Navigation – Plan du site

Weak definites as bound relational definites

Francis Corblin
p. 91-122


Cet article introduit une proposition nouvelle pour intégrer les définis faibles (Carlson et Sussman 2005) dans une théorie globale de la définitude. Les définis faibles sont analysés comme définis relationnels, et comparés aux autres membres de ce paradigme : les définis « associatifs » (Hawkins 1978) et les définis relationnels complets à génitif (Barker 1993). La thèse principale de l’article est que les définis faibles ont un contenu relationnel intégrant des variables liées à l’intérieur de leur domaine propositionnel. Cette thèse conçoit la définitude faible comme une signification constructionnelle impliquant la proposition dans son ensemble et déclenchant un enrichissement du sens qui repose sur le quale télique (Pustejovsky 1995) du nom-tête.

Haut de page

Texte intégral

The present paper is an extension of some previous works devoted to a sub-class of weak definites called telic definites: Corblin (2011), Asic and Corblin (2012). I have benefited of very fruitful discussions with many colleagues on these topics and in particular M. Aurnague, T. Asic, C. Beyssade, and with all the participants to the workshop Déterminants et inférences in which the first version of this paper was presented. I am very grateful to them for their criticisms and suggestions.

I am particularly grateful to an anonymous reviewer of the paper. All the comments made by this reader bear on the key parts of the proposal, and my only regret is that I did not get enough time, space, and cleverness for addressing seriously all the issues raised in the review within the limits of this paper. I did my best, nevertheless, to make the proposal more explicit on all the points pointed to by this reviewer.

1“Weak definites” is a term used in the literature for two kinds of definite NPs: “long” definite NPs, exemplified in (1), and “short” definite NPs, exemplified in (2):


John got these data from the student of a linguist.

Poesio (1994)


Mary went to the store.

Carlson & Sussman (2005)

2The major reason why both NPs are called “weak” is negative: they do not have all the properties which are most often considered to be essential properties of (regular?) definite NPs.

3This paper argues that what these two categories of NPs have in common at a deeper level of abstraction is that they illustrate the same semantic category, namely the category of relational definites. In order to establish this, we show that the missing link from (1) to short weak definites of (2) is another kind of short definite which has been extensively studied in the literature, namely the “associative” (Hawkins 1978) or “functional” (Löbner 1985) short definite, illustrated in (3):


Ils entrèrent dans un bar et appelèrent le garçon.

They entered a bar and called for the waiter.

4Our main claim is that the difference between the weak definites of (2) and the associative definites of (3) is comparable to the difference between bound and free anaphora introduced in the 1980’s in the generative framework. Although the definite of (3) has to be licensed by some extra-clausal contextual anchor, the definite of (2) is licensed by referential anchors to be found within the clause-mate sentence itself.

5This means that weak definites should be considered to be definite NPs that are no less regular than associative definites, and they should be conceived, then, as a standard manifestation of definiteness. The main claims of the paper challenge the view suggested in Carslon and Sussman (2005) taking weak definite NPs to be a very special (if not marginal) use of definiteness. Moreover, in contrast to authors considering such definite NPs as generic or kind-denoting NPs, e.g. in the literature on English (Aguilar and Zwarts 2011 and Schwartz 2012), and in the literature on French (Furukawa 1986), I argue that weak definite NPs are no more generic than the whole class of definite NPs called “associative” in the pioneering work of Hawkins (1978).

The present work is based of French data.

1. Distinctive properties of short weak definites

  • 1 The presentation is based on Carlson and Sussman (2005). There is a long-standing interest for the (...)

6Carlson and Sussman (2005) sums up the properties leading to consider weak definites exemplified in (4) as exhibiting a behavior that is very different from regular definites.1


Mon fils est à l’école.

My son is at the-school.

‘My son is at school.’

7a. The identity of the individual school is not under discussion. Weak definites are fine when used “out of the blue”, with no previous information regarding the individuals and the situation referred to in the sentence, and it is very likely that no more information will be asked afterwards as to the particular school my son is in.

8b. In the continuation (5), it is not implied that my son and yours are in the same school:


Mon fils est à l’école et le vôtre y est aussi.

My son is at the-school, and yours-is-there too.

9c. Weak readings emerge only in some clausal contexts:

Only some nouns give rise to the relevant reading:


Pierre va à la banque.


Pierre va à l’immeuble.

Peter goes to the bank.

Peter goes to the building.

Only some embedding contexts trigger weak readings:


Pierre est à la banque.


Pierre est derrière la banque.

Peter is at the bank.

Peter is behind the bank.


Pierre consulte le calendrier.


Pierre déchire le calendrier.

Peter checks the calendar

Peter tears the calendar.


Mon père va à la banque.


Mon chien va à la banque.

My father went the bank.

My dog went to the bank.

10d. It is implied in the weak reading of l’école in (1) that the agent of the sentence is not only located to a particular place (a school), but is there in order to receive the kind of service a school is designed for, i.e. it is implied that my son is a school boy. Following Roberts (2003) and other authors like Aguilar-Guevara and Zwarts (2011) and Aguilar-Guevarra and Schulpen (2011) we will call this a meaning “enrichment” of the interpretation.

2. Full relational definites and associative definites

11Two other kinds of definite NPs should be compared to weak definites as distinguished in the previous section: associative definites (ADs) and full relational definites (FRDs).

12Associative definites are short definite NPs which are licensed by a contextual anchor indicating that the considered NP refers to an individual which is a typical part of a larger frame, or script. This kind of definite have been extensively studied in the literature on French under the heading “associative anaphora” (see a.o., Fradin 1984, Kleiber 1993 and Schnedecker et al. 1994), and in the literature on English as a case of “bridging” (from Clark 1975). ADs are exemplified in (3) above and (10):


Dans un village français, il est facile de trouver l’église.

In a French village, it is easy to find the church.

Full relational definites are definite NPs formed using the prepositions de, as exemplified in (11):


La mère de Jean est professeur.

The mother of John is a professor.


Le flanc du bateau était déchiré.

The side of the boat was broken.

13FRDs have also been extensively studied in the literature, including the literature on weak definites in the Poesio’s (1994) extension of the term (a sub-class of FRDs); Poesio himself draws a clear link between FRDs and bridging definites (ADs) based on their relational meaning.

14Are weak definites so special? Weak definites (WDs), as grouped together under §1, share properties with ADs and FRDs.

2.1 Weak definites/Associative definites

15Shared properties between WDs and ADs:

A. the head noun must be a relational noun, not a sortal one (See Löbner 1985);

B. both can introduce a new individual in the discourse (no familiarity required);

C. there is no strong constraint regarding uniqueness;

D. anaphora to such NPs is not necessarily strict anaphora (co-reference).

Non-shared properties between WDs and ADs:

E. ADs cannot be uttered “out of the blue”; WDs can.

F. ADs can be hosted by any kind of clausal environment. WDs cannot (see §1, c-property).

16The number and nature of the shared properties are a sufficient condition for taking both of them as special definites, if the standard of comparison is as usual, a notion of regular definite for which it cannot be the case that A-D hold.

17Some comments are in order about D and E. What D stands for is that ADs normally requires some extra information (not inferable from the host clause itself) indicating that the clause refers to a sub-part of a situation of a special kind in which, typically, the relational meaning of the head noun has at least one referent. For instance, in (3), the first proposition indicates that the discourse refers to a bar, and is crucial for licensing the waiter.

2.2 Weak definites/ Full relational definites

18If weak definites are compared to full relational definites (FRDs) like (1), we get the following results:

Shared properties:

G. WDs and FRDs can be used out of the blue; see (1).

H. WDs and FRDs come with some meaning enrichment which does not arise for their indefinite counterparts; see (4), and (9)/(10).

Non-shared properties:

I. FRDs can be hosted by any predication, WDs cannot; see (11)/(12).

J. The meaning enrichment mechanism of WDs and FRDs are not identical. See infra.

19A crucial shared property of WDs and FRDs is that they can be accepted and fully interpreted without any prior information regarding the eventuality and individuals referred to in the host clause.

20But as noted under G, WDs and FRDs will then come with some meaning enrichment which seems a reflex of definiteness, since it does not occur in the indefinite versions of the NPs.

Consider the contrast (13)/(14):

  • 2 La robe de Marie is in my view a FRD, and for me, the weak definites of the Poesio type are just a (...)


La robe de Marie était rouge.2

The dress of Mary was red.


Une robe de Marie était rouge.

A dress of Mary was red.

21In (13), it is implied that the FRD refers to the dress Mary was ‘wearing at t’, the time event of the sentence. If accepted, (14) refers to one of Mary’s dresses (a dress she owns, or a dress she once wore).

22And as already said for (4), WDs come also with some meaning enrichment: it is implied in (4) that my son is a school-boy.

The H-contrast is illustrated by (15)/(16) :


Le docteur de Marie était amoureux d’elle.

The doctor of Mary was in love with her.


Marie était amoureuse du docteur.

Mary was in love with the doctor.

23The initial NP of (15) is fine as a FRD, can be interpreted out of the blue, and preserves the most typical meaning enrichment for le docteur de Marie (Mary’s lover is her regular doctor) although it is associated to a property (‘to be in love’) having nothing to do with a doctor’s typical role.

24But in (16) the definite NP is not accepted as a genuine WD. It requires some special prior information: was it the doctor of the village, of the firm, a doctor previously mentioned?

25This contrast reflects the common observation that WDs are licensed only if their host sentence refers to a stereotypical activity related to the function or relation denoted by the definite NP’s nominal head.

26It leads to the idea that if FRDs rest on a semantic mechanism operating within the NP boundaries, and independent of their host sentence, WDs’ reading appears as a constructional semantic mechanism taking as input the definite NP and its host clause.

27The I-contrast reinforces this view. Works by Barker (1993) for English, and Milner (1982), and Corblin (1987, 2001) for French, have analyzed the mechanism leading to the contextual specification of the underspecified meaning of FRDs. The general idea is as follows: the lexical content of FRDs contains an under-specified relation R:

Docteur de Marie (x)

docteur(x) & R(x,m)

28The introduction of this free variables over relations is analyzed as being tied to the presence of the preposition de; it is not possible with any other preposition imposing a specific relation (for instance docteur près de Marie).

29Theory independent observations show that in FRDs the R relation can be contextually specified, by virtually any kind of relation provided by the context, as for instance in (17):


Marie dessinait une robe pour sa collection. La robe de Marie était rouge.

Mary was drawing a dress for her collection. The dress of Mary was red.

30Most French speakers I tested, when asked to interpret (17), say that the intended interpretation of la robe de Marie is ‘the dress Mary was drawing’, although it is not absolutely impossible that it denotes the dress Mary was wearing at the time of utterance. This gives arguments for analyzing de as introducing a free variable over relations, exactly as pronouns are analyzed by Montague as free variables over individuals. Free pronominal variables over individuals are either interpreted as constants recoverable in context, or as bound variables.

31In (17) the preferred interpretation picks out the relation between a dress and Marie as the intended content of the free variable on relations introduced by de. The analogue of this case for pronouns is the relation of co-reference: the pronominal free variable over individuals introduced by a pronoun is interpreted as denoting an individual mentioned in the close context.

  • 3 Lexical specification of free variables might be illustrated by some cases of “nominal anaphora”: i (...)

32The free variable R can also be specified on the basis of the lexical information provided by the NP (e.g. dress(x) & R(x,y), if y is a woman, = ‘x is a dress wore by y’). At face value, there is no close counterpart in the behavior of pronouns, i.e. cases in which a pronoun would be interpreted as a constant on the sole basis of its lexical content, instead of borrowing its reference from a contextual mention. But pronouns have a very weak lexical content, and this might be enough for explaining this impossibility.3

33The general idea about expressions used as free variables is that free variables can be used in human languages only when the intended substitution or binding of the variable is easily recoverable by the hearer. The mechanisms by which speakers interpret the intended free variable on relations introduced by de in robe de Marie include picking up a contextual relation, as in (17), or inferring that the intended relation is one of the most typical relations existing between individuals of the relevant type: woman/dress will derive for R the ‘wearing’ relation, human/material objects can also derive the ‘possessor’ relation.

34For FRDs this kind of lexically-based specification of R is inferred from the information provided by the NP itself. The host sentence has no influence on it.

  • 4 After all, the standard interpretation of “Pierre a pris l’escalator” can be worded as: “Pierre too (...)

35The view that WDs are also based on relational interpretations of their head looks sound,4 although they are short definites, and as compared to FRDs, they are deprived of any explicit lexical instantiation for the R relation and of any explicit indication of what the argument(s) of this relation are. But if the head of WDs is relational, it is sound to see its meaning enrichment as a specification in context of the underlying R-relation and as a contextual selection of the relevant argument(s) of the relation.

36Under this view, the fact that WDs are only licensed in restricted lexical contexts makes sense once it is admitted that WDs are interpretable if their host sentence supplies enough information for specifying an adequate (i.e. licensing the use of a definite NP) R-relation and providing its arguments.

And such an assumption would answer at once many general questions regarding WDs:

  1. Why can WDs be used out of the blue?
    Because the host clause is enough for providing a specification of R and of its arguments.

  2. Why are WDs restricted to some host clauses?
    Because not all host clauses provide the resources required for specifying the R-relation and finding an argument of the required sort.

  3. What is the difference with ADs?
    ADs find the relevant information for specifying R and its arguments in the discourse context; WDs find this information in the host clause.

  4. What is the difference with FRDs?
    FRDs find a part of the relevant information within the NP (selection of an argument), and the other part (specification of the R-relation) either in the discourse context, either in the lexical resources found within the definite NP itself.

In guise of illustration we offer the following contrasts:


Ce service médical est excellent. Le docteur est très réputé.

This medical service is excellent. The doctor is very famous.


Le docteur de Marie vient juste de se marier.

Mary’s doctor is just married.


Le docteur a dit à Marie d’arrêter de fumer.

The doctor asked Mary to stop smoking.

  • 5 To define doctor as a relational noun, means that if x is a doctor, x must have some relation to ot (...)

37The example (18) is a clear case of AD. The definite NP le docteur finds in the previous sentence a clue for considering that the situation referred to is a medical institution in which there is, in general, at least one doctor. The definite NP refers to a doctor playing this role in the medical service referred to in the first sentence. If we choose to assume that relational nouns like doctor5 always come with an unspecified R-relation, the resulting interpretation is roughly : doctor(x) & R(x,medical-service), and R is specified as the kind of relation associating a doctor to the service in which she is working.

38In (19), we have a case of FRD. The definite NP le docteur de Marie is a full relational NP specifying the genitive as denoting an argument of the R-relation. In absence of any contextual clue, the R-relation is supposed to be derivable from the lexical information associated to the NP: the argument being a person, R is specified as the relation between a doctor and her patient. An empirical observation about (19) is that R is most often interpreted by speakers as ‘the doctor Mary used to consult for her health’, i.e. her regular doctor at the time of the utterance.

39It is not obvious that (20) would be recognized as a WD by all authors dealing with the matter. It does not fit straightforwardly some criteria: for instance it occurs as a subject and many authors insist that WD are in general governed by a verb or a preposition (Aguilar and Zwart 2011), and it seems moreover that it refers to a very specific individual, and could not be confused with a generic or kind-denoting NP. I suspect that some scholars would say that it is more like an AD than like a genuine WD, and I agree that one may hesitate.

40What this shows is that there is not always a very clear difference between ADs and WDs, which gives argument for treating them as related phenomena in the theory.

41Now if we take strictly the distinctions made previously in this paper, we have arguments for claiming that the definite in (20) is a weak definite. The main argument is that a sentence like (20) can be uttered without any previous information, which means that it is the content of the clause itself, which helps to specify the relevant R-relation and its arguments. (20) is usually interpreted as ‘the doctor Mary consulted for her health at t, asked her to stop smoking in the course of this consultation’. It is worth noting that this interpretation is not identical to the interpretation of le docteur de Marie: it is not implied by (20) that the doctor is her regular doctor: it is only implied that this doctor is not any doctor but the one Mary consulted for her health, and that t, the time of the event ‘demander’, is a subpart of this consultation.

42What is clear, then, and requires some explanation, is that a FRD like le docteur de Marie, comes with a meaning enrichment (regular doctor of Mary) which is different from the one coming with le docteur in (20).

43What I conclude is that WDs are not so special, and they share many properties with other relational definites, ADs and FRDs: and the shared properties are precisely those which set the whole class apart from regular definites.

44If we take seriously the idea that WD are relational definites, many of their distinctive properties follow (see above). However, we are left with two issues.

45If WDs are a small part of a very large subset of definite interpretations, it is this large subset which has to be accommodated in the theory of definiteness. WDs are not a marginal use, but are representative of a typical use of the category.

46WDs should be accommodated as relational definites in such a way that what distinguishes them from ADs and FRDs should be explained.

3. Relational definites and the theory of definiteness

47Most, if not all, scholars assume that definite NPs form a natural class unified by a common semantic content. In other words, they think that definiteness can be defined, and this term covers more than a bunch of different cases sharing a family “resemblance” as Wittgenstein might have put it.

48In languages like French, with a morphological definite article, the same scholars assume that most, if not all, occurrences of the definite article convey the same meaning or are licensed by the same set of conditions of use.

49In a Russellian approach, a definite NP implies that its lexical content applies to one and only one individual. This view is not counterintuitive when applied to FRDs, although it encounters the problem of long weak definites of the Poesio type, in which the lexical content of the definite NP applies to more than one individual (see above).

50For short definites, the approach based on uniqueness is even more problematic, and theories based of familiarity (Heim 1982) look more in accordance with the linguistic intuitions.

51As a working hypothesis for trying to unify the two approaches, and for accounting for relational definites as a standard manifestation of definiteness, I suggest that definiteness consists in the presupposition that the hearer has enough resources for making the lexical content of the NP “functional” i.e. returning a single individual. I do not see this as an original proposition, but instead, as a way of formulating many ideas found in the literature, especially in the work of Hawkins (1978) and Löbner (1985).

52The familiarity use of definite NPs are uses in which, considering the set of entities made salient by the on-going discourse, the content of the definite NP returns one and only one entity satisfying the content.


Schematic representation of the familiarity use:

Given a subset of old discourse referents: {a, b, c, d…},

le x take for granted that only one of a, b, c, d, … satisfies x.

The definite NP le x returns this old (familiar) discourse referent.

53This use is based on a sortal interpretation of the definite head noun, and I suspect it is this use which is taken as the regular use of definite NPs.

54This use can be said to be “functional” because using the x takes for granted that the hearer, given the resources she has access to, is able to return one and only one individual. In order to work properly, this interpretative mechanism needs a pragmatic auxiliary device designed to restrict the interpretation domain of the definite NP.

55The relational use of definite NPs is even closer to the classical notion of function. The use of the takes for granted that the content of the NP can be, given the resources accessible to the hearer, interpreted as a function returning a single entity when taking other entities provided by the context as arguments.


Schematic representation of the relational use:

Le x takes for granted that x can be interpreted as a function-argument complex returning a single individual.

x can specify a relation and its argument (FRDs) or just a relation(ADs and WDs).

56In order to work properly, this interpretative mechanism must be completed by a pragmatic device called “pragmatic enrichment”. What pragmatic enrichment does is just specifying the free variables of the semantic content, if there are any, in such a way that the enriched lexical content becomes “functional”, i.e. returns a single individual. This process is constrained by the general rule associating definiteness to accessible resources: any pragmatic enrichment is supposed to be accessible given the resources of the hearer and preferred in virtue of general pragmatic principles.

3.1 Relational definites and relational meaning of nouns

57A crucial part of the present proposal is based on the notion of “relational interpretation” (vs sortal interpretation) of the definite’s head noun. A full discussion of the notion is beyond the scope of the paper, but some clarifications might be useful.

58In my view, a relational interpretation of a lexical noun is a way of using it. In the lexicon, nouns can be divided in three categories : relational nouns (e.g. mother, double,…), sortal nouns (e.g. stone, sand,…) and unspecified ones (e.g. school, train, dog…). Making a relational use of a noun (if possible) is using it for denoting individuals by means of the relation they entertain to other individuals.

59The thorny point is of course related to unspecified nouns but I cannot discuss it at length due to space considerations. For example, an anonymous reader of the paper makes the following comment : “it is not clear that WDs involve relational nouns. In what sense are store, school, train, relational nouns, but not mall, class, bus?” My observation is that the French equivalent of bus has WD readings (prendre le bus), and that if cours is the French equivalent of class, it has WD readings (aller au cours). Another comment of the reader is: “It is rather unclear why train is relational but car is not, and only train give rise to a weak interpretation. Thus one cannot understand why take the train can be a WD whereas take the car cannot”. Again, I must say that French data does not make this difference : prendre le train and prendre la voiture are used in French with WD readings of the definite NP. A conclusion might be that the relational uses of nouns are rather pervasive, especially for artefacts.

A side remark is that we should carefully distinguish two properties for nouns:

a. to accept a relational use, which means to be used as ADs or as FRDs, or as WDs;

b. to be used as WDs.

  • 6 A full discussion on “gaps” in the WD paradigm cannot be undertaken in this paper. It is clear that (...)

60I do not make any prediction that accepting a relational use implies occurring as WD. WD is conceived in my proposal as a constructional meaning, and more than the properties of the head noun is required for licensing a WD reading.6

3.2 Pragmatic enrichments and unicity

61The two uses of definite NPs (familiar/relational) rest on meaning enrichments which, in my view, are not of the same nature, although they play the same role.

62For familiar definites (Roberts 2003), the general idea is that the domain of possible referents is restricted to a subset of known individuals made salient by the discourse context. This is an instance of a more general phenomenon: contextual domain restriction for natural language quantifiers.

63For relational definites, the mechanism, as I see it, is different. Relational definites are associated to a free variable over relations, and what definiteness requires is interpreting this relation in such a way that it selects a single individual. This instruction is realized not by restricting the domain of possible referents to a subset of known individuals, but by choosing for R a functional content. For instance for la robe de Marie, I do not think that we can accommodate the data by extending the contextual domain restriction view, or by weakening the notion of familiarity (see the weak familiarity of Roberts 2003). The use of relational definites for future events like in “For this wedding, Mary’s dress will be white”, which are perfect FRDs, is a strong argument, for me, against the idea that some notion of familiarity and of restriction domain can accommodate relational definites.

64This conception of pragmatic enrichment for relational definites holds that it leads to a functional content, which may seem a paradox, since the hallmark of WDs (weak FRDs of the Poesio type or genuine WDs of the Carlson and Sussman type) is that there is more than one individual satisfying the content of the NP. But a correct description is just that the explicit literal content of the definite NP applies to more than one individual: what we claim is that definiteness requires the accommodation of an enrichment making this content functional. For instance ‘robe de Marie’ is an explicit literal content which is not functional : neither existence nor unicity is implied; but using la robe de Marie in an episodic sentence requires the accommodation of a specific relation between a dress and Marie such that a single dress is selected (see infra).

  • 7 I am grateful to an anonymous reviewer for insisting that such examples should be explicitly addres (...)

65Another known argument against proposals relying on unicity is based on examples like “John took the train to come here” which is fine even if John took more than one individual train.7 As we will see later, I do not think this a very strong argument, since relational definites are associated to a free variable on times, which can be bound by quantifiers of their context. What is assumed in the train example is just that at any time John took trains, he took an unique train, which is somewhat trivial, since it does not seem possible to take more than one train at a given time t.

3.3 An illustration

66Let us come back on (13) to illustrate the schematic representation (22). La robe de Marie is interpreted as a relational use. The content of the NP is unspecified as to R, but specifies one of R arguments’, namely Marie. The use of le (instead of the indefinite un) means that given the resources accessible to the speaker, she is able to turn up the R-relation to a function. Given what is known about dresses and women, and in the absence of any specific information, a pragmatic enrichment is triggered, which is the most accessible one making the relation functional: the process ends up as interpreting la robe de Marie as: ‘dress wore by Mary at t, t being the time reference of the sentence.’

67Short definites interpreted as relational take for granted that both the precise functional specification content of R and the identity of its argument(s) makes no doubt given the discourse context. Consider an AD case like le docteur in (18). No old discourse is accessible for licensing a familiarity interpretation, and docteur, then, is interpreted as a relational noun. What the use of le implies is that, given the resources accessible to the hearer, docteur can be enriched in such a way that a single individual will be returned. The initial sentence refers to a particular medical service. It is part of the resources available that in a medical service, in general, the role of doctor is played by one (or more) individual(s). So the hearer has access at once to a way of restricting R (R as the relation between a doctor and the service she is working in), and to find an argument for this relation (a particular medical service).


Ce service médical est excellent.

Le docteur est très réputé.

Medical servicei

doctor(x) & R(x, i)

68Some comments are in order. First, the interpretation of short relational definites (ADs) triggers a mechanism comparable to the interpretation of anaphoric pronouns, which searches the discourse context to find what are the intended arguments of the relation. This is probably why, in the literature on French, ADs are called “anaphores associatives”. And it is true that the considered arguments, in general, must be found in the close discourse context with constraints comparable to those restricting the antecedent of actual pronouns: the relation docteur, imposes constraints of the sort of potential antecedents required for being accepted as a missing argument of the relation: service médical is fine, but not, for instance boulangerie (bakery). And most of the information about these constraints is part of the lexical knowledge associated to docteur in the lexicon, hence an accessible resource.

69We will see later that considering these cases of so-called “bridging” as anaphoric (a free argument variable associated to the relational meaning is unified with a constant of the required sort found in the context) offers a way of contrasting ADs as free relational definites to genuine WDs, which are comparable to bound anaphora.

70The issue of uniqueness deserves also some comments. The existence of doctors as part of a medical service is only typical enough for being mentioned in both lexical entries, but it is not the case that medical services have typically one and only one doctor. The use of a definite NP the doctor might require from the hearer to accommodate that the medical service has only one doctor. But this is only one possibility.

71A second way of making the relational description functional is to appeal to a time variable and interpret : ‘the only individual playing the role doctor of the medical service i at t, t being the time of the event referred to in the sentence’.

72A third way of making the description functional is by admitting that the medical service may have many doctors and even that more than one can play the role doctor at t, and by relativizing the role doctor to a given patient. This is possible since we know that for a given patient p, and for a given t, if an individual plays the role doctor for p, typically there is no other such individual. So (18) can be naturally interpreted if the medical service is composed of many doctors who are active at t, provided that the considered doctor is the one that a figure of the story is consulting for her health at t.

73The whole mechanism is an example of what is described above as “pragmatic enrichment”. The general idea is that the use of a relational definite NP le x, assumes that the hearer has enough resources for making x functional by enriching the under-specified content in virtue of general pragmatic principles like:

a. if a relation is left underspecified and must be specified, consider that the specific context and what you can infer in the absence of any specific information are enough for finding the relation intended by the speaker;

b. if an argument of a relational meaning is not explicitly identified, it works as a free variable to be identified with an individual of the required sort mentioned in the context;

c. the process of specifying the intended relation and the process of finding its arguments are related: e.g. for docteur, a contextual mention of a medical service, makes accessible the relation between a doctor and a medical service.

74Although “pragmatic”, this enrichment is semantically driven, exactly as many kinds of anaphoric interpretations are. We make, more precisely, the hypothesis that definiteness triggers for relational head-nouns a special kind of enrichment, i.e. an enrichment which ends up as a functional description, i.e. a description applying to one and only one individual.

And this is a crucial difference with what happens e.g. for indefinite NPs using the same relational Ns.

75For long indefinite NPs like une robe de Marie (a dress of Marie), un docteur de Marie (a doctor of Marie), the R-relation explicitly introduced by de has to be identified, but there is no requirement that the relation should be specified in such a way that the lexical content become functional. On the contrary, indefinites come with a “plurality presupposition”. This is why une robe de Marie will not be interpreted as ‘a dress Mary is wearing at t’ the utterance time, but more likely as a dress owned by Mary or as a dress that she once wore. It is definiteness in la robe de Marie which imposes the selection of the most accessible pragmatic enrichment leading to a functional interpretation of the relation.

76For short indefinites counterparts of ADs and WDs like a doctor, a school, contextual identification of arguments for the relation is possible:


Ils entrèrent dans le service médical. Un docteur les accueillit.

They entered the medical service. A doctor received them.

77It is likely that most speakers would understand ‘a doctor of the medical service received them’. But I think it is only in virtue of a defeasible implicature, as confirmed by (23):


Ils entrèrent dans le service médical. Un docteur les accueillit. Mais il ne travaillait pas dans le service.

They entered the medical service. A doctor received them. But she/he was not working in the medical service.

This a clear difference with ADs, as shown by (24):


Ils entrèrent dans le service médical. Le docteur les croisa, #mais il n’était pas du service.

They entered the medical service. The doctor passed them. But she/he was not working in the service.

  • 8 This suggests that in short indefinite NPs the head noun is not obligatorily interpreted as relatio (...)

78For definites, thus, the pragmatic enrichment of the relation is not an implicature, as it is for indefinites, but a semantically triggered process.8

  • 9 “I will use the term pragmatic enrichment to describe this phenomenon, and assume that this is just (...)

79The way I conceive pragmatic enrichment looks slightly different from Roberts (2003)’s proposal, even if I owe the term to her. Roberts sees the kind of pragmatic enrichment associated to definiteness as an instance of (implicit) domain restriction.9 My intuition is that Roberts’ view fits well for the familiarity use of definite NPs, but is not the most adapted for relational uses. My view is that for relational definites, pragmatic enrichment is a kind of anaphoric process which replaces free variables (relations and arguments) by constants, either provided by the context or derived from general lexical knowledge.

80I will try to show, in the next section, that genuine WDs can be accounted for by mechanisms which are partly similar to those required by ADs; I do not think that there are so strict boundaries between the three varieties of relational definites considered in this paper.

81The process being seen basically as an anaphoric process, the claim I will try to substantiate is that genuine WDs are distinguished by the fact that they are bound within their host clause.

4. Weak definites as bound relational definites

82There are two major obvious differences between ADs and WDs noted in § 2 above:

E. ADs cannot be uttered “out of the blue”; WDs can.

F. ADs can be hosted by any kind of clausal environment. WDs cannot (see §1 c-property).

83The E property is well known and makes WDs the strangest kind of definite NP, since they can be used without any prior information. The second one is more complex to formulate adequately.

4.1 Restrictions on the host sentence of definite NPs

84It is striking that each individual WD occurs only with a very small set of predicates, and does not survive (as WD) any time an arbitrary predicate is substituted (see § 1 above).

85This leads to see WDs as a constructional interpretation in which definiteness plays a role, but which requires many other ingredients in the clause itself, since the interpretation process is context-free, in the sense that the whole phenomenon is entirely blind to its extra-clausal context.

86The constraints on the licensing predicate select in the end a very tiny set of licensing contexts for a given WD, and it might be the case that the smallness of the set per se plays a role in the phenomenon, or at least in its perception by speakers and scholars. A strong limitation to a restricted set of lexical environments, together with regular pragmatic enrichment, may explain that it is possible to see any complex predicate + WD as some sort of idiom. But the regularity of the phenomenon, i.e. the fact that there are many such complexes involving different WDs, is rather an argument for analyzing it as a constructional emergence, and not as a set of genuine idioms.

87The precise nature of the constraints on licensing predicates is not easy to formulate in a unified way. For WDs called “telic definites” in Corblin and Asic (2012), the constraint is that only very underspecified localization verbs and prepositions can be used (see the contrast (25)/(26):


Mon père est à la banque.

My father is at the bank. (WD reading preferred).


Mon père est dans la banque.

My father is in the bank. (WD reading ruled out).

  • 10 For such complex predicate + NP, the intuition is that it is the NP which contains the main part of (...)

88Some other WDs also use a very underspecified verb like prendre (take) like in prendre le train, prendre l’ascenseur. The combination of this under-specified verb prendre with any NP headed by theses nouns, including indefinites NPs, results in the meaning: ‘to use an individual of sort N for beneficiating of its typical function’. Prendre un train means : ‘to use a train for being transported by it’. Such cases, then, are not so different from telic definites: they are based on an underspecified predicate (e.g. prendre),10 which is interpreted, when taking an argument of the sort N, as implying that the agent of the verb uses as a beneficiary the telic function of the kind of entity denoted by N. Of course, one can look at, listen to, or draw a train: but a train is designed for being used by people and goods for their transportation. In the generative lexicon theory of Pustetjovsky (1995), this information would appear as a telic quale.

89I will not go into the details on this for space consideration, in particular, I will not try to show that this way of defining licensing contexts covers all cases of WDs. It will be enough for what follows to observe that the licensing contexts of many WDs are based on a very restricted set of lexical predicates which activates the telic quale of the head-noun of the WD complement.

  • 11 But see footnote 13.
  • 12 I use an indefinite in the presentation for commodity reasons although I will give arguments latter (...)

90As we will see soon, WDs, if conceived as relational definites, are precisely cases in which the “pragmatic enrichment” of the R-relation is the one which is derivable from the specific function of an object of the kind : to go to the bank, for instance, implies (at least) to go to11 an12 individual bank for using the specific function (money issues) a bank is designed for.

At this point, thus, and taking stock, we have three converging arguments for taking WDs as relational definites:

  1. WDs are not interpreted by familiarity, like relational definites (ADs and FRDs).

  2. WDs are blind to their extra-clausal context, as can be FRDs, a kind of relational definite.

  3. WDs emerge in lexical contexts activating the lexical telic quale (the typical role they are designed for) of their lexical head; they are thus based on a relational meaning.

91Sticking to the view that WDs are relational definites, but, as short definites, they are deprived of any explicit expression of their arguments (similar to ADs in this respect), it is expected that they trigger some anaphoric-like process for instantiating their arguments and the associated R-relation.

92This expectation is borne out, and again it points to the fact that WDs are bound within their own clause. Although ADs typically seek for their arguments in the extra-clausal discourse context, WDs take as arguments only individuals denoted in their clause.

This process is detailed for telic definites in Corblin (2011). Consider for illustration the contrast (27)/(28):


Mon fils est à la banque.

My son is a the bank.


Mon chien est à la banque.

My dog is at the bank.

93Only (27) licenses a genuine WD reading and (28) is a mere locative sentence in which la banque has to be either a familiarity definite, or an AD, specified on the basis of previous information. If banque is analyzed as a relational predicate, its underspecified representation is:


→ banque(x) & R(x,y,…).

94As usual we assume that the use of le takes for granted that the relation can be turned up to a function on the basis of accessible resources. What the lexicon provides is a telic quale defining a bank as something designed for dealing with their money issues by humans. The agent of the clause my son being human, it qualifies as a possible beneficiary of the kind of service provided by a bank, which is not the case for my dog. By a process rather similar to the one used by ADs on discourse context, the human agent is chosen as an argument for R specified as the telic quale of bank. As an intermediary step, on the way of making the relation functional, we get something like:

banque(x) & R(x, my son)

R(x,y) = using x for dealing with money issues for y.

95Is a relation so-defined a function? Of course not! There are many individual banks which can be used by a single individual for dealing with her money issues.

96So it is not only by relation to the argument my son that the above specification of the function can be made functional, but this is not a peculiarity of WDs. A comparable situation arises, for instance for la robe de Marie, even when specified as ‘dress wore by Marie’. The assumption we made then is that the definite is used only when there is in the available resources some regular means of accommodating what is needed for making the description functional. In the case of robe de Marie, we accommodate very often ‘dress wore by Mary at t, t being the time of the event, and the only reason for making this accommodation rather than anything else is just that we know that it is the best way, in general, to single out a dress with respect to a woman.

97In the case of (27), we do not only understand that my son is at an individual bank for dealing with his money issues, we understand, in addition, that the individual bank he is at is the individual bank which is the most relevant, according to him for doing so properly. And we understand this because we know that individual banks are more or less convenient for what we need.

98So I do think that (27) says much more than a mere indefinite interpretation: it says that my son selected an individual bank as being the most convenient for what he wanted to get, and was not in just any bank. And we could make the same comments when comparing, for instance, prendre le train (to take the train) to prendre un train (to take a train). I will come back on this contrast.

99A comparison with “Mon fils est à l’école” establishes that what we accommodate for making the interpretation functional is completely different from one lexical unit to the other. L’école will select, if my son is young, the R-specification: to-be-a-pupil-in(x,y). In this case, the mere knowledge that at a given t one is a pupil of one and only one school will be enough for making the relation functional.

4.2 WDs as bound relational definites

100If the specification in context of the arguments of short relational definites is conceived on the model of pronominal reference (a free variable is identified to a contextual antecedent, or is bound by a quantifier of the context), WDs behave as bound pronouns i.e. pronouns which must find their antecedent in their c-commanding domain. And they can be contrasted to ADs, which can find their antecedent in any part of the discourse context. This contrast between bound and free anaphors dates back to Chomsky (1981), and was replaced quickly by more sophisticated versions of the binding theory. It was used for distinguishing reflexive pronouns, but also some parts of idioms from referential “free” pronouns and applied to examples like “Johni lost hisi way” and “Johni is only interested by himselfi”.

My claim is that WDs are bound anaphoric expressions:

  1. WDs are anaphoric expressions: they are associated to a free variable on relations and free variables as arguments.

  2. WDs are bound within their clausal context:

a. their free-argument variable can only be interpreted as co-referential with an expression of their clausal domain: the individual referred to by the subject in most of our previous examples.

b. their free variable over relations can only be derived from the lexical content of the definite head-noun, and, in most cases discussed in this paper, from the telic quale of this noun.

101To formulate things simply: all short relational definite NPs instantiate the semantic content ‘N(x) & R(x,y,…)’; R is a variable on relations and R must be specified in context, together with its free variable argument: an individual must be find in the context satisfying the particular relation to x.

102For ADs, the specification of R and its arguments is free, since all of the discourse context can be explored.

103WDs are bound expressions, in the sense that the R-relation and its arguments must be found within the resources offered by the clausal domain.

104Note that this is the only way for explaining the fact that WD clauses, although they contain incomplete or unsaturated NPs like short relational definites, can be blind to their discourse context but nevertheless be fully interpretable. A short relational definite is an expression which refers to an x and says that x have some relation to something else of the right sort which helps to identify it. If the sentence is accepted without any help of its discourse context, it can only be that its free variables are bound.

I will now try to present informally this approach on the example (29):


Pierre est à l’hôpital.

Pierre is at the hospital.

  • 13 An anonymous reviewer mentions the sentence “He goes to the school at home” for a child being taugh (...)

105The noun hôpital is specified in the lexicon as having two interpretations: a concrete interpretation (a building) and a social institution interpretation; a hospital as a social institution designed for taking care of injured or ill people. It has doctors, nurses, medical services, etc. as its parts.13

106FRDs can use both interpretations, l’hôpital de cette ville will be interpreted as ‘hospital(x) & R(x, this-city)’; it will require accommodating that this city has only one hospital.

107L’hôpital de Pierre, if it is known that Pierre is a doctor, will be interpreted as: ‘hospital(x) & working-in(x, Pierre, at t)’.

108Now consider short definites, exemplified in (29). L’hôpital can be a familiarity definite. The discourse may have mentioned a given hospital, and the sentence, then, will not trigger any pragmatic enrichment as part of its interpretation construal. Suppose Pierre is a taxi driver and one of his colleagues is asked where Pierre is. She may answer (30):


Pierre attend un client à l’hôpital. Pierre est à l’hôpital, maintenant.

Pierre is waiting for a client at the hospital. Pierre it at the hospital, now.

109If l’hôpital is not a familiarity definite, it is a relational definite. The context may license an AD interpretation. It is required then that the discourse context contains some anchor licensing some specification of the associated R-relation: if it is known that Pierre is a surgeon, or if the previous sentence tells us that Pierre has been injured in an accident, etc.

110If the discourse context contains no such clue, the only accessible option leads to interpreting l’hôpital as a bound relational definite. As any definite, it assumes that given the resources accessible to the hearer she can make the description functional (denoting a single individual).

111This implies that hospital specifies the R-relation on the basis of its telic quale: a hospital is designed for taking care of injured or ill people. The free argument variable of this relation is co-indexed with Pierre since Pierre denotes an individual of the appropriate sort (human).

112Up to this point, the pragmatic enrichment of the description triggered by definiteness gives: hospital(x) & used-for-taking-care-of-as-injured-or-ill(x, Pierre).

113One may object that this description is not functional, since Pierre may be cured in different hospitals during his life; but exactly as for robe de Marie, the R-relation comes with a free variable over times which is bound by the quantifier out-scoping the main predicate (to be at), which gives : t x is-at(p,x,t) & hospital(x) & used-for-taking-care-of-as-injured-or-ill(x,p,t).

114It is otherwise motivated to postulate this free variable over times for explaining the dependence of (all) relational definites to quantifiers of their clausal contexts. Let us compare (31) to (32):


Aujourd’hui, la robe de Marie était rouge.

Today, the dress of Marie was red.


La robe de Marie était toujours rouge.

The dress of Marie was always red.

115In (31) an existential quantifier binds the time variable, which implies that la robe de Marie denotes a single individual, but in (32) the time variable is bound by a universal quantifier and la robe de Marie denotes a single dress at any t, but, since it is universally quantified, the interpretation ends up involving a whole set of dresses.

116Considering that, in (29) the main quantifier is existential, it binds the t variable introduced by the R-relation as shown above. And this binding is enough to make the relation functional, because we know that when an injured or ill person is cured in a hospital x, at a given t, there is no other such hospital.

4.3 (In)-definiteness and specificity

117In the above proposal, WDs are not generic, but refer to a specific individual. This explains why, in some cases, as in (33), a referential expression of the next sentence takes it as its antecedent:


Pierre est à l’hôpital, et cet hôpital est tout près d’ici.

Pierre is at the hospital and this hospital is very close.

118Explaining why this kind of anaphora to the specific entity involved is not that frequent, and why there is no special interest in most cases towards this individual entity is also straightforward: the specific entity is just a role holder and it is defined only as the hospital Pierre is being cured in at t. The main information conveyed by the sentence is that Pierre is ill or injured, not that there is a specific hospital hosting Pierre.

119In what respect do WDs contrast with their indefinite counterparts? It is important to be precise on this because there is a continuous tendency in the literature to see WDs are close to indefinites, if not indefinite themselves, while the present proposal takes them as genuine definites.

120It is difficult to use minimal pairs for all cases in French because the indefinite NPs cannot freely combine with the spatial preposition à which is required for telic definites (see for the restrictions on à Vandeloise (1987) and Aurnague (2004)). Let us contrast just the following acceptable pair:


Pierre s’est arrêté à une banque à son retour.

Pierre stopped at a bank on his way back.


Pierre s’est arrêté à la banque à son retour.

Pierre stopped at the bank on his way back.

121A consequence of the above proposal is that (34) is a logical consequence of (35).

122That Pierre intended to deal with his own money issues is part of the meaning of (35) but a mere implicature of (34). This can be shown by the following test:


Pierre s’est bien arrêté à une banque à son retour, mais simplement pour demander son chemin.

It is true that Pierre stopped to a bank on is way, but just for asking his way.


Pierre s’est bien arrêté à la banque à son retour, ?mais simplement pour demander son chemin.

It is true that Pierre stopped to the bank on is way back, but just for asking his way.

123In (36) a goal having nothing to do with the telic quale of bank (to ask for one’s way) is introduced, but the sentence is perfectly natural. In (37), the sentence is not natural unless some other contextual information licenses definiteness, either familiarity or uniqueness of the bank w.r.t. a given spatial domain (town, district…). In other words, in (37) the WD reading is no longer accessible.

124In addition, (37) conveys the meaning that the considered bank is for Pierre the most relevant bank for dealing with his money issues at t, and not any bank as conveyed by (35).

125Some contrasts between WDs and corresponding indefinites are more subtle to explain. For instance, consider the following one, showing that un autre (another one) is licensed by indefinite antecedents, but not by WDs:


Pierre a pris un ascenseur jusqu’au 15ème, puis un autre pour monter au 19ème.

Pierre took an escalator up to the 15th floor, then another one up to the 19th.


Pierre a pris l’ascenseur jusqu’au 15ème, ?puis un autre pour monter jusqu’au 19ème.

Pierre took the escalator up to the 15th floor, then another one up to the 19th.


Pierre a pris un train et Marie un autre.

Pierre took a train, and Marie another one.


Pierre a pris le train et ?Marie un autre.

Pierre took the train, and Marie another one.

126The first thing to note is that this kind of contrast confirms that WDs are not indefinites, but behave in this respect exactly as other short relational definites, since ADs behave exactly in the same way.


Pour payer, ils firent signe à un garçon, puis à un autre.

For paying, they called a waiter, then another one.


Pour payer, ils firent signe au garçon, ?puis à un autre.

For paying, they called the waiter, then another one.

127A somewhat theory independent analysis of the relation of another one to its anchor NP is as follows: the anchor refers to individual(s) of sort N, and another one refers to an individual of sort N, which was not referred to by the anchor.

128It is easy to explain the observation that indefinite anchors license another one: they refer to at least one individual N, and in addition come with the implicature that there is more than one N (the plurality implicature).

But there is more than one way for explaining that WDs do not license another one.

129A first option is that WDs do not because they do not refer to a specific individual N. I do not think that this hypothesis is correct because there are many other contexts in which we are forced to admit that a WD refers to a specific individual, since it can be picked up by a pronoun. See (44) and (45):


Marie prit l’ascenseur. Il était vieux et vétuste.

Marie took the elevator. It was old and slummy.


Ils firent signe au garçon. Il vint très vite.

They called the waiter. He came immediately.

130A second possible explanation is that a WD does not license another one because, as a genuine definite NP, it has no plurality implicature, and on the contrary, makes salient the fact that there is one and only one such N which is accessible in this discourse context.

131This may look at first a better explanation, but one might object that the McCawley’s example “The dog met another dog” casts doubts on it. In other words it is not completely impossible in general, for a definite NP the x to be in construction with an indefinite another x.

132So I think each of these options should be considered more carefully. There is possibly a difference to be made between two notions:

  1. Introducing an individual in the discourse so that it can be picked up by a referential pronoun.

  2. Introducing an individual N in the discourse so that it can be contrasted to other Ns.

133Definite NPs are typically deprived of the implicature of plurality associated with indefinites. When the information that there is many such Ns (The Poesio type weak definites) is lexically implicated, as in (46),


Pierre a réparé le pied de la table.

Pierre repaired the foot of the table.

the information that there is more that one such N is, so to speak, “back-grounded” by the use of le, the difference between individual feet being irrelevant, and one and only one foot is accessible for the discourse (the one which is being repaired), and another one is not licensed as shown in (47):


Pierre répara le pied de la table, ?puis un autre.

Pierre repaired the foot of the table, then another one.

134This looks to be a strong dynamic property of definiteness, which can be observed, thus, not only for WDs. Examples like (47) confirm that the non-licensing of another one is independent from the capacity of the NP to receive a specific interpretation and to license pronominal anaphora.

  • 14 In the same contexts there is a difference between anaphoric versions of another-NPs and complete o (...)

135Although we do not have a fully explicit analysis of the phenomenon to propose, we have shown that the non-licensing of another one cannot be an argument for holding that WDs are not specific NPs, and remains a strong argument for holding that WDs are not indefinite NPs. We suggest moreover that the non-licensing of another one is linked to the uniqueness requirement associated to definiteness. In other words, if we are correct, a speaker faced with a relational definite is asked to use resources she has for accommodating a description selecting one and only one individual (in the world), even though she knows as in (46) that there is a plurality of objects satisfying the description. Maybe this amounts, in such cases, to accommodate that one and only one of the objects of the sort is relevant for the discourse, the other ones, if any, being irrelevant. It looks awkward to ask her immediately to make as if there were more that one N relevant for the immediate discourse context.14

136As for McCawley’s example, it is important to note that it concerns familiarity definites. If this example is accepted as fully natural, we get a difference between familiarity definites (licensing another one) and relational definites (not licensing another one) which would remain to be explained.

4.4 On sloppy readings and bound definites

137A distinguishing property of WDs as opposed to regular definites was reported as D above: “anaphora to such NPs is not necessarily a strict anaphora (co-reference).”

138It deserves some comments, since, again, it might be interpreted as an argument for seeing WDs as close to indefinite NPs.

139There is a large class of anaphoric or elliptic constructions which can take an indefinite as antecedent, and can refer to an individual different from the antecedent.


Pierre prit un escalator, et Marie aussi.

Pierre took an escalator, and so did Marie.

And this a property which is not verified by familiarity definites:


Pierre lut le livre, et Marie aussi.

Pierre read the book, and so did Marie.

140The fact that WDs do accept to be antecedent of non-co-referring elliptic expressions might thus be interpreted in favor of the thesis that they are indefinite.

141But a closer inspection of related data reveals that WDs, but not indefinites, can be the antecedent of non-co-referring pronouns. This is illustrated in (50):


Pierre va à l’école. Marie y va aussi

Pierre goes to the-school, Marie there-goes too.


Pierre a pris le train pour aller à Anvers, Jeanne l’a pris pour aller à Lyon.

Pierre took the train to go to Antwerp, Jeanne takes-it for going to Lyon.

142It is easy to verify that the corresponding indefinites are only compatible with a co-referential reading of the anaphoric pronoun, which makes (53) awkward because we know it is impossible to use the same train for going to both destinations:


Pierre va à une école. Marie y va aussi.

Pierre goes to a school. Mary to-it goes too. Same school.


Pierre a pris un train pour aller à Anvers, Jeanne l’a pris pour aller à Lyon.

Pierre took a train to go to Antwerp, Jeanne takes-it for going to Lyon. Same train.

143A first conclusion is that some non-co-referring cases of anaphora prove that WDs are not indefinites.

144Going a step further, I will show that the existence of such data is a strong argument for the analysis of WDs as bound relational definites.

145If we search the literature for cases similar to (50) and (51) supra, what we find is the extensive literature of the 1980s on the sloppy identity reading of anaphoric referential pronouns started by the pioneering work of J. Ross (1967).

146The best known example of this reading is known as “the paycheck example”, and exemplified in (54):


The man who gives his paycheck to his wife is wiser than the man who gives it to his mistress.

147In (54) his paycheck is the antecedent of the pronoun it although, in the only plausible reading, it does not refer to the same paycheck; it is this reading that Ross called a sloppy identity reading.

148What is relevant for the present discussion on the nature of WDs is that the typical antecedent licensing a sloppy identity reading is a relational definite (his paycheck). In the terminology used in this paper, his paycheck is a FRD, since his instantiates the argument of the R-relation :

His paycheck(x)

paycheck(x) & R(x,y=his)

149The literature on sloppy identity pronouns establishes that some necessary ingredients for licensing the SR reading are:

  • a definite antecedent; no SR with indefinite antecedents;

  • a relational definite antecedent; no SR with familiarity definites;

  • a relational definite with an argument realized as a pronoun; no SR, for instance, for Pierre’s paycheck as antecedent?

  • a pronoun taking as antecedent a c-commanding NP of its own clause (the subject for instance); no SR if the possessive pronoun antecedent’s is found outside the clause.

For space consideration I will not illustrate these constraints.

150Now suppose the proposal defended in this paper is correct. It analyzes WDs as genuine definites (not as indefinites), and as relational definites. Of course, WDs differ from the classical paycheck examples because they are short relational definites, not FRDs. But I defended the view that short relational definites come with an implicit free-variable for the argument of the R relation, which has to be contextually specified. And I also claimed that WDs are bound expressions, in the sense that they have to find the free-variable argument within their clause.

  • 15 A full discussion of the issue of sloppy reading of pronouns is far beyond the scope of this paper. (...)

151If the proposal is correct, WDs realize all the necessary conditions for licensing sloppy reading of pronouns, and a prediction of the proposal is thus that the SR of pronouns should be licensed, a prediction that is borne out, and which is thus a strong argument in favor of the present proposal.15

152A further confirmation comes from the fact that the other kind of short relational definites distinguished in this paper, namely ADs, should, if we are correct, license SR for pronouns. Remember that ADs are, so to speak, free versions of WDs: they can find the argument of their associated relation in any part of the discourse context. A straightforward prediction is that they will license SR of pronouns if they find their argument within their clause. And again, this prediction is borne out as illustrated by (55):


Dans ce village, l’école est au centre. Dans celui-ci, elle est en périphérie.

In this village, the school is in the center. In that one, it is in the periphery.

153The antecedent of the pronoun elle, in (55) is the definite NP l’école , but the pronoun does not refer to the same school. In (55) the school is a relational definite, and it finds its argument within the clause (this village); it realizes, thus, the necessary conditions for licensing the SR of pronoun, which is actually observed in (55).

154It can be concluded, then, that the licensing of sloppy readings of pronouns, which is generally open by WDs and only by ADs bound within their clause, is a strong confirmation that the analysis of WDs as bound relational definites is on the right track.

5. Conclusion

155In this paper, I have given arguments for analyzing a whole set of definite NPs (including the varieties of “Weak Definites”) as relational definites, as opposed to familiarity definites. In general, and for reasons that are not completely clear, familiarity definites are taken as regular definites, which leads to considering the distinctive properties of relational definites as symptoms of “weakness”.

156By definition, relational definites do not require any acquaintance with the individual referred to, and resembles, in this respect to indefinites. The way they reach the capacity to refer to an individual is as the value of a function. They rest, then, on the general knowledge that some Ns have a relational meaning which can be turned up to a function if the free variables introduced by this relational meaning are identified with some constants that are contextually given or accommodated. Definiteness implies that the speaker has enough resources for retrieving the functional meaning intended by the speaker.

  • 16 The idea that WDs are generic-like NPs is defended by rather sophisticated theories that cannot be (...)

157This kind of designation of an individual may explain why some analyses of weak definites treat them as generic or kind-denoting. But the idea that weak definites NPs are generic or kind-denoting NPs16 is not at all necessary for explaining those peculiarities, and it is moreover a step one might hesitate to make for many reasons: it faces many empirical objections (WDs do introduces specific discourse referents), it leads to weaken the notion of kind-denoting or generic NP, and it should probably come with a “cut”; some but not all relational definites would be considered kind-denoting (neither associative definites nor full relational definites would be, I suspect). In contrast, the present proposal draws a very strong link between all these varieties.

158The paper emphasizes the “constructional” nature of the WD’s emergence: weak definiteness is an interpretation of a whole clause containing a definite NP. It imposes constraints on any part of the clause (subject, verb, preposition), and one of the consequences of the constraints is that the licensing verbs form a very small set, which may explain that weak definites are often perceived as quasi-idiomatic. But taking seriously the idea is not realistic considering the productivity of the construction, and again is not necessary.

159Some issues remains to be considered more at length. One of them is the relevance of the telic quale of the head-nouns of WDs. In all the examples considered, it is easy to reconstruct the specification of the underlying R-relation as derived from what is considered as the basic function, or use of objects of the sort. This is a neat difference between WDs and ADs which can be based on a large variety of relations. This issue requires more investigations and discussions.

160A final remark concerns the theory of definiteness and the dichotomy familiar/relational. The main working hypothesis of this paper is that familiarity definites and relational definites are somewhat different, although they can be conceived as two ways for a lexical content to be functional, i.e. to return a single individual. There are two main alternatives on the dichotomy: one can try to reduce it, by conceiving, for instance relational definites, as exemplifying what Roberts (2003) calls “weak familiarity” or symmetrically, as in Corblin (1987) to conceive familiarity definites as a special case of associative definites. The other alternative is to emphasize the dichotomy and to investigate the consequences of the differences.

161A test-case is accommodation. All theories accept that the lexical content of an NP can never in itself be enough for returning a single individual, which means that all theories of definiteness have recourse to the accommodation of restrictions allowing the lexical content to pick up a single individual. For instance for the guy to return a single individual, it must be accommodated that the domain of potential candidates is restricted to a very small set of individuals with a unique guy in it. The previous sentence applies rather well to familiarity definites, and the associated accommodation mechanisms have been studied extensively; they are generally conceived as pragmatic domain restrictions. But less is known regarding relational definites and the principles governing accommodation which are relevant for them. Is it true, as Roberts suggests, that they can also be conceived on the model of pragmatic domain restriction, or should we set up a rather different model for the pragmatic enrichment of relational definites? My own intuition is that enriching a relational meaning in such a way that it ends up as functional is a mechanism very different from the kind of domain restriction relevant for familiarity definites and I think that these issues might be worth addressing in further research.

Haut de page


Aguilar-Guevara, Ana; Schulpen, Maartje (2011). Understanding the Meaning Enrichment of Weak Definites. In Proceedings of the 2011 ESSLLI Student Session.

Aguilar-Guevara, Ana; Zwarts, Joost (2011). Weak definites and reference to kinds. In Li, N. & Lutz, D. (eds.) Proceedings of SALT 20: 179-196. Cornell University, Ithaca, NY: CLC Publications.

Asic, Tijana; Corblin, Francis (2012). Telic definites and their preposition(s): French vs Serbian. Paper presented at the workshop Languages with and without articles 2012, Paris.

Aurnague, Michel (2004). Les structures de l’espace linguistique : regards croisés sur quelques constructions spatiales du basque et du français. Leuven/Paris: Peeters.

Barker, Chris (1993). Definite Possessives and Discourse Novelty. In Canakis, C.; Chan, G. & Denton, J. (eds.) Chicago Linguistic Society 28: Papers from the Twenty-Eighth Regional Meeting, volume I: 26-41.

Barker, Chris (2005). Possessive weak definites. In Kim, J.; Lander, Y. & Partee, B. H. (eds.) Possessives and Beyond: Semantics and Syntax : 89-113. Amherst, MA: GLSA Publications.

Beyssade, Claire (2012). Weak definites between indefiniteness and genericity.  Talk given at the  workshop Determiners and Inference, Paris.

Birner, Betty; Ward, Greg (1994). Uniqueness, familiarity, and the definite article in English. In Proceedings of the Berkely Linguistic Society annual meeting 20: 93-102.

Bosch, Peter (2010). Weak definites and German preposition-determiner contractions. Talk given at the workshop Specificity from theoretical and empirical points of view, Institut für Linguistik, Universität Stuttgart.

Carlson, Greg; Sussman, Rachel (2005). Seemingly indefinite definites. In Kepsar, S & Reis, M. (eds.) Linguistic Evidence: Empirical, theoretical, and computational perspectives: 71-86. Berlin: de Gruyter.

Chomsky, Noam (1981). Lectures on Government and Binding. Dordrecht: Foris.

Clark, Herbert H. (1975). Bridging. In Schank, R. C. & Nash-Webber, B. L. (eds.) Theoretical issues in natural language processing: 169-174. New-York: Association for Computing Machinery.

Corblin, Francis (1987). Indéfini, défini et démonstratif. Construction linguistique de la référence. Paris/Genève: Droz.

Corblin, Francis (2001). Défini et génitif : le cas des définis défectifs. In Marandin, J-M. (ed.) Cahier Jean-Claude Milner : 19-54. Paris: Verdier.

Corblin, Francis (2011). Des définis para-intensionnels : être à l'hôpital, aller à l'école. Langue française 171/3: 55-76.

Epstein, Francis (1999). Roles and non-unique definites. In Proceedings of the 25th annual meeting of the Berkeley Linguistics Society (BLS 20): 122-133. Berkeley (CA): BLS.

Flaux, Nelly (1992). Les syntagmes nominaux du type le fils d’un paysan : référence définie ou indéfinie ? Le Français Moderne XL/1: 23-45.

Flaux, Nelly (1993). Les syntagmes nominaux du type le fils d’un paysan : référence définie ou indéfinie ? Le Français moderne XLI/2: 113-140.

Fradin, Вernard (1984). Anaphorisation et stéréotypes nominaux. Lingua 64: 325-369.

Furukawa, Naoyo (1986). L’article et le problème de la référence en français. Tokyo: Tosho.

Hawkins, John A. (1978). Definiteness and Indefiniteness. A Study in Reference and Grammaticality Prediction. Londres: Croom Helm.

Heim, Irene (1982). The semantics of definite and indefinite noun phrases. Doctoral dissertation, University of Massachusetts.

Kehler, Andrew; Shieber, Stuart M. (1997). Anaphoric dependencies in ellipsis. Computational Linguistics 23/3: 457-466.

Kleiber, Georges (1993). Anaphore associative, pontage et stéréotypie. Lingvisticae 1nvestigationes 17: 38-82.

Löbner, Sebastian (1985). Definites. Journal of Semantics 4/4: 279-326.

Milner, Jean-Claude (1982). Ordres et raisons de langue. Paris: Seuil.

Poesio, Massimo (1994). Weak definites. In Harvey, M. & Santelmann, L. (eds.) Proceedings of the Fourth Conference on Semantics and Linguistic Theory: 282-299. Cornell University, Ithaca, NY: CLC Publications.

Pustejovsky, James (1995). The Generative Lexicon. Cambridge, MA: MIT Press.

Roberts, Craige (2003). Uniqueness in definite noun phrases. Linguistics and Philosophy 26/3: 287-350.

Ross, John Robert (1967). Constraints on variables in syntax. Doctoral dissertation, MIT.

Schnedecker, Catherine; Charolles, Michel; Kleiber, Georges; David Jean (eds.) (1994). L'anaphore associative. Aspects linguistiques, psycholinguistiques et automatiques. Paris: Klincksieck.

Schwarz, Florian (2012). How weak and how definite are weak definites? Ms., University of Pennsylvania & Semantic Archives.

Stvan, Laurel S. (1998). The semantics and pragmatics of bare singular noun phrases. Doctoral dissertation, Northwestern University, Evanston, Illinois.

Vandeloise, Claude (1987). La préposition à et le principe d’anticipation. Langue Française 76: 77-111.

Haut de page


1 The presentation is based on Carlson and Sussman (2005). There is a long-standing interest for the weakness of some definite NPs as compared to “regular ones” in the literature on French: Milner (1982), Corblin (1987; 2001), Flaux (1992; 1993), Beyssade (2012) among others. For the literature on English, the issue has been discussed from different points of views in many works, and among others Birner and Ward (1994), Poesio (1994), Stvan (1998), Epstein (1999), Roberts (2003), Barker (in press).

2 La robe de Marie is in my view a FRD, and for me, the weak definites of the Poesio type are just a subset of FRDs (FRDs for which no lexically triggered enrichment can achieve unicity — eg. The corner of a busy intersection). So La robe de Marie is not a WD of the Poesio type.

3 Lexical specification of free variables might be illustrated by some cases of “nominal anaphora”: in French for instance, out of the blue, les grands (the tall ones) is interpreted as “the tall persons”, but can be interpreted as picking up a Noun of the context, for instance in: “En matière de formats, Goran Rakic préfère les grands” (as for formats, Goran Rakic prefers the great ones).

4 After all, the standard interpretation of “Pierre a pris l’escalator” can be worded as: “Pierre took the escalator he needed to go up at the time of utterance”, which is based on a relation of an escalator to Pierre. This relation is, moreover what we would find in the lexical definition of the noun escalator: an escalator should be defined not on the basis of what actual escalators are, or look like, but on the basis of what they are designed for.

5 To define doctor as a relational noun, means that if x is a doctor, x must have some relation to other individuals : x is doctor of some place, of some service, of someone, etc. This is not true for “sortal nouns” like stone or sand. See §3.1.

6 A full discussion on “gaps” in the WD paradigm cannot be undertaken in this paper. It is clear that even after the adjunction of the constructional requirements formulated below, the whole system “over-generates”. A theory independent observation is that in English for instance, any theory required for deriving the WD reading of to go the bank will derive a WD reading for to go to the school, which is not licensed. Of course, we guess that the existence in English of to go to school plays a role for ruling out this reading, but the theory explaining this intuition remains to be done.

7 I am grateful to an anonymous reviewer for insisting that such examples should be explicitly addressed.

8 This suggests that in short indefinite NPs the head noun is not obligatorily interpreted as relational, but can be interpreted as sortal as well. In contrast, for short definites, if not interpreted by familiarity, the head noun can only be interpreted as relational, which triggers the introduction of a free variable R and hence, the necessity to specify R in context.

9 “I will use the term pragmatic enrichment to describe this phenomenon, and assume that this is just an instance of the pervasive phenomenon of domain restriction in the interpretation of logical operators”. Roberts (2003).

10 For such complex predicate + NP, the intuition is that it is the NP which contains the main part of the meaning, whereas the predicate itself has a very weak content. This reminds the notion of “verbe-support” introduced by Maurice Gross and colleagues and the notion of “light verb” introduced by Jespersen.

11 But see footnote 13.

12 I use an indefinite in the presentation for commodity reasons although I will give arguments latter for analyzing this interpretation as a full definite.

13 An anonymous reviewer mentions the sentence “He goes to the school at home” for a child being taught at home and argues that it is not compatible with the claim that school triggers the relation ‘to be a pupil in (x,y)’. I do not think that the sentence does provide such an argument. The point of the example is based on the polysemy of nouns like school: place (a building), and institution (something devoted to teaching). What the example shows is that the weak reading is based on the ‘institution’ facet; by default, we infer that this institution is located in a building designed for its activities. But this is only a default as shown by the example. One may add, in the same vein, French expressions like hôpital à domicile (hospital at home). My analysis of the example is thus : ‘he goes to the unique school (institution) in which he is a pupil at t, and this takes place at home’.

14 In the same contexts there is a difference between anaphoric versions of another-NPs and complete ones. Anaphoric versions (exemplified in our previous examples) are worse than complete ones:

(i) Pierre a pris le train et Marie un autre train.
(ii) Pierre prit l’ascenseur jusqu’au 15ème, puis un autre ascenseur pour monter jusqu’au 19ème.

It might be the case that when an explicit noun is present, there is no doubt regarding the extension of the lexical material to be used for.

15 A full discussion of the issue of sloppy reading of pronouns is far beyond the scope of this paper. I am just using some necessary conditions licensing the reading for arguing that some NPs, including WDs are relational definites. For a synthesis of the literature see Kehler and Shiebert (1997).

16 The idea that WDs are generic-like NPs is defended by rather sophisticated theories that cannot be discussed in this paper: see for instance Aguilar-Guevarra and Zwarts (2011), Schwartz (2012). In the literature on French, Furukawa argues if favour of this view since Furukawa (1986). For a discussion, see Corblin (2011).

Haut de page

Pour citer cet article

Référence électronique

Francis Corblin, « Weak definites as bound relational definites », Recherches linguistiques de Vincennes [En ligne], 42 | 2013, mis en ligne le 01 mars 2016, consulté le 20 février 2017. URL : ; DOI : 10.4000/rlv.2184

Haut de page


Francis Corblin

Université Paris-Sorbonne, Institut Jean Nicod (CNRS-ENS-EHESS) & IUF

Haut de page

Droits d’auteur

© Presses universitaires de Vincennes

Haut de page