The use of articles to express referring cases:
towards a mentalist explanation

Andrei Popescu-BelisISSCO, University of Geneva
40, Bd du Pont d’Arve
CH-1211 Geneva 4

Since the pioneering works of Frege and Russell, it has been common to use logic and set-theory to explain how articles express the referential properties of noun phrases, in the languages that make use of articles. Such views have been embedded, for instance, in the Discourse Representation Theory (Kamp & Reyle 1993), and have been used quite successfully in Natural Language Generation programs (Dale 1992).

However, logic-based approaches have been challenged on philosophical, cognitive and linguistic grounds. First, the "descriptive theories" of reference suffer from several drawbacks (Devitt & Sterelny 1999). Second, it is still unclear whether human mind uses logic-form representations. Third, as functionalists have pointed out, there is considerable variation in the expression of each referential property. We will first summarize the functionalist model, then describe our own proposal for a cognitive account of the link between expression, referring status and representations.

One of the most interesting functionalist models of reference has been proposed by D. Bickerton and T. Givón, using two features to describe the referential properties of the entity the speakers refer to. The entity can be a specific individual or not (+S or -S) and it can be presupposed or not (i.e., known or not) from previous discourse (+P or -P). Bickerton (1981) explains the use of articles in several Creole languages by representing the four cases as quadrants in the S/P coordinates (+S+P, +S-P, -S-P, -S+P), and associating to each article one or several contiguous quadrants. Despite the superficial mapping between Creole articles and the articles in the superstrate language (here, English), the two systems are used in a different way to express the various cases. Whereas in English there is considerable overlapping between the domains (quadrants) associated with each article, Guyanese Creole for instance makes a more straightforward use of the three articles, "di" for +S+P, "wan" for +S-P and "no article" for -S-P and -S+P.

Givón (1984:406-435) used more evocative names for these cases : "definite" (+S+P) > "referential indefinite" (+S-P) > "non referential" (-S-P) > "generic" (-S+P). He situates them on a helix, in which the two extreme cases are contiguous, albeit on a different level (individuals vs. categories). He also defends a cognitive model of reference understanding that gives a central role to the activation of the referents in the working memory of the speakers. This is also one of the conclusions arrived at by Hopper and Thompson (1984): in a wide variety of cases, over several languages, the articles and other nominal markers are assigned on the basis of discourse salience or preeminence rather than following strictly the logical status of the referent.

The strength of this functionalist model lies in the blending of a semantic and a discursive (pragmatic) factor: specificity (S) and presupposition (P). The four cases constitute a cognitive account of the use of articles, rather than a formalization of their "meaning", showing that articles cannot be translated simply from one language to another. However, in a language such as French, severe overlapping between article use tends to show that such a model must be refined to account for more than four cases. We believe also that the model should separate more clearly the objective referential properties from their cognitive representation (the quadrant representation amalgamates both).

We propose here a model based on a clear division between reality, representation and expression. The model suggests a cognitively plausible representation of the "referring cases" that an entity may occupy, then a link from representation to expression through articles. The model deals for the moment with non-plural entities.

The analysis of a variety of French noun phrases in context provides the basis to distinguish eleven referring cases (henceforth, RC), representing the real-world status of entities the speakers might refer to. Indeed, we do not believe that reality contains only here-and-now entities: an animal searching for a mate is really looking for any individual of the same species and opposite sex; a herd chased by predators flees one or more attackers. The RCs are part of a continuum, hence their boundaries overlap. They make implicit use of the (S) and (P) properties, but they are at this point defined using words, not logical symbols (their list is given in the Annex 1).

We are not looking for a mathematical representation of the RCs, but for a model of their cognitive representation in the minds of the speakers. We made an attempt to situate the RCs on the S/P axes (cf. Popescu-Belis 1999:185), quantifying the activation of the (S) and (P) features, but this did not help to find a better link between representation and expression; also, the cognitive bases of (S) and (P) are unclear. We developed instead a model inspired by G. Edelman’s studies about categorization and the difference between individuals and categories (Edelman & Reeke 1982). We suppose that the speakers are able to perceive and to represent the entities of the discourse, using two parallel pathways: the (I) pathway detects and represents the (fine-grained) properties of instances, while the (C) pathway detects and represents the category of each instance. Of course, the two representations, as well as the more elementary features that constitute them, are necessarily subject to binding and integration. Also, the (I) pathway is rather linked to a repertoire of proper names, while the (C) pathway is linked to common names.

The important point here is our proposal to represent the RC of an entity in terms of differential activation between a representation of instance features and a representation of category features. The total activation is the salience of the entity in the discourse. The part of the C/I space containing the possible values of activation is divided among the various RCs. For instance, the representation of a well-known unique instance, or one related to a known one, or of a new instance, corresponds to a high (I) activation ([UNIQ] > [1-KNOWN] > [1-REL] > [1-NEW], cf. definitions in the Annex). On the other hand, the representation of a category, or of "any instance", or of random instances, has a high (C) activation ([CATEG] > [ALL] > [N-RND]). Finally, [ZERO] and [PROP] cases have low activation on both scales. There are no clear-cut boundaries between the regions of the C/I space corresponding to the various RCs; it is also possible to assign to each larger region one of the labels from the S/P space or from Givón’s model.

Finally, we model the expression of RCs as a mapping from the C/I space to the set of available articles. The C/I space is divided according to the total activation: in French, the highest activation zone corresponds to a dominant use of the definite article, the next zone to the indefinite article, and the lowest activation zone to the use of no article. As we have shown in an experiment with real discourse (Popescu-Belis 1999:175-178), there is considerable variability, in French, in the use of articles. The mapping described above provides thus a propensity to use either of the articles, rather than a constraint. The role of conventional constructions, sometimes exhibiting little regularity, must not be underrated here.

The result of our model is thus a link from the real-world entities, or their low-level perception, to the related expressions of their referring case using articles; the link is mediated by cognitive representations. The relevance of the model will be studied through its implementation in a simple communicating agent.



Bickerton D. (1981) - Roots of Language, Ann Arbor, Mich.: Karoma Press.

Dale R. (1992) - Generating Referring Expressions: Building Descriptions in a Domain of Objects and Processes, Cambridge, Mass.: MIT Press.

Devitt M. and Sterelny K. (1999) - Language and Reality: An Introduction to the Philosophy of Language, Cambridge, Mass.: MIT Press.

Edelman G.M. et Reeke G.N. (1982) - "Selective networks capable of representative transformations, limited generalizations and associative memory", PNAS, 79, pp. 2091-2095.

Givón T. (1984) - Syntax: a functional-typological introduction, vol. 1, Amsterdam: John Benjamins.

Hopper P.J. and Thompson S.A. (1984) - "The discourse basis for lexical categories in universal grammar", Language, 60, 4, pp. 703-752.

Kamp H. and Reyle U. (1993) - From Discourse to Logic: Introduction to Model-theoretic Semantics of Natural Language, Formal Logic and Discourse Representation Theory, Dordrecht: Kluwer.

Popescu-Belis A. (1999) - Modélisation multi-agent des échanges langagiers: application au problème de la référence et son évaluation, PhD thesis, LIMSI-CNRS and Université de Paris XI (





Poster on display at the workshop (gzipped postscript)

Annex 1 – Definition of the referring cases in which a
non-plural non-proper noun or noun phrase may be found


the NP evokes a property of the category rather than an instance;


the NP is used to assert that a property is true for no instance of the category;


the NP evokes the unique and known instance of the category;


one instance, already introduced in the discourse;


an instance that is attached by association to an already introduced instance (that complements the NP, or appears in a relative clause);


a new instance;


a new instance that is still undetermined;


one and only one random instance of the category;


one or more random instances;


any instance or all instances;


the category itself, as an instance of a higher-order category.



Annex 2 – Examples of article use grouped according to the
estimated referring cases (Popescu-Belis 1999:181-183)


  1. Le chien est depuis longtemps l’ami de l’homme.
  2. L’occasion fait le larron.

  • « Un chien errant entra dans notre vie. … Nous l’appelions Bobby, comme il convient à un chien chéri. »
  • Paul est un docteur qui connaît son métier.

  • Marco Polo arrivait à pied depuis la Chine.
  • Le prévenu affirma sous serment qu’il ne savait pas tirer au pistolet.
  • Il fait un temps de chien.
  • Paul est docteur.
  • Trêve de philosophie ! Parlons concrètement.
  • Je l’obtiendrai de gré ou de force.
  • [zéro]

  • Jean n’a pas vu le moindre lion en Afrique.

  • Vraiment, on ne mettrait pas un chien dehors !
  • Il n’y avait plus un chien dehors.
  • Jean n’a pas vu un lion en Afrique.
  • Si Jean avait un âne, il le battrait.
  • Où veux-tu trouver une licorne ?

  • Jean n’a pas vu de lion en Afrique.
  • [uniq]

  • Le soleil se couche.
  • Ses yeux étaient verts comme la mer.
  • « Les animaux malades de la peste. »
  • Je vais d’abord imposer le silence.
  • « L’amour n’a plus de goût, non plus que la dispute. »
  • « Ces idées peuvent vous venir aussi à la table familiale, quand on enfonce la fourchette dans le rôti. »


  • Contentement passe richesse. Noblesse oblige. Prudence est mère de sûreté.
  • L’orateur imposa d’abord silence.
  • Trêve de philosophie ! Parlons concrètement.
  • [1-conn]

  • J’ai acheté un chien et un chat. Le chien est très amical.



  • Pierre s’est cassé la jambe.
  • Ulysse donna l’ordre de fermer les portes.
  • « Après l’apprentissage sur le segment, nous avons transféré Ève sur une droite. »
  • Jean est parti chasser le renard qui mangeait ses poules.

  • Pierre s’est cassé une jambe.
  • « Après un apprentissage sur un segment, nous avons transféré Ève sur une droite. »
  • Pierre n’avait pas vu une erreur qui s’était glissée.

  • …Le procès est fini. Accusateur et accusé ont été renvoyés dos à dos par le juge.
  • Ulysse donna ordre de fermer les portes.
  • « Après apprentissage sur un segment, nous avons transféré Ève sur une droite. »
  • [1-nouv]

  • Ma radio ne marchait plus. J’ai dû changer la batterie.
  • Je suis allé hier voir le docteur : tout va bien. Il m’a dit…
  • « Nul ne le vit débarquer dans la nuit unanime, nul ne vit le canot de bambou s’enfoncer dans la fange sacrée… » [première phrase d’un récit]

  • J’ai acheté un chien et un chat. Le chien est très amical.
  • Jean a chassé hier un sanglier.
  • « Un chien errant entra dans notre vie. … Nous l’appelions Bobby, comme il convient à un chien chéri. »
  • Paul lit un journal.
  • Ces idées peuvent vous venir aussi à la table familiale, quand on enfonce une fourchette dans un rôti.

  • Nous nous sommes mis à table.
  • Pierre lui a demandé pardon.
  • [1-indét]

  • « Ces idées peuvent vous venir aussi à la table familiale, quand on enfonce la fourchette dans le rôti. »
  • Paul lit le journal.
  • Il s’est assis au volant et a démarré.

  • Tu devrais voir un docteur.
  • Bill cherche une femme.
  • Jean est parti chasser un sanglier pour le repas de Noël.
  • Que serions-nous sans une femme ?

  • Que serions-nous sans femme ?
  • [1-aléa]

  • Le chien est depuis longtemps l’ami de l’homme.
  • « Le médecin peut réussir là où le prêtre échoue. »
  • Qui se sert de l’épée périra par l’épée.
  • L’occasion fait le larron.

  • Anne ne supporte pas de voir un chien dans sa maison.
  • Vous m’avez reçu comme un chien dans un jeu de quilles.

  • « La théorie de la fonction indicative défendue par Dretske réduit le lien entre concept et symbole à leur corrélation stable pour l’individu. »
  • La mère esquimau enveloppe son enfant dans une peau de renard.
  • [n-aléa]

  • Le chien est depuis longtemps l’ami de l’homme.
  • Jacques ne sait pas tirer au pistolet, et encore moins à l’arc.
  • Jean est parti chasser le renard.
  • « Vous abandonnerez au chien » le fruit de votre chasse. 

  • « Je suis épais comme un sandwich SNCF. »

  • Je sens une odeur de chien dans cette pièce.
  • [tout]

  • Le lion est carnivore, contrairement à la gazelle.
  • Le chien est un quadrupède.
  • Que serions-nous sans la femme ?

  • Une mère peut-elle ne pas aimer son enfant ?
  • Un chien est un mammifère.


  • Le chien fait partie de l’embranchement des vertébrés.




    Annex 3 – Poster on display at the workshop (gzipped postscript)