Integrating when and what information in the left parietal lobe allows language rule generalization

Abstract

A crucial aspect when learning a language is discovering the rules that govern how words are combined in order to convey meanings. Because rules are characterized by sequential co-occurrences between elements (e.g., "These cupcakes are unbelievable"), tracking the statistical relationships between these elements is fundamental. However, purely bottom-up statistical learning alone cannot fully account for the ability to create abstract rule representations that can be generalized, a paramount requirement of linguistic rules. Here, we provide evidence that, after the statistical relations between words have been extracted, the engagement of goal-directed attention is key to enable rule generalization. Incidental learning performance during a rule-learning task on an artificial language revealed a progressive shift from statistical learning to goal-directed attention. In addition, and consistent with the recruitment of attention, functional MRI (fMRI) analyses of late learning stages showed left parietal activity within a broad bilateral dorsal frontoparietal network. Critically, repetitive transcranial magnetic stimulation (rTMS) on participants' peak of activation within the left parietal cortex impaired their ability to generalize learned rules to a structurally analogous new language. No stimulation or rTMS on a nonrelevant brain region did not have the same interfering effect on generalization. Performance on an additional attentional task showed that this rTMS on the parietal site hindered participants' ability to integrate "what" (stimulus identity) and "when" (stimulus timing) information about an expected target. The present findings suggest that learning rules from speech is a two-stage process: following statistical learning, goal-directed attention-involving left parietal regions-integrates "what" and "when" stimulus information to facilitate rapid rule generalization.

Fig 1. Schematic overview of the protocol.

Sessions 1 and 2 were conducted a week apart. Session 1 assessed learning using a single language. In Session 2, generalization was assessed with 2 new languages that followed an analogous structure to the language learned in Session 1. Languages were counterbalanced across the protocol. rTMS intervention order (lPL/POz) was counterbalanced between participants in the intervention group and performed on the same day. The control group followed the exact same protocol but had no fMRI or rTMS intervention. L1, L2, L3 = Languages 1–3. fMRI, functional MRI; lPL, left parietal lobe; POz, midline posterior location according to the 10–20 system electrode location; rTMS, repetitive transcranial magnetic stimulation.

Fig 2. Hypothesized RT slopes for rule and no-rule blocks over repetitions of the incidental rule-learning task.

For each artificial language learned, participants were exposed to blocks with rules, in which the initial word determined the identity of the last word of the phrase, and no-rule blocks, in which the final word could not be predicted based on the first one. (A) Part 1: Reflecting statistical learning, rule blocks are expected to exhibit a greater gain in RTs across trials than no-rule blocks as a consequence of the ability to predict the upcoming occurrence or absence of the target word. The difference between rule and no-rule slopes (learning slope) is thus a measure of statistical learning indicating progressive rule learning in the early stages. (B) Part 2: If participants can benefit from previous learnings to orient attention to the initial element to consistently anticipate the final one, their RTs for rule blocks should plateau in later learning stages and show a sustained difference compared to no-rule blocks throughout (rule effect, that is, the mean difference in RT between rule and no-rule trials). (C) A plateau should also be observed for participants that generalize their attentional focus on initial and final elements to a new language with the same type of dependencies (that is, rule). RT, reaction time.

Fig 3. Incidental rule-learning task results for Session 1.

Slopes for rule and no-rule blocks (N = 54) over task repetitions derived from the mixed model analysis conform to the expected pattern (Fig 2), with a significant learning slope in Part 1 and a significant rule effect with a flat (that is, nonsignificant) learning slope in Part 2. Actual data shown averaged into 6 trial bins (for visualization purposes only; the analysis did not bin the data) with the SEM over the slopes for rule and no rule derived from the mixed model analysis. Data used to generate Fig 3 can be found in S1 Data.

Fig 4

(A) BOLD signal over a ventral frontoparietal network in rule blocks significantly covaries with their corresponding measure of statistical learning (learning slope); the more activity, the greater (that is, more negative) the slope (see also S4 Fig). (B) BOLD signal activity over a dorsal frontoparietal network (rule blocks minus no-rule blocks) covarying with the measure of goal-directed attention (Part 2 rule effect minus Part 1 rule effect); the more activity, the larger the effect. Only significant results (p < 0.05 FWE-corrected at the cluster level, with an additional p < 0.005 at the voxel level and 50 voxels of cluster extent) are shown for both analyses. Neurological convention is used with MNI coordinates shown at the bottom right of each slice. https://identifiers.org/neurovault.collection:8592. BOLD, blood-oxygenation-levelamily-wise error; IFG, inferior frontal gyrus; INS, insula; iPL, inferior parietal lobule; MFG, middle frontal gyrus; MidCing, midcingulum; MNI, Montreal Neurological Institute; PostCG, postcentral gyrus; Precu, precuneus; SMG, supramarginal gyrus; SPL, superior parietal lobe.

Fig 5. Incidental rule-learning task results for Session 2.

Both rTMS POz intervention (left panel) and control group Session 2 (right panel) show the expected pattern of rule learning with a significant rule effect and a nonsignificant learning slope, indicating that attentional focus generalized to the learning of the new languages. In contrast, a significant learning slope under rTMS lPL effects (center panel) suggests a return to the progressive rule learning of early learning stages (that is, statistical learning). Actual data shown averaged into 6 trial bins (for visual purposes only; the analysis did not bin the data) with the SEM over the slopes for rule and no rule derived from the mixed model analysis. Data used to generate Fig 5 can be found in S2 Data. lPL, left parietal lobe; POz, midline posterior location according to the 10–20 system electrode location; rTMS, repetitive transcranial magnetic stimulation.

Fig 6. Illustration of the experimental design of the attention task.

Participants had to judge the pitch of a target syllable presented after a sequence of alternating syllables. The pitch of the target syllable could be either higher or lower than that of the preceding sequence of syllables. Sequences of syllables were presented either rhythmically to engage temporal orienting (attention to “when”) or nonrhythmically (50%), with an otherwise constant trial length. At the same time, the initial syllable of each sequence could be informative or noninformative of the identity of the target syllable (50%), manipulating identity-based attention (that is, attention to “what”). ISI, interstimulus interval.

Fig 7. Attention Task results.

Black and gray shaded bars represent mean RTs with their SEM for informative and noninformative trials, respectively, for the phases with rTMS on POz and lPL, separated into rhythmic and nonrhythmic conditions (***p < 0.0001, pairwise comparison; main effect of identity; *p < 0.006). Data used to generate Fig 7 can be found in S3 Data. lPL, left parietal lobe; POz, midline posterior location according to the 10–20 system electrode location; RT, reaction time; rTMS, repetitive transcranial magnetic stimulation.