Semantics, artificial intelligence and collective intelligence

Art: M.C. Escher

How does language work? On the receiving end, we hear a sequence of sounds that we translate into a network of concepts. This is what we call « understanding the meaning » of a statement we hear. On the transmit side, we have in mind a network of concepts – a meaning to be conveyed – that we translate into a sequence of sounds. A language interfaces sound sequences with concept networks. Instead of phoneme chains, we can have sequences of ideograms, letters, gestures (in the case of sign languages), etc. What remains invariant among the changes of languages and writing systems is this quasi-automatic interfacing between a sequence of sensible images (sound, visual, tactile), and a graph of abstract concepts (general categories). And let us note that the relations between concepts are themselves concepts. 

The Meaning-Text school of linguistics, initiated by Igor Melchuck, has clearly underlined this duality and correspondence between the text (the sequence of images) and the meaning (the network of concepts). Networks of concepts cannot be transmitted directly. Only language allows the communication and the dynamic coordination between the networks of concepts of the members of a human community, from the scale of the family or the small team to the scale of the largest political or economic units. Language enables storytelling, dialogue and questioning. Language supports not only communication but also thought, that is to say the construction and recognition of complex concept networks, as well as the conceptual organization of memory, complementary to its emotional and sensorimotor organization.

This reciprocal translation function between sequences of images (the signifiers) and networks of abstract categories (the signifieds) suggests a hypothesis about the kind of mathematical structure we need to model language: a category that organizes the correspondence between an algebra and a graph structure. The algebra regulates reading and writing operations on texts, while the graph structure orders operations on nodes and oriented links. To each text corresponds a network of concepts, and vice versa. Operations in the world of texts reflect operations in the world of conceptual graphs. 

We need a regular language to encode strings of signifiers, and we can transform the sequences of symbols of a regular language into syntagmatic trees (syntax being the order of the syntagm) and vice versa. However, if its syntagmatic tree – its internal grammatical structure – is indispensable to the understanding of the meaning of a sentence, it is not sufficient. For each linguistic expression lies at the intersection of a syntagmatic axis and a paradigmatic axis. The syntagmatic tree represents the internal semantic network of a sentence, the paradigmatic axis represents its external semantic network and in particular its relations with sentences having the same structure, but from which it is distinct. To understand the phrase « I choose the vegetarian menu », it is of course necessary to recognize that the verb is « to choose », the subject « I » and the object « the vegetarian menu » and to know moreover that « vegetarian » qualifies « menu ». But one must also recognize the meaning of words and know, for example, that vegetarian is opposed to meaty and to vegan, which implies going beyond the sentence itself to locate its components in the semantic opposition systems of the language. The establishment of semantic relations between concepts thus presupposes that we have not only recognized the syntagmatic trees internal to sentences, but also the belonging of the concepts or their components to paradigmatic matrices external to the sentence, be they specific to the language or to certain practical domains.

Because natural languages are ambiguous and irregular, I have invented a mathematical language (IEML) translatable into natural languages that algebraically encodes not only syntagmatic trees, but also the paradigmatic matrices that give meaning to words or more complex concepts. Each sentence of this language (each concept) is located precisely at the intersection of a syntagmatic tree and paradigmatic matrices. 

Based on the regular syntagmatic-paradigmatic grid of IEML, it becomes possible to generate and recognize semantic relations between concepts in a functional way: knowledge graphs, ontologies, and data models… Still on the AI side, an encoding of labels or data categorization in this algebraic language that is IEML would certainly facilitate machine learning.

Technically, it is a lightweight and decentralized project: an IEML-natural language dictionary, an open-source parser supporting computable functions on language expressions, and a platform for collaborative editing and sharing of concepts and ontologies. On the intellectual level, the formalization of meaning would surely require a serious collective research and training effort.

Beyond the AI and technical aspects, my political vision – in the noble sense of the word – is the following: IEML should be used to foster the semantic interoperability of digital memories as well as a synergy between increasing personal cognitive empowerment and increasing the transparency and reflexivity of collective intelligence, from the scale of the small group to that of humanity.

Publié par Pierre Lévy

Assiociate Prof. at the University of Montreal (Canada), Fellow of the Royal Society of Canada. Author of: Collective Intelligence (1994), Becoming Virtual (1995), Cyberculture (1997), The Semantic Sphere (2011) and several other books translated in numerous languages.

2 commentaires sur « Semantics, artificial intelligence and collective intelligence »

Votre commentaire

Entrez vos coordonnées ci-dessous ou cliquez sur une icône pour vous connecter:


Vous commentez à l’aide de votre compte Déconnexion /  Changer )

Image Twitter

Vous commentez à l’aide de votre compte Twitter. Déconnexion /  Changer )

Photo Facebook

Vous commentez à l’aide de votre compte Facebook. Déconnexion /  Changer )

Connexion à %s

%d blogueurs aiment cette page :