You Can’t Store Meaning In Software

I’ve had some recent conversations at work which made me realize I needed to make some of the implications of my other posts more obvious and explicit. In this case, while I posted awhile ago about How Meaning Attaches to Data Structures I never really carried the conversation forward.

Here is the basic, fundamental mistake that we software developers make (and others) in talking about our own software. Namely, we start thinking that the data structure and programs actually and directly hold the meaning we intend. That if we do things right, that our data structures, be they tables with rows and columns or POJOs (Plain Old Java Objects) in a Domain layer, just naturally and explicitly contain the meaning.

The problem is, that whatever symbols we make in the computer, the computer can only hold structure. Our programs are only manipulating addresses in memory (or disk) and only comparing sequences of bits (themselves just voltages on wires). Now through the programming process, we developers create extremely sophisticated manipulations of these bits, and we are constantly translating one sequence of bits into another in some regular, predictable way. This includes pushing our in-memory patterns onto storage media (and typically constructing a different pattern of bits), and pushing our in-memory patterns onto video screens in forms directly interpretable by trained human users (such as displaying ASCII numbers as characters in an alphabet forming words in a language which can be read).

This is all very powerful, and useful, but it works only because we humans have projected meaning onto the bit patterns and processes. We have written the code so that our bit symbol representing a “1” can be added to another bit symbol “1” and the program will produce a new bit symbol that we, by convention, will say represents a value of “2”.

The software doesn’t know what any of this means. We could have just as easily defined the meaning of the same signs and processing logic in some other way (perhaps, for instance, to indicate that we have received signals from two different origins, maybe to trigger other processing).

Why This Is Important

The comment was made to me that “if we can just get the conceptual model right, then the programming should be correct.”  I won’t go into the conversation more deeply, but it lead me to thinking how to explain why that was not the best idea.

Here is my first attempt.

No matter how good a conceptual model you create, how complete, how general, how accurate to a domain, there is no way to put it into the computer. The only convention we have as programmers when we want to project meaning into software is that we define physical signs and processes which manipulate them in a way consistent with the meaning we intend.

This is true whether we manifest our conceptual model in a data model, or an object model, or a Semantic Web ontology, or a rules framework, or a set of tabs on an Excel file, or an XML schema, or … The point is the computer can only store the sign portion of our symbols and never the concept so if you intend to create a conceptual model of a domain, and have it inform and/or direct the operation of your software, you are basically just writing more signs and processes.

Now if you want some flexibility, there are many frameworks you can use to create a symbollic “model” of a “conceptual model” and then you can tie your actual solution to this other layer of software. But in the most basic, reductionist sense, all you’ve done is write more software manipulating one set of signs in a manner that permits them to be interpreted as representing a second set of signs, which themselves only have meaning in the human interpretation.

Q&A: Meaning Symbol Sign and Mind (Part 1)

On one of my recent posts, a commentor named “psycho” asked me some very good questions. I decided I needed to respond in more detail than just a single comment reply. I respond in pieces below, so just for context, here is psycho’s entire original comment.

But if you take more meanings, and put them together to get yet another meaning. Don’t you feel like those meanings were again like symbols creating a new meaning?

In my understanding, every bit of information is a symbol – what is represented by the invididual neurons in the brain. And if you take all related bits (that is neurons, symbols), and look at it as a whole, what you get is meaning.

The sentence is a symbol, and it is made of word-symbols. And the list of word-symbols makes a meaning. Which, when given a name (or feeling), becomes a symbol, that can be further involved in other meanings.

I’ll respond to each paragraph in a separate post, in order to get all of my thoughts down in a reasonably readable fashion. Here is part one.

Construction of Symbols

But if you take more meanings, and put them together to get yet another meaning. Don’t you feel like those meanings were again like symbols creating a new meaning?

I try to make a very strong statement of the difference between symbols, signs and their “meanings”. Perhaps I’m being too analytical, but it allows my to think about certain types of information events in a way I find useful in my profession as a data modeller. So let me try to summarize here the distinctions I make, then I’ll try to answer this question.

First, in my writings, I separate the thing represented by a symbol from the thing used as the representation. The thing represented I call the “concept” or “meaning”. The thing which is used to represent the concept I have termed “the sign”.  A symbol is the combination of the two. In fact, a specific symbol is a discrete object (or other physical manifestation) built for the express purpose of representing something else. That specific symbol has a specific meaning to someone who acts as the interpreter of that symbol.

As I have come to learn as I continue reading in this subject area, this is a somewhat ideosyncratic terminology compared to the formal terms that have grown out of semiology and linguistics. To that I say, “so be it!” as I would have  a lot of re-writing to do to make my notions conform. I think my notions are comparable, in any case, and don’t feel I need to be bogged down by the earlier vocabulary, if I can make myself clear. You can get a feel for some of my basic premises by poking around some of my permanent pages, such as the one on Syntactic Media and the Structure of Meaning.

There is obviously a lot of nuance to describing a specific symbol, and divining its specific meaning can be a difficult thing, as my recurring theme concerning “context” should indicate. However, within my descriptive scheme, whatever the meaning is, it is not a symbol. Can a symbol have several meanings? Certainly. But within a specific context at a specific time, a specific symbol will tend to have a single specific meaning, and the meaning is not so fluid.

How do you express a more complex or different idea, then? It is through the combination of SIGNS which each may represent individual POTENTIAL concepts that I am able to express my thoughts to you. By agreement (and education) we are both aware of the potential meanings that a specific word might carry. Take for example this word (sign):

blue

When I show you that word in this context, what I want you to recognize is that by itself, I am merely describing its “sign”-ness. Those four letters in that combination form a word. That word when placed into context with other words may represent several different and distinct ideas. But by itself, it is all just potential. When you read that word above, you cannot tell if I’m going to mean one of the colors we both might be able to see, or if I might be about to tell you about an emotional state, or if I might describe the nature of the content of a comedian’s act I just saw…

While I can use that sign when I describe to you any of those specific meanings, in and of itself, absent of other symbols or context, it is just a sign with all of those ambiguous, potential meanings, but in the context of our discussion, it has no specific meaning.

It has a form, obviously, and it has been constructed following rules which

Photo of an Actual Stop Sign In Its Normal Context

you and I now tacitly understand. Just as a stop sign has been constructed following rules we have been trained to recognize.

Imagine now a warehouse at the Department of Transportation where a pile of new stop signs has been delivered. Imagine they are laid flat and stacked on a pallet, just waiting to be installed on a corner near you.

While they lay in that stack, they certainly have substance, and they each have the potential to mean something, but until they are placed into a proper context (at a corner by a road) their meaning is just as ambiguous as the word sign above. If you were driving a fork lift through the warehouse and came upon the pallet, would you interpret the sign right then as applying to you? Probably not! Could you say, just be looking at an individual instance of a sign, exactly which cars on which road it is intended to stop? No, of course not.

So this is the distinction between the sign and the meaning of a symbol. The sign is a physical construct. When placed into a recognized context, it represents a specific meaning. In that context, the sign will only carry that one specific meaning. If I make another instance of the sign and put it in a different context, while the signs may look the same, they will not mean the same, and hence I will have made two different symbols.

Just to be perfectly clear on the metaphor I’m presenting, here is a “pile” of signs (words) which I could use in a context to express meaning:

blue

blue blue

blue blue blue blue

Now let me use some of them and you will see that given a context (which in this case consists of other word signs and some typcal interpretations) I express different meanings (the thoughts in your head when you read them together):

once in a blue moon

blue mood

blue sky project

blue eyes crying in the rain

But make no mistake, while i have now expressed several different ideas to you using the same sign in different contexts, they are each, technically, NOT THE SAME SIGN AT ALL! Rather they are four examples of a type of sign, just as each of the stop signs on that pallet at the DoT are examples of a type of sign, but each is uniquely, physically its own sign! This subtlety is I think where a lot of people’s thinking goes awry, leading to conflation and confusion of the set of all instances of a sign with all of the concepts which the SET of signs represents.

To make this easier to see, consider the instance of the word (sign) “blue” above which I have colored red. That is a specific example of the “blue” sign, and it has a specific, concrete meaning which is entirely different from the word (sign) “blue” above which I have colored green.  The fact that both phrases have included a word (sign) of “blue” is almost coincidental, and does not actually change or alter the individual meanings of the two phrases on their own.

Finally, since I have belabored my nit-picking a bit, if I were to re-word your initial statement slightly to use the terminology I prefer on this site, It would change to:

But if you take more [signs], and put them together to get yet another meaning. Don’t you feel like those [signs] were again like symbols creating a new meaning?

And to this question, it should be clear, that my answer is “Yes, precisely: when you put other signs together, you create new meaning”.

Meaning Over Transformation

This entry is probably ahead of the story, but I wanted to start moving into this subject and I’m not yet organized. It should make more sense later on when I’ve explained such things as the “magical” function M() more thoroughly.

Review: The Magical Function “M()”

As a review for those who may not have seen this function previously on this site, I have invented a mysterious and powerful function over all things used as signs by humans. Named the “M()” function, I can apply it to any symbol or set of symbols of any type and it will return what that symbol represents. I call it the “M()“ function because it takes something which is a symbol and it returns its meaning (that’s all of its meaning).

How Meaning Carries Over Symbol Transformations

When we move information from one data structure to another, we may or may not use a reversible process. By this I mean that sometimes a transformation is a one-way operation because some of the meaning is lost in the transformation. Sometimes this loss is trivial, but sometimes it is crucial. (Alternatively, there can be transformations which actually add meaning through deductive reasoning and projection. SFAT (story for another time))

Whether a transformation loses information or not, there are some interesting conclusions we can illustrate using my magical, mysterious function M(). Imagine a set β of data structure instances (data) in an anchor state. The full meaning of that data can be expressed as M(β). Now imagine a transformation operation T which maps all of the data in β onto a second set of data Δ.

T : β |–> Δ such that for each symbol σ in β, there is a corresponding symbol δ in Δ that represents the same things, and σ <> δ

By definition, since we have defined T to be an identity function over the meaning of β, then we can conclude that if we apply M() before and after the transformation, we will find ourselves with an equivalence of meaning, as follows:

By definition: T(β) = Δ

Hence: M( T(β) ) ≡ M( Δ )

Also, by definition of T(), then M( β )  ≡ M( T(β) )

Finally, we conclude: M( β ) ≡ M( Δ )

Now, obviously this is a trivial example concocted to show the basic idea of M(). Through the manner by which we have defined our scenario, we get an obvious conclusion. There are many instances where our transformation functions will not produce equivalent sets of symbols. When T() does produce an equivalence, we call it a “loss-less” transformation (borrowing a term from information theory) because no information is lost through its operation.

Another relationship we claim can also be defined in this manner is namely that of semantic equivalence.  This should be obvious as well, from reflection, as I was careful above to refer to “equivalence of meaning”, which is really what I mean when I say two things are semantically equivalent. In this situation, we defined T() as an operation over symbols such that one set of symbols were replaced with a different set of symbols, and the individual pairs of symbols were NOT THE SAME (σ <> δ)! In a most practical sense, what is happening is that we are exchanging one kind of data structure (or sign) with another, such that the two symbols are not syntactically equivalent (they have different signs)  but they remain semantically equivalent. (You can see some of my thoughts on semantic and syntactic equivalence by searching entries tagged and/or categorized “equivalence” and “comparability“.)

A quick example might be a data structure holding a person’s name. Let’s say that within β the name is stored as a string of characters in signature order (first name  middle name  last name) such as “John Everett Doe”. This symbol refers to a person by that name, and so if we apply M() to it, we would recognize the meaning of the symbol to be the thought of that person in our head. Now by applying T() to this symbol, we convert it to a symbol in Δ, also constructed from a string data structure, but this time the name components are listed in phone directory order (last name, first name middle name) such as “Doe, John Everett”. Clearly, while the syntactic presentation of the transformed symbol is completely different, the meaning is exactly the same.

T(“John Everett Doe”) = “Doe, John Everett”

M( T(“John Everett Doe”) ) ≡ M( “Doe, John Everett” )

M( “John Everett Doe” ) ≡ M( T(“John Everett Doe”) )

M( “John Everett Doe” ) ≡ M( “Doe, John Everett” )

“John Everett Doe” <> “Doe, John Everett”

When the transformation is loss less, there is a good chance that it is also reversible, that an inverse transformation T ‘ () can be created. As an inverse transformation, we would expect that T ‘ () will convert symbols in Δ back into symbols in β, and that it will also carry the meaning with complete fidelity back onto the symbols of β. Hence, given this expectation, we can make the following statements about T ‘ ():

T ‘ (Δ) = β

M( T ‘ (Δ) ) ≡ M( β )

By definition of T ‘ (), then M( Δ )  ≡ M( T ‘ (Δ) )

And again: M( Δ ) ≡ M( β )

Extending our example a moment, if we apply T ‘ () to our symbol, “Doe, John Everett”, we will get our original symbol “John Everett Doe”.

Meaning Over “Lossy” Transformation

So what happens when our transformation is not loss-less over meaning? Let’s imagine another transformation which transforms all of the symbols σ in β into symbols ε in Ε. Again, we’ll say that σ <> ε, but we’ll also define T ‘ ‘ () as “lossy over meaning” – which just indicates that as the symbols are transformed, some of the meaning of the original symbol is lost in translation. In our evolving notation, this would be stated as follows:

T ‘ ‘ (β) = Ε

M( T ‘ ‘ (β) ) ≡ M( Ε )

However, by the definition of T ‘ ‘ (), then M( β )  !≡ M( T ‘ ‘ (β) )

Therefore: M( β ) !≡ M( Ε )

In this case, while every symbol in β generates a symbol in Ε, the total information content of Ε is less than that in B. Hence, the symbols of the two sets are no longer semantically equivalent. With transformations such as this, the likelihood that there is an inverse transformation that could restore β from Ε becomes more unlikely. Logically, it would seem there could be no circumstances where β could be reconstituted from Ε alone, since otherwise the information would have been carried completely across the transformation. I don’t outright make this conclusion, however, since it depends on the nature of the information lost.

An example of a reversible, lossy transformation would include the substitution of a primary key value for an entire row of other data which in itself does not carry all of the information for which it is a key, but which can be used in an index fashion to recall the full set of data. For example, if we created a key value symbol consisting of a person’s social security number and last name, we could use that as a reference for that person. This reference symbol could be passed as a marker to another context (from β to Ε, say) where it could be interpreted only partially as a reference to a person. But which person and what other attributes are known about that person in the new context Ε if we define the transformation in such a way that all of the symbols for these other attributes stay in β? Not much, making this transformation one where information is “lost” in Ε.  However, due to its construction from β, the key symbol could still be used on the inverse transformation back to β to reconstitute the missing information (presuming β retains it).

An example of a one-way transformation might be one that drops the middle name and last name components from a string containing a name. Hence, T ‘ ‘ ( “John Everett Doe” ) might be defined to result in a new symbol, “John”. Since many other symbols could map to the same target, creating an inverse transformation without using other information becomes impossible.

Functions On Symbols

Data integration is a complex problem with many facets. From a semiotic point of view, quite a lot of human cognitive and communicative processing capabilities is involved in the resolution. This post is entering the discussion at a point where a number of necessary terms and concepts have not yet been described on this site. Stay tuned, as I will begin to flesh out these related ideas.

You may also find one of my permanent pages on functions to be helpful.

A Symbol Is Constructed

Recall that we are building tautologies showing equivalence of symbols. Recall that symbols are made up of both signs and concepts.

If we consider a symbol as an OBJECT, we can diagram it using a Unified Modeling Language (UML) notation. Here is a UML Class diagram of the “Symbol” class.

UML Diagram of the "Symbol" Object

The figure above depicts how a symbol is constructed from both a set of “signs” and a set of “concepts“. The sign is the arrangement of physical properties and/or objects following an “encoding paradigm” defined by the members of a context. The “concept” is really the meaning which that same set of people (context) has projected onto the symbol. When meaning is projected onto a physical sign, then a symbol is constructed.

Functions Impact Both Structure and Meaning

Symbols within running software are constructed from physical arrangements of electronic components and the electrical and magnetic (and optical) properties of physical matter at various locations (this will be explained in more depth later). The particular arrangement and convention of construction of the sign portion of the symbol defines the syntactic media of the symbol.

Within a context, especially within the software used by that context, the same concept may be projected onto many different symbols of different physical media. To understand what happens, let’s follow an example. Let’s begin with a computer user who wants to create a symbol within a particular piece of software.

Using a mechanical device, the human user selects a button representing the desired symbol and presses it. This event is recognized by the device which generates the new instance of the symbol using its own syntactic medium, which is the pulse of current on a closed electrical circuit on a particular wire. When the symbol is placed in long term storage, it may appear as a particular arrangement of microscopic magnetic fields of various polarities in a particular location on a semi-metalic substrate. When the symbol is in the computer’s memory, it may appear as a set of voltages on various microscopic wires. Finally, when the symbol is projected onto the computer monitor for human presentation, it forms a pattern of phosphoresence against a contrasting background allowing the user to perceive it visually.

Note through all of the last paragraph, I did not mention anything about what the symbol means! The question arises, in this sequence of events, how does the meaning of the symbol get carried from the human, through all of the various physical representations within the computer, and then back out to the human again?

First of all, let’s be clear, that at any particular moment, the symbol that the human user wanted to create through his actions actually becomes several symbols – one symbol for each different syntactic representation (syntactic media) required for it to exist in each of the environments described. Some of these symbols have very short lives, while others have longer lives.

So the meaning projected onto the computer’s keyboard by the human:

• becomes a symbol in the keyboard,
• is then transformed into a different symbol in the running hardware and operating system,
• is transformed into a symbol for storage on the computer’s hard drive, and
• is also transformed into an image which the human perceives as the shape of the symbol he selected on the keyboard.

But the symbol is not actually “transforming” in the computer, at least in the conventional notion of a thing changing morphology. Instead, the primary operation of the computer is to create a series of new symbols in each of the required syntactic media described, and to discard each of the old symbols in turn.

It does this trick by applying various “functions” to the symbols. These functions may affect both the structure (syntactic media) of the symbol, but possibly also the meaning itself. Most of the time, as the symbol is copied and transferred from one form to another, the meaning does not change. Most of the functions built into the hardware making up the “human-computer interface” (HCI) are “identity” functions, transferring the originally projected concept from one syntactic media form to another. If this were not so, if the symbol printed on the key I press is not the symbol I see on the screen after the computer has “transformed” it from keyboard to wire to hard drive to wire to monitor screen, then I would expect that the computer was broken or faulty, and I would cease to use it.

Sometimes, it is necessary/desirable that the computer apply a function (or a set of functions called a “derivation“) which actually alters the meaning of one symbol (concept), creating a new symbol with a different meaning (and possibly a different structure, too).

Packaged Apps Built in Domains But Used In Contexts

Packaged applications are software systems developed by a vendor and sold to multiple customers. Those applications which include some sort of database and data storage especially are built to work in a “domain”.

The “domain” of the software application is an abstract notion of the set of contexts the software developers have designed the software to support. While the notion of “domain” as described here is similar to and related to the notion of “context”, the domain of the software only defines the potential types of symbols that can be developed. In other words, the domain defines a syntactic medium (consisting of physical signs, functions and transformations on those signs, and the encoding paradigm).

But the software application domain is NOT its context. Context, when applied to software applications, is defined by the group of people who use the software together.

There’s a difference, therefore, between how developers and designers of business software think about and design their systems, and how those systems are used in the real world. No matter how careful the development process is, no matter how rigorous and precise, no matter how closely the software matches the business requirements, and no matter how cleanly and completely the software passed its tests, the community using the software will eventually be forced to bend it to a purpose for which it was never intended.

This fact of life is the basis of several relatively new software development paradigms, including Agile and Extreme Programming, and the current Service-Oriented Architecture. In each of these cases, the recognition that the business will not pause and wait while IT formally re-writes and re-configures application systems.

One of the shared tenets of these practices is that because the business is so fluid, it is impossible to follow formal development methods. In SOA, the ultimate ideal is a situation where the software has become so configurable (and so easy to use), that it no longer requires IT expertise to change the behavior. The business users themselves are able to modify the operation of the software daily, if necessary.

Software Applications As Perception

“The agent has a scheme of individuation whereby it carves the world up into manageable pieces.”                    K. Devlin, “Situation Theory and Situation Semantics”, whitepaper, 2004, Stanford University.

A software application creates and stores repeated examples of symbols defined within the context of a particular human endeavor, representing a perceived conceptual reality, and encoded into signs using electro-magnetic syntactic media. While the software may be linked through automated sensors to an external environment, it is dependent on human perception and translation to capture and create these symbols. Business applications are almost entirely dependent on human perception to recognize events and observations. That said, while the original “perceptions” are made by human agents, the software, by virtue of the automation of the capture of these perceptions, can be said to “perceive” such events (although this should be considered a metaphor).

Application design is in large part the crystallization of a particular set of perceptions of the world for purposes of providing a regular, repeatable mechanism to record a set of like events and occurrences (data). In essence, the things important to be perceived (concepts) either for their regularity or their utility by some human endeavor (context) will determine the data structures (signs) that will be established, and therefore the data (symbols) that can be recorded by the software system.

The aspects important to the recognition and/or use of these repeated events (e.g., the inferences and conclusions to be derived from their occurence) determines the features or qualities and relationships that the application will record.

Good application design anticipates the questions that might be usefully asked about a situation, but it also limits the information to be collected to certain essentials. This is done purposefully because of the fundamental requirement that the attributes collected must be perceived and then encoded into the symbology within the limited power of automated perceptual systems (relative to human perceptual channels).

In other words, because a human is often the PERCEIVER for an application, the application is dependent on the mental and physical activity of the person to capture (encode) the events. In this role, while the human may perceive a wealth of information, the limits of practicality imposed by the human-computer interface (HCI) guarantees that the application will record only a tiny subset of the possible information.

This does not pose any particular problem, per se (except in creating a brittleness in the software in the face of future contextual change), but just illustrates further how the context of the application is more significantly constrained than either the perceived reality or even the boundaries formed from the limits of human discourse of the event. This inequality can be represented by this naive formulation:

Μ(Ac) << Μ(Hc)

The meaning contained in the Application A defined by the context c is much less than the meaning (information) contained in the Human H perception of the context.

It is important also to note that:

Μ(Ac) is a subset of Μ(Hc)

The meaning contained in the Application A is a subset of the meaning contained in the Human H.

No aspect of the application will contain more information than what the human can perceive. This is not to imply that the humans will necessarily be consciously aware of the information within the application. There are whole classes of applications which are intended to collect information otherwise imperceptible to the human directly. In this manner, applications may act as augmentations of human perceptual abilities. But these applications do not of themselves create new conceptions of reality posteriori to their development, but rather are designed explicitly to search for and recognize (perceive) specific events beyond the perception of natural human senses. Even in these situations, the software can only recognize and record symbols representing the subset of possible measurements/features that their human designers have defined for them.

Hence, while software applications may be said to perceive the world, they are limited to the perceptions chosen a priori by their human designers.

Charting The Semantic Stream

…what Man touches, he stains with meaning…
…it flows like water, permeating everything…

Each computer system presents a different view of the world. Like water sitting in bottles, bowls and buckets with different shapes, meaning fills every available nook and cranny. But just as there are not enough bottles to hold all of the water of the world, no system can hold all of the meaning man can project.

This site is dedicated to capturing my ideas regarding the movement of concepts through human-defined artifacts. The goal is to capture an impression of the nature and properties of meaning and symbol, and to discover and map how meaning flows – what changes and what remains constant – as it passes from one end of civilization to the other.