E-mail zhcho36 gmail.com Human Cognition and Language Processing with Neural -Lexicon Hypothesis

2025-05-03 0 0 4.49MB 27 页 10玖币

侵权投诉

*E-mail: zhcho36@gmail.com

Human Cognition and Language Processing with Neural-Lexicon

Hypothesis

Zang-Hee Cho1†, Sun-Ha Paek 2, Young-Bo Kim 3, Taigyoun Cho4, Hyejin Jeong1, Haigun Lee 5†

1Neuroscience Convergence Center, Institute of Green Manufacturing Technology, Korea University, Seoul, South Korea

2Department of Neurosurgery, School of Medicine, Seoul National University, Seoul, South Korea

3Department of Neurosurgery, School of Medicine, Gachon University, Incheon, South Korea

4Department of Industrial Design, Hong Ik University, Seoul, South Korea

5Department of Material Science, Institute of Green Manufacturing Technology, Korea University, Seoul, South Korea

†These authors contributed equally to Co-Corresponding author

ABSTRACT

Cognition and language seem closely related to the human cognitive process, although they have not been studied

and investigated in detail. Our brain is too complex to fully comprehend the structures and connectivity, as well

as its functions, with the currently available technology such as electro-encephalography, positron emission

tomography, or functional magnetic resonance imaging, and neurobiological data. Therefore, the exploration of

neurobiological processes, such as cognition, requires substantially more related evidences, especially from in-

vivo human experiments. Cognition and language are of inter-disciplinary nature and additional methodological

support is needed from other disciplines, such as deep learning in the field of artificial intelligence, for example.

In this paper, we have attempted to explain the neural mechanisms underlying “cognition and language processing”

or “cognition or thinking” using a novel neural network model with several newly emerging developments such

as neuronal resonance, in-vivo human fiber tractography or connectivity data, Engram and Hebbian hypothesis,

human memory formation in the high brain areas, deep learning, and more recently developed neural memory

concepts, the neural lexicon. The neural lexicon is developed via language by repeated exposure to the neural

system, similar to multilayer signal processing in deep learning. We have derived a neural model to explain how

human “cognition and language processing” or “cognition and thinking” works, with a focus on language, a

universal medium of the human society. Although the proposed hypothesis is not fully based on experimental

evidences, a substantial portion of the observations in this study is directly and indirectly supported by recent

experimental findings and the theoretical bases of deep learning research.

Key words: Neural Network Modeling, Neural-Lexicon, Neural Resonance, Cognition and Language

Processing, Cognition and Thinking

1. INTRODUCTION

Human cognition and language are the two most

difficult subjects due to their wide variely and the

lack of evidence especially on their neural basis.

Nevertheless, it is worthwhile to study based on

contenparary neuroscience perspective with the

modern logic behind deep learning.

For this study, we have chosen “cognition” and

“language,” or slightly differently “cognition” and

“thinking,” as the central topics. We attempted to

answer the questions from neuroscience

perspectives while leveraging AI tools, specifically

deep learning with multiple signal processing layers,

which we modified with neural bases and termed as

neural lexicons. Neural lexicons are memory units

believed to be developed via languages and other

related neuroscience developments such as newly

developed fiber tractography or connectivity data

neuronal Resonance, Hebbian neural processes, or

the Engram1–6. Language, in particular, appears to

play a major role and is a unique human property

with which we have postulated and developed neural

lexicons, which are similar to the signal processing

layers in deep learning7–15. A general sketch of the

proposed neural pathways or the network for

“cognition and language” or “cognition and thinking”

is provided in Fig. 1, where three neural Lexicons

and two memory units are shown together with

connecting fibers from the sensory areas to the

motor areas. We have provided details of the neural

circuit shown in Fig. 1 subsequently.

We begin with the basic neuronal process as shown

in Fig. 2. With this process, a conceptual neuronal

activity involving previously excited and newly

excited neurons, especially the neurons that are

excited at the same time by the external stimulation,

such as the Hebbian neurons, seem to play

significant a role in the early stages of the neural

signal processing of the sensory cortex.

As a simplified example, the neuronal signal, which

we coined as the “sensogram” is assumed to

represent the total number of activated neurons at a

given moment, which is shown as (see also Fig. 2(a)),

arXiv:2210.12960v1 [q-bio.NC] 24 Oct 2022

2022

*E-mail: zhcho36@gmail.com

𝓐(𝒕) = 𝒇𝟎𝟎+𝒇𝟏𝟏 + 𝒇𝟐𝟏 + 𝒇(𝒕𝟏𝟐) + 𝒇(𝒕𝟑𝟒𝟓):

Total Activated Neurons …. {Eq.1 } Sensogram

where 𝑓

00, 𝑓

11, and 𝑓

21 are the zeroth, first, and

second order noises, respectively, while

𝑓(𝑡12) and 𝑓(𝑡345) are the signal components at the

same time, respectively. From Eq. 1, the total

number of concurrently activated neurons, the

Hebbian neurons or Cluster, can be represented as

neuronal resonance at the primary sensory cortex as

follows,

𝓐′(𝒕) = 𝒇𝟏𝟏+𝒇𝟐𝟏 + 𝒇(𝒕𝟏𝟐) + 𝒇(𝒕𝟑𝟒𝟓): Activated

Neurons by Hebbian rule…. {Eq.2} Pre-Engram

Figs. 2(a) and (b) show the transition from the

sensogram to pre-Engram and, finally, the Engram

formation. From Eq. 2, the Hebbian neurons, from

which one can derive the Engram via the Engram

pattern 𝓑(𝒕) at the secondary sensory area, can be

written as,

𝓑(𝒕) = 𝒇𝟐𝟏 + 𝒇(𝒕𝟏𝟐) + 𝒇(𝒕𝟑𝟒𝟓) …. {Eq.3}

Engram

where 𝑓(𝑡12) and 𝑓(𝑡345) are the components of

the phoneme while 𝑓

21 is the system noise.

Subsequently, the Engram produced is sent to the

higher cortical areas, such as the inferior parietal

cortex (IPC) or lobe. In these areas, it is assumed

that some form of higher-order memory units,

termed the neural Lexicons, are developed via

language, as shown in Fig. 2(c). In the n-Lexicon,

the incoming Engram signal is absorbed by

neuronal resonance with a pattern stored in the n-

Lexicon, which is developed via language. This

results in the pre-Langram or phoneme, which

consists of phonetic components as shown on the

right of the figure[2,5,6]. The sum of phonetic

components that form a phoneme or the pre-

Langram is given as:

𝓒(𝒕) = 𝒇(𝒕𝟏𝟐) + 𝒇(𝒕𝟑𝟒𝟓): Phoneme …. {Eq.4}

Pre-Langram

where 𝑓(𝑡12) and 𝑓(𝑡345) are the essential phonetic

components of the Phoneme. A phoneme is

considered a phonologic element that forms the

basis of a Morpheme or Word, the most diminutive

linguistic form or the semantic atom in the

linguistic field[7,9,10,12,13,15]. As will be shown, with

these bases, we will describe how words, sentences,

cognition, decisions, and, eventually, thinking can

be formulated.

Figure 1. Proposed

overall scheme of the

human cognition and

language processing with

neural Lexicon

hypothesis. Here we

assumed three neural

Lexicons with different

functions, such as neural

code conversion,

grouping, sentence

formation, and decision.

Sentence formation leads

to cognition and,

therefore, “thinking,”

which is the highest

cognitive function. * and

** represent the neural

codes which comprises

pure neuronal code and

language based neural

code, respectively. This

model is based on

recently obtained fiber tractography or connectivity, neuronal resonance, Engram and Hebbian neuronal cluster

formation hypothesis, the neural Lexicon hypothesis or language-based memory formation we have proposed,

and recently developed deep learning logic. These developments, although further experimental evidence is

needed, allow us to hypothesize and conclude consistently and also explain how our human brain works during

cognition and thinking, which are the highest cognitive functions.

Legends; STG: Superior Temporal Gyrus, IPC: Inferior Parietal Cortex, PFC: Prefrontal Cortex, OFC:

*E-mail: zhcho36@gmail.com

Figure 2. (a). A simplified neuronal

model of activated and inactivated

neurons that will generate neural

codes such as the Engram is

illustrated. From this, one can derive

neuronal memories and neural

Lexicons. Shown is a simple

neuronal excitation model in the

primary sensory area. Excited

neurons represent noises and signals

used for the cognition and thinking

process. 𝑓

00, 𝑓

11, and 𝑓

21 are the

noises while the 𝑓(𝑡12) and 𝑓(𝑡345)

are the signal components. Note also

that the Hebbian neuronal cluster,

𝒜′(𝑡), will further be processed to

form an Engram.

(b). Noisy input signal Sensogram,

𝒜(𝑡), which includes the neuronal

cluster formed by the Hebbian rule,

is shown. When the input signal,

𝒜(𝑡) or Sensorgram, including the

Hebbian neuronal cluster (𝒜′(𝑡)), is

entered into Memory–II, where

some form of a memory pattern,

ℬ(𝑡), is stored (ex. Engram pattern),

incoming signals will be absorbed

by neuronal resonance. This

produces a new neuronal cluster

called the Engram. Here we

assumed 𝒜′(𝑡) as a “pre-Engram,”

a neuronal cluster formed by the

Hebbian rule.

(c). A certain memory pattern is

formed and works as a filter for the

incoming neuronal signals by

neuronal resonance. Here the neural

Lexicons assume a certain memory

pattern which is formed by repeated

language exposure. An exemplary

n-Lexicon is shown, which involves

a simple conversion of the incoming

signal by neuronal resonance with

help of a neural pattern, 𝒞(𝑡) ,

formed via a language for example,

pre-Langram for the conversion of

Engram to phoneme. The functions

of the n-Lexicons may vary with

circumstances. n-Lexicons perform

various major functions, from

filtering or selection to organized

grouping based on memory patterns

developed via language.

Legends; 𝒜(𝑡) : Sensogram, total

activated neurons including the

Hebbian cluster and non-Hebbian neurons representing noises. 𝒜′(𝑡): Hebbian Cluster. ℬ(𝑡): A memory pattern

of the Engram. 𝒜′(𝑡): Hebbian Neuronal Cluster. ℬ(𝑡): Engram (Pre-Phoneme). 𝒞(𝑡): A memory pattern of a

Phoneme (Pre-Langram). Langram is a temporary name assigned for “Language-Engram.” ℬ(𝑡): Pre-Phoneme

(Engram). 𝒞(𝑡): Phoneme. 𝑓(𝑡12) and 𝑓(𝑡345): Components of a Phoneme. 𝑓

21: System Noise

*E-mail: zhcho36@gmail.com

2. MATERIALS AND MERHOD

Neuronal Model for the Neuronal Code

Processing - from Sensogram to Engram -

Memory-I and II

The overall model of the neural signal processing

and the related human brain surface (left hemisphere

or cortical surface) is shown in Fig. 3-5. Similar to

the multi-layer deep learning process, we propose a

neural process based on neuro-biological

perspectives using newly proposed memory units,

which are the neural Lexicons or n-Lexicons[14]. We

have already discussed the basis in Fig. 2(c), where

we assumed that the input signal produces a

neuronal cluster known as the Engram. The Engram

was proposed as early as the beginning of the

twentieth century by Semon, later by Hebb, and

more recently by Tonegawa2,5,6. The Engram is

conceived as the result of the input signal, which is

the sensogram 𝓐(𝒕); the sensogram includes the

Hebbian cluster (𝓐′(𝒕))) and other noises. When the

filtered sensogram, which is the pre-Engram 𝓐′(𝒕),

enters the secondary sensory cortex, it is filtered by

the memory (𝓑(𝒕)) an Engram pattern, resulting in

the Engram (𝓑(𝒕)), as shown in Fig. 3(a).

Numerical Approach from the Engram to

Cognition via neural Lexicons, n-Lexicon-I and

After we have obtained the Engram, as shown in

Figs. 3(a) and (b), we begin obtaining n-Lexicons

with which those neural codes such as the Engram

are converted to pseudo-language-based neural

codes such as the phonemes or pre-Langrams. This

first conversion process from the Engram to

phoneme will take place at the front part of n-

Lexicon-I (Ia), as shown in Fig. 3(b). This phoneme

is composed of 𝑓(𝑡12) and 𝑓(𝑡345). In the following

stage, n-Lexicon-I (Ib), a group of phonemes will

take place. This will form a word or morpheme,

which is a true language-based neural code; the

“Word” with the neural pattern which is developed

via language in n-Lexicon-I (Ib). The n-Lexicon-I,

therefore, involves two signal processings, namely,

the conversion of the Engram to the phoneme in n-

Lexicon-I (Ia) and the subsequent formation of the

“Word” by a grouping of the phonemes in n-

Lexicon-I (Ib), which can be written as:

𝓒′(𝒕) = 𝓒𝟏(𝒕) + 𝓒𝟐(𝒕) + 𝓒𝟑(𝒕): Word (Group of

Phonemes)…{Eq. 5} Langram or word

where 𝒞1(𝑡), 𝒞2(𝑡), and 𝒞3(𝑡) are the phonemes.

Figure 3. (a). The first parts of the sensory memory units, such as Memory-I and -II, are shown in the primary

and secondary auditory areas. In the primary auditory area, excited neurons produce signals as shown in the

bottom-left (𝒜(𝑡)). Among the excited neurons, the Hebbian neurons (𝒜′(𝑡)) transmit to the secondary sensory

area, memory-II where memory is formed as an intermediate, long-term, or working memory. In memory-II, it

is assumed that a certain memory pattern, ℬ(𝑡), is formed which should resonate with input signals and further

consolidate the memory. This memory is coined as an “Engram pattern” and will produce neural code, the

Engram. The Engram is the basic neural code playing important roles in the neural circuit for cognitive signal

processing.

*E-mail: zhcho36@gmail.com

(b). The formation of the Engram is followed by the neural–Lexicons, which are the memory units developed via

repeated language exposure and assumed to be located in higher cortical areas. The first neural Lexicon (n-

Lexicon-I) is assumed to be located in the tertiary area in sensory sites, such as the inferior parietal cortex (IPC),

the sensory input integrating area where the supramarginal and angular gyri are located. For this neural Lexicon,

two major functions are envisaged; one for the absorption of the Engram for the conversion of the neural signal

(Engram) into language-based neural code, the phonemes, and subsequent summing or grouping of the phonemes

to form a Langram, word, or morpheme. These processes are illustrated at the bottom, and the resulting word or

Langram is shown on the right. An interesting feature of the Langram is that it is the first neural code based on

language used throughout the neural circuit leading to cognition and thinking.

Legends; ℬ(𝑡): Engram pattern, a memory pattern for Pre-Phoneme. ℬ(𝑡): Engram (Pre-Phoneme). 𝒞(𝑡), 𝒞′(𝑡):

Memory patterns for the Pre-Langram (Morpheme) and Langram (Word). 𝒞(𝑡): Langram or word.

𝑓(𝑡12) and 𝑓(𝑡345): Components of a Phoneme. STG: Superior Temporal Gyrus. IPC: Inferior Parietal Cortex.

Subsequently, n-Lexicon-II involves another

grouping process, as shown in Fig. 4. The result of

the grouping of the words in n-Lexicon-II leads to a

sentence or cognition, which we termed as

“Cognogram” as shown in Fig. 4, which can be

noted as,

𝓓(𝒕) = 𝓒′𝟏(𝒕) + 𝓒′𝟐(𝒕) + 𝓒′𝟑(𝒕): Phrase or

Sentence (Group of words)…{Eq. 6} Cognogram

where 𝒞′1(𝑡), 𝒞′2(𝑡) and 𝒞′3(𝑡) are the words.

It should be noted that the n-Lexicon-II is a memory

unit again with a pattern formed or developed via

language. Incoming Langrams are grouped via

neuronal resonance with a pattern formed at the n-

Lexicon-II. We have drawn a group of fibers on top

of the brain surface, which are the connecting

pathways to the frontal areas from the sensory

integration area, the inferior parietal cortex (IPC) to

the prefrontal cortex (PFC), which we have obtained

with 7T MRI fiber tractography (Cho[1] and see

supplement III. Tractography). This shows how the

sensory integration centers (IPC-supramarginal and

angular gyri) are connected to the high cortical areas

(PFC and orbitofrontal cortex [OFC]) in the frontal

cortices, which are believed to be the central

cognitive center[16–18].

Approach to decision and thinking via neural

Lexicons – n-Lexicon-III and beyond

Further processing of the Cognogram in the n-

Lexicon-III, located in the orbito-frontal cortex

(OFC), constitutes the selection process. OFC is

known to be closely related to the limbic system and,

therefore, emotion. The limbic centers, such as the

ACC (BA32, BA24), are believed to comprise old

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

*E-mail:zhcho36@gmail.comHumanCognitionandLanguageProcessingwithNeural-LexiconHypothesisZang-HeeCho1†,Sun-HaPaek2,Young-BoKim3,TaigyounCho4,HyejinJeong1,HaigunLee5†1NeuroscienceConvergenceCenter,InstituteofGreenManufacturingTechnology,KoreaUniversity,Seoul,SouthKorea2DepartmentofNeurosurgery,Schoolo...

展开>> 收起<<

E-mail zhcho36 gmail.com Human Cognition and Language Processing with Neural -Lexicon Hypothesis.pdf

共27页,预览5页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

E-mail zhcho36 gmail.com Human Cognition and Language Processing with Neural -Lexicon Hypothesis

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: