Negative polarity in quantifiers evokes greater activation in language-related regions compared to negative polarity in adjectives

Abstract

The processing of sentences with negative quantifiers (e.g., few) is more costly than of sentences that contain their positive counterparts (e.g., many). While this polarity effect is robust and reliably replicable, its neurological bases are not well understood. In this study, we use functional magnetic resonance imaging (fMRI) paradigm for 30 participants to assess the polarity effect in sentences with polar quantifiers, and compare it with the polarity effect of polar adjectives. Both in quantifiers and in adjectives, the polarity effect manifests in the anterior insula bilaterally. The polarity effect in quantifiers, however, shows greater activation in the left hemisphere than it does for adjectives. In particular, left inferior frontal gyrus (IFG) and left superior temporal sulcus (STS) show increased activation for polarity in quantifiers than in adjectives, which is the evidence for the specific involvement of the language network in this type of polarity processing. Using the polarity effect in adjectives as a control, we provide further evidence for the linguistic complexity that negative quantifiers implicate on processing.

Appendix

In the main text we illustrate different facets of the Polarity effect through the modality of fMRI. However, we also collected diffusion MRI (dMRI) data from our participants, with a goal of quantifying the relationship of two MRI modalities vis a vis the Polarity effect and hemispheric laterality. We did not find any relationship between the Polarity effect and white-matter pathways. Yet, we report here the dMRI analysis and results due to the importance of publishing also null results (van Assen et al. 2014).

Specifically, we used the dMRI data to identify the arcuate fasciculus, a white-matter pathway known to be important for language and which is known to be left lateralized (Nucifora et al. 2005; Thiebaut de Schotten et al. 2011; Bain et al. 2019). We hypothesized that arcuate asymmetry (dMRI) would correlate with activation asymmetry (fMRI) as well as with the Polarity effect (behavior). We also used the dMRI data to identify the uncinate fasciculus as a control fascicle.

The dMRI data were collected in the same session as the fMRI data (see below for data collection and processing specifics). We identified the arcuate and uncinate for our subjects (example in Fig. 

Fig. 4

Relating dMRI asymmetry and polarity measurements. a Left hemisphere of a sample subject showing the arcuate (blue) and uncinate (pink) fasciculi. b The laterality index (L.I.) of the arcuate fasciculus shows a positive correlation with the ΔRT = ($\frac{{\mathrm{RT}}_{\mathrm{neg}}-{\mathrm{RT}}_{\mathrm{pos}}}{\stackrel{-}{\mathrm{RT}}}$), though not significant: r = 0.2, p = 0.2. c Taking the uncinate fasciculus as a control region, we show no correlation of the L.I. of the uncinate fasciculus with the ΔRT. d Taking the RT as a control, we find no correlation between arcuate L.I. and RT

4), and calculated a laterality index (L.I.) for the arcuate and uncinate in each subject. The relationship between arcuate laterality and the polarity effect was non-significant (Fig. 4b). Our controls showed no relationship between uncinate laterality and the polarity effect (Fig. 4c), nor for arcuate laterality and reaction time (Fig. 4d).

dMRI data collection and analysis

Diffusion data were collected with a diffusion-weighted spin-echo echo-planar imaging sequence, with acceleration factor of 3 and partial Fourier of 6/8. Slice thickness was 1.7 mm, field of view 220 × 220 mm² and matrix size 128 × 128, yielding voxel size 1.7 × 1.7 × 1.7 mm³. The diffusion gradient strength was 45 mT/m and diffusion-weighted gradients were applied for 64 noncollinear directions with b = 2000 s/mm². The repetition time was 4000 ms and the echo time was 96.2 ms. The total acquisition time was about 5 min.

To analyze the diffusion data, we used the Vistasoft (http://github.com/vistalab/vistasoft/mrDiffusion) and MRtrix (version 0.2.12, https://www.nitrc.org/projects/mrtrix/) software packages. Preprocessing steps include corrections for subject motion and eddy currents. From the preprocessed diffusion data, we created a set of streamlines, called the tractogram, which represents the large white-matter pathways that cover the brain. We computed a whole-brain tractogram using MRtrix with a maximum harmonic order (l_max) of 6, and then generated 500,000 streamlines using the sd_prob method (Tournier et al. 2012). To compute the whole-brain tractogram in MRtrix, we used a white-matter mask as both the seed and mask regions, with the default MRtrix parameters (step size 0.2 mm, curvature radius 1 mm, stopping criterion FA < 0.1).

To extract the arcuate fasciculus from the whole-brain tractogram, we used Matlab (2017) to run the software package Automated Fiber Quantification (AFQ; Yeatman et al. 2012 and https://github.com/yeatmanlab/AFQ). We used the same pipeline to identify the left and right uncinate fasciculi, which we used as a control fascicle. To evaluate fascicle laterality, we counted the streamlines in the left and right fascicle and computed the laterality index, L.I. = (right minus left) ÷ (right plus left), where a positive L.I. indicates rightward laterality and a negative L.I. indicates leftward laterality.

We define $\Delta RT=\frac{R{T}_{neg}-R{T}_{pos}}{R{T}_{pos}}$, which reflects the percent change in processing time implicated by the negative antonym. We identify a correlation between ΔRT for the two sentence Types (r = 0.59, p = 0.0006), providing further support for our assumption that quantifier negation and adjective negation share to some extent their underlying cognitive mechanisms.