On Information in Historical Sources

[ Track I-1 of this “blog”; see “read me first” for the description of tracks. ]

[ Warning: This is published as a blog; it obviously avoids the recommendation of restricting posts to 600 words or less. While I consider the form of a blog convenient to share developing ideas with some friends and colleagues as they arise, I do not follow the blogging spirit. This is effectively a series of lightly written papers. ]

1. Information in general

We live in the information age; nevertheless, when you ask what “information” is, the answers you get are so far apart, that it might have a good chance for the least defined term of the decade. So, if one embarks upon a study of what is different between the kind of information one encounters in contemporary documents and that which is used for historical research, one has to give a definition, or at least a description, of how one understands the term.

There are definitions of the term, provided by and used within different disciplines. I would like to start with one, which has been used under various names – Knowledge Pyramid, Ladder of Knowledge, DIKW model – in various disciplines: information science, philosophy of information, cybernetics and a few others, though rarely in information theory [Ackoff 1989, Ashenhurst 1996, Rowley 2007, Frické 2009, Saab 2011, Baskarada 2013, Jifa 2014, Duan 2017]. And usually not, or not much so, in computer science and information technology. The latter two are frequently quite satisfied with being able to represent information in data structures and process them by algorithms, without having to define what it actually is. Two comments on the model I start with: (a) It is a compromise between the numerous varieties which have been proposed by different researchers in different disciplines, most closely following [Favre-Bull 2001]. (b) I start with a rather naïve version; a significant adaption follows later in this paper.

In a nutshell the model assumes, that there is a hierarchical relationship between various phenomena related to “information”, as in graphic 1.

Graphic 1: Knowledge Pyramid

The definitions of these layers will follow soon, let me start with an intuitive example first. Let’s look at some medieval tally sticks. These were short wooden sticks, which were given as acknowledgment of some economic relationship, described in writing on the stick. Notches at the top ensured on the one hand, that the form of the stick was unique, at the other the notches in many cases representing also the amount of what was owed or simply counted. The stick was than split lengthwise and the two parts were kept by the two parties of the transaction. Having an irregular shape, the individual parts could not be tampered with. Graphic 2 shows such a tally stick, holding information. The notches on top represent the amount of what is described in the text.

Graphic 2: Information recorded on tally stick

Graphic 3, on the other hand holds only data. Whether the writing was never there, as the two partners of the transaction were illiterate, or whether it has eroded: We see notches, which counted something – or something else. Which something, well …

Graphic 3: Data recorded on tally stick

More formally: Data are marks in some representational system, which can be stored. Information results, when these marks are put into some context. So “22°” are data. “The temperature of this room is 22°” is information. Knowledge arises, when this information encounters the ability to draw advice for action from it. “Oh well, I do not feel overly warm just because I had to run to get here in time. I really should get out of my jacket.”

Seemingly a side track at the moment, but important later: Most researchers – notable exception: Luciano Floridi [Floridi 2011, 182 ff.] – do not think, that truth has to do anything with all this. Assume your knowledge of the world includes the “truth”, that vampires exist. The data represented by a color change of the garlic in your garden becomes the information that it is ready for harvest. Within your view of the world, it is valuable knowledge to deduct from that information the plan to use some of it to surround your windows.

This is not the only reason, that many people feel less than perfectly comfortable with the notion of wisdom in this context; which is why I ignore that concept in the remainder of this paper. How can we represent the relationship between the remaining conceptual layers more stringently? The usual approach is represented in graphic 4.

Graphic 4: Information among agents

Cognitive agents – you, me, a component of a “smart” piece of software – can perform its activity, as it has other information or knowledge as a context in the background: That a number you see on a thermometer represents temperature and how to translate that into action requires information or knowledge beyond the number indicated by the device. Both processes have in common, therefore, that they put a specific chunk of data or information into a context. There is a very big difference between the two levels, however.

To convert the data “22°” to the information “the temperature of this room is 22°”, requires contextual information, that in our sociocultural environment this is the way to read a thermometer. This is a context, which is common to all cognitive agents, which operate successfully in that environment. The contextual knowledge, that this temperature should trigger the action of getting rid of your jacket, is restricted to one specific cognitive agent, as it is bound to the biological entity supporting that agent, you. So the context required to convert data into information is shared between a larger number of agents, the context required to convert information into knowledge, is private to a specific agent.

If we try to use this model to describe what is happening in communication, we get the following graphic, inspired by [Favre-Bull 2001, p 87, ill. 47]. The icons of graphs, which represents the contexts, hint at the widely spread opinion, that networks are a useful model to understand the relationships in a semantic environment. With this in mind, we can describe the situation of two people discussing a possible purchase as illustrated by graphic 5.

Graphic 5: Discussing the price of a purchase

This graphic is to be read as follows: The person in the top left corner has been asked, what the price of an item is, which the person on the top right hand corner wants to purchase. Based on his knowledge about the price he fixed before or, if none has been fixed, the situation of his business and the going rates for such merchandise, he selects an amount. This is converted into information, combining an amount and a currency, and in the next step into data which can be transmitted. Whether the signal used for that purpose is a set of sound waves at a market stall or a string of bits transmitting from an internet shop via the WWW is irrelevant for our purpose. In both cases, the purchaser at the right hand side will receive that signal, hopefully undistorted by the acoustic noise at the market or the static electricity around the connecting lines. The signal reconstituted as data gets transformed into information as amount and currency and allows the purchaser to act upon the offer, contemplating it in terms of the desirability of the object and her overall budget.

That all of this works requires that the background knowledge of the two participants in the conversation and the common context of sociocultural conventions between them overlaps sufficiently, that they have the feeling to “understand” the other party. This overlap does not need to be complete: Whether the seller in a developing country believes to ask an outrageous price, while the purchasing tourist understands she’s been made a gracious offer, does not really hinder the action (even if the communication does not get everything across). But incompatible contexts may destroy the communication: If the tourist believes she must haggle, as this is allegedly required in the sociocultural context of the country where the exchange takes place, the seller offers a fixed price, however, the two parties will not get together.

So there are two levels at which communication may break down: The transmission of the signal amidst noise at the bottom of the diagram and the compatibility of the knowledge context – semantic context, for short – on top. May break down – but most of us have frequently encountered that two parties in a conversation do misunderstand what the other party says, so this is a more serious problem, than the speculative “may” indicates.

One might even say preliminarily, that this is the most crucial problem when applying this model to information as contained in historical sources. If you replace in graphic 5 the present day icons of seller and purchaser by photographs of statues of Pericles and Thucydides and the string “5 €” by the string “πόλεμος” the diagram is still completely appropriate to describe what Thucydides understood, when listening to Pericles explaining his policy, as both shared the same sociocultural context. If you replace the seller by Pericles and assume that the lady’s icon at the right hand side represents a modern historian, you immediately see that this will not work – a modern historian simply does not share the sociocultural context with Pericles. Which is an incomparably more serious problem, than the question how close the text of Thucydides we have is to what Olorus’ son actually wrote, leave alone what words the offspring of Xanthippus really had chosen.

Nevertheless, information technology in its broadest sense until quite recently had the tendency to focus on how to overcome the signal noise, not so much the semantic noise of a process of communication; therefore much of the discussion of information theory proper is concentrating on signal handling. We will try to point out, that this is not helpful to handle information as it occurs in historical sources in the next section and propose some concrete technical consequences of that in section 3. Just to clarify things, let us point out, why this somewhat distorted focus may have arisen.

Much of the information technology we have today and more of what constitutes information theory as a contemporary discipline rests upon the research paper by Claude Elwood Shannon published in two parts in 1948, entitled A Mathematical Theory of Communication [Shannon 1948]. Without the concepts presented there, electronic communication and signal technology would simply not be what it is today: no TV free of distortions, no phone connection without crackling background, most certainly no internet. Nevertheless Shannon was very clear on what he thought could be done and what could not be done. The second paragraph of his paper starts:

The fundamental problem of communication is that of reproducing at one point either exactly or approximately a message selected at another point. Frequently the messages have meaning; that is they refer to or are correlated according to some system with certain physical or conceptual entities. These semantic aspects of communication are irrelevant to the engineering problem.” ([Shannon 1948, 379] “Meaning” italicized by Shannon; last sentence by me.)

This as an introduction to the basic diagram on the model of communication employed, as reproduced in graphic 6, the traces of which you easily recognize in graphic 5.

Graphic 6: Shannon’s model of communication

Shannon’s paper was most fundamental to modern information technology; but it was not exactly easy to read. Being recognizably fundamental already at the time it was republished as a book already one year later in 1949, entitled The Mathematical Theory of Communication this time [Shannon 1949]. Not exactly easy to read, it was introduced by an introduction by Warren Weaver, as mathematician highly qualified to understand the original argument and personally highly qualified to write transparent prose, so his text probably influenced the perception of Shannon’s concepts outside of signal engineering much more than the original text. Now Weaver was fully aware of the semantic problems of communication. He lists three levels of communication problems:

Level A. How accurately can the symbols of communication be transmitted? (The technical problem.)

Level B. How precisely do the transmitted symbols convey the desired meaning? (The semantic problem.)

Level C. How effectively does the received meaning affect conduct in the desired way? (The effectiveness problem.)” [Weaver 1949, 24]

It should be obvious, that with rather limited effort our concepts of data, information and knowledge could be related to these three levels.

Two pages later, Weaver even writes:

So stated, one would be inclined to think that Level A is a relatively superficial one, involving only the engineering details of good design of a communication system; while B and C seem to contain most if not all of the philosophical content of the general problem of communication.” [Weaver 1949, 4] Which basically rephrases Shannon’s statement.

But barely two pages later Weaver continues:

Part of the significance of the new theory comes from the fact that levels B and C, above, can make use only of those signal accuracies which turn out to be possible when analyzed at Level A. Thus any limitations discovered in the theory at Level A necessarily apply to levels B and C. But a larger part of the significance comes from the fact that the analysis at Level A discloses that this level overlaps the other levels more than one could possible [sic] naively suspect. Thus the theory of Level A is, at least to a significant degree, also a theory of levels B and C.” [Weaver 1949, 6] (Capitalization “Level A” vs. “levels B and C” is Weaver’s.)

And he starts his conclusion triumphantly:

It is the purpose of this concluding section to review the situation and see to what extend and in what terms the original section was justified in indicating that the progress made at Level A is capable of contributing to levels B and C, was to indicating [sic] that the interrelation of the three levels is so considerable that one’s final conclusion may be that the separation into the three levels is really artificial and undesirable.” [Weaver 1949, 25]

How this is disclosed did always beat me, since I read it the very first time. It seems to be at the root of the popular perception of Shannon’s model, however.

Let me start another comment on the relevance of Shannon for the fundamental understanding of communication with an anecdotal chronological observation. One of the best known undocumented quotations from the early world of computer technology is the statement which Thomas J. Watson jr. of IBM allegedly made in 1943: “I think there is a world market for maybe five computers”. Is it not astonishing, that only five years later a signal technician created a theoretical model which is supposed to be at the bottom of digital technology, when signal technology at the time was still almost exclusively analog? Again, looking at what Shannon actually wrote helps.

He starts his reasoning with a model based on discrete signals. But in chapter III of his paper he extends it to continuous signals, starting with:

We now consider the case, where the signals or the messages or both are continuously variable, in contrast with the discrete nature assumed heretofore. To a considerable extent the continuous case can be obtained through a limiting process from the discrete case by dividing the continuum of messages and signals into a large but finite number of small regions and calculating the various parameters involved on a discrete basis. As the size of the regions is decreased these parameters in general approach as limits the proper values for the continuous case.” [Shannon 1948, 623]

So, discrete and continuous computations can relatively easily be mapped unto each other. As discrete and continuous are the appropriate technical terms for “analog” and “digital” which in common parlance – not the least in the so called Digital Humanities – have taken on an almost eschatological importance, let me reflect on that a little bit longer, even if it may seem to be a detour from the main line of argument in this paper.

Analog mechanical calculations are best envisaged by a thought experiment. Assume you want to add the two numbers 1.00134 and 0.48723. What you have are three precisely polished glass containers with extremely fine grained scaling at the side and a supply of water. To perform an addition, you fill 1.00134 liters of water into container A and fill 0.48723 liters of water into container B. Next you put the content of both containers into container C and read from the scale at its side, that it now contains 1.48857 liters. This is an analog computation – to the best of my knowledge never implemented with water, but an important line of research in the early days of electronic computing, when amperage and voltage were used instead of water, still used for specialized purposes. (The classically trained Humanist is right in reflecting on Archimedes’ method to measure the volume of an irregular object. The anecdote with the bath tub, for the not so trained reader. The fervent reader of alternative history science fiction, may, I vaguely remember, actually have encountered an hydro-analog computer somewhere.)

Now a digital mechanical calculation is best envisaged by thinking of a child’s building blocks. To add the numbers 3 and 4 you build two towers, one out of three, the other of four building blocks, put the one above the other and read the correct answer, 7. This principle can of course be extended quite considerably; there is no reason, why you should not build two towers out of 100134 and 048723 building blocks, putting one above another resulting in a tower of 148857 to which a decimal point would have to be applied. No reason why not to do so, except you would want to use considerably smaller building blocks than in the 3 plus 4 case.

If we doubt that it is possible to handle hundred thousands of building blocks: do not forget, that in our analog thought experiment we had accepted, it would be possible to etch a scale into glass, which would be fine enough to read one-hundred-thousandth liter.

If we continue our continuous (i.e., analog) thought experiment, we will eventually arrive at a scale, which allows us to read the amount of water represented by one molecule; if we continue our discrete (i.e., digital) thought experiment, we will ultimately arrive at building blocks the size of one molecule of an appropriate medium, water, for example. With other words: Both thought experiments ultimately converge. The difference between a discrete and a continuous model of computations by physical devices is simply derived from the precision of the instrument we use to measure the physical units. “The digital” is no category of eschatology but one of plain instrument making. If not a term for eschatology, one should even doubt, whether it really should be one of epistemology.

Bluntly speaking: A digital watch is a much better analogy of time, than a mechanical one. Its timer cuts time up into slices which are orders of magnitude more finely grained than the ones a mechanical watch can produce. Many “analog” films use crystals for light recording, which are coarser than the resolution of a digital camera. Their resolution is lower, therefore, than that of the digital camera which in turn is more analog, than the allegedly analog device.

Digital technology is frequently more analog, than analog one.

I beg the patience of the reader for this seeming fixation on the low relevance of the “digital” for the handling of information. The reason will become apparent later in the argument, when we grapple with one misunderstanding which is ultimately derived from a misapprehension of this principle – the ubiquitous “Computers know only yes or no; they are unable to handle nuances at all. (And that is ultimately, why they are unfit for historians and humanists.)”

2. Information in historical sources

One of the reasons, why the discussions about the application of computational techniques to the Humanities frequently leave one rather confused, particularly if they are organized under the label of “Digital Humanities” is, that their applicability depends heavily on the methodological position of the speaker. If, as F.R. Ankersmit assumes, the point of history is an exercise in literary presentation of reflections about high level politics, where even anthropological reflections on village communities, highly focused on narrative as they tend to be, are a deviation from the true path, there is little computer technology can do beyond replacing the type writer. If a philosophy or methodology of history is derived from the narrative to be produced, sources and how to extract information from them, are a secondary phenomenon. If a philosophy or methodology of history is derived from the problems posed by gaining knowledge from objects and documents that have survived, they are central.

My own position – clearly thinking, that historical research is defined by the way we get knowledge about the past, not how we present that knowledge to the public – will be the topic of another of these papers offered as blog posts, but the following considerations would be incomprehensible, if that position would not be stated briefly.

Let us start our consideration of the information contained in historical sources with graphic 7, modified from graphic 5 to apply that model to the situation a historian finds herself in, when she tries to understand Pericles’ reasons to implement his policy regarding the war with Sparta, as reflected by Thucydides.

Graphic 7: Reasoning about Pericles’ policies

Our friend faces two problems: The first we already mentioned, when discussing information as such. We do not share the silent assumptions of the sociocultural context (question mark in the middle), so we have no access to the context in which Pericles did formulate his policies. But we have ignored another problem so far: The information we have – or some of it in any case – did not originate from Pericles, but from Thucydides, being the result of an earlier communication process. Estimating the distortions produced by such earlier communication processes is the bread and butter of historical research, as far as it is focused on the content and not the literary qualities of historical writing.

The topic has a long tradition, therefore. Let us look at two sources on Hannibal Barkas, trying to understand, which of the three possible routes he took across the Alps with his elephants. Let us imagine two potential sources to solve that riddle. The one shall be a statue or bust of Hannibal. We use it to get insight into his character, whether he impresses us as restrained overall, having a character which implies that despite his undisputed audacity he was governed by a tendency to select of the risky alternatives the safest one over the advantages the other alternatives may possess. The other source shall be a remarkably deep layer of horse manure, found at one of the possible routes, indicating by various archaeometric methods that a very large number of horses and other producers of manure passed a specific Alpine pass at about the time of the Hannibalian crossing [Mahaney 2017].

Judging the relative value of these sources, historians of my stripe would argue for the horse manure. The artist who shaped Hannibal’s statue had almost certainly never seen him; and even if so, he almost certainly tried to express those features of his character which were emphasized by the contemporary discussion of the man. The horses producing manure had no intention whatsoever to express any opinion on their path.

Since the 19th century, this has been clearly fixed in historical methodology. Droysen [Droysen 1937, 38-50] clearly emphasized that traditions (Traditionen), like statues, which were the result of an intentional effort to leave a specific view of an event, were less valuable as sources than remainders (Überreste) which resulted from processes which were not controlled by an intention to leave a specific image for the coming generations. Albeit Droysen was thinking less of statues and horse manure, but rather of earlier historiography vs. the results of administrative documentation.

For us this implies, that a historian is not interested in the message the author of a source wanted to transmit, but rather in such insights about the situation, which a source provides as independent of the intentions of the author.

An example. Let’s look at Römischer Käyserlicher Majestät Ordnung und Reformation gutter Policey, im Heiligen Römischen Reich, zu Augspurg Anno 1530 auffgericht, usually quoted as Reichspolizeiordnung of 1530 [Weber 2001], in English best described, rather than translated, as constitutional law of the Holy Roman Empire, after some major revisions in force until 1806.

Observation 1: At a rather prominent place, this constitutional law contains the following section: IX. Von unordentlicher und köstlicher Kleidung. (IX. On irregular and costly clothing.) Extremely detailed regulations regulating members of which social group would be allowed to spend how much on their garments, a dress code allegedly preventing insensible luxury, at the same time carefully preserving the visible differences between social ranks.

What will a historian learn from the presence of that section? “OK, the last attempt at regulating expenditure of clothing did not work, again.” (Similar dress codes being a textbook example for history students that the permanent re-issuing of regulations is a sure sign they were ignored.)

Observation 2: § 2 of the subsection of this dress code which speaks “Von Bürgern und Inwohnern in Städten” (citizens and other people living in towns) prescribes: Deßgleichen sollen sie kein Tuch, die Elen über zween Gülden werth, ihnen anmachen lassen, oder einig Marder, Zobel, Hermlin und dergleichen Futter antragen. Wol mögen sie zum höchsten Marderkehln, und ihre Haußfrauen Fehine Futter gebrauchen. (Furthermore they shall not use any cloth which costs more than 2 guilders a yard for their garments, or any marten, sable or ermine or similar lining. At the most, they may use scraps of marten and their housewives squirrel for lining.)

What will a historian learn from that text? “Some plain citizens could occasionally afford rather expensive furs, if they did set their mind to it.” – “Marten were still sufficiently numerous in the woods, to make them a borderline case, when luxury was discussed.”

Observation 3: § 4 of the same section on townspeople reads: Wäre es aber Sach, daß ein solcher Handwerker in einer Stadt in Rath wird erwählt, alsdann soll derselb mit Kleidung sich nicht anderst, dann hernach von Kauffleuten gemeldt wird, zu halten Macht haben. (In case that a craftsman is elected into the council of the town, he shall keep for his garments the same rules which further below are defined for merchants.)

What will a historian learn from that text? “There were sufficiently many towns where craftsmen had achieved the right to enter the council of the city, that special regulations for them were needed.”

Let’s look at these historical findings again:

  1. OK, the last attempt at regulating expenditure of clothing did not work, again.
  2. Some plain citizens could occasionally afford rather expensive furs, if they did set their mind to it.
  3. Marten were still sufficiently numerous in the woods, to make them a borderline case, when luxury was discussed.
  4. There were sufficiently many towns where craftsmen had achieved the right to enter the council of the city, that special regulations for them were needed.

What have these four historical results in common? This is information which is implied by the source; it is in no way the message, which the authors of the law wanted to send out to their contemporary recipients. Shannon’s focus on the ability to transmit a message is therefore not a sensible model to understand the information contained in a historical source. A historian is an interpreter of messages, not a recipient. He or she may interpret it correctly, but if the original context of the messages is lost, all such interpretations have to be kept tentative. Which graphic 8 tries to visualize.

Graphic 8: Transmission model for historical sources

If we change the basic model of our understanding of how we process information, does that have implications on the technological solutions we should use? I think so. The following and final section will try to explore, where we have to deviate from the current implementations of information systems. To prepare for that, let me close this section with a high level view of the requirements of historical information systems, which I consider to be derived from our musings on the nature of information above.

An information technology appropriate for historical sources:

  1. Represents the artifacts as free from any interpretation as possible in the technical system,
  2. embeds them, however, in a network of interpretations of what they imply,
  3. provides tools which help to remove contradictions between such interpretations,
  4. accepts, however, that such contradictions may prove to resist resolution
  5. as well as that all interpretations always represent tendencies, no certainties.

3. Consequences for information systems

While to the best of my knowledge the above attempt to intertwine musings about historical sources with considerations of the nature of information as handled by technical systems is a rather rare exercise, musings about the nature of information in such systems are anything but rare. As mentioned at the very beginning, the knowledge pyramid has left traces in information science, cybernetics, the philosophy of information and a few others. But with all respect due to the excellent work done by all these disciplines within themselves, it is hard to see, how information technology would have developed differently from the way it did, if they would never have existed (extremely rare exceptions – Norbert Wiener, e.g. – proving the rule). The main systematic exception is cognitive science, which contributes its own share of theoretical reflections on the nature of information, frequently interconnects with the implementation of technical solutions however, and therefore actually influences technical development. If thinking about the nature of information in historical sources does not influence the development of information technology, at first look that just hurts the vanity of the thinker. At second look, one reaches the conclusion, however, that information technology serves historical research less than it should, if it implements assumptions which violate the principles formulated at the end of the last section. If historical information can only be handled by systems which cannot handle contradictory information, such systems are ultimately inappropriate for this usage.

In software engineering, we are familiar with the term “technology stack” or “software stack” describing the selections of technologies made at various levels to implement a system. A restrictive definition usually defines a specific stack as the choice of an operating system, a web server, a database system and a programming language. LAMP (Linux, Apache, MySQL, {Perl, PHP or Python} ) is the best known example. The term “stack” leads to the association, that once the bottom level has been chosen, the choice of the upper levels is severely restricted. The existence of the WAMP (Windows, Apache, MySQL, {Perl, PHP or Python} ) stack shows, however, that the metaphor fails here: The different levels are related, but not strictly hierarchical. In a wider sense, we speak of a technology stack also, if we try to describe the selection of software components used to implement a solution: say the selection of JAR packages employed to build a JAVA based system.

I propose conceptual stack as a new term for the combination of general concepts which go – implicitly or explicitly – into the design of information systems. A conceptual stack in that sense more abstract than the software stack, but sufficiently concrete to determine specific properties and capabilities of all information systems build upon that stack. As in the technical case, we assume these concepts to depend on each other, but not in a strictly hierarchical sense.

In contemporary information systems, I identify at least five conceptual decisions, which restrict their usefulness for the handling of information contained in historical sources as discussed above. These are:

(1) The interpretation of the signals used for communication with the concept of granular units of information. What in my understanding is Weaver’s distortion of Shannon’s concepts.

(2) The believe that, as bits can be used to conveniently implement Boolean logic, computer systems necessarily have to be based on binary logic. What in my understanding is the binary fallacy.

(3) The assumption that the language of historical documents can best be approached by analyzing their syntax. What in my understanding is Chomsky’s dead end.

(4) The approach to embed interpretations of an object into their representation. What in my understanding is the markup fallacy.

(5) The principle, that variables in a programming language are conceptually independent of each other, as long as they are not explicitly connected into a structure or object. What for reasons not immediately apparent in my understanding is the Gorilla syndrome.

To each of these conceptual layers a paper will be presented on this blog. For now, I restrict myself to describe them briefly and try to give hints how implementations of solutions could look like.

3.1 Weaver’s Distortion

My accusation against Weaver that in the attempt to make Shannon’s model easier to understand he mixed up different conceptual levels, can probably be made a bit more transparent by an analogy. Norbert Wiener said famously “Information is information, not matter or energy”. In the world of matter and energy we are perfectly aware, that there exist two closely connected but in many ways independent sub-worlds: A Newtonian one and a world that is ruled by Quantum physics. They are closely connected; nevertheless the confusing habits of quarks do not prevent the Earth to circle the sun in an encouragingly reliable way – even if gravitation, responsible for the reliability, can probably be understood only on the sub-nuclear level. My proposal is, that a similar separation can be used to understand the relationship between the world of data, turning into information and knowledge in the context of other data, and the signals constituting those data. On the possibility to use more than one theory of information in parallel see [Sommaruga 2009].

There exist such differences in the interpretation of signals in quite a few fields of computing. When, e.g., you look at image processing, the JFIF encoding of JPEG compression mixes two views: Most of the steps interpret bytes as numbers which are conceptually points in the continuum, so the fact that their representation is made up of bit strings is an irrelevant accident. In the final step of the compression algorithm, however, the blocks of numbers are handled as bit strings, the fact that they consist of numbers completely irrelevant, when compressed by Huffman encoding.

The computational legacy of Weaver’s distortion is, therefore, that the programming paradigms we use today inherit it. So operations which shall interpret information are handled by data types, which are exactly that: data.

One of the reasons for this is, that what you do with numbers can be understood and validated by the formal apparatus provided by analysis in the mathematical sense and particularly by numerical analysis among its branches. For strings similarly stringent formalisms have been developed.

But anything related to meaning seems to be more slippery. Though there have been attempts to change that: Keith Devlin specifically proposed a mathematical theory (well, a pre-mathematical one, in his own words) which addresses that problem:

“… whereas in this essay I am taking information itself as the basic entity under consideration. More precisely, I am seeking a specific conceptualization of ‘information’ as a theoretical ‘commodity’ that we can work with, analogous to (say) the numbers that the number-theorist works with or the points, lines and planes the geometer works with.” [Devlin 1991, 17] (All italics are Devlin’s.)

Staying at the lowest illustrative level, and knowingly ignoring, that the bulk of Devlin’s theory is built upon units one level of complexity higher [cf. Devlin 2009], we can describe his approach by the notion of an “infon” as the atomic unit of an information system. An infon is defined as

<< P, a1, …, an, l, t, i >>

The parameters are defined as:

P – an n-ary relation

a1, …, an – objects between which P holds

l – a spatial location

t – a temporal location

i – a truth value

Two comments:

(a) Objects a1, …, an: Best understood as other infons, as the whole system is most conveniently be understood as completely recursive. Keep in mind, that the recursion can always end by an infon, where everything consists of nil – though I am not sure, whether this understanding is Devlin’s or mine.

(b) Strictly speaking, Devlin defines an infon on p. 22 of Logic and Information [Devlin 1999] as << P, a1, …, an, i >> and adds l and t only in the following paragraphs. As it is obviously simple to replace these both parameters by nil when not applicable, I recommend to use the more complete definition for easier understanding from the start.

As an example Devlin [Devlin, 1991, 24] gives:

<< marries, Bob, Carol, l, t, 1 >>

To describe the information, that Bob marries Carol at a location l and a date t. As this is true, the final parameter is 1. If it would be an alternative fact, a.k.a. a lie, it would be 0.

Considering this as a basic notion of information has many attractions. On the one hand, it acknowledges, that information grows out of data in context; on the other, it reflects that knowledge does not have to be true. Zeus and Hera are married on Olympus, even if they do not exist and the spatial location of Olympus requires an interesting extension of the concept of space.

Two short comments:

(a) I strongly propose – see section 3.2 below – to replace Devlin’s binary truth values by continuous ones in the unit interval [0, 1].

(b) Infons remind one immediately and obviously of RDF triples. While an n-ary relation can obviously be represented by a set of binary ones, I would recommend to avoid this simplification. An n-ary relation holding at a given spatial and temporary location is something rather different from a set of binary relations with slightly different temporal and spatial coordinates between them.

Research proposal in software technology 1:

Implement infons for seamless usage in main stream programming languages. (Or possibly situations, the slightly more complex abstraction which actually is the subject of the bulk of Devlin’s theory, omitted here to simplify the argument.) ⁋

But our argument for the basing of information systems on other building blocks than the current data types does not end here. We have derived this requirement from our initial consideration that information should be understood in the sense of one or the other version of the knowledge pyramid, not on the level of signals.

For simplicity’s sake, we have so far avoided the discussions of the shortcomings of the knowledge pyramid itself, which has led to various criticisms against it and occasional calls for its abolishment. Let’s look again at the starting example.

Information results, when these marks are put into some context. So “22°” are data. “The temperature of this room is 22°” is information.

Well, yes. However: What about the number “22”? Before it turns up on a thermometer, it might measure the length of a room in feet, the weight of truck in tons, the distance of two towns … So: “22” is data; by being contextualized as “22°” it becomes information. And what about the bit string “0000001000000010”? It could be the “device control character 2” from the ACII table, the first half of the Unicode code for the universal quantifier symbol, … So: “0000001000000010” is data; by being contextualized as “the value of an integer variable” it becomes information. And so on.

At first look one can react in two ways to this discovery. Argue it contradicts the knowledge pyramid model; or argue, that it emphasizes its validity as a confirmation of the overwhelming importance of context, even if the transitions between interpretative levels are more complex than the simple 3 level model – data, information, knowledge – indicates.

I would recommend that we use it as an encouragement to look at another approach for the understanding of the concept of information, which unfortunately has left only very few traces [e.g. Kettinger 2010] in its discussion: Langefors’ “infological equation” [Langefors 1973]. According to him, the information communicated by a set of data, is understood to be a function i() of the available data D, the existing knowledge structure S and the time interval t which is allowed for the communication, given by the formula

I = i(D,S,t)

It should be understood, that “formula” here just stands as a help for conceptualization, not as an object in any calculus. Nevertheless, on can use it for a stimulating exercise in reasoning, where we assume that S represents existing knowledge, the context of the transmission of information. The interesting thing about Langefors’ equation is, that it introduces the time a communication – or the process of generating information out of data – takes. If information is derived from data in a continuous process, we must assume, that the longer that process may take, the more information we may extract – “more information” easily conceptualized as “information in a more complex context”. Formulated differently: If this is a process represented by a function, rather than a timeless transition, there is not a discrete, but a continuous relationship between data and information. That is, what has been data at one stage, is information at another. Still using a formula as a notation to help reasoning, we can therefore write

I2 = i (I1, S2, t)

to represent the notion, that the information available at time “2” is a function of the information available at time “1” and the knowledge available at time “2”, depending on the length of time we have available for the execution of that function. As I have discussed this in detail somewhere else [Thaller 2009, 345ff.] I will cut the argument short and just mention, that from this stage we can develop the argument further, representing knowledge, S, also as a function s(), rather than a static entity and arrive at

Ix = i (Ix-α, s(Ix-β, t(x- β)), t(x- α))

To be read as: The information available at time x is the result of an interpretative process i() which has interpreted the information available at an earlier point of time x-α over the time span t between x and α, in the context of a knowledge generating process s(). This knowledge generating process in turn has been running over the time span t between x and β, using the available information at the point of the time preceding x by β.

What I find fascinating about this model, are its implications. Computer science usually assumes today, that we are representing “information” statically in data structures, on which dynamic functions – algorithms – operate. The implication of the ideas above is, that no such thing as static information exists; “representing it” just captures a snapshot of a continuously running algorithm.

Research proposal in software technology 2:

Represent information as a set of conceptually permanently running algorithms, the state of which can be frozen and stored. ⁋

In a sense, this is the most ambitious proposal contained in this paper. I assume, that it closely connects with neural networks. How to realize the simple task of comparing the similarity of two strings in a context, where the strings are represented by nodes of an active network is not entirely trivial, however.

3.2 The Binary Fallacy

It is a long and in many ways successful tradition, to start the practical training of computer scientists and generally that of software technologists by the introduction of the concepts of bit and byte. But this focus has a tendency to create a distorted look upon what is actually happening in software systems.

If you look at the software fragment

int parameter;

if (parameter) doSomething();

most programmers would spontaneously translate the condition as “if parameter is true or one, execute the function. If it is false or zero, don’t.” Only if pressed, they would give the correct reading “if parameter has any other value but zero – that is: -215+1 till -1, 1 till 215-1 execute the function.

That is, a Boolean interpretation of the code fragment actually underuses what it represents.

Similarly, one should point out, that the “natural” representation of a floating point number – 1 sign bit, 8 bits exponent and 23 bits fraction, is an arbitrary decision by IEEE; historically different decisions by individual hardware manufacturers have been made. And together with the handling of subnormal numbers and handling of infinity / NaN floating point arithmetic, which appears as a completely “natural” operation, is actually a bundle of short algorithms on a lower level of the software / hardware stack.

Therefore, there is no technical reason why computations should be restricted to binary logic, nor is there any compelling reason, why a “number” must be a zero-dimensional point on the continuum.

Both of these preliminary observations should be kept in mind, when we think about the inherent fuzziness of information derived from our metaphor of the historian as observer of signals once exchanged between participants, computing in a context which has been lost.

There are a number of phenomena which this inherent fuzziness encompasses. Without claiming completeness: (1) Fuzziness in a narrower sense, i.e., the impossibility to give a crisp truth value for a statement. (2) An inherent imprecision of a semantic concept, as in “old people”. (3) An item which conceptually is a scalar, but goes beyond our current datatypes. E.g. a price for a commodity, for which we do not have a precise value, but a minimum and a maximum, plus possibly hints at the distribution of the data points between these.

I do not claim this list to be complete; but I am so far working on the suspicion, that all other phenomena of the general fuzziness of sources can be handled by a combination of the techniques needed to solve these three. Specifically the two most obviously missing ones: (a) the problem of missing data or decisions with incomplete information and (b) the problem of decisions based on contradictory information.

Now, for all three of the problems mentioned partial solutions exist. These solutions exist usually so high up the technology stack however, that they are applicable only under very special circumstances.

  • If you solve any of these problems at the level of an application, available to the end users of that application – e.g. a data base “Spurious people of the 13th century”, <http://sp10c.someuniversity.terra> – every other application to be developed, ever, has to reinvent the solution.
  • If you solve any of these problems at the level of an application system – e.g. a specific data base system like Neo4J – it is easily available for all applications realized with the help of that system; every application realized in another application system to be developed ever, has still to reinvent the solution.
  • If you solve any of these problems at the level of a programming language – e.g. C++ or Java – it is easily available for all applications realized with the help of any application system realized with the help of that programming language; though every application realized in another programming language to be developed ever, has still to reinvent the solution.

The solutions for the problems described below, are therefore assumed to be provided at the level of (higher) programming languages – with implicit cross-relations to the solutions for the problems described in section 3.5 below.

Increasing complexity, the solutions required should be presented in inverse order from their introduction.

(3) “An item which conceptually is a scalar, but goes beyond our current datatypes. E.g. a price for a commodity, for which we do not have a precise value, but a minimum and a maximum, plus possibly hints at the distribution of the data points between these.”

For this problem, Julong Deng’s Grey System Theory, or rather the systematic treatment of the building blocks for such systems as described for Western readers by Sifeng Liu and Yi Lin [Liu 2006, 2011], seems to provide almost a blueprint for implementation.

Research proposal in software technology 3:

Implement grey numbers, or a derivation from them, and integrate them seamlessly into main stream programming languages. ⁋

(2) “An inherent imprecision of a semantic concept, as in ‘old people’.”

It is obvious, that this problem closely relates to Zadeh’s [1965, 1975, 1978, 1999] seminal work on Fuzzy Sets and Systems, and the later concept of “Computing with Words” based on linguistic variables, which has found widespread applications in many branches of computation. One is tempted to say, in almost all of them outside of the Humanities – which is a bit baffling, as Zadeh himself has said, that originally he expected them to respond more eagerly to his proposals than any other disciplines [Blair 1994, Termini 2012]. Nevertheless, a systematic approach to make these concepts useful for the handling of historical sources, has to be a bit broader.

The smaller generalization required is, that since 1965 Zadeh’s Fuzzy Sets have been joined by Rough Sets [Pawlak 1982, Pawlak 1985], Evidence or Believe Theory [Shafer 1976], and an almost endless list of modifications of the basic approaches, like Intuitionistic Fuzzy Sets [Atanassov 1986], Hesitant Fuzzy Sets [Torra 2010, Herrera 2014], Rough Fuzzy Sets [Nanda 1992, Jiang 2009] etc. etc. have arisen. Zadeh himself in his later years has tried to combine some of these approaches into a Generalized Theory of Uncertainty [Zadeh 2005] but this is too restricted in scope and leaves aside quite a few of the approaches to handle problems of fuzziness besides Fuzzy Sets (divergent capitalization intentional). As Barr has noted “One of the problems with fuzzy sets is that the meaning of the term has been left vague (one might say fuzzy).” [Barr 2010, 400] Whether category theory is the right approach for such an attempt – and how far that eases implementation – seems to me a not yet completely decided question, even if Barr and Wells have shown that at Least Fuzzy Sets can be shown to be covered by topos theory [Barr 2010, 400-403, within context of 383-411].

A more general problem is, that most of the approaches I have encountered, still assume fuzziness to be the exception, rather than the rule, as it has to be, if we accept the model of an observer of signals exchanged in a lost context. The current logic of embedding an approximately reasoned decision into an information system, is illustrated by graphic 9, an attempt to generalize the similar graphics contained in the literature.

Graphic 9: General logic of “computing with words”

With other words: From an information system, which is crisp, some information is transferred into a fuzzy box, the result of the decision made crisp again for the major parts of the larger embedding system.

Research proposal in software technology 4:

Implement linguistic variables and integrate them seamlessly into main stream programming languages, as permanently accessible data type in all parts of the flow of execution. Base the implementation on a generalized concept of uncertainty, which broadens the scope of Zadeh’s theory of that name. ⁋

(1) “Fuzziness in a narrower sense, i.e., the impossibility to give a crisp truth value for a statement.”

This is in some ways the most puzzling problem for software technology. At first look it seems to be rather simple, as logics with multiple values of truth, preferably continuous truth functions, are well understood and an ample literature exists. The engineering problem appears, however, when the evaluation of an expression in multivalued logic is the base of a control structure.

In the code fragment

if (condition_is_valid) doSomething();

else doSomethingElse();

what happens, if “condition_is_valid” has a truth value of 0.75?

To the best of my (admittedly incomplete) knowledge a very early proposal for the inclusion of fuzzy logic into a programming language – Adamo’s LPL [Adamo 1980] – is the only one, where for such cases a combination of the execution of both branches is contemplated in detail.

Research proposal in software technology 5:

Design genuinely fuzzy control structures and integrate them seamlessly into main stream programming languages. ⁋

Two short remarks:

(a) This is probably a generalization of the preceding problem (embed fuzziness as a principle into the general program structure, rather than as an “island” of approximate reasoning into a crisp program).

(b) There are probably cross-relations to the problem of a “frozen algorithm” mentioned with the comment on the consequences of trying to implement the infological model of Langefors in section 3.1 above (research proposal 2).

3.3 Chomsky’s Dead End

Linguistics and programming languages have shared an intimate relationship for a long time; Backus-Naur form, one of the major break-throughs on the way from tinkering to designing languages, acknowledges a debt owed to Chomsky’s early work. Software technology in the meantime certainly has paid for that debt, as the numerous programs show which analyze the syntax of linguistic expressions.

At least for historians this has led to a focus of linguistic work which is at the least not very productive, if not outright counter-productive. Syntax is certainly a very useful tool, to formulate correct sentences, but it obscures the fact, that many linguistic remainders have at best a doubtful syntax and it may encourage a belief that we understand text”s, which we do not.

Igitur Carolus Magnus a Leone III. Pontifice Romae anno 800 coronatus, …” (“So Charlemagne was crowned in the year 800 in Rome by Pope Leo III, …”; Robert Bellarmine (1542-1621), De Translatione Imperii Romani, liber secundus, caput primum)

Charlemagne, Leo III and cardinal Bellarmine all would have been able to “understand” this sentence, as all of them were quite familiar with Latin grammar. But what would they really have understood? The chief of a network of post-tribal German loyalties and the son of a modest family in Southern Italy, where Byzantium was still looming large, might have had at least similar notions of what the title “Imperator” implied. The idea which the highly educated 16th/17th century cardinal connected with the title would certainly have been incomprehensible for both – and vice versa.

The problem is, that this title is meaningful only within some understanding of the roles within a specific political system. (Which is the reason, why any good history teacher at university level will spend at least about 20 times as long explaining what exactly the term implied in the year 800, than with the story of the actual coronation.)

Whether the fixation on syntax is good for linguistics is a question linguists will have to answer. Why a semantic understanding requires an understanding of the syntax of a message, at least I have never understood. There are linguists which doubt it – for a historian the notion of Roy Harris, that much of modern linguistics is based upon a language myth, that holds to the erroneous belief, that we communicate in precise statements, where we understand all implications of what we say, is eminently attractive. “Do we always know what we mean?” [Harris 1998, 14] Indeed, do we? Do we ponder before we speak, are we aware of all the assumptions we make in formulating a sentence and all the implications our linguistic choices make for the listener? If so, how can we hope ever to understand an utterance where the last native speaker has died a few hundred years ago?

For historical studies technological support for keeping track of semantic changes would be much more important in any case. And, not on top of several conceptual layers which emphasize independence of context and syntactic bias, before “semantic technologies” are plastered on as a second thought at the very top of the non-semantic conceptual building.

Of the five problems I mention, this is probably the one, where it is hardest to plausibly describe in brief a direction for a technical – algorithmic – solution.

Conceptual alternatives exist: The notion of Lakoff and Johnson, that understanding is based upon metaphors [Lakoff 1980] is immediately intuitive for a historian who has tried to look through the meaning hidden behind historical texts. And the notion, that the possibility to construct associations between different concepts, to blend concepts [Fauconnier 2003] is actually a feature distinguishing the human mind in a much more fundamental way, than the I-Language and the Universal Grammar [Isac 2008], is highly convincing for the same historian.

But, as I said, while for most of the other proposals to solve a concrete technical problem in this paper, a sensible starting point and the first few steps of the solution are clear, how to implement metaphors and conceptual blending is much harder to see. What might be very useful as the starting point would be semantic graphs, which allow the handling of seemingly contradictory relationships between nodes. Blending of two concepts means, that an edge connects two nodes, where from the point of view of any of the two concepts, no edge should exist. This would require a class of graphs were both nodes and edges are labelled and there exist bundles of edges, where the acknowledgment of one implies the rejection of one or more of the others. Under the label of “co-edges” I have discussed in other contexts that these would also be useful for the implementation of the type of graphs discussed in the next section.

Research proposal in software technology 6:

Provide tools for the easy handling of such graphs in main stream programming languages. ⁋

3.4 The Markup Fallacy

From a theoretical point of view markup languages are not a very central subject of computer science; for many humanists and historians, however, they seem to be the quintessence of the so called Digital Humanities. In my opinion, they current usage of markup in the handling of historical documents has two methodological weaknesses, however, which are not easily overcome, unless software technology provides support for a new class of concepts.

The first of these weaknesses is rather straightforward: Embedding markup into a text goes directly against the principle mentioned earlier, that a software system handling historical information “Represents the artifacts as free from any interpretation as possible in the technical system …”. On the surface that is violated by some principles, which the TEI has propagated strongly in its early days, when the principle, that markup should signal meaning, not display layout features, resulted in the idea that e.g. italics should result in an <emph> </emph> tag. How an encoder knows, that this was the intention in inserting italics, without parapsychological powers, did always beat me. Though I have to admit, that many historians have opted for normalized instead of diplomatic editions, it should be apparent from the earlier parts of this paper, that this violates my methodological understanding of historical research. From the methodological position formulated, any mixture of representation and interpretation is a sin.

But there is a much more fundamental problem with this sort of markup, when one looks at it from the point of view of processing historical data in information systems. “Markup” according to the current paradigm, applies to text; adding explanatory or analytic comments to an image, a 3D reconstruction or any other non-textual material is considered an “annotation”. (Though annotations have recently also appeared related to text.) I can see no epistemological reason whatsoever, why texts and other forms of source representations are handled differently.

In principle it is quite possible to define standoff annotations which provide a homogeneous solution for one-dimensional (textual), two-dimensional (images) … n-dimensional data. Indeed, in the context of long term preservation, we have proven that this is technically viable in one of my earlier projects [Thaller 2009].

However, while Desmond Schmidt [Schmidt 2009] has proposed a solution for preparing standoff markup for a text, in a way which allows editing of the text independent of the markup, I am not aware of any solution, which would allow this for a data object of higher dimensionality.

Research proposal in software technology 7:

Develop a representation of “information objects”, where a data object of arbitrary dimensionality can be combined with interpretative layers in such a way, that the data object can be changed without damaging these layers. ⁋

There is a small caveat to be added to the above. All of these considerations relate to the situation, where a source is converted by a 1 : 1 operation into a technical representation, be it a human transcription or the scanning operation of an image. Despite the emphasis on leaving a source as undistorted as possible, there is of course the need to handle data objects which represent attempts to create a common representation of more than one such object, e.g. the reconstruction of the commonalities between the witnesses of a an abstract text surviving as different manuscripts. In principle the “leave the source unchanged” principle should apply here as well. For such nonlinear texts the equation “one source is represented as a string, i.e., an array of characters” obviously does not hold. I have myself proposed a model for representing texts not as arrays, but as graphs [Thaller 1993]. Similar situations may become important in the future in data objects of other dimensionality, when the explosive spread of scanning techniques emphasizes more strongly the need to represent relationships between families of images or other objects of higher dimensionality.

Research proposal in software technology 8:

Generalize the solution of research proposal 7 to handle graphs of objects of inhomogeneous dimensionality. ⁋

3.5 The Gorilla Syndrome

Of all the problems I mention here, this is probably the one most confusingly named. Let me start by quoting a text which is completely focused on software technology, historical sources not even remotely concerned. It comes from a book with interviews of people important for the development of programming languages [Seibel 2009].

Interviewer: So you started out saying software reuse is “appallingly bad”, but opening up every black box and fiddling with it all hardly seems like movement towards reusing software.

[ Joe ] Armstrong [inventor of the programming language Erlang]: I think the lack of reusability comes in object-oriented languages, not in functional languages. Because the problem with object-oriented languages is they’ve got all this implicit environment that they carry around with them. You wanted a banana but what you got was a gorilla holding the banana and the entire jungle.” (My emphasis.)

What is described here, at first look seems to be a purely technical problem with modern programming languages. One of the major breakthroughs of software technology has been the invention of object orientation, which assumes that programs should not be defined as operations on numbers, characters or bits, but as interactions between more abstract units, objects, which hide the fact, that ultimately on some lower logical level numbers, characters or bits are manipulated. In the context of historical sources you could envisage an object of the class “currency”, which allows a convenient handling of expressions like “1.4.2” (one pound, four shillings, two pence). Convenient handling implying an expression like “1.4.2 * 25” to allow you to multiply that amount by 25, resulting in “30.4.2”. Of course, in other historical sources “1.4.2” would stand for one gulden, four kreuzer, two heller, and “1.4.2 * 25” should therefore result in “26.46.2”. Solutions for such problems have of course been implemented, including solutions implemented by me. And as this is exactly the kind of simple abstraction which can be hidden perfectly within the class of an object, you would expect, that all you have to do to apply such a solution would be to import an appropriate JAR file into your software stack (or the equivalent in another object oriented programming language, my weapon of choice being C++, but not refered to directly here, as there are presumably more JAVA-aware readers). So you simply include JAR files implementing the classes “currencyPound” and “currencyGulden” into your program.

Unfortunately, however, the implementation of such a class would almost certainly use some other class to handle e.g. label strings “pound”, “shilling”, “pence” and “gulden”, “kreuzer”, “heller” to format such amounts for printing. String classes are very important to handle anything resembling a text; therefore they are usually quite complex and refer to many other classes. Unless the two classes “currencyPound” and “currencyGulden” (the bananas) are implemented in exactly the same software stack, they may likely use different string classes (the gorillas) which in turn use other classes forming a different technology stack (the jungle).

As many historical sources contain meaningful units, which can easily be handled algorithmically, but cannot be mapped directly unto the basic data types used in software technology, such classes allowing their handling would be extremely useful. Some of them being considerably less trivial: A class for geographical locations, e.g., which accepts that some of them are related to specific locations and some of them are purely mythical. (Remember Mount Olympus from section 3.1.)

Research proposal in software technology 9:

Look upon possibilities to extend the object oriented paradigm of programming into a context oriented one. Intuitively speaking by two approaches: (1) Augmenting the “private” and “public” sections of classes by a “context” section, which provides an interface between classes outside of their lines of inheritance. (2) Providing a possibility for “virtual system calls”, which provides interfaces into tools, which can be shared between programs in different languages. Apologies to readers which do not find this paragraph intuitive. ⁋


Russel L. Ackoff: “From Data To Wisdom”, Journal of Applied Systems Analysis 15 (1989) 3-9.

J.M. Adamo: “L.P.L. A fuzzy Programming Language: 1 Syntactic Aspects,” and “L.P.L. A fuzzy Programming Language: 2 Semantic Aspects,” Fuzzy Sets and Systems 3 (1980) 151-179, 261-289.

Robert L. Ashenhurst: “Ontological Aspects of Information Modeling”, Minds and Machines 6 (1996) 287-394.

Krassimir T. Atanassov: “Intuitionistic Fuzzy Sets”, Fuzzy Sets and Systems 20 (1986) 87-96.

Michael Barr and Charles Wells: Category Theory for Computing Science, Montréal, 2010.

Sasa Baskarada and Andy Koronios: “Data, Information, Knowledge, Wisdom (DIKW): a Semiotic Theoretical and Empirical Exploration of the Hierarchy and its Quality Dimension”, Australasian Journal of Information Systems 18 (2013) 5-24.

Betty Blair: “Interview with Lotfi Zadeh”, Azerbaijan International 2 (Winter 1994) 46-47, 50.

Keith Devlin: Logic and Information, Cambridge, 1991.

Keith Devlin: “Modeling Real Reasoning”, in: Giovanni Sommaruga (ed.): Formal Theories of Information, (= Lecture Notes in Computer Science 5363), Berlin-Heidelberg, 2009, 234-252.

Johann Gustav Droysen: Historik. Vorlesungen über Enzyklopädie und Methodologie der Geschichte, ed. by Rudolf Hübner, München, 1937.

Yucong Duan et al.: „Specifying Architecture of Knowledge Graph with Data Graph, Information Graph, Knowledge Graph and Wisdom Graph“, presented at SERA 2017, accessed on April 23rd 2018 at: doi.org/10.1109/SERA.2017.7965747.

Gilles Fauconnier and Mark Turner: The Way We Think. Conceptual Blending and the Mind’s Hidden Complexities, New York, 2003.

Bernard Favre-Bull: Information und Zusammenhang. Informationsfluß in Prozessen der Wahrnehmung, des Denkens und der Kommunikation, Springer, 2001

Luciano Floridi: The Philosophy of Information, Oxford, 2011.

Martin Frické: “The Knowledge Pyramid: A Critique of the DIKW Hierarchy”, Journal of Information Science 35 (2009) 131-142.

Roy Harris: Introduction to Integrational Linguistics, Oxford, 1998.

Francisco Herrera et al. (eds.) “Special Issue on Hesitant Fuzzy Sets”, International Journal of Intelligent Systems 29 (2014) 493-595.

Daniela Isac and Charles Reiss, I-Language, Oxford, 2008.

Yuncheng Jiang et al.: “Reasoning with Expressive Fuzzy Rough Description Logics” Fuzzy Sets and Systems 160 (2009) 3403-3424.

Gu Jifa and Zhang Lingling: “Data, DIKW, Big Data and Data Science”, Procedia Computer Science 31 (2014) 814-821.

William J. Kettinger and Yuan Li: “The infological equation extended: towards conceptual clarity in the relationship between data, information and knowledge”, in: European Journal of Information Systems 19 (2010), p. 409-421.

George Lakoff and Mark Johnson: Metaphors We Live By, Chicago 1980, with a substantial afterword reprinted 2003.

Sifeng Liu and Yi Lin: Grey Information. Theory and Practical Applications, London, 2006.

Sifeng Liu and Yi Lin: Grey Systems. Theory and Practical Applications, London, 2011.

Börje Langefors: Theoretical Analysis of Information Systems, Studentlitteratur, 1973.

W.C. Mahaney et al.: “Biostratigraphic Evidence Relating to the Age-Old Question of Hannibal’s Invasion of Italy”, Archaeometry, 59 (2017), pp. 164-178 and 179-180.

S. Nanda and S. Majumdar: “Fuzzy Rough Sets”, Fuzzy Sets and Systems 45 (1992) 157-160.

Zdzisław Pawlak: “Rough Sets”, International Journal of Parallel Programming, 11 (1982), 341-356.

Zdzisław Pawlak: “Rough Sets and Fuzzy Sets”, Fuzzy Sets and Systems 17 (1985) 99-102.

Jennifer Rowley: “The Wisdom Hierarchy: Representations of the DIKW Hierarchy”, Journal of Information Science 33 (2007) 163-180.

David J. Saab and Uwe V. Riss: “Information as Ontologization”, Journal of the American Society for Information Science and Technology 62 (2011) 2236-2246.

Desmond Schmidt and Robert Colomb: “A Data Structure for Representing Multi-Version Texts Online”, International Journal of Human-Computer Studies 67 (2009) 497-514.

Peter Seibel: Coders at Work, Apress, 2009, 213.

Glenn Shafer: A Mathematical Theory of Evidence, Princeton, 1976.

Claude E. Shannon: “A Mathematical Theory of Communication”, Bell System Technical Journal, 27 (1948) 379–423, 623–656, 1948.

Giovanni Sommaruga: “One or Many Concepts of Information?”, in: Giovanni Sommaruga (ed.): Formal Theories of Information, (= Lecture Notes in Computer Science 5363), Berlin-Heidelberg, 2009, 253-267.

Settimo Termini: “On some ‘Family Resemblances’ of Fuzzy Set Theory and Human Sciences”, in: Rudolf Seising and Veronica Sanz (eds.): Soft Computing in Humanities and Social Sciences (= Studies in Fuzziness and Soft Computing 273), Berlin-Heidelberg, 2012, 39-54.

Manfred Thaller: “Historical Information Science: Is there such a Thing? New Comments on an Old Idea.”, In: Seminario discipline umanistiche e informatica. Il problema dell’ integrazione, ed. Tito Orlandi, 51-86 (= Contributi del Centro Linceo interdisciplinare ‘Beniamino Segre’ 87), Rome, 1993. Reprinted under the same title in: Historical Social Research Supplement 29 (2017), 260-286.

Manfred Thaller: “The Cologne Information Model: Representing Information Persistently”, In: The eXtensible Characterisation Languages – XCL, ed. Manfred Thaller, Hamburg, 2009, 223-39. Reprinted under the same title in: Historical Social Research Supplement 29 (2017), 344-356.

Vincenç Torra: “Hesitant Fuzzy Sets”, International Journal of Intelligent Systems 25 (2010) 529-539.

Warren Weaver: “Introductory Note on the General Setting of the Analytical Communication Studies”, in: Claude E. Shannon and Warren Weaver: The Mathematical Theory of Communication, 1949.

Matthias Weber (ed): Die Reichspolizeiordnungen von 1530, 1548 und 1577, (= Ius Commune, Sonderheft 146), Frankfurt am Main, 2001.

Lotfi A. Zadeh: “Fuzzy Sets”, Information and Control 8 (1965) 338-353.

Lotfi A. Zadeh: “The Concept of a Linguistic Variable and its Application to Approcimate Reasoning”, I – III, Information Sciences 8 (1975) 199-249, 301-357, 9 (1975) 43-80.

Lotfi A. Zadeh: “Fuzzy Sets as a Basis for a Theory of Possibility”, Fuzzy Sets and Systems 1 (1978), 3-28.

Lotfi A. Zadeh and Janusz Kacprzyk (Eds.): Computing with Words in Information / Intelligent Systems I and II (= Studies in Fuzziness and Soft Computing 33 and 34 (1999) ).

Lotfi A. Zadeh: “Toward a Generalized Theory of Uncertainty (GTU) – an outline”, Information Sciences 172 (2005), 1-40.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.