico-semantic condition; Blue: Syntactic condition; Green: other experimental conditions; Black: control condition. [NB1: For Experiment 1, the task was passive reading for the critical materials, but, as described in the text, a small number of (filler) trials contained a memory probe task. NB2: For Experiment 2, three versions of the same base item (corresponding to the Lexico-semantic, Syntactic, and Global meaning conditions) are presented for illustrative purposes. As detailed in the text, in the actual materials, distinct sets of base items were used for the three critical conditions in order to match the number of trials across conditions while avoiding sentence repetition.]

Figure 3:. Trial structure for Experiments 1–3.

One sample trial is shown for each experiment.

Figure 4:. Summary of the behavioral results from Experiments 1–3.

Each column shows the results from a different experiment: left - Experiment 1: Violations; middle - Experiment 2: Recovery from Adaptation; right - Experiment 3: Meaning Judgment. Top graphs: error rates; bottom graphs: reaction times. Here and in all subsequent figures, for each condition and each measure, dots correspond to individual participants; the bar shows the average across these participants, and the error bar shows standard error of the mean (across participants). Significant differences between the critical, Lexico-semantic (red) and Syntactic (blue), conditions are marked with *’s.

Figure 5:. Responses in language fROIs to the conditions in Experiments 1–3.

Responses (beta weights from a GLM) are measured as PSC relative to the fixation baseline. Each panel shows the results from a different experiment: (A) Experiment 1: Violations; (B) Experiment 2: Recovery from Adaptation; (C) Experiment 3: Meaning Judgment. Within each panel, each group of bars shows data from a different fROI (legend at the bottom). Significant differences between the critical, Lexico-semantic (red) and Syntactic (blue), conditions are marked with *’s. (The significance of the reality-check analyses that establish the sensitivity of the language fROIs to the lexico-semantic and syntactic manipulations relative to the control conditions is not marked; please see Results and Table 3.) Brain images at the bottom show the broad masks used to constrain the selection of the individual fROIs (these are not the fROIs themselves, which were defined using the sentences>nonwords contrast separately in each participant, as described in the text, and could thus vary within the borders of the masks depicted here).

Figure 6:. Responses in fROIs defined by the Syntactic>Lexico-semantic contrast to the critical conditions in Experiments 1–3.

Participant-specific fROIs were defined, within the borders of each mask (Figure 5), as the top 10% of voxels showing the strongest Syntactic>Lexico-semantic contrast effect in the corresponding experiment. These fROIs were defined based on half the data from that experiment, and then the other (independent) half were used to estimate the effect size of this same contrast (i.e., estimate the replicability of the contrast effect). Conventions are the same as in Figure 5, with one exception: in panels A and C, parts of the y-axis at the top or bottom have been cut out (marked by two parallel horizontal tick marks) in order to stretch the bars more and accentuate differences across conditions when those appeared. In these visually edited cases, distance between the most extreme 1–2 data points and their corresponding bars are not at scale. Those data points are colored in gray. Differences between the Lexico-semantic (red) and Syntactic (blue) conditions are marked with *’s.

Figure 7:. Responses in fROIs defined by the Lexico-semantic>Syntactic contrast to the critical conditions in Experiments 1–3.

This figure depicts data from a parallel analysis to that depicted in Figure 6; here, participant-specific fROIs were defined as the top 10% of voxels showing the strongest Lexico-semantic>Syntactic contrast effect in the corresponding experiment, and the size of this contrast was then estimated in an independent part of the data (this is the opposite contrast to the one used in Figure 6). Conventions are the same as in Figures 5,6, with the addition of the following: non-significant effects with p<0.10 are marked with *s above tildes.