Difficult to evaluate, with potential yellow flags.
Read full explanation
Public pre-print. Submitted to Psychological Review as a theory of human intelligence, figure copyright from cited researchers obtained during revision.
This paper claims that internally deployed language functions as the symbolic operating system of human cognition, mediating not just formalized reasoning but subconscious processing, perception, memory encoding, and conscious experience itself. This is not a claim that thought is reducible to language, but that language is the primary symbolic interface through which the brain's distributed systems are networked, accessed, and coordinated.
Abstract
This paper uses previously unexplained findings to outline the causal mechanisms underlying human reasoning as a consequence of language. Internal deployment of language functions as the symbolic system underlying human cognition, with a dual use in formal reasoning and in directing the subconscious. Sufficiently sublimated language functions within the mind to build an abbreviated representation of reality and the self, with language functioning as the internal symbols manipulated to activate the relevant neurology for internal simulations of varying resolution and conscious awareness. This synthesis coherently explains previously disparate phenomena in intelligence research: the Flynn Effect and its reversal, cognitive deficits correlating with language-impairment, the function of the Default Mode Network, reasoning emergence (and limits) in language models, the split brain’s ‘alien hand’, and the mechanisms driving human consciousness.
I. Background/Introduction
‘Intelligence’ for this paper is defined as having an internal model of reality and having the ability to update that model with new information.
This synthesis builds an argument indirectly. Through retrodictively explaining anomalies in the current intelligence literature through this framework it eliminates competitor theories through their inability to cleanly account for the other listed anomalies. This relies on a reinterpretation of findings from pharmacology research, sociology, developmental psychology, literacy research, neuroscience, and artificial intelligence.
Language models (LMs) have exploded in prominence after the deployment of OpenAI’s GPT-3, which demonstrated what humans approximate to “intelligence” and reasoning outside of the initial domains of training, for reasons that were previously unknown. Popular models are OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude. They are trained by using machine learning to generate statistical associations with language from a corpus of data composed of books, forum posts, and research papers that are online. The rise of human-like reasoning in machines through next-word prediction models, with these models developing reasoning in domains beyond language itself, and exclusivity to language models specifically, is considered unexplained.
The Flynn Effect is approximately 3 to 7 points gained per generation in IQ across a variety of measurements. Discovered by James Flynn in the 1950’s in his rebuttal to The Bell Curve and race essentialist theories of human intelligence. (Flynn, 1987) The Flynn Effect was discovered as a consistent rise in IQ over generations that has since leveled off in the modern day, and in developed nations, started to reverse. Both the rise and fall of intelligence measures (with one exception) at the societal level is considered unexplained.
The “Reading Wars” (Seidenberg, 2013, Hanford, 2018) were a series of educational policy implementations in the United States and conflicts from the 1970s to the present focused around the displacement of phonics and grammar education in favor of ‘guessing’ words as the dominant method to teach children how to read. Phonics universally won out in education research at the time, but the “Whole Language” methods of education took hold and dominated for decades. The ‘Whole Language’ education is considered responsible for a significant portion of the partial collapse in modern literacy (Seidenberg, 2013).
The human brain has significant anatomy specifically dedicated to linguistics and has unique structure and size compared to primates. Regions of the brain specifically dedicated to interpreting and expressing language are known as the Broca’s area and Wernicke’s area. These regions are suspiciously active in non-verbal tasks (Amalric, 2017). The damage in the form of lesions and strokes to the linguistic centers of the brain cascades into language agnostic learning and is considered unexplained.
II. Hypothesis: Language as the Cognitive Substrate
Internal deployment of language functions as the principal foundation for human cognition, with a dual use in formal reasoning and directing the subconscious. Language functions as the ‘operating system’ of the brain, and networks the different systems of the brain, from which the thing we consider a ‘conscious’ experience manifests as inner dialogue, personal narrative, meaning, and patterns of thought. Language is a symbolic system that is used to process and filter sensory information and encode memory.
If it is the case that humans use language to structure their own internal thought processes then as modern societies became less language oriented, it logically follows that the ability for the average person to engage with and deploy sophisticated language atrophied in accordance with these technological develops, as did their ability to deploy language internally to architect their consciousness. Extending this theory of human cognition, it is impossible for human text to not leave traces of our reasoning patterns within written communication. When a statistical model is trained on human text it will partially reconstruct the reasoning capability of human minds, but because language models lack of our biological subconscious processing this predicts a hard limit of language model intelligence.
The rest of this paper is extrapolated from this line of reasoning.
III. Constellation of Evidence
Psychotherapy:
Why does therapy make people more intelligent?
The nature of this theory presupposes that, if language is deployed internally to shape thought, there must be evidence that using language internally could be trained to consistently increase human intelligence on the individual level. A 2013 RCT from Bastos et al. followed patients on anti-depressants while tracking their intelligence and accidentally found exactly that pattern.
272 adult patients with depression were randomized to receive Long-Term Psychodynamic Psychotherapy (LTPP), fluoxetine monotherapy, or their combination for a 24 month period. The Wechsler adult intelligence scale version III (WAIS-III) was the primary neuropsychological measure, and the trends in figure 1 are striking (Bastos, Guimarães, & Trentini, 2013).
Fig 1.
Psychotherapy was the only factor that consistently improved intelligence across all measures over two years amongst a 200+ person RTC, and temporally led the dual intervention treatment. The trend to note is that there is no measure where pharmacological intervention shows intelligence gains prior to the psychotherapy, despite fluoxetine being among the first reliable antidepressants (Wong, Bymaster, & Engleman, 1995) to reduce depression symptoms. The ‘brain fog’ reported by users of anti-depressants is reflected in the combination treatment, which lags behind the pure psychotherapy in intelligence gains.
The gains in treatment groups are able to be directly and causally attributed to cognitive therapy – a focused approach on using language internally to redirect thought. The purely pharmacological therapy, despite being reliably able to reduce depressive symptoms, had no increase in intelligence measures, which rules out depressive symptoms as a cause of reduced cognition, or regression to baseline.
The fact that these results were found in the LTPP and fluoxetine combination (COM) may suggest that the interaction of both treatments in moderate depressed patients may cause improvement, not only in cognition, but also in metacognition, once apparently there is an increase in the capacity to think about actions, consequences and judge their implications, controlling one’s own knowledge. This has important connections with psychodynamic concepts, such as capacity of abstraction, insight, empathy and representation of self.
The researchers noted an abbreviated hypothesis of the theory themselves, but lacked the breadth presented here for a fuller picture. With a causal experiment to validate language at the heart of intelligence then it should be possible to see gains in literacy reflected in gains of intelligence on the scale of an entire society.
The Flynn Effect:
This theory not only predicts the Flynn Effect as a society becomes more literate, it makes specific predictions about the causes and variables that are behind the stagnation and reversal in the modern day. Pietschnig et al.’s (2015) meta analysis of intelligence data (Fig 2.) shows the previously reliable gains made in various tests of intelligence since the 1940’s that was dubbed the Flynn Effect:
Fig 2.
The Similarities Subset – an explicit test of linguistic ability – had already started flattening from 1990 to 2000.
In the modern day, the Flynn Effect in developed countries (Teasdale, 2008) is now in reverse. Research shows there are “minimal positive and negative Flynn Effects in developed countries” while the Flynn Effect continues to trend upward in developing countries (Dworak, 2023). This implies there is a divergence in the past 20 years in developed nations that stunts previously consistent generational gains in human intelligence.
With the additional context of the Reading Wars and the compounding effect of Whole Language teaching practices on perpetuating illiteracy in America (Seidenberg, 2013), we can explain the recent drop in intelligence measurements as negative environmental impacts on the ability to use and practice sophisticated language.
Per Figure 3, measures from the US Department of Education show adult illiteracy (Below Level 1) has tripled from 2017 to 2023, with no signs of reversing…
Fig 3.
If this framework is correct, then as populations gained basic literacy they would produce reliable gains in intelligence. In modern day America all intelligence metrics are falling, with visual intelligence as a notable exception. Measures of visual intelligence have diverged from other intelligence measures and continue their rise in developed countries while the rest stagnate or decrease, consistent with this framework.
“There is debate about what’s causing it, but not every domain is going down; one of them is going up,” Dworak said. “If all the scores were going in the same direction, you could make a nice little narrative about it, but that’s not the case. We need to do more to dig into it.”
— Elizabeth Dworak, Health Psychologist & Statistician “Americans’ IQ scores are lower in some areas, higher in one”
A society that is now dominated by TV, smartphones, gaming, and short form video content would produce the initial rise with literacy and the decline from reading and writing becoming deprioritized on the societal scale. The Flynn Effect was accidentally capturing the intelligence gained by an increasingly literate population, and as we have an increasingly illiterate society the intelligence trend begins to reverse. There is no other theory that can explain all measurements of intelligence rising for decades, and falling — with a specific exception for visual intelligence continuing to rise — in any way that is remotely coherent.
Aphasia:
Aphasia (or dysphasia) is an impairment in a person’s ability to comprehend or formulate language due to dysfunction in specific brain regions. This impairment can be due to strokes, neurological degradation, brain tumors, or physical trauma. If this theory of cognition is correct, then direct damage to linguistic centers of the brain should have knock-on effects to measurably diminish a person’s cognition, even in language agnostic tasks.
Research consistently shows aphasia has co-occurring non-linguistic cognitive deficits in areas beyond language such as attention, memory, executive functions and learning.
Participants complete two computer-based learning tasks that require them to categorize novel animals based on the percentage of features shared with one of two prototypes. As hypothesized, healthy controls showed successful category learning following both methods of instruction. In contrast, only 60% of our patient population demonstrated successful non-linguistic category learning. Patient performance was not predictable by standardized measures of cognitive ability. Results suggest that general learning is affected in aphasia and is a unique, important factor to consider in the field of aphasia rehabilitation.
– Non-linguistic learning and aphasia: Evidence from a paired associate and feedback-based task (Schuchard, Middleton, & Thompson, 2013)
The researchers understate the case, the difference data is staggering:
Fig 4.
The researchers were surprised by their own findings:
“With respect to patient characteristics, language profile and learning ability, results suggest that learning ability is unrelated to demographic variables such as age, months post onset of stroke and years of education. We had predicted that learning ability might be predicted by scores of executive function. Instead, learning scores did not correlate with any of the standardized measures obtained (AQ, BNT, or CLQT scores of memory, executive function, attention and visuospatial skills). These findings are consistent with previous studies that have failed to find a predictable relationship between verbal impairments or demographic variables and skills in nonverbal domains (Basso et al., 1973; Chertkow et al., 1997; Helm-Estabrooks, 2002). Findings further suggest that category learning ability is distinct from skills measured by the CLQT.”
This is consistent with the model of language as a substrate for cognition, and previous research (Murray, 2012) into aphasia’s impact on cognition, memory, and reasoning. The linguistic centers are the focal point of learning, memory, and consciousness itself. Inevitably, when linguistic systems are damaged the effects will cascade far beyond what should be a mere communication system.
The Angular Gyrus:
If this theory is correct, then the sensory systems of the brain must be constantly connected to language centers at the level of neuroanatomy. From this perspective, this explains the nature of the Default Mode Network (DMN), but also the angular gyrus. The relationship to the angular gyrus is the bridge between the language areas of the brain and other sensory regions for the language to organize the inner narrative as the continuous experience of “consciousness”.
Fig 5.
This brief review aimed to bring together previous findings to construct a unified picture of the AG during all processes, from perception to action. It highlights the integrative role of the AG in comprehension and reasoning—for instance, when manipulating conceptual knowledge, reorienting the attentional system toward relevant information, retrieving facts for problem solving, and giving meaning to external events based on stored memories and prior experiences.
– The Angular Gyrus: Multiple Functions and Multiple Subdivisions (Seghier, 2013)
This also explains why the anatomy of the human angular gyrus is so distinct from other primate species (Niu 2023) and why language centers are so suspiciously active in non-linguistic activities (Seghier, 2013). This model of human cognition maps cleanly onto the findings of neuroscience both in what we experience and what separates us from our primate siblings. As we sleep the language centers of the brain engage in consolidation and linguistic compression to encode our waking memories into long term memories, which explains why sleep is so critical to learning (Peigneux, 2001). The phonological loop and memory correlating with linguistic capability is not unexpected, but obvious from the perspective of this framework.
Suppressing the Substrate:
If language is the substrate of cognition, then it should be easy to interrupt experimentally, and should be interrupted with a systemic pattern biased by language. In fact, this is a consistent finding — known as articulatory suppression (Hanley 2003).
A 1964 paper (Conrad, 1963) discussed two experiments that hinted toward the linguistic substrate:
Sequences of 6 letters of the alphabet were visually presented for immediate recall to 387 subjects. Errors showed a systematic relationship to original stimuli. This is held to meet a requirement of the decay theory of immediate memory.
The same letter vocabulary was used in a test in which subjects were required to identify the letters spoken against a white noise background. A highly significant correlation was found between letters which confused in the listening test, and letters which confused in recall.
The role of neurological noise in recall is discussed in relation to these results. It is further argued that information theory is inadequate to explain the memory span, since the nature of the stimulus set, which can be defined quantitatively, as well as the information per item, is likely to be a determining factor.
– ACOUSTIC CONFUSIONS IN IMMEDIATE MEMORY
Essentially why did errors in immediate memory show systematic patterns related to acoustic similarity?
387 subjects viewed sequences of 6 letters (Letter set: B C P T V F M N S X) presented visually, wrote sequences immediately after presentation and their results from trials with exactly one substitution error showed consistent systemic patterns of substitution. When subjects made memory errors, substitutions weren’t random—they systematically confused acoustically similar letters.
This is due to the internal substrate being slightly tangled as a feature of the language itself, not a deficit of the brain. The irrelevant speech effect (Leist, 2025) is another finding that human speech sounds in the background of experiments reliably degrades recall and learning, similarly tangling the linguistic systems within the mind.
Fig 6.
Why would a linguistic system that should be a simple communication module have such broad and consistent impacts on learning, memory, perception, and recall?
Large Language Models:
Large Language Models (LLMs/LMs) are the most immediately testable tool of this framework, and a strong casual connection that proves language is the fundamental foundation of human cognition. An LLM has no body, is pure next word token prediction, has no internal model, and has no parents from which genetic analysis can be done, and so most competing explanations of intelligence die here.
In 2022 Google Research tested the scaling law behind theories of machine learning, and found emergent properties unique to language models when they were scaled far beyond what the conventional wisdom in the field considered reasonable. Researchers found language models scaled to billions of parameters could engage in a level of novel reasoning. Not partial reasoning, not hallucination, but genuine reasoning in domains that the text models were not trained for.
The cognitive patterns we implicitly encoded in our language weren’t anticipated by machine learning scientists, which is why they underestimated how intelligent a token prediction language model could be when scaled. Despite the GPT-4 model being trained purely on text, without having any experience of a visual world, it developed a visual intuition in order to draw with LaTex (Bubeck et al., 2023):
Per the authors of “Emergent Abilities of Large Language Models”(Wei et al., 2022):
“Overall, more work is needed to tease apart what enables scale to unlock emergent abilities.”
[...]
“Emergent abilities can span a variety of language models, task types, and experimental scenarios. Such abilities are a recently discovered outcome of scaling up language models, and the questions of how they emerge and whether more scaling will enable further emergent abilities seem to be important future research directions for the field of NLP.”
Working backwards from this novel interpretation of language as the operating system for human cognition, the causal root of LM reasoning emergence becomes an obvious consequence of training on language. These models are capturing the gestalt patterns of human reasoning that humans cannot help but imprint in our text. In the years since this reasoning emergence, technology companies have assumed that continuing to scale language models further could achieve what is colloquially called “general intelligence”, or human-level intelligence, in these language models.
This linguistic theory of cognition not only explains the emergent capabilities of language models, but also explains the inability to scale from their own outputs, their inability to engage in human-trivial reasoning tasks (like making valid moves in a game of chess) and their inability to be scaled beyond the plateau they’ve already begun to hit.
Not only does this framework predict reasoning emergence in LMs, it predicts their limits.
Consider this passage from Redburn, describing ascending a ship’s mast to unfurl sails:
“It is surprising, how soon a boy overcomes his timidity about going aloft.
For my own part, my nerves became as steady as the earth’s diameter… I took great delight in furling the top-gallant sails and royals in a hard blow, which duty required two hands on the yard.
There is a wild delirium about it, a fine rushing of the blood about the heart; and a glad, thrilling, and throbbing of the whole system, to find yourself tossed up at every pitch into the clouds of a stormy sky, and hovering like a judgement angel between heaven and earth.”
It is an example of many artists’ greatest aspiration; to wield language so effectively that it can evoke embodied experience. It is not specific diction, or subject matter, or allusions to the divine; it’s the symphony of it all recreating something to be felt.
Since humans collectively inherit some overlap of our embodied experience we can reduce the amount of required formal logic for interpretation and have greatly reduced formal (grammar) and informal rules (culture) for a shared symbolic system of language. Language in humans is the natural manifestation of the theoretical symbolic reasoning system that is considered a primary requirement for genuine artificial intelligence.
When a statistical model is trained on human language it is trained only on what humans felt compelled to communicate. Language model scaling was an experiment that accidentally served as a unique causal piece of evidence for this framework. Language was distilled down to pure statistical associations; completely divorced from culture, biology, genetics, and the myriad of interfering variables that plague human experiments and from these statistical associations in text – the shadow of reasoning emerged.
Imagine the following:
You are about to leave and automatically begin tapping your pockets, because the subconscious notices you are 120 grams lighter on your right side than usual, are missing a pressure in a back pocket, and so it bubbles up from the subconscious that ‘something important isn’t quite right’. You slow as you walk to the door and realize you forgot your wallet just before you leave.
You begin to verbalize what you did when you got home last night in order to recreate events from your memory, and remember that you put groceries in the fridge. You open the fridge and find your wallet sitting next to the carton of eggs.
Since language models are only capturing and training on the verbalized patterns of reasoning absent the embodied machinery managed by our subconscious models can only scratch the surface of human reasoning. Most of what humans consider the “conscious” experience is riding atop our biology, habits, and an ocean of subconscious operations coordinating and filtering what we experience, which bubbles up information that we formalize and interrogate.
Evidence of model reasoning being an empty pattern is in their inability to train on their own data. When LMs generate their own data (generating text exchanges, articles, fiction, essays, etc.) for their training or use data generated from other models, their reasoning abilities atrophy dramatically. This phenomenon is known as ‘model collapse’ (Shumailov et al., 2023). In humans the ‘conscious’ intelligence formally thinking about the wallet is the “last mile” of millions of years of evolution. It is that ‘last mile’ that LMs have only partially captured.
Artificially intelligent systems remain a difficult and unsolved problem.
Unringing The Bell Curve:
If this theory is correct it must be able to coherently explain the persistent racial gaps in measures of intelligence in a desegregated America, regardless of the social inconvenience. Most research focuses on adults, economics, lead, heritability, and family support. There is no hypothesis from any fields that fully explain the inconsistent correlations between race and IQ at the individual level, while also explaining the cumulative population differences, and the variance by regions that is only partially correlated with socioeconomics.
This theory, as it applies to race and IQ, makes an inevitable prediction; the racial differences in intelligence must somehow be tied back to language. The literacy research on black students is where this theory can explain patterns already found by intelligence researchers and their educational data.
Linguistic interference between AAVE (African American Vernacular English) and SE (School/Standard English). The hypothesis that originally motivated this research was that children who are more familiar with SE forms would suffer less interference from mismatches between oral and written language when learning to read and, hence, less confusion about the regularities of written English (Baratz, 1969b; Labov, 1972). Two versions of this interference hypothesis can be entertained regarding the mechanisms underlying the presumed relationship. First, the major effect of encountering such mismatches might be motivational and attitudinal changes, such that the child becomes less eager to read and less receptive to instruction, impeding academic progress. Although student attitudes were not measured in this study, we think the age differences we observed are not entirely consistent with this view. That is, we would have expected that these attitudinal changes would have been cumulative and, thus, that the magnitude of the correlations between reading scores and familiarity with SE would have grown substantially from kindergarten to second grade. There was no suggestion of that trend in our sample, although the possibility remains that a longitudinal study might reveal the expected increase over time.
Alternatively, the greater disparity between oral and written forms of English for children who are less familiar with SE might make it harder for them to discover and learn particular correspondences between spellings and spoken words. For a child familiar only with AAVE, for instance, the oral counterpart of the written sentence ‘‘Their hands are cold’’ could legitimately be /dejr hæn a co/ (‘‘Deir han’ a’ co’’’). If so, this would give rise to several potentially confusing mismatches between dialects that would not be encountered by a non-AAVE speaking child [...]
No particular mismatch, on its own, would pose a serious impediment to learning to decode, but the accumulation of such discrepancies between oral and written forms could make grapheme – phoneme correspondences seem far less regular than they are (for SE) and, hence, more difficult to master.
– Familiarity With School English in African American Children and Its Relation to Early Reading Achievement. (Anne 2004)
This study examined 217 black school children (K-2nd grade) and measured their familiarity with School English (SE) through sentence imitation tasks and found a strong and consistent correlation with SE familiarity and reading achievement.
Researchers found School English familiarity correlated with reading scores at magnitudes (r=.42 - .59), and these correlations:
Held across all reading measures (word identification, decoding, comprehension)
Remained significant after controlling for memory ability, city, and socioeconomic status (SES)
Consistent for both phonological and grammatical features
While use of AAVE varied by SES, reading achievement showed much stronger correlations with individual SE familiarity than with school-level prevalence of AAVE or SE. Under this framework the SE familiarity is the causal mechanism through which socioeconomic factors affect academic outcomes. This model of human cognition coherently explains gaps between black and white populations when measuring intelligence, variation at the individual level, and the causal mechanism at play to explain differences at both ends of the scale. The explanation is not what is socially convenient, but is what inevitably be the case.
The authors were correct, at the time, that their findings are correlational and do not explain the nature of the relationship between early reading and dialect knowledge. However, if the deployment of language internally is required for human cognition, and that internal use of language can be improved at the individual level, it logically follows that to become adept with internal linguistic deployment requires a frictionless automation of internal linguistic systems. Anything that interrupts the sublimation of language during the learning and development of language will cascade into reasoning capabilities, and explains why linguistically deprived children remain cognitively stunted (Vyshedskiy 2017).
This theory validates, explains, and extends their dialectical interference hypothesis.
The children using AAVE aren’t struggling from just phoneme mismatches, it’s constant code-switching and inconsistent structural rules overlapping on the same language that prevents internal consolidation of the linguistic systems. Children cannot develop fluent deployment of language as a symbolic reasoning tool when they’re switching between systems with different rules, adding friction by translating between dialects, and learning a written language that encodes a different system than their primary speech. Every reading task is exponentially more complicated, which results in a frustrating and difficult experience learning to read for AAVE students. Additional feedback mechanisms in educational settings can compound the experience in either direction. Students that feel reading is a confusing and frustrating experience try to minimize their time reading and become the frustrating students that fall behind. Students that feel reading is enjoyable and make reliable gains in their abilities engage with reading for enjoyment and become ‘good’ students whom the education system elevates and rewards.
Racial gaps in intelligence are real, measurable, and consequential.
These differences are not genetic.
These differences are not immutable.
Cognitive Technology
Given the causal links coupling cognition to our linguistic system it stands to reason that there must be a way for words themselves to impart novel capability into the human mind. This theory does not strain to explain the Chinese children’s advantage in mathematics, which has long been hypothesized as a consequence of counting and numerical representation in Chinese as a language overlapping with mathematical logic.
In Chinese, 13 is “十三” (ten-three), which implicitly embeds the relationship of the base 10 number system 十 (ten) and an addition of 三 (three). For 23 in Chinese it’s “ten-ten-three” instead of the idiosyncratic “twenty three” or “thirteen” in English.
Due to the regularity of the number system, once young Chinese children learn the first ten numbers, they are likely to learn numbers beyond ten quickly and systematically. Furthermore, they are likely able to connect the numbering to later mathematics computations such as regrouping. In contrast, learning numbers beyond ten is more effortful and time-consuming for US children (Miller et al., 1995; Ng & Rao, 2010; Zhang et al., 2017). A comparison of three-, four-, and five-year-old Chinese children’s and US children’s abstract counting revealed language differences starting after the number ten, in which Chinese children demonstrated rapid number learning but US children displayed a noticeable drop-off (Miller et al., 1995). Specifically, most Chinese children could count to 20 but only about half of the US children could (Miller et al., 1995), implying that the Chinese numbering system better supports development of counting than does the US system.
– The role of mathematical language in mathematics development in China and the US (Kung, 2019)
The advantage Chinese children consistently demonstrate in mathematics is one of the most robust findings in early education research and is found before children actually receive any formal education (Miller, 1995). This theory makes the differences in mathematical achievement for the Chinese and Americans an expected outcome of how these languages relate to mathematics, and predicts that education of young students in English requires more explicit linguistic scaffolding and deliberate practice to improve their early grasp on mathematics.
Mathematics is sublimated by the best mathematicians as a language in and of itself, which in turn, reshapes their cognition as it becomes augmented by a secondary language. Expert mathematicians will literally “think mathematically” as a consequence of their expertise, and this theory implicitly rules out the educational myth of a ‘mind for numbers’. Mental arithmetic is implied as an inevitable pathway for children to learn, practice, and sublimate mathematical thinking.
1984:
“Every year fewer and fewer words, and the range of consciousness always a little smaller.”
In the most extreme extrapolation of this theory and its consequences it logically follows that highly divergent species of language within the linguistic taxonomy should have a measurable impact on the cognition ‘in kind’ due to the constraints or catalysts on thought from internalizing specific linguistic features. This proposition is beyond what we measure as an increase or decrease in an abstract reasoning or numeracy, but a change in the fundamental qualia of the consciousness that manifests downstream of the mind’s linguistic habitat.
Consider the following example of valid English:
The man sat, chewing on a reed, pondering what he just read from the little red book. He picked up a new reed and continued to read.
The mental structure required to parse the example requires a tremendous capacity for flexibility and ad-hoc contextual re-interpretation. If language is as deeply connected to cognition as this paper suggests, is it possible that the English language can imprint upon the mind the mental flexibility and creativity required to understand the language? Conversely, would a strictly tonal and character based language like Mandarin enforce a thought pattern of structured perfectionism and pattern recognition?
The experiments have already been done, and the findings are what this framework would predict:
A total of 103 United States students (53 female, 24 male, two non-binary, and 24 non-reporting) and 166 Mainland Chinese students (128 female, 30 male, one non-binary, and seven non-reporting) completed an online survey. The survey includes the STEAM-related creative problem solving, Sternberg scientific reasoning tasks, psychological critical thinking (PCT) exam, California critical thinking (CCT) skills test, and college experience survey, as well as a demographic questionnaire. A confirmatory factor analysis (CFA) yields a two-factor model for all creativity and critical thinking measurements. Yet, the two latent factors are strongly associated with each other (r=0.84). Moreover, Chinese students outperform American students in measures of critical thinking, whereas Americans outperform Chinese students in measures of creativity. Lastly, the results also demonstrate that having some college research experience (such as taking research method courses) could positively influence both United States and Chinese students’ creativity and critical thinking skills.
[...]
The results of this study partially confirmed our second hypothesis and replicated the findings from past studies (Niu et al., 2007; Lun et al., 2010; Wong and Niu, 2013; Tang et al., 2015). As predicted, there was a significant main effect for culture in students’ performance for all six measures in the two-C analysis model. United States students performed better than Chinese students in all three creativity measures, and Chinese students performed better than United States students in all critical thinking measures. Given the diversity in the type of measures used in this study, the results suggest that United States and Chinese students’ performance aligns with the stereotype belief found in the study of Wong and Niu (2013).
– Fostering Creativity and Critical Thinking in College: A Cross-Cultural Investigation. (Park, 2021)
Further research will refine the impact language has on the shape and scope of conscious thought, but this study implies an inevitable connection between the shape of language and the shape of thought. When a language is further from automation and sublimation we don’t just find changes in reasoning ability, but changes in moral reasoning. Researchers find that thinking in a second language consistently changes personality and social reflexes to a more ‘utilitarian’ thinking (Dylman 2025) despite the “person” being fundamentally the same. Until this theory, what other theory of human cognition could make such a finding expected?
As this theory is explored experimentally, future researchers will uncover the deeper impact of languages, or even what qualifies as a language (mathematics, programming languages, musical chords), and how to further probe human consciousness.
The Lies That Bind:
To bring back the definition of the ‘internal model’ specified at the start of this paper, there exists a model within the mind of the world, a model in mind of the self, and a language that is used as a symbolic system to represent the world and ourselves within our internal world. Due to a filtered perception riding atop of our biology, the representation of ourselves can be built of incorrect assumptions resulting in conflating our internal world representations as an accurate reflection of reality, occasionally resulting in the painful experience of ‘human error’.
This narrative construction is not optional, and is constantly building a consistent cause and effect relationship to map the experience of ourselves with our perceptions and decisions. With this theory, informed by lateralization of the mind, we can explain what may be the most important research on human consciousness – the split brain “hallucinations”.
These studies tested patients primarily in the two perceptual domains where processing is largely restricted to the contralateral hemisphere, that is vision and touch. In these early studies, stimuli, for instance objects, that were presented to the left hemisphere either physically in the right hand or as an image in the right visual half-field, could be readily named (as the left hemisphere is dominant for language) or pointed out with the right hand (which is controlled by the left hemisphere).
The patient’s behavior became intriguing when the stimuli were presented in the left visual field or in the left hand. Now the patient, or at least the verbal left hemisphere, appeared oblivious to the fact that there had been a stimulus at all but was nevertheless able to select the correct object from an array of alternatives presented to the left hand or the left visual half-field (see Fig. 1). In a particularly dramatic recorded demonstration, the famous patient “Joe” was able to draw a cowboy hat with his left hand in response to the word “Texas” presented in his left visual half field. His commentary (produced by the verbal left hemisphere) showed a complete absence of insight into why his left hand had drawn this cowboy hat. Another astonishing example involved the same patient.
– Split-Brain: What We Know Now and Why This is Important for Understanding Consciousness (de Hann, 2020)
Lateralization is the preference for a specific hemisphere of the brain to be the dominant controller over a specific function where cross-talk would be destructive, overlapping fine motor control from both hemispheres would result in overlapping and contradicting commands to muscle movement and is relegated to a single hemisphere for simplicity and efficiency. This is not exclusive to humans, frogs have lateralization (Robins, 2006), but what is unique to humans is our advanced linguistic processing that lives in the left brain.
For the purpose of generating a coherent and stable narrative there can only be one narrator, and it makes sense for the narrator to live in the left hemisphere in addition to our linguistic processing hardware. In the case of split brains where the body takes action on behalf of the right brain, the left brain will “hallucinate” causes to maintain coherence. The right brain reading a sign to stand will cause the body to stand, and the left brain will post-hoc justify with a need to stretch as the cause instead of the command it didn’t perceive. This framework’s explanation for the right brain’s ability to still process language after severance is the existence of a subconscious mind and the sublimation of the native language as an intact symbolic system. The fact that the subconscious is such an observably coherent, intelligent, deliberate, and sophisticated intelligence in the human mind running alongside our conscious selves is profound when fully appreciated. If this test was using a new language that hadn’t been sublimated or had slightly scrambled words like ‘stadn up’, Joe’s subconscious would be unable to work without explicit linguistic processing from the left brain. This explains why highly literate individuals can nearly instantly understand the meaning of familiar text without “reading”; the pattern recognition of the symbolic system is the domain of the right hemisphere.
The human experience we define is consciousness is a perpetual, internal, linguistic narrative imbued with meaning by an obligate organ in the left brain.
IV. Discussion
The constellation of evidence presents an unmistakable image when you no longer expect complexity.
Damage the linguistic substrate of an individual (aphasia), weaken the linguistic substrate of a society (screens/Whole Language), strengthen the deployment (psychotherapy), add friction to its sublimation (AAVE), disrupt it with audible language (articulatory suppression), compress it into statistical associations (language models) - and intelligence tracks accordingly, without exception.
From the scale of individuals to societies, this theory is unique in the unification of these findings and revealing the fundamental mechanisms of thought. This framework makes unavoidable predictions at how modules of the brain must network through language areas at the anatomical and cellular level. The simplicity in the mechanism cuts through the Gordian knot of half explanations and unexpected findings, and rhymes with discovering a fundamental truth.
An obvious and immediate experiment would be training a language model purely on human fiction. If this theory is correct, then a language model with no connection to reality will still pick up the reasoning ability humans embedded in our fiction and will be capable of basic reasoning. To my knowledge, this experiment has never been attempted. An immediate biological analog is focal interference on linguistic centers of the brain during learning tasks to see how artificial disruption of language can augment real-time learning and perceptions. Additional important neurological experiments should be run on likely candidates of the obligate organ, a likely candidate being the claustrum as the ‘meaning maker’ of the conscious experience for the left brain and the symmetrical right claustrum functioning for the independent subconscious, extending the work of Koubessi et. al and their research on the disruption of consciousness (Koubessi 2014).
If this theory is correct, I suspect our memory system recycles much of the language system for a ‘hash-like’ compression of personal narrative and the hippocampus plays a role in synchronizing the sensory inputs for a linear experience for memory and simulation. Since internal deployment of language is used to re-simulate events through internal modules like vision or emotion, it follows that various stimuli will not have the same temporal consistency and require adjustment by the hippocampus to be coherent experiences both internally and externally.
Learning is viewed through this framework as the brain is encoding the symbolic ‘linguistic hash’ of the memory for storage, and repeated experience improves the availability and accuracy of that hash. There are many experiments where testing our ability to relate linguistic input to our sensory systems would likely alter both the ability to recall events in memory along with the quality of recollection; it’s not a coincidence that speech, recall and cognition degrade with alcohol consumption. I speculate that individual identity is a constantly updating internal narrative with a mix of subconscious and semi-conscious linguistic components generating and coloring the perception of personal history, and substances that weaken the thought loops of personal narratives can change personality and self perception.
By understanding the mechanisms to train intelligence we can develop novel interventions to improve and predict the outcomes of brain surgery, map brain anatomy to human cognition, and more rigorously test the impact pharmacology has on the mind and conscious experience. Knowledge of linguistic mechanisms deteriorating in the brains of dementia and Alzheimer’s patients changes how interventions can be targeted to strengthen and preserve their linguistic abilities. Most importantly, intelligence can be reliably developed and trained at the individual level and many of the cognitive gaps we see today can be closed at very little cost. The critical window of language acquisition will be taken with the utmost seriousness by parents, institutions and educators.
The most intelligent generation in human history is in our future.
V. Acknowledgements
I’d like to thank my friend Paul for his patience and feedback on this paper. I’d like to thank my father for always reminding me “people aren’t random”.
References:
Amalric M, Dehaene S. Cortical circuits for mathematical knowledge: evidence for a major subdivision within the brain’s semantic networks. Philos Trans R Soc Lond B Biol Sci. 2017 Feb 19;373(1740):20160515. https://doi.org/10.1098/rstb.2016.0515
Baratz, J. C. (1969). A bi-dialectal task for determining language proficiency in economically disadvantaged Negro children. Child Development, 40(3), 889–901. https://doi.org/10.2307/1127197
Basso, A., De Renzi, E., Faglioni, P., Scotti, G., & Spinnler, H. (1973). Neuropsychological evidence for the existence of cerebral areas critical to the performance of intelligence tasks. Brain, 96(4), 715–728. https://doi.org/10.1093/brain/96.4.715
Bastos, A. G., Guimarães, L. S. P., & Trentini, C. M. (2013). Neurocognitive changes in depressed patients in psychodynamic psychotherapy, therapy with fluoxetine and combination therapy. Journal of Affective Disorders, 151(3), 1066–1075. https://doi.org/10.1016/j.jad.2013.08.036
Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (pp. 610–623). https://doi.org/10.1145/3442188.3445922
Bone, J. K., et al. (2025). The decline in reading for pleasure over 20 years of the American Time Use Survey. iScience, Volume 28, Issue 9, 113288. https://doi.org/10.1016/j.isci.2025.113288
Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., Lee, P., Lee, Y. T., Li, Y., Lundberg, S., Nori, H., Palangi, H., Ribeiro, M. T., & Zhang, Y. (2023). Sparks of artificial general intelligence: Early experiments with GPT-4. arXiv preprint arXiv:2303.12712. https://arxiv.org/abs/2303.12712
Charity, A. H., Scarborough, H. S., & Griffin, D. M. (2004). Familiarity with school English in African American children and its relation to early reading achievement. Child Development, 75(5), 1340–1356. https://doi.org/10.1111/j.1467-8624.2004.00744.x
Chertkow, H., Bub, D., Deaudon, C., & Whitehead, V. (1997). On the status of object concepts in aphasia. Brain and Language, 58(2), 203–232. https://doi.org/10.1006/brln.1997.1771
de Haan, E. H. F., Corballis, P. M., Hillyard, S. A., Marzi, C. A., Seth, A., Lamme, V. A. F., Volz, L., Fabri, M., Schechter, E., Bayne, T., Corballis, M., & Pinto, Y. (2020). Split-Brain: What We Know Now and Why This is Important for Understanding Consciousness. Neuropsychology Review, 30(2), 224–233. https://doi.org/10.1007/s11065-020-09439-3
Dylman, A. S., & Zakrisson, I. (2025). The effect of language and cultural context on the BIG-5 personality inventory in bilinguals. Journal of Multilingual and Multicultural Development, 46(2), 392–405. https://doi.org/10.1080/01434632.2023.2186414
Dworak, E. M., Revelle, W., & Condon, D. M. (2023). Looking for Flynn effects in a recent online U.S. adult sample: Examining shifts within the SAPA Project. Intelligence, 98, 101734. https://doi.org/10.1016/j.intell.2023.101734
Hanley, J.R., Bakopoulou, E. Irrelevant speech, articulatory suppression, and phonological similarity: A test of the phonological loop model and the feature model. Psychonomic Bulletin & Review 10, 435–444 (2003). https://doi.org/10.3758/BF03196503
Huang, F., Richter, E., Kleickmann, T., Wiepke, A., & Richter, D. (2021). Fostering creativity and critical thinking in college: A cross-cultural investigation. Frontiers in Psychology, 12, 760351. https://doi.org/10.3389/fpsyg.2021.760351
Koubeissi, M. Z., Bartolomei, F., Beltagy, A., & Picard, F. (2014). Electrical stimulation of a small brain area reversibly disrupts consciousness. Epilepsy & behavior : E&B, 37, 32–35. https://doi.org/10.1016/j.yebeh.2014.05.027
Labov, W. (1972). Language in the inner city: Studies in the Black English Vernacular. University of Pennsylvania Press. ISBN: 0812276582
Leist, L., Lachmann, T. & Klatte, M. Impact of irrelevant speech and non-speech sounds on serial recall of verbal and spatial items in children and adults. Sci Rep 15, 1951 (2025). https://doi.org/10.1038/s41598-025-85855-w
Lessmoellmann, A. “Don’t Count on It.” Scientific American Mind 17, no. 5 (October/November 2006): 74-77.
Lun, V. M.-C., Fischer, R., & Ward, C. (2010). Exploring cultural differences in critical thinking: Is it about my thinking style or the language I speak? Learning and Individual Differences, 20(6), 604–616. https://doi.org/10.1016/j.lindif.2010.07.001
Murray, L. L. (2012). Attention and other cognitive deficits in aphasia: Presence and relation to language and communication measures. American Journal of Speech-Language Pathology, 21(2), S51–S64. https://doi.org/10.1044/1058-0360(2012/11-0067)
Niu, M., Palomero-Gallagher, N. Architecture and connectivity of the human angular gyrus and of its homolog region in the macaque brain. Brain Struct Funct 228, 47–61 (2023). https://doi.org/10.1007/s00429-022-02509-7
Niu, W., Zhang, J. X., & Yang, Y. (2007). Deductive reasoning and creativity: A cross-cultural study. Psychological Reports, 100(2), 509–519. https://doi.org/10.2466/pr0.100.2.509-519
Peigneux P, Laureys S, Delbeuck X, Maquet P. Sleeping brain, learning brain. The role of sleep for memory systems. Neuroreport. 2001 Dec 21;12(18):A111-24. doi: 10.1097/00001756-200112210-00001. PMID: 11742260.
Pietschnig, J., & Voracek, M. (2015). One century of global IQ gains: A formal meta-analysis of the Flynn Effect (1909–2013). Perspectives on Psychological Science, 10(3), 282–306. https://doi.org/10.1177/1745691615577701
Robins, A., & Rogers, L. J. (2006). Lateralized visual and motor responses in the green tree frog, Litoria caerulea. Animal Behaviour, 72(4), 843–852. https://doi.org/10.1016/j.anbehav.2006.01.022
Rhoades, E. A., Luckner, J. L., Shearer, A. E., & O’Connell, L. (2018). Bilateral hearing loss is associated with decreased nonverbal intelligence in US children ages 6 to 16 years. Otolaryngology–Head and Neck Surgery, 158(3), 562–568. https://doi.org/10.1177/0194599817742954
Seghier ML. The angular gyrus: multiple functions and multiple subdivisions. Neuroscientist. 2013 Feb;19(1):43-61. doi: 10.1177/1073858412440596. PMID: 22547530; PMCID: PMC4107834.
Schuchard, J., Middleton, E. L., & Thompson, C. K. (2013). Non-linguistic learning and aphasia: Evidence from a paired associate and feedback-based task. Neuropsychologia, 51(1), 79–90. https://doi.org/10.1016/j.neuropsychologia.2012.10.024
Seidenberg, M. S. (2013). The Science of Reading and its Educational Implications. Language Learning and Development, 9(4), 331–360. https://doi.org/10.1080/15475441.2013.812017
Shumailov, I., Shumaylov, Z., Zhao, Y., Gal, Y., Papernot, N., & Anderson, R. (2023). The curse of recursion: Training on generated data makes models forget. arXiv preprint arXiv:2305.17493. https://arxiv.org/abs/2305.17493
Storfer, M. D. (1990). Intelligence and giftedness: The contributions of heredity and early environment. Jossey-Bass.
Tang, M., Werner, C., Cao, G., Tumasjan, A., Shen, J., Shi, J., Kauffeld, S., & Spörrle, M. (2015). Creative expression and its evaluation on work-related verbal tasks: Comparison of Chinese and German samples. The Journal of Creative Behavior, 49(2), 129–148. https://doi.org/10.1002/jocb.134
Teasdale, T. W., & Owen, D. R. (2008). Secular declines in cognitive test scores: A reversal of the Flynn Effect. Intelligence, 36(2), 121–126. https://doi.org/10.1016/j.intell.2007.01.007
Vyshedskiy A, Mahapatra S, Dunn R (2017) Linguistically deprived children: meta-analysis of published research underlines the importance of early syntactic language use for normal brain development. Research Ideas and Outcomes 3: e20696. https://doi.org/10.3897/rio.3.e20696
Wei, J., Tay, Y., Bommasani, R., Raffel, C., Zoph, B., Borgeaud, S., Yogatama, D., Bosma, M., Zhou, D., Metzler, D., Chi, E. H., Hashimoto, T., Vinyals, O., Liang, P., Dean, J., & Fedus, W. (2022). Emergent abilities of large language models. Transactions on Machine Learning Research. https://arxiv.org/abs/2206.07682
Wong, D. T., Bymaster, F. P., & Engleman, E. A. (1995). Prozac (fluoxetine, Lilly 110140), the first selective serotonin uptake inhibitor and an antidepressant drug: Twenty years since its first publication. Life Sciences, 57(5), 411–441. https://doi.org/10.1016/0024-3205(95)00209-O
Wong, W. C., & Niu, W. (2013). Cultural difference in stereotype perceptions and performances in nonverbal intelligence and creativity. The Journal of Creative Behavior, 47(1), 41–59. https://doi.org/10.1002/jocb.22
Public pre-print. Submitted to Psychological Review as a theory of human intelligence, figure copyright from cited researchers obtained during revision.
This paper claims that internally deployed language functions as the symbolic operating system of human cognition, mediating not just formalized reasoning but subconscious processing, perception, memory encoding, and conscious experience itself. This is not a claim that thought is reducible to language, but that language is the primary symbolic interface through which the brain's distributed systems are networked, accessed, and coordinated.
Abstract
This paper uses previously unexplained findings to outline the causal mechanisms underlying human reasoning as a consequence of language. Internal deployment of language functions as the symbolic system underlying human cognition, with a dual use in formal reasoning and in directing the subconscious. Sufficiently sublimated language functions within the mind to build an abbreviated representation of reality and the self, with language functioning as the internal symbols manipulated to activate the relevant neurology for internal simulations of varying resolution and conscious awareness. This synthesis coherently explains previously disparate phenomena in intelligence research: the Flynn Effect and its reversal, cognitive deficits correlating with language-impairment, the function of the Default Mode Network, reasoning emergence (and limits) in language models, the split brain’s ‘alien hand’, and the mechanisms driving human consciousness.
I. Background/Introduction
‘Intelligence’ for this paper is defined as having an internal model of reality and having the ability to update that model with new information.
This synthesis builds an argument indirectly. Through retrodictively explaining anomalies in the current intelligence literature through this framework it eliminates competitor theories through their inability to cleanly account for the other listed anomalies. This relies on a reinterpretation of findings from pharmacology research, sociology, developmental psychology, literacy research, neuroscience, and artificial intelligence.
II. Hypothesis: Language as the Cognitive Substrate
Internal deployment of language functions as the principal foundation for human cognition, with a dual use in formal reasoning and directing the subconscious. Language functions as the ‘operating system’ of the brain, and networks the different systems of the brain, from which the thing we consider a ‘conscious’ experience manifests as inner dialogue, personal narrative, meaning, and patterns of thought. Language is a symbolic system that is used to process and filter sensory information and encode memory.
If it is the case that humans use language to structure their own internal thought processes then as modern societies became less language oriented, it logically follows that the ability for the average person to engage with and deploy sophisticated language atrophied in accordance with these technological develops, as did their ability to deploy language internally to architect their consciousness. Extending this theory of human cognition, it is impossible for human text to not leave traces of our reasoning patterns within written communication. When a statistical model is trained on human text it will partially reconstruct the reasoning capability of human minds, but because language models lack of our biological subconscious processing this predicts a hard limit of language model intelligence.
The rest of this paper is extrapolated from this line of reasoning.
III. Constellation of Evidence
Psychotherapy:
Why does therapy make people more intelligent?
The nature of this theory presupposes that, if language is deployed internally to shape thought, there must be evidence that using language internally could be trained to consistently increase human intelligence on the individual level. A 2013 RCT from Bastos et al. followed patients on anti-depressants while tracking their intelligence and accidentally found exactly that pattern.
272 adult patients with depression were randomized to receive Long-Term Psychodynamic Psychotherapy (LTPP), fluoxetine monotherapy, or their combination for a 24 month period. The Wechsler adult intelligence scale version III (WAIS-III) was the primary neuropsychological measure, and the trends in figure 1 are striking (Bastos, Guimarães, & Trentini, 2013).
Fig 1.
Psychotherapy was the only factor that consistently improved intelligence across all measures over two years amongst a 200+ person RTC, and temporally led the dual intervention treatment. The trend to note is that there is no measure where pharmacological intervention shows intelligence gains prior to the psychotherapy, despite fluoxetine being among the first reliable antidepressants (Wong, Bymaster, & Engleman, 1995) to reduce depression symptoms. The ‘brain fog’ reported by users of anti-depressants is reflected in the combination treatment, which lags behind the pure psychotherapy in intelligence gains.
The gains in treatment groups are able to be directly and causally attributed to cognitive therapy – a focused approach on using language internally to redirect thought. The purely pharmacological therapy, despite being reliably able to reduce depressive symptoms, had no increase in intelligence measures, which rules out depressive symptoms as a cause of reduced cognition, or regression to baseline.
The researchers noted an abbreviated hypothesis of the theory themselves, but lacked the breadth presented here for a fuller picture. With a causal experiment to validate language at the heart of intelligence then it should be possible to see gains in literacy reflected in gains of intelligence on the scale of an entire society.
The Flynn Effect:
This theory not only predicts the Flynn Effect as a society becomes more literate, it makes specific predictions about the causes and variables that are behind the stagnation and reversal in the modern day. Pietschnig et al.’s (2015) meta analysis of intelligence data (Fig 2.) shows the previously reliable gains made in various tests of intelligence since the 1940’s that was dubbed the Flynn Effect:
Fig 2.
The Similarities Subset – an explicit test of linguistic ability – had already started flattening from 1990 to 2000.
In the modern day, the Flynn Effect in developed countries (Teasdale, 2008) is now in reverse. Research shows there are “minimal positive and negative Flynn Effects in developed countries” while the Flynn Effect continues to trend upward in developing countries (Dworak, 2023). This implies there is a divergence in the past 20 years in developed nations that stunts previously consistent generational gains in human intelligence.
With the additional context of the Reading Wars and the compounding effect of Whole Language teaching practices on perpetuating illiteracy in America (Seidenberg, 2013), we can explain the recent drop in intelligence measurements as negative environmental impacts on the ability to use and practice sophisticated language.
Per Figure 3, measures from the US Department of Education show adult illiteracy (Below Level 1) has tripled from 2017 to 2023, with no signs of reversing…
Fig 3.
If this framework is correct, then as populations gained basic literacy they would produce reliable gains in intelligence. In modern day America all intelligence metrics are falling, with visual intelligence as a notable exception. Measures of visual intelligence have diverged from other intelligence measures and continue their rise in developed countries while the rest stagnate or decrease, consistent with this framework.
A society that is now dominated by TV, smartphones, gaming, and short form video content would produce the initial rise with literacy and the decline from reading and writing becoming deprioritized on the societal scale. The Flynn Effect was accidentally capturing the intelligence gained by an increasingly literate population, and as we have an increasingly illiterate society the intelligence trend begins to reverse. There is no other theory that can explain all measurements of intelligence rising for decades, and falling — with a specific exception for visual intelligence continuing to rise — in any way that is remotely coherent.
Aphasia:
Aphasia (or dysphasia) is an impairment in a person’s ability to comprehend or formulate language due to dysfunction in specific brain regions. This impairment can be due to strokes, neurological degradation, brain tumors, or physical trauma. If this theory of cognition is correct, then direct damage to linguistic centers of the brain should have knock-on effects to measurably diminish a person’s cognition, even in language agnostic tasks.
Research consistently shows aphasia has co-occurring non-linguistic cognitive deficits in areas beyond language such as attention, memory, executive functions and learning.
The researchers understate the case, the difference data is staggering:
Fig 4.
The researchers were surprised by their own findings:
This is consistent with the model of language as a substrate for cognition, and previous research (Murray, 2012) into aphasia’s impact on cognition, memory, and reasoning. The linguistic centers are the focal point of learning, memory, and consciousness itself. Inevitably, when linguistic systems are damaged the effects will cascade far beyond what should be a mere communication system.
The Angular Gyrus:
If this theory is correct, then the sensory systems of the brain must be constantly connected to language centers at the level of neuroanatomy. From this perspective, this explains the nature of the Default Mode Network (DMN), but also the angular gyrus. The relationship to the angular gyrus is the bridge between the language areas of the brain and other sensory regions for the language to organize the inner narrative as the continuous experience of “consciousness”.
Fig 5.
This also explains why the anatomy of the human angular gyrus is so distinct from other primate species (Niu 2023) and why language centers are so suspiciously active in non-linguistic activities (Seghier, 2013). This model of human cognition maps cleanly onto the findings of neuroscience both in what we experience and what separates us from our primate siblings. As we sleep the language centers of the brain engage in consolidation and linguistic compression to encode our waking memories into long term memories, which explains why sleep is so critical to learning (Peigneux, 2001). The phonological loop and memory correlating with linguistic capability is not unexpected, but obvious from the perspective of this framework.
Suppressing the Substrate:
If language is the substrate of cognition, then it should be easy to interrupt experimentally, and should be interrupted with a systemic pattern biased by language. In fact, this is a consistent finding — known as articulatory suppression (Hanley 2003).
A 1964 paper (Conrad, 1963) discussed two experiments that hinted toward the linguistic substrate:
Essentially why did errors in immediate memory show systematic patterns related to acoustic similarity?
387 subjects viewed sequences of 6 letters (Letter set: B C P T V F M N S X) presented visually, wrote sequences immediately after presentation and their results from trials with exactly one substitution error showed consistent systemic patterns of substitution. When subjects made memory errors, substitutions weren’t random—they systematically confused acoustically similar letters.
This is due to the internal substrate being slightly tangled as a feature of the language itself, not a deficit of the brain. The irrelevant speech effect (Leist, 2025) is another finding that human speech sounds in the background of experiments reliably degrades recall and learning, similarly tangling the linguistic systems within the mind.
Fig 6.
Why would a linguistic system that should be a simple communication module have such broad and consistent impacts on learning, memory, perception, and recall?
Large Language Models:
Large Language Models (LLMs/LMs) are the most immediately testable tool of this framework, and a strong casual connection that proves language is the fundamental foundation of human cognition. An LLM has no body, is pure next word token prediction, has no internal model, and has no parents from which genetic analysis can be done, and so most competing explanations of intelligence die here.
In 2022 Google Research tested the scaling law behind theories of machine learning, and found emergent properties unique to language models when they were scaled far beyond what the conventional wisdom in the field considered reasonable. Researchers found language models scaled to billions of parameters could engage in a level of novel reasoning. Not partial reasoning, not hallucination, but genuine reasoning in domains that the text models were not trained for.
The cognitive patterns we implicitly encoded in our language weren’t anticipated by machine learning scientists, which is why they underestimated how intelligent a token prediction language model could be when scaled. Despite the GPT-4 model being trained purely on text, without having any experience of a visual world, it developed a visual intuition in order to draw with LaTex (Bubeck et al., 2023):
Per the authors of “Emergent Abilities of Large Language Models”(Wei et al., 2022):
Working backwards from this novel interpretation of language as the operating system for human cognition, the causal root of LM reasoning emergence becomes an obvious consequence of training on language. These models are capturing the gestalt patterns of human reasoning that humans cannot help but imprint in our text. In the years since this reasoning emergence, technology companies have assumed that continuing to scale language models further could achieve what is colloquially called “general intelligence”, or human-level intelligence, in these language models.
This linguistic theory of cognition not only explains the emergent capabilities of language models, but also explains the inability to scale from their own outputs, their inability to engage in human-trivial reasoning tasks (like making valid moves in a game of chess) and their inability to be scaled beyond the plateau they’ve already begun to hit.
Not only does this framework predict reasoning emergence in LMs, it predicts their limits.
Consider this passage from Redburn, describing ascending a ship’s mast to unfurl sails:
It is an example of many artists’ greatest aspiration; to wield language so effectively that it can evoke embodied experience. It is not specific diction, or subject matter, or allusions to the divine; it’s the symphony of it all recreating something to be felt.
Since humans collectively inherit some overlap of our embodied experience we can reduce the amount of required formal logic for interpretation and have greatly reduced formal (grammar) and informal rules (culture) for a shared symbolic system of language. Language in humans is the natural manifestation of the theoretical symbolic reasoning system that is considered a primary requirement for genuine artificial intelligence.
When a statistical model is trained on human language it is trained only on what humans felt compelled to communicate. Language model scaling was an experiment that accidentally served as a unique causal piece of evidence for this framework. Language was distilled down to pure statistical associations; completely divorced from culture, biology, genetics, and the myriad of interfering variables that plague human experiments and from these statistical associations in text – the shadow of reasoning emerged.
Imagine the following:
Since language models are only capturing and training on the verbalized patterns of reasoning absent the embodied machinery managed by our subconscious models can only scratch the surface of human reasoning. Most of what humans consider the “conscious” experience is riding atop our biology, habits, and an ocean of subconscious operations coordinating and filtering what we experience, which bubbles up information that we formalize and interrogate.
Evidence of model reasoning being an empty pattern is in their inability to train on their own data. When LMs generate their own data (generating text exchanges, articles, fiction, essays, etc.) for their training or use data generated from other models, their reasoning abilities atrophy dramatically. This phenomenon is known as ‘model collapse’ (Shumailov et al., 2023). In humans the ‘conscious’ intelligence formally thinking about the wallet is the “last mile” of millions of years of evolution. It is that ‘last mile’ that LMs have only partially captured.
Artificially intelligent systems remain a difficult and unsolved problem.
Unringing The Bell Curve:
If this theory is correct it must be able to coherently explain the persistent racial gaps in measures of intelligence in a desegregated America, regardless of the social inconvenience. Most research focuses on adults, economics, lead, heritability, and family support. There is no hypothesis from any fields that fully explain the inconsistent correlations between race and IQ at the individual level, while also explaining the cumulative population differences, and the variance by regions that is only partially correlated with socioeconomics.
This theory, as it applies to race and IQ, makes an inevitable prediction; the racial differences in intelligence must somehow be tied back to language. The literacy research on black students is where this theory can explain patterns already found by intelligence researchers and their educational data.
This study examined 217 black school children (K-2nd grade) and measured their familiarity with School English (SE) through sentence imitation tasks and found a strong and consistent correlation with SE familiarity and reading achievement.
Researchers found School English familiarity correlated with reading scores at magnitudes (r=.42 - .59), and these correlations:
While use of AAVE varied by SES, reading achievement showed much stronger correlations with individual SE familiarity than with school-level prevalence of AAVE or SE. Under this framework the SE familiarity is the causal mechanism through which socioeconomic factors affect academic outcomes. This model of human cognition coherently explains gaps between black and white populations when measuring intelligence, variation at the individual level, and the causal mechanism at play to explain differences at both ends of the scale. The explanation is not what is socially convenient, but is what inevitably be the case.
The authors were correct, at the time, that their findings are correlational and do not explain the nature of the relationship between early reading and dialect knowledge. However, if the deployment of language internally is required for human cognition, and that internal use of language can be improved at the individual level, it logically follows that to become adept with internal linguistic deployment requires a frictionless automation of internal linguistic systems. Anything that interrupts the sublimation of language during the learning and development of language will cascade into reasoning capabilities, and explains why linguistically deprived children remain cognitively stunted (Vyshedskiy 2017).
This theory validates, explains, and extends their dialectical interference hypothesis.
The children using AAVE aren’t struggling from just phoneme mismatches, it’s constant code-switching and inconsistent structural rules overlapping on the same language that prevents internal consolidation of the linguistic systems. Children cannot develop fluent deployment of language as a symbolic reasoning tool when they’re switching between systems with different rules, adding friction by translating between dialects, and learning a written language that encodes a different system than their primary speech. Every reading task is exponentially more complicated, which results in a frustrating and difficult experience learning to read for AAVE students. Additional feedback mechanisms in educational settings can compound the experience in either direction. Students that feel reading is a confusing and frustrating experience try to minimize their time reading and become the frustrating students that fall behind. Students that feel reading is enjoyable and make reliable gains in their abilities engage with reading for enjoyment and become ‘good’ students whom the education system elevates and rewards.
Racial gaps in intelligence are real, measurable, and consequential.
These differences are not genetic.
These differences are not immutable.
Cognitive Technology
Given the causal links coupling cognition to our linguistic system it stands to reason that there must be a way for words themselves to impart novel capability into the human mind. This theory does not strain to explain the Chinese children’s advantage in mathematics, which has long been hypothesized as a consequence of counting and numerical representation in Chinese as a language overlapping with mathematical logic.
In Chinese, 13 is “十三” (ten-three), which implicitly embeds the relationship of the base 10 number system 十 (ten) and an addition of 三 (three). For 23 in Chinese it’s “ten-ten-three” instead of the idiosyncratic “twenty three” or “thirteen” in English.
The advantage Chinese children consistently demonstrate in mathematics is one of the most robust findings in early education research and is found before children actually receive any formal education (Miller, 1995). This theory makes the differences in mathematical achievement for the Chinese and Americans an expected outcome of how these languages relate to mathematics, and predicts that education of young students in English requires more explicit linguistic scaffolding and deliberate practice to improve their early grasp on mathematics.
Mathematics is sublimated by the best mathematicians as a language in and of itself, which in turn, reshapes their cognition as it becomes augmented by a secondary language. Expert mathematicians will literally “think mathematically” as a consequence of their expertise, and this theory implicitly rules out the educational myth of a ‘mind for numbers’. Mental arithmetic is implied as an inevitable pathway for children to learn, practice, and sublimate mathematical thinking.
1984:
In the most extreme extrapolation of this theory and its consequences it logically follows that highly divergent species of language within the linguistic taxonomy should have a measurable impact on the cognition ‘in kind’ due to the constraints or catalysts on thought from internalizing specific linguistic features. This proposition is beyond what we measure as an increase or decrease in an abstract reasoning or numeracy, but a change in the fundamental qualia of the consciousness that manifests downstream of the mind’s linguistic habitat.
Consider the following example of valid English:
The mental structure required to parse the example requires a tremendous capacity for flexibility and ad-hoc contextual re-interpretation. If language is as deeply connected to cognition as this paper suggests, is it possible that the English language can imprint upon the mind the mental flexibility and creativity required to understand the language? Conversely, would a strictly tonal and character based language like Mandarin enforce a thought pattern of structured perfectionism and pattern recognition?
The experiments have already been done, and the findings are what this framework would predict:
Further research will refine the impact language has on the shape and scope of conscious thought, but this study implies an inevitable connection between the shape of language and the shape of thought. When a language is further from automation and sublimation we don’t just find changes in reasoning ability, but changes in moral reasoning. Researchers find that thinking in a second language consistently changes personality and social reflexes to a more ‘utilitarian’ thinking (Dylman 2025) despite the “person” being fundamentally the same. Until this theory, what other theory of human cognition could make such a finding expected?
As this theory is explored experimentally, future researchers will uncover the deeper impact of languages, or even what qualifies as a language (mathematics, programming languages, musical chords), and how to further probe human consciousness.
The Lies That Bind:
To bring back the definition of the ‘internal model’ specified at the start of this paper, there exists a model within the mind of the world, a model in mind of the self, and a language that is used as a symbolic system to represent the world and ourselves within our internal world. Due to a filtered perception riding atop of our biology, the representation of ourselves can be built of incorrect assumptions resulting in conflating our internal world representations as an accurate reflection of reality, occasionally resulting in the painful experience of ‘human error’.
This narrative construction is not optional, and is constantly building a consistent cause and effect relationship to map the experience of ourselves with our perceptions and decisions. With this theory, informed by lateralization of the mind, we can explain what may be the most important research on human consciousness – the split brain “hallucinations”.
Lateralization is the preference for a specific hemisphere of the brain to be the dominant controller over a specific function where cross-talk would be destructive, overlapping fine motor control from both hemispheres would result in overlapping and contradicting commands to muscle movement and is relegated to a single hemisphere for simplicity and efficiency. This is not exclusive to humans, frogs have lateralization (Robins, 2006), but what is unique to humans is our advanced linguistic processing that lives in the left brain.
For the purpose of generating a coherent and stable narrative there can only be one narrator, and it makes sense for the narrator to live in the left hemisphere in addition to our linguistic processing hardware. In the case of split brains where the body takes action on behalf of the right brain, the left brain will “hallucinate” causes to maintain coherence. The right brain reading a sign to stand will cause the body to stand, and the left brain will post-hoc justify with a need to stretch as the cause instead of the command it didn’t perceive. This framework’s explanation for the right brain’s ability to still process language after severance is the existence of a subconscious mind and the sublimation of the native language as an intact symbolic system. The fact that the subconscious is such an observably coherent, intelligent, deliberate, and sophisticated intelligence in the human mind running alongside our conscious selves is profound when fully appreciated. If this test was using a new language that hadn’t been sublimated or had slightly scrambled words like ‘stadn up’, Joe’s subconscious would be unable to work without explicit linguistic processing from the left brain. This explains why highly literate individuals can nearly instantly understand the meaning of familiar text without “reading”; the pattern recognition of the symbolic system is the domain of the right hemisphere.
The human experience we define is consciousness is a perpetual, internal, linguistic narrative imbued with meaning by an obligate organ in the left brain.
IV. Discussion
The constellation of evidence presents an unmistakable image when you no longer expect complexity.
Damage the linguistic substrate of an individual (aphasia), weaken the linguistic substrate of a society (screens/Whole Language), strengthen the deployment (psychotherapy), add friction to its sublimation (AAVE), disrupt it with audible language (articulatory suppression), compress it into statistical associations (language models) - and intelligence tracks accordingly, without exception.
From the scale of individuals to societies, this theory is unique in the unification of these findings and revealing the fundamental mechanisms of thought. This framework makes unavoidable predictions at how modules of the brain must network through language areas at the anatomical and cellular level. The simplicity in the mechanism cuts through the Gordian knot of half explanations and unexpected findings, and rhymes with discovering a fundamental truth.
An obvious and immediate experiment would be training a language model purely on human fiction. If this theory is correct, then a language model with no connection to reality will still pick up the reasoning ability humans embedded in our fiction and will be capable of basic reasoning. To my knowledge, this experiment has never been attempted. An immediate biological analog is focal interference on linguistic centers of the brain during learning tasks to see how artificial disruption of language can augment real-time learning and perceptions. Additional important neurological experiments should be run on likely candidates of the obligate organ, a likely candidate being the claustrum as the ‘meaning maker’ of the conscious experience for the left brain and the symmetrical right claustrum functioning for the independent subconscious, extending the work of Koubessi et. al and their research on the disruption of consciousness (Koubessi 2014).
If this theory is correct, I suspect our memory system recycles much of the language system for a ‘hash-like’ compression of personal narrative and the hippocampus plays a role in synchronizing the sensory inputs for a linear experience for memory and simulation. Since internal deployment of language is used to re-simulate events through internal modules like vision or emotion, it follows that various stimuli will not have the same temporal consistency and require adjustment by the hippocampus to be coherent experiences both internally and externally.
Learning is viewed through this framework as the brain is encoding the symbolic ‘linguistic hash’ of the memory for storage, and repeated experience improves the availability and accuracy of that hash. There are many experiments where testing our ability to relate linguistic input to our sensory systems would likely alter both the ability to recall events in memory along with the quality of recollection; it’s not a coincidence that speech, recall and cognition degrade with alcohol consumption. I speculate that individual identity is a constantly updating internal narrative with a mix of subconscious and semi-conscious linguistic components generating and coloring the perception of personal history, and substances that weaken the thought loops of personal narratives can change personality and self perception.
By understanding the mechanisms to train intelligence we can develop novel interventions to improve and predict the outcomes of brain surgery, map brain anatomy to human cognition, and more rigorously test the impact pharmacology has on the mind and conscious experience. Knowledge of linguistic mechanisms deteriorating in the brains of dementia and Alzheimer’s patients changes how interventions can be targeted to strengthen and preserve their linguistic abilities. Most importantly, intelligence can be reliably developed and trained at the individual level and many of the cognitive gaps we see today can be closed at very little cost. The critical window of language acquisition will be taken with the utmost seriousness by parents, institutions and educators.
The most intelligent generation in human history is in our future.
V. Acknowledgements
I’d like to thank my friend Paul for his patience and feedback on this paper. I’d like to thank my father for always reminding me “people aren’t random”.
References:
Amalric M, Dehaene S. Cortical circuits for mathematical knowledge: evidence for a major subdivision within the brain’s semantic networks. Philos Trans R Soc Lond B Biol Sci. 2017 Feb 19;373(1740):20160515.
https://doi.org/10.1098/rstb.2016.0515
Baratz, J. C. (1969). A bi-dialectal task for determining language proficiency in economically disadvantaged Negro children. Child Development, 40(3), 889–901.
https://doi.org/10.2307/1127197
Basso, A., De Renzi, E., Faglioni, P., Scotti, G., & Spinnler, H. (1973). Neuropsychological evidence for the existence of cerebral areas critical to the performance of intelligence tasks. Brain, 96(4), 715–728.
https://doi.org/10.1093/brain/96.4.715
Bastos, A. G., Guimarães, L. S. P., & Trentini, C. M. (2013). Neurocognitive changes in depressed patients in psychodynamic psychotherapy, therapy with fluoxetine and combination therapy. Journal of Affective Disorders, 151(3), 1066–1075. https://doi.org/10.1016/j.jad.2013.08.036
Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (pp. 610–623). https://doi.org/10.1145/3442188.3445922
Bone, J. K., et al. (2025). The decline in reading for pleasure over 20 years of the American Time Use Survey. iScience, Volume 28, Issue 9, 113288.
https://doi.org/10.1016/j.isci.2025.113288
Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., Lee, P., Lee, Y. T., Li, Y., Lundberg, S., Nori, H., Palangi, H., Ribeiro, M. T., & Zhang, Y. (2023). Sparks of artificial general intelligence: Early experiments with GPT-4. arXiv preprint arXiv:2303.12712.
https://arxiv.org/abs/2303.12712
Charity, A. H., Scarborough, H. S., & Griffin, D. M. (2004). Familiarity with school English in African American children and its relation to early reading achievement. Child Development, 75(5), 1340–1356. https://doi.org/10.1111/j.1467-8624.2004.00744.x
Chertkow, H., Bub, D., Deaudon, C., & Whitehead, V. (1997). On the status of object concepts in aphasia. Brain and Language, 58(2), 203–232.
https://doi.org/10.1006/brln.1997.1771
CONRAD, R. (1964), ACOUSTIC CONFUSIONS IN IMMEDIATE MEMORY. British Journal of Psychology, 55: 75-84. https://doi.org/10.1111/j.2044-8295.1964.tb00899.x
de Haan, E. H. F., Corballis, P. M., Hillyard, S. A., Marzi, C. A., Seth, A., Lamme, V. A. F., Volz, L., Fabri, M., Schechter, E., Bayne, T., Corballis, M., & Pinto, Y. (2020). Split-Brain: What We Know Now and Why This is Important for Understanding Consciousness. Neuropsychology Review, 30(2), 224–233.
https://doi.org/10.1007/s11065-020-09439-3
Dylman, A. S., & Zakrisson, I. (2025). The effect of language and cultural context on the BIG-5 personality inventory in bilinguals. Journal of Multilingual and Multicultural Development, 46(2), 392–405. https://doi.org/10.1080/01434632.2023.2186414
Dworak, E. M., Revelle, W., & Condon, D. M. (2023). Looking for Flynn effects in a recent online U.S. adult sample: Examining shifts within the SAPA Project. Intelligence, 98, 101734.
https://doi.org/10.1016/j.intell.2023.101734
Flynn, J. R. (1987). Massive IQ gains in 14 nations: What IQ tests really measure. Psychological Bulletin, 101(2), 171–191. https://doi.org/10.1037/0033-2909.101.2.171
Hanford, E. (2018). Hard Words: Why American kids aren’t being taught to read. APM Reports. https://www.apmreports.org/episode/2018/09/10/hard-words-why-american-kids-arent-being-taught-to-read
Hanley, J.R., Bakopoulou, E. Irrelevant speech, articulatory suppression, and phonological similarity: A test of the phonological loop model and the feature model. Psychonomic Bulletin & Review 10, 435–444 (2003). https://doi.org/10.3758/BF03196503
Helm-Estabrooks, N. (2002). Cognition and aphasia: A discussion and a study. Journal of Communication Disorders, 35(2), 171–186.
https://doi.org/10.1016/S0021-9924(02)00063-1
Huang, F., Richter, E., Kleickmann, T., Wiepke, A., & Richter, D. (2021). Fostering creativity and critical thinking in college: A cross-cultural investigation. Frontiers in Psychology, 12, 760351.
https://doi.org/10.3389/fpsyg.2021.760351
Koubeissi, M. Z., Bartolomei, F., Beltagy, A., & Picard, F. (2014). Electrical stimulation of a small brain area reversibly disrupts consciousness. Epilepsy & behavior : E&B, 37, 32–35.
https://doi.org/10.1016/j.yebeh.2014.05.027
Labov, W. (1972). Language in the inner city: Studies in the Black English Vernacular. University of Pennsylvania Press.
ISBN: 0812276582
Leist, L., Lachmann, T. & Klatte, M. Impact of irrelevant speech and non-speech sounds on serial recall of verbal and spatial items in children and adults. Sci Rep 15, 1951 (2025).
https://doi.org/10.1038/s41598-025-85855-w
Lessmoellmann, A. “Don’t Count on It.” Scientific American Mind 17, no. 5 (October/November 2006): 74-77.
Lun, V. M.-C., Fischer, R., & Ward, C. (2010). Exploring cultural differences in critical thinking: Is it about my thinking style or the language I speak? Learning and Individual Differences, 20(6), 604–616.
https://doi.org/10.1016/j.lindif.2010.07.001
Murray, L. L. (2012). Attention and other cognitive deficits in aphasia: Presence and relation to language and communication measures. American Journal of Speech-Language Pathology, 21(2), S51–S64. https://doi.org/10.1044/1058-0360(2012/11-0067)
Niu, M., Palomero-Gallagher, N. Architecture and connectivity of the human angular gyrus and of its homolog region in the macaque brain. Brain Struct Funct 228, 47–61 (2023). https://doi.org/10.1007/s00429-022-02509-7
Niu, W., Zhang, J. X., & Yang, Y. (2007). Deductive reasoning and creativity: A cross-cultural study. Psychological Reports, 100(2), 509–519.
https://doi.org/10.2466/pr0.100.2.509-519
Peigneux P, Laureys S, Delbeuck X, Maquet P. Sleeping brain, learning brain. The role of sleep for memory systems. Neuroreport. 2001 Dec 21;12(18):A111-24. doi: 10.1097/00001756-200112210-00001. PMID: 11742260.
Pietschnig, J., & Voracek, M. (2015). One century of global IQ gains: A formal meta-analysis of the Flynn Effect (1909–2013). Perspectives on Psychological Science, 10(3), 282–306.
https://doi.org/10.1177/1745691615577701
Robins, A., & Rogers, L. J. (2006). Lateralized visual and motor responses in the green tree frog, Litoria caerulea. Animal Behaviour, 72(4), 843–852. https://doi.org/10.1016/j.anbehav.2006.01.022
Rhoades, E. A., Luckner, J. L., Shearer, A. E., & O’Connell, L. (2018). Bilateral hearing loss is associated with decreased nonverbal intelligence in US children ages 6 to 16 years. Otolaryngology–Head and Neck Surgery, 158(3), 562–568. https://doi.org/10.1177/0194599817742954
Seghier ML. The angular gyrus: multiple functions and multiple subdivisions. Neuroscientist. 2013 Feb;19(1):43-61. doi: 10.1177/1073858412440596. PMID: 22547530; PMCID: PMC4107834.
Schuchard, J., Middleton, E. L., & Thompson, C. K. (2013). Non-linguistic learning and aphasia: Evidence from a paired associate and feedback-based task. Neuropsychologia, 51(1), 79–90.
https://doi.org/10.1016/j.neuropsychologia.2012.10.024
Seidenberg, M. S. (2013). The Science of Reading and its Educational Implications. Language Learning and Development, 9(4), 331–360.
https://doi.org/10.1080/15475441.2013.812017
Shumailov, I., Shumaylov, Z., Zhao, Y., Gal, Y., Papernot, N., & Anderson, R. (2023). The curse of recursion: Training on generated data makes models forget. arXiv preprint arXiv:2305.17493.
https://arxiv.org/abs/2305.17493
Storfer, M. D. (1990). Intelligence and giftedness: The contributions of heredity and early environment. Jossey-Bass.
Tang, M., Werner, C., Cao, G., Tumasjan, A., Shen, J., Shi, J., Kauffeld, S., & Spörrle, M. (2015). Creative expression and its evaluation on work-related verbal tasks: Comparison of Chinese and German samples. The Journal of Creative Behavior, 49(2), 129–148.
https://doi.org/10.1002/jocb.134
Teasdale, T. W., & Owen, D. R. (2008). Secular declines in cognitive test scores: A reversal of the Flynn Effect. Intelligence, 36(2), 121–126. https://doi.org/10.1016/j.intell.2007.01.007
Vyshedskiy A, Mahapatra S, Dunn R (2017) Linguistically deprived children: meta-analysis of published research underlines the importance of early syntactic language use for normal brain development. Research Ideas and Outcomes 3: e20696. https://doi.org/10.3897/rio.3.e20696
Wei, J., Tay, Y., Bommasani, R., Raffel, C., Zoph, B., Borgeaud, S., Yogatama, D., Bosma, M., Zhou, D., Metzler, D., Chi, E. H., Hashimoto, T., Vinyals, O., Liang, P., Dean, J., & Fedus, W. (2022). Emergent abilities of large language models. Transactions on Machine Learning Research.
https://arxiv.org/abs/2206.07682
Wong, D. T., Bymaster, F. P., & Engleman, E. A. (1995). Prozac (fluoxetine, Lilly 110140), the first selective serotonin uptake inhibitor and an antidepressant drug: Twenty years since its first publication. Life Sciences, 57(5), 411–441.
https://doi.org/10.1016/0024-3205(95)00209-O
Wong, W. C., & Niu, W. (2013). Cultural difference in stereotype perceptions and performances in nonverbal intelligence and creativity. The Journal of Creative Behavior, 47(1), 41–59.
https://doi.org/10.1002/jocb.22