Max Jensen

 

     In my first paper for this class I outlined stemmatic syntax in general as a way to structure language. What I decided will be more useful for this class is to show how stemmatics can be applied more generally and give a general sketch of what stemmatics is doing with cognition. I would like to propose the use of stemmatics as a phenomenological account of physical events, and relate it to what the robot is doing “out-there-in-the-world.” I will relate this to some ideas that Professor Brandt discussed in the short note on robotics he sent out over the break. I will begin the paper by again discussing stemmatics, going further into depth, and discussing some of the points I merely glossed over in the first paper. I will discuss how at a basic level, the robot might be able to use this structure to give a report (in stemmatic terms) of what it has experienced. I will conclude by tying these ideas to language to how we might take these reports as the basics of human language.

 

    Stemmatics can be seen as a system that orders an event into a structure that is hierarchical (different meanings are attributed to different levels) as well as canonical (there are not an unlimited number of levels). In order to have an event we must have a relation or an action, hence, we must start with a verb. A relation would be marked by a copulative verb (to be) and an action would be marked by an action verb (to think, to swim, to give, etc.).

 

    The first node structure is that of a subject. The subject is primary because without a subject no action or relation can take place. This constitutes the “doer” of the action or the thing that is being related to another idea. With robotics, however, we cannot yet have the robot identify objects over time, and hence the subject must always be egocentric, i.e. the robot himself. I see a loose analogy here to Jean Piaget’s theory of cognitive development. The first stage of child cognitive development he terms “sensorimotor” stage of development.  In this stage, from birth until approximately two years of age, the baby understands the world through physical actions that s/he performs on the world. I consider this to be much like our robot in that its representations are minimal and generally egocentric.

 

    The stemmatic node pertains to a predication of the subject. This can be seen as a further specification of the subject, typically manifested as adjectives. Until the robot can represent other objects in space, all predications will necessarily be about itself. Take Figure 1 for example. Note that a joining preposition (when necessary) goes to the end of the branch.

   

The robot, so far, can only identify himself in space, and therefore his locative constructions will be confined to this stemma. The robot may also be able to make claims such as “I am low on energy,” “I am stuck.” The drawback to this is of course we will have to assign the words in an ungrounded fashion. The strongpoint of this is that the robot will be using the stemmatic system (even though it may not know what exactly that means).

 

    The third node is the direct object, it also includes infinitive verbs that complement modal verbs (must, should, be, have to, etc.). Yet again we are constrained to egocentricism until object consistency is developed, i.e. the robot can only operate on itself. Take for example figure 2.

 

“I move myself”

  

This would be the basic structure of the event: some action (i.e. the verb) with an egocentric subject (the robot) acting upon something (in this case, itself). We can also add in modal meanings to this, or even future states:

                 

“I want to move myself”           “I will move myself”

To add yet another level meaning to this structure, we can add a destination for the robot. This is the fourth mode, or the dative. The dative case in language is the case of giving, and is manifested most typically in an indirect object, and typically signals motion from an unspecified beginning to a specified end. For example, “I gave her the book” we do not know where the book started, but we do know where it ended (in the female’s hands). For the robot, the end location can be a specified point in space:

 

“I want to move myself to a coordinate”         “I move myself to a coordiante"          

The modal version of these two stemmata the dative branch (#4) falls under the verb “move” and not under “want.” We do not “want coordinate {x,y,z}” but rather we “move (to) coordinate {x,y,z}.” Also, the “myself” is emphatic and not necessary, but I use it to show how a verb can take a direct object.

 

    The fifth node is a little more complicated. It is typically referred to as co-agent. The diagram I used in my first paper shows the basic schema for this node:

 

Another addition that I can make to this diagram is that the prepositions “by” and “through” are also typically significant of a co-agent structure. Through, in a sense, is very similar to “with:” think of the sentence “I traveled by (using a) plane” meaning something very similar to “I traveled with the use of a plane.” The co-agentive structure, as shown by the above diagram, can profile both origin and path. Both origin and path for the robot are a set of numbers, but even this could be put into a stemmatic structure. Here are two examples:

              

“I move myself from (x,y,z)` to (x,y,z)”       “I move myself from (x,y,z,)` through a path to (x,y,z)”

                                                    

In the first example we have merely starting point and ending point. The way a stemma becomes linearized (i.e. turned into a normal sentence) depends on the language’s specific rules, hence the co-agent comes before the dative in English. In the second example, the “path” could be considered simply as the list of points that the robot moves itself through to reach a certain goal location. Another example of the co-agent is tool use. This could only come, yet again, once the robot is able to recognize objects over time. An example of this might be as follows:

 

“I opened the door with the key”

Here we have the robot (1subject) doing an action (verb, opening) to something (3direct object, the door) with the use of something (5co-agent, the key). This is just to show what other types of events are described by the co-agent node; we are still confined by the fact that the robot cannot yet recognize objects over time.

 

    The sixth node is the locative (i.e. location). This is a very easy node to work with in the sense that it has relatively limited meanings. The locative node is used primarily for establishing a topos, or the place, in which the event occurs. This could be thought of as the map that the robot is in currently, as compared to possible other maps the robot might have stored. For example:

 

“In a (specific) map, I moved myself to (x,y,z)”

Once a system of object recognition is developed, this will allow for construction such as “I put the key inside the contained.”

 

    The seventh node, epistemic, generally pertains to truth conditions and frequency, typically as adverbs. Another important aspect of the epistemic node is the conditional value. This is useful as it might give us a report on the how often the robot does something, or why it does it. For example:

              

“I move myself to (x,y,z) daily”           “I move myself to (x,y,z) if my energy is low,”

                           

The eighth node is simple: it merely connects one sentence to another by using a conjunctive preposition: and, or, but, however, etc.

   

    So far this paper has been a general guide to understanding how stemmatic trees can describe events. I will now attempt to describe how stemmatics may offer a way for the robot to use language, or at least a rudimentary for of it.

 

 As I have shown, stemmatics offers basic schemata of movement, most simply, source-path-goal. The idea of a schema is that it allows for compression. This means that many separate, seemingly dissimilar events can be described by one commonality (cf. they are schematized). These schemas then, are representations of different types of actions. In the note that Professor Brandt sent out over the weekend, he makes a claim that the difference between classical robotics and cognitive robotics is that there is an initial calculation which gives rise to representations. These representations can then be used for cognitive functions, such as creating narratives, goals, and so forth. These cognitive functions are then re-implemented through further calculations so that the robot can act intelligently.

 

Perhaps these stemmatic structures, which represent schemata, can be used in the representational-cognitive phase of processing. Most of what the human brain does is it ignores differences between events, this allows for the creation of categories. Different events can have identical stemmatic structures; hence we may even be able to make stemmata into categories of events, which can then be selectively employed to solve problems. For example, all different types of moving from one location to another can be described generally as:

 

    Language, unfortunately, is much more complicated than this. Construal problems, metaphor, idiomatic phrases, semantics, frames, blends, background information, etc., are all problems yet to be solved. What I do feel this model accomplishes is that it gives a basic sketch of how stemmatics can describe basic physical events, which no doubt came before more complicated mental events. With this laid down as a framework, we may be able to create more and more complicated schemata, which in turn would lead to more and more complex narrative structures. As for now we could perhaps work on having the robot give specific information about its movements around a map into a stemmatic tree, even if it is only a list of numbers. This may give us a start in plowing through the symbol grounding problem, as the symbols we generate will be integrated, by a rule, into an inherently meaningful structure.