SciELO - Scientific Electronic Library Online

vol.17 issue28Editorial PresentationO espírito histórico-filosófico do pensamento hispânico author indexsubject indexarticles search
Home Pagealphabetic serial listing  

Services on Demand



Related links

  • On index processCited by Google
  • Have no similar articlesSimilars in SciELO
  • On index processSimilars in Google


Discusiones Filosóficas

Print version ISSN 0124-6127

discus.filos vol.17 no.28 Manizales Jan./June 2016 

DOI: 10.17151/difil.2016.17.28.2

An assessment of the argument from convention

Una evaluación del argumento de la convención

Andrei Moldovan*
Universidad de Salamanca, España,


Recibido el 31 de Marzo de 2016, Aprobado el 30 de Mayo de 2016


This paper focuses on what is known in the literature on the semantics and pragmatics of definite descriptions as "the argument from convention". This argument purports to show that referential uses of definite descriptions are a semantic phenomenon. A key premise of the argument is that none of the pragmatic alternatives (any one of a variety of Gricean accounts of referential uses) is successful. I argue that no good reason is offered to support this claim. I conclude that the argument from convention fails to be compelling.

Key words

Ambiguity, convention, definite descriptions, Michael Devitt, reference.


Este artículo se centra en lo que se conoce en la literatura sobre la semántica y la pragmática de las descripciones definidas como "el argumento de la convención". Este argumento pretende demostrar que los usos referenciales de las descripciones definidas son un fenómeno semántico. Una premisa clave del argumento es que ninguna de las alternativas pragmáticas (variedades de las aproximaciones griceanas a los usos referenciales) es exitosa. Sin embargo, no se ofrecen buenas razones para apoyar esta afirmación. Concluyo que el argumento de la convención no consigue ser convincente.

Palabras clave

Ambigüedad, convención, descripciones definidas, Michael Devitt, referencia.

Referentialism about definite descriptions

A classic debate in natural language semantics concerns the correct account of the meaning of definite descriptions (DDs henceforth). While the Russellian quantificational theory was the standard account for a long time, it has been the subject of intense criticism. In particular, Donnellan's ("Reference") distinction between referential and attributive uses of DDs is the starting point of a heated debate. Donnellan's distinction is usually formulated in a framework of structured propositions. Let me briefly introduce it with the help of sentence (1):

1. The woman wearing a white watch is a spy.

Someone might use (1) to communicate something about the person who is easily identifiable in the context as the woman wearing a white watch. In this case the DD is used referentially, and it is "merely one tool for doing a certain job—calling attention to a person or thing" (285). Alternatively, the speaker might have no individual in mind, and might intend to convey the general proposition that there is a unique woman wearing a white watch in the context of utterance and that she is a spy. In this case, the use is attributive.

The distinction in itself is not a semantic one (i.e., it does not directly concern literal meaning), but concerns the way these expressions are normally used. While Donnellan carefully avoids drawing semantic conclusions from his observation concerning the two uses, others have argued that his distinction is semantically relevant. Referentialism is the thesis that DDs (the definite article, to be more precise) are ambiguous, having two linguistic meanings. If used with the referential meaning, the contribution of the DD to the proposition literally expressed is the object the speaker has in mind. This linguistic meaning accounts for referential uses of DDs. If used with the Russellian meaning, the use is attributive and the contribution of the DD is a quantifier. Peacocke ("Proper"), Reimer ("Donnellan's"), Devitt ("The Case") are among those who have defended the Referentialist thesis. For instance, Reimer writes:

My view... is that a referential utterance of the form The F is G expresses a singular proposition provided the intended referent satisfies the linguistic meaning (the 'sense') of the definite description: provided it is the (contextually) unique F. In cases where this is not met, a singular proposition may well be communicated, but no proposition (singular or general) will be literally expressed. (93)1

The defenders of a unitary semantic account of DDs argue that Donnellan's distinction offers no reason to abandon the Russellian analysis of sentences with DDs. Grice ("Vacuous"), Kripke ("Speaker's"), Neale (Descriptions), Bach ("Referentially") and others consider that referential uses are to be accounted for pragmatically. They advance varieties of a Gricean pragmatic account of referential uses of DDs. For instance, according to Neale's (Descriptions 81) version of the account, the literal meaning (semantic content) of an utterance of the form 'The F is G' is a general quantifier proposition, but, in the right circumstances, the speaker might mean and intend to convey a singular proposition about the unique individual that fulfils the description. The singular proposition is the content of a generalized conversational implicature (or GCI). As opposed to the particularized ones, which depend heavily on the details of the context of utterance, GCIs occur by default, or depend much less on the context (Grice, "Logic" 37). Other examples of GCI are the implication of unfamiliarity of certain uses of indefinite descriptions (as in 'I saw John talking to a man'), or the temporal use of 'and' (as in 'John grabbed the microphone and started to talk').

The argument from convention

Much of the debate between Referentialists and Russellian-Griceans2 focuses on what is known as "the argument from convention3". The argument from convention appeals to considerations concerning the use of linguistic expressions to draw a conclusion about the semantic properties of these expressions. It has received a lot of attention in the literature, and has been the focus of a long debate between Michael Devitt and Kent Bach. It also earned a reputed victory: Stephen Neale, one of the main defenders of the Russellian-Gricean approach in Neale (Descriptions), abandoned the view due of the argument from convention. He writes that the fact that referential uses are common, standard, and cross-linguistic undermines their "standard, wooden, Gricean, explanation" ("This" 173). However, others have shrugged their shoulders. For instance, Elbourne (Definite) concludes his discussion of the argument with: "I find little to choose between the Russellian theory augmented by Gricean manoeuvring and Devitt's ambiguity theory" (109). The argument from convention is the subject of current debates also because it has wider implications concerning the methodology of natural language semantics and the distinction between semantics and pragmatics. Therefore, it is not without interest to carefully assess its merits and flaws.

But what is the argument exactly? What are its premises and what is the conclusion? This is one of Devitt's formulations of it:
The basis for the thesis that descriptions have referential meanings is [...] that we regularly use descriptions referentially. [...] Furthermore, there is no good pragmatic explanation of this regularity. The best explanation is a semantic one: there is a convention of using descriptions referentially. (Devitt, "What" 107-108)
Starting from this formulation we could reconstruct the argument as follows:
i. DDs have a standard and regular referential use.
ii. There is no good pragmatic explanation of this regularity.
iii. If an expression has a standard and regular use then this is strong evidence that the use is literal.
iv. Therefore, referential uses of DDs are best explained as literal uses; in particular, there is a linguistic convention of using descriptions referentially.

Probably other premises are implicit in the argument, such as a premise saying that any linguistic explanation of phenomena relating to meaning is either pragmatic or semantic. With this additional premise, and given (ii), we obtain a disjunctive syllogism, and the conclusion that referential uses of DDs are a semantic phenomenon (i.e., they are literal) follows necessarily. However, this does not make the entire argument from convention valid, as the conclusion not only says that referential uses are literal, but also that there is a linguistic convention of using descriptions referentially. But an account of referential uses of DDs might take them to be a semantic phenomenon without postulating a special linguistic convention, and without subscribing to the thesis of the ambiguity of the definite determiner. Neale ("This") has shown how this is possible. According to Neale ("This" 171-173), the truth-conditions of an utterance of a sentence of the form 'The F is G' where the DD is used referentially are: true if and only if there is a unique x such F(x) & x = a, and x is G (where a is the individual the speaker refers to).

On the other hand, (i), (iii) and (iv) do not have the form of a modus ponens argument, and do not make the argument valid either. This is because the consequent of the conditional is that regular referential use of DDs is strong evidence for literal use. As the forthcoming discussion of premise (iii) shows, the textual evidence suggests that this is the best way to reconstruct it. However, this is not exactly what the conclusion of the argument says. The conclusion of the argument is that Referentialism is the best theoretical option.

In conclusion, the argument from convention is not a valid argument. Instead, it is best reconstructed as an argument to the best explanation of referential uses. But then the force of the argument depends on what are the alternatives to Referentialism that we consider. The only alternative explicitly considered in premise (ii) is the pragmatic one. Is Referentialism a better alternative than a pragmatic account? In what follows I consider the premises one by one and I finally conclude that the argument is not compelling.

Premise (i)

There are good reasons to accept premise (i). Both Reimer and Devitt argue that, although other expressions such as quantifiers formed with 'every', 'some', 'both', but also indefinite descriptions, have referential uses, only in the case of DDs this use is regular and standard. As Reimer notes, "no other quantifiers are standardly used as referring expressions viz., to communicate object-dependent propositions" (96). Moreover, referential uses are probably more frequent than attributive uses of DDs. Devitt writes that "the vast majority of uses of descriptions are referential" ("What" 108). This seems intuitively plausible. However, Schoubye (526) warns against making claims about statistical frequency simply on the basis of one's own intuitions.

This is not good methodology in supporting claims that have a sociological dimension. I know of no empirical study that proves that referential uses are more common than attributive uses. Nevertheless, there is indirect evidence in this sense coming from experimental research in cognitive science. According to a study mentioned in Gibbs (57), people take longer to understand attributive uses, and to identify them as such, than referential uses of DDs, a conclusion which suggests that we are more familiar with referential uses, and so that they are more common, or at least very common.

Premise (iii)

The discussion of premise (ii) is central to this paper, so I leave it for the next section. Concerning premise (iii) there are differences between Reimer's and Devitt's respective formulations of it. Devitt ("Meanings" 126; "The Case" 283) maintains that regular use constitutes "strong evidence" that there is a linguistic convention. He also writes,

What are the conventions that constitute the system? How do we answer this key question? [...] we look for evidence from regularities in behavior (usage). Is this expression regularly used to express a certain speaker meaning? ("What" 106-107)
Mark Sainsbury formulates what seems to be the same idea as follows: An expression standardly and conventionally used on occasion with referential intentions is properly counted as a referring expression (as used on that occasion). The connecting principle is that semantic theory should reflect how the expressions it treats are used (182).
In turn, Reimer is more moderate: she notes that if an expression is standardly used as a referring expression "that surely suggests that it has at least one interpretation according to which it is a referring expression" ("Donnellan's" 97, emphases added). But she adds that "standard use is no guarantee of literal use" (95), as there are cases of standard use that are not literal: "While sentences of the form Could you do x? are standardly used to mean Do x, such is not their literal meaning, which concerns a query as to the hearer's ability to do x". (95)

Leaving aside the subtle differences, the authors mentioned maintain that there is an important connection between standard and regular use, on the one hand, and literal use, on the other. This is not only an intuitive claim, but also one that Reimer argues for on the basis of considerations relating to the phenomenon of lexical change. Reimer focuses on dead metaphors, which are expressions such as 'world wide web', 'kidney bean' or 'seeds of doubt'. These expressions had a metaphorical use at a certain moment in the course of evolution of the language, but when the metaphors "died", their previous metaphorical senses became new lexical meanings of the expression. As Reimer (97-98) explains, the verb 'to incense' was once used with the meaning to make fragrant with incense although it is now more commonly used to mean make very angry. The latter use is no longer metaphorical. However, the only change the relevant use of this expression has suffered is that it has become standard. And so the only plausible explanation of the fact that the previous metaphoric use is now a literal use is that standardization of use led to literality in these cases.

In reply to this claim, Griceans point out that if regular use is systematically explained as conventional the theory overgenerates predictions of literality. Bach ("Descriptions" 227-228) argues that such a theory rules out a plausible pragmatic explanation of certain phenomena such as scalar implicatures (e.g. the use of 'some' to mean some but not all), or the temporal and the causal use of 'and'. Schoubye (523-525) notes that the referential use of quantifier expressions such as 'every F' is also frequent and regular, but it is implausible to suppose that the correct account is semantic rather than pragmatic. If standard and regular use were sufficient to support the conclusion that that use is literal, then, by analogy, the Referentialist about DDs should run similar arguments for referential uses of quantifiers such as 'every' or the temporal use of 'and', and conclude that they are also literal. After all, it might very well turn out that the frequency of the temporal use of 'and' is equally high, or even higher, than that of the simple truth-functional use.

So, the Gricean argues that considerations concerning regularity of use are not sufficient to support the conclusion that referential uses of DDs are literal. Instead, Griceans such as Kent Bach argue that the effect of frequency and regularity of certain uses of expressions on their interpretation is "standardization". He explains standardization as a case of inference "compressed by precedent":

The familiarity of the form of words, together with a familiar inference route from their literal meaning to what the speaker could plausibly be taken to mean in using them, streamlines the process of identifying what the speaker is conveying. ("Standardization Revisited")

Standardization is a pragmatic phenomenon, and not a semantic one, closely related to that of GCIs. In Bach's view, referential uses of DDs are a case of standardization, rather than conventionalization.

Premise (ii)

If an explanation of referential uses of DDs that appeals to standardization or to GCIs is a live alternative, then it is one that needs to be rejected by the Referentialist in order to establish her claim on the basis of the argument from convention, which is an argument to the best explanation of these uses. We see now why premise (ii) of the argument is important: precisely because it aims to rule out such a pragmatic explanation of referential uses. So, for the argument from convention to support its conclusion, it is necessary to offer a defence of premise (ii).

There are formulations of the argument from convention in which premise (ii) is not mentioned as part of it. In Reimer ("Donnellan's" 97) and Devitt ("The Case" 283) the argument aims to support the conclusion that referential uses are conventional simply on the basis of their standard use, and the pragmatic account is discussed only as a reply that the Gricean might give to the argument, and not as part of the original argument4. But without premise (ii) the argument from convention is considerably weaker. That is because the evidence offered in support of the Referentialist thesis is compatible with pragmatic accounts as well, as we have seen, so the Referentialist conclusion is not warranted. As a result, I think it is more charitable to focus on the version of the argument from convention that does include premise (ii).

Let me now turn to the argument in favour of premise (ii) that Devitt offers.

Devitt's argument for premise (ii)

Premise (ii) of the argument says that there is no good pragmatic explanation of the regularity of the referential use of DDs. However, in his effort to establish this point Devitt does not consider all the pragmatic accounts available in the literature. For instance, he does not discuss contextualist approaches, such as the one in Recanati ("Contextual"). This fact in itself is sufficient to conclude that premise (ii) has not been established. I discuss in what follows the pragmatic accounts of referential uses that Devitt does consider, i.e., varieties of the Gricean one, and I argue that he fails to provide good reasons that they are unsuccessful.

Devitt rejects the Gricean alternatives to Referentialism for reasons similar to the ones invoked by Bach, Schoubye and others to reject Referentialism, namely, for an alleged overgeneration problem. In "The Case" (284-285), Devitt argues that the Gricean approach, unless it is significantly modified, falsely predicts non-literality about uses of expression that should clearly be treated as literal. He starts from Grice's claim that "[t]he presence of a conversational implicature must be capable of being worked out" ("Logic" 31), even if it is grasped intuitively. Devitt ("Referential" 13) then points out that if the requirement is that we as theorists must be capable of working out a derivation, it is too weak. After all, for all dead metaphors such a derivation can be imagined. Even after a metaphor dies and no speaker relies any longer on inferential processes in interpreting it, a Gricean derivation of the new meaning from the old one is still available. On the basis of this interpretation of the derivability requirement, together with Grice's "Modified Occam's Razor" (MOR, henceforth), which reads that "Senses are not to be multiplied beyond necessity" (Grice "Further" 47), it looks like the Gricean must conclude that there are no dead metaphors. The Gricean is committed to saying that metaphors are always alive and are to be explained pragmatically. But this is absurd, as a great part of our vocabulary is made up of metaphors that are now lexicalized. It is even worse, Devitt suggests, as the Gricean is committed to the claim that we should never postulate a new meaning of a word corresponding to a particular use if that use can be derived pragmatically. If intended meanings "can all be derived from assumptions about the context and other minds" ("The Case" 284), then "there are no conventional meanings at all" ("Referential" 13, emphasis in original), and all communication is to be explained pragmatically5.

In order to avoid this radical consequence, Devitt argues, the Gricean might interpret the derivation condition as saying that speakers must have the competence to provide a derivation. And she might argue that after a new use of an old word becomes standard, speakers lose their ability to derive the new use from the old meaning. This is what Bach ("Standardization vs." 683) suggests. But this will not do, as this lack of knowledge is, in Devitt's words,

a contingent and seemingly irrelevant matter of ignorance. Suppose that there was a massive breakthrough in the study of the origins of English so that we could now all recreate the original derivations. That would surely not eliminate conventional meanings! ("The Case" 284)

If the distinction between semantics and pragmatics is dependent on the information that speakers have, "we could make semantic meanings pragmatic simply by removing ignorance of their original derivations" (285). In order to avoid the fundamentalist position, Devitt argues, it is not enough that the theorist, or the speaker himself, be able to provide a Gricean derivation of the new use. "We need the speaker and hearer to be somehow involved with the derivation" ("Referential" 15).

So, the argument goes, the Gricean condition of derivability must neither refer to the possibility of such a derivation, nor to the speaker's competence to perform it, but instead must describe an actual process of interpretation. Devitt argues that whether a use is literal or not depends not on whether the users can derive the meaning pragmatically, but whether they (both speaker and hearer) actually do so. Grice's condition of derivability initially reads that the "speaker thinks... that it is within the competence of the hearer to work out, or grasp intuitively," ("Logic" 31, emphasis added) that q is required in order to make his saying that p consistent with the presumption that he is observing the principle of cooperation and the maxims (where q is the content of the implicature). In Devitt's modified version, it says that "the speaker thinks or grasps intuitively that the hearer will work out or grasp intuitively" (Devitt, "Referential" 16, emphasis added) that q is required etc. The derivation, that is, must have psychological reality, although, he clarifies, this does not mean that such processes are "conscious rational ones "in the central processor" (15). Instead, he suggests that the speaker's and the hearer's "speedy subconscious process "mirrors" the rational process of a Gricean derivation" (15).

A consequence of this formulation is that the speaker and hearer must distinguish p from q themselves (that is, what is said from the implicatum) in the process of communication. But, in the case of referential uses of DDs speakers do not intuitively distinguish what is said from the implicatum. The object-dependent proposition seems to be arrived at directly, and not by a derivation of the intended meaning by appealing to norms of rationality and cooperativeness. Communication with definite descriptions used referentially is as direct and immediate as that involving indexicals, demonstratives, proper names and other paradigmatic devices of direct reference6. Therefore, the derivation requirement, understood in the way Devitt proposes, is not fulfilled for referential uses of DDs. Given that, according to Grice ("Logic" 31), this is a necessary requirement that any conversational implicature must fulfil, the conclusion is that referential uses of DDs are not a Gricean pragmatic phenomenon.

To sum up, Devitt argues that the Gricean account of the derivation of implicatures is incompatible with the lexicalizations of metaphors. To avoid this overgeneration problem, Devitt argues that the Gricean needs to psychologize the derivation condition on implicatures. The psychologized derivation criterion is not fulfilled by referential uses of DDs. Therefore, these uses do not involve the generation of an implicature, contrary to what Neale (Descriptions) and others Russellian-Griceans argue. So, such a pragmatic account fails. This argument is meant to refute any Russellian-Gricean account of referentially used DDs, be it one that invokes particularized conversational implicatures, or one that involves GCIs, as both these implicatures must be derivable7.

This argument aims to establish premise (ii). Given that the other premises are acceptable, as I argued above, the strength of the argument from convention ultimately depends on whether Devitt's criticism of the pragmatic approach is successful. I think it is not. In the next section I argue that Devitt's criticism fails to be compelling. However, my purpose is not to defend the Russellian-Gricean approach to the semantics of DDs8. It is only to show that the argument from convention in support of Referentialism fails to establish its conclusion.

No overgeneration worry for the Gricean

Devitt's argument for the need to psychologize the Gricean derivation of implicatures is based on the claim that the Gricean has an alleged overgeneration problem. But she does not. The Gricean does not hold that derivability is a sufficient condition for the correctness of a pragmatic explanation of the use of an expression. Grice's view is much more moderate and balanced. To begin with, there are other necessary conditions on conversational implicatures, such as cancellability and nondetachability. But the crucial point here is that the standard methodology of truth-conditional semantics relies primarily on data from competent speakers' truth-value judgements to test hypotheses. This is a basic methodological observation, at least since Carnap ("Meaning"). It is not at all clear how the project of truth-conditional semantics could be pursued if such data is systematically ignored. And truth-value intuitions are a prima facie indication of whether a particular meaning is literal or non-literal, for instance, whether it is part of what is said or a conversational implicature.

To say this is not to admit through the back door the psychologizing of the distinction between literal and non-literal content. Devitt's psychologized derivation condition is meant to be fulfilled by all conversational implicatures. As a result, an implication counts as a conversational implicature only if the speaker and hearer consciously distinguish, in the process of interpretation, the literal content of an utterance from the content of the implication. In contrast, I am not (and surely Grice is not) claiming that for all conversational implicature we are able to distinguish intuitively literal from non-literal content. But it is plausible to think that in many cases we are. And this, as I will argue, is sufficient to show that the Gricean has no overgeneration problem.

Le me insist some more on this point. The original Gricean account of conversational implicatures does not suffer from the overgeneration problem that Devitt thinks he has identified if a certain value is placed on intuitions of literal truth and intuitions of what is indirectly and nonliterally conveyed by an utterance of a sentence. This is not to claim that truth-value intuitions are a reliable guide to semantic content. They are not, and Bach ("Seemingly" 23), among others, warns us that truth-value intuitions should be treated cautiously, as "they are often responsive to non-semantic information, to what is implicit in what is said but not part of it." But it is one thing to say that we, as competent speakers, are not always able to distinguish semantic content from pragmatics effects on it, and something very different to say that we are never able to do so. The examples of conversational implicatures that Grice gives in order to illustrate the concept are cases in which we are able to do so. Otherwise, we could never understand the point that Grice makes in presenting them. So, I take it that it is uncontroversial that some reliance can be placed on such intuitions, at least in those cases in which they are strong and do not vary too much from one speaker to another.

This is a point that Grice himself makes, although in somewhat different terms: "We must of course give due (but not undue) weight to intuitions about the existence or nonexistence of putative senses of a word (how could we do without them?)" ("Logic" 49). These intuitions about literal senses of words can easily be translated into judgments about the literal truth-value of utterances of sentences, and so made to comply with the standard methodology. The resulting data from truth-value intuitions do distinguish between live and dead metaphors, and shows beyond doubt that the Gricean does not have a massive overgeneration problem. Consider the use of 'fabulous' in sentence (2):

2. The results of the last research project are fabulous.

Now, the Oxford Dictionary of Word Origins explains that in early use 'fabulous' meant "known thought fable" or "not based on fact" (160). A derivation of the use of the word to mean wonderful or amazing is still possible. But truth-value intuitions indicate that (2) is literally true or false depending on the quality of the research results mentioned, and so that the use of 'fabulous' in (2) is literal.

The point made here about dead metaphors could also be made with respect to metonymies (also known as "predicate transfer"): there are metonymies that are "dead", i.e., lexicalized. Ruhl (97) calls them semantic metonymies (e.g., 'tongue' when it designates a certain linguistic capacity, 'orange' when it designates a fruit, or 'china' when it designates Chinese porcelain), as opposed to pragmatic metonymy (e.g., the use of 'crown' to mean royalty, or Recanati's (Literal 26) example of a waiter's use of 'ham sandwich' to mean the customer who asked for, or consumed, a ham sandwich). While it might be controversial whether the latter examples are pragmatic, what is relevant to our purposes here is that the former examples are clearly literal uses. One example is sufficient to make the point:

3. That orange smells good.

Competent speakers judge (3) as true if and only if a particular fruit smells good, not a particular shade of the colour orange. These are the literal truth-conditions of an utterance of (3), which means that the literal meaning of 'orange' refers to a fruit.

These considerations indicate that a Gricean who relies minimally on intuitions about literal truth does not predict that there are no dead metaphors or no dead metonymies, or that there is no lexicalization. The Gricean has no overgeneration problem, and so there is no need to psychologize the derivation of implicatures.

Another source of the alleged overgeneration problem that Devitt discusses is Grice's Modified Occam's Razor. According to MOR, senses are not to be multiplied beyond necessity. But MOR does not lead to overgeneration of pragmatic explanations either. Grice does not mean by his MOR that whenever a sense of the word is derivable (as well as intended and cancellable), it is not literal. His idea is not that we should shrink the lexicon as much as we can. MOR is not supposed to replace the traditional methodology of truth-conditional semantics. Instead, MOR is better understood as a contribution to this methodology, and in particular as a principle of theoretical simplicity and economy that is meant to be relevant for cases in which the main source of data does not help choose between literality and non-literality. If intuitions do not provide a means to decide whether a use is literal or not, and if we have no other alternative way of deciding this issue, then (and only then) Grice's principle of parsimony becomes relevant. Considerations of parsimony are only relevant where the empirical data do not serve to choose between alternative competing accounts of the same phenomenon9. In such cases, the question whether the use is standardized (in Bach's sense) or conventionalized (in Devitt's sense) is in place. This is the case for the referential use of DDs, which triggers diverging intuitions: while Referentialists, but also contextualists such as Recanati (Direct 283), have the intuition that the referential use of DDs is literal, Russellians disagree. The main source of data for semantic theories, i.e., truthvalue intuitions, do not lead to clear results in this case10. Tentatively, one might appeal to MOR, and argue that, until new data is found, a Russellian-Gricean account is preferable. But MOR should be used tentatively, as new empirical evidence might surface which decides the issue one way or another.

Final considerations

If Devitt's defence of premise (ii) is not successful, then the argument from convention fails to support its conclusion. The Referentialist might insist that an argument in favour of the conclusion that referential uses are best explained semantically could still be made after dropping premise (ii), and so without mentioning pragmatic accounts at all. This argument might be thought to run as follows:

i. DDs have a standard and regular referential use.
ii. If an expression has a standard and regular use then this is strong evidence that the use is literal.
iii. Therefore, referential uses of DDs are best explained as literal uses; in particular, there is a linguistic convention of using descriptions referentially.

I have already discussed this version of the argument above, and concluded that it cannot be successful: an argument for the best explanation of a certain fact cannot be made in absence of a discussion of the alternative proposals for explaining it. Moreover, metasemantic considerations concerning conventionalization cannot be appealed to independently of the main source of semantic data (i.e. data from truth-value judgments), as well as of other kinds of metasemantic considerations, such as lexical parsimony. Similarly, MOR and the derivability criterion cannot be used in isolation to draw the opposite conclusion, that referential uses are not literal. As we have seen, derivability (in the sense of the existence of a derivation), and regularity of use do not exclude each other. A use might be regular, and still a derivation of it from a different use of the same word might be available. But not all uses that are derivable are pragmatic, and not all uses that are regular are conventional and semantic. So, one must take on board both kinds of considerations when addressing the issue of whether a certain use of an expression is literal or not. That is because both kinds of considerations lead to overgeneration of false predictions if taken in isolation.

The domain of application of these considerations pertaining to the existence of derivations and the regularity of use is the class of cases for which truth-value intuitions are insufficient to decide the issue one way or another. Does any one of the two kinds of considerations have priority with respect to the other, relative to this class of borderline cases? Grice and Griceans such as Bontly (289) suggest that, given MOR, the default position is the pragmatic one. Grice writes: "Though it may not be impossible for what starts life, so to speak, as a conversational implicature to become conventionalized, to suppose this is so in a given case would require special justification ("Logic" 39). Devitt replies: "there is no basis for thus putting the onus on showing that conventionalization has taken place rather than on showing that it has not, so we still have an implicature" ("Referential" 17). Instead, Devitt suggests that the default position for the difficult cases is the conventionalist one. He proposes the following formulation of MOR: "If a Gricean derivation of the thought conveyed by some use of an expression is available and positing a convention does not better explain the data pertaining to this use, then one ought to explain the use in pragmatic terms." (Devitt unpublished manuscript, quoted in Philips (375)). According to Devitt, we should first consider the virtues of the conventionalist explanation, and only afterwards appeal to MOR. But why? I find no good reason to choose either the pragmatic view as default for borderline cases, or the conventionalist view.

In conclusion, in this paper I aimed to show that Devitt's argument to the effect that the Gricean needs to reformulate the derivation requirement on implicatures in psychologistic terms in order to avoid an overgeneration problem fails, because there is no overgeneration problem to begin with. As a result, the argument from convention fails to be compelling. If the argument from convention is reconstructed without the problematic premise (ii) it is even less convincing, as it aims to establish Referentialism exclusively on the basis of the regularity of the referential use of DDs. But such considerations relating to the conventionalization of standard uses of expressions always need to be considered in tandem with MOR and the availability of Gricean derivations. The two kinds of considerations pull in opposite directions, and no convincing reason has been given so far to give priority to one over the other. Therefore, while the argument from convention does offer a reason in favour of Referentialism, this is not a sufficient reason to establish it.


1 Not all Referentialists agree that in case the intended referent does not satisfy the descriptive meaning no proposition is expressed. For instance, Devitt ("The Case" 282) and Marti ("Direct") disagree. In what follows I ignore the difference between these two versions of Referentialism, as it is not essential to the forthcoming discussion.
2 In fact there are other positions in the debate that I do not discuss here. For instance, not all Russellians think referential uses are semantically irrelevant (e.g., Neale ("This") does not). Also, not all those who take DDs when used referentially to be referential expressions subscribe to the ambiguity thesis. Contextualists such as Recanati ("Contextual") reject Referentialism as formulated here.
3 Devitt ("Referential" 8) credits Neale ("This" 71) for coming up with this label. Maybe 'the argument from standard use' is a better label.
4 However, in "Referential" (8) Devitt writes that in his "The Case" "the focus of my presentation of the Argument from Convention was on the idea that referential uses are particularized conversational implicatures", an idea which he rejected. This suggests that the inexistence of a correct pragmatic account of referential uses is part of the very argument, and so, that my reconstruction is the one Devitt ("The case") had in mind as well.
5 It might be replied that this is an uncharitable reading of Devitt's argument, as he is not attributing the overgeneration problem to Grice or to any reasonable Gricean, but to a "fundamentalist Gricean". But then Devitt's argument is a mere straw man fallacy, given that it is irrelevant in a discussion with a Gricean to criticize a radical position that no reasonable Gricean holds. If the argument is not a straw man fallacy, then Devitt must mean that, in general, this is a problem for any Gricean that subscribes to the above formulation of the derivation conditions for implicatures.
6 Reimer makes the same point: "it seems to be that... there is no divergence between what is literally said and what is meant: Intuitively, the speaker said precisely what he meant" (Reimer, "Demonstrating" 893).
7 Devitt's discussion is part of a long debate with Bach. Bach ("Referentially" 38) replies that his account does not invoke an implicature, but what he calls an "impliciture". Bach relies heavily on the distinction between the content of the locutionary act of saying and that of the illocutionary act of asserting. The Russellian theory of DD characterizes the content of the former, even when the content of the latter is a singular proposition. However, Bach ("Referentially" 40) does call his account Gricean, without giving details, and writes that implicitures are "akin" to GCIs. So, Devitt ("Note" 50) is right to be puzzled. And Devitt is right that if implicitures are also derivable from what is said then the above objection applies to Bach's pragmatic account of referential uses as well.
8 I do not believe that this is the best theoretical option. If one wants to account for referential uses semantically, Neale's ("This" 171-173) proposal mentioned above is superior to Referentialism in various respects. First of all, although it is a semantic approach, it does not require postulating a new linguistic convention for referential uses, so it is in line with MOR. Second, it has the theoretical advantage that it can be conceived of as an application of any of a variety of widely accepted theories of quantifier domain restriction (see Moldovan).
9 Suppose that I am wrong about how Grice meant to use his MOR. Still, it is possible to combine a non-psychologizing reading of Grice with a standard methodology that uses truthvalue judgments as the main source of data for testing semantic hypotheses, and, at the same, relegate MOR to a secondary place, relevant only for those cases in which the empirical data is inconclusive.
10 See also Schoubye (522-523), who expands Heim's (1016) point that it is questionable that we can test the difference between the Russellian and the Referentialist truth-conditions by appeal to truth-value judgement tasks.


Bach, Kent. "Standardization vs. Conventionalization". Linguistics and Philosophy. 18, 1995: 677-686. Print.         [ Links ]

___. "Standardization revisited". Pragmatics: Critical Concepts. Vol. IV: 712-722. A. Kasher (ed.). London and New York: Routledge, 1998. Print.         [ Links ]

___. "Seemingly Semantic Intuitions". Meaning and Truth. Investigations in Philosophical Semantics. J. K. Campbell, M. O'Rourke, D. Shier (eds.). New York, NY: Seven Bridges Press, 2002. 21-33. Print.         [ Links ]

___. "Descriptions: Points of Reference". Descriptions and Beyond. M. Reimer and A. Bezuidenhout (eds.). Oxford: Oxford University Press, 2004. Print.         [ Links ]

___. "Referentially Used Descriptions: A Reply to Devitt". European Journal of Analytic Philosophy. Vol. 3, No. 2. 2007. Print.         [ Links ]

Bontly, Thomas D. "Modified Occam's razor: Parsimony, pragmatics, and the acquisition of word meaning". Mind and Language. 20 (3). 2005: 288-312. Print.         [ Links ]

Carnap, Rudolf. "Meaning and Synonymy in Natural Language". Meaning and Necessity. Chicago: University of Chicago Press, 1947. Print.         [ Links ]

Cresswell, Julia. Oxford Dictionary of Word Origins. Oxford: Oxford University Press, 2009. Print.         [ Links ]

Devitt, Michael. "Meanings and psychology: A response to mark Richard". Noûs. 31 (1) 1997: 115-131. Print.         [ Links ]

___. "The Case for Referential Descriptions". Descriptions and Beyond. M. Reimer and A. Bezuidenhout (eds.). Oxford: Oxford University Press, 2004. Print.         [ Links ]

___. "Referential Descriptions and Conversational Implicatures". European Journal of Analytic Philosophy. Vol. 3, No. 2. 2007. Print.         [ Links ]

___. "What Makes a Property "Semantic"?" Perspectives on Pragmatics and Philosophy. A. Capone, F. Lo Piparo, and M. Carapezza, (eds.). Switzerland: Springer International Publishing, 2013. Print.         [ Links ]

___. "Good and Bad Bach". Croatian Journal of Philosophy. 13 (2) 2013: 169-200. Print.         [ Links ]

Donnellan, Keith. S. "Reference and definite descriptions". Philosophical Review, 75 (3) 1966: 281-304. Print.         [ Links ]

Elbourne, Paul. Definite Descriptions. Oxford: Oxford University Press, 2013. Print.         [ Links ]

Gibbs, Raymond W. Jr. "Psycholinguistic Experiments and Linguistic-Pragmatics". Experimental Pragmatics. Ira A. Noveck and Dan Sperber (eds.). Palgrave Macmillan, 2004. 50-71. Print.         [ Links ]

Grice, Paul H. "Vacuous Names". Words and Objections. D. Davidson, J. Hintikka (eds.) Dordrecht: Reidel, 1969. 118-145. Print.         [ Links ]

___. "Logic and Conversation". Studies in the Way of Words. Cambridge MA: Harvard University Press, 1989. 22-40. Print.         [ Links ]

___. "Further Notes on Logic and Conversation". Studies in the Way of Words. Cambridge MA: Harvard University Press, 1989. 41-57. Print.         [ Links ]

Heim, Irene. "Definiteness and Indefiniteness". Semantics: An International Handbook of Natural Language Meaning. C. Maienborn, K. von Heusinger and P. Portner (eds.). De Gruyter Mouton. 2011. Print.         [ Links ]

Kripke, Saul. "Speaker's Reference and Semantic Reference". Midwest Studies in Philosophy. 2, 1977: 255-276. Print.         [ Links ]

Marti, Genoveva. "Direct Reference and Definite Descriptions". dialectica. 62(1), 2008: 43-57. Print.         [ Links ]

Moldovan, Andrei. "Incomplete Descriptions and the Underdetermination Problem". Research in Language. Vol 13. No 4. 2015: 352-367. Print.         [ Links ]

Neale, Stephen. Descriptions. Cambridge, MA: MIT Press, 1990. Print.         [ Links ]

___. "This, That, and the Other". Descriptions and Beyond. M. Reimer and A. Bezuidenhout (eds.). Oxford: Oxford University Press, 2004. Print.         [ Links ]

Peacocke, Christopher. "Proper Names, Reference and Rigid Designation". Meaning, Reference and Necessity. S. Blackburn (ed.). Cambridge: Cambridge University Press, 1975. Print.         [ Links ]

Phillips, Ben. "Modified Occam's Razor". Australasian Journal of Philosophy. 90 (2). 2012: 371-382. Print.         [ Links ]

Recanati, François. "Contextual Dependence and Definite Descriptions". Proceedings of the Aristotelian Society. 87. 1987: 57-73. Print.         [ Links ]

___. Direct Reference: from Language to Thought. Oxford: Blakwell, 1993. Print.         [ Links ]

___. Literal Meaning. Cambridge: Cambridge University Press, 2004. Print.         [ Links ]

Reimer, Marga. "Demonstrating with descriptions". Philosophy and Phenomenological Research. 52 (4). 1992: 877-893. Print.         [ Links ]

___. "Donnellan's Distinction/Kripke's Test". Analysis. 58(2). 1998: 89-100. Print.         [ Links ]

Ruhl, Charles. On Monosemy: A Study in Linguistic Semantics. Albany: SUNY Press, 1989. Print.         [ Links ]

Sainsbury, Mark. Reference without Referents. Oxford: Clarendon Press, 2005. Print.         [ Links ]

Schoubye, Anders J. "Against the argument from convention". Linguistics and Philosophy. 35(6). 2012: 515-532. Print.         [ Links ]

Strawson, Peter. F. "On Referring". Mind. 59. 1950: 320-344. Print.         [ Links ]

Como citar:
Moldovan, A. "An assessment of the argument from convention". Discusiones Filosóficas. Ene-Jun. 2016: 15-34. DOI: 10.17151/difil.2016.17.28.2

Creative Commons License All the contents of this journal, except where otherwise noted, is licensed under a Creative Commons Attribution License