According to Noam Chomsky, the purpose of language is not to communicate (given that Chomsky did not want to confuse the communication of animals and plants with that of humans), but rather to establish a cognitive mechanism for thoughts to be generated by the human brain (Bolhuis et al. 2014; Chomsky 2012, 2013, 2019). This mechanism takes two syntactic elements ‘a’ and ‘b’ and merges them to form ‘a + b’: ‘the’ and ‘apple’ are combined to yield ‘the apple’. This process can apply to the results of its own output such that ‘ate’ can be combined with ‘the apple’ to yield ‘ate the apple’. Language is thus built-up from component parts using a process called ‘Merge’.

In a like manner, Merge happens when two sites are stimulated concurrently in the neocortex to establish a Hebbian connection (Hebb 1949). As mentioned in an earlier chapter, electrical stimulation of M1 (i.e., the motor cortex) yields a muscle twitch, but if stimulation of M1 is paired (over many trials) with the stimulation of V1 (i.e., the visual cortex, a non-motor area) then the stimulation of V1 evokes a muscle twitch (Baer 1905; Bartlett, Doty et al. 2005; Doty 1965; Giurgea and Raiciulescu 1957). This output configuration should now be readily combined with paired stimulations of other sites in the neocortex to generate a more complex output configuration to act as a neural integrator for Merge[1]. In short, the work of Pavlov (1927) has provided a neural platform to support Merge. Now, whether language is solely for thinking in humans is a ‘rationalist’ anthrocentric idea going back to the 17 Century, as embraced by Chomsky (Bolhuis et al. 2014; Chomsky 1965, 2012, 2013, 2019); the modern view is that the telencephalon (the neocortex in mammals) was design through evolution for thinking (i.e., for planning and making decisions, Darlington and Lisberger 2020; Glimcher 2004; Gold and Shadlen 2007; Penfield and Roberts 1966), as well as for communicating with conspecifics, which is also part of the thinking process from fish to mammals (Antunes et al. 2011; Arriaga et al. 2012; Fox et al. 2017; Fukutomi and Carlson 2020; Goldman and Nottebohm 1983; Haesler et al. 2007; Herculano-Houzel et al. 2014; Nelson et al. 2013; Ojemann 1991; Prat et al. 2015, 2017; Prather et al. 2017; Ridgeway et al. 2019; Romanski 2012; Scharff and Haesler 2005; Sharif et al. 2024)[2]. That the neocortex mediates thinking in all animals was well appreciated by Pavlov (Michaud 2019).

Dickey et al. (2022) has established that 70-ms duration, 90 Hz ripples are synchronized to within 500 ms throughout the human neocortex. This, memory replay-activity (Wilson and McNaughton 1994), occurs within and between hemispheres (including the parietal, the temporal, and the frontal association cortices) and it involves the rhythmical spiking of pyramidal neurons. This process is presumed to merge information between different sensory modalities during memory consolidation, as required for the acquisition of language (Chomsky 2012, 2013, 2019). Even though declarative information is stored in fragments throughout the neocortex (Tehovnik 2017), as evidenced from single-unit recording and focal electrical stimulation (Brecht and Freiwald 2012; Bruce et al. 1981; Doty 1969; Freiwald and Tsao 2010; Ibayashi et al. 2018; Metzger et al. 2023; Ojemann 1983, 1991; Penfield and Rasmussen 1952; Penfield and Roberts 1966; Rolls 2013; Schwarzlose et al. 2005), each single unit representing a complex percept, such as a visual object or a word, is connected synaptically to an array of neurons that provide a context for the stored information at the single cell level (Doty 1969; Dickey et al. 2022; Lu and Golomb 2023). A single declarative conscious unit, which includes a single ‘consolidated’ neuron plus all its connections within and across the association cortices is established serially via the hippocampus during the consolidation of new information that takes place during immobility and sleep (Dickey et al. 2022; Wilson and McNaughton 1994). Thus, the order of consolidation is critical to memory recall. The hippocampus concatenates the neocortical neurons to facilitate the retrieval process. Indeed, the flow of retrieved information is highly fragmented in hippocampal patients (Corkin 2002; Hassabis et al. 2007ab), much like the percepts evoked by stimulating a single site in temporal or parietal cortex (Penfield and Rasmussen 1952; Penfield 1958, 1959, 1975). Even when stimulating regions of the brain such as the medial eye fields that have been implicated in the learning and generation of movement sequences (Tanji and Shima 1994; Tanji 2001) never does continuous electrical stimulation of neurons here evoke a highly organized sequence of body movements but merely one fragment of movement (Tehovnik et al. 1993, 1994).

When adding new information to the brain (e.g., Chen and Wise 1995ab; Hikosaka et al. 2002), the first task is to determine if there are any existent information stores that can be updated with the new information or whether a new topic must be started afresh. Neocortical libraries of the brain are laid down early in childhood development: young children can acquire language fragments after one trial (Chomsky 1959), whereas adults need more practice to consolidate the fragments, especially when learning a foreign language.[3] By elementary grades two to three, children are introduced to different topics such as language, mathematics, social studies, art, music, and physical education, and maybe even a second language[4]. At these early grades, children are six to seven years of age which means the brain is still adding new neurons via mitosis, a process that begins to slow by the age of twelve (Charvet and Finlay 2018; Sanai et al. 2011; Sorrells et al. 2018), but synaptic rewiring, which is critical for information storage, continues for the duration of one’s life and is therefore central to adult learning (Hebb 1949; Kandel 2006).

Patient HM whose hippocampus was severely damaged was able to engage in dialogue with no difficulty (Corkin 2002). A rapid exchange of speech is dependent on an efference-copy representation, which is mediated by the cerebellum (Bell et al. 1997; De Zeeuw 2021; Guell, Schmahmann et al. 2018; Levinson and Torreira 2015; Loyola et al. 2019; Shadmehr 2020; Tehovnik et al. 2021; Wang et al. 2023)[5]. To carry out a volitional act that requires memory, neocortical-cerebellar loops must be intact as verified with anatomy and optogenetics (Hasanbegović 2024). Thus, it should not be surprising that complete language capability depends on an intact neocortex, hippocampus, and cerebellum (Corkin 2002; Guell, Schmahmann et al. 2018; Hassabis et al. 2007ab; Kimura 1993; Mariën et al. 2017; Penfield and Roberts 1966; Scoville and Milner 1957; Squire and Knowlton 2000; Squire et al. 2001), and it is the neocortical ripple activity that merges the fragments of consciousness, as stored in the neocortex, to produce a continuous stream of thought and/or stream of movement (James 1890). Finally, since the merging of information via ripple activity occurs within 500 ms it can be assumed that this represents one temporal unit of consciousness, which is consistent with previous estimates (based on binocular rivalry, Varela 1999ab).

On the point of the hippocampus being involved in the consolidation of serialized information as required for language, Purandare and Mehta (2022) discovered that a third of the neurons (33%, 3379/10263) in the hippocampal–dentate gyrus, CA1, and the subiculum of (immobile) rodents exhibited selectivity for movie fragments, with elevated firing in specific movie sub-fragments, termed movie-fields (also see: Djksterhuis et al. 2024). Neurons in the hippocampus were more selective for movie fragments than were neurons in the early visual pathways, i.e., in the lateral geniculate nucleus and V1. This is consistent with the observation that transmission of information in the hippocampal fibres is independent and parallel (Knight 1964), thereby shunting the information to specific locations of the neocortex. Furthermore, the hippocampal formation is activated during the learning of new associations even for simple stimuli that are temporally overlapped using a classical conditioning paradigm (Berger and Thompson 1976; Swain et al. 2011). Therefore, all new information is transmitted through the hippocampal formation no matter how simple or complex. What this means is that every time you attend a lecture (and you remain immobile in your seat), the hippocampus is activated according to the serial order of that lecture and the information is most likely retrieved in the same order but only in reverse. Of course, the order of retrieval does not need to match the order of the lecture or the order of pre-exam memorization, but matching either should facilitate the speed at which an examination of the retained information is completed[6]. Finally, without a neocortex language processing in humans is impossible (Kimura 1993; Ojemann 1983, 1991; Penfield and Roberts 1966) and without a hippocampus (but with an intact neocortex and cerebellum), new language associations cannot be consolidated into long-term memory (Corkin 2002).

Consciousness is continuous and it never seems to stop during wakefulness (Chomsky 2013, 2019; James 1890), and as mentioned this process demands lots of energy per neuron in the neocortex (Herculano-Houzel 2011). The neocortex has stored within it a lifetime of conscious streaming that is spontaneously generated throughout one’s life, and it is designed to be constantly active metabolically, even when one is immobile (Shulman et al. 2009; Sokoloff 1977; Troubat et al. 2009). The stream of consciousness is related to what one is learning on a particular day (Hebb 1949, 1968), but two individuals confronted with the same problem on a given day will solve it differently, since each has a distinct neural constitution based on a differential history of learning even if the genetic profile is identical as with identical twins (Noble and Noble 2023)[7]. An ‘ordered’ stream of consciousness is established when preparing to deliver a lecture, for example, which can entail providing a verbal description of the contents of visual images on a slide, and then manually advancing to the next slide until the lecture is complete. Every time one delivers a lecture it is never the same, since there will be interruptions by audience members requesting points of clarification, and by the end of the lecture it is common-place for a lecturer to address questions, which are always varied. Each time a lecture is delivered, the declarative information stored in the neocortex is updated along with modifications to the cerebellar efference-copy code to sculpt motor routines related to the lecture to optimize the movements of the vocal musculature and other body parts; all this is accomplished through never-ending feedback (until death[8]) by way of the body’s sensory receptors.

Consciousness, as well as body movement, is programmed serially and it is dependent on neural loops that engage both the neocortex and cerebellum (Figure 21, Hasanbegović 2024). Therefore, in the case of language, words are concatenated serially in the neocortex, as cerebellar circuits are summoned to evoke corresponding muscle contractions in serial order via the vocal apparatus (Aboitiz 2018; Kimura 1993; Mariën et al. 2017; Ojemann 1983, 1991; Penfield and Roberts 1966; Schmidt and Wild 2014). If one is polylingual each language is stored independently in the neocortex, such that a specific network of neurons with access to a specific neocortical-cerebellar language loop is engaged (Figure 22, Ojemann 1991). Finally, to utter the expression ‘I want to be a scientist’ the following scheme has been proposed based on the neocortex and cerebellum as discussed (Figure 23). The details of how the hippocampal formation facilitates the process of word retrieval as synchronized with a loop will need to be established empirically. Clearly, patient HM (who had a damaged hippocampus) would be able to utter the sentence ‘I want to be a scientist’, but narrating a coherent story from long-term memory would be impossible (Corkin 2002). Most of what we (and other animals) do in life that is of significance requires long-term memory.

Summary:

1. Chomsky’s Merge is mediated by the neocortex by creating associations between the various fragments of language through the process of repetition, even though a single trial may be sufficient to acquire a language fragment as a child[9].

2. The hippocampal formation consolidates serialized information for independent storage in various locations of the neocortex, and this process is established by memory replay-activity that occurs during immobility and sleep.

3. Based on the estimated period of temporal overlap between elements for memory consolidation (occurring within 500 ms), it is presumed that the minimal duration of consciousness is about 500 ms. Indeed, any response that occurs at a latency of below 125 ms (in awake primates) cannot be detected using consciousness.

4. The retrieval of language information is also dependent on the hippocampal formation, such that damage of the hippocampus fragments the details of a narration, since this structure links the fragments distributed and stored in the neocortex.

5. Information is stored throughout the neocortex according to individual declarative conscious units, each of which is embedded within a network of neurons that allow for the restoration of the context at the time of memory consolidation.

6. Language is dependent not only on the neocortex and the hippocampus, but also on the cerebellum for the rapid (and automatic) exchange of information when one is having a dialogue; the efference-copy representations for language are stored in the cerebellum, which convert the declarative code into an executable motor code.

7. Neocortical-cerebellar loops are engaged when writing or speaking, i.e., when one’s thoughts need to be expressed using the motor system. Such loops are used by all vertebrates for the induction of behavior.

Footnotes:

[1] Even though Chomsky believes that language is ready-to-go genetically from birth (Chomsky 1965, 2012, 2019), namely, that language is ‘imprinted’ onto human beings; this imprinting, however, cannot be done without a neocortex (Kimura 1993; Ojemann 1991; Penfield and Roberts 1966; Periera, Fedorenko et al. 2018).

[2] “It was Pavlov who showed that language was a consequence of the human cerebral complexity, and that it objectified the superiority and specificity of the human brain with respect to animal brains. He perceived language as a special type of conditioned reflexes, a second system of signalization, the first one being that of gnosis and praxis of direct thinking by images. To each image will be substituted through education its verbal denomination. Since they name everything, instead of associating images, human beings can directly associate the corresponding names, a system more efficient in maximizing the abstraction capabilities of the human brain” (Chauchard 1960, p. 122, from Michaud 2019).

[3] Many faculties of the brain need to be tuned by the environment during a critical period, and if missed a faculty will not develop (Fine et al. 2003; Hubel and Wiesel1977). In the case of language, the first year of life is critical for the establishment of syntax, and other language attributes develop before puberty (Friedmann and Rusou 2015).

[4] Both Albert Einstein and Noam Chomsky were no fans of the Western educational system, complaining that rote learning is not creative learning (Chomsky 2023; Kremer 2015). But in defense of having a standardized educational system using rote learning (significantly birds learn to sing through repetition with guidance from a teacher, Carouso-Peck, and Goldstein 2018), the human population of the world is now more literate and smarter than ever before. One can object to this standardization of education, but many of our outstanding scientists came from institutions with a standard educational program and it did not seem to interrupt their creative process: e.g., Sir Isaac Newton (education: Trinity College), Hermann von Helmholtz, (education: Friedrich Wilhelm Institut), Nikola Tesla (education: Graz University of Technology), Albert Einstein (education: Federal Polytechnic School), Lord Edger Adrian (education: Trinity College), Sir Charles Sherrington (education: University of Cambridge), Wilder Penfield (education: John Hopkins University), Donald Hebb (education: University of Chicago), Noam Chomsky (education: University of Pennsylvania), Vernon Mountcastle (education: Roanoke College), Edward O Wilson (education: Harvard University), David Hubel (education: McGill University), Thorsten Wiesel (education: Karolinska Institute), Case Vanderwolf (education: McGill University), and Peter H Schiller (education: Clark University), to mention but a few.

[5] As for Chomsky’s ‘Merge’ to be expressed automatically in children (Chomsky 2019), it would require that the cortico-cerebellar loops be programmed genetically such that all the gains at the Purkinje neurons are able to anticipate a linguistic understanding and expression with minimal adjustments of the gains once a child begins hearing and making sounds. That all this is finalized syntactically by the age of one, will need to be verified quantitatively using information theory (Tehovnik and Chen 2015).

[6] Significantly, AI models can learn by the repetition of sequences of ‘neural discharges’, thereby facilitating the retrieval of familiar sequences (citation: Google AI, Aug. 14, 2024) to suggest that AI, like the human brain, has a hippocampal analogue.

[7] Science is built on counter-intuitive thinking that is why MIT searches throughout the world for the best talent; closing one’s borders to this thinking to enhance security diminishes a country’s development. The greatest societies in the world had a porous border for talent, which is independent of race and class.

[8] Patients suffering from ALS (amyotrophic lateral sclerosis) cease to exist consciously once the feedback loops from the body to the brain are completely silenced, and often these patients die shortly thereafter (Birbaumer 2006; Nicolas-Alonso and Gomez-Gil 2012; Tehovnik and Chen 2015).

[9] Chomsky (2019) claimed that the neurons of the brain are much too slow to execute Merge, suggesting that this process is carried out at the subcellular/molecular level as described by Roger Penrose. The transmission duration of a chemical synapse is about one millisecond, and it can take tens of milliseconds for signals to be transmitted from caudal to rostral parts of the neocortex (Schiller and Tehovnik 2015; Yeomans and Tehovnik 1988). If Merge is mediated by a classical-conditioning mechanism (as argued here), a millisecond range would be sufficient to support the process (Gallistel et al. 2022).

Figure 21. Neocortical-cerebellar loops over time. A command is issued from the neocortex (cortex), which has access to the cerebellar cortex at the Purkinje cells (Purkinje) via the pons (Pons). The return portion of the loop passes through the cerebellar nuclei (Nuclei) and thalamus (Thal) en route to the neocortex. Based on the work of Varela (1999ab), the loop duration for the mediation of a fragment of consciousness is about 500 ms, but the information flow for watching a film, reading a book, delivering a speech, or playing a musical instrument is continuous. The loop configuration is based on the anatomical, unit recording, and optogenetic experiments of Hasanbegović (2024), all performed on the mouse and generalized to the primate (Thach et al. 1992).

Figure 22. Neocortical-cerebellar loops per language. A language command to speak is issued from the neocortex (cortex), which has access to the cerebellar cortex at the Purkinje cells (Purkinje) via the pons (Pons). The return portion of the loop passes through the cerebellar nuclei (Nuclei) and thalamus (Thal) en route to the neocortex. According to Ojemann (1983,1991) every language is stored separately in the neocortex according to electrical inactivation experiments done on human subjects. This idea is supported by work on stroke patients, whereby primary languages are often preserved over secondary languages (based on cerebellar stroke, Mariën et al. 2017) and it is well-known that doing language interpretations in real time is difficult, likely because of the neural segregation between the language loops. Irrespective of the type of language, every language is transmitted at a comparable bit-rate of 39 bits per second, which is short of 1 trillion possibilities per second (Coupé et al. 2019). This dovetails with the idea of Chomsky that all humans have a universal-grammar capability (Chomsky 1965), further supporting the idea that we are one species. Whether other mammals have a similar capability is not known, but machine learning is now being used to decipher communications between whales, which have neocortical neurons in excess of 40 billion, e.g., the killer whale (Ridgway et al. 2019; humans have 16 billion neocortical neurons by comparison, Herculano-Houzel 2009). The loop configuration in the figure is based on the anatomical, unit recording, and optogenetic experiments of Hasanbegović (2024), all performed on the mouse and generalized to the primate (Thach et al. 1992).

Figure 23. Pools of declarative conscious units concatenated to produce the phrase ‘I want to be as scientist’. Here both the neocortex and cerebellum are engaged. The neocortex triggers the automated response, and the cerebellum preserves an efference-copy of the response, which can be altered at the level of the Purkinje neurons via sensory feedback. Catecholamine schedule control refers to the transitions between wakefulness and sleep using locus coeruleus norepinephrine, as well as regulating the speed of behavior by way of the midbrain dopaminergic system. The bit rate for transfer of information has been set at 40 bits per second, which is a commonly accepted average value for English (Reed and Durlach 1998). For other details consult earlier chapters on the neocortex and cerebellum.