Human brain theory

ISBN 978-3-00-068559-0

Monografie von Dr. rer. nat. Andreas Heinrich Malczan

16.   The emergence of language understanding

 

Having a memory means being able to memorise something. For example, the image of a particular bird in an ornithology book.

But it can also be more complicated: remembering a scene from a film means that you don't just save one image, but a sequence of images. In this case, the individual images are interlinked - that's what we want to call it here. Chained may mean that, apart from the first and last frame of the scene, each frame has exactly one predecessor and each frame has exactly one successor. The first image naturally has no predecessor and the last image has no successor. This means that concatenated signals are not simple signal sets, but signal sequences that have a well-defined order.

Another special feature is that each image can only be seen for a certain, often well-defined time. The duration of visibility of each individual frame is more or less the same if the film runs at a constant speed. In the first silent films, around 20 images were shown per second, so each of them was only visible for a twentieth of a second.

It is similar when learning language. Here, the temporal sequence of sounds must be learnt, which must later be filled with meaning. However, before recognising the meaning of a word, each word must first be learned as a temporal sequence of acoustic signals.

The question is how the vertebrate brain can memorise such concatenated signals (signal sequences). What algorithm is used for this, and which neuronal substructures are involved and how? This is the question we want to address in this chapter. It is assumed that we have extensive knowledge of the neuronal substructures of the vertebrate brain.

 

The various chapters on the pontocerebellum in my previous publications have already explained how the pontocerebellum learns a complex signal that is associated with a set of active elementary signals. It is important to note that the processes of long-term potentiation and long-term depression are at work here, which (among other things) require a tetanic excitation with a duration of one second. The signals must also act on the Purkinje cells for this length of time so that they can be linked (imprinted) there. A signal with a duration of one twentieth of a second or less cannot cause any learning processes in the pontocerebellum.

 

 

In order to understand the learning process for signal sequences, we need to analyse the three subsystems of the brain that are jointly necessary for learning signal sequences. According to the author, these are the cerebellum, the basal ganglia system and the hippocampus with its limbic rotation loops. We will refer to the ability to learn signalling sequences as temporal memory for short.

Since language consists of a chain of sounds, we will use language as an example to explain how language is learnt in the brain.

Firstly, the individual sounds of a language must be learnt. This has already been briefly explained in chapter 12.1 of this monograph.

 

16.1   Learning the sounds of a language

 

The basis for speech is the sense of hearing on the one hand and the motor sense on the other. We start with the sense of hearing.

When we hear, the air vibrations of the sound move the eardrum. These vibrations are transmitted to the cochlea via the ossicles and cause the hair cells there to vibrate. Reflection creates a travelling wave, as a result of which the inner hair cells of the basilar membrane are neurally excited when the fluid there is set into vibration. This is a different location for each sound frequency, so that the output of the hair cells already contains a frequency analysis of the sound.

 

In vertebrates, these auditory signals (among others) take the path to the auditory cortex and from there via the bridge nuclei into the parallel fibre system to the Purkinje cells of the pontocerebellum. The auditory cortex corresponds to an auditory area in the cerebellum, which receives these signals.

At the same time, there are large average neurones in layer VI of the auditory cortex that form cortical average signals. These descend via the nucleus ruber to the nucleus olivaris, where they are switched to the climbing fibres of the auditory area in the cerebellum.

If, for example, the vowel "a" from the word "mama" is heard, a small child who has not yet learned this sound will analyse the frequencies of the sound in the cochlea, send them to the cortex and transmit this frequency mixture typical for this sound to the parallel fibres of the auditory area of the cerebellum. As this sound is still unknown, the first free Purkinje group of the area will be shaped by this input and store this signal as an intrinsic signal if the necessary tetanic excitation for the LPT and LTD is simultaneously activated by the mean value signal on the climbing fibres.

However, both types of signal must be present for around one second, the minimum time for LTD and LTP in the cerebellum. This is why the mother has to stretch the sound a over time and speak particularly slowly. Mothers do this automatically anyway with their very young children who are just learning to speak.

This is how the first free Purkinje cell (or Purkinje group) learns the sound "a". As soon as it is heard later, the associated output neuron in the dentate nucleus fires and its recognition signal reaches an associated neuron in the cortical Wernicke centre via the axon.

The child learns all important speech sounds according to the same algorithm. Each sound is assigned a Purkinje cell or Purkinje group in the auditory area of the pontocerebellum, which has stored this sound and sends the recognition signal to the cortical Wernicke centre via the assigned output neuron of the dentate nucleus whenever it is spoken. There is exactly one cortex neuron for each sound, which is active and fires precisely when the associated sound is recognised in the cerebellum.

This is why there is a so-called sound area in Wernicke's centre, in which each learned sound excites a cortical pyramidal cell located there as soon as it is heard. This happens with incredible speed, as the hair cells in the cochlea react almost inertia-free and their signals only have to activate a few neurons involved on the myelinated axons.

The cerebellum initially recognises individual sounds in this way, but now the hippocampus comes into play and enables the recognition of sequences of sounds.

What is special, however, is that although the Purkinje cells need a period of around one second to learn a sound, they only need a few milliseconds to recognise it later. So if the sound is only heard for 10 milliseconds after learning, the pontocerebellum still recognises it and reports the recognition to the Wernicke's speech centre. This is why the mother can increase the speed of speech the more speech comprehension her child develops. Fast talkers can manage well over 100 letters per second without any loss of speech intelligibility. The cerebellum therefore learns relatively slowly, but recognises extremely quickly!

After learning the individual sounds of a language, we still need to learn syllables, words and sentences. However, the issue here is not how the meaning of a syllable, word or sentence is recognised, but how the brain learns syllables, words or sentences from the mixture of sounds that we call language and is able to recognise them acoustically. The meaning of what is spoken is understood much later and is not under discussion here.

 

 

16.2   Learning the spoken language

 

The starting point is the phonetic area, which receives the individual sounds of speech in the Wernicke centre of the pontocerebellum. For the sake of simplicity, we assume that this is the speech of an infant's mother, i.e. always the same mother. This is because speech differs from person to person in terms of pitch and intonation. It is only much later that a child learns to "tune out" these differences.

 

When learning the syllables and words of a language, the brain has to link signals that are not active at the same time, but at different times. According to the author, this is where the limbic system comes into play.

 

The output of the phonetic area in Wernicke's centre reaches the hippocampus. There is an input neuron there for every sound learnt so far, probably in the amygdala. This projects into the hippocampus, which converts this signal into a higher-frequency signal sequence and rotates it in a closed signal loop through feedback. This signal rotation in the limbic system was recognised as early as 1937 by American neurologist James Papez.

 

Each speech sound recognised by the cerebellum triggers a signal rotation in a loop assigned to it in the Papez circuit, which would never end without a necessary stop signal.

Each of these Papez circuits delivers the signal rotating in it to an output neuron at each turn. For the time being, we will refer to this neuron as the Papez neuron of the speech sound. It is active as soon as the sound is pronounced and recognised. And it remains active because the signal in the associated Papez circuit is constantly rotating.

All Papez neurons of all sounds project descending to the nucleus ruber, because acoustic signals, like all other modalities, were originally used for motor control. They travelled from the nucleus ruber via the nucleus olivaris to the opposite side. In this way, they were originally able to cause contralateral inhibition there because both halves of the body were in neuronal competition with each other.

However, when the cerebellum developed, whose input nucleus was the nucleus olivaris, the cerebellum received (among other things) the limbic signals of the Papez neurons via the climbing fibre projection. Each climbing fibre received the output of exactly one Papez neuron, i.e. the signal of an associated sound when it was heard.

We assume here that this climbing fibre in the cerebellum supplied a whole chain of Purkinje groups, partly because of its greater length, but also because it could split into several climbing fibre offshoots.

Thus, for each speech sound there was a chain of Purkinje groups (up to three Purkinje cells each with identical input), which were always followed by a Golgi cell, all activated by the same climbing fibre.

We assume that each active Purkinje group activated an associated excitatory output neuron in the dentate nucleus, as well as an inhibitory output neuron.

The output neuron with the excitatory transmitter may project to the cortex and end there in Wernicke's centre in an area that we will call the first-level language area.

 

Purkinje cells now require a parallel fibre input in addition to the climbing fibre input. We postulate here that both should be identical. Therefore, each Papez neuron, which indicates a recognised sound through its activity, may also project via the bridge nuclei and the mossy fibres into the parallel fibres of the relevant cerebellum area, in which the climbing fibres also end with the same signal information.

 

The circuit is now complete and can learn the first sound combination of two sounds.

We analyse how the brain learns the first sound combination from two different sounds. For example, the sound combination "au" in the word car.

The mother pronounces the word "ow", for example, to make the child realise that it has just bumped itself and is in pain.

 

The pronounced sound "a" is recognised in the pontocerebellum and sent into its limbic rotation loop. There it initially rotates permanently in the associated Papez circle. The output reaches both the Purkinje cell for the sound "a" and the mossy fibres for the sound "a" in descending order. These activate a whole group of granule cells, which ascend to the cerebellum cortex and each activate a whole group of parallel fibres. The Purkinje cell excited by the climbing fibre signal receives the parallel fibre signals of the sound "a" and is shaped by them. It learns the sound "a", which in this case already stands for the word a.

Now the sound "u" is heard because the mother pronounces the word "au". The sound "a" has already faded, now the sound "u" is heard.

The recognising Purkinje cell sends the recognition signal for "u" to the Papez neuron in the limbic system, whereupon the continuous rotation signal for the sound "u" is started.

The signal line from the limbic Papez circuit that belongs to the already learned sound u now activates the climbing fibre that belongs to the sound "u". The first Purkinje group on this climbing fibre receives this climbing fibre excitation.

At the same time, the rotation signals of the two sounds "a" and "u" from the limbic system reach the parallel fibres via the mossy fibres, so that it is precisely this Purkinje group that learns these two sounds through imprinting. If both signals are later rotated at the same time, with the signal for "a" rotating first and the signal for "u" following later, this Purkinje group will recognise this and report the recognition of the syllable "au" to the Wernicke language centre.

It is characterised by this input and in future always reacts precisely when the sound combination "au" is heard. The pontocerebellum has thus learnt the syllable "au" and will always recognise it later. The output neuron of the dentate nucleus of this syllable must now send its axon towards the Wernicke centre, where it docks onto a newly formed cortical pyramidal cell and activates it as soon as the syllable "au" is heard.

It immediately reports the recognition of this syllable to the cortex by contacting a Purkinje cell in a new area in Wernicke's centre, which in future will always be excited when the syllable "au" is heard.

Of course, this process is associated with the formation of new Purkinje cells, cortex cells and new neurons in the limbic Papez circuit. We already know that new neurons are formed in the cerebellum until around the second year of life, whose axons in turn have to contact newly formed target neurons in the cortex, hippocampus and amygdala so that the neuronal circuitry described above can develop step by step.

It can be assumed that the cortical projection neuron in Wernicke's centre, which represents the new syllable "au", seeks contact with the limbic system again and its axon contacts a new input neuron there, which in turn completes the Papez circuit. If the syllable "au" is now heard, this new rotation loop is activated in the Papez circuit and represents the syllable "au" just heard.

It can be assumed here that there is also a receptive neighbour inhibition in the limbic system, e.g. in the amygdala, which is also passed through with each signal rotation. In this case, it can be assumed that the rotation signal for the syllable "au" cancels out the two rotation signals "a" and "u", as these are related to it. Then only the signal for the syllable "au" rotates.

Its axon sends the output signal of this rotation loop back to the pontocerebellum in the cerebellar Wernicke centre. This again occurs in two ways: once as a climbing fibre signal and also via the mossy fibre as a parallel fibre signal.

If the sound sequence "auf" is heard, the rotation loop for the sound "f" also becomes active and activates the parallel fibres for the sound "f" via the mossy fibres. As the signal for the syllable "au" also rotates, it also activates the parallel fibres belonging to the syllable "au". The first Purkinje group on the climbing fibre for the syllable "au" is tetanically excited by the rotation signal of the syllable "au". As this Purkinje group simultaneously receives the parallel fibres of the syllable "au" and the sound "f", it is imprinted with these signals. If the new syllable "auf" is heard later, this Purkinje group recognises this and reports it to the corresponding cortical Wernicke centre, where it docks onto a new (free?) neuron.

This process can be continued indefinitely so that ultimately the word "recording" is learnt and recognised.

But the learning continues. Let's assume the mother wants to teach her toddler the new word "am". The syllable "au" and the word "auf" have already been learnt.

When the new word am is heard, the sound "a" is recognised as the first part of the word. In Wernicke's centre, the neuron for the sound "a" is activated, and the climbing fibre that belongs to this sound is activated via its axon in the pontocerebellum. The Purkinje cell that belongs to the sound "a" is already attached to this climbing fibre. At the same time, the rotation loop for the sound "a" is activated and reaches this Purkinje group via the mossy fibre. This immediately reports to the cortex neuron in Wernicke's area: "Sound a recognised!"

When the second sound component of the word "am" is heard, i.e. the "m", two signals are involved in the signal rotation at the same time, namely "a" and "m". At the same time, two climbing fibres are also active, the first for the sound "a" and the second for the sound "m". Both signal components are also active on the parallel fibres. The Purkinje group for the sound "a" signals the recognition of this sound. The Purkinje group on the climbing fibre for the sound "m" receives the strong climbing fibre excitation and the parallel fibre excitation of the sounds "a" and "m". This leads to imprinting with these signals. If the word "am" is later heard at some point, the associated neuron in the dentate nucleus fires and excites a newly contacted cortical neuron in Wernicke's centre, which represents the syllable or the word "am".

It should be noted that the previous limbic rotation signals are also cancelled here, in this case the signal for the sound "a" and the sound "m" is cancelled, while the (new) rotation loop for the syllable "am" is activated instead.

In the limbic system, the last syllable or the last word rotates as a complex signal. If speech continues, either a further sound is added to the rotating syllable or a new syllable is analysed. In this case, two syllables are combined to form a new word using the same algorithm. Here too, the new "sum signal" is used to delete both partial signals in the rotation loops. In this way, multi-syllabic words or even entire sentences can be stored in an associated Purkinje group, for example entire poems.

It should be noted that rotating signals in the limbic loop are most likely cancelled by longer word pauses. This prevents syllables that belong to different words from being directly linked to new "syllables" that do not even exist in the language in question. Such a cancellation signal is obtained by silence during the pause in speech. The auditory mean value signal is then the zero signal. If it is inverted, e.g. by the spinocerebellum, it can be used as a cancellation signal. It must then reach the amygdala and activate the GABAergic interneurones in the auditory loop area so that all speech signals are cancelled. This resets the auditory system to zero.

This is how the child learns to hear the language. The meaning of syllables and words requires further learning in which syllables and words are linked to the images of new objects. Cerebellar learning is also used here. These images can be olfactory, gustatory, visual, tactile or include other modalities; they are synaptically linked to the acoustic images and allow the meaning of the word to be inferred.

 

Finally, we still have an important question to answer. If language comprehension depends on the limbic system and the hippocampus, then we need to clarify why language comprehension is often retained when the hippocampus is disrupted or destroyed. The so-called long-term memory can also be preserved, only the ability to learn something new, i.e. new syllables, words or even poems, is lost?
This is where the basal ganglia system comes into play.
Figure 52 in Chapter 13 "The three subsystems of the human brain" shows that in each of the three subsystems of the brain, the basal ganglia and the limbic Papez circuits are connected in parallel. In a parallel connection, each subsystem receives the cortical output to it and the thalamus receives the output of these two subsystems connected in parallel.
This means that the basal ganglia also receive the speech input and can generate a short, time-delayed echo for each sound and send it to the cerebellum, both via the climbing fibres and the mossy fibres. These are the same climbing fibres and mossy fibres that the limbic system also uses, as both are connected in parallel.
However, the excitatory, short echoes from the basal ganglia are sufficient for the Purkinje cells to recognise the sounds, syllables and words and report the recognition to the cortex. Although the pontocerebellum learns slowly, it recognises very quickly and in fractions of a second.
Therefore, if the hippocampus or limbic system fails, speech comprehension is usually retained. However, a disruption to the basal ganglia system cannot be compensated for so easily.

 

But how does a child learn to speak? How does a child learn to control the muscle groups involved in the speech process in such a way that it speaks itself?


Monografie von Dr. rer. nat. Andreas Heinrich Malczan