Evidence from formal logical reasoning reveals that the language of thought is not natural language — part 6/10
Cleaned-text code points [68153, 79848). Bibliography omitted here; source.md includes it.
GS SA Age 50 78 Time post-onset 2.5 years 33 years Etiology Left hemisphere stroke Left subdural empyema and meningitis Previous occupation and educational background Graduate degree; worked in cybersecurity Left school at 15; police sergeant Cognitive assessments - Forward digit span (Wechsler Adult Intelligence Scale) - Pyramid & Palm Trees (3 -picture version) - 3 items - 50/52 - 2 items - 49/52 Grammatical processing Comprehension spoken reversible sentences: - Total (chance = 40/80) - Active (20/40) - Passive (20/40) Comprehension written reversible sentences: - Total (chance = 40/80) - Active (20/40) - Passive (20/40) Auditory grammaticality judgment: - Total (chance = 32/64) Written grammaticality judgment: - Total (chance = 32/64) - 22/80 - 20/40 - 2/40 - 38/80 - 32/40* - 6/40 - 36/64 - 44/64* - 30/80 - 16/40 - 14/40 - 46/80 - 34/40* - 12/40 - 48/64* - 29/64 Lexical processing ADA spoken-word picture matching (total): ADA written-word picture matching (total): PALPA Picture Name (spoken or written): - High-frequency set - Mid-frequency set - Low-frequency set ADA Synonym Judge (combined written and auditory presentation): - Total - High imageability errors - Low imageability errors - High frequency errors - Low frequency errors - 60/66 - 63/66 - 19/20 - 15/20 - 13/20 - 137/160 - 2 - 10 - 4 - 7 - 51/66 - 53/66 - 1/20 - 0/20** - 0/20** - 135/160 - 4 - 12 - 6 - 3 Supplementary Table 3. Extended demographic and language assessment information for participants with aphasia. S.A. was premorbidly a police sergeant and G.S. was premorbidly in cybersecurity. S.A. suffered a subdural empyema, followed by meningitis, leading to a secondary vascular lesion in the left middle cerebral artery and damage to the perisylvian areas. GS suffered a left-hemisphere stroke, also resulting in a vascular lesion in the left m iddle cerebral artery and damage to the perisylvian areas. This table further shows cognitive assessments include the WAIS digit span (Wechsler, 1981), Pyramid and Palm Trees test (3-picture version), identifying semantic relations between pictures (Howard and Patterson, 1992) and WASI matrices (Wechsler, 1999). Grammatical processing assessments from .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 29, 2025. ; https://doi.org/10.1101/2025.07.26.666979doi: bioRxiv preprint Zimmerer et al., (2014) include comprehension of spoken and written reversible sentences and from Linebarger et al. (1983), grammaticality judgments, with rep etition penalties for spoken/auditory tasks. Lexical processing includes the picture matching and synonym judge tasks from the ADA comprehension battery (Franklin et al. 1992) and picture naming task from the PALPA test battery (Kay et al. 1992). * indicates performance above chance under a binomial test at p = 0.05. ** a cut -off was applied for SA’s PALPA naming test: with 1/20 on the high frequency set, the mid - and low -frequency sets were automatically scored 0/20 without further testing. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 29, 2025. ; https://doi.org/10.1101/2025.07.26.666979doi: bioRxiv preprint Supplementary Methods and Materials fMRI data acquisition Whole-brain structural and functional data were collected on a whole -body 3 Tesla Siemens Trio scanner with a 32 -channel head coil at the Athinoula A. Martinos Imaging Center at the McGovern Institute for Brain Research at MIT. T1 -weighted structural image s were collected in 176 axial slices with 1 mm isotropic voxels (repetition time (TR) = 2,530 ms; echo time (TE) = 3.48 ms). Functional, BOLD data were acquired using an EPI sequence with a 90° flip angle and using GRAPPA with an acceleration factor of 2; the following parameters were used: 31 4.4 mm thick near axial slices acquired in an interleaved order (with 10% distance factor), with an in -plane resolution of 2.1 × 2.1 mm, FoV in the phase encoding (A >> P) direction 200 mm and matrix size 96 × 96 voxe ls, TR = 2,000 ms and TE = 30 ms. The first 10 s of each run were excluded to allow for steady state magnetization. fMRI data preprocessing fMRI data were preprocessed and analyzed using SPM12 (release 7487), CONN EvLab module (release 19b), and other custom MATLAB scripts. Each participant’s functional and structural data were converted from DICOM to NIFTI format. All functional scans were coregistered and resampled using B -spline interpolation to the first scan of the first session (Friston et al., 1995). Potential outlier scans were identified from the resulting subject-motion estimates as well as from BOLD signal indicators using default thresholds in CONN preprocessing pipeline (5 standard deviations above the mean in global BOLD signal change, or framewise displacement values above 0.9 mm; Nieto-Castañón, 2020). Functional and structural data were independently normalized into a common space (the Montreal Neurological Institute [MNI] template; IXI549Space) using SPM12 unified segmentation and normali zation procedure (Ashburner and Friston 2005) with a reference functional image computed as the mean functional data after realignment across all timepoints omitting outlier scans. The output data were resampled to a common bounding box between MNI coordinates (−90, −126, −72) and (90, 90, 108), using 2 mm isotropic voxels and 4th order spline interpolation for the functional data, and 1 mm isotropic voxels and trilinear interpolation for the structural data. Last, the functional data were smoothed spatiall y using spatial convolution with a 4 mm FWHM Gaussian kernel. fMRI data modeling For all experiments, effects were estimated using a general linear model (GLM) in which each experimental condition was modeled with a boxcar function convolved with the canonical hemodynamic response function (HRF) (fixation was modeled implicitly, such that all timepoints that did not correspond to one of the conditions were assumed to correspond to a fixation period). Temporal autocorrelations in the BOLD signal timeser ies were accounted for by a combination of high -pass filtering with a 128 s cutoff, and whitening using an AR (0.2) model (first-order autoregressive model linearized around the coefficient a = 0.2) to approximate the observed covariance of the functional data in the context of restricted maximum likelihood estimation. In addition to experimental condition effects, the GLM design included first -order temporal derivatives for each condition (included to model variability in the HRF delays), as well as nuisance regressors to control for the effect of slow linear drifts, subject -motion parameters, and potential outlier scans on the BOLD signal. Details on the Language Localizer. In this task, participants silently read sentences and lists of nonwords in a blocked design. The sentences > nonwords contrast targets cognitive processes related to high -level language comprehension, including understanding word meanings and combinatorial linguistic processing, and has been shown to effectively isolate language areas from nearby functional areas (see Fedorenko et al., 2024 for a review). The task is available for download from https://www.evlab.mit.edu/resources. Following much past work (e.g., Lipkin et a. 2022; Malik-Moraleda et al., 2022; Tuckute et al., 2024), to define the language functional regions of interest (fROIs), we used a set of group -level parcels —derived from a large group of independent .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 29, 2025. ; https://doi.org/10.1101/2025.07.26.666979doi: bioRxiv preprint participants performing the same task —and intersected these parcel s with individual activation maps; within each parcel, we selected the 10% of most responsive voxels in each participant as that participant’s fROI. The responses in these language fROIs to each condition of the critical tasks were estimated; to estimate the responses to the conditions of the language localizer, we used a split-half approach where one run is used to define the fROIs, and the other run to estimate their responses, so as to avoid non - independence (e.g., Kriegeskorte, 2011). Details on the Spatial Working Memory MD Localizer. In this task, participants are presented with 3x4 grids within which sequences of locations flash up (one at a time for a total of four locations in the easy condition, and two at a time for a total of eight locations i n the hard condition). Participants are asked to keep track of the locations, and at the end of each trial, they are shown two sets of locations and asked to choose the set they just saw. The hard > easy contrast targets cognitive processes broadly related to performing demanding tasks —what is often referred to by an umbrella term ‘executive function processes’. To define the MD fROIs, we used a set of group-level parcels and, similar to the language localizer, selected the 10% of most responsive voxels in each participant as that participant’s fROI. In-scanner behavioral data. All fMRI participants were engaged during the logical reasoning tasks: the overall response rate was 100% for the induction task, 99.63% for the verbal deductive reasoning task, and 100% for the deductive matrix reasoning task. The accuracies were high: induction task —0.77 (0.74 for the pre -solution induction trials and 0.89 for the post-solution application trials); verbal deductive reasoning task—0.81 (0.67 for the harder Modus Tollens condition and 0.95 for the easier Modus Ponens condition ); deductive matrix reasoning task—72.13% (47% for the harder condition and 97.25% for the easier condition). And the reaction times (RTs) mirrored the accuracy data, with longer RTs for the harder conditions: induction task—4.17s (4.49s for the pre -solution induction trials and 2.88s for the post -solution application trials); verbal deductive reasoning task—11.31s (13.52s for the harder Modus Tollens condition and 9.10s seconds for the easier Modus Ponens condition); deductive matrix reasoning task —2.364s (3.73s for the harder condition trials and 0.997s for the easier condition trials). .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted July 29, 2025. ; https://doi.org/10.1101/2025.07.26.666979doi: bioRxiv preprint