in Kybernetes (2000), vol 29, Number 7/8, 836-845


Abduction in Language interpretation and Law making


Evelyne Andreewsky  &  Danièle Bourcier  








          The complexity of any given cognitive phenomenon, such as “ scientific discovery ”, “ technical expertise ”, or “ natural language understanding ”, requires a multidisciplinary approach. We present, within the framework of such an approach, some visible evidences of how these very different phenomena are closely rooted in the same highly inventive cognitive process, abduction. These evidences will be provided out of examples from both everyday language interpretation and law making expertise.



Abduction, Cognition, Complexity, Language Interpretation, Law Making. 





            The computer metaphor has taken the lead in conceptualizing the mind and providing explanations for cognitive phenomena. With this metaphor, grounded in the most prestigious technique of our time (i.e. computing), the human mind becomes “computational”: it processes information. Psychology thus adopts the key-concepts of Computer Sciences (such as information, representation or computation). Let us recall that American Psychology (that of Behaviorism, which emphasized the study of the responses of rats to sensorial stimuli) switches, as Arthur Koestler claims, from “an anthropomorphic view of the rat to a ratomorphic view of man” (before disappearing altogether). We are now dealing with a “computomorphic” view of ourselves! This view has generated a new Science of the Mind, consisting of a vast disciplinary enterprise, born in the 70’s. It involves Artificial Intelligence, Neurosciences, Linguistics, Experimental Psychology, etc.; its aim is to build the intelligibility of mind and knowledge – hence its label of Cognitive Sciences.  


Cognitive Sciences and Systems Science have fundamentally the same interdisciplinary roots. In the mid-century, the need to explore complex systems and to counterbalance fragmented and compartmentalized approaches in scientific research led to interdisciplinary attempts to model complex phenomena. Typical of such attempts was Cybernetics, for modeling communication and control. Systems Science and Cognitive Sciences were born of Cybernetics. The explicit aim of Cybernetics was managing complex systems – opening therefore a way towards Systems Science. It was also a means (though more implicitly), of gaining some understanding of an extremely complex system - the cognitive one. This system is altogether fascinating for researchers (in keeping with the old injunction: know thyself), and most difficult to explain; indeed, according to Bateson, the human mind has explained everything – but not itself. That explanation is the whole project of Cognitive Sciences.


Aristotelian Logic and Cognitive Paradigms


            Cognitive Sciences are growing along multiple ways and paradigms. Their object, the study of human knowledge and of cognitive abilities such as thinking or reasoning, is obviously not a new one. Some thousand years ago in Greece, the highest form of thinking, “Logic”, was extensively investigated, and logic became the alpha and omega of the whole history of scientific progress. Now, since logic is, so to speak, embodied in computers, Cognitive Sciences are the heir of the logical approaches to reasoning. We may track the origin of these approaches to Aristotelian classical thought, and its three modes of reasoning (“reasoning” meaning “finding out the truth”) - deduction, induction, and retroduction (or “abduction”):


-        “To think is to compute” - the main cognitivist metaphor – may be linked to deduction;

-        “Thinking is an emerging property of neural networks” - the “Connectionist” paradigm - may be linked with induction.

-        Finally, a variety of cognitive paradigms - such as the “auto” ones (auto-organization, autopoïese, etc.) - which reject the hypothesis of any “objective external reality” providing information to the cognitive system - may be linked with abduction.





The main Cognitive Science paradigm, the symbolic - or computo-representational - one, which takes the computer metaphor as a functional model of mind, is close to deduction. This is particularly clear in the prototypical computo-representational approach, that is, the system-expert model, in terms of if…then (with the pattern of the well known Socratic deductive reasoning: All men are mortals; if x is a man, then x is mortal).

In this framework, the human cognitive system is characterized by both its internal (or “mental”) states and by the processes used to move from one state to another. These states are representational[1] and point to external (objective) entities.

The computational theory of mind comes directly from the cybernetic tradition. It is based on analogies between organisms and machines and, more precisely, on an equivalence between hardware and software, on one hand, and body and mind on the other. An equivalence which was proclaimed as early as 1943, in keeping with the renown W. Pitts and W. McCulloch experiment (McCulloch, 1965). It has been taken for granted by the cyberneticians, strongly relying upon a “computomorphic view of man”. Such a view induced many exciting hopes! First, the hope to peer into the mysterious “black box” of the behaviorists (that is, the cognitive system) and, conversely, to design some electronic golem (in keeping with the feeling of those outstanding scientists such as John von Neuman, or Norbert Wiener, when designing the first computer in the forties).




Induction, in contrast to deduction, goes from facts to theory. It is a very usual reasoning process, being, for instance, the (implicit) way we learn our mother tongue (the theory - here the rules of the language - being implicitly inferred out of day to day linguistic interactions). The connectionist paradigm is close to this framework. Indeed, this paradigm is rooted in “learning”. A connectionist system in the process of learning is a network of interconnected formal neurons which modifies its connections. Patterns (in other words, facts) are presented to the input “retina” of the system. It “learns” (for instance in the framework of a pattern recognition problem) how to discriminate and to class these patterns through successive modifications of the weight of the connections between its “neurons”. At the end of the learning process the system is provided, out of its weighted network, with a true functional “classification theory”; it is therefore able to label any new pattern presented to its retina as belonging to one or another class of patterns defined through learning.

The first "neuronic network", Rosenblatt’s Perceptron, was designed at the beginning of the 60’s. The approach was then abandoned for a variety of reasons, namely the weakness of its formalization (Minsky & Papert, 1968). It took fifteen years to see its re-emergence, under the name of Connectionism (or “neo-Connectionism”), in a more elaborate theoretical framework, thanks to the progress in non-linear Mathematics and in Systems Dynamics (Haken, 1983).

Several simulations of cognitive activities such as character recognition, language perception, etc., have been attempted in the framework of the connectionist paradigm. These simulations (McClelland et al, 1986) are no longer expressed, as are classical computational systems, in terms of messages transmitted between some modules of the system, but imply the activated state of the whole system, through weighted values - numbers, and not symbols.

Connectionism aims to explain behavior endogenously, by means of physiologically plausible hypotheses, with a direct relation to Neurosciences. Its type of explanation is opposed to the classical A.I. approaches, since within these latter approaches no behavior is produced unless the set of rules underlying this behavior is specified to the system.


The classical computo-representational paradigm and the connectionist one are quite different. Yet they share a “computomorphic” view of man (be it sequential, like in the former case, or “massively parallel”, like in the later one). In both cases, mind is defined as an information processing device operating upon an external “objective” reality.



-        Abduction


Abduction (“retroduction”, for Aristotle, cf. Peirce, 1995) has multiple definitions. In short, it is a creative process (Bourgine, 1989) which consists of finding a plausible hypothesis to fit a given “strange” phenomenon. Indeed, to quote von Foerster (1974), there is no information, or anomalies in our environment. Hence, if a given phenomenon looks strange, this only means that the theoretical framework used to interpret this phenomenon must be revisited! The revisiting cognitive process is labeled abduction, and its aim is to “normalize” anomalies.


The abduction pattern refers to the fairly common experience of observing an unexpected, anomalous and strategic datum, which becomes the occasion for developing a new theory or for extending an existing one.

The datum is, first of all, unexpected. It is for instance the case when a research directed towards the test of one hypothesis yields a fortuitious by-product, an unexpected observation which bears upon theories not in question when the research begun.

Secondly, the observation is anomalous, surprising, either seemingly inconsistent with the prevailing theory or with other established facts. In either case, the inconsistency provokes curiosity; it stimulates the investigator to make sense of the datum, to fit it into a broader frame of knowledge.

Thirdly, in noting that the unexpected fact must be strategic, i.e., that it must permit implications bearing on generalized theory, we are of course referring rather to what the observer brings to the datum, than to the datum itself. Here it obviously requires a theoretically sensitized observer to detect the universal in the particular.


According to Piaget & Garcia (1983) there is a  functional continuity between every day cognitive elaboration and scientific ones, given that we rely on our expertise to deal with the circumstances under hand, both in everyday life and in scientific research. Cognitive every day interpretation somehow recalls reasoning, featuring highly scientific domains. Indeed, both are dealing with the production and testing (falsifying) of hypothesis and theories, both resort to abduction, the process aiming to suggest a hypothesis able to explain a given unexpected phenomenon. According to Peirce (1995), such hypotheses or theories are “ new suggestions, even if all their elements were already in mind, since we never dream to put these elements together ”. In such a framework, interpreting everyday unexpected sentences, or thinking of new laws to handle unexpected behaviors, is something like suggesting scientific theories. It leads -like in theories - to the emergence of something new, i.e., some relevant hypotheses on what is at hand within given circumstances.

            The abductive framework emphasizing pragmatic hypotheses (that is, hypotheses capable of verification or justification) fitting circumstances, is a main creative cognitive process, as  will be illustrated hereafter by Language Interpretation and Judicial examples.


1.  Abduction and language interpretation


Several anomalies (or problems) emerge from the traditional “objective” approaches to meaning. These cognitive problems, insofar as they are inherent to understanding, may be highlighted through the difficulties of natural language understanding systems (A. I.). This type of framework presupposes an “objective” approach to meaning postulating namely, that all sentences have a meaning independent of the interlocutors… This is obviously not the case since, for instance, a statement such as the following: it’s better to give than to get, “takes” two completely opposite meanings, depending on whether it’s made by a priest during a sermon, or by a pugilist in the context of a boxing match! How can one then “construct” the meaning of this statement without being dependent on these circumstances (potentially infinite), that is to say without taking into account the creative dimension of understanding?


            Along this dimension, the abductive framework best fits everyday life examples, showing that “ the ” meaning of a given word may be strongly driven by the domain of experience in which this word occurs. Word meaning may indeed entirely change from one occurrence to the other, according to each new domain at hand, and is therefore likely to differ from any earlier meaning. This is obviously the case of a word occurring in a new “ metaphor ” (Lakoff & Johnson, 1980). But it can also be the case for a word occurring in any sentence. Thus the need to "create" and to recreate these words meanings. This may be illustrated with very common words and sentences, such as for instance “ water ” in the claim:

There is no more water!

            Such a claim, interpreted in the framework of many different domains of experience, is endowed with many different meanings. When uttered in a Supermarket, it usually means that there are no more bottles of mineral water left on the shelves - the lexical item water is therefore to be interpreted as “ bottles ”. If one is lost in the desert, this item comes to mean something like “ a drink vital for survival ”, and the claim is a cry of despair. It would turn into a cry of joy when bailing out a boat; water being now an undesirable ballast slowing down the boat. In the framework of a scientific article, commenting some “ memory of water ” data (likely to change our conceptualization of the liquid), water is something like “ our traditional concept of H2O ”, and so forth.

Such a strong “ updating ” of the meaning of water in each new domain of experience implies a recurrent creation of these meanings, and therefore a recurrent abduction.


Such an endless creation of meaning stresses the complexity of language understanding. Word meaning, far from pointing to any objective external reality, is made up of our complex subjective and inter-subjective internal one. In this framework, it stands for a nice metaphor of cognition, in keeping with Wygotski's (1995) claim: the meaning of a word reflects, in its simplest mode, the unity of Language and Thought; individual Cognition[2] is reflected in a given word as the sun in a drop of water. A word endowed with meaning is a microcosm of Human Cognition.



2. Judicial abduction: legal proof, legal presumption and experimental norms


            The judicial system is obliged to anticipate but it cannot foresee all: neither future events, nor the necessary rules, ror the effect of the implemented norms. We will focus on three cases where law relies upon abductive reasoning for the establishment of facts, the invention of principles, or the creation of norms.


            2. 1.            Legal proof and abductive reasoning


            The detective, the policeman, the paleontologist reconstitute the past by means of clues from which they put forward hypotheses. Certain authors have studied the relations between scientific reasoning and the detectives' procedures. The story of Oedipus could for example be interpreted as that of an examining magistrate inquiring on his personal case. Umberto Eco has proven that Conan Doyle used abductive reasoning as if it possessed an absolute logic. Sherlock Holmes in The Sign of Four constructs the following type of reasoning: the fact that Watson enters with a certain color of mud on his shoes allows the former to "deduce" that Watson has sent off a telegram from a certain post-office... The aim is to produce rigorously linked arguments, whose exposition is in the interest of the story. The policeman, who is Darwin's brother-in-arms, discovers the criminal via his foot-tracks by means of a mental process identical to the one used by Cuvier to reconstitute the aspect of his Montmartre fossils by the simple means of their bone debris (Huxley, 1988).


            Policemen and examining magistrates do not learn de visu about facts relevant to a case or a file.  The "truth" can hence only be established after traces, fingerprints on objets, memories of witnesses, or by experiment.  The knowledge obtained remains indirect, however, and depends on the aptitude to decipher and interpret the traces which may under certain conditions become judicial proofs.


            This notion of proof must be distinguished from that of proof in Logic. Logic has thought as its basis, that is, reasoning and not facts. In Logic, proof (or deduction) is the result of a reasoning, which serves to demonstrate the accuracy of a thesis. This is why Logic does not recognize abduction as valid reasoning. The separation between factual proof, logical proof and judicial presumption deserves our special attention.



            2. 2.            Presumption as an abductive reasoning


            Law is not merely preoccupied with facts: the judge, the administrator, "qualifies" - i.e., matches facts and hypotheses with the legal condition of the application of a rule.  Facts can be infinite and the judge only takes into consideration those which produce a judicial consequence.  Using a comparison with scientists: the chemist or physician is only interested in details as far as they help to confirm or infirm a certain theory or hypothesis (Perelman, 1975)


            A second characteristic is that in the search for facts, law has regulated the administration of proof, subject to a specific procedure. If this procedure is not followed, the proof provided is not acceptable even if the "truth" is glaring. For instance, law has created a presumption which makes it difficult to search for facts leading to a proof of fatherhood. In fact, the husband is the presumed father of a child conceived during marriage: he is the only person who can disown the child and he only has six months to do so. What purpose do these presumptions serve? They allow to conclude without proof. It devolves on the person who refutes the results to bring the proof.


            In law, presumption is defined as the technique allowing to provide solutions under the uncertainty of proofs.


            What reasoning will the judge follow in the case of contention? He will draw from a known fact (the date of birth, or of marriage) the consequences of the unknown fact: the fatherhood or non-fatherhood of x. Such a fact, if not true, is at least becoming likely out of a first group of facts or clues. This idea of likelihood is interesting in our case. It limits impossible induction and delirious interpretations (the famous  "interpretation delirium").


            One may thus know the likelihood of a hypothesis and the acceptability of a conclusion. 


            Let us consider the schemas of Peirce in the case of fatherhood:


                                         The father is the mother's husband (plausible)

x is y's child (known fact)

ABDUCTION              y is the husband of x's mother (possible fact)



                                        y is the husband of x's mother                            

                                          x is y's child

INDUCTION              The father is the mother's husband (probable)



                                       The father is the mother's husband (legal presumption)

                                        y is the husband of x's mother

DEDUCTION             x is y's child



            The abduction which started off from a simple presumption becomes thus a legal presumption, that is, a rule of law which, while basing itself on factual truth, forces the truth mechanism.



            2. 3. Law making and experimental norms


            The legislator may have foreseen the facts and the rules but not the citizens' reactions to a new law. Any normative system can produce effects which are unexpected and even conflicting with the initial aim. In the framework of law making, unexpected effects can indeed emerge. They can be positive (improving the law), or negative (bypassing the law). In this framework, we must rely upon abduction to cope with such law effects.

            Legal technique has already foreseen this possibility of palliating the unexpected effects of texts, or of finding solutions to new situations, by developing so-called experimental laws. These laws aim at testing the effects of new norms or institutions concerning social or technical evolutions (such as abortion in France, pre-legal euthanasia in the Netherlands), or likely to have important economical effects (new taxes). These laws are evaluated, extended, modified or definitely adopted, following the ensuing results.


            Let us observe, for example, how, in the case of euthanasia, new rules based on practice are emerging in the Netherlands.


            How can one use the present social system and normative framework to anticipate or interpret a new, surprising, controversial and complex phenomenon? Progress in science and medical technology has developed the situation of euthanasia. Formerly, in the Netherlands, like in France today, euthanasia was practiced in "petit cabinet" by means of an overdose of morphine and registered as "natural death". At present euthanasia is "legally" organized in the following way (Griffiths, 1998):

            Euthanasia has always been considered as murder in the existing juridical system. Nonetheless, since 1990, a partial "legalization" has been installed: this is assorted with the physicians' duty to prepare a case report and to transmit it to the public prosecutor for evaluation of all necessary conditions. This procedure was even validated by the Parliament in 1993.  The court either definitely files and disposes of the case, or prosecutes the physician who is then judged. Though considered guilty, he may not, however, be sentenced.


            Social laboratories and professional communities are used to help promote the process of learning and to encourage the emergence of rules representing maximal consensus.


            The same type of evaluative approach was used in France in the case of child abortion (I.V.G.: Voluntary pregnancy interruption). Abortion was "experimented" by a temporary law in 1975. This phase was considered as very important, not so much for the uncertainty of its effects (an increase in abortions, for example), as for the social resistance and the strong political opposition in Parliament (the project was only approved by 99 members of the majority which then counted 291).


            In the Netherlands, in the case of euthanasia (or aid in committing suicide, its other alternative) we observe a different type of elaboration and adjustment process.  There is no preliminary law, but experimental practice allowing abductive approaches of a problem considered case by case, following protocols established by physicians, and the personal demands of patients. A new law was then proposed. The social corpus (in this case, the physicians) took the risk of elaborating the first rules of the game.


            The behaviors and reactions (unknown at the start) instigated by the new facts will be interpreted in order to allow the emergence of an adequate procedure leading to the establishment of a new rule.  This procedure develops the emergence of shared knowledge (the aim being to render the procedure visible and hence to incite the physicians to prepare medical reports) when faced with unknown situations. This knowledge is mutually co-determined, and it facilitates consensus via temporary abductive mechanisms.





We have tried to present how abduction patterns refer to the fairly common experience of dealing with an unexpected phenomenon, which calls for developing a new interpretation or for extending an existing one.

Anyone testing a given hypothesis and observing an anomaly that surprises because it conflicts with his knowledge and views, normally reacts by suspecting a mistake. When he has excluded this possibility, his second sensible reaction is to find an ad hoc interpretation to explain the anomaly. If this interpretation is interesting, probable, simple, elegant and testable enough, he can experiment to verify this new hypothesis. A scientific investigation runs on two legs, one leg for hypothesis testing, and a second leg for anomaly explaining, out of abduction (Merton R. K., 1957). 

The same holds for Language understanding, either in everyday life or within the Legal framework: clearly, one cannot simply rely on ready made meanings, neither for usual nor for Legal Language where, in both cases, understanding is a creative mechanism able to cope with unexpected phenomena.

Scientists have long been regarded as using mainly the method of hypothesis testing; yet, from Aristotle to Peirce, a number of them have espoused the method of anomaly explaining (involving abduction processes). This is how both the scientist and the layman reason.






- Andreewsky E. (ed.), 1991, Systémique et cognition, Dunod, Paris.

- Bourcier D., 1999, Interprétation et abduction. La science du droit est-elle concernée par la sérendipité?, Revue interdisciplinaire d’Etudes juridiques, Facultés universitaires Saint Louis, Bruxelles, Special Issue “ Interpretation ” (forthcoming).

- D. Bourcier, 1992, "Peut-on faire émerger des normes d'un réseau? Quelques réflexions sur l'approche connexionniste du droit" in Lire le droit; Langue texte cognition, D. Bourcier & P. Mackay  (éds), LGDJ, Paris, 307-331.

- Bourgine P., 1989, "Connaissances abductives et compilation de connaissances", Proceedings of the First European Systems Sciences Congress, Vol. 2, 995-1002.

- Chevallier J., 1996, "Les lois expérimentales" in L'écriture du droit face aux technologies de l'information, Bourcier D. & Thomasset  C. (éds), Diderot, Paris.

- Cote P.-A., 1992, "L'interprétation de la loi, une création sujette à contraintes"  in Lire le droit Langue texte cognition, D. Bourcier & P. Mackaay (éds),  LGDJ, Paris,135-143.

- Dupuy J.-P., 1994, Aux origines des sciences cognitives, Editions La Découverte, Paris.

- Eco U., 1981, "Guessing: from Aristotle to Sherlock Holmes", Versus, n° 30, September-December, 14.

- Griffiths J., Bood A. & Weyers H., 1998, Euthanasia & Law in the Netherlands, Amsterdam University Press, Amsterdam.

- Grize J-B., 1990, "Mais au fond qu'est-ce que raisonner"?, Archives de Psychologie, 58, 118.

- Haken H., 1983, Synergetics: An Introduction, Springer, Berlin.

- Huxley Th, 1988, Science, Sculpture and other Essays, Mac Milan, London.

- Lakoff G. & Johnson M., 1980, Metaphors we Live By, Univ. of Chicago Press.

- McClelland J., Rumelhart D. & the PDP Research Group, 1986, Parallel Distributed Processing: Explorations in the Microstructure of Cognition, MIT Press, Cambridge.

- McCulloch W., 1965, Embodiment of Mind, MIT Press, Cambridge.

- Maturana H. & Varela F. J., 1980, Autopoiesis & Cognition, D. Reidel Publ. Company, Dordrecht.

- Merton R. K., 1957, Social Theory and Social Structure, New-York, The Free Press.

- Minsky M. and Papert S., 1969, Perceptrons, MIT Press, Cambridge.

- Piaget J. & Garcia R., 1983, Psychogenèse et histoire des Sciences, Paris, Flammarion.

- von Foerster, H., 1974, "Notes pour une épistémologie des objets vivants", in Morin E. & Piattelli-Palmerini M. (eds.), L'unité de l'homme : Invariants biologiques et universaux culturels, Seuil, Paris.

- Peirce C.S., 1995, Le raisonnement et la logique des choses (The Cambridge Conferences; lectures of 1898), Editions du Cerf, Paris.

- Rompen C., Meek M. F. &. van Andel M.V., "A cause célèbre: the so-called Bal-point murder", Journal of Forensic Sciences (accepted for publication)

- Shanon, B., 1993, The Representational and the Presentational, Harvester Wheatsheaf, New-York.

- Vygotski L., 1985, Pensée et langage, trad. F. Sève, Paris, Messidor.

- Winograd T. & Flores F., 1986, Understanding Computers and Cognition, Ablex.



[1]The representational theory of mind goes back at least to Descartes. For the Cognitive Sciences, the representation of knowledge in human memory constitutes a main research problem; one of the first models of semantic memory (Quillan, 1968) was built on the hypothesis of an hierarchical organization of the concepts stored in memory, according to some cognitive economy. From that time, many cognitive representational models have been developed in classical Artificial Intelligence (I. A.), to cope with behavioral data (Shanon, 1993).

[2] "Cognition" is "sosnanie" in Russian; and means also "consciousness".