[For an English version of this post, click here.]
Le langage permet une coordination dynamique entre les réseaux de concepts entretenus par les membres d’une communauté, de l’échelle d’une famille ou d’une équipe, jusqu’aux plus grandes unités politiques ou économiques. Il permet également de raconter des histoires, de dialoguer, de poser des questions et de raisonner. Le langage soutient non seulement la communication mais aussi la pensée ainsi que l’organisation conceptuelle de la mémoire, complémentaire de son organisation émotionnelle et sensorimotrice.
Mais comment le langage fonctionne-t-il ? Du côté de la réception, nous entendons une séquence de sons que nous traduisons en un réseau de concepts, conférant ainsi son sens à une proposition. Du côté de l’émetteur, à partir d’un réseau de concepts que nous avons à l’esprit – un sens à transmettre – nous générons une séquence de sons. Le langage fonctionne comme une interface entre des séquences de sons et des réseaux de concepts. Et gardons en tête que les relations entre les concepts sont eux-mêmes des concepts.
Les chaînes de phonèmes (des sons), peuvent être remplacées par des séquences d’idéogrammes, de lettres, ou de gestes comme dans le cas de la langue des signes. L’interfaçage quasi-automatique entre une séquence d’images sensibles (sonores, visuelles, tactiles), et un graphe de concepts abstraits (catégories générales) reste constant parmi toutes les langues et systèmes d’écriture.
Cette traduction réciproque entre une séquence d’images (le signifiant) et des réseaux de concepts (le signifié) suggère qu’une categorie mathématique pourrait modéliser le langage en organisant une correspondance entre une algèbre et une structure de graphe. L’algèbre réglerait les opérations de lecture et d’écriture sur les textes, tandis que la structure de graphe organiserait les opérations sur les nœuds et les liens orientés. A chaque texte correspondrait un réseau de concepts. Les opérations sur les textes reflèteraient dynamiquement les opérations sur les graphes conceptuels.
Nous avons besoin d’un langage régulier pour coder des chaînes de signifiants et nous pouvons transformer les séquences de symboles en arbres syntagmatiques (la syntaxe étant l’ordre du syntagme) et vice versa. Cependant, si son arbre syntagmatique – sa structure grammaticale interne – est indispensable à la compréhension du sens d’une phrase, il n’est pas suffisant. Parce que chaque expression linguistique repose à l’intersection d’un axe syntagmatique et d’un axe paradigmatique. L’arbre syntagmatique représente le réseau sémantique interne d’une phrase, l’axe paradigmatique représente son réseau sémantique externe et en particulier ses relations avec des phrases ayant la même structure, mais dont elle se distingue par quelques mots. Pour comprendre la phrase ” Je choisis le menu végétarien “, il faut bien sûr reconnaître que le verbe est “choisir”, le sujet “je” et l’objet “le menu végétarien” et savoir en outre que “végétarien” qualifie “menu”. Mais il faut aussi reconnaître le sens des mots et savoir, par exemple, que végétarien s’oppose à carné et à végétalien, ce qui implique de sortir de la phrase pour situer ses composantes dans les systèmes d’oppositions sémantiques de la langue. L’établissement de relations sémantiques entre concepts suppose que l’on reconnaisse les arbres syntagmatiques internes aux phrases, mais aussi les matrices paradigmatiques externes à la phrase qui organisent les concepts, que ces matrices soient propres à une langue ou à certains domaines pratiques.
Parce que les langues naturelles sont ambiguës et irrégulières, j’ai conçu une langue mathématique (IEML) traduisible en langues naturelles, une langue calculable qui peut coder algébriquement non seulement les arbres syntagmatiques, mais aussi les matrices paradigmatiques où les mots et les concepts prennent leur sens. Chaque phrase du métalangage IEML est située précisément à l’intersection d’un arbre syntagmatique et de matrices paradigmatiques.
Sur la base de la grille syntagmatique-paradigmatique régulière d’IEML, il devient possible de générer et de reconnaître des relations sémantiques entre concepts de manière fonctionnelle : graphes de connaissance, ontologies, modèles de données… Toujours du côté de l’IA, un codage des étiquettes ou de la catégorisation des données dans cette langue algébrique qu’est IEML faciliterait l’apprentissage machine. Au-delà de l’IA, ma vision pour IEML est de favoriser l’interopérabilité sémantique des mémoires numériques et de développer une synergie entre l’autonomisation cognitive personnelle et la réflexivité de l’intelligence collective.
Sur le plan technique, il s’agit d’un projet léger et décentralisé: un dictionnaire IEML-langues naturelles, un analyseur syntaxique (parseur) open-source supportant les fonctions calculables sur les expressions de la langue et une plate-forme d’édition collaborative et de partage des concepts et ontologies. Le développement, la maintenance et l’utilisation d’un protocole sémantique basé sur l’IEML nécessitera un effort de recherche et de formation à long terme.
Le but de ce texte est de présenter une vue générale des limites de l’IA contemporaine et de proposer une voie pour les dépasser. L’IA a accompli des progrès considérables depuis l’époque des Claude Shannon, Alan Turing et John von Neumann. Néanmoins, de nombreux obstacles se dressent encore sur la route indiquée par ces pionniers. Aujourd’hui, l’IA symbolique se spécialise dans la modélisation conceptuelle et le raisonnement automatique tandis que l’IA neuronale excelle dans la catégorisation automatique. Mais les difficultés rencontrées aussi bien par les approches symboliques que neuronales sont nombreuses. Une combinaison des deux branches de l’IA, bien que souhaitable, laisse encore non résolus les problèmes du cloisonnement des modèles et les difficultés d’accumulation et d’échange des connaissances. Or l’intelligence humaine naturelle résout ces problèmes par l’usage du langage. C’est pourquoi je propose que l’IA adopte un modèle calculable et univoque du langage humain, le Métalangage de l’Économie de l’Information (IEML pour Information Economy MetaLanguage), un code sémantique de mon invention. IEML a la puissance d’expression d’une langue naturelle, il possède la syntaxe d’un langage régulier, et sa sémantique est univoque et calculable parce qu’elle est une fonction de sa syntaxe. Une architecture neuro-sémantique basée sur IEML allierait les forces de l’IA neuronale et de l’IA symbolique classique tout en permettant l’intégration des connaissances grâce à un calcul interopérable de la sémantique. De nouvelles avenues s’ouvrent à l’intelligence artificielle, qui entre en synergie avec la démocratisation du contrôle des données et l’augmentation de l’intelligence collective.
Après avoir été posté sur ce blog, le texte a été publié par le Giornale Di Filosofia numéro 2. Lien vers –> le texte complet en PDF publié sur le site du Giornale di Filosofia. Ou bien lisez le texte ci-dessous 🙂.
Today, artificial intelligence is divided between two major trends: symbolic and statistical. The symbolic branch corresponds to what has been successively called in the last 70 years semantic networks, expert systems, semantic web and more recently, knowledge graphs. Symbolic AI codes human knowledge in the form of networks of relationships between concepts ruled by models and ontologies which give leverage to automatic reasoning. The statistical branch of AI trains algorithms to recognize visual, linguistic or other forms from large masses of data, relying on neural models roughly imitating the learning mode of the brain. Neuro-mimetic artificial intelligence has existed since the beginnings of computer science (see the work of McCulloch and von Foerster) but has only become useful because of the increase in computing power available since 2010. In the early 2020s, these two currents are merging according to a hybrid or neuro-symbolic model which seems very promising. Though many problems still remain, in terms of the consistency and interoperability of metadata.
Big tech companies and a growing number of scientific, economic and social sectors use knowledge graphs. Despite the availability of the WWW Consortium metadata standards for marking classifications and ontologies (RDF, OWL) the different sectors (see the slide below) do not communicate with each other and – even worse – divergent systems of categories and relationships are most often in use within the same domain. The interoperability of metadata standards – such as RDF – only addresses the compatibility of digital files. It should not be confused with true semantic interoperability, which addresses concept architectures and models. In reality, the problem of semantic interoperability has yet to be solved in 2021, and there are many causes for the opacity that plagues digital memory. Natural languages are multiple, informal, ambiguous and changing. Cultures and disciplines tend to divide reality in different ways. Finally, often inherited from the age of print, the numerous metadata systems in place to classify data are incompatible like thesauri, documentary languages, ontologies, taxonomies, folksonomies, sets of tags or hashtags, keywords, etc.
The Conundrum of Semantic Interoperability
There is currently no way to code linguistic meaning in a uniform and computable way, the way we code images using pixels or vectors for instance. To represent meaning, we are still using natural languages which are notoriously multiple, changing and ambiguous. With the notable exception of number notation and mathematical codes, our writing systems are primarily designed to represent sounds. Their representation of categories or concepts is indirect (characters → sound → concepts) and difficult for computers to grasp. Computers can handle syntax (the regular arrangement of characters), but their handling of semantics remains imperfect and laborious. Despite the success of machine translation (Deep L, Google translate) and automatic text generation (GPT3), computers don’t really understand the meaning of the texts they read or write.
Now, how can we resolve the problem of semantic interoperability and progress towards a thorough automatic processing of meaning? Many advances in computer science come from the invention of a relevant coding system making the coded object (number, image, sound, etc.) easily computable. The goal of our company INTLEKT Metadata Inc. has been to make concepts, categories or linguistic meaning systematically computable. In order to solve this problem, we have designed the Information Economy MetaLanguage: IEML. This metalanguage has a compact dictionary of less than 5000 words. IEML words are organized by subject-oriented paradigms and visualized as keyboards. The grammar of this metalanguage is completely regular and embedded in the IEML editor. Thank to this grammar, complex concepts and relations can be recursively constructed by combining simpler ones. It is not a super-ontology (like Cyc) but a programmable language (akin to a computable Esperanto) able to translate any ontology and to connect any possible categories. By using such a semantic code, artificial intelligence could take a giant step forward feeding collective intelligence. Public health data from all countries would not only be able to communicate with each other, but could also harmonize with economic and social data. Occupational classifications and different international labour market statistics would automatically translate into each other. The AI of smart contracts, international e-commerce and the Internet of Things would exchange data and execute instructions based on automatic reasoning. Government statistics, national libraries, major museums and digital humanities research would feed into each other. On the machine learning side, we would reach a system of uniform and precise labels and annotations that would help AI to become more ethical, transparent, and efficient. A common semantic code would make it finally possible to achieve a de-fragmentation of the global memory and an integration of symbolic and statistical AI. The only price to pay for reaching neuro-symbolic collective intelligence would be a concerted effort for training specialists to translate metadata into IEML.
Once you are on the site, on the top right you can choose between french and english
“USL” (Uniform Semantic Locator) allows the search for words and paradigms in the dictionary
“Tags” gives you some examples of USLs groups by domain
If you are in “USL” the search for IEML expressions (instead of natural language translations) is done by typing * at the beginning of the query
Type: choose “all”
Class: filters nouns verbs or auxiliaries
Cardinality: choose “root” paradigms (big tables, or multi-tables paradigms), or the (small) tables, or singular = individual words. It is recommended to explore the dictionary by “roots”
When you click on a search result, the corresponding paradigm appears on the right.
The right panel present certain relations according to the selected words.
More than 60% of the human population is connected to the Internet, most sectors of activity have switched to digital and software drives innovation. Yet Internet standards and protocols were invented at a time when less than one percent of the population was connected. It is time to use the data flows, the available computing power and the possibilities of interactive communication for human development… and to solve the serious problems we are facing. That is why I will launch soon a major international project – comparable to the construction of a cyclotron or a voyage to Mars – aiming at an augmentation of the Internet in the service of collective intelligence.
This project has several interrelated objectives:
Decompartmentalize digital memory and ensure its semantic (linguistic, cultural and disciplinary) interoperability.
Open up indexing modes and maximize the diversity of interpretations of the digital memory.
Make communication between machines, but also between humans and machines, more fluid in order to enforce our collective mastery of the Internet of Things, intelligent cities, robots, autonomous vehicles, etc.
Establish new forms of modeling and reflexive observation of human collective intelligence on the basis of our common memory.
IEML
The technical foundation of this project is IEML (Information Economy MetaLanguage), a semantic metadata system that I invented with support from the Canadian federal government. IEML has :
the expressive power of a natural language,
the syntax of a regular language,
calculable semantics aligned with its syntax.
IEML is exported in RDF and is based on Web standards. IEML concepts are called USLs (Uniform Semantic Locators). They can be read and translated into any natural language. Semantic ontologies – sets of USLs linked by a network of relationships – are interoperable by design. IEML establishes a virtual knowledge base that feeds both automatic reasoning and statistical calculations. In short, IEML fulfills the promise of the Semantic Web through its computable meaning and interoperable ontologies.
For a short description of the IEML grammar, click here.
Intlekt
The URLs system and the http standard only become useful through a browser. Similarly, the new IEML-based semantic addressing system for the Internet requires a special application, called Intlekt, whose technical project manager is Louis van Beurden. Intlekt is a collaborative and distributed platform that supports concept editing, data curation and new forms of search, data mining and data visualization.
Intlekt empowers the edition and publishing of semantic ontologies – sets of linked concepts – related to a field of practice or knowledge. These ontologies can be original or translate existing semantic metadata such as: thesauri, documentary languages, ontologies, SKOS taxonomies, folksonomies, sets of tags or hashtags, keywords, column and row headings, etc. Published semantic ontologies augment a dictionary of concepts, which can be considered as an open meta-ontology.
Intlekt is also a data curation tool. It enables editing, indexing in IEML and publishing data collections that feed a common knowledge base. Eventually, statistical algorithms will be used to automate the semantic indexing of data.
Finally, Intlekt exploits the properties of IEML to allow new forms of search, automatic reasoning and simulation of complex systems.
Special applications can be imagined in many areas, like:
the preservation of cultural heritage,
research in the humanities (digital humanities),
education and training
public health,
informed democratic deliberation,
commercial transactions,
smart contracts,
the Internet of things,
and so on…
And now, what?
Where do we stand on this project in the summer of 2020? After many tests over several years, IEML’s grammar has stabilized, as well as the base of morphemes of about 5000 units which enables any concept to be built at will. I tested positively the expressive possibilities of the language in several fields of humanities and earth sciences. Nevertheless, at the time of writing, the latest state of the grammar is not yet implemented. Moreover, to obtain a version of Intlekt that enables the semantic ontology editing, data curation and data mining functions described above, a team of several programmers working for one year is needed. In the coming months, the friends of IEML will be busy pursuing this critical mass.
IEML (the Information Economy Meta Language) has four main directions of research and development in 2019: in mathematics, data science, linguistics and software development. This blog entry reviews them successively.
1- A mathematical research program
I will give here a philosophical description of the structure of IEML, the purpose of the mathematical research to come being to give a formal description and to draw from this formalisation as much useful information as possible on the calculation of relationships, distances, proximities, similarities, analogies, classes and others… as well as on the complexity of these calculations. I had already produced a formalization document in 2015 with the help of Andrew Roczniak, PhD, but this document is now (2019) overtaken by the evolution of the IEML language. The Brazilian physicist Wilson Simeoni Junior has volunteered to lead this research sub-program.
IEML Topos
The “topos” is a structure that was identified by the great mathematician Alexander Grothendieck, who “is considered as the re-founder of algebraic geometry and, as such, as one of the greatest mathematicians of the 20th century” (see Wikipedia).
Without going into technical details, a topos is a bi-directional relationship between, on the one hand, an algebraic structure, usually a “category” (intuitively a group of transformations of transformation groups) and, on the other hand, a spatial structure, which is geometric or topological.
In IEML, thanks to a normalization of the notation, each expression of the language corresponds to an algebraic variable and only one. Symmetrically, each algebraic variable corresponds to one linguistic expression and only one.
Topologically, each variable in IEML algebra (i.e. each expression of the language) corresponds to a “point”. But these points are arranged in different nested recursive complexity scales: primitive variables, morphemes of different layers, characters, words, sentences, super-phrases and texts. However, from the level of the morpheme, the internal structure of each point – which comes from the function(s) that generated the point – automatically determines all the semantic relationships that this point has with the other points, and these relationships are modelled as connections. There are obviously a large number of connection types, some very general (is contained in, has an intersection with, has an analogy with…) others more precise (is an instrument of, contradicts X, is logically compatible with, etc.).
The topos that match all the expressions of the IEML language with all the semantic relationships between its expressions is called “The Semantic Sphere”.
Algebraic structure of IEML
In the case of IEML, the algebraic structure is reduced to
1. Six primitive variables
2. A non-commutative multiplication with three variables (substance, attribute and mode). The IEML multiplication is isomorphic to the triplet ” departure vertex, arrival vertex, edge ” which is used to describe the graphs.
3. A commutative addition that creates a set of objects.
This algebraic structure is used to construct the following functions and levels of variables…
1. Functions using primitive variables, called “morpheme paradigms”, have as inputs morphemes at layer n and as outputs morphemes at layer n+1. Morpheme paradigms include additions, multiplications, constants and variables and are visually presented in the form of tables in which rows and columns correspond to certain constants.
2. “Character paradigms” are complex additive functions that take morphemes as inputs and characters as outputs. Character paradigms include a group of constant morphemes and several groups of variables. A character is composed of 1 to 5 morphemes arranged in IEML alphabetical order. (Characters may not include more than five morphemes for cognitive management reasons).
3. IEML characters are assembled into words (a substance character, an attribute character, a mode character) by means of a multiplicative function called a “word paradigm”. A word paradigm intersects a series of characters in substance and a series of characters in attribute. The modes are chosen from predefined auxiliary character paradigms, depending on whether the word is a noun, a verb or an auxiliary. Words express subjects, keywords or hashtags. A word can be composed of only one character.
4. Sentence building functions assemble words by means of multiplication and addition, with the necessary constraints to obtain grammatical trees. Mode words describe the grammatical/semantic relationships between substance words (roots) and attribute words (leaves). Sentences express facts, proposals or events; they can take on different pragmatic and logical values.
5. Super-sentences are generated by means of multiplication and addition of sentences, with constraints to obtain grammatical trees. Mode sentences express relationships between substance sentences and attribute sentences. Super-sentences express hypotheses, theories or narratives.
6. A USL (Uniform Semantic Locator) or IEML text is an addition (a set) of words, sentences and super-sentences.
Topological structure of IEML: a semantic rhizome
Static
The philosophical notion of rhizome (a term borrowed from botany) was developed on a philosophical level by Deleuze and Guattari in the preface to Mille Plateaux (Minuit 1980). In this Deleuzo-Guattarian lineage, by rhizome I mean here a complex graph whose points or “vertices” are organized into several levels of complexity (see the algebraic structure) and whose connections intersect several regular structures such as series, tree, matrix and clique. In particular, it should be noted that some structures of the IEML rhizome combine hierarchical or genealogical relationships (in trees) with transversal or horizontal relationships between “leaves” at the same level, which therefore do not respect the “hierarchical ladder”.
Dynamic
We can distinguish the abstract, or virtual, rhizomatic grid drawn by the grammar of the language (the sphere to be dug) and the actualisation of points and relationships by the users of the language (the dug sphere of chambers and galleries). Characters, words, sentences, etc. are all chambers in the centre of a star of paths, and the generating functions establish galleries of “rhizomatic” relationships between them, as many paths for exploring the chambers and their contents. It is therefore the users, by creating their lexicons and using them to index their data, communicate and present themselves, who shape and grow the rhizome…
Depending on whether circuits are more or less used, on the quantity of data or on the strength of interactions, the rhizome undergoes – in addition to its topological transformations – various types of quantitative or metric transformations.
* The point to remember is that IEML is a language with calculable semantics because it is also an algebra (in the broad sense) and a complex topological space.
* In the long term, IEML will be able to serve as a semantic coordinate system for the information world at large.
2 A research program in data science
The person in charge of the data science research sub-program is the software engineer (Eng. ENSIMAG, France) Louis van Beurden, who holds also a master’s degree in data science and machine translation from the University of Montréal, Canada. Louis is planning to complete a PhD in computer science in order to test the hypothesis that, from a data science perspective, a semantic metadata system in IEML is more efficient than a semantic metadata system in natural language and phonetic writing. This doctoral research will make it possible to implement phases A and B of the program below and to carry out our first experiment.
Background information
The basic cycle in data science can be schematized according to the following loop:
1. selection of raw data,
2. pre-processing, i.e. cleaning data and metadata imposition (cataloguing and categorization) to facilitate the exploitation of the results by human users,
3. statistical processing,
4. visual and interactive presentation of results,
5. exploitation of the results by human users (interpretation, storytelling) and feedback on steps 1, 2, 3
Biases or poor quality of results may have several causes, but often come from poor pre-treatment. According to the old computer adage “garbage in, garbage out“, it is the professional responsibility of the data-scientists to ensure the quality of the input data and therefore not to neglect the pre-processing phase where this data is organized using metadata.
Two types of metadata can be distinguished: 1) semantic metadata, which describes the content of documents or datasets, and 2) ordinary metadata, which describes authors, creation dates, file types, etc. Let us call “semantic pre-processing” the imposition of semantic metadata on data.
Hypothesis
Since IEML is a univocal language and the semantic relationships between morphemes, words, sentences, etc. are mathematically computable, we assume that a semantic metadata system in IEML is more efficient than a semantic metadata system in natural language and phonetic writing. Of course, the efficiency in question is related to a particular task: search, data analysis, knowledge extraction from data, machine learning, etc.
In other words, compared to a “tokenization” of semantic metadata in phonetic writing noting a natural language, a “tokenization” of semantic metadata in IEML would ensure better processing, better presentation of results to the user and better exploitation of results. In addition, semantic metadata in IEML would allow datasets that use different languages, classification systems or ontologies to be de-compartmentalized, merged and compared.
Design of the first experience
The ideal way to do an experiment is to consider a multi-variable system and transform only one of the system variables, all other things being equal. In our case, it is only the semantic metadata system that must vary. This will make it easy to compare the system’s performance with one (phonetic tokens) or the other (semantic tokens) of the semantic metadata systems.
– The dataset of our first experience encompasses all the articles of the Sens Public scientific journal.
– Our ordinary metadata are the author, publication date, etc.
– Our semantic metadata describe the content of articles.
– In phonetic tokens, using RAMEAU categories, keywords and summaries,
– In IEML tokens by translating phonetic tokens.
– Our processes are “big data” algorithms traditionally used in natural language processing
– An algorithm for calculating the co-occurrences of keywords.
– A TF-IDF (Term Frequency / Inverse Document Frequency) algorithm that works from a word / document matrix.
– A clustering algorithm based on “word embeddings” of keywords in articles (documents are represented by vectors, in a space with as many dimensions as words).
– A user interface will offer a certain way to access the database. This interface will be obviously adapted to the user’s task (which remains to be chosen, but could be of the “data analytics” type).
– Result 1 corresponds to the execution of the “machine task”, i.e. the establishment of a connection network on the articles (relationships, proximities, groupings, etc.). We’ll have to compare….
– result 1.1 based on the use of phonetic tokens with
– result 1.2 based on the use of IEML tokens.
– Result 2 corresponds to the execution of the selected user-task (data analytics, navigation, search, etc.). We’ll have to compare….
– result 2.1, based on the use of phonetic tokens, with
– result 2.2, based on the use of IEML tokens.
Step A: First indexing of a database in IEML
Reminder: the data are the articles of the scientific journal, the semantic metadata are the categories, keywords and summaries of the articles. From the categories, keywords and article summaries, a glossary of the knowledge area covered by the journal is created, or a sub-domain if it turns out that the task is too difficult. It should be noted that in 2019 we do not yet have the software tools to create IEML sentences and super-phrases that allow us to express facts, proposals, theories, narratives, hypotheses, etc. Phrases and super-phrases, perhaps accessible in a year or two, will therefore have to wait for a later phase of the research.
The creation of the glossary will be the work of a project community, linked to the editors of Sens-Public magazine and the Canada Research Chair in Digital Writing (led by Prof. Marcello Vitali-Rosati) at the Université de Montréal (Digital Humanities). Pierre Lévy will accompany this community and help it to identify the constants and variables of its lexicon. One of the auxiliary goals of the research is to verify whether motivated communities can appropriate IEML to categorize their data. Once we are satisfied with the IEML indexing of the article database, we will proceed to the next step.
Step B: First experimental test
1. The test is determined to measure the difference between results based on phonetic tokens and results based on IEML tokens.
2. All data processing operations are carried out on the data.
3. The results (machine tasks and user tasks) are compared with both types of tokens.
The experiment can eventually be repeated iteratively with minor modifications until satisfactory results are achieved.
If the hypothesis is confirmed, we proceed to the next step
Step C: Towards an automation of semantic pre-processing in IEML.
If the superior efficiency of IEML tokens for semantic metadata is demonstrated, then there will be a strong interest in maximizing the automation of IEML semantic pre-processing.
The algorithms used in our experiment are themselves powerful tools for data pre-processing, they can be used, according to methods to be developed, to partially automate semantic indexing in IEML. The “word embeddings” will make it possible to study how IEML words are correlated with the natural language lexical statistics of the articles and to detect anomalies. For example, we will check if similar USLs (a USL is an IEML text) point to very different texts or if very different texts have similar USLs.
Finally, methods will be developed to use deep learning algorithms to automatically index datasets in IEML.
Step D: Research and development perspective in Semantic Machine Learning
If step C provides the expected results, i.e. methods using AI to automate the indexing of data in IEML, then big data indexed in IEML will be available. As progress will be made, semantic metadata may become increasingly similar to textual data (summary of sections, paragraphs, sentences, etc.) until translation into IEML is achieved, which remains a distant objective.
The data indexed in IEML could then be used to train artificial intelligence algorithms. The hypothesis that machines learn more easily when data is categorized in IEML could easily be validated by experiments of the same type as described above, by comparing the results obtained from training data indexed in IEML and the results obtained from the same data indexed in natural languages.
This last step paves the way for a better integration of statistical AI and symbolic AI (based on facts and rules, which can be expressed in IEML).
3 A research program in linguistics, humanities and social sciences
Introduction
The semiotic and linguistic development program has two interdependent components:
1. The development of the IEML metalanguage
2. The development of translation systems and bridges between IEML and other sign systems, in particular…
– natural languages,
– logical formalisms,
– pragmatic “language games” and games in general,
– iconic languages,
– artistic languages, etc.
This research and development agenda, particularly in its linguistic dimension, is important for the digital humanities. Indeed, IEML can serve as a system of semantic coordinates of the cultural universe, thus allowing the humanities to cross a threshold of scientific maturity that would bring their epistemological status closer to that of the natural sciences. Using IEML to index data and to formulate assumptions would result in….
(1) a de-silo of databases used by researchers in the social sciences and humanities, which would allow for the sharing and comparison of categorization systems and interpretive assumptions;
(2) an improved analysis of data.
(3) The ultimate perspective, set out in the article “The Role of the Digital Humanities in the New Political Space” (http://sens-public.org/article1369.html in French), is to aim for a reflective collective intelligence of the social sciences and humanities research community.
But IEML’s research program in the perspective of the digital humanities – as well as its research program in data science – requires a living and dynamic semiotic and linguistic development program, some aspects of which I will outline here.
IEML and the Meaning-Text Theory
IEML’s linguistic research program is very much based on the Meaning-Text theory developed by Igor Melchuk and his school. “The main principle of this theory is to develop formal and descriptive representations of natural languages that can serve as a reliable and convenient basis for the construction of Meaning-Text models, descriptions that can be adapted to all languages, and therefore universal. ”(Excerpt translated from the Wikipedia article on Igor Melchuk). Dictionaries developed by linguists in this field connect words according to universal “lexical functions” identified through the analysis of many languages. These lexical functions have been formally transposed into the very structure of IEML (See the IEML Glossary Creation Guide) so that the IEML dictionary can be organized by the same tools (e.g. Spiderlex) as those of the Meaning-Text Theory research network. Conversely, IEML could be used as a pivot language – or concept description language – *between* the natural language dictionaries developed by the network of researchers skilled in Meaning-Text theory.
Construction of specialized lexicons in the humanities and social sciences
A significant part of the IEML lexicon will be produced by communities having decided to use IEML to mark out their particular areas of knowledge, competence or interaction. Our research in specialized lexicon construction aims to develop the best methods to help expert communities produce IEML lexicons. One of the approaches consists in identifying the “conceptual skeleton” of a domain, namely its main constants in terms of character paradigms and word paradigms.
The first experimentation of this type of collaborative construction of specialized lexicons by experts will be conducted by Pierre Lévy in collaboration with the editorial team of the Sens Public scientific journal and the Canada Research Chair in Digital Textualities at the University of Montréal (led by Prof. Marcello Vitali-Rosati). Based on a determination of their economic and social importance, other specialized glossaries can be constructed, for example on the theme of professional skills, e-learning resources, public health prevention, etc.
Ultimately, the “digital humanities” branch of IEML will need to collaboratively develop a conceptual lexicon of the humanities to be used for the indexation of books and articles, but also chapters, sections and comments in documents. The same glossary should also facilitate data navigation and analysis. There is a whole program of development in digital library science here. I would particularly like to focus on the human sciences because the natural sciences have already developed a formal vocabulary that is already consensual.
Construction of logical, pragmatic and narrative character-tools
When we’ll have a sentence and super-phrase editor, it is planned to establish a correspondence between IEML – on the one hand – and propositional calculus and first order logics – on the other hand –. This will be done by specifying special character-tools to implement logical functions. Particular attention will be paid to formalizing the definition of rules and the declaration that “facts” are true in IEML. It should be noted in passing that, in IEML, grammatical expressions represent classes, sets or categories, but that logical individuals (proper names, numbers, etc.) or instances of classes are represented by “literals” expressed in ordinary characters (phonetic alphabets, Chinese characters, Arabic numbers, URLs, etc.).
In anticipation of practical use in communication, games, commerce, law (smart contracts), chatbots, robots, the Internet of Things, etc., we will develop a range of character-tools with illocutionary force such as “I offer”, “I buy”, “I quote”, “I give an instruction”, etc.
Finally, we will making it easier for authors of super-sentences by developing a range of character-tools implementing “narrative functions”.
4 A software development program
A software environment for the development and public use of the IEML language
Logically, the first multi-user IEML application will be dedicated to the development of the language itself. This application is composed of the following three web modules.
1. A morpheme editor that also allows you to navigate in the morphemes database, or “dictionary”.
3. A navigation and reading tool in the IEML library as a whole, or “IEML database” that brings together the dictionary and lexicon, with translations, synonyms and comments in French and English for the moment.
The IEML database is a “Git” database and is currently hosted by GitHub. Indeed, a Git database makes it possible to record successive versions of the language, as well as to monitor and model its growth. It also allows large-scale collaboration among teams capable of developing specific branches of the lexicon independently and then integrating them into the main branch after discussion, as is done in the collaborative development of large software projects. As soon as a sub-lexicon is integrated into the main branch of the Git database, it becomes a “common” usable by everyone (according to the latest General Public License version.
Morpheme and word editors are actually “Git clients” that feed the IEML database. A first version of this collaborative read-write environment should be available in the fall of 2019 and then tested by real users: the editors of the Scientific Journal “Sens Public” as well as other participants in the University of Montréal’s IEML seminar.
The following versions of the IEML read/write environment should allow the editing of sentences and texts as well as literals that are logical individuals not translated into IEML, such as proper names, numbers, URLs, etc.
A social medium for collaborative knowledge management
A large number of applications using IEML can be considered, both commercial and non-commercial. Among all these applications, one of them seems to be particularly aligned with the public interest: a social medium dedicated to collaborative knowledge and skills management. This new “place of knowledge” could allow the online convergence of the missions of…
– museums and libraries,
– schools and universities,
– companies and administrations (with regard to their knowledge creation and management dimension),
– smart cities, employment agencies, civil society networks, NGO, associations, etc.
According to its general philosophy, such a social medium should…
– be supported by an intrinsically distributed platform,
– have the simplicity – or the economy of means – of Twitter,
– ensure the sovereignty of users over their data,
– promote collaborative processes.
The main functions performed by this social medium would be:
– data curation (reference and categorization of web pages, edition of resource collections),
– teaching offers and learning demands,
– offers and demands for skills, or employment market.
IEML would serve as a common language for
– data categorization,
– description of the knowledge and skills,
– the expression of acts within the social medium (supply, demand, consent, publish, etc.)
– addressing users through their knowledge and skills.
Three levels of meaning would thus be formalized in this medium.
– (1) The linguistic level in IEML – including lexical and narrative functions – formalizes what is spoken about (lexicon) and what is said (sentences and super-phrases).
– (2) The logical – or referential – level adds to the linguistic level…
– logical functions (first order logic and propositional logic) expressed in IEML using logical character-tools,
– the ability of pointing to references (literals, document URLs, datasets, etc.),
– the means to express facts and rules in IEML and thus to feed inference engines.
– (3) The pragmatic level adds illocutionary functions and users to the linguistic and logical levels.
– Illocutionary functions (thanks to pragmatic character-tools) allow the expression of conventional acts and rules (such as “game” rules).
– The pragmatic level obviously requires the consideration of players or users, as well as user groups.
– It should be noted that there is no formal difference between logical inference and pragmatic inference but only a difference in use, one aiming at the truth of propositions according to referred states of things, the other calculating the rights, obligations, gains, etc. of users according to their actions and the rules of the games they play.
The semantic profiles of users and datasets will be arranged according to the three levels that have just been explained. The “place of knowledge” could be enhanced by the use of tokens or crypto-currencies to reward participation in collective intelligence. If successful, this type of medium could be generalized to other areas such as health, democratic governance, trade, etc.
Je voudrais répondre dans cette petite entrée de blog à quelques questions qui m’ont été posées par des amis Turcs (du site Çeviri Konusmalar) au sujet de l’intelligence artificielle et de l’autonomie des machines. Voir ici sur Twitter…
Un des rôles de la philosophie est de catégoriser l’expérience humaine de façon à réduire le plus possible l’illusion, ou si l’on préfère à trouver les concepts qui vont nous permettre de comprendre notre situation et de mieux guider notre action. Cela amène souvent les philosophes à contredire l’opinion courante. Aujourd’hui cette opinion est propagée par le journalisme et la fiction. Aussi bien les journalistes que les auteurs de roman ou de série TV présentent les robots ou l’intelligence artificielle comme capable d’autonomie et de conscience, que ce soit dès maintenant ou dans un futur proche. Cette représentation est à mon avis fausse, mais elle fonctionne très bien parce qu’elle joue…
ou bien sur la peur d’être éliminé ou asservi par des machines (sensationnalisme ou récit dystopique),
ou bien sur l’espoir que l’intelligence artificielle va nous aider magiquement à résoudre tous nos problèmes ou – pire – qu’elle représenterait une espèce plus avancée que l’homme (dans le cas de certaines publicités ou d’utopies naïves).
Dans les deux cas, espoir ou peur, le ressort principal est la passion, l’émotion, et non pas une compréhension exacte de ce que c’est que le traitement automatique de l’information et du rôle qu’il joue dans l’intelligence humaine.
Afin de recadrer cette question de l’autonomie des machines, je voudrais répondre ici le plus simplement possible à trois questions:
Qu’est-ce que l’intelligence humaine?
Qu’est-ce que l’informatique, ou les machines à traiter l’information?
Est-ce que les machines peuvent devenir autonomes?
Qu’est-ce que c’est que l’Intelligence humaine?
D’abord il faut reconnaître que les humains sont des animaux et que les animaux ont déjà des capacité de mémoire, de représentation interne des situations, de résolution de problèmes, d’apprentissage, etc. Les animaux sont des êtres sensibles, qui ressentent attraction et répulsion, plaisir et douleur, voire empathie. Les plus plus intelligents d’entre eux ont la capacité de transmettre certaines connaissances acquises dans l’expérience à leur progéniture, d’utiliser des outils, etc. Ensuite, l’intelligence animale se manifeste de manière particulièrement frappante sur un plan collectif ou social et, pour ce qui nous intéresse, notamment chez les primates (les grands singes), dont nous faisons partie. Les primates ont des structures sociales avec des rôles sociaux fort différenciés et des stratégies collectives élaborées pour se défendre, se nourrir, contrôler leur territoire, etc. Nous partageons bien sûr toute cette intelligence animale. Mais nous avons en plus la manipulation symbolique.
Ce qui différencie l’intelligence humaine de l’intelligence animale c’est d’abord et avant tout l’usage du langage et des systèmes symboliques. Un système symbolique c’est un moyen de communication et de pensée dont les éléments – les symboles – ont deux aspects: un aspect sensible (visible, audible) et un aspect invisible, abstrait, la catégorie générale. Et le rapport entre le signifiant sensible – le son – et le signifié intelligible – le sens – est conventionnel, décidé par la société. Il n’y a aucune autre raison que la convention et l’usage pour que le concept de raison, par exemple, se représente par les deux syllabes ré et zon en français, et la preuve c’est que ça se dit autrement dans d’autres langues. Tous les animaux communiquent mais seuls les êtres humains parlent, posent des questions, reconnaissent leur ignorance, dialoguent et surtout racontent des histoires. L’usage du langage donne aux humains non pas la conscience (que les autres animaux ont déjà), mais la conscience réflexive. La capacité de réfléchir sur les concepts nous est donnée par la manipulation des symboles.
Avec cette capacité de manipulation symbolique et cette réflexivité viennent deux caractéristiques spéciales de l’humanité: les systèmes techniques et les institutions sociales, tous deux d’une grande complexité et en constante évolution historique.
Une énorme partie de l’intelligence humaine est réifiée dans l’environnement technique et vécue dans des institutions sociales (rituels, politique, droit, religion, morale, etc.). La partie individuelle de notre intelligence est marginale, mais essentielle, c’est elle qui nous permet d’innover, de progresser et d’améliorer notre condition.
Qu’est-ce que l’informatique, ou le traitement automatique de l’information?
L’intelligence artificielle est une expression de type « marketing » pour designer en fait la zone la plus avancée et toujours en mouvement des techniques de traitement de l’information.
Quand je dis que l’intelligence humaine a toujours été artificielle, je ne veux pas dire que les humains sont des robots ou des machines, je veux dire que les humains ont toujours utilisé des procédés techniques pour augmenter leur intelligence, qu’il s’agisse de l’intelligence personnelle ou collective. L’écriture nous a donné le moyen d’étendre notre mémoire individuelle et nos capacités critiques. Aujourd’hui l’Internet nous permet un accès rapide à une quantité d’information que nos ancêtres n’auraient jamais pu imaginer. Mais ce n’est pas seulement une question de mémoire, nous avons aussi des capacités de calcul, de simulation de systèmes complexes, d’analyse automatique des données, voire de raisonnement automatique qui amplifient les capacités cognitives “purement biologiques” des premiers homo sapiens. Nous avons le même cerveau que les hommes préhistoriques, avec la même capacité de manipuler les symboles et de raconter des histoires, mais nous avons en plus un énorme appareillage d’enregistrement, de communication et de traitement des symboles qu’ils n’avaient pas et qui se branche sur la partie purement biologique de notre intelligence.
L’informatique, le traitement automatique des données, avec sa pointe avancée et mouvante qu’on appelle l’intelligence artificielle, est apparue dans la seconde moitié du 20e siècle, mais elle poursuit un effort multi-séculaire d’augmentation cognitive qui a commencé avec l’écriture, s’est poursuivi avec le perfectionnement des systèmes de codage de la connaissance, la notation des nombres par position et le 0, l’imprimerie et les médias électriques…
La partie névralgique du nouvel appareillage de traitement automatique des symboles se trouve aujourd’hui dans d’énormes centres de calculs qu’on appelle le “cloud” et dont nos ordinateurs et smartphones ne sont que des terminaux. Mais dans ce réseau de machines, le traitement automatique des données se fait uniquement sur la forme sensible des symboles, sur le signifiant ramené à des zeros et des uns. Les ordinateurs n’ont pas accès au signifié, au sens.
Puisqu’on m’interroge sur le machine learning, oui, parmi toutes les techniques de calcul utilisées aujourd’hui par les ingénieurs en informatique, le machine learning, et le deep learning qui en est un cas particulier, sont en plein développement depuis une dizaine d’années. Mais il faut se garder d’attribuer à l’apprentissage automatique plus qu’il ne peut donner. Il s’agit essentiellement d’algorithmes de traitement statistique auxquels on soumet en entrée d’énormes masses de données et qui produisent en sortie des modèles de reconnaissance de formes ou d’action qui sont “appris” des données. Or non seulement l’apprentissage machine dépend des algorithmes qui sont programmés et continuellement débogués par des humains, mais en plus ses résultats en sortie dépendent des masses de données qui leur sont fournies en entrée. Or ce sont encore des humains qui choisissent les données, les filtrent, les classent, les catégorisent, les organisent, les interprètent, etc. Aussi bien les approches logiques que les approches statistiques de l’intelligence artificielle condensent dans des machines logicielles et matérielles des connaissances et des finalités humaines. Leur autonomie, si autonomie il y a, ne peut être que locale et momentanée. A moyen et long terme, les machines ne peuvent évoluer qu’avec nous et vice versa: nous ne pouvons évoluer qu’avec elles.
La question de l’autonomie des machines
Le traitement automatique des données prolonge l’ensemble du système technique contemporain et il baigne dedans. Il est totalement dépendant de la production d’énergie, de la distribution d’électricité, de la production des matériaux, etc. On ne peut absolument pas imaginer le système technique contemporain sans l’informatique mais pas non plus l’informatique sans toute cette infrastructure technique.
De la même manière, le système technique s’effondrerait rapidement si les humains disparaissaient. Notre environnement technique est conçu, construit, utilisé, entretenu, réparé, interprété par des humains: il n’a aucune autonomie d’aucune sorte.
la technique nous *apparaît* autonome parce que nous projetons sur elle les effets émergents des interactions sociales et des inerties socio-techniques que nous ne pouvons pas contrôler à l’échelle individuelle. Nous avons tendance à réifier les effets de nombreuses décisions et actions humaines agrégées dans les machines et à prêter aux machines une volonté propre. Mais c’est une illusion. Une illusion qui nous décharge de nos responsabilités personnelles et collectives: “c’est la faute de la machine”.
Qu’on utilise une interface pseudo-humaine ou des robots androïdes autant qu’on veut, mais c’est un artifice, un décor. Le robot ou la machine est toujours susceptible d’être éteint ou débranché, quant à son logiciel dans le cloud, il doit sans cesse être déboggué et de nouvelles versions doivent être téléchargées périodiquement. Pour moi, cette idée de la machine autonome relève du fétichisme : on donne une personnalité à un appareil qui n’est pas un être sensible et qui a été – encore une fois – conçu, fabriqué, marqueté, vendu, utilisé, réparé et qui va finalement être jeté à la poubelle au profit d’un nouveau modèle.
Nous avons des machines capables de traitement automatique des symboles. Et nous ne les avons que depuis moins d’un siècle. A l’échelle de l’évolution historique, trois ou quatre générations, ce n’est presque rien. A la fin du XXe siècle, 1% de la population humaine avait accès à l’Internet et le Machine Learning était confiné dans des laboratoires scientifiques. Aujourd’hui plus de 60% de la population est branchée et le machine learning s’applique à grande échelle aux données entreposées dans le cloud. Face à cette mutation si rapide, nous avons la responsabilité d’orienter, autant que possible, le développement technique, social et culturel. Plutôt que de s’égarer dans le fantasme de la machine qui prend le pouvoir, pour le meilleur ou pour le pire, Il me semble beaucoup plus intéressant d’utiliser les machines pour une augmentation de l’intelligence humaine, intelligence à la fois personnelle et collective. C’est plutôt dans cette direction qu’il faut travailler parce que c’est la seule qui soit utile et raisonnable. Et c’est d’ailleurs ce que font en silence les principaux industriels du secteur, même si la “singularité” attire plus l’attention des foules.
Si vous visez le divin, ou le dépassement, ne tentez pas de remplacer l’homme par une machine prétendument consciente et ne craignez pas non plus un tel remplacement, parce qu’il est impossible. Ce qui est peut-être possible, en revanche, est un état de la technique et de la civilisation dans lequel l’intelligence collective humaine pourra s’observer scientifiquement, déployer et cultiver sa complexité inepuisable dans le miroir numérique. Faire travailler les machines à l’emergence d’une intelligence collective réflexive, un pas apres l’autre…
This blog post offers a simple guide to the landscape of signification in language. We’ll begin by distinguishing the numerous elements that construct meaning. We’ll start by having a look at signs, and how they are everywhere in communication between living beings and how a sign is different from a symbol for instance. A symbol is a special kind of sign unique to humans, that folds into a signifier (a sound, an image, etc.) and a signified (a category or a concept). We’ll learn that the relationship between a signifier and a signified is conventional. A bit further, I’ll explain the workings of language, our most powerful symbolic system. I will review successively what grammar is: the recursive construction of sense units; semantics: the relations between these units; and pragmatics: the relations between speech, reference and social context. I’ll end this chapter by recalling some of the problems in fields of natural language processing (NLP).
Sign, symbol, language
Sign
Meaning involves at least three actors playing distinct roles. A sign (1) is a clue, a trace, an image, a message or a symbol (2) that means something (3) for someone.
A sign may be an entity or an event. What makes it a sign is not its intrinsic properties but the role it plays in meaning. For example, an individual can be the subject (thing) of a conversation, the interpreter of a conversation (being) or he can be a clue in an investigation (sign).
A thing, designated by a sign, is often called the object or referent, and – again –what makes it a referent is not its intrinsic properties but the role it plays in the triadic relation.
A being is often called the subject or the interpreter. It may be a human being, a group, an animal, a machine or whatever entity or process endowed with self-reference (by distinguishing self from the environment) and interpretation. The interpreter always takes the context into account when it interprets a sign. For example, a puppy (being) understands that a bite (sign) from its playful sibling is part of a game (thing) and may not be a real threat in the context.
Generally speaking, communication and signs exist for any living organisms. Cells can recognize concentrations of poison or food from afar, plants use their flowers to trick insects and birds into their reproductive processes. Animals – organisms with brains or nervous systems – practice complex semiotic games that include camouflage, dance and mimicries. They acknowledge, interpret and emit signs constantly. Their cognition is complex: the sensorimotor cycle involves categorization, feeling, and environmental mapping. They learn from experience, solve problems, communicate and social species manifest collective intelligence. All these cognitive properties imply the emission and interpretation of signs. When a wolf growls, no need to add a long discourse, a clear message is sent to its adversary.
Symbol
A symbol is a sign divided into two parts: the signifier and the signified. The signified (virtual) is a general category, or an abstract class, and the signifier (actual) is a tangible phenomenon that represents the signified. A signifier may be a sound, a black mark on white paper, a trace or a gesture. For example, let’s take the word “tree” as a symbol. It is made of: 1) a signifier sound voicing the word “tree”, and 2) a signified concept that means it is part of the family of perennial plants with roots, trunk, branches, and leaves. The relationship between the signifier and the signified is conventional and depends on which symbolic system the symbol belongs to (in this case, the English language). What we mean by conventional is that in most cases, there is no analogy or causal connection between the sound and the concept: for example, between the sound “crocodile” and the actual crocodile species. We use different signifiers to indicate the same signified in different languages. Furthermore, the concepts symbolized by languages depend on the environment and culture of their speakers.
The signified of the sound “tree” is ruled by the English language and not left to the choice of the interpreter. However, it is in the context of a speech act that the interlocutor understands the referent of the word: is it a syntactic tree, a palm tree, a Christmas tree…? Let’s remember this important distinction: the signified is determined by the language but the referent depends on the context.
Language
A language is a general symbolic system that allows humans to think reflexively, ask questions, tell stories, dialogue and engage in complex social interactions. English, French, Spanish, Arabic, Russian, or Mandarin are all natural languages. Each one of us is biologically equipped to speak and recognize languages. Our linguistic ability is natural, genetic, universal and embedded in our brain. By contrast, any language (like English, French, etc.) is based on a social, conventional and cultural environment; it is multiple, evolving and hybridizing. Languages mix and change according to the transformations of demographic, technological, economic, social and political contexts.
Our natural linguistic abilities multiply our cognitive faculties. They empower us with reflexive thinking, making it easy for us to learn and remember, to plan in the long-term and to coordinate large-scale endeavors. Language is also the basis for knowledge transmission between generations. Animals can’t understand, grasp or use linguistic symbols to their full extent, only humans can. Even the best-trained animals can’t evaluate if a story is false or exaggerated. Koko the famous gorilla will never ask you for an appointment for the first Tuesday of next month, nor will it communicate to you where its grandfather was born. In animal cognition, the categories that organize perception and action are enacted by neural networks. In human cognition, these categories may become explicit once symbolized and move to the forefront of our awareness. Ideas become objects of reflection. With human language comes arithmetic, art, religion, politics, economy, and technology. Compared to other social animal species, human collective intelligence is most powerful and creative when it is supported and augmented by its linguistic abilities. Therefore, when working in artificial intelligence or cognitive computing, it would be paramount to understand and model the functioning of neurons and neurotransmitters common to all animals, as well as the structure and organization of language, unique to our species.
I will now describe briefly how we shape meaning through language. Firstly, we will review what the grammatical units are (words, sentences, etc.). Secondly, we will explore the semantic networks between these units, and thirdly, what are the pragmatic interactions between language and extralinguistic realities.
Grammatical units
A natural language is made of recursively nested units: a phoneme which is an elementary sound, a word, a chain of phonemes, a syntagm, a chain of words and a text, a chain of syntagms. A language has a finite dictionary of words and syntactic rules for the construction of texts. With its dictionary and set of syntactic rules, a language offers its users the possibility to generate – and understand – an infinity of texts.
Phonemes
Humans beings can’t pronounce or recognize several phonemes simultaneously. They can only pronounce one sound at a time. So languages have to obey the constraint of sequentiality. A speech is a chain of phonemes with an acoustic punctuation reflecting its grammatical organization.
Phonemes are meaningless sounds without signification1 and generally divided into consonants and vowels. Some languages also have “click” sounding consonants (in Eastern and Southern Africa) and others (in Chinese Mandarin) use different tones on their vowels. Despite the great diversity of sounds used to pronounce human languages, the number of conventional sounds in a language is limited: the order of magnitude is between thirty and one hundred.
Words
The first symbolic grammatical unit is the word, a signifier with a signified. By word, I mean an atomic unit of meaning. For example, “small” contains one unit of meaning. But “smallest” contains two: “small” (meaning tiny) and “est” (a superlative suffix used at the end of a word indicating the most).
Languages contain nouns depicting structures or entities, and verbs describing actions, events, and processes. Depending on the language, there are other types of words like adjectives, adverbs, prepositions or sense units that orient grammatical functions, such as gender, number, grammatical person, tense and cases.
Now let’s see how many words does a language hold? It depends. The largest English dictionary counts 200,000 words, Latin has 50,000 words, Chinese 30,000 characters and biblical Hebrew amounts to 6,000 words. The French classical author Jean Racine was able to evoke the whole range of human passions and emotions by using only 3,700 words in 13 plays. Most linguists think that whatever the language is, an educated, refined speaker masters about 10,000 words in his or her lifetime.
Sentences
Note that a word alone cannot be true or false. Its signifier points to its signified (an abstract category) and not to a state of things. It is only when a sentence is spoken in a context describing a reality – a sentence with a referent – that it can be true or false.
A syntagm (a topic, sentence, and super-sentence) is a sequence of words organized by grammatical relationships. When we utter a syntagm, we leave behind the abstract dictionary of a language to enter the concrete world of speech acts in contexts. We can distinguish three sub-levels of complexity in a syntagm: the topic, the sentence, and the super-sentence. Firstly, a topic is a super-word that designates a subject, a matter, an object or a process that cannot be described by just a single word, i.e., “history of linguistics”, “smartphone” or “tourism in Canada”. Different languages have diverse rules for building topics like joining the root of a word with a grammatical case (in Latin), or agglutination of words (in German or Turkish). By relating several topics together a sentence brings to mind an event, an action or a fact, i.e., “I bought her a smartphone for her twentieth birthday”. A sentence can be verbal like in the previous example, or nominal like “the leather seat of my father’s car”. Finally, a super-sentence evokes a network of relations between facts or events, like in a theory or a narrative. The relationships between sentences can be temporal (after), spatial (behind), causal (because), logical (therefore) or underline contrasts (but, despite…), and so on.
Texts
The highest grammatical unit is a text: a punctuated sequence of syntagms. The signification of a text comes from the application of grammatical rules by combining its signifieds. The text also has a referent inferred from its temporal, spatial and social context.
In order to construct a mental model of a referent, a reader can’t help but imagine a general intention of meaning behind a text, even when it is produced by a computer program, for instance.
Semantic relationships
When we hear a speech, we are actually transforming a chain of sounds into a semantic network, and from this network, we infer a new mental model of a situation. Conversely, we are able to transform a mental model into the corresponding semantic network and then from this network, back into a sequence of phonemes. Semantics is the back and forth translation between chains of phonemes and semantic networks. Semantic networks themselves are multi-layered and can be broken down into three levels: paradigmatic, syntagmatic and textual.
Figure: Hierarchy of grammatical units and semantic relations
Paradigmatic relationships
In linguistics, a paradigm is a set of semantic relations between words of the same language. They may be etymological, taxonomical relations, oppositions or differences. These relations may be the inflectional forms of a word, like “one apple” and “two apples”. Languages may comprise paradigms to indicate verb tenses (past, present, future) or mode (active, passive). For example, the paradigm for “go” is “go, went, gone”. The notion of paradigm also indicates a set of words which cover a particular functional or thematic area. For instance, most languages include paradigms for economic actions (buy, sell, lend, repay…), or colors (red, blue, yellow…). A speaker may transform a sentence by replacing one word from a paradigm by another from the same paradigm and get a sentence that still makes sense. In the sentence “I bought a car”, you could easily replace “bought” by “sold” because “buy” and “sell” are part of the same paradigm: they have some meaning in common. But in that sentence, you can’t replace “bought” by “yellow” for instance. Two words from the same paradigm may be opposites (if you are buying, you are not selling) but still related (buying and selling can be interchangeable).
Words can also be related when they are in taxonomic relation, like “horse” and “animal”. The English dictionary describes a horse as a particular case of animal. Some words come from ancient words (etymology) or are composed of several words: for example, the word metalanguage is built from “meta” (beyond, in ancient Greek) and “language”.
In general, the conceptual relationships between words from adictionary may be qualified as paradigmatic.
Syntagmatic relationships
By contrast, syntagmatic relations describe the grammatical connections between words in the same sentence. In the two following sentences: “The gazelle smells the presence of the lion” and “The lion smells the presence of the gazelle”, the set of words are identical but the words “gazelle” and “lion” do not share the same grammatical role. Since those words are inversed in the syntagmatic structure, the sentences have distinct meanings.
Textual relationships
At the text level, which includes several syntagms, we find semantic relations like anaphoras and isotopies. Let’s consider the super-sentence: “If a man has talent and can’t use it, he’s failed.” (Thomas Wolfe). In this quotation “it” is an anaphora for “talent” and “he”, an anaphora for “a man”. When reading a pronoun (it, he), we resolve the anaphora when we know which noun – mentioned in a previous or following sentence – it is referring to. Onthe other hand, isotopies are recurrences of themes that weave the unity of a text: the identity of heroes (characters), genres (love stories or historical novels), settings, etc. The notion of isotopy also encompasses repetitions that help the listener understand a text.
Pragmatic interactions
Pragmatics weave the triadic relation between signs (symbols, speeches or texts), beings (interpreters, people or interlocutors) and things (referents, objects, reality, extra-textual context). On the pragmatic level of communication, speeches point to – and act upon – a social context. A speech act functions as a move in a game played by its speaker. So, distinct from semantic meaning, that we have analyzed in a previous section, pragmatic meaning would address questions like: what kind of act (an advice, a promise, a blame, a condemnation, etc.) is carried by a speech? Is a speech spoken in a play on a stage or in a real tribunal? The pragmatic meaning of a speech also relates to the actual effects of its utterance, effects that are not always known at the moment of the enunciation. For example: “Did I convince you? Have you kept your word?”. The sense of a speech can only be understood after its utterance and future events can always modify it.
A speech act is highly dependent on cultural conventions, on the identity of speakers and attendees, time and place, etc. By proclaiming: “The session is open”, I am not just announcing that an official meeting is about to start, I am actually opening the session. But I have to be someone relevant or important like the president of that assembly to do so. If I am a janitor and I say: “The session is open”, the act is not performed because I don’t have any legitimacy to open the session.
If an utterance is descriptive, it’s either true or false. In other cases, if an utterance does something instead of describing a state of things, it has a pragmatic force instead of a truth value.
Resolving ambiguities
We have just reviewed the different layers of grammatical, semanticand pragmatic complexity to better understand the meaning of a text. Now, we are going to examine the ambiguities that may arise during the reading or listening of a text in a natural language.
Semantic ambiguities
How do we go from to the sound of a chain of phonemes to the understanding of a text? From a sequence of sounds, we build a multi-layered (paradigmatic, syntagmatic and textual) semantic network. When weaving the paradigmatic layer, we answer questions like: “What is this word? To what paradigm does it belong? Which one of its meanings should I consider?”. Then, we connect words together by answering: “What are the syntagmatic relations between the words in that sentence?”. Finally, we comprehend the text by recognizing the anaphoras and isotopies that connect its sentences. Our understanding of a text is based on this three-layered network of sense units.
Furthermore, ambiguities or uncertainties of meaning in languages can happen on all three levels and can multiply their effects. In the case of homophony, the same sound can point to different words like in “ate” and “eight”. And sometimes, the same word may convey several distinct meanings like in “mole”: (1) a shortsighted mouse-like animal digging underground galleries, (2) an undercover spy, or (3) a pigmented spot or mark on the skin. In the case of synonymy, the same meaning can apply to distinct words like “tiny” and “small”. Amphibologies refer to syntagmatic ambiguities as in: “Mary saw a woman on the mountain with a telescope.” Who is on the mountain? Moreover, who has the telescope? Mary or the woman? On a higher level of complexity, textual relations can be even more ambiguous than paradigmatic and syntagmatic ones because rules for anaphoras and isotopies are loosely defined.
Resolving semantic ambiguities in pragmatic contexts
Human beings don’t always correctly resolve all the semantic ambiguities of a speech, but when they do, it is often because they take into account the pragmatic (or extra-textual) context that is generally implicit. It’s in a context,that deictic symbols like: here, you, me, that one over there, or next Tuesday, take their full meaning. Let’s add that, comparing a text in hand with the author’s corpus, genre, historical period, helps to better discern the meaning of a text. But some pragmatic aspects of a text may remain unknown. Ambiguities can stem from many causes: the precise referents of a speech, the uncertainty of the speaker’s social interactions, the ambivalence or concealment of the speaker’s intentions, and of course not knowing in advance the effects of an utterance.
Problems in natural language processing
Computer programs can’t understand or translate texts with dictionaries and grammars alone. They can’t engage in the pragmatic context of speeches like human beings do to disambiguate texts unless this context is made explicit. Understanding a text implies building and comparing complex and dynamic mental models of text and context.
On the other hand, natural language processing (a sub-discipline of artificial intelligence) compensates for the irregularity of natural languages by using a lot of statistical calculations and deep learning algorithms that have been trained on huge corpora. Depending on its training set, an algorithm can interpret a text by choosing the most probable semantic network amongst those compatible within a chain of phonemes. Imperatively, the results have to be validated and improved by human reviewers.
I was happily surprised to be chosen as an “IBM influencer” and invited to the innovation and disruption Forum organized in Toronto the 16th of November to celebrate the 100th anniversary of IBM in Canada. With a handful of other people, I had the privilege to meet with Bryson Koehler the CTO of the IBM Cloud and Watson (Watson is the name given to IBM’s artificial intelligence). That meeting was of great interest to me: I learned a lot about the current state of cloud computing and artificial intelligence.
Image: Demonstration of a robot at the IBM innovation and disruption forum in Toronto
Contrary to other big tech companies, IBM already existed when I was born in 1956. The company was in the business of computing even before the transistors. IBM adapted itself to electronics and dominated the industry in the era of big central mainframes. It survived the PC revolution when Microsoft and Apple were kings. They navigated the turbulent waters of the social Web despite the might of Google, Facebook, and Amazon. IBM is today one of the biggest players in the market for cloud computing, artificial intelligence and business consulting.
The transitions and transformations in IBM’s history were not only technological but also cultural. In the seventies, when I was a young philosopher and new technology enthusiast, IBM was the epitome of the grey suit, blue tie, black attache-case corporate America. Now, every IBM employee – from the CEO Dino Trevisani to the salesman – wears jeans. IBM used to be the “anti-Apple” company but now everybody has a Mac laptop. Instead of proprietary technology, IBM promotes open-source software. IBM posters advertise an all-inclusive and diverse “you” across the specter of gender, race, and age. Its official management and engineering philosophy is design thinking and, along with the innovative spirit, the greatest of IBM’s virtues is the ability to listen!
Toronto’s Forum was all about innovation and disruption. Innovation is mainly about entrepreneurship: self-confidence, audacity, tenacity, resilience and market orientation. Today’s innovation is “agile”: implement a little bit, test, listen to the clients, learn from your errors, re-implement, etc. As for the disruption, it is inevitable, not only because of the speed of digital transformation but also because of the cultural shifts and the sheer succession of generations. So their argument is fairly simple: instead of being disrupted, be the disruptor! The overall atmosphere of the Forum was positive and inspirational and it was a pleasure to participate.
There were two kinds of general presentations: by IBM clients and by IBM strategists and leaders. In addition, a lot of stands, product demonstrations and informative mini-talks on various subjects enabled the attendees to learn about current issues like e-health and hospital applications, robotics, data management, social marketing, blockchain and so on. One of the highlights of the day was the interview of Arlene Dickinson (a well known Canadian TV personality, entrepreneur, and investor) by Dino Trevisani, the CEO of IBM Canada himself. Their conversation about innovation in Canada today was both instructive and entertaining.
From my point of view as a philosopher specialized in computing, Bryson Koehler (CTO for IBM cloud and Watson) made a wonderful presentation, imbued with simplicity and clarity, yet full of interesting content. Before being an IBMer Bryson worked for the Weather Channel, so he was familiar handling exabytes of data! According to Bryson Koehler, the future is not only the cloud, that is to say, infrastructure and software as a service, but also in the “cloud-native architecture“, where a lot of loosely connected mini-services can be easily assembled like Lego blocks and on top of which you can build agile and resilient applications. Bryson is convinced that all businesses are going to become “cloud natives” because they need the flexibility and security that it provides. To illustrate this, I learned that Watson is not a standalone monolithic “artificial intelligence” anymore but is now divided into several mini-services, each one with its API, and part of the IBM cloud offer alongside other services like blockchain, video storage, weather forecast, etc.
Image: Bryson Koehler at the IBM innovation and disruption Forum in Toronto
Bryson Koehler recognizes that the techniques of artificial intelligence, the famous deep learning algorithms, in particular, are all the same amongst the big competitors (Amazon, Google, Microsoft and IBM) in the cloud business. These algorithms are now taught in universities and implemented in open source programs. So what makes the difference in IA today is not the technique but the quality and quantity of the datasets in use to train the algorithms. Since every big player has access to the public data on the web and to the syndicated data (on markets, news, finance, etc.) sold by specialized companies, what makes a real difference is the *private data* that lies behind the firewall of businesses. So what is the competitive advantage of IBM? Bryson Koehler sees it in the trust that the company inspires to its clients, and their willingness to confide their data to its cloud. IBM is “secure by design” and will never use a client’s dataset to train algorithms used by this client’s competitors. Everything boils down to confidence.
At lunchtime, with a dozen of other influencers, I had a conversation with researchers at Watson. I was impressed by what I learned about cognitive computing, one of IBM’s leitmotiv. Their idea is that the value is not created by replicating the human mind in a computer but in amplifying human cognition in real-world situations. In other words, Big Blue (IBM’s nickname) does not entertain the myth of singularity. It does not want to replace people with machines but help its clients to make better decisions in the workplace. There is a growing flow of data from which we can learn about ourselves and the world. Therefore we have no other choice than to automate the process of selecting the relevant information, synthesize its content and predict, as much as possible, our environment. IBM’s philosophy is grounded in intellectual humility. In this process of cognitive augmentation, nothing is perfect or definitive: people make errors, machines too, and there is always room for improvement of our models. Let’s not forget that only humans have goals, ask questions and can be satisfied. Machines are just here to help.
Once the forum was over, I was walking in front of the Ontario lake and thought about the similarity between philosophy and computer engineering: aren’t both building cognitive tools?
Image: walking meditation in front of the Lake Ontario after the IBM innovation and disruption Forum in Toronto
Image: Bryson Koehler at the IBM innovation and disruption Forum in Toronto
Image: walking meditation in front of the Lake Ontario after the IBM innovation and disruption Forum in Toronto 
Pierre Levy's Blog 