Publishing the Forbidden. All Rights Reserved. © Samizdat 2014-2023.

Tag: a priori languages

Eco: Artificial Intelligence

The Arecibo Message.svg

Frank Drake (1930-), Carl Sagan (1934-96), et al, The Arecibo Message, 1974. The Arecibo Message was broadcast into space via FM radio waves from the Arecibo radio telescope in Puerto Rico on 16 November, 1974. Aimed at the globular star cluster M13, the message comprised 1,679 binary digits. Total broadcast time was less than three minutes in duration. This representation of the message is by Arne Nordmann, licensed under the Creative Commons Attribution-Share Alike 3.0 Unported license. 

Lincos does furnish us with an image of a language that is almost purely “mental” (its level of expression is supported by nothing more than electromagnetic phenomena). This reminds us of other languages which are, in one way or another, the heirs of the ancient search for the perfect language.

Computer languages, like BASIC or Pascal, are, in fact, a priori languages. They are not full languages because their syntax, though rigorous, is simplified and limited, and they remain parasitic on the natural languages which attach meanings to their empty symbols, which, for the most part, serve as logical connectors of the type if . . . then.

None the less, they are universal systems: they are comprehensible to speakers of differing natural languages and are perfect in the sense that they permit neither error nor ambiguity.

They are a priori, in that they are based not on the rules which govern the surface structures of natural languages, but rather, ideally, on a presumed deep grammar common to all natural languages.

They are, finally, philosophical because they presume that this deep grammar, based on the laws of logic, is the grammar of thought of human beings and machines alike. They also exhibit the two limitations inherent in philosophical a priori languages:

(1) their rules of inference are drawn from the western logical tradition, and this may mean, as many have argued, that they reflect little more than the basic grammatical structures common to the Indo-European family of languages;

(2) their effability (sic) is limited; that is, they are capable of expressing only a small proportion of what any natural language can express.

The dream of a perfect language which covers all the meanings and connotations of the vocabulary of a natural language, and in which human beings and machines can engage in “meaningful” conversations (or machines can draw inferences as happens in natural languages), underlies much of contemporary research into artificial intelligence.

Machines are provided, for example, with rules of inference by which they can “judge” whether or not a certain story is coherent, or decide that, if someone is ill, then someone needs medical assistance–and so on.

By now, the literature on this subject is vast, and the proposed systems are many: they run from those that still adhere to the ideal of a componential semantics based on primitives, to those that furnish the machine with schemes of action or a typology of “frames,” “scripts” and “goals.”

In general all of these projects succeed in solving certain problems only through imposing ad hoc solutions, which work only for local portions of the range of action of natural languages.”

Umberto Eco, The Search for the Perfect Language, translated by James Fentress, Blackwell. Oxford, 1995, pp. 310-2.

Eco: Side-effects


Ernst Hähnel, statue of Leibniz, 1883, Leipzig, Germany. Photo © Ad Melkens / Wikimedia Commons.  

“Thus all of the ingenuity expended upon the invention of philosophic a priori languages allowed Leibniz to invent a language of a radically different type, which–though remaining a priori–was no longer a practical, social instrument but rather a tool for logical calculation.

In this sense, Leibniz’s language, and the contemporary language of symbolic logic that descended from it, are scientific languages; yet, like all scientific languages, they are incapable of expressing the entire universe, expressing rather a set of truths of reason.

Such languages do not qualify as a universal language because they fail to express those truths that all natural languages express–truths of fact. Scientific languages do not express empirical events.

In order to express these we would need “to construct a concept which possesses an incalculable number of determinations,” while the completely determined concept of any individual thing or person implies “spatial-temporal determinations which, in their turn, imply other spatial-temporal successions and historical events whose mastery is beyond the human eye, and whose control is beyond the capacity of any man.” (Mugnai 1976: 91).

None the less, by anticipating what was to become the language of computers, Leibniz’s project also contributed to the development of programs well adapted for the cataloguing of the determinations of individual entities, which can tell us that there exists an entity called Mr. X such that this entity has booked a flight from A to B.

We may well fear that by controlling our determinations so well the computer eye has begun to infringe on our privacy, checking on the hour in which we reserved a room in a certain hotel in a certain city. This, then, is one of the side-effects of a project that commenced with the idea of expressing a merely theoretical universe populated with universal ideas such as goodness, angels, entity, substance, accidents, and “all the elephants.”

Dalgarno could never have imagined it. Passing through the mathematical filter of Leibniz, renouncing all semantics, reducing itself to pure syntax, his philosophical a priori language has finally managed to designate even an individual elephant.”

Umberto Eco, The Search for the Perfect Language, translated by James Fentress, Blackwell. Oxford, 1995, pp. 287-8.

%d bloggers like this: