Gold Fish 1

home *** CD-ROM | disk | FTP | other *** search

/ Gold Fish 1 / GoldFishApril1994_CD2.img / d4xx / d411 / mind / mind.lzh / AITheory < prev next >

Wrap

Text File | 1990-11-21 | 176KB | 2,633 lines

Artificial Intelligence Theory Journal, Part Three of Three (1978 - 1979) Standard Technical Report Number: MENTIFEX/AI-3 by Arthur T. Murray Mentifex Systems Post Office Box 31326 Seattle, WA 98103-1326 USA (Not copyrighted; please distribute freely.) 13 NOV 1978 This evening at Vaierre I am trying to do some more work on bringing the whole Nommultic system together out of the various components. I am trying to attach the language-work of last year to the recent overview of an experiential and motor system. I am beginning to suspect that I don't need a centrally located, pyramidal-type language decoder with the "ultimate-tags" as denominated on 1.OCT.1977. It looks like it may work just to let the known and recognizable words exist in the vastly long "pipeline" of auditory memory. Idea of the moment: Node-slices could be kept thin and compact by having just axons within the slice. Then cell bodies could be arranged in tiers above the "pipeline." Who knows, maybe "carrier" neurons could send their axons lengthwise through the pipe, while "nodular" neurons would introduce perpendicular axons. A single such perpendicular cell might be able to have many, many "yes-or-no" synapses within a slice. In fact, such a perpendicular cell, or groupings of them, might CONSTITUTE THE VERY ACTUAL ASSOCIATIVE TAG. But you sure would have to have flat synaptic branching from a cell body. I am beginning to realize lately how passive the experiential side and how active the motor side must be. I suppose the habituation and learning mechanisms must all be on the motor side. The whole perceptual and passive apparatus exists just to serve the motor side in its decision-making deliberations. I've had a possible insight tonight on how the motor mind may perhaps reflect on things: through variation by the aid of engram fatigue. When the motor mind first associates towards a passive memory, such as a visual image slice, the following scenario may take place. The accessed image has been fetched through its associative tag. The nodular image slice becomes "energized" within the visual "pipeline." Up and down the pipeline, similar images become stimulated. One such image wins out and is the first to send out a signal over its own associative tag. This signal keeps the associative process rolling. It may evoke further information from any perceptual sense or from language-memory. At any rate, the consciousness has accessed visual memory and has had a certain output. Now here's where "engram fatigue" or "neuron fatigue" enters in. The motor mind may get shunted back to the same original visual memory slice. In this second instance, however, the fatigue of the formerly first responding engram may allow a competing different engram to respond, with the process being repeated many times over. Thus the same neuronal input can yield a wide variety of successively different outputs. You know, it may be that verbal thought really does take place in the passive experiential auditory pipeline, and that each word or grammar-ending just always happens to have a habituated motor sequence attached to it. So that when we feel we are hearing our own verbal thought, we are really just remembering what it sounded like when we formerly spoke each sound. Likewise if we pronounce in our minds a seen combination of alphabetic letters, we are quickly just joining together the remembered motor sounds of phonemes. I guess we have to have pipeline word-decoding, or else there could never be any variations from a strictly spoken standard for a word. (Nolarbeit) AI Theory Journal 17 NOV 1978 Remembrance of Verbs to Describe Actions Our current impasse is at a point where we are trying to bring together all the accumulated subsystems of our intended automaton. Recently we have tended to simplify several elaborate designs of ours, so that for the final design we luckily end up with a choice of either the elaboration or the simplification, which might be too simple to work. We have tended to simplify the volition system from last March and the verbal decoding system from a year ago. The impasse rests most pointedly in such questions of a grammar system as how the automaton will observe actions and then recall verbs to name the actions. That verb-problem stands out because we see rather readily now how we can at least find nouns to go with perceived objects. Actually, the verb- problem grew out of a narrower problem from several days ago, when I was trying to figure out how the automaton would assign the concept of plurality to perceived objects, so as to be able to form noun-plurals. I was making a little progress on the plurality-question. For instance, I realized that to perceive plurality is not a one-step process, because just to perceive the unity of one entity is a conceptual step in itself. To recognize two creatures, for instance, a mind can recognize first the one and then the other, but not both at the same time. So I was recently tending towards the conclusion that use of the concept of plurality involves (the processing of) multiple slices of perception. However, that quasi-conclusion caused me trouble because a system of unitary associative tags from percept to word didn't seem sufficient to handle plurality. I even started hypothesizing that our minds project plurality onto things, with greater or lesser success. Then late the night of 13NOV1978, I got the idea that maybe there should be an additional memory alongside the others (sensory and motor), a memory which held perceptual content but not sensory content, a memory which would handle conceptual associations beyond the linear scope of the purely sensory memories: an abstract memory. So for several days I revelled in this possibility of a new insight, but meanwhile I came to focus on this problem of assigning verbs to perceived actions. I sensed an analogy here with the work of 16.OCT.1978 on the skin- surface. When we verbally name a perceived action, we automatically tend to select the most aptly differentiated verb available to us. In so doing, we automatically pass over many less apt verbs which would nevertheless have correctly described the action. For example, "He ruined it" is correctly within the meaning of "He destroyed it." Another idea which I have been getting is that the remembrance of verbs is perhaps the function of a rather elevated "abstract language-domain." (Nolarbeit) AI Theory Journal FRI 19 JAN 1979 More on the Verb Problem Perhaps a verb can be viewed as follows. A verb is a punctiform expression of relationships ramifying from the logical punctum of the verb itself. We can then think of a non-specified verb, which, in the history of its being known to us, has developed some very highly ramified main branches beneath the punctum. A verb will describe an event of greater or lesser complexity. In accordance with its complexity, a verb residing in a mind's semantic knowledge will have "main branches" as divisions leading to the (probably quite numerous) minuscule ramifications. When we perceive an event and recognize its nature as being properly described by a certain verb, what we do is find and connect logical categories which satisfy the main-branch logical-input requirements of the specific verb. When two verbs are somewhat similar in their semantic meanings, we discriminate between the two verbs by means of differences between the two groups of main branches. We must have thousands of categories into which we can classify things when we perceive them. Perhaps we even classify things into multiple categories as a prelude to the connection of main branches. The same sort of system could perhaps serve to assign prepositions according to perceived relationships. When we perceive a thing, our mind seeks to attach to it both a name and a set of one or more semantic categories. The greater the discrimination we achieve in our attaching of semantic categories, the greater the discrimination we can also achieve in the selection of verbs. Finding a semantic category for a thing is not the same as finding a noun with which to name the thing. Right now I am mentally deep into the present subject, the verb problem. Some strange possibilities are opening up. Since this Nolarbeit Theory Journal is ipso facto a journal, I think I will override my tendency to keep personally out of my discussion, and instead just ramble on with topical thoughts. The strange possibility is also a slightly disconcerting one. It points to that "abstract memory" which I was mentioning in the work of 17NOV1978. Suppose there had to be an extra memory like a long bundle of minuscule fibers. Each fiber would represent one of the semantic categories presently under discussion. The disconcerting idea is that our minds might be limited by such a system in which we can access verbs only through this intermediate subsystem of semantic category-fibers. In other words, we are always limited in our ability to describe the events which we perceive, inasmuch as we must break down each percept into a set of minuscule semantic categories coming from a larger set already resident in our mind, after which breakdown we can then re-assemble the information-flow via "main branches" or "semantic trunks" to fetch a specific and (hopefully) appropriate verb. For weeks or months now I have been stewing on this problem of how we recall appropriate verbs. I have been pondering this problem while keeping in mind my 12APR1978 diagram of a visual-memory channel. I was telling myself that I had enough theory to recall appropriate nouns for things, but I could not even begin to figure out how multiple image-slices, taken in succession, would lead a mind to recall a verb. It was as if I was trying to imagine extra hardware and extra processes into the system of the diagram. But I knew I wanted to do it all with just the associative tags coming from the image-slices as originally planned. Even now I may not have the solution, but I will describe how I began today's writing. In my search for verb-recall, I was picturing an entity lying on a surface. If I were asked to describe with a verb the action of the entity while lying there, I would say that it is lying there. Now, how do I arrive at that verb from just a still picture? Obviously, I am detecting a relationship between the entity and the surface. I would recognize the entity all by itself, but in this case its side is orientated to the surface in such a way as to help me recall the verb "lying." My pondering mind seized upon the idea of the side of the entity as being in a special, semantic category. From the notion of concentrating upon the side of the entity, as opposed to the total entity, I got the idea of the leg of a system in which multiple legs had to be "satisfied" so that a common summit would be reached where a verb stood. I also got the idea that the semantic legs (trunks, main branches) could have very many minuscule categories attached to them, but that it would take only one activated category per leg to satisfy the recall- requirements for a given verb. It was at around that point in my thinking that I began writing the body of today's work. It is always thus; I usually wait until I have the rudiments of a solution before I start writing down thoughts. But I have been so stymied by this problem of verbs that today (on Seattle's Pier 51) I have gone back and written down even my preliminary thought. Now I can go on. If the broad "trunk" requirements for selection of a typical verb can be satisfied on each semantic trunk by any one of many numerous semantic categories, then obviously a verb is typically a very generalized notion. Highly specific verbs would probably tend to ramify into relatively few categories, but, on the other hand, some categories must be so general that they encompass the trunks of almost all verbs. At any rate, we have posited today a practice of "intermediation" between percepts and verbs. Verbs are to be visualized as like an octopus or a furcated carrot. A percept can summon a verb only by generalizing into semantic categories and then un-generalizing along semantic trunks to reach a specific verb. Scratch-Leaf - Time division: verbs in infancy vs. in maturity. - Verbs become categorized? - How do we recognize that someone is sitting, or lying down? It's a relational thing. - A conceptualization as legerdemain. 12 MAR 1979 Ways of Approaching the Verb Problem - Look at how the first verbs are learned in infancy. - Infancy learning of first verbs. - Future learning of new verbs. - Look at how verbs are recalled to describe perceptions. - Look at how verbs are used for internal mental states. - Look at the tie-in of motor memory to verbs. - Avalanche the problem by writing down all possible ideas. - Study the transformation of nouns into verbs, as in "booking" a flight, or "chaining" things together. - Treat the problem as that of a chain of complexities. As the perception of action is processed in the visual channel, what complex transformations can the process go through without losing the full information necessary to reach an appropriate verb? - [26MAR1979] Consider how modal, auxiliary verbs work. As I further ponder the verb problem, all kinds of preliminary propositions come to mind. With so many of them, they can't all be very correct, but from enough of them true directions should eventually emerge. As visual images come down the visual memory channel in infancy, objects are perceived, and we can easily imagine how nouns are learned and recalled for such discrete objects. The noun-words are learned phonetically, and then linked up with the visual images. When things are in a class, they all share a relationship, namely their mutual belonging to that class. Our perception system for visual images attaches nouns to perceived objects. "Other than nouns it does not attach" - dare I say that? Because I don't think the out-tagging system can handle just jumbles of visual haphazardness. When perceiving a scene or image, we either relate it through one of its ingredients to a previously tagged item, or else we learn a new word or make a new association so that the novel image can itself become an archetype to serve in the recognition of re-occurrences of such an image. But probably all such novel tagging is done in the early phases of language-acquisition, so that subsequent tagging and word-learning probably amount to re-groupings and re-classifications of previous archetypes rather than to the novel formation of new archetypes. After infancy, we learn many new nouns and verbs, but not new archetypes of visual perception. There must be a classification process which goes on in an area which can be thought of as perpendicular to the visual memory channel. When we read a storybook, our mind conjures up its own image to go with each noun or verb. Numerous individual examples of each noun have been classified as expressions of each particular noun. Now when our mind encounters the nouns in a story, it uses the whole class behind the noun for understanding the story, but our "mind's eye" conjures up some specific visual example which happens to present itself most fittingly for our interpretation of the story. In fact, it may be that particular example that yields access to the logical associativity of the whole class behind the noun. So the bare noun reminds us of a specific example, but each specific example has full access to the whole class. Now I think that there are some psycholinguistic classes that are so abstract that they go beyond visual images. Every perceptual recognition yields access to at least that one class of which the recognition is being made. For instance, perception of one mouse yields access to the psycholinguistic class of "mouse," and then in turn to the class of "animal," and so on through myriad other classes. I suspect that oftentimes access is gained to some logicoconceptual classes that are so abstract that they lead not to nouns but rather to logical conditions and functions, such as the condition of plurality and the function of subject of a verb. It may be that these logicoconceptual classes can not be traced backwards to specific examples, as the word "mouse" could. Perhaps every slice of perception must lead at some level to (at least) one of these logicoconceptual classes, of which, understandably, there might be a relatively small number: certainly under a thousand and probably at least half a dozen. In fact, the possible binary permutations of the full number of logicoconceptual classes might give a hint as to the upward limit on the number of verbs which we could possibly access. However, that large number might in turn be obviated if the unitary logicoconceptual classes can be used more than once in the formulation of the recall-apparatus for a verb. It's possible that there must be a "re-affirmation" mechanism with the memory channels to keep these logicoconceptual classes valid and vigorous. Suppose a slice of visual perception entails recall through a previous slice in the distant past, a slice which grants access to a line running parallel to the visual memory channel, which line constitutes a logicoconceptual class. At the moment of new perception, the new slice first gains access to the logicoconceptual class by the roundabout recall route. There should perhaps be a "re-affirmation" mechanism of creating a new direct tag from the new slice to the parallel line representing or constituting the logicoconceptual class. If there is non-retraceability, if a logicoconceptual line can not go backwards to activate its afferents, then it is perfectly fine if all new perceptions find their (roundabout) way to a direct tie-in with one or more of the logicoconceptual classes. In fact, this theory being developed may be equivalent to saying that we forcibly attach logicoconceptual interpretation to each and every perception, and that we can not do otherwise. To perceive is to interpret. This present theory may be solving two problems at once: both the action-to-verb problem and the problem of accessing function-lines in the operation of grammar rules. 19 MAR 1979 A brain cannot even begin to recall a verb without first seizing upon some entity within its perception-slice as the subject or object of the verb-to-be-recalled. And the brain cannot seize upon a perceived entity unless it successfully makes a comparison with an old slice. It's as if to say that we never really perceive anything new, just recombinations of elemental old things. So a first step in fetching a verb is to recognize an entity by making a connection with a stored record. According to the theory that is presently being developed, such connection is how access is gained (and maintained) to "psycholinguistic classes." A constellation of accessed classes will approximate the recall-requirements for a stored verb. I say "approximate" because the process does not have to be, and perhaps even cannot be, exact and certain - it just selects the most likely, the most apt and fitting verb. The reader or re-reader of these notes may begin to suspect that the theory is calling for a great superabundance of class-structures. So when a verb-related entity is perceived, it could activate a plethora of class-structures. However, any action perceived and calling for a verb will probably involve several or many separately perceived entities. For argument, let's say that five perceived entities are necessary for the recall of a particular verb. Each of the five entities being perceived might individually activate dozens of class-structures, but only the congruence of five specifically required classes would fetch the verb. Remember, from perception these class structures operate only forwards and not backwards. Of course, it is not yet clear whether this process operates by "strict" "voting" or by summation-type "voting." 23 MAR 1979 A "re-affirmed" (See NTJ 12MAR1979.) perception slice can feed into even a large number of psycholinguistic classes. These classes themselves do not feed back into the perception channel. If the perception channel did not have an ulterior purpose, obviously these classes in the "abstract memory" would be useless. I suppose that a main function of the abstract memory is to achieve "intermediation" between raw percepts and such complexities of language as verbs. When the first verbs are learned by an infant, probably the connection between the raw perception and the learned verb is originally a very tenuous one, but that doesn't matter very much, because there are not many verbs to cause confusion in the child's vocabulary. If semantic inputs to a verb can be called "radices," then each early verb of a child might be learned with just one radix. For a more mature speaker, a more discriminating assembly of radices would be required. By the function of the abstract memory, the classes of the intermediation "vote" for which of all (however loosely) connected verbs will be fetched for recall. We can make a case now for the need of adept speakers to teach a neophyte. The neophyte's internal selection of the correct verb is not ratified internally, but rather by the approval of the teaching speakers. 27 MAR 1979 In selecting a verb for recall, it is obvious that a mental mechanism, while perceiving a stream of input, must initially seize upon one significant percept as the linguistic subject of the verb to be recalled. Such is probably the case even when we use impersonal expressions like "It is raining." This selection of tentative verb-subjects may be a function of an attention-mechanism. At any rate, it matters indeed to theorize that such selection occurs. To coin a phrase, the "nominator-mechanism" which seizes upon tentative verb-subjects can also serve to provide a logic-line for the grammar area which differentiates between subjects and objects of verbs. Once something has been perceived as a tentative verb-subject, the mind can go to work expressing that verb-subject in the proper grammatical form. It is quite likely that, for selection as a verb-subject, all a thing has to do is be noticed first in a series. Or there could be a level-of- associativity trigger-mechanism which fires when a percept is highly significant enough. When things are expressed initially as direct objects, it would probably be not in the course of raw visual perception, but probably rather in a more abstract situation where the subject or verb or both are already understood. Later, in the evening, I've had an insight into what it means to say that we "know" something. Knowledge is a composite of both the original acquisition and the subsequent consideration of information. The preceding sentence suggests how knowledge or information is stored in the mind. Suppose we first hear a piece of knowledge as a statement through our ears. That linguistic statement is laid down in our auditory memory channel, where it remains as a record both of experience and of knowledge. However, true possession of the information as knowledge comes from the subsequent processing we do of the information. In accordance with how much we believe the information and are affected by it, we develop traces of the information in the memory channel of our own internal reflections. The more we tend to believe a statement, the more broadly we will associatively associate it with the main corpus of our knowledge and belief. Therefore significant knowledge becomes widely anchored within a mind, because it reverberates so deeply in our memory channels. 28 MAR 1979 I have been reviewing the NTJ work of 9NOV1978 on motor memory and I have had an insight or two concerning volition. Instead of having prolonged associativity constitute inhibition of motor initiative, I would now like to reverse that notion and argue instead that positive (i.e., any) associativity above a threshold level actually causes motor initiation. Indeed, the theory is becoming quite clear right now. The passive, experiential side of the mind knows (from experience) its own motor capabilities. To contemplate any such capability is to "nudge" the threshold of its execution. My first of two insights is that the passive mind can't actually look ahead and feel or foresee each motor initiation. No, the mind is just blindly confident that the motor initiations are available. So the so-called "desire" to activate emerges on the passive side in the context of belief and knowledge as discussed yesterday. My second insight concerns the relative natures of verbal and non- verbal volition. Non-verbal volition works fine, as in reaction to sudden danger. Enough compelling association towards an action simply causes it. What's more, I would like to place the activation-thresholds at the point where the "Motor Memory Activation Channel" enters the motor habit tagging system. Thus there need not be an elongated threshold system in between the passive and active sides of the mental automaton. Verbal volition, however, can be much more refined, precise and delicate, because there are such intricate pathways of verbal cogitation. The ego, referring to itself in English as "I," wanders amid its verbal memory and feels confidently in command of its motor options. Now I am getting an insight on how generation of sentences may actually occur in the passive experiential side instead of within the motor system. If so, this theory would mean that the motor system habituates the basic phonemic sequences of the words and that the passive side manipulates all the grammatical changes worked upon words. Nolarbeit Theory Journal 28 MAR 1979 M u s c l e s o o o o o o o o o o .---. .------------. \ \ \ \ \ \ \|_|_|_ (< EYE >) '--. EAR ,---' \ \ \ \ \ \/ | \"---"/ \ / | \ \ \ \ \/ | """// \ ( \ \ \ \/ | // \ \ | \ \ \/ | // \ \ \ \/ | // \ \ | \/ | ___/(______ \ \ / Cerebellum | / \ \ \ | /_____________| | | \ \ // \\ | | \ \ | // \\ | | \ \ // \\ \ / _______) \__ | _______/(__ _______)\_____ | | | | | | | | | | | | | | | | Motor | | | | | | | | | | Visual | | Auditory | | | | | Memory | | | | | | | | | | Memory | | Memory | | | Concept | | Activation | | | | | | | | | | Channel | | Channel | | | Nodes | | Channel | | | | | | | | | 4 APR 1979 The system diagram of our automaton changed considerably between 10SEP1977 and 28MAR1979. In September of 1977 we were glad just to have our first system-wide diagram. It gave us a holistic basis against which to react, and we have reacted so thoroughly that the diagram (28MAR1979) is really in a state of high flux. The main difference is that the system has become simplified, streamlined, and highly orthogonal. The perception and motor channels are now seen as running in parallel. In the pristine diagram of 10SEP1977, heavy black lines represented unknown, "black-box" processing-channels. In the recent new diagram of 28MAR1979, no such lines have been drawn in, because broad interaction is envisioned at right angles all across the various perception and motor channels. In the diagram of 10SEP1977, there had been separate boxes set up to organize the elongated memory channels, as if perceptions would be assigned associative tags as they were filtered through such separate, modular tagging-systems. Now the perception channel itself is seen as paramount, with orthogonal tagging going on all along the length of the perception channel. A process of theorizing has perhaps become clear, that of "dimensionalizing" complex systems. The visual channel of 12APR1978, despite all its complexity, becomes just one linear dimension within our total system diagram of 28MAR1979. When we lay down all the perception and motor channels in parallel, our notion of "dimensionality" suggests that in the subsequently orthogonal direction we can include as many different sensory and motor channels as are feasible. For example, if it were possible to have a "sixth sense" that registered dangers and perils, we could just lay it down in a groove alongside the other channels of our automaton. Such a sense might be used only rarely, but it would not overcomplicate the host system, because its dimensionality fits right in with the host design. In futuristic automata, we could have some really exotic senses present. 6 APR 1979 The Acquisition and Function of Grammar Rules Now that on 28MAR1979 a new system-wide diagram has been developed, the problem of grammar has returned to the eminence it held in fall of 1977. At that time I did develop a complex system of grammar, but I ended up with the feeling that my grammar system was too much in isolation from the (as yet undeveloped) other portions of the automaton. The more sweeping the grammar rules I allowed for in my system, the more I had to conjure up extraneous inputs necessary for the function of my grammar system. I felt that I was ending up trying to tackle the problems just by transposing the problems outwards to the perceptual system. Since the perceptual system had not been designed, the new grammar system stood in isolation while great burdens had been heaped upon an almost non-existent perceptual system. However, the feeling of accomplishment in the grammar area did set the stage for the perceptual work of April 1978. Meanwhile, in March of 1978 work had also been done on the motor side of the automaton. It remained to simplify the volition system in November of 1978. From then on, enough major subsystems had been roughly designed that we could gain a new look at the total system. As I tried to integrate the perception subsystem with the total system, I focussed on the verb problem on 17NOV1978. Over the winter I felt stymied by the verb problem, but I am holding in abeyance the tentative solution through an abstract memory. Before designing in detail that abstract memory, I want to go back to the isolated grammar work of late 1977 and modify my design which used extraneous inputs from the perceptual system. There are two forms of habituation necessary in the language system: the habituated linking of phonemes to form words, and the habituation of grammar rules. In my present work I have been tending to physically separate those two domains of habituation: to put phoneme habituation into the VMHTS "cerebellum" and to let grammar-rule habituation develop right within the auditory memory channel itself. I am nudged towards using the auditory memory channel because of the problem of how to enable the mind to "hear itself think." In the system diagram of 10 SEP 1977, there was an "Internal Verbal Perception Line" as a sort of internal return loop so that the motor mind could hear its own output and simply make a choice as to whether or not that output would actually be spoken. Now, however, it may prove radically simpler to let the very sound-volition system be hearable unto itself and serve as its own self-perception system. The way to achieve that self-perception would simply be to have the rather wide-spread sentence-generation process deposit its pantothenic results in the succinct form of an utterance-capsule at the freshest extremity of the self- lengthening auditory memory channel. Indeed, it sometimes seems to me as if people like R.D. Palmer have a well-developed "pre-elocution register," because they seem to hold before their mind's eye their intended utterance with the additional ability of altering it quickly in the light of their own flashing reaction to it. The trick of such a pseudo-register would be as follows. The "pantothenic" procedure is probably sufficient to generate sentences for immediate, unpremeditated utterance followed by immediate deposition at memory-extremity. Such is probably the glib manner in which young children speak. However, mature speakers with the "pre-elocution" skill can perhaps let their intended utterances go into memory-extremity before exceeding volition-thresholds for actual speaking. Thus the sentence-formulations can be rapidly adjusted several quick times in the brief moment before utterance. Of course, the mechanism being described is by no means simple. There still has to be passage of control lines from the auditory memory channel to the "effatory" motor channel. I haven't decided whether those would be old "pantothenic" lines or newly re-affirmed memory- extremity lines, although the old pantothenic lines are necessary to generate the reaffirmations. In describing the system in this paragraph, I haven't yet described what mechanism would be generating those sentences right within the self-perceiving auditory memory channel. The beauty of today's proposal is that several things fit together quite nicely. The habituated grammar knowledge resides ubiquitously within the auditory memory channel and generates sentences from verbal material interspersed amid itself. Simultaneously the auditory memory channel perceives and understands the sentences being generated within itself. The auditory memory channel thus becomes an arena through which all the conscious knowledge of a mind can interact for such purposes as equilibrium, synthesis, and communication. It is beginning to look as if "deep structure" is going to be little or nothing more than the classifications which arise in the abstract memory channel. Just how flexible and "habituable" is this channel? Its inputs from perception are constantly variable through the mechanism of "re- affirmation." Its outputs to the auditory memory channel are variable with respect to their destinations within the auditory memory channel. In other words, there can be a two-tiered process, which nicely keeps the abstract memory isolated between perception and cogitation. The first tier is the perceptual re-affirmative inputs. The second tier, that of outputs to auditory destinations, will probably function by a normal associative tagging mechanism. That is to say, abstract memory lines will perpendicularly acquire their output destinations on the basis of associative tagging through simultaneity. No, I take that back. The assigning of their destinations must probably occur by conscious learning. You see, ordinary simultaneity would create too much of a jumble. The abstract memory is supposed to be aloof and isolated. Perception fetches no theta-word without simultaneously conveying through the abstract memory all the concomitant grammatical influences upon that word. Perception fetches a theta-word directly, but the abstract concomitance governs the form and syntax in which the word will emerge to be thought or spoken. Suppose a nonsense-word like "kred" is to be pluralized into "kreds" from perception. The basic theta-word is fetched by recognition. At the same time the abstract line for plurality is accessing the extremely frequently used suffix for plurality in the auditory memory channel. But at the same time this procedure must be being organized by lines governing syntax. Of course, syntax is part of the "concomitance." That is to say, syntax arises from the perception itself. Rules of syntax have been learned. This "learning" has been habituated through the process of establishing to which destinations the abstract-memory outputs will go. Now, I have been getting the idea that each whole line for syntax will actually be a node upon a whole "tree" of syntax nodes. In other words, each line will be truly elongated and not punctual, but also as a node each line will lead up from itself to a particular sentence-delta at the summit of the tree, from whence branches will go down leading to other nodes crying for imposition of satisfactory "fillers," be they words or concepts. Now, it is rather clear how a low node-line for a particular percept can come into play simply as a result of logicoconceptual classification. It's not so clear how a sentence-delta will come into play. I would like to make it hinge upon the total state of mind of the speaker. For example, a playful person could express all of his declarative observations in the form of questions. Of course, there would not be very many sentence-deltas to choose from, anyway. The sentence-delta could come from perceptual classification, but at a time possibly either equal or prior to the low node-line. 9 APR 1979 The Evolution of Mind. The new system diagram of 28MAR1979 leads me to speculate that now there is an obvious possibility for the origin of mind. Since the channels in the diagram are mainly in parallel and coming from the joint area of perception and motor function, I tend to see that stimulus- response area as a source both in the diagram and possibly also in evolution. Until recently I had never viewed the evolution of mind as possibly so simple a process. But now I can imagine certain steps. An early step would be the differentiation of cells necessary for stimulus and response. With such cells, many sorts of complex systems can arise even before memory is introduced. Now, I don't claim to know how instinct functions, unless it is a form of quasi-memory pre-established genetically. But the next step in the evolution of mind might be the appearance of memory capabilities. Two subdivisions in this memory-stage might be instinctual memory and learning- memory. My diagram of 28MAR1979 suggests to me that to add memory to a neuronal system is to open a real floodgate of possibilities. It makes sense to add memory only in enormous quantities suitable for the whole natural lifetime of the organism. Of course, with all these parallel memories there must also be the associative cross-linkage. The third step in the evolution of mind might necessarily be the development of an abstract, logicoconceptual memory, as opposed to merely sensory and motor memories. In this regard it is perhaps significant that, in my design of the abstract memory, I have had to derive it from what we might call the "apex" of the perceptual memory channels. Logicoconceptual classification does not go directly sideways, but occurs "apically" through the roundabout route of recognition. Contemplating the 28MAR1979 diagram, I receive the strong impression that the whole works is just an outgrowth in temporal extension of the original stimulus-response apex. Prototype Construction. Although consideration of actual hardware construction should not be allowed to influence the theoretical design of our automaton, construction insights may be included in this journal. For a long time I have hoped to be able to use variable loops of memory channels in a prototype so that, on the one hand, I could have a real-time machine without absurd simulations of time or its environment. Now it (again) looks as though it may be possible to use loops to recirculate the various parallel channels, and to add increments as they become available. Of course, the sideways associative networks would also have to be recirculated. 11 APR 1979 The Abstract Memory It should be possible to use the 4APR1979 method of "dimensionalizing" the abstract memory channel. The abstract channel stands out alongside so many "concrete" sensory and motor channels. The first delineation of it is that its inputs probably come only from sensory channels, and not from motor channels. The second delineation is that its outputs are only to the auditory memory channel as the vehicle of language. Therefore, in terms of inputs and outputs, the abstract memory channel is an organizer of all sensory perception channels for special presentation to the auditory perception channel. Significantly, the abstract memory channel can have both input from and output to the auditory memory channel. Although I have been designing the abstract memory in terms of the visual memory channel, a blind person can certainly learn the use of language, and therefore the abstract memory must be open to most or all of perception. The possible need for an abstract memory arose in the effort to solve the problem of access to verbs. I may now use the abstract memory to carry or mediate the whole grammar system of language. For grammar, the question is, must abstract memory lines be simple and unconnected, or can they feed into one another and cause structures to arise within the abstract memory itself? In other words, what is the "dimensionality" of the abstract memory? To start out with, it is at least two-dimensional because it has the length and width of its group of multiple lines. Each "planar" line represents a classification of elements from perception. These are not classifications made consciously by the mind, they are automatic. It seems to me that the most difficult classifications are the ones that fetch verbs. But it is not enough just to find a root verb in storage; for Indo-European languages, the proper modification has to be worked upon the form of the verb. Such modifications are almost always worked according to standard rules; otherwise, specific irregularities are recalled. But the rationale for the functioning of rules has to be present already when the fetch-order is made for a verb, so I refer to the extra request-structure as the "concomitance." I would now like to suggest that the total verb-fetching structure is itself a concomitance to the perception-driven generation of a noun phrase perceived as the subject of an incipient utterance. The mind unconsciously knows that noun phrases call for verb phrases to form sentences. So perhaps a "first-filler" mechanism latches on to a subject-function noun phrase and then tries to fill in the role of the verb phrase. Perception or selection of a subject noun-phrase initiates a coordinated process. It is not hard to imagine how a classification-line in the abstract memory can get hold of a subject noun-phrase. There can be a kind of "primacy-hook" so that an attempt is made to force any new discrete percept into the role of a subject noun-phrase. The fact that thereafter begins the search for the concomitance means that "primocapture" (or the "primacy- hook") has broken into a kind of absorption-structure. 19 APR 1979 Ideas on Grammar Habituation - The bridge from a perception to auditory memory involves three things: word, syntax, and inflection. - In a way, a salient feature of syntax is probably already present when a percept is seized upon as one meant to lead to an initial word. - It is important to distinguish between the two domains which give rise to the generation of sentences: external perception and internal reflection. It is possible that all sentence-generation occurs only under the control of internal reflection through a consciousness-continuity-mechanism. Such a notion is attractive because it allows a syntax-mechanism to be always dominant. The idea is, let nothing be perceived (or thought) unless the syntax-mechanism is attentive to it and ready to latch onto it. - We expect the more convoluted, more contorted sentences (such as these journal sentences) to arise from the domain of internal reflection. There syntax can become quite complex because of the way in which one thought (or reflection) leads to another. External perception we expect to give rise to short, simple, direct sentences made in observation of the external world. This discussion matters significantly, because we are faced with the question of deciding what causes the lead-off to a sentence, or what causes the initial syntactic assertion. You see, we may want there to be a natural tendency for subjects and nominative case to be the first to assert themselves in the domain of external perception. The oblique cases we expect to be much more likely to start sentences (such as this one) in the domain of internal reflection. Such expectations are reasonable, because the continuity of internal thought can cause each internal sentence in a chain to pass a syntactic departure onto the succeeding sentence. Oblique departures can allow internal sentences to begin with or hinge upon oblique constructions. On the other hand, observations made about the external domain can be expected to spring so directly from "prime movers" as subjects that oblique constructions would not be called for. - In the auditory memory channel, an "onset-tag" would serve to fetch a word, while an ultimate-tag would serve to recognize a word. Such close bifurcation to a word gives the idea of looping through a word before returning to pronounce a word. 27 APR 1979 More on Grammar The bridge by which a word crosses from perception to auditory memory involves three things beyond the word itself: part of speech, syntax, and inflection. The part of speech is a spontaneous concomitance of the word. That is, each perception is originally channeled as a particular part of speech. So a perception heads towards a word along two vectors: the parse- vector towards syntax and the recall-vector towards the basic, stored form of the word. The part of speech (plus perhaps also the "syntactic departure") leads to the syntax, and then the syntax governs the inflection. Nolarbeit Theory Journal 28 APR 1979 diagram number one _______ / \ /syntactic\ \ model / \_______/ __ round-about connection /\ /--------------------------------\ / | | /part-of-speech | | /vector | | / | | / __|__ ___V___ / _____ / \ re-affirmation line / \/ recall-vector / \ /percept\<----------------------/abstract \-------------------->/stored \ \ / \ memory / \ word / \_____/ \_______/ \_____/ 28 APR 1979 For sentence-trees, we want a structure which is like a filter in abstract memory. A percept has various vectors in its concomitance, such as part-of-speech vector, function-vector, recall-vector, and any of various modification-vectors for such things as plurality, negation, or conditionality. An S-structure (or sentence-structure) in abstract memory looks like a ladder which has slots instead of rungs, and furthermore there are multiple apparitions of the ladder for multiple levels below the "S." When the concomitance vectors of a percept "address" any slot on any ladder- level of any S-structure, then the very re-affirmation line quickly wends its way up through the S-structure to the apical S. Well, perhaps not; let's investigate further, because meanwhile I have drawn today's diagram number two. Nolarbeit Theory Journal 28 APR 1979 diagram number two \ \ \ \ \ \ \ \ \ \ \ ... \ \ \ \ = = \ \ \ \ \ \ \ \ = SM = \ \ \ \ \ ... \ ... \ \ =...= \ \ \ \ \ = = \ = = \ \ \ \ \ \ \= SM = \ = SM = \ \ ... \ \ \___ \ \=...= \ =...= \ \ = = \ \ / \ \ \___ \ ... \ \ = NP = \ \< SM > \ / \ \ = = \ \ =...= \ \\___/ \ < NP > \ = NP = \ \ * \___ \ \___/ \ =...= \ \___ S / \ \ \ \ / \ < Nuc > \ \ \ < V > \___/ \___ \___ \ \___/ / \ / \ \ < VP > < MV > \ \___/ \___/ \___ / \ < NP > \___/ I need a word to describe that roundabout passage of a percept back through old recognitions in a cross-over to elongations in the abstract memory channel. We could call it "anocatamnesis" for "up-and-down memory," that is, "supratraversial" memory which flows up, across, and then down again in an "anocatamnemic" way to achieve an "anocatothenic" tie-in to an abstract memory line. Forsooth, I need such words so as to think more readily. Anyway, in today's diagram number two we see the various slots in the various levels of an S-structure. The concomitance-vectors of a percept will reach any such slot supratraversially. It is readily obvious from this diagram that certain superior (abstract-memory) lines must feed into certain inferior lines. In fact, the whole rationale of addressing the anocatothenic S-line is to generate a sentence out of parts tagged by perception. The question is, what taps the S-line? Perhaps it does not matter whether the S-line is tapped by perception or internal reflection. Nevertheless, a subapical percept gets to its slot through the anocatamnemic filter. Notice that "NP" for "noun phrase" occurs twice in today's diagram number two, but that the two noun phrases differ greatly with respect to their function, which is "subject of verb" in the superior instance and "direct object of verb" in the inferior instance. Obviously, as part of the anocatamnemic filter, a function-vector has to help select which anocatothenic noun phrase will be activated. So the concomitance of perception filters through to the proper slots in the S-structure. (Let each elongated abstract-memory line be an "anocat.") We might have to think of the S-anocat as freezing or steadying mental activity in such a way as to permit all available and relevant percepts to percolate through the supra- filter to their proper slots in the S-structure. After all, these anocats are actually controlling or manipulating the recall-vectors of the words tagged to the percolating percepts. So when a word is being recalled, say, a noun or a verb, the source-percept (or a series of them) is like a stanchion to which the various filter-vectors are tied. In fact, it is possible to glimpse a scene and then close one's eyes and generate a sentence about the scene. The various percepts become the stanchions. I just got a possible insight on the inflection-mechanism. Perhaps it functions because the word-recall-mechanism is two-tiered. The first tier is indirect through anocatamnesis, and then the second tier is direct through the re-affirmation mechanism. The first tier could be subconscious, and the second tier, with the inflections added, could be conscious. Anyway, in the case of a momentarily "frozen" S-structure, a percept- vel-word such as a subject NP has activated its proper slot. The system now has the time of the "freeze," during which all possible percepts can drive vectors through the supra-filter to reach and fill up the appropriate slots in the S-structure. Meanwhile, the S-structure effortlessly functions perpendicularly (to the anocats) to drive slot-fillers away from deep- structure and out towards the ready-made sentence of surface structure. However, what does each slot have hold of? Each anocat-slot has been arrived at via the concomitance-vectors of a percept-stanchion. Somehow the anocat has to keep hold of the underlying word-recall-vector. So far, the anocat-slot "knows" what part of speech the word is, what its grammatical function is, and, per se, where syntax will put it in the sentence. Aha, what it does not know is what the inflection of the word will be. But the inflection is highly dependent upon the function-vector, which was used, for instance, to choose which of several possible NP-slots would be selected in the S-structure. Now, do we want the function-vector to continue up in the supra-filter, or to be subsumed in the branchings away from any relevant anocat-slot? Remember, we still haven't figured out how the anocat-slot has hold of the word-recall-vector. The time-order of perception is not necessarily going to be the same as that of the resulting utterance. The S-structure creates a specific time- order for the utterance. The question is, at what point do the words mesh with the anocat- clusters of information concerning the words? In general, manipulative control over a word must probably be considered to operate upon that word as it sits there in the auditory memory channel. Then the control of inflections of that word is a separate but related process. The S-structure that we see in today's diagram number two - its sole purpose is to generate syntax, temporal word-order. It is not concerned with the forms of the words, the inflections, or the intonations and accents. But since the S-structure is indeed declaring the syntax, it has hold of the words while other processes determine the inflections. Perhaps the supra-filter fetches the stem of a word plus a vectored approach to its possible inflections. If so, there is an interstitial return from the auditory memory channel to the abstract memory channel. Perhaps the stem of the word is fetched by an onset-tag, and then an ultimate-tag leads back to the abstract processing-mechanism which vectors towards the proper inflection. Note that when the ultimate-tag leads to the abstract channel, it (the tag) has to join up with a function-vector if it is going to lead to the proper inflection in the auditory channel. Perhaps the S-structure actually decrees which recall-vector will first fetch a stem from the auditory memory channel. When we say that the S-structure has control of the underlying recall- vectors, perhaps what we mean is that signals from the anocat-slots have to be AND-gated to the processing-vectors if their progress is to proceed. I suspect that initial word-order hinges upon perception, and that subsequent word-order to round out the sentence comes from the S-structure. Here's a possible scenario. An active component signal flashes at the primordial point of the surface-structure syntax. This signal cannot re- trace backwards, so it has to have been "escorted" by an AND-gated vector to the word-fetch (word-recall) line. So therefore let us say that the word recall-line in the supra-filter has to go through its own quasi-ladder- levels to reach the auditory memory channel. A specific percept will pass through each quasi-ladder-level only if it is AND-gated from the S- structure. Even, or finally, at the level of surface-structure a fetch-line (recall-line) has to get the go-ahead of an AND-gate from the S-structure. If the first item in an utterance gets its surface-structure go-ahead, then a stem is accessed in the auditory memory channel. However, accessing that stem could possibly send an ultimate-tag back into the abstract-processor in search of an inflection. Now, it is obvious that the ultimate-tag would lead to an abstract area specific to the stem just fetched. For instance, it might lead to a decision-juncture of five possible case-endings. The proper selection would then be done by AND-gating with the proper function- vector back along the recall-path. A word recall-line, then, is a "knotty" path with abstract AND-nodes along it. The AND-nodes correspond to quasi-ladder-levels. When input goes into a cluster of recall-lines that will coalesce into a sentence, each recall-line has to be helped along by successive go-ahead AND-signals from the S-structure. It may be that all but the surface-structure go-aheads are very quickly and haphazardly forthcoming, so that sentence-generation proceeds quickly and in a manner of parallel processing. However, the release of surface-structure items is a serial operation. (Stuttering may be due to a foul-up in this operation.) When a point on the surface-structure releases its go-ahead AND-signal, we might say that that point is de-potentiated. It summons a stem, and the stem activates an ultimate-tag. The ultimate-tag AND-gates with the function-vector to address an appropriate inflection-engram. Perhaps the inflection-engram also activates an ultimate-tag to send serial control back to the S- structure. In fact, control may pass straight to the S-anocat, whence it quickly percolates to the serially next undepotentiated surface-structure item. During this process, all accessed engrams of the auditory channel are being thought or spoken. 2 MAY 1979 In the work of 27APR1979 I was theorizing that the grammar-bridge involves three vectors: part of speech, syntax, and inflection, above and beyond the recall-vector. I would now like to pare that number of four down to just two: a function-vector and a recall-vector. Syntax doesn't have to be a vector, because the S-structure is there waiting for the perceptions. Part-of-speech, and inflection, are really just aspects of a word's function, so a function-vector should cover both those aspects. The grammar system really doesn't care what part of speech a word is; it just cares what function it serves. We scholars have artificially classified the parts of speech. As for inflections, the function-vector will go into the grammar system, and then it will encounter a mechanism which takes care of inflections. It is good to pare down the number of bridge-vectors, so that the S- structure of syntax will have to keep track of fewer underlying items in tandem. So therefore, when a word is going to be recalled from, say, visual perception to auditory memory, the process has to bridge the chasm filled with lines of abstract memory. The abstract memory is operationally divided into parts. The logicoconceptual "cable" allows rudimentary percepts to join together to fetch such complex items as verbs. This "L-C cable" (logicoconceptual cable) probably also works to fetch such work-words as all conjunctions and prepositions. It would be safe to theorize that nouns and adjectives also can be mediated by the LCC (L-C cable). So word-fetching recall-vectors are rooted in the LCC of the abstract memory. The problem is, we need a grammar system that will handle both classes and specifics. (Later, in the evening.) There could be a way of increasing the number of logicoconceptual lines in an abstract memory cable if bunched lines were allowed to subdivide or if free, unused lines were allowed to join with dedicated lines and then unjoin, leaving only dedicated lines. Then the dedications could diverge. To pursue the syntax question, let's assume that the L-C cable had as its righthand "wall" a plane of around twenty thousand L-C lines each representing a word-stem onset-tag in the auditory memory channel. Obviously, I am establishing a syntax-gulf between the L-C cable and the auditory memory channel. Instead of directly accessing the auditory engrams, I am positing lengthy lines over which the syntax-system can exercise some control. Strangely enough, my concept of AND-gating is re-emerging. I want a way for the syntax system to control each logicoconceptual word line whether as a class or as a specific. The problem is that each of the 20,000 L-C lines will probably have multiple, even multitudinous outputs over to historical engrams in the auditory channel. Therefore it becomes imperative for syntax to control the vertical L-C line itself rather than the numerous horizontal fetch-lines. So we must set up a system where, sure, the L-C line can be accessed and be ready to fire over into auditory memory, but where an enabling (and sequencing) input from syntax must be present if the L-C line is to fire. Felicitously, once we require an AND-gate-type input from syntax, we allow the process also to work backwards, in a way. Looking for a subject or object of a verb, by blanket AND-gating, syntax can prime the L-C cable to fire the L-C line of whatever word is solely or most decidedly active. Nolarbeit Theory Journal 3 MAY 1979 ___________________ M O---| / _________________________________________ ___| Cerebellum / | u O __| / | / _| /-------' s O / | /--------, / / \ / | Motor Memory Activation Channel c O / / \ /\ |_________________________________________ / / / \ /\ \ l O / / / \ / \ \ / / / / \ / \ \ _________________________________________ e O / / / / \ / \ \ | / / / / \/ \ \____| s O / / / \_____ / / / | O / / | Concept Nodes / / |_________________________________________ O / / O _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ /\ | | _________________________________________ | | | | |____ | | \_________________| | EAR_____________________ | | | / / | Auditory Memory Channel |__/ |_________________________________________ _____________ ___/ \ ___/ | ___/ ______ ______ | ___/ | | | | | ___/ /| MV |---| NP | | syntax cable / _____ / |______|\ |______| | ==========================| | |/ \ ______ | ==========================| | S | \| | | ==========================| |_____|\ ______ | V | | \___ \ | | |______| | \___ \| NP |\ ______ | \___ |______| \ | | | \___ \| NP | | \___ |______| | \_____________/ ========================================================\ ======================================================== \ ======================================================== | ======================================================== | ======================================================== / \ logicoconceptual cable \ / '---------------------------------------------------------' __ ___________ / \\ / \______________________________________________ / \\ | | EYE |\_____| \ //\_____ \__// | Visual Memory Channel | ______________________________________________ \___________/ 3 MAY 1979 The firing of an L-C line may not have to be strict AND-gating with strictly two inputs. Instead, it may be a kind of summation or integration process where the second input just pushes it to threshold, but where even strong primary inputs might suffice to cause firing apart from syntax. One way to achieve the sequencing desired of syntax or an S-structure would be to have a system in which control-lines fed into one another. That is to say, each line could be both a control line for the L-C cable and an input line to its successor. I am now beginning to visualize a hierarchy of the "syntax-cable" lying above and dominating the L-C cable. However, the L-C cable is different in that it does not have internal sequences. Fibers of the syntax cable would use "backwards-blanketing" to prime for firing whole classes such as nouns or verbs in the L-C cable. So we can visualize a surface-structure level of the syntax cable resting right above the L-C cable. This surface structure holds a series of, say, seven syntax lines. Each such syntax-line is in itself an elongated abstract memory fiber. It has multitudinous historical one-way connections to the L-C cable class which it governs by backwards-blanketing. It is not important now whether it goes directly to all the members of the class or to an intermediate collective fiber. The seven syntax lines are chained together in a successive way. Now, I don't want to rush the firing- sequence, so I think I should introduce a moderating mechanism to let each word get sent into pronunciation and actually get activated before the next word is crammed down the pipeline. It is one task to activate the words in the proper sequence, and it is another, equally important task to await the successful activation of each preceding word before initiating the activation of each succeeding word. I would like the ready-for-next-word signal to come out of the accessed auditory memory channel and into the syntax-cable, perhaps in the form of an ultimate-tag fed onto a special-purpose abstract bus coming from all ultimate-tags. For the moment, I am ignoring the interstitial problem of inflection. Anyway, I see the need now to hierarchize the insides of the syntax- cable. Let the bottom level, just above the L-C cable, be the properly sequenced surface-structure. Then let other levels pyramid upwards to go to the S-apex of the sentence-tree. Note that I have finally found a way to fit in the dimensionality of a transformational-grammar tree-structure. First the logicoconceptual cable rests like a flat wall of around twenty thousand fibers between, say, visual perception and the auditory memory channel. But when I want to pyramid the S-structure of the syntax-cable over the flatness of the logicoconceptual cable, I take the liberty of imagining that the flatness of the L-C cable is rotated through ninety degrees so as to interface with the bottom plane of the S-structure. But actually these cables can just as well be thought of as round and as internally jumbled, so I make up the following Nommultic guideline: The internal dimensionality of a quasi-neuronal cable does not matter with respect to the external dimensionalities which the cable can enter into. My reasons for hierarchizing the syntax-cable are not yet totally clear. It is convenient to get the tree-structure built in already now. I suspect that the syntax cable must be so hierarchized so that later I can plan in how transformations will occur. For the moment, I want to theorize that the ready-for-next-word signal will come in at the apical S-level and go down all branches at once so as to move the firing locus over by one line. For weeks now I have been imagining a syntax-tree and trying to fit it into my general scheme of things based on the diagram of 28MAR1979. So now I see it as a pyramidal epiphysis to a rotated plane. 4 MAY 1979 Preparing to Assault the Inflection Problem This morning on Seattle's Pier 51 I have been reading back over the NTJ from 4APR1979 onwards. I want now to record certain side-ideas without necessarily following a specific track of thought. Yesterday it became especially clear that my present theorizing calls for elements of verbal thought to surface in consciousness in the following way. We consciously perceive a lot of quick sideways loops made by the grammar system through or in and out of the auditory memory channel. The grammar system strings together word-stems and inflections to form sentences. Those phonetic elements, the word-stems and inflections, are actually stored and operated upon within the auditory memory channel. All the complex linguistic processing goes on within the grammar system, and yet the focus of linguistic processing slips in and out of auditory memory as the sentence-elements are strung together. Having thought up the idea of a "linguistic focus" in the preceding paragraph, I get the idea of an imaginary little ball or bead being moved around as the focus of linguistic control. For instance, when the control- focus moves into the auditory memory channel to flow through a word-stem, the rest of the process has to wait until the control-focus comes back out, perhaps in the form of an ultimate-tag designating the end of the engram. But of course, no little black bead is bouncing back and forth in and out of the auditory channel. Instead, the grammar system is undergoing various internal states, one of which involves accessing an auditory engram and then waiting for a return-signal before resumption of operation. Of course, it is not yet certain that return-signals exit the auditory channel, but it is presently convenient to think so. This recent work on the grammar system has been quite novel. Previously I was designing systems of information-flow where one flow of information did not govern another. For my perception and motor channels, I was simply deciding what the basic automatic routing would be. Now for the grammar system I have begun to design systems of control, where one mechanism (syntax) must govern the procedures of other mechanisms. I have had to adjust my thinking to handle combinatorial flows where widely disparate inputs yield hybrid outputs, as for instance in the case of syntax governing word-recall. At some times, I have had to think about a line representing a specific, and at other times about a line representing a whole class. I have had to think about ways to control both types of lines without violating their logical integrity. My long-standing "pull-string" theory of transformational grammar probably has to do with the selection of which transformation will operate. By "pull-string" I meant that the very route of access to the sentence- structure would determine or select the transformation. The basic criteria are the elements in a normal, untransformed sentence-structure. When the mind seizes upon any such element, the element can possibly act as a "pull- string" to yank even the weirdest syntactic transformation into operation. For instance, if the mind seizes upon an element which it is presently regarding as a direct object, such a seizure might yank the passive transformation into operation. Haste or urgency in expression might compel many kinds of deletion-transformations. 5 MAY 1979 (Function-Cable Plus Ultimate-Tag) Times Inflection-Cable Yields Appropriate Inflection It is shortly after midnight and I have decided to sit down and try to figure out at least a rudimentary instance of inflection. Let's say that in the first declension of a Latin-type language we are dealing with inflection as to whether a noun is subject or direct object of a verb. First of all, I suspect that my so-called "function-vector" is a function of the syntax system. That is to say, the S-structure, not perception, determines whether a noun is construed as subject or object. If a noun is going to be direct object instead of subject, then it will be led into by a different transformation, a different form of the S-structure. I'm not absurdly saying that perception plays no role at all; I'm saying that the information (of subject or object) is transmitted more by how we perceive than by what we perceive. Let me review how the concatenated surface-structure will operate. Each node on it "backwards-blankets" a whole class of potential fillers. Theoretically, an individual filler thus gets released over to the auditory memory channel, where it will activate an engram. The grammar system is meanwhile poised to receive an ultimate-tag out of the auditory channel, which tag will send a blanket "next-signal" down the S-apex. It therefore follows that all ultimate-tags feed into an abstract memory line which can pass the control-focus back to the syntax-system. Such a state of affairs is not difficult or unpleasant to imagine. In fact, it may be a way of causing sentences to arise in the first place. However, such ultimate tags must also carry the burden of seeking out inflection. Now, I don't yet know what my hoped-for solution is going to be, but I do know the available inputs for this black-box mechanism. One input is the nature of the syntax-node that has prompted the engram-fetch. Somehow that information has to be kept available so that it can influence what happens inflection-wise to the fetched engram. The other available input is inherent in the nature of the fetched engram. Somehow, over the history of the organism, it must have slowly or quickly come to be the case that, from engrams of the particular fetched stem, ultimate-tags would not only activate the return-control bar, but they would also prime for activation a declension-peculiar slate of possible case-endings to be selected depending upon the "function-vector" emanating from the syntax- node. In other words, all ultimate-tags from a certain declension have to go and "half-activate" or "poise-activate" first a collective abstract line and through it a set of abstract lines, each holding multitudinous specific cross-over lines into the auditory memory channel, where they activate an inflection-engram. Now the problem remains of how to get the syntax-node function-vector information over to the single line in the poise-activated cluster of abstract inflection-lines. Note that each abstract line has multitudinous concrete cross-over lines. So far that function-vector is just kind of hanging there. It doesn't go into the L-C cable, because it isn't needed there. Of course, we could set up an abstract "function-cable." The syntax- cable would be "trained" to set the status of the function-cable at each time when the syntax-cable were firing one of its function-valent nodes, or even any node. Yes, the surface-structure nodes could become "hard-wired" to trip the function-cable with each node-firing. Then the function-cable in turn sends out specific-case blanket-signals to all possible declension- clusters of case-lines. Well now, dare I cry "eureka"? If this tentative solution works, it has been achieved by some methodologically noteworthy inclusions in the system. It seems as though I just throw in one or several abstract lines whenever I want to create an isolation-buffer between two mechanisms. I don't think that I would want the syntax-nodes directly to access and trip the declensional case-clusters. Anyway, it makes more sense first to centralize and then centrally to distribute the control lines. It is also noteworthy that I have had two outputs issue from a single syntax-node. One output blanket-accesses part of the L-C cable so as to flush out a word-recall. The other output now sets the abstract function- cable, which in turn trigger-blankets all associated case-lines regardless of declension. There is perhaps a third output from the surface-structure syntax-node telling the next node that it will fire when co-triggered by the signal percolating in from the ultimate-tag return-control bar. Let's go over now how words will be fetched and their inflectional endings be put on. The syntax-node of the surface-structure sends out two signals. The first signal, a kind of recall-vector, blankets a portion of the logicoconceptual cable containing all the abstract vocabulary-lines for a certain part of speech. If one of those abstract lines has been pre- poised from perception, it now tries to activate all or the freshest of its concrete cross-over lines. At least one concrete cross-over line succeeds in activating an engram in the auditory memory channel. That engram is now blipped consciously throughout the auditory memory channel. Without pause, its ultimate-tag is activated and outputs to two destinations. No, wait a minute; it does not go to the return-control bar, it goes only to the cluster of abstract case-ending-lines peculiar to its declension. It can not fire a case-line yet, because it does not know which case is appropriate; so it merely pre-poises all its cases. Meanwhile, the second signal from the syntax-node has unerringly gone into the "function-cable" and activated an abstract collective bar which in turn branches out distributively to pre-poise all appropriate case-lines of all declensions. Of course, it is the grammatically same case in all the declensions. Once the two pre-poisings from syntax and ultimate-tag come together, they co- operate to select one specific abstract case-ending line, from which the concrete cross-over lines now reach into the auditory channel to activate the inflectional engram. At the end of the inflectional engram, now finally the ultimate-tag outputs solely to the return-control bar, to let the syntax-cable know that the next node can now generate the next word in the utterance. It has been convenient to make groupings of related lines in the abstract memory and call the groupings "cables." I can group together the various clusters of abstract inflectional lines and call them the "inflection-cable." The inflection-cable is totally under the selective pre-poise control of the function-cable, which is itself directed or set by the node-lines of the syntax-cable. By the way, does it seem true that perpendicular concrete-lines are necessary to interconnect abstract lines? Does it also seem true that the multitudinous concrete-lines with their re-affirmation lines are the vehicle of the mechanism of habituation-learning? I feel bold enough to start theorizing that even plurality-signals could work their way through the grammar system by much the same mechanisms as I have already employed today. Number-signals would come from the L-C cable and they would "start-poise" either the function-cable or the inflection-cable. Of course, with number you also get the problem of agreement between subject and verb. Maybe it would be good to have the number-vector go from the L-C cable to the function-cable, where it could perhaps be used to affect the number of both subjects and verbs. 8 MAY 1979 /^^^^^^^^^^^^\ /^^^^^^^^^^^^^^\ / Visual \ ______ / Auditory \ | Memory Channel | / \ ultimate-tag | Memory Channel | | | / syntax \<---------------|---------------\ | | | \ node / ______ | | | | | \______/------>/ \ | | | | round-about| connection | /function\ | | | | /-----------|-------------\ |flush \ cable / | | | | | | | |vector \______/ | ______ | | | | | _V__V_ | | / \ | | | _|___ | / \ recall-vector| | / stored \ | | | / \ re-af|firmation /logico- \-------------|----|-->\ stem / | | | /percept\<-----|-------->/conceptual\ (onset-tag)| | \______/ | | | \ / line | \ cable / | | | | | | \_____/ | \________/ ______V_ | ultimate|-tag | | | | / \<-|---------' | | | | /inflection\ | ________ | | | | \ cable /-|--->/ \ | | | | \________/ | / stored \| | | | | \inflection/ | | | | \________/ | 9 MAY 1979 Ideas on the Habituation of Grammar It may be premature for me to start attacking the problem of grammar habituation, because I still have important problems in deciding just what there is that should be habituated. For instance, I have the problem of habituating the various transformations of sentence-structures. But I may be able to list the various ingredients and then devise a kind of "universal habituator" for abstract memory channels. I will probably have to get down into quasi-neuronal function to achieve habituation. In my model, I don't want to use the growth of any portions of neurons, unless I am forced to. I would rather use the post- growth logical interconnection of neurons that have already grown into a sort of tabula-rasa network. The unusual thing here is going to be the tenuousness of the habituating influence: language-patterns of experienced, adept speakers as perceived by a neophyte. I may try first to plan the habituation of a simple, normal syntactic structure as found in the diagram of 3MAY1979. I should keep in mind three important functions of neurons: positive transmission, inhibition, and frequency-coding. Indeed, I was most eager to mention inhibition just now, but then frequency-coding occurred to me as also worthy of mention. Then the idea hit me, after I wrote the first sentence of this paragraph, that I could perhaps use frequency-coding as a way of letting emphasis on a pattern be a way of habituating the pattern. I should just keep this idea in mind. I must also consider the right degree of play-off between genetic hard- wiring and habituated learning. When an infant is first saying nouns to refer to percepts, I can imagine a primitive S-structure causing each instance of the enabling of a firing-link from a perception channel over to the auditory engram. In other words, the pyramidal S-structure acts like an element of volition, perhaps only by operating whenever attention is directed to things that will be verbalized. For present purposes, we can assume that the S-structure either always tries to operate or is involved subconsciously by some such mechanism as attention or volition. In establishing the logicoconceptual cable, it seems safe to theorize that first the perpendicular recall-lines develop, going from perception to the auditory channel. We can then think of a mass of L-C fibers pervading and inhibiting the recall-lines. The L-C fibers could come in for all nouns as a part-of-speech class. It would then be the normal dormant function of the L-C fibers to inhibit the noun-recall lines. Activation of the L-C fibers would disinhibit, but not activate, the noun-recall-lines. The recall-lines and the L-C fibers would be running at right angles to each other. Each recall-line would adopt the L-C fiber or fibers closest to and inhibiting the recall-line as the incipient logicoconceptual line for that recall-line and its related percepts, both historical and future. Now we run into the problem of the re-affirmation lines. A re- affirmation line from visual perception is a normal associative tag that goes from the visual channel over to the L-C cable but not necessarily directly over to the auditory channel. We could think of the L-C cable for nouns as being essentially flat. The recall-line associative lines are already present but dormant before the advancing front of visual perception and memory gets down the channel to those lines. A new percept through anocatamnesis stimulates a pristine logicoconceptual line, during the absence of inhibition. It is rather easy for us to say by way of design that the extremity-fresh associative-tag recall-line will now form a rather permanent connection with whichever pristine logicoconceptual line is now being supratraversially stimulated. Since the fresh associative tag flows over a theoretically flat array of logicoconceptual lines, it can easily bond with the proper line among all the lines. Furthermore, this idea of re-affirmation is not necessary for the one-time function of the grammar system, but rather it serves constantly to update the system and perhaps even to allow gradual changes. Now we have made clear a technique of "habituating" the logicoconceptual lines based on the idea that the recall-line associative tags are single-shot, use-only-once bonding-devices. It's a habituation system which is essentially pre-hard-wired. The logicoconceptual fibers for nouns can be controlled as a class. So far, we are saying that they work by resting-state inhibition. That is, the L-C fibers will not access auditory memory unless disinhibited. The same sort of associative-tag system that reaches from the visual channel to the L-C cable should also in turn reach from the L-C cable into auditory memory. Lines that go from percepts to fetch words are "onset-tags," because they fetch the word at its beginning. If a disinhibited L-C line fetches a word, it is through anocatamnesis. Now, the auditory channel is a serial memory. We might as well envision it as flat, so that extremity-fresh, unused onset-tags can be bonded from the flat L-C cable over to the flat auditory channel. We plan that this system shall re-affirm a word each time it is thought. The information of the phonemic series of the word shall duplicate itself at the freshest extremity of the auditory memory as a temporal series of nodes. Of course, our grand design from 5MAY1979 for inflections specifies that word-engrams shall have ultimate-tags. These ultimate-tags would be needed just for recognizing heard words, let alone facilitating grammar. The question is, during re-affirmation, can an old ultimate-tag be passed forward to become a new ultimate-tag? If it can be, it would not be done in isolation; that is, the new ultimate-tag would be created (bonded) only if it had a place to go to outside of the auditory channel. The trouble is, we are now getting into re-affirmation lines that have to be link-bonded at both ends, although I suppose even the visual ones were that way. What I suppose we could do is to have the ultimate-tag be re-affirmed only after its destination had been re-affirmed. In other words, wherever the historical ultimate-tag went, it would be to some kind of abstract- memory line, such as in an inflection cable or a return-control bar. With regard to the auditory channel, the ultimate-tag is an output line. It is a juncture from the last phoneme in an engram to an outgoing associative tag. I have thought of a reason why it may be better to have ultimate-tags be re-affirmed through their destinations rather than through their sources. If we did it through their sources in the auditory memory channel, then we might get interference from interstitial ultimate-tags residing somewhere higher in the channel at the end of syllables that chanced to be components of a word that we are dealing with. Still, I am uneasy with the prospect of slippage of the exact bondage- point where the ultimate-tag terminates an engram. There may have to be a kind of ultimate-tag generating-system in which an ultimate-tag issues forth from any phoneme followed by a pause, a non-phonemic interlude. Actually, every temporal node-level of the auditory memory channel could have a tabula-rasa potential ultimate-tag, but perhaps only re-affirmed ones would get into the grammar system. Anyway, if we ignore the prospect of slippage for a while, we can proceed while counting on re-affirmation by destination. When a fiber of the auditory memory channel is serving as a departure- point for a pristine ultimate-tag, that tagging might cause that fiber to remain activated long enough for the bonds of re-affirmation to form at both ends of the new extremity-fresh ultimate-tag. 10 MAY 1979 Ideas on a "Universal Class-Habituator" Actually, what I started trying to design yesterday is a "Universal Class-Habituator," or "UCH." The idea is to habituate classes of things into a syntactic structure, and then specific elements activated within the classes will follow the habituated pattern so as to create a grammatical sentence. I think I might be able to devise a science of universal habituators as a help in designing one. (If I may speak facetiously, it would be nice even to have a "meta-habituator.") The science can develop if I specify a lot of things about habituators and their design. The term "classes," in the sort of automaton which I am designing, refers to elongated abstract-memory lines, each of which can singly represent a whole class of specifics. So, even though these abstract lines might be hierarchized in a two-dimensional array, their elongation means that they always have that extra dimension which is that of time. One idea which I have gotten in this field of habituator-science is that of a certain reciprocity between a class and its specific elements. I want to use this reciprocity to make a rather novel suggestion about how a mind might habituate a pattern of classes: A mind sets up a proper example of a target pattern using specific elements from the classes, and then proceeds to habituate the pattern through a process of re-affirmative repetition. This suggestion is based upon the reciprocity-notion that, in a way, you can get at each class by getting at any one of its elements. In my work of the earlier days of this month, I developed the idea that all active vocabulary items peculiar to a language are bundled into separate classes according to their possible grammatical functions, so that they can be controlled by syntactic nodes. If the items are bundled, it may be possible temporarily to use a specific element to get hold of the whole bundle in habituation. Another habituator-science idea is that it may be necessary to solidify or consolidate each habituated instance of abstract interconnection before proceeding to a new habituation. Or, more generously, it may be necessary just to solidify each level of habituation before proceeding to a higher level. It may be a general rule in habituator-science that inter-abstract habituation must both occur and be solidified by means of the so-called multitudinous perpendicular concrete lines, otherwise known as "associative tags" or "re-affirmation lines." But we then have a problem figuring out how the first concrete line gets strung from one abstract line to another: it evokes the image of a spider trying to spin a web. A couple of other, distant possibilities arise for first habituations: the splitting of a single abstract line to yield two divergent but related ones; and the joining of abstract lines first by (educated) chance and then subsequently by re-affirmation. I would now like to return to yesterday's meandering ideas while keeping in mind the above ideas on "habituator science." It may be necessary to theorize that the auditory channel has the ability to "fuse together" a reasonably short series of sounds. A fused series could then be fetched by an onset tag, be activated in full, and then come to a natural halt, with a built-in tendency to bond-generate an ultimate-tag. With such fusion, we might avoid interference from interstitial ultimate-tags. The fusion might be inaugurated whenever sufficiently intense sounds are recorded in a rapid and unbroken series. Or the fusion might be controlled by an attention-mechanism. Such fusion might allow nominative cases to be treated differently and more directly than oblique cases. The fusion could work by concatenating node-slices of the channel. Recall could cause a new fusion at the extremity of the channel. So far, we have been trying yesterday to design how an infant would habituate first nouns and then verbs. We used re-affirmative associative tags to let noun-recall-lines accrete onto logicoconceptual fibers in a flat array of abstract memory. We can now envision a single, rudimentary abstract line serving as the primordial line of the first nascent sentence-structure. Without worrying about how thick or thin that line is, let us envision it as the first linguistic control-line employed by the infant. We know how infants like to point at a thing and blurt out the word that they have learned as the name of the thing. Such use of isolated nouns is what we are presently trying to describe and explain. We might as well draw up plans for a two-tiered volition system for verbalization. One volition line will access a sentence-structure to generate verbal thought, and the other will both generate thought and cause it to be spoken aloud. In other words, what I am trying to do is refrain from designing the S-structure in a vacuum. If it is part of our "habituator-science" that single abstract lines must be accessed by multitudinous concrete lines, then we had better make provisions for that process at the pyramidal top of any large or small S-structure. It is beginning to look as though the physical flow of "candidate" abstract lines will be provided genetically. Thus the infant's first quasi- syntactic control-line will already be available in the general vicinity where it is needed. What I have not yet decided is whether or not there will be epitaxial layering of the flat arrays in the abstract memory channel. This decision may matter, if it matters whether the infant's first control-line is at the highest or lowest level. Anyway, we want to use a kind of "random dynamics" (q.v.) method to start the infant blurting out its first nouns. I suppose that we can first acknowledge and then ignore that random-dynamics mechanism by shoving it off into the motor areas. But it will cause the first laying-down of the concrete-fibers going from the speech-control and thought-control (mirabile dictu) volition-lines over to the S-structure. I suppose it is perfectly harmless to imagine the infant's first noun- control-line as being at a middle level, rather than at a highest or lowest level. Then we have the option of building either upwards or downwards. Notice that we have required the S-structure to be subject input-wise to a flat associative-tag array, just as we did in the case of the visual and auditory memory channels. Two days ago I was reading in the Engineering Library of the University of Washington that the cortex of the brain, if laid out flat, would measure approximately fifty centimeters by fifty centimeters by two millimeters. If the human brain can be thought of as flat, then I am encouraged when I find myself designing such a flat automaton. The class of logicoconceptual lines for nouns develops control of all the recall-lines for nouns. Although it might have been easy to let the first noun-recall-lines operate individually in motor-control of nouns, we want to shift the control upwards to the rudimentary S-structure. So we are speculating that the first control from above operates downwards to inhibit the function of the recall-lines below. As I recall, such operation is considered to be typical in the brain: lower functions tend positively to operate, unless inhibited from above. So we might have to envision a kind of "spread-out cell" ("SOC") that operates vertically so as to let a superior abstract-line exercise widespread collective control over all the logicoconceptual noun-lines in the flat array below. Actually, the SOC-cells above could also come from a (superior) flat array, as long as they always acted as a group. If we design a built-in tendency for downwards inhibition to occur, then we have a natural, unawkward control-link which the rudimentary S- structure can easily gain access to. So when the infant's rudimentary S- structure fires downwards through the SOC-lines, it disinhibits the flat array of logicoconceptual noun-lines. We have next to figure out how verbs would come into the infant's system. We want the infant to learn to say such things as "Bird fly" or "Train go." Verbs will have one ingress into the auditory memory channel, in that the infant will certainly hear a lot of verbs, even if he or she doesn't understand them. Verbs will start to be learned when some sort of initial link-up is made between perception and the stored sound of a verb. Over this past winter I did a lot of theorizing on the verb-problem. Here is where I want the logicoconceptual cable to really do its work. Perhaps I can create a new flat array in the L-C cable, as a place for verbs. Let's say that the verb-array lies below the noun-array. However, the verb-array has no concrete lines coming in directly from perception. Instead, the verb-array gets its inputs through multiple contacts with the noun-array above it. Now, the verb-array is a flat array of abstract lines. Each abstract verb-line develops associative-tag recall-lines over to verb- words stored in the auditory channel. Instead of a concrete-line reaching "leftwards" into perception, each verb-line will have a kind of concrete "trailer" line that goes underneath the (superior) noun-array for its whole breadth. Of course, each verb-line will have multitudinous such "trailers," as re-affirmative lines along the time-dimension. Now, each inferior trailer-line can make contact with any number of superior noun-lines. Thus the noun-lines can "vote" to select a verb-line. The most appropriate verb can be selected, perhaps through frequency-coded propagation, by whichever verb-line got the most positive-votes and the fewest inhibitions. You see, we could design it so that a trailer-line, getting some but not all of its requisite nodes filled, would be inhibited somewhat by each unfilled node. Thus a simple verb of few, but fully filled, inputs could override a complex verb of many, but some unfilled, inputs. Of course, a complex verb of fully filled, many inputs would override a simple verb of fully filled, but few, inputs. Now, the need for these trailers to have access to the full superior noun-array poses some question as to where the actual logicoconceptual verb- lines should be placed so as not to get in the way of the trailers. Notice that this system seems to allow the neophyte over time to learn refinements in the selection of verbs, via the re-affirmative function. Wherever the logicoconceptual verb-lines are located, we next have the problem of how to control them as a class. 11 MAY 1979 More Ideas on Grammar Habituation in Infants The idea that higher-level functions in the human brain typically govern lower-level functions by inhibiting them may prove useful if we think of this phenomenon as an evolutionary building-block in the genesis of complex systems. Specifically, we have the idea that percepts will tend to recall words. Then we are going to inhibit the recall from on high. Once we establish inhibition as the normal state of affairs, we acquire positive control through negation of the inhibition. Yesterday I was getting the idea that processes which I have been thinking of as from top to bottom may actually be upside-down in the cortex of the human brain. That is, I have been designing syntactic control structures as governing flat arrays of data-flow located on lower levels. All these putative structures may actually be upside-down in the human cortex, so that the relatively more significant control structures may occupy an inner, more centralized position as opposed to the relatively less significant associative arrays. This is a tentative prediction of the Nommultic theory that is possibly susceptible to empirical verification. Anyway, for the nonce I will continue in my present mode so that I can think of structures as being both superior in control and superior in location. Reading over yesterday's work, I get the idea that the logicoconceptual cable may be stratified into as many layers as there are parts of speech. It might not be good for me to lend credence too soon to such a formulation, but I can always re-arrange it. Even if there are part-of-speech strata, the visual and auditory channels would probably not have multiple strata. If there are part-of-speech strata, I suppose that superior disinhibition- lines would send down disinhibition "runners" from above. It is not yet clear whether the disinhibition "descender-runners" would physically access only a distinct stratum while actually permeating all the strata. Growth would have happened locationally, while learning would happen dynamically. It may be advisable to establish a notion that various neuronal networks occupying a given volume can be highly transparent and permeable to one another. Although I have been relying upon orthogonality in my design, I have tended to design networks as physically separate, lest they interfere with one another. Thus I planned certain arrays as being flat, and then perhaps stratified, while I kept in mind how things would fit together. But now I have to keep in mind that such caution might not be necessary if the arrays and networks can physically intermingle while preserving their logical integrities. Anyway, I am on the verge of adding in the control of verbs in infants. Yesterday I designed how the logicoconceptual verb-array could be a layer right below the noun-array (or even a more general array). If we assume that the noun-array was set up first temporally, then we can have it over and done with when it is time to establish superior control over the verb- array. The noun-array came to be controlled as follows. Horizontal volition runners activate the rudimentary, single S-line. This S-line rests amid the SOC-cells that descend to counterinhibit the logicoconceptual noun-lines. Let's say the infant gets pretty adept at blurting out nouns and is starting to learn verbs. The activation of counterinhibiting SOC-cells down to the verb-array will be causing verbs to be spoken. We can imagine that, from the original S-line for nouns, to which volitional fibers had been reaching, now additional associative fibers will be advancing further in the same direction as the volitional fibers had been going, and these newly activated associative fibers will be bond-attaching to a collective cable that abstractly flows through and controls the SOC-cells for verbs. So now we have designed that horizontal volition will activate first the noun S-line and then the verb S-line. The verb S-line develops as something which automatically follows the noun S-line in time. But now I want to do something which differentiates the volition signals for the noun S-line and the verb S-line. You see, the mind must try to say both types of words, nouns and verbs, immediately at once. So the volition system will try to slow down and differentiate. My insight here is that the flat, horizontal, volitional sentence-inputs will wander up one level higher and that in response a higher S-line will be established. The original, lower volitional fibers will cease to be true volition and will now function as return-lines carrying the information that the subject-noun has been expressed and that it is now the turn of the predicate-verb to be expressed. So therefore those return-fibers will go to the verb S-line, which has remained low while the controlling noun S-line has shifted up one level along with the volitional fibers. These arrangements may not be exactly correct, but the general idea of the insight is clear: Let volition wander upwards and let the former volition-fibers become informational return-lines. Today's earlier work was done on Pier 51. The last several paragraphs actually took some mental effort and I was pacing around a bit trying to hold on to the train of thought and seeking the right sentences to express my thought. But I finally had enough written down so that I felt I could safely leave the continuance for this evening or tomorrow morning. However, after I left off writing, the "insight" deepened considerably and now, in the evening, I feel that I may have a major structural development. The deeper insight involves closing today's loop of control-lines on top, counterinhibition-lines going down the side, recall-lines back across the bottom, and then ultimate-tags back up to the top again, so that a very dynamic square or rectangle is created. I have not thought out this hollow- square scheme in great detail, but, once the idea occurred to me at the conclusion of this afternoon's writing, I began to visualize several possible major features of the scheme. The idea is to have the whole generative grammar system operate as a kind of hollow square. (My next sentence may sound facetious, but there is sincerity in it.) If we visualize looking at the square from the future back into the past, the flow is counterclockwise. The logicoconceptual cable is at the bottom left, and the auditory memory channel is at the bottom right. Habituation of syntactic sentence-structures is to take place across the top of the square. The control-nodes will be at the top left, and volition will be coming in at the top right. Nolarbeit Theory Journal 14 MAY 1979 _____ __________ / \ | | / \<--------------------------------| Volition | \ S / |__________| Sentence \_____/ Structure | | _____ _____ | / \ B | | return-control +------------->/ Verb \<-------------| R-C | line | \ / |_____| | \_____/ /|\ A | | | _V____ | | / \ | __|__ / Noun \<-----------+ | | \ Phrase /<-----------|------------------| R-C | \______/ C | |_____| | | /|\ /|\ /|\ /|\ /|\ | | | | | | | Noun O O Verb | | | | | Spread- /|\ / \ Spread- | | | | | Out Cells / | \ O O Out Cells | | | | | / | \ | | | | | | | O O O O O | | | | | / / \ \ | | | | | | | O O O O O O | | | | | / / \ \ | | | | | | | O O O O O O | ultimate-tags | / / \ / \ \ | | | | | | | O O O O O O O O | | | | | /W\ /W\ /W\ /W\ /W\ /W\ | | | | | | | ||||||||||||||||||||||||| /W\ /W\ | | | | | logicoconceptual noun-lines ||||||||| recall-lines ______________________ ooooooooooooooooooooooooooo========================/ \ ===========================ooooooooooo=============\______________________/ verb-lines auditory memory channel 17 MAY 1979 The Comprehension of Natural Language When a sentence comes in to be understood, it is, so to speak, "captured" by the auditory memory channel. The mind will attempt to understand the sentence immediately, but, if an error is made, the sentence still remains in memory for additional attempts at understanding it. Usually, the mind tries to understand a sentence as it comes in, without waiting for the end of the sentence before processing the initial elements. I personally have had many sporadic instances in which I have raced ahead too quickly in taking meaning out of sentences, with the result that I have had to readjust my comprehension as the rest of the sentence came in. I have even made erroneous comprehensions by quickly processing just the first few syllables of a word, such as a compound noun. I assert that I was correctly processing what I heard up to each point, but I was often surprised when a sentence kept coming in in a way that invalidated my initial comprehension of the incomplete sentence. In several surprising instances I have comprehended the (incomplete) utterance and started thinking out a reaction to it, seemingly in the natural interval as a speaker was poised between syllables of a word. Actually, I was probably allowed the interval measurable by the amount of time it took for the corrective syllables to register. My point is, that processing advances as each morpheme comes in and is recognized. Some morphemes therefore convey a temporary ambiguity for the brief time that those morphemes have been spoken but not the immediately following morphemes. Of course, the falsely comprehended, incomplete utterance can be completely free of ambiguity. For now, I will call this ambiguity "subset-ambiguity," because, during a very brief interval, a subset of the morphemic string is erroneously but validly processed as if it were a complete and independent utterance. The following scenario shows how subset-ambiguity might operate. Suppose a traveler returns to his home town after a period of being incommunicado. Hungry for news, he is greeted upon arrival by an acquaintance whose countenance seems to grow sad and troubled at seeing the traveler. The acquaintance says, "Oh, my poor friend, I was so very sorry to hear about the sudden death of your brother Tom's racehorse." This example is rather extreme, but it illustrates two things. Firstly, various initial subsets of the sentence can obviously be comprehended in such a way as to describe a catastrophe, namely the death of the brother. Secondly, the example shows how expectancy or uncertainty may play a role in how an utterance is comprehended. It will become necessary to reflect upon whether expectancy plays only a psychological and epistemological role, or also a role in syntax and grammar. Anyway, I am at the point now where I want to start theorizing about how syntax and grammar operate in the comprehension of language. I have a general guideline in mind. Whatever language-comprehension is, it involves the establishing of the proper associative connections among the elements of a perceived utterance. Merely to deposit a sentence in memory is not to know the information contained within the sentence: witness a talking parrot. Is it possible that expectancy or urgency may use frequency-coding to strengthen the associative "valence" of an important sentence? Frequency- coding might be used to enhance permanently the associative bonding of an associative tag. An incoming sentence can drastically change and update our knowledge about the world or the universe. Let's discuss change in our knowledge of familiar elements from the external world. Our associative knowledge is clustered about the words naming the elements. Thus the same word for an element may exist at many different points within the stream of the auditory memory channel. However, those different points are accessible through internal associativity up and down the auditory channel itself. Of course, that internal associativity establishes only identity (or perhaps also similarity), not the information-rich relationships among separate words. Other than by temporal contiguity, separate words can not be associatively related within the auditory channel alone. Within my Nommultic theory as developed so far, separate words can be related only by the associative tags, of which there are two types in the auditory memory channel: onset- tags and ultimate-tags. I am now reserving judgment about all the various features of the onset-tags and ultimate-tags. I am getting the idea that it is possible that there are two ways for extraparietal association among separate words of the auditory channel. The first way would be through the avenue of one of the other senses, such as vision. For example, a visual scene could serve to associate many separate words describing things within the scene. The second way would be through associations brought about in language-comprehension. The question arises, can the comprehension-mechanism manipulate associations without getting entangled in them? I suspect that it can, by being abstract. When a sentence, either fresh or stale, is run through the comprehension-mechanism, associations are generated which remain as a momentary part of the historical record of that moment in the history of the organism. The sentence itself, whenever re-activated, goes through the comprehension- mechanism and re-establishes the "momentary but permanent" associations. (Wouldn't it be awful if we had to re-activate each original sentence of knowledge in order to make use of the information in the sentence?) The more important a sentence is, the more frequently we are likely to run it through the comprehension-mechanism so as to associate it quite broadly with our general knowledge. The next question is, can the comprehension-mechanism lay down associations among separate words in a word-to-word fashion so that it is not immediately necessary to go into the other sensory channels or even into the auditory channel in a nonverbal way? (For all I know, perhaps this wonderful performance is what it can only do.) Although it may thus string words together into relationships, the meaning (and subsequent logical generativity) of the string depends upon the associativity of each constituent word out among the total sensorium. If words are strung together but there is a paucity, a dearth, of sensory information about the words, then it will be difficult for richly specific mentation to ensue. A clue is developing now as to what provides the "motivation" or impetus for sentence-generation (thought-generation). An angular, crankshaft-like process operates. When one sentence, whether from outside or inside, goes through the comprehension-mechanism, associative relationships are established among words regardless of the prior associative import of the words. However, that prior import can immediately come into play as a kind of logical tension which prompts the generation of new sentences. Thus the mechanism of syntax and grammar mediates thought all up and down the system, and the verbal product of thought is deposited at the freshest extremity of memory. 18 MAY 1979 As the first word of an utterance comes into the auditory memory channel, the mind commences processing that word through its grammar-filter. Obviously, just as I did with sentence-generation, I will want the incoming sentence to submit to the control of a syntactic structure. I will disregard the problem of psychological expectancy by imagining that the incoming sentence is being perceived in relative isolation, so that the hearer does not know what to expect. If each word in the vocabulary of the language could serve as only one part of speech, then, clearly, we could use that part of speech as a criterion or selector for entry into any of multiple syntactic structures. We could then concentrate on the inter-related concerns of function and inflection. Perhaps I should establish a principle of widest-possible spread-out in the through-filtering of a perceived word into the comprehension-mechanism. I have in mind a pair of English sentences such as the following: "This man likes music." "This man I like." In accordance with the (presently emerging) "saturation principle," an initial English noun-phrase, such as "this man" above, should try to run through all fitting and available syntactic structures, in a process of being impeded on branches where things don't fit. A syntactic path which absorbs the whole utterance "wins the day," so to speak, because associativity saturates and continues from the winning path, while it dies out along any obstructed branches. I would now like to suggest that ambiguity can be seen in the light of the saturation-principle. An utterance is ambiguous as long as it fully traverses two separate syntactic paths unto naturally ending destinations. Notice that this description of ambiguity also covers the "subset-ambiguity" mentioned yesterday, because subset-ambiguity is fleetingly ambiguous, until post-subset morphemes carry the true and full utterance to the end of the true path. Now, is there a problem of going back and negating or invalidating the momentary subset-ambiguity? I suspect that such a problem does not exist, because there is only a momentary disturbance insofar as the hearer's psychological belief-structure is momentarily shaken by the erroneous comprehension. The belief-structure is a self-adjusting network of extremely free associativity. An erroneous comprehension stemming from ambiguity can therefore just ripple away or be cancelled out within the "panpsychic" belief-structure. In the language-learning of an organism, comprehension must actually be learned before generation can be learned. In recent weeks I have worked on generation first, because it lent itself to analysis. I knew that I had much to finish in the theory of generation, but yesterday I suddenly realized that I might as well move into comprehension, if it seemed presently easier. So yesterday I started with sweeping generalizations, because I hoped thereby to move deftly in upon the quasi-neuronal switching functions at the heart of comprehension. I expect to make the comprehension-process more or less a reversal of the generation-process. I am not yet sure whether I will have to create the analog of a two-way street with separate lanes, or whether the generation/comprehension flow will be bidirectional along the same transmission lines. It is quite possible that the various associative recall-tags will be bidirectional, but that there will have to be separate, co-mirroring syntactic structures for generation and comprehension. Whereas inflectional endings had been added on by syntax-nodes in the generation-process, inflectional endings in the comprehension-process will probably serve to guide words to the proper syntax-nodes, during the saturation-process. I may use "neuron-fatigue" as a way of shifting through the variant possibilities for comprehension of an ambiguous utterance. Any malappropriate, initial explication of an ambiguity would slow down in its firing on secondary or postsecondary comprehension-passes, so as eventually to yield in favor of a fresher, perhaps more appropriate explication. You see, it is important that comprehension not have to be consciously labored at. 6 JUN 1979 Disinertiativity through Transabstractivity I need two clear and succinct terms to describe two important processes in the mind. I think the first term will be "transabstraction." A non-thinking central nervous system can be quite efficient at receiving inputs and bridging them over to the various outputs. However, depending upon the relative level of evolution of the CNS, the bridging of inputs to outputs will probably have to be rather direct and in strict registry. On the other hand, in a mind capable of abstract thought, the conscious mind has all its outputs freely at its disposal. Even if the outputs are kept in registry at lower levels (of reflex or habituation), the conscious mind has an overriding particularistic control of the outputs. The mind can prolongedly contemplate its options for output, and then freely select and initiate the desired motor activity. Likewise the conscious mind has free access to its permanent experiential memory stretching back through its lifetime. The concept of ego, ensconced within the conscious mind, wanders freely up and down the memory channels, creates new verbal thought, and freely reviews its options for motor initiation. Perhaps I should use the word "transabstractivity" to refer to the free linkability of practically all data and information within the thinking conscious mind. The second term which I need will probably be "disinertiativity," to describe the mind's process of potentiating the action of previously inert objects. Many things, such as the surface of the moon of Earth, would remain extremely inert, were it not for the transabstractional or transabstractive disinertiativity of the mind. This process was discussed in the Nolarbeit of 28JUN1975. Transabstractivity in disinertiation releases the logical tensions potentially existing between inert objects. 14 JUL 1979 The Implicit Mechanism of Attention The distinction between inherent and solely operational qualities. When we build a transabstractive mind, as described so far by the Nolarbeit, many of its functions are designed explicitly into the system. For instance, it shall be able to associate memories and to habituate rules of grammar. However, I am beginning to suspect that there may be additional functions which become available even though we did not try to design them into the system. My first idea of what to call them is "implicit functions," but I wish I had a more descriptive term. Anyway, the idea here is that these unexpected implicit functions spring into being just because we have built such a powerful transabstractive system. Although the examples which I am going to present may be erroneous, the idea may still be valid. First of all, there may be some really unexpected implicit phenomena, such as the ability to dream, to sleepwalk, or to be hypnotized. But what I had in mind first was the possible explanation of a mechanism of attention. In trying to design the uppermost intellectual level of a conscious, thinking mind, I have operated on a plane where there must be freedom and nothing but freedom. That is to say, the fact, that the design was of the very highest level of a thinking mind, has perhaps certain consequences. One consequence is perhaps that the design at the top is actually much simpler and less complicated than many systems operating at lower levels, such as perhaps in the cerebellum. [See Albus in BYTE, July, 1979.] Such simplicity would perhaps be due to the rationale that freely thinking consciousness must of necessity be quite isolated and protected from possible interference from lower levels. If lower levels must interfere, they should interfere only in special or extreme circumstances, such as those of instinctual drives or serious dangers. It may be that a mechanism of attention results unexpectedly and implicitly from the design of the topmost, transabstractive level. Attention would work in the following manner. We all know that we can be attending to one voice among many, and end up really hearing and understanding only that voice to which we "paid" attention. Well, part of Nommultic theory says that we actually heard and retained as engrams the whole milieu of sounds, but that only those sounds were remembered to which access was gained and maintained via associative tags. (Now, as ideas are beginning to dawn on me, will it be the idea of selective fixation of associative tags, or the idea of the importance of the employment of reaffirmation tags?) This explanation of attention may have to go hand-in- hand with development of ideas about language-comprehension. Anyway, when we are listening to one speaker among, say, three speakers, we came to be listening to that speaker by a process of association, and we remain listening to and attentive to that speaker by a process of association. Perhaps, then, any vortex of associativity actually constitutes a so-called "mechanism of attention." If we have been attending intensely to a phenomenon in our perception, then far back into our memory channels there are instances of associativity going on with respect to aspects of our attention. As each new slice of perception is perceived, it is associated, not just with other incoming perceptions by simultaneity, but also with recall-fetched memories that have quickly moved to the freshest extremity of the memory channel. Thus a past engram-slice becomes duplicated at the freshest extremity, and this process enables attended percepts quickly to become associated with vast past experience. After all, each duplicated engram-slice is now like a node present in two places at once. Since association is happening so intensively to each previous slice of the attended perception, each fresh slice is avidly taken up by the same state of affairs. Of course, if we hear something shocking or surprising, we may stop paying attention as our thinking drifts away through associations from that which shocked or surprised us. How can I say that associative attention puts us into a state of expectancy, expecting more of the same from whatever has seized our attention? Well, I can make use of my old friend, the concept of neuron- fatigue. As fresh and old engram-slices are being associated with what we are (probably with one main sensory channel) paying attention to, many old engram-slices are going into semi-activation but then losing out to those engram-slices which reach full activation. Then neuron-fatigue eliminates the fully activated engram-slices, but leaves many semi-activated memories in a sufficient state of residual activation as to constitute, psychologically, a state of expectancy. The mind is expecting certain associations simply because it is readier to make them than others. So a mechanism of attention is implicit when you build a conscious, intelligent, associative, transabstractive mind. 21 NOV 1979 Comprehension We can simplify the attack on language-comprehension by imagining a society of minds that enjoy the generativity of only one syntactic tree: subject - verb - direct object. We can say that the language contains only nouns and verbs. Actually, language-comprehension will start whenever the perceiving mind seizes upon an initial morphemic word as fitting into any one of perhaps many syntactic lead-offs. From now on here I want to posit a logicoconceptual cable for purposes of comprehension. Let us reflect as though the comprehension L-C cable were separate from the generation L-C cable, and then later we can see if we would want both L-C cables to be one and the same. When the initial word of a sentence is perceived, it goes to the freshest extremity of the auditory channel. As it filters down through the auditory channel, it will stimulate some one ultimate-tag, or group of same- word ultimate-tags, most strongly. That one ultimate-tag, with or without its fellows, will "supratraversially" access and activate a conceptual fiber in the logicoconceptual cable for comprehension. Reaffirmation will then occur as the new ultimate-tag of the fresh auditory engram, through simultaneity, attaches itself to the same abstract fiber which was supratraversially stimulated in the L-C cable. Thus updating, and perhaps even initial language-learning, occur. Now, what good does it do the incoming word to have stimulated an abstract fiber in the comprehension L-C cable? Well, that process tentatively establishes the part of speech of the word, because the abstract fibers of the L-C cable are layered (bundled) according to part of speech. Note that nothing has been established as to the grammatical function of the word, such as subject or direct object. Function will be determined either by interpretation (based perhaps on word order) or by clues from inflection. Now let's say that a second word, such as a verb, comes in. The incoming verb will access and reaffirm an abstract fiber in the verb-layer of the L-C cable. Then a second noun, a direct object of the verb, will access and reaffirm a second abstract fiber in the noun layer of the L-C cable. Of course, the comprehension-processing of the first noun as subject will have to have taken place quickly already so that no confusion occurs between the two nouns. Now, this sort of sentence states a relationship involving two nouns and a verb. The sentence can convey knowledge of the relationship. Mere activation and reaffirmation of L-C cable abstract fibers do not set up the internal representation of the relationship carried in the sentence. The relationship is carried by the syntactic structure of the sentence, and therefore the relationship must be reconstructed internally through activation and reaffirmation of a syntactic structure residing in the "syntax cable" of the abstract memory channel serving comprehension. Let's looks at that dynamic bundle of potential meaning and information, the verb. So far we have merely accessed an abstract conceptual fiber representing the concept of the verb. In the experiential history of the organism, knowledge of the meaning of the verb must henceforth (either long or briefly) be associated both to the subject and the direct object. In other words, the direct object acquires some of the semantic data underlying the verb. But the direct object is epitomized only as a morphemic word occurring here and there in the experiential stream of the auditory memory channel. Those happenstance morphemic-word loci are not going to be the fastening point to which semantic data coming from a verb are now henceforth tied as a result of comprehension of the sentence. No, those auditory engrams are too phantom-like; instead, the abstract conceptual fiber for the direct object will be the steadfast point to and from which all semantic connections are made. The auditory memory engrams are just a means of rapid access to the abstract conceptual fiber. The abstract conceptual fiber for a noun can lead semantically to data stored in all five sensory memory channels. Thus, when the semantic background of a noun is invoked, perhaps during time-extended or lingering comprehension, all sorts of associations fan out into the memory channels of all the senses. Then, within the individual memory channels, the comparator-effect allows circuitous re-entry back into the abstract domain. In the case of verbs, there is an escalation away from the five raw sensory memory channels. The abstract conceptual fiber for a verb has its fanning-out "feeler-apparatus" which has hold of the semantic bundles underlying the verb. Dare I say the following? When semantic contact is to be made from a verb to its subject and direct object, the quasi-reaffirmation process causes the feeler-apparatus of the verb to connect to its subject and to its direct object. Of course, the reaffirmation process works only when the two things to be affirmed have been stimulated separately. It is the syntactic structure of the syntax cable which will proffer abstract fibers as candidate elements for the feeler-apparatus to take hold of and affirm. It is semantically essential that the syntax cable guide these affirmations, because other plausible guides, such as juxtaposition of word-order, are not reliable or strict enough. So inflection guides the syntax cable, and the syntax cable guides the affirmative constructions of relationship, and the resulting structures constitute the new condition of knowledge engendered by the comprehension of the sentence. It is possible, during reflective thought, that an abstract conceptual fiber may perhaps "vibrate" in concert with its dually sensory and semantic background out among the various sensory memory channels. Suppose that one concept is momentarily of supreme importance and interest to the mind. The abstract fiber of that concept will be undergoing heavy use. It will not be important of itself, but rather its situational ambience will be making it important. Thus, over and over again, the conceptual fiber will semantically project out into the psychic ambience, and the psychic ambience will circuitously rush back to the important fiber. Meanwhile, all sorts of words will be coming to mind in a brainstorm of verbal mentation. Some of the words and psychic currents will trigger generation of sentences of thought. In fact, when a sentence is generated, it immediately goes through the above described comprehension-process, so that its peculiar semantic relationship can be reaffirmed through syntactically guided structuring within the logicoconceptual cable of comprehension, which cable may or may not be the same cable as that of generation. The generating of a sentence is the verbal expressing of a relationship. Before generation of the sentence, the relationship is only pregnantly available to the mind. If the mind is generating a sentence about what it is perceiving externally, then the relationship is coming from without the mind. Such an external relationship can be very clearly perceived and yield a strong formulation as a sentence. However, suppose that a sentence is being generated as a result of internal meditation or reflection. Then a new relationship is about to be discovered and is about to come into expression in the sentence. Or perhaps merely an old relationship is about to be reiterated. At any rate, the about-to-emerge relationship is pregnantly available within the interior psychic ambience. Logical "tension" builds up and finds release in the generation of the new sentence. Such logical tension can come from one new external fact entering a mind. If one bombshell of a fact enters a mind, then the equilibria of many old relationships can be disturbed, so that a vortex of thought quickly flares up and slowly subsides. On the one hand, generation of a sentence sets up a relationship, but it is the immediate comprehension of the new sentence that affirms the relationship and leaves a structured memory trace of it. The fortuitous network of associations that generates the sentence does not have to show up again unchanged as that network which remains structured in memory after comprehension of the sentence. Typically, a very tenuous network might generate the sentence, and then, after immediate comprehension, there might remain a network as a structure carrying a broader and fuller semantic impact. That is to say, a tenuous, fleeting relationship finds expression in a sentence, but thenceforth the full impact of the relationship stands formalized in the sentence, ready to operate strongly each time the sentence is run through comprehension. The shift in semantic impact between generation and comprehension of a sentence is perhaps where mental creativity arises. The fortuitous network that generates a sentence is not in itself creative, or is it? Well, we could say that the initial, fortuitous seizing upon the relationship is creative because the relationship was perhaps never perceived before. And each time that the new sentence is run through the mind, the success of the creativity can show itself, because new aspects of the relationship can be realized with each comprehension. 23 NOV 1979 In the months that I have been thinking now and then about language- comprehension, I have been concerned with determining just what "trace" is left after a sentence has been comprehended. Quite early it was clear that there would have to be an episodic trace of the sentence as a sequence of sounds in auditory experiential memory. However, it is now seeming clear that the purely sensory recording of the sentence in auditory memory probably plays little role in the organism's structure of knowledge. Even if the words and morphemes of the sentence have their full complement of onset-tags and ultimate-tags, those tags just lead to abstract conceptual fibers in the abstract memory. The import of the sentence is in the relationship which it asserts among the concepts. We might say that a sentence adjusts the background of all its concepts. Although primitive or rudimentary concepts may be rooted in sensory memory, more complex or abstract concepts are probably rooted in a structure of relationships to other concepts. The concept of a transitive verb is probably strictly furcated into a set of relationships with subjects and a set of relationships with direct objects. During comprehension, each set is probably attached on the basis of syntax or inflection. Each transitive verb implies attachments based on relationship. A verb as a concept is comprised in its relational history. The temporal progression of conceptual knowledge is constantly being "woven" as verbal thought momentarily specifies the varying relationships among concepts. A comprehensional relationship of a concept cannot ultimately be distinguished from its semantic "definition." However, the semantic definition is like the statistical "average" of all the previous relationships of the concept. When an incoming sentence asserts a relationship, the heavy preponderance of past relationships is used in the process of comprehending the sentence, and at the same time the gist or scope of past semantic definition can henceforth be slightly altered if there is any novelty in the newly asserted relationship. The comprehensibility of a concept grows each time the concept is newly comprehended. Either a standard meaning is reinforced once again, or a shift in meaning is adduced. If someone tries to use a verb in an improper way, then the asserted relationship will not be believed. A believed relationship will be affirmed in subsequent associative thought, but a discredited relationship will die out for lack of reassertion. The picture which I am painting here is perhaps hard to believe. Thousands of unitary abstract fibers hold the concepts in a mind. The rudimentary concepts are defined in terms of sensory memory, and the complex concepts are defined in terms of one another, or of rudimentary concepts, or of sensory memory. The actual attachments of relationship to a concept are effected by means of the myriad concrete associative tags which are singly available to the conceptual fiber at each pulsed moment of ratiocination. The length, the long dimension of the abstract conceptual fiber signifies its temporal aspect and its changeability over time. Any new relationship is understood in terms of potentially all previous relationships. 26 NOV 1979 It is really quite a claim to state that intelligent consciousness operates not in the sensory memory channels, but in an immense conduit of single-concept fibers. It is quite a jump from orderly, even rigid, sensory memory into the stick-forest of conceptual fibers. At first blush, it might sound ridiculous to assert that a stick-forest of fibers can think. But the thinking results from or as the interconnections among the fibers. In this light, I would like to mention copulative verbs such as "be" or "become." It looks as though verbs are going to have to have "portals" or partitions for at least subject and direct object, if not also for semantic attributes of the action expressed by the verb. It is possible that there could be two main classes of the semantic tags attached to a verb-fiber: subject-linkable tags and object-linkable tags. When a verb-concept is invoked, those two linkage-classes have to be assigned. When the action of a verb upon a direct object is comprehended, what supposedly happens is that the object-linkable tags become linked to whatever concept is momentarily the direct object. The momentariness of interconceptual taggings is very important. It may be that the tags are extraordinarily strong only when first established. When a non-transitive, copulative verb such as "be" or "become" is used, what we probably get is the unimpeded through-linkage of the subject- portal to the quasi-object-portal. The various forms of the English verb "to be" are limited in number, and each form probably has its own concept fiber. During comprehension, the syntax cable probably causes tagging of the subject-fiber to the verb-fiber and of the verb-fiber to the quasi- object-fiber. Once these tags have been set up, the subject-fiber is recentissime linked, via the rather meaningless verb-form, to its predicate nominative as a quasi-object. Now, the linking by the syntax cable is momentary and transitory, but these affirmatory links are permanent. Henceforth, invocation of that one-time subject-fiber must also tend to cause invocation of that one-time predicate nominative. But I think I see now that the residual cross-linkage must not go via the verb-fiber of the form of "to be." No, the verb-fiber only established the linkage. Thereafter, the verb-fiber must be free to service other accounts, so to speak. Suppose that the mind comprehends and believes the sentence, "John is a midget." Henceforth, the concept-fibers for "John" and "midget" should both be capable of invoking each other. Each of the two major fibers has become like a subtag to the other. There is no interference from the verb-fiber for the word "is," because it did not really attach anything of its own nature to either the subject or the predicate nominative. Thus forms of the verb "to be" serve as manipulative or copulative instruments of the syntax cable. The next question is, what associative linkage-tags are established by a transitive verb? I just got an interesting idea about the complementary roles perhaps played by abstract memory and sensory memory. I would hazard a guess that a transitive verb does not establish a tripartite link involving subject, verb, and direct object. During comprehension, the syntax cable establishes various links through simultaneity. However, it looks as though the links to the verb have to be directional in nature. Suppose that the only abstract link established between subject and direct object is that of simultaneity. In other words, the subject is linked to the subject-portal of the verb, and the object-portal of the verb is linked to the direct object, but there is no direct, permanent link, via the verb, between subject and direct object. There ought not to be any link via the verb, if the verb is to operate independently. But the various words are linked to the auditory memory channel by simultaneity. We can view the comprehension-process in the light of dimensionality. The work of the last year or two has tended to map out a very simple dimensionality, namely: that past experience is along one dimension, while all present experience takes place on a plane or slice which processes information along pathways lying at right angles to the dimension of the past. This idea of a single plane or slice for present experience keeps things simple by allowing an indefinite number of slices of present experience to be added on as time passes. Each slice of present processing can be interacting with memory and control traces lying anywhere along the past dimension, but past slices are never altered, and new connections are made only in the plane of each fresh, momentary slice. Information can flow in and out of past slices, but past pathways are basically unalterable, except for the possibility that shifts due to neuron-fatigue can shift the routing of information. So there we have the picture. The preterite mind is a firm structure of belief and knowledge, and only the present extensions of that structure can admit of change. When we comprehend a sentence in the present, we are just reweaving the ties among fibers parallel to the past dimension. The grammar of a sentence operates through the syntax cable to tell us what connections to make among the concepts named in the sentence. 28 NOV 1979 Probably the most substantial concept in a mind is the concept of self or ego. It is almost taxing to concede or realize that this present theory of the mind requires the concept of ego to be physically located as a unitary fiber in the abstract memory channel. We can try to justify the elongation of the "punctum" of ego by saying that a concept has to remain relatively constant over time. So even if we did not think of a concept as a unitary point, we would be forced into positing and attaching a unitary line or fiber to achieve continuity or constancy over time. For instance, if we tried to think of a concept as a ringlet of points joined into a rough circle, we would still have to attach a unitary fiber somewhere so as to carry the concept out of the past and into each new moment of the present. So the concept of ego is one fiber among many in the abstract memory channel. The ego-fiber has both experiential and linguistic associations. In my writing of this year I have distinguished between the single abstract fiber and its numerous "concrete" associative tags. In a way, the tags are really "concrete" because they are laid down by a concrete happening in present experience. At any rate, I may have to introduce a concept of "neuroresistance" to keep an abstract fiber separate and buffered from its concrete tags. Let us say that an abstract fiber has no neuroresistance and that each concrete tag has some small but important measure of resistance to signal-propagation. The idea is not to slow signals down, but to gate them. We don't yet care whether the resistance is in the lines or at the synapses. Once we have the idea of neuroresistance, we can develop the idea of "proximolocality." Whenever several associative lines happen to converge on any point or locus, a process of summation gives added or enhanced significance to that point. For instance, a signal reaching it will branch out in multiple directions. If associative lines had no neuroresistance due to either time or distance, then any point in a network would be logically the same as any other point. Indeed, that situation is what we want for an abstract fiber, which is just an elongated point. But for associative links we want differences to arise in the conceptual import of various points. Ratiocination proceeds under the constraint of time. By dint of "proximolocality," a dense, concentrated point can briefly function as a unit appreciably distant from its surrounding points. Now, I am describing a topography as if it were in a flat plane. It is hard to imagine the physical accretion of indefinitely many associative links onto a thereby dense point in a single slice of topography. So we use myriad successive slices that function logically as if they were, in many but not all ways, a single slice. The abstract fibers ensure that there is no resistance between the slices, and the associative links introduce enough resistance so that points can become conceptually distinct. Now, there is a certain complementarity having to do with whether a concept is associated to abstract fibers or sensory memory fibers. An ultimate measure of the power of a neuronal mind lies in its ability to discriminate, to detect differences. The linguistic use of a large vocabulary relies upon the ability to discriminate. "Quaeritur": Does abstract ratiocination tend to merge concepts or to differentiate them? The primary and ultimate differentiation comes from the sensory memory channels. Many of the concepts in the abstract memory channel are associated with words stored in the auditory memory channel. Those unique words themselves are a means of sharp discrimination. The conceptual word- fibers receive discriminating information from all the sensory memory channels. Thus words of a reasonably concrete nature are sporadically kept differentiated by the senses. However, highly abstract words have no direct reference to the sensory memory channels. The abstract notions are not perceived directly by the senses, they are only comprehended in thought. Since we assume that all thought has its roots in the senses, we expect to find that even the most abstract of concepts are just built up through association with the more concrete concepts. A baby could assemble a concept of ego around the concept-fiber for its own name. Then, for sentence-generativity, it could gradually link its self-concept to the abstract fiber controlling the word "I" in the English- speaking auditory memory channel. Thus we have a rationale by which the acme of conceptual density can gradually travel among the points of the stick-forest of the abstract conceptual plane. No matter what abstract fiber the baby originally uses to gather information about itself, the incipient use of language will formalize a specific linguistic fiber as the center of the ego-concept. Subsequent reaffirmations will consolidate the primacy of the linguistic fiber. Is it possible that some concepts can operate only via syntactic relationships? Take an abstract concept such as "honesty" or "courage." What associations will lead away from the word-fiber? For one thing, there will be associations to specific episodes when the word was used. But it was used for an idea rather than a direct perception. We do not necessarily have to design into a mind the stipulated ability to entertain an abstract concept such as "honesty." Instead, a linguistic mind may have the implicit ability to handle such concepts. Suppose that we designed a linguistic mind to handle concrete perceptions and words naming them. What would such a mind do with words naming abstract concepts? For one thing, the mind would implicitly be able to record each word in audition and to set up an abstract conceptual fiber to govern the stored word. My claim that abstract concepts can operate only on the transabstractional level is reinforced if we agree that such concepts are never introduced to the individual by any other route than the quasi- transabstractional route of language. I mean, abstract concepts are introduced through the domain of language, and that domain is where they always remain. How primitive people would first develop and name such concepts, is another question. 3 DEC 1979 The Vault of the Mind Lately I've been taking pains to specify notions which I should keep in mind as I continue my design for artificial intelligence. For instance, there is the notion of neuronal prodigality, by which I should never shy away from positing a structure just because it seems to use an awful lot of neurons. Prodigality is both legitimate and mandatory if there is no other way to get the job done. Today I want to discuss the notion of the vault of the mind. Over the last two years I have often entertained the notion that the thinking mind seems to be separated from the physical universe by an almost impenetrable chasm. Among the floating ideas in the academic literature is the idea about how small is the largest number which the mind can know directly. I suppose that that number is less than eight, and around three. It has a bearing on how the mind perceives aggregates with many parts. My feeling is that that small number of knowable elements serves us as an indispensable bridge between the vault of the mind and the universe-at-large. I have lately had the following line of thought about concepts. A certain amount of perception of the physical world is necessary to get language going in the mind. Language involves variety, so there must be variety in that physical experience. But aspects of language operate both away from and towards variety. When language makes a concept out of things and names them, it is merging variety into unity. The environment ought to contain the raw material for a variety of concepts, so that not everything merges into unity under the scrutiny of mind. Now, somehow in the acquisition of language, the mind becomes capable of thinking up new concepts and thereby increasing internally the variety of its concepts, the basic set of which had had originally to come from external sources. I want to link this notion of conceptual fecundity with the notion of "proximolocality," the notion that local densities arise on the conceptual plane. We want to examine the mechanism by which conceptual variety is increased within the mind. I hypothesize that variety is increased by the formulation of sentences of thought. A concept, although unitary in its comprising nature, is nothing more than a focus of relationships. To establish or change a concept, you must establish or change the relationships. When logical tension builds up, a sentence flashes into being as a statement of a possibly old or possibly new relationship. Now, what is logical tension? Suppose that associative activity is frequently accessing two (or more) loci at the same time on the conceptual plane. Those loci thereby organize themselves as ingredients for a nascent sentence. If the two points can be thought of as points emanating tension, we can think of a diffuse percolation of associativity between the two points. The diffusion touches points satisfying the input requirements for selection of a verb. But once the verb is activated in a sentence of thought, the relationship is no longer diffuse; it is now direct via the verb. Now, a few days back I was unwilling to say that any recorded link from subject to direct object actually went through the verb. However, we might use the notion of simultaneity-freezing to make sure in the future that each verb gets associated with its one-time (same-time) subject and direct object. If the simultaneity-freeze establishes one direct link from subject to verb and another direct link from verb to object, then any subsequent associative quasi-recall of the event should cause the verb to spring to mind properly embedded in a sentence of thought somehow paralleling the original sentence. I say "quasi-recall" because we are not in the sensory memory channels where true recall occurs. We are making "quasi-recall" of assimilated information. The verb is still free and independent. However, if the verb is accessed by a subset of the elements of the simultaneity, then all of the elements will receive an impetus towards coming into play. Notice that this discussion bypasses those phonetic records of thought laid down in the auditory memory channel. The phonetic records are unwieldy and unnecessary. The thought occurred, and henceforth the conceptual mappings are altered. The phonetic channel can serve to communicate the thought or to record the exact wording, if necessary. The conceptual plane is a vast topography containing tens of thousands of concepts. The mind is pregnant with innumerable possible thoughts. Suppose that the conceptual plane mirrors relationships actually existing or potentially existing in the external world. The mental expression of all possible relationships can not happen all at once. As the mind cogitates, it produces some relationships which can then serve as the building-blocks for further relationships. Variety need not be reduced through cogitation, it can be enhanced and embellished. So a certain initial population of concepts has to arise along with the use of language. But then the mind with language becomes free within its "vault" to think transabstractively about any available element of knowledge. When I say "free," I mean that the broad flows and forces are free to converge in passing at any conceptual point on the network of the vast topography. The mind could even set up for itself the task of examining all available points one after another. The notion of the vault of the mind has to do with the presence or absence of certain concepts within the mind. The initial concepts are probably the least easy to establish, because the mind is so alien to the physical world. But, as the mind quickens within its vault, it operates ever more freely within its own realm of ideas. Words, a form of code, belong truly to the immaterial world of ideas rather than to the physical world-at-large. Once the vault of the mind has been established during infancy, incoming communications via language tend to reinforce the abstract uniqueness of each concept. Suppose that the mind hears a sentence purported to define or explain a certain concept. That sentence relates the target concept to various other concepts. The sentence is itself an abstraction, and it manipulates abstractions, namely, its component words. Being a statement of the relationships among concepts, a sentence is an abstraction of abstractions. However, each abstraction as a concept has a dynamic interactive potential. Belief and knowledge have accreted onto each concept. During comprehension, the semantic import of a sentence is absorbed by the conceptual topography of the comprehending mind. The credence granted to a sentence during comprehension is a function of the very process of associative absorption of the sentence. A mind is free to accept or reject any statement. A few weeks ago I was having difficulty figuring out the differences in comprehension of sentences believed and not believed. I was wondering how a mind could knowingly take in a lie, comprehend it, and not suffer damage to the conceptual apparatus of the mind. But a transabstractive mind tends to guarantee full dissemination of information. At any time that the comprehension of an obvious lie is operating somewhere on the conceptual plane, at the same time the massive operation of verisimilitude is operating elsewhere on the plane and in contravention of the lie. The lie does not enter the conceptual plane in an isolated way; if it did so, it might not be recognized as a lie. Whatever logic brands a statement as untrue also maintains the status quo of belief during assimilation of the semantic content of the statement. 5 DEC 1979 Phrases, Methods, and Categories For some time I have been wondering how the mind might use some sort of analog of "forward looking radar" to work the proper modifications upon the initial elements of phrases which are going to culminate in a noun at the end. For instance, in English, how does the mind know to put "this" or "these" in front of some adjectives followed by a noun? Yesterday I had a possible insight based on the German phrase, "An der schoenen, blauen Donau." The problem has long been quite poignant to me with respect to German, because the mind has to select the proper gender for an article well in advance of the utterance of the noun. Then adjectival endings depend upon whether or not the article was even used. From my own speaking of German acquired in my teenage years, I feel that my mind unconsciously launches into the correct form of article for an upcoming noun. Yesterday's explanation involves the slice of conceptual topography. Suppose that a German prepositional phrase is going to be constructed. The main element must necessarily be the noun at the end. The transitory, associative "valence" of that noun looms large in the topography. That is to say, the ongoing thought processes of association are just about to push that noun into conscious realization in a sentence of thought. However, the noun is not being associated to in isolation. The very nature or character of the association is about to be expressed by the preposition. The article and the adjectives are also being associated to in conjunction with the associative build-up of valence on the conceptual noun-node. What I have in mind is that the syntax cable will seize upon the noun-node as a starting point but that there will first be a backwards motion away from the noun and in search of a definitely primal element from which the process can then turn around and go forwards through conscious activation of the properly modified words in the auditory memory channel. In other words, the article- node of the syntax cable has to be addressed or dealt with before the forward swing of the sentence-construction can proceed. There is a mixture of inflectional information-flow along two routes. Only declensional information goes through ultimate tags in the auditory memory channel. Information as to the case, number, and gender of, say, an article has to be dealt with before the point is reached where auditory ultimate tags are involved. The gender of a German noun is somehow closely associated with the concept-node of the noun. The singularity or plurality of number is a pervasive concept that comes into play in the associative embroglio of the nascent phrase. The case required by the preposition is either fixed conceptually or chosen conceptually. We end up with a group of conceptual variables that are all going to interact before the conscious forward swing ensues. The fact that some of these concepts are pervasive, while others are unique, will bear upon the dimensionality of my solution to the problem. Now I would like to discuss my current methodology. Lately I have been working on language-comprehension. On several days I have written long sequences of rambling thought. Such verbiage may not seem valuable in its own right, but it has value within my method, which is to go back at many points in the future and look for the germination of ideas that are ready to bear fruit. The backdrop of so much verbiage is meant as a fertile panorama to react against creatively. If my general ideas about an abstract conceptual channel are on the right track, then I hope gradually to become able to work from a position of initially rough completeness. That is, if I am developing a correct structural framework for the inner workings of the mind, I should start finding certain felicitous results. For one thing, it should become possible successfully to turn my attention to minor but bothersome details. Successful accommodation of various details will bode well for the validity of the general theoretical structure. New work should come more easily and more quickly if a correct theoretical framework is yielding me a comprehensive overview. I would like eventually to join the worldwide circle of minds and to author two sorts of publications on my project: a factual, technical, highly descriptive and specific exposition of my results, and a more searching, questing generalization of all the philosophy involved in the project. You have to produce the hard results for the first kind of book to earn the right to expatiate in the other kind of book. Under the idea of the generalization-style of book I have in mind "The Phenomenon of Man" by Teilhard de Chardin. Such a book is rich in general ideas and does not have to present the specific solution to all our problems. My project has been broadening out lately because I have begun collecting certain categories on index cards: AI Notions, AI Questions, and Brain Information. Gradually I want to acquire an overview in those three areas. If I tabulate my guiding notions, I may be able to apply them better. At present they include such notions as dimensionality, disinertiativity, prodigality, and transabstractivity. I might just arbitrarily list the accumulated categories at certain times in this theory journal. By tabulating the questions, I should enhance my control of the project in many ways, such as in furthering the internal cohesiveness of the theoretical structure. By accumulating cards with units of brain information, I hope to conceptualize a mapping-out of the brain so that I can look for the physical manifestations of the things about which I am theorizing.