Stigw.dk

1.1. The Formal Philosophy of Language The Problem of Reference
The Formal Philosophy of Language
The Problem of Reference: Frege & Russell How can words be “about” things? What is the relation between words and the entities they stand for? The problem is probably as old as philosophy itself. It can at least be traced back to Aristotle's De Interpretatione. In much philosophy prior to this century, the focus of attention was the relation of objects to thoughts, rather than language. The problem, in its modern incarnation, where attention is predominantly switched to words and signs, has become known as the problem of
reference. In chapter 1.1, I will trace the problem from its modern origins in the works of Frege and
Russell at the end of the 19th century, to the present day. This brief history of 100 years of reference will set an important part of the background that allows me to place Kronfeld's theory in its proper The starting-point for Gottlob Frege's famous paper “Über Sinn und Bedeutung” (Frege 1892/1986) is the relation of identity (Gleichheit). If identity is a relation at all, is it a relation between objects or between the names and signs representing these objects? The proposition a=a is necessarily true, whereas a=b is not. If this was merely a relation between objects, the two propositions would not differ in nature. Obviously, it is a relation between signs, but one whose existence depends on the signs having a referent. a=b merely expresses that we attach two different names to the same object.
If this was the whole story, propositions like a=b would contain little information, and it would not explain why propositions of this sort are so frequently expressed in natural language. Frege's answer is that the difference between the two signs expresses a difference in the way the object is presented to us (die Art der Gegebensein).
It is insufficient to operate with a sign and the object to which it refers. The reference relation has three elements: Zeichen (sign), a linguistic expression, Sinn (sense), the conceptual
content, fixed by convention, allowing us to determine reference, and Bedeutung (reference), the
1.1. The Formal Philosophy of Language referent, the object referred to. (Frege's term for object is Gegenstand, which covers “objects” in a wide sense but excludes concepts and relations). This tripartite, Fregean model is constitutive of what can be called the indirect view of reference, which supposes the existence of some connecting link
between the word and the entity. Frege believes the indirect reference relation to hold even for proper names. Every name has its Sinn.
The Fregean Sinn is not identical to thoughts or beliefs. Thoughts and beliefs belong to one person at one time, where the Sinn is shared by everyone who uses the same language. Furthermore, language is ambiguous. One word can have more than one Sinn, and for every Sinn there can be more than one Bedeutung — or none at all! We can express this lack of reference as Bedeutung having a This holds not just for single words, but for whole sentences. The Sinn of a sentence is equal to the thought expressed by it. On some occasions of language use, e.g. when we're reading a work of fiction, this is enough. On other occasions the sentence also has a Bedeutung, which Frege takes to be a determinate truth value (Wahrheitswert). This does not hold for all sentences. The relative clause “Der die elliptische Gestalt der Planetenbahnen entdeckte” (“[he] who discovered the elliptical shape of the planetary orbits”) has no truth value, but an object for its Bedeutung, fully similar to a noun. A sign can lack its referent, but the sentence seems to postulate the existence of an object Someone did discover the elliptical shape of the planetary orbits. The fact that this is not stated explicitly, Frege believes to be a defectiveness of language; a defectiveness one might even detect in the formal language of analysis (“[ein] Unvollkommenheit der Sprache, von der übrigens auch die Zeichensprache der Analysis nicht ganz frei ist”, Frege 1892/1986: 55).
Bertrand Russell translates the Fregean Sinn and Bedeutung into meaning and denotation. In his paper
“On Denoting” (Russell 1905/1977), he dissociates himself from those ideas of Frege that he has previously shared, and proposes a new theory. Russell's analysis concerns what he calls denoting
phrases. These are characterized by form, and judging by Russell's examples they correspond largely
Denoting phrases can refer either to a single object, refer ambiguously (e.g. the indefinite “a man”) or refer to nothing at all. “The King of France” has no denotation, since there is no (present) The term Frege uses about the example is “Nebensatz” (subordinate clause). A modern reader is of course likely to insist that as this example simply is an NP, it has nothing to say about the reference of a“Satz” (sentence, or main clause).
1.1. The Formal Philosophy of Language King of France. The trouble arises with sentences like “the King of France is bald”. This could be solved by saying that the expression concerns the meaning, not the denotation of “the King of France”, but if we say that “the King of England” is bald (and there is a King of England, as was the case at Russell's time of writing), this is obviously a statement about the denotation, i.e. the man himself.
Consequently, “The King of France is bald” should be pure nonsense, but according to Russell's intuitions “it is not nonsense, since it is plainly false” (ibid.: 46).
Two solutions are available for this problem. One solution that was proposed by the psychologist Alexius Meinong is to attribute a special kind of existence to entities like “the King of France”.2 Analogous to the way we accept the existence of mathematical objects, we will have to accept impossible constructs like “the round square” by virtue of the fact that we can talk about them.
Meinong is not afraid to say that objects exist which have non-existence as a predicate (“es gibt Gegenstände, von denen gilt, daß es dergleichen Gegenstände nicht gibt”, Meinong 1904/1971: 490).
To Russell, this is simply an unacceptable paradox. The other solution is that of Frege: a null value for Bedeutung. But in that case, “the King of France” should be nonsense, and instead Russell proposes a third, radical solution: “we must abandon the view that denotation is what is concerned in propositions which contain denoting phrases” (Russell 1905/1977: 47).
Russell sees a clear difference between propositions of the type “Louis was a man” and “the king of France was a man”. The proper name has a denotation (direct reference) and in sentences,
it performs the role of subject to a predicate. But according to Russell an expression like “the King of France” is not the subject of the sentence, but an existential postulate (cf. Frege's “defectiveness of language”). So the sentence, “the King of France is bald”, really means: There is one (and only one) King of France, and he is bald, i.e.
✂☎✄✝✆✟✞✡✠☞☛✍✌✏✎☞✑☞✒✔✓✕✑✔✓✕✖✘✗✙✌✏✚✜✛✔✞✍✄✣✢✥✤✧✦✩★✪✆✟✞✡✠☞☛✍✌✣✎✫✑☞✒✔✓✕✑✔✓✕✖✘✗✙✌✣✚✜✛✔✞✡★✫✢✭✬✮✄✰✯✣★☞✢✥✤✲✱✣✗✜✳✡✴✵✞✍✄✏✢✘✢ Russell says that “the King of France” in this case has a primary occurrence and that the proposition
is false because the existential postulate is false. Correspondingly, we can interpret the negation, “the Meinong's solution can be seen as a third strategy, alternative to both the direct and indirect views of reference discussed below. The strategy consists of sticking to a simpler theory at the cost ofaccepting a much richer ontology. A modern incarnation of this approach is David Lewis' insistence thatpossible worlds really exist (Lewis 1973), which can be read as a rival approach to that of Kripke reported insection 1.1.3.
1.1. The Formal Philosophy of Language ✂☎✄✝✆✟✞✡✠☞☛✍✌✏✎☞✑☞✒✔✓✕✑✔✓✕✖✘✗✙✌✏✚✜✛✔✞✍✄✣✢✥✤✧✦✩★✪✆✟✞✡✠☞☛✍✌✣✎✫✑☞✒✔✓✕✑✔✓✕✖✘✗✙✌✣✚✜✛✔✞✡★✫✢✭✬✮✄✰✯✣★☞✢✥✤✷✶✸✱✣✗✜✳✹✴✵✞✍✄✣✢✘✢ whereby the proposition is false. But we can also let “the King of France” have a secondary
occurrence:
✶✺✂✵✄✝✆✟✞✡✠✫☛✍✌✣✎☞✑☞✒✻✓✕✑✔✓✕✖✘✗✙✌✣✚✜✛✔✞✍✄✣✢✥✤✲✦✩★✰✆✟✞✡✠☞☛✼✌✣✎☞✑☞✒✔✓✕✑✻✓✕✖✘✗✙✌✣✚✜✛✔✞✡★☞✢✭✬✮✄✰✯✏★☞✢✥✤✧✱✣✗✜✳✹✴✵✞✍✄✣✢✘✢ whereby the proposition is true. The secondary occurrence is defined by the denoting phrase in question being part of a proposition that is a mere constituent of the entire expression. In the logical interpretation of “the King of France is not bald” (with a secondary occurrence), the scope of the negation covers the entire expression of which “the King of France” is a constituent.
According to this theory, referring expressions are really existential postulates, whereas proper names seem to have direct reference. With this view Russell is obviously opposed to Frege. However, Russell still supports an indirect view of reference in the sense that he largely believes in the role of descriptive content: “Common words, even proper names, are usually really descriptions” (Russell
1910/1918: 216). When we say “Julius Caesar”, we mean “the man whose name was Julius Caesar” and thus describe Julius in a way not essentially different from the description “the author of The Gallic Wars”. The direct reference only occurs in cases of deixis, and Russell finally reaches the conclusion that the only logically proper names are demonstratives like “this”. The demonstratives
do not point to objects, only to the sensory data we receive from these objects. This is Russell's epistemology (Russell 1910/1918): sensory data are the only things of which we have knowledge by
acquaintance, defined as a direct cognitive relation. To all other things we are only indirectly related,
i.e. we have knowledge by description. All descriptions must, in order for us to understand them, be
reducible to constituents of which we have knowledge by acquaintance.
The schism between Russell and Frege seems to diminish, when one considers how narrow the Russellian concept of proper name is. However, Russell's view on descriptions as existential postulates (technically speaking, functions) that only make sense in context is still a radically different view on reference, since it rejects the importance of Bedeutung or denotation.
1.1.2. Referring: Strawson and Searle 1.1. The Formal Philosophy of Language Frege and, more explicitly, Russell express a certain dissatisfaction with human language as it is. For instance, the fact that copula can express a subject-predicate relation as well as a relation of identity leads Russell to the following exclamation: “It is a disgrace to the human race that it has chosen to employ the same word is' for these two entirely different ideas — a disgrace which a symbolic logical language of course remedies” (Russell 1919/1985: 215). It is Russell and, among others, the younger Wittgenstein of Tractatus logico-philosophicus (Wittgenstein 1921/1993) that form the backdrop of the 1930’s philosophical movement of logical positivism that writes off as meaningless any linguistic
utterance that cannot be assigned the value true or false. Among the great rebellions against this philosophy is Wittgenstein's own Philosophische Untersuchungen in which he emphasizes that the meaning of a word is its use in the language (“Die Bedeutung eines Wortes ist sein Gebrauch in der Sprache”, Wittgenstein 1953/1993: 262). This focus on language use is something quite new compared to Frege and Russell. The first one to explicitly treat reference as the action of referring is P.F.
Strawson, in his paper “On Referring” (Strawson 1950/1985).
Frege and Russell discuss how expressions refer. But words do not refer. We refer by words.
This is Strawson's main insight: “ Mentioning', or referring', is not something an expression does; it is something someone can use an expression to do. Mentioning, or referring to, something is characteristic of a use of an expression, just as being about' something, and truth-or-falsity, are characteristics of a use of a sentence” (ibid.: 224). Russell's theory can be criticised for making postulates about expressions per se that really only apply to uses of these expressions. Strawson points out that distinctions are needed between: 1) expressions and sentences, such as “the King of France”
and “the King of France is bald”; 2) uses of expressions and sentences — the reference and truth value
differs according to whether “the King of France is bald” is said during the reign of the Sun King or at the age of Strawson; thus, reference is not tied to the expression and truth value not to the sentence; both depend on the use; 3) utterances — two people simultaneously stating that “the King of France
is bald” would be making two distinct utterances, but since the reference and the truth value would be the same, this would be the same use.
The Russellian analysis, as I have reported it here, is seen as wrong in two ways. Firstly, because Russell believes expressions like “the King of France is bald” to be simply false. If there is no reference, says Strawson, there is no truth value — but the lack of truth value does not mean that the sentence is meaningless, like Russell (and logical positivism) would have it. Secondly, Russell is wrong in taking the existential postulate to be part of the logical form of the expression. It is correct that by using the expression “the King of France” you imply the existence of the said regent, but this 1.1. The Formal Philosophy of Language is not a case of logical implication, and the implication is not part of the sentence which has an ordinary subject-predicate format. Even Frege talked of existence as a precondition (Voraussetzung) which isn't part of the thought being expressed, and like Strawson he emphasized that the negation of the sentence isn't a negation of the existential postulate (Frege 1892/1986: 54-55). To distinguish this sort of precondition from logical implication/entailment, Strawson introduces the term presupposition
(Strawson 1952), now of course a standard item of linguistic terminology. The definite article in a referring expression is a signal that the existence of a given object is presupposed.
Similarly, Strawson renounces Russell's theory of demonstratives as logically proper names which by direct reference make the objects themselves part of the proposition. The reference of “this” changes with every new use — and it is the meaning of “this” that enables us to fix the reference. In Strawson's terms, every sentence as well as every single word, including demonstratives, has a meaning (quite similar to Frege's Sinn): a set of general directions for how the expression can be used to refer or, in the case of a sentence, express true or false propositions (Strawson 1950/1985: 224).
Contemporary with Wittgenstein's reflections on language as an activity, a set of games (in the Philosophische Untersuchungen) we find J.L. Austin's posthumously published works on speech acts
(Austin 1963/1971, a.o.). Some utterances are propositions and assertions to which we indeed can assign truth values. Austin terms these as constatives. But language is more than these utterances. We
also have expressions like “I apologize”, “Welcome”, and “I christen this ship Liberté”. These utterances are actions in themselves (or parts of actions), and Austin calls them performatives. By
saying “I apologize” one performs the act of apologizing. Such utterances have no truth values but felicity conditions — you can't christen a ship by simply going down to the harbour at any time with
a bottle of champagne and declaring the christening utterance. At a later stage of the argument, Austin reaches the conclusion that the distinction between constatives and performatives aren't that important: In direct continuation, John R. Searle in his book Speech Acts extends and systematizes Austin's ideas (Searle 1969/1974). Searle adopts Austin's terminology which distinguishes between three kinds of speech acts: locutionary act, the production of the utterance; illocutionary act, the
proposition, offer, promise, etc. the speaker makes by producing the utterance: the basic speech act; and the perlocutionary acts, the effects on the hearer, which contrary to the locutionary act are not
fixed by convention, but fully dependant on circumstances.
To this, Searle adds the two propositional acts, referring and predicating, as the building
1.1. The Formal Philosophy of Language blocks of which illocutionary acts are made. These building blocks are acts themselves, and speech acts are compositional (one can promise to refer, etc.). Like Strawson, Searle maintains that referring expressions don't refer — the language user refers.
Several main passages of Speech Acts are devoted to reference, more specifically referring expressions in the definite singular. Searle proposes three axioms of referring, in agreement with the tradition from Frege and Strawson: 1) The thing you refer to must necessarily exist; 2) A predicate which is true for a given object, holds for everything identical to this object, no matter what expression is used to refer to it; and 3) In referring to an object, you identify — or are able to identify — this object (i.e. distinguish it from others).
The first two of these axioms are not without their problems, for reasons already touched upon in the foregoing discussion. Searle acknowledges this. If e.g. 1) holds, things have to exist before you can deny their existence, like in the sentence “The round square doesn't exist”. As we have seen, Russell's solution was to say that such a proposition does not express a reference, but an existential postulate (and since existence is not a property, there is no subject-predicate relation). Searle's solution is to distinguish between talk about the real world and talk about an imaginary universe. These are two different ways of using language. In a discourse on fiction, we can seriously discuss the doings of Sherlock Holmes, but we can't mix these with discourse on reality and say, “Sherlock Holmes will pay me a visit tonight”. For Sherlock Holmes does not exist — he does not even have some kind of idealized existence in Alexius Meinong's sense.
The third axiom is original to Searle. He classifies three ways of identifying: Either the expression contains a predicate that only holds for one object, or the utterance together with the context distinguishes a single object, or the utterance contains sufficient description for the object to be identified. Searle himself equates this to Frege's Sinn, which he translates as sense (ibid.: 80). The necessary condition for the speech act of referring to be fully consummated is that the object is
uniquely identified for the hearer. For the reference to be successful it is only needed that the speaker
is potentially capable of making the identification. In any case, this happens by the aid of an identifying description, which can be a demonstrative (“that guy over there”), purely descriptive (“the
first man to run a mile in less than 3 minutes and 53 seconds”), or a mixture of both (“the man we In real life, however, the situation is not always that ideal. In ordinary conversation, A can e.g. mention Jones, whom B does not know. B will then ask, “Who's Jones?”. In this reference to Jones, B only has a pseudo-identifying description, namely “the person A calls Jones”. And A can 1.1. The Formal Philosophy of Language answer, “An Air Force lieutenant”, which is only a partly identifying description. Searle calls this partially consummated reference.
Searle finds it much more interesting that referring can succeed in cases where the description is wrong. He takes an example from A.N. Whitehead: “That criminal is your friend”, to which the opponent answers, “He is my friend and you are insulting” (ibid.: 89). Searle sees an explanation in the demonstrative “that” which indicates that either the person is physically present or has been mentioned previously by use of another referring expression. Finally, all identifying descriptions aren't equally useful. The purpose of referring is primarily identifying — a correct description (e.g. “the only man in Montana with exactly 8432 hairs on his head”) isn't enough, if it is not relevant in context and to the participant in the discourse.
In all this, Searle seems to follow the Fregean idea of a Sinn, a “descriptive content” (ibid.: 92). In addition, he repeats Strawson's point that this general Sinn/Sense must be separated from the meaning of the utterance in a specific context (use). As a plain descriptivist, Searle also denies the Russellian idea of logically proper names with direct reference. He simply rejects the possibility that anything can refer without communicating a conceptual content. And even if Searle in his first axiom acknowledges an existential postulate as the basis of reference, he dislikes the traditional reading of ✂☎✄✝✆✟✞✡✠☞☛✍✌✏✎☞✑☞✒✔✓✕✑✔✓✕✖✘✗✙✌✏✚✜✛✔✞✍✄✣✢✥✤✧✦✩★✪✆✟✞✡✠☞☛✍✌✣✎✫✑☞✒✔✓✕✑✔✓✕✖✘✗✙✌✣✚✜✛✔✞✡★✫✢✭✬✮✄✰✯✣★☞✢✥✤✲✱✣✗✜✳✡✴✵✞✍✄✏✢✘✢ For if the predicates have a variable as argument which has been introduced with the existential quantifier, the properties seem to apply to an already identified object, whereas in fact the object is only identified by the description. “For these reasons reference is — in one sense of “logical” — of no logical interest whatsoever” (ibid.: 94).
I say some caution should be given to this argument. Searle (loc.cit.) uses wordings like “it is tempting to regard” etc., so what he is criticizing is really the natural language paraphrase of the logical notation. The existential quantification does not imply that the variable has been “identified”, and the place of the existential quantifier at the beginning of the formula does not postulate a temporal relationship between quantification and predication. In this way, Searle is correct in warning about misunderstandings originating in the reading of the formula — but this can't at the same time be used in an argument that reference is “of no logical interest” (especially not without specifying what is meant by “one sense of logical”).
Searle goes on to rebel against Russell's theory of descriptions and proper names (partly a 1.1. The Formal Philosophy of Language repetition of Searle 1958/1985). Even if Frege and Strawson are wrong, and “the King of France is bald” is a plainly false proposition, Russell is still mistaken! The existence of an object (which is something different from uttering the existential postulate) is a precondition for referring, not a part of the reference itself. Similarly, hitting a person presupposes his existence, but still the act of hitting is not the same as expressing his existence. Furthermore, referring occurs in other speech acts, such as interrogative and imperative sentences. Searle neatly continues Russell's reasoning ad absurdum in saying that in that case any question about something would be a question about that thing's existence, and any imperative sentence a command to call something into being! Searle gives three arguments why proper names must have sense/descriptive content. For one thing they do form part of postulates of existence (“Cerberus does not exist”), where it makes no sense to say that they refer directly. Russell is simply cheating when he says that (only) demonstratives are proper names. Secondly, there's the Fregean argument of identity. If it turns out that “Shakespeare is Bacon”, this sentence would mean something more than just “Shakespeare is Shakespeare”. Thirdly, direct reference is simply impossible according to Searle's own theory. Accordingly, proper names must be short-hand descriptions in the sense that they presuppose the existence of a description. Proper names are not equivalent to descriptions, but they are connected to the set of possible descriptions “in a loose sort of way”, whereby the strict dichotomy Sinn - Bedeutung is partly dissolved. If proper names really represented the set of possible descriptions, they would lose their function in language.
Instead, they represent an unspecified subset: the uniqueness and immense pragmatic convenience of proper names in our language lie precisely in the fact that they enable us to refer publicly to objects without being forced to raise issues and come to agreement on what descriptive characteristics exactly constitute the identity of the object. They function not as descriptions, but as pegs on which to hang descriptions (Searle 1958/1985: 273).
Searle obviously introduces reference into the realm of pragmatic phenomena. The question of which referring expression to use becomes a question of contextual relevance, in a way compatible with the
conversational maxims of H.P. Grice (Grice 1975/1985). Every referring expression, every identifying description, represents a possible mode of presentation — a rather precise translation of Frege's Art
1.1. The Formal Philosophy of Language Against Descriptivism: From Donnellan to Kripke What is common to the views of Frege, Strawson, and Searle? Firstly, there is the claim that there is a connecting link between the referring expression and the entity to which it refers. To Frege, this link is a Sinn: a conceptual content common to all users of the expression. To Strawson it is general directions: guidelines for the use of the word. To Searle, it is defined as an identifying description, deixis being regarded as a kind of description as well.
The second important claim of what we might call the descriptivist tradition is that proper
names refer in the same way other referring expressions do. They are just a kind of short-hand or These claims have not been left unchallenged. The turning point comes with Keith Donnellan's paper “Reference and Definite Descriptions” (Donnellan 1966/1985). While this article in itself does not invalidate the descriptivist claims, it has become the basis for most later criticism Donnellan criticizes Russell as well as Strawson for failing to see that a referring expression can serve two functions. Donnellan's example is the sentence “Smith's murderer is insane”. Imagine this sentence uttered in two different situations. In the first situation Smith's dead body is lying on the floor. Since it is a brutal killing and Smith was a likeable person, we exclaim “Smith's murderer is insane”, by which we mean: Whoever murdered Smith (we do not know his identity, only this attribute) must be a madman. Donnellan terms this the attributive use of a referring expression. In
the next situation we're in court. Jones is charged with the murder of Smith. We do not know Smith, or whether Jones is guilty or not, but we notice that Jones is acting strangely, which makes us say, “Smith's murderer is insane”. Everybody knows who we are talking about, no matter what their opinion is on Jones' guilt. Reference succeeds, even if Jones isn't Smith's murderer at all. The referring expression serves as a part of pointing out the referent, and seemingly it doesn't matter much whether the description is correct. Donnellan calls this the referential use.
Like Strawson's presupposition, Donnellan's distinction between referential and attributive has become a standard piece of linguistic terminology.
According to Donnellan, Russell's analysis is only true for the attributive use of descriptions: There is a person who murdered Smith, and he is such-and-such. There's nothing wrong with the Russellian idea of logically proper names with direct reference — but this holds for descriptions as 1.1. The Formal Philosophy of Language Strawson, on the other hand, does describe the referential use, but in a way Donnellan believes to be insufficient. Even if Strawson emphasizes use (contrary to expression), he still believes that a referring expression as such can be identified in a given sentence, “just as we can speak of a function of a tool that is not at the moment performing its function” (ibid.: 237). Donnellan believes that it only makes sense to analyse a sentence in its use on a certain occasion.
Both Russell and Strawson accept the idea that reference implies an existential postulate.
Even if this to Russell is a logical implication and to Strawson a presupposition, they agree that the truth value of the existential postulate affects the truth value of the proposition (even if they disagree on how the truth value is affected). Donnellan says that the truth value of the proposition is affected differently according to whether the use is referential or attributive.
For expressions of the type “The is ”, we can say the following: In the attributive use, ∃ is presupposed, i.e. something has to fit the description. If nothing is , then nothing has been said to be . Strawson's claim that the proposition has no truth value in this case, is correct for this usage. In the referential use, however, reference can succeed without the presupposition ∃ holding, i.e. without anything fitting the description . Even if nothing is , something has still been asserted to be . Both Strawson and Russell are wrong in their ideas about the truth value of the proposition in this case. Furthermore, a unique entity must be identified. (E.g.
“Who is the man drinking a Martini?” is a question about a single person, even if there are other Donnellan separates referring from denoting (in the Russellian definition of denotation: a unique entity fitting the description). Descriptions and referring expressions are not one and the same With his claim that (what seems to be) descriptions has direct reference in the referential use, just like Russell's logically proper names, Donnellan has prepared the way for a departure from descriptivism, and much later work in the field has consisted of reactions to “Reference and Definite Descriptions”. One such is David Kaplan's “Dthat” (Kaplan 1975/1985) in which he points to a weakness in Donnellan's analysis of the referential use: Donnellan talks of “having something in mind” without contemplating what this really means. The idea (from Searle and others) that pointing is a kind of describing, Kaplan would rather turn the other way around: “If pointing can be taken as a form of describing, why not take describing as a form of pointing?” (ibid.: 322). Another philosopher working in a similar direction is John Perry (Perry 1979).
Donnellan's original paper does not establish an alternative paradigm to descriptivism, for his discussion does not suggest how reference works when descriptive content is no longer taken to 1.1. The Formal Philosophy of Language be crucial. A real suggestion comes with Saul Kripke's Naming and Necessity (excerpt: Kripke As the title suggests, Naming and Necessity has two main themes: The reference of proper names, and As regards proper names, Searle (and possibly Wittgenstein 1953/1993) saw them as having to do with a subset of possible descriptions. This makes it a necessary truth that e.g. Aristotle possesses the logical sum of properties ascribed to him — logical sum meaning that it can be a disjunction of the properties. Kripke believes this to be simply wrong. He defines proper names as rigid designators: expressions that point out the same entity in all possible worlds. This does not
mean that the entity has to exist in all possible worlds. It is possible to say: “Imagine if Hitler had never been born!”, or similarly, “Imagine if Aristotle hadn't had any of the properties we ascribe to Such sentences have lead to many arguments over cross-world or transworld identification.
How many properties can we change and still claim identity? We have no qualms about changing gender in counterfactual conditions (“If Shakespeare had been a woman.”), but how about number? (“If I had been born as twins, I would have been two nice guys”?). According to Kripke, such arguments put the cart before the horse. By using the name we originally make a reference to something in our own world. This reference is rigid — it's all the other circumstances that can be changed in the construction of an imaginary world. So it's also possible to say, “Imagine if Nixon had not been called Nixon!”. We use the proper name from our world. This is not essentially different from using English to express a proposition about a world where the English language doesn't exist (or the English words have different meanings). This independence from properties indicates that Neither does Kripke accept Russell's idea that a name, e.g. Socrates, is similar to the description “the man called Socrates”. The claim that Socrates was called Socrates is not as obvious and commonplace as it seems, for the modern English Socrates (or the Danish Sokrates, for that matter) are not identical to ancient Greek . If we change the sentence to “the man I call ❁❃❂❅❄✏❆❈❇❊❉☎❋❊● Socrates”, we get a simple circular definition. This kind of circularity can include more than one name.
A person may use the names Cicero and Catiline without knowing anything else about the historical figures than that Cicero had Catiline executed and Catiline was executed by Cicero.
To many people, the name Cicero may have no more conceptual content than the imprecise “a Roman orator” which does not refer in the Russellian sense of denoting a unique object. Still, 1.1. The Formal Philosophy of Language referring succeeds — when you say Cicero, you talk of Cicero and not of anyone else. How is this possible when there is no identifying description or any sufficiently precise conceptual content? Kripke's answer — which he presents as an illustration, not a full-blown theory — is the idea of a causal chain. At a certain time, a naming takes place. This act of baptism can include a description
or a definition, but it can also be simply deictic. From this point, the name is spread by communica- Someone, let's say, a baby, is born; his parents call him by a certain name. They talk about him to their friends. Other people meet him. Through various sorts of talk the name is spread from link to link as if by a chain (.) On our view, it is not how the speaker thinks he got the reference, but the actual chain of communication, which is relevant (Kripke 1972/1985: What Kripke lacks in this explanation are the specifications for this chain of communication. E.g. there is such a causal chain from the historical Saint Nicholas to the modern Santa Claus, but still “Santa Claus” does not refer to the historical person. However, the very idea of causal chains present an alternative explanation of reference — a paradigm where descriptive content is no longer of central 1.1.4. Twin Earth: Putnam & Putnam Kripke's argument for direct reference applies to proper names. Hilary Putnam, in continuation of Kripke, extends the argument to cover most nouns, and some words from other classes.
In “The Meaning of Meaning'” (Putnam 1975), Putnam talks of extension and intension.
The extension of a property is all the entities for which a certain predicate holds. So, the extension of “cow” is the set of all cows. Extension in this case is similar to Frege's Bedeutung, but has the advantage of being technically defined in terms of truth-values and set theory. Truth values can be problematic, since everything in the real world is not a case of either/or. To a certain extent, this can be remedied by the use of so-called fuzzy sets where an element's membership in the set is stated with a degree or probability. This is not where Putnam sees the greatest problems.
If we assume that all creatures with a heart also have a kidney, the phrases “creatures with 1.1. The Formal Philosophy of Language a heart” and “creatures with a kidney” will have the same extension. But we wouldn't want to say that the expressions mean the same thing. For this reason Putnam works with an intension, similar to the Fregean Sinn. In this argument, only extension, not intension, is technically defined. One widespread idea about intension is that it's a concept, which means it's equivalent to a psychological state. Even
though Frege insisted that Sinn is a shared property and thus not identical to a single individual's thoughts and beliefs, understanding the Sinn still involves a psychological state. The idea that such a psychological state is identical to “knowing the meaning of a word” is one of the postulates Putnam wants to argue against. The other is the idea that extension is determined by intension. Isn't it possible to imagine cases where the intension is the same but the extension varies? The idea of psychological states has, according to Putnam, been characterized by what he calls methodological solipsism. This refers to the idea that the existence of psychological states does
not presuppose the existence of other individuals, maybe not even the individual's own body (“the disembodied mind-position”). This is rather silly if you consider e.g. the psychological state of “being jealous”. When Frege wanted to separate Sinn from psychological phenomena, it was exactly because of the individual nature of thoughts and ideas. But, says Putnam, psychological states can indeed be common properties: If two people understand a word the same way, then they are in the same psychological state. On the traditional view, “knowing the meaning of a word” must be the psychological state of knowing an intension and knowing that the intension is connected to a certain word. (A person who does not know German might know the intension of Rad, if his native language has a similar word like “wheel”, but he still doesn't know the meaning of the German word.) From the postulates follow that a psychological state determines extension. To refute this, Putnam must present an example where identical psychological states lead to different extension. He does so with his celebrated Twin Earth argument.
Twin Earth is a planet somewhere else in the universe. Down to the molecular level, it is identical to Earth, and every Earthling has his Twin Earth double. But there is one difference: Wherever on Earth we find water (H O), on Twin Earth we find a liquid whose chemical substance we can represent as XYZ. XYZ looks like water, tastes like water, has all the properties of water, and on Twin Earth you refer to it by using the word “water”. At a first glance this poses no problems: The word “water” has different meanings on Earth and Twin Earth, in the sense of having different extensions. On Earth the extension of “water” can be defined as the set of H O, on Twin Earth as the set of XYZ. But how was this in the year 1750, when the chemical structure of “water” was not known on either of the two planets? When an Earthling and his Twin Earth double used the word 1.1. The Formal Philosophy of Language “water”, they would be in exactly the same psychological state — but the extension would not be the Another example would be a present-day person who couldn't tell elm trees from beech trees.
Still, the words “elm” and “beech” have different extensions. Now imagine that on Twin Earth “elm” means beech and “beech” means elm. In that case the words, when used by the equally botanically ignorant Twin Earth double, would have another extension. But he who says “beech” on Earth and his Twin Earth double would be in the same psychological state. To Putnam, these examples are sufficient for the exultant conclusion: “Cut the pie any way you like, “meanings” just ain't in the How is reference determined then? Putnam's answer has two parts: the meaning is partly determined indexically (equivalent to the act of baptism), partly socially. Putnam's theory of
indexicality is Kripke's theory of rigid designators extended from proper names to words for natural
kinds (like water). Water is H O in all possible worlds, even if the word “water” can have different
meanings in other possible worlds. Putnam terms this “world-relative constant meaning”, expressed ✞✡❍✪✒☞✖✥✛✜■✫✛✜✖✘★❑❏▲✒✫✖✘✳✡✴◆▼❖✢✭✞✡❍✪✒☞✖€✛✜■☞✛✜✖✘★❅✄❑☛✼✌❑▼❖✢✭✞✍✄❑☛✡◗❘❏▲✗✙❙❚✛✜✖❈❯❱✄◆✱✏✛✜✗✜✖✘◗▲◗✸✗✜❲❳✛✜❨❩❙❚✒❅❙✡❬✣✛❭✛✙✌✫❙❚☛✍❙❚★◆✖✘✛✙✓✕✛✜✖✘✖✘✛✜✴ ❙❪✒◆✗✜◗❈❫✙❙✹❬✣☛✡◗❵❴❜❛✍❝◆❞✡❡✣❢❤❣✏✐✙❞✡❥☞❣✣❦♠❧♦♥☞♣✸❦✍q❱▼r✢ (ibid.: 149). So reference is determined indexically (“this”). The sameL relation is read “same liquid as” and does (also) hold between entities from different worlds as a cross-world relation. The criteria for such relations depend on context: H O is not XYZ — but if we had had both H O and XYZ on Earth without distinguishing between them, chemists would at a certain time have discovered that there are “two kinds of water”. An authentic example is “jade”, which is in fact two different minerals.
Putnam acknowledges that the meanings of absolutely indexical words like “I” are concepts,
while on the other hand denying that this meaning determines the extension. Proper names, and nouns for natural kinds, have their meanings determined through a process equivalent to Kripke's idea of causal chains. This is also true for words from other classes: colour adjectives and verbs like “growing”. Maybe it applies to most nouns, including those denoting artefacts. A few “one-criterion words” seem to carry an unambiguous definition; verbal derivations like hunter = one who hunts.
The second part of Putnam's hypothesis takes meaning to be a sociolinguistic phenomenon.
A meaning is known by the language community as a whole — the individual does not necessarily know the precise definition of gold, but there are experts who can tell real gold from other stuff. So, 1.1. The Formal Philosophy of Language there is a division of linguistic labour which reflects the actual division of labour in society. Kripke's idea of an act of baptism followed by a causal chain is one example among others of reference as a social phenomenon. Putnam claims that “acquiring a word” isn't the same thing as learning its meaning. The speaker has acquired the word if he fulfills his society's minimal criteria for knowledge about the word's usage. The knowledge forms a stereotype — a tiger is a big striped cat, etc. — and
has nothing to do with the precise definition used by a biologist (in the case of tigers). The stereotype does have a conceptual content, but that is a question of the competence of the individual speaker.
Extension is still determined by other factors than psychological states! Putnam, in his 1975 incarnation, is a realist philosopher. His natural kinds are classical, definable categories, and the meaning of a word is established by direct reference.
But his idea of stereotypes points in a new direction. As opposed to the word, the stereotype has no extension — the concept need not relate to the world. A few years later, in Reason, Truth, and History (Putnam 1981), Putnam takes leave of traditional realism entirely.
Initially, he dismisses what he calls magical theories of reference — those theories that
postulate a mysterious, unanalyzable connection between the sign and the entity. A word or a sign does not refer all by itself. If a chimpanzee clatters away at a typewriter, incidentally producing a sentence, this sentence does not refer. In his dismissal of magical theories, Putnam takes his inspiration In the traditional view, there is a connection between the truth value of a sentence and the reference of the elements of that sentence. The meaning of the whole is derivative of the meaning of the parts — the principle of compositionality, usually ascribed to Frege. Putnam sets out to prove that it is possible to change the reference without affecting the truth value. Putnam's Theorem (Putnam
1981: 33-35, and 217) says that for any interpretation of a language (interpretation meaning an assignment of an intension to every predicate of the language, with intension now used in the technical sense of a function from predicates to extensions in possible worlds), there is another, different interpretation which makes the same sentences true. That means that extensions and truth values in possible worlds will not work as a formalization of meaning.
With Reason, Truth, and History, Putnam renounces what he calls metaphysical realism:
the view that the world consists of separate objects whose existence is independent of the mind and of which we can construct true and complete descriptions. One of the consequences of metaphysical realism is that truth becomes a matter of correspondence between words or thoughts (symbols) and entities in the external world. Putnam calls this externalism — reality is seen from the outside, from
1.1. The Formal Philosophy of Language a godlike perspective, or from no perspective at all.
Obviously, Putnam seeks to replace externalism by an internalism, where truth is seen as
“some sort of ideal coherence of our beliefs with each other and with our experiences as those experiences are themselves represented in our belief system” (ibid.: 49-50, Putnam's italics). Objects do not exist independently of our conceptual schemes. Both objects and signs are part of the way we perceive things — both are inside, not outside our conceptual system. So the solution to the problem of reference is simply that “rabbit” refers to rabbits, “space aliens” to space aliens, and so forth. The externalist idea of causal chains would cause trouble, at least in the case of the aliens.
Internalism is necessarily relativistic, but that does not mean anything goes (like in radical relativism, such as Feyerabend 1975). All input, all sensory data, are angled by our conceptual schemes — but they are still real input. There is an objectivity, but it is an objectivity for us, not the metaphysical objectivity abstract from the observer. The affinity to the Kantian tradition is obvious.
Putnam the internalist (1981) repeats the Twin Earth argument from his younger, externalist self (1975). But it must be read differently in this new context. When presented by an externalist, the argument is a rejection of the role of psychological states in determining extension. When presented in an internalist context, it is rather a reductio of the very idea that extension is of importance to meaning. Putnam's 1981 position is a dismissal of the “problem of reference” as a pseudo-problem of Putnam's move away from traditional realism, with his rejection of externalism and subsequent dismissal of the problem of reference, is in some ways reminiscent of the Continental tradition in philosophy, as represented by, inter alia, Roland Barthes, Jacques Lacan and Jacques Derrida. I will not deal with these thinkers, but in section 1.2 on the linguistic tradition, I will briefly touch upon the way their rejection of reference has its roots in the work of Saussure. However, to the delight of those who believe in the value of rational inquiry, the formal approach from Frege and Russell lives on, and a substantial body of literature on reference is still being published in this tradition.
The work of Kaplan (notably Kaplan 1977), Perry (Perry 1993), and others, is distinctly Russellian in nature. The neo-Russellian point of view, in its essence, is that some expressions, such as proper names, are referential (or directly referring) by nature, and whatever descriptive meaning 1.1. The Formal Philosophy of Language such an expression may contain, this meaning is not part of the proposition expressed by the utterance in which the referential expression occurs. As opposed to this, others, notably Gareth Evans (Evans 1982) and Christopher Peacocke (Peacocke 1983), have maintained a more Fregean view: the content of the expression is not just the referent itself, but also the mode of presentation.
Recent developments have been characterized by two tendencies. One is a weakening of the opposition between Fregeans and Russellians, a search for compromises which include insights from both camps. The other is a tendency to reject the traditional separation of linguistics of language, or at least a will to bring together the two communities for fruitful interaction (see e.g.
Künne, Newen & Anduschus 1997). Searle's speech act theory, Grice's work on conversational implicature, as well as later developments of these theories, are common property to the linguistic and philosophical traditions. Modern theories of reference take both semantic and pragmatic properties into A prominent exponent of these tendencies is François Recanati, whose Direct Reference. From Language to Thought (Recanati 1993) is probably the most important monograph on reference published in the Nineties. A speech act theorist, Recanati rejects any distinction between linguistics and philosophy when it comes to semantics and pragmatics (Recanati 1993: 20). He offers a complex picture of the reference of indexicals and proper names, and provides a pragmatic analysis of how definite descriptions can be used referentially, and how indexicals can be used in a descriptive fashion.
While the, by now, traditional idea of direct reference assumes that the referential term has no descriptive content, or that reference of this kind is a direct relation which involves no mediating entity (such as a meaning or a concept), Recanati construes direct reference as the cases where the mode of presentation does not enter the truth-conditions of the utterance, without making any claims that there is no mediating mode of presentation. On the contrary, he postulates two different kinds of modes of presentation, one linguistic (the meaning of the expression), one psychological (the thought associated with the expression). The modes of presentation help us identify the referent. Far from being a simple relation, so-called direct reference involves a mechanism that blocks the admittance of the mode of presentation into the proposition. Recanati describes terms that apply this mechanism as having a feature, REF (for directly REFerential).
The argument for the two different kinds of mode of presentation goes as follows (Recanati, Unfortunately, this does not imply a unity of thought. Devitt and Sterelny give an enlightening account of the many possible “hybrid theories” (1999: 96-101).
1.1. The Formal Philosophy of Language chapter 4). Central to the Fregean idea of senses is the “cognitive significance criterion”: if it is possible for someone to understand two utterances, S and S', and take one to be true and the other false, S and S' must have different senses. Russellians don't agree with this. On the contrary, the tradition from Russell takes sentences like “Cicero is bald” and “Tully is bald” to have the same content. The Russellian move is to distinguish between proposition and thought. The propositional content of the two utterances is the same, but the associated thoughts differ.
Recanati takes the distinctions further. On the one hand, we have the proposition expressed by the utterance (i.e. the truth-conditionally relevant content), and the meaning of the sentence (the non-truth-conditional content). On the other hand, the associated thought can be similarly divided into its truth-conditional content, and the rest, which is called the narrow content. (The somewhat confusing terminology stems from a tradition where “narrow” means “internal to the individual”).
In literal communication, the proposition expressed by the utterance is equal to the truth- conditional content of the thought. Kaplan, who uses the word character about the sentence meaning
(Kaplan 1975/1985), and Perry (Perry 1977), seem furthermore to imply that the linguistic meaning can be equated with the narrow content of the thought. The truth-conditionally irrelevant properties Recanati rejects this conjecture. If we look at an indexical like “I”, the linguistic mode of presentation 1) is determined by convention (“I” refers to the speaker); 2) does not vary from context to context (the reference, of course, varies); 3) is token-reflexive, meaning the reference of a token bears a relation to this token (“I” = “the utterer of this token of I”). None of these properties apply to the thought associated with the indexical. You do not identify yourself indirectly, by convention, as “the speaker”. You are simply, in Russell's terminology (Russell 1910/18), directly acquainted with yourself. The thought associated with expressions like “I”, “now”, and “here”, cannot be token- reflexive. Neither is it constant. The linguistic mode of presentation associated with “I” is “the speaker”, but the speaker thinks of himself as myself, where the hearer thinks of him as he. For these reasons, there must be psychological modes of presentation separate from the linguistic modes.
These psychological modes of presentation explain how we can accept “Cicero is bald” as true, while at the same time taking “Tully is bald” to be false. Frege's Constraint (Schiffer 1978) says
that if there is a mode of presentation m, under which a person believes a thing to be F, then the same person cannot also believe the same thing not to be F under m. Recanati takes this to apply to the psychological modes of presentation. The linguistic modes do not satisfy Frege's Constraint, and are constant from occurrence to occurrence.
Proper names are construed as a kind of, or at least related to, indexicals (Recanati, chapter 1.1. The Formal Philosophy of Language 8). They carry the feature REF, meaning they are directly referential expressions. They are part of language in the sense that they form a grammatical category, but the conventions assigning bearers to names are social, not linguistic conventions. What is a linguistic convention is the knowledge that there are such social or contextual conventions — in other words, the knowledge that a name has a bearer. The sense of a name, NN, is consequently, “bearer of the name NN”. Like (other) indexicals, the meaning of names refer the hearer to some relation which holds between the expression and its reference, but contrary to (other) indexicals, proper names are not necessarily token-reflexive, and in a sense all proper names have the same meaning (REF + “bearer of”).
The view that NN means “the bearer of NN” was, as we saw, one of the Russellian postulates criticized by Kripke for circularity. Recanati's answer to this (chapter 9), is that this “metalinguistic” view is only a partial theory of reference. It is a theory about the sense, or character, of proper names. It explains the reference of a proper name token, by appealing to the “bearer of” or “is called” relation, which is another kind of reference (where Kripke's causal explanation might well come into play). This is also in accordance with Putnam's division between individual competence (what is “in the head”) and the social factors which ultimately determine reference.
Finally (chapter 10), Recanati sees a difference between proper names and indexicals when it comes to psychological modes of presentation. Indexicals like “I”, “here”, and “now”, are associated with unique linguistic modes of presentations, as well as unique psychological modes of presentation (the “egocentric concepts” Ego, Nunc, and Hic). Proper names are not associated with unique psychological modes of presentation — there is not a single mode of presentation which is thought of every time a certain proper name is used. Neither are the modes associated with proper names egocentric concepts. Recanati tends to see them as associated with encyclopedia entries. An
encyclopedia entry could contain information such as “my new neighbour”, “man met in the staircase on Jan. 1st, 1992”, “wears glasses”, “called Peter Jones”.
Indexical and proper names are referential terms by nature; they are “born” with feature REF.
Definite descriptions do not have this feature. But this is mutable due to pragmatic factors.
Descriptions can be used referentially (Recanati, chapter 15).
Donnellan said that in the referential use of “The is ”, reference can succeed even if nothing is “the ”. But the description can be used referentially even if something does fit the description, and indeed, that will be the case in most examples of referential use. Donnellan was merely highlighting the difference, and Recanati believes that the improper uses should be downplayed in the discussion, as they are not representative and cannot provide the basis for a theoretical account.
According to Recanati, a description, “the F”, does express a descriptive content. What happens in the referential use is a synecdochic transfer from the concept of the definite description to a directly
referential concept, what Recanati calls a de re-concept. This is a dossier, a file of information, like
the encyclopedia entries associated with proper names. So, the descriptive concept “man who succeeded Reagan” will transfer to a concept of George Bush. In normal cases, the description “the F” will be part of the information in the dossier. But the content of the dossier is truth-conditionally irrelevant. The concept of Bush might include “has a wife called Barbara”, but the truth of “George Bush is President” is independent of the truth of this and other information in the dossier.
There seems to be an inherent asymmetry at play when it comes to referential expressions.
Descriptions are attributive by nature, but can be used referentially. Proper names and indexicals, on the other hand, cannot that easily be used attributively. Descriptive names, such as “Jack the Ripper” (whoever he is), seem to form an exception. But Recanati rejects these to be any different from other names. They are not essentially descriptive, and we use and create them in the anticipation of “a richer state of knowledge in which we will be able to think of the referent non-descriptively” (Recanati 1993: 180). An example of this is the planet “Neptune”, which was named before it was actually observed.
There also seems to be a descriptive use of indexicals (chapter 16), like the impersonal use of “you” to mean “whoever”. This involves a transfer from the object which is initially referred to, to another object — a deferred reference, similar to pointing to a picture to indicate the man in the picture.
Recanati sees this as happening on a different level from the one involved in the transfer that makes a description referential. Indexicals are still inherently referential. The theory of direct reference works very well, but you have to delimit its field of operation. It does not explain deferred reference.
In Chapter 2 we will see how Amichai Kronfeld's work on reference, though independent of Recanati's, represents a similar synthesis of Fregean and Russellian views, and a similar integration of pragmatics into the philosophy of language.
The Linguistic Tradition
Strawson's presuppositions, Donnellan's distinction, Searle's speech act theory, and Grice's implicatures, though all published as works of philosophy, have been assumed by general linguistics. While this interaction has taken place, as well as the recent, deliberate cross-fertilization alluded to in Chapter 1.1.5., there is of course a vast linguistic tradition untouched by the formal philosophy of language.
Mainly, this has to do with delimitations of the research field. To many a linguist, the question of how language connects to the outside world is seen as a topic that falls outside linguistics proper. However, some traditions in linguistics, while not directly concerned with the problem of reference, assume a tacit view on the word-object relation. In this chapter, I will provide a brief overview.
Ferdinard de Saussure, as we meet him in the traditional 1916 edition of the Cours de linguistique générale, stresses that the relation between a name and a thing is a complicated affair: For some people a language, reduced to its essentials, is a nomenclature: a list of terms corresponding to a list of things. (.) This conception is open to a number of objections. It assumes that ideas already exist independently of words. It does not clarify whether the name is of a vocal of a psychological entity, for ARBOR might stand for either. Furthermore, it leads one to assume that the link between a name and a thing is something quite unproblematic, which is far from being the case. (The Saussure's point, that the sign (signe) is the totality of a sound pattern or signal (signifiant), and a
content or signification (signifié), seems deceptively similar to e.g. Fregean notions of terms and their
reference. But this is not so: the sound pattern is “not actually a sound”, but “the hearer's psychological impression of a sound”, and the signification is a concept. So the view held by “some people”, in the quote above, is not just naïve for assuming the name-thing correspondence to be a simple relation, but for the very invocation of external, physical manifestations such as word and object: “the two elements involved in the linguistic sign are both psychological. (.) A linguistic sign is not a link between a thing and a name, but between a concept and a sound pattern” (ibid.: 66).
Linguistics based on Saussurian principles has no place for reference. Its sole object is the relations internal to language. This can be seen as a methodologically motivated demarcation of the field of investigation, but at least in some cases of European structuralism and semiotics (including
the French thinkers mentioned at the beginning of section 1.1.5.), it implies a blunt rejection of the “Pour certaines personnes la langue, ramenée à son principe essentiel, est une nomenclature, c'est- à-dire une liste de termes correspondant à autant de choses. (.) Cette conception est critiquable à bien deségards. Elle suppose des idées toutes faites préexistant aux mots; elle ne nous dit pas si le nom est de naturevocale ou psychique, car arbor peut être considéré sous l'un ou l'autre aspect; enfin elle laisse supposer quele lien qui unit un nom à une chose est une opération toute simple, ce qui est bien loin d'être vrai.” (Saussure1916/1974: 97) problem of reference as anything of relevance to the study of language. Never has this been put more The word dog' exists, and functions within the structure of the English language, without reference to any four-legged barking creature's real existence. The word's behaviour derives from its inherent structural status as a noun rather than its referent's actual status as an animal. Structures are characteristically closed' in this way. (Hawkes 1977: 17) Leonard Bloomfield's American “structuralism”, on the contrary, was of a behaviourist persuasion: no psychological entities, only their physical manifestations, were legitimate objects of study for the (empirical) science of linguistics. American structuralism, too, had nothing to say about reference — not because of anti-realist tendencies like in European structuralism, but because it restricted itself to the description and classification of the observable manifestations of phonemes and morphemes. Questions of meaning and language use were never on the agenda.
The Chomskyan revolution brought back mental representations as legitimate entities in a theory of language. But again, Chomskyan linguistics is predominantly concerned with syntax, and consequently has little to say about how words hook up with things in the world. Among the few linguists who join Chomskyanism to semantics, Ray Jackendoff, writing explicitly on sense and reference, refutes the idea that the information language conveys is about the real world: “the world as experienced is unavoidedly influenced by the nature of the unconscious processes for organizing environmental input. One cannot perceive the real world as it is'” (Jackendoff 1983: 26). Reference is projection. We do not refer to things in the real world, but to our projection of the world; referring expressions map into conceptual structures. With this essentially Kantian picture, Jackendoff reveals a view on reference surprisingly similar to that of one of Chomsky's severest critics, Hilary Putnam, in his internalist incarnation, as presented in 1.1.4.
Noam Chomsky himself has only touched briefly on the problem of reference. His “Language and thought” lecture (Chomsky 1993) dismisses the philosophers' notions of thoughts, common public language and “a relation of reference between words and things and a mode of fixing it” as obscure notions of no service to the empirical study of language and its use. His main dissatisfaction with the tradition from Frege is the unwarranted conclusion that because people use words to refer to things, words themselves refer. This idea of fixed reference is connected to the philosophical notions of public shared meanings that Chomsky believes to be empirically unfounded and conceptually unrequired.
Part of Chomsky's point about shared meanings and fixed reference is, I believe, well taken.
No modern linguist would claim an objective criterion for, say, correct pronunciation, and it seems similarly unlikely that the meaning of a word should have an abstract existence independent of the individual speaker. As acknowledged by Kripke himself in his discussion of proper names, the causal chain theory cannot account for meaning change. A rigid application of the causal theory of reference — such, perhaps, as Putnam's claim that the meaning of natural kind terms ultimately derives from experts — implies a normative, linguistic conservatism (“people use the word that way, but it really means.”) of a kind that modern linguistics has deliberately dissociated itself from. (See also Marconi 1997: 97-105 for a good discussion).
Furthermore, Chomsky seems to support Jackendoff's view that we do not refer to things in is there an object London to which I am referring? If so, it is a very curious thing. (.) Referring to London, we can be talking about a location, people who sometimes live there, the air above (but not too high), buildings, institutions, etc., in various combinations. A single occurrence of the term can serve all these functions simultaneously, as when I say that London is so unhappy, ugly, and polluted that it should be destroyed and rebuilt 100 miles away. No object in the world could have this collection of properties. (Chomsky 1993: 22- Of course, Chomsky would not deny that there is an external world of things, or that we use language to talk about these things. Like in earlier linguistic traditions, the omission of reference studies is methodological. To judge from the quoted lecture, it is Chomsky's conviction that it is presently impossible to deal with such problems in a scientific manner, and that some questions of mind and meaning may well be among the “mysteries that will be forever beyond our cognitive reach” (ibid.: So, while Chomskyan linguistics is deliberately biased towards conceptualism (i.e. towards psychological objects), there is, as Pollard & Sag have it, less uniformity of opinion among semanticists on the issue of conceptualism vs. realism. (.) Since the mid-1970’s, the dominant paradigm for linguistic semantics has been that of Richard Montague. But Montague's approach was to characterize natural language meaning in terms of truth and entailment in abstract set-theoretic models (.); and there is no obvious sense in which Montague's possible-worlds models favor either a conceptual or a realist interpretation. (Pollard & Sag 1987: 4) The picture indeed is muddled. Newer semantic frameworks such as Situation Theory (Barwise & Perry 1983) favour a realist approach (one of the chief architects of this framework is John Perry whom we mentioned in the last chapter as a proponent of a neo-Russellian view on reference), but at the same time the idea of situations, as opposed to possible worlds, offer a more psychologically plausible account of how we process information. But there are also staunch realists who have embraced Chomsky's generative syntax and the Montagovian interpretations based on it as a possible safeguard against Continental anti-realist philosophy of language (Devitt & Sterelny 1987). As w shall see in a moment, one of the latter-day alternatives to Montague-style theories of meaning, Lakoff's cognitive semantics (Lakoff 1987), rejects the model-theoretic, possible-world approach for
being too metaphysically realist and failing to explain reference. This criticism seems a bit misguided, for it is not the task of formal semantics, the way it is being put to use in linguistics, to explain reference — rather, it is its strength that it allows us to disregard the issue in order to study the The main linguistic question has not been what a term refers to in the outside world, but which terms in an utterance refer to the same thing, i.e. which terms are coreferential, or stand in an
anaphoric relationship. Of modern semantic theories, Hans Kamp's Discourse Representation Theory
(Kamp 1984, Kamp & Reyle 1993) is the one most consciously developed to account for problems of this nature. It presents, among other things, a solution to the problem that pronouns can behave as genuinely referential terms as well as as variables bound by quantification. Kamp's theory is also an interesting compromise between the truth-functional approach to semantics and the approach that seeks to represent what a language user grasps in response to a linguistic input. But no more than other model-theoretic frameworks does Discourse Representation Theory seem to favour any particular view on how reference is established. (In chapter 3.4, I will use the DRT notation to demonstrate how a correct assignment of coreferentiality is not always enough to capture the truth-functional properties of a discourse containing referring expressions).
The whole formalist enterprise has become subject to intense criticism in later years. Model- theoretic semantics is characterized by operating sub specie aeternitatis — we choose to see the world It should be mentioned, though, that in the second edition “our views on the role of Chomsky's theory of grammar in a theory of mind and meaning have been completely reworked” (Devitt & Sterelny1999).
“from the outside” and evaluate the truth of the expression against the objective reality represented by the model. It is, as it has been put polemically, a semantics for God.
Putnam calls this idea of an objective reality metaphysical realism. George Lakoff simply calls it objectivism. In his rebellion against the objectivist tradition in linguistics, the difference
between descriptive and causal theories of reference is, along with much else, seen as “minor variations” within objectivism (Lakoff 1987: 174).
Lakoff's criticism consists of both empirical and logical arguments. The logical argument is a repetition of Putnam's Theorem (Putnam 1981), which was mentioned in 1.1.4.: compositionality does not work, for it is possible to change the reference (so that “cat” for instance means “cherry”) without affecting the truth-value of the sentence. Lakoff spends a number of pages refuting attempts to relativize or disprove the theorem. Two seem particularly relevant in this context: The meanings of words are fixed by convention. “Cat” does not mean cherry. Lakoff's reply: If we allow “cat” to mean cherry (and maintain that really it doesn't), then we don't get the right reference. What Putnam has proven is exactly that the incorrect reference does not change the meaning of the sentence — if meaning is solely defined by truth values in possible worlds.
We can modify model-theoretical semantics to define the meaning of a sentence as truth value plus something else. Reply: David Kaplan (Kaplan 1975/1985) has proposed the term character for this missing aspect of the meaning. The addition to the model must be a mathematical function from the meaning of each element to the character of the sentence. But a function can only operate on structure, and cannot differentiate between different input with identical structure. If character is determined by something other than a function, this is a de facto goodbye to formal logic/model The conclusion Lakoff draws from this is uncompromising. A passage of his exposition is titled, “Why There Can Be No Objectivist Account of Reference” (Lakoff 1987: 238). It seems, though, that what he is really challenging is formal semantics as a theory of lexical meaning. Like Putnam, Lakoff is a relativist without being an absolute relativist. He maintains a “basic realism”, meaning that there is a real world (including human experience) and that truth is not just consistency within a system, but the system must relate to the world. In collaboration with Mark Johnson (Lakoff & Johnson 1980), Lakoff has developed a philosophy of “experientialism” compatible to Putnam's internalism. Central to experientialism is the concept of embodiment: Structure gets its meaning from
being related to our preconceptual experience of being bodies, not abstract mental states.
Cognitive linguistics (of which Lakoff's cognitive semantics is only a particular branch) is not the only modern movement in response to Chomskyan and formal linguistics. Another alternative 1.3. Reference and Computational Linguistics is the functional approach, which takes language use to be the main object in the study of language
– or, in other words, makes pragmatics, rather than syntax, the core discipline of linguistics. It is also in the field of pragmatics that the aforementioned influence from and interaction with philosophers like Strawson, Donnellan, Searle, Grice, and Recanati have been felt. Sperber and Wilson's Relevance
theory (Sperber & Wilson 1986), extending the Gricean notion of relevance (which was mentioned
in section 1.1.2. and will make another appearance in 2.1.), has been put to use in explaining reference assignment of both anaphoric and more complex kinds (Matsui 1998, 1999). But there is also a functional/pragmatic school that is strictly anti-formalist and rely more on a Wittgensteinian view on language as activity, which implies that you cannot specify just one relation of reference.
It is an obvious conclusion that there is no consensus on reference within the field of linguistics. In the case of many linguistics inquiries, the problem of reference is simply irrelevant. In those cases where it is not, the opinions are divided between a cognitivist approach which takes reference to be merely a mapping into conceptual structure, and a pragmatic approach that in some cases takes inspiration from the realist philosophers of language. Be that as it may, the fact remains that we do use language to talk about things in the world. There may be important methodological arguments why some aspects of this problem will, at least for the time being, remain more suitable to philosophical than empirical investigation. Still, there is no fundamental reason why reference should not be considered a legitimately linguistic issue. In the next section, we shall see how, at least, computational linguistics cannot ignore it.
Reference and Computational Linguistics
The problem of reference enters into computational linguistics in a variety of ways. The easiest way to see its relevance is to imagine a robot with a natural language interface, maybe a sort of robot storeman that can respond to verbal orders such as, “Put that box on shelf number five”. Maybe we would also prefer such a robot to be able to answer us, report on tasks performed (“I placed the red box on shelf number three”), ask for clarification (“Do you mean the big box or the small box?”), or ask for further instructions (“Where shall I place this box?”). In designing such a system, we would very much be faced with the problem of relating expressions to objects in the environment. In 1.3. Reference and Computational Linguistics computational linguistics, the abstract problem of reference suddenly becomes a quite practical problem. In that case, it is tempting and natural to seek a pragmatic solution, and that is indeed what has been going on in much work within computational linguistics. Theories of reference as a mapping into conceptual structure would seem of little immediate help, a bit of a philosophical luxury, for the box is there, and we want our robot to move it. But in the really long run, a theory of reference is unavoidable — pragmatic solutions will take us only so far.
We need not go into robotics for the problem to become relevant. Much of the same applies, if we merely wish to refer to objects on the screen, or in a database. In look at Terry Winograd's classic Lisp programme SHRDLU (Winograd 1972), which performs a task much like that of the imaginary robot, but in a simulated “microworld”. In the following subchapters, I will present other relevant work in computational linguistics: models of discourse, an important tool
for the restriction of possible referents, this having to do with the anaphoric relation, and computa-
tional models of referring; generation of referring expressions as part of text generation systems,
including the development of algorithms for choosing the descriptive content; and referring phenomena in multimedia. A surprising number of these issues is anticipated in Winograd's pioneering work.
Winograd's “natural language understanding programme” — which only later became known under the nonsense acronym SHRDLU — is perhaps the single most impressive breakthrough that ever occurred in natural language processing. While not purposely designed to address or discuss problems of reference, it nevertheless deals with, and provides solutions to, a number of relevant issues (as well as many more topics of importance to computational linguistics and artificial intelligence).
SHRDLU operates in a simulated “blocks world” containing a table, a box, and several blocks and pyramids of different sizes and colours. SHRDLU can “see” everything in the scene, and has a hand with which it can manipulate the objects: stack the blocks, put them in the box, etc. I will quote extracts from a sample dialogue with SHRDLU, including Winograd's explanatory remarks (Winograd 1972: 8-15), and comment upon the programme's treatment of references. I keep Winograd's original ⑧✫⑨✵⑩✵❶❸❷✵❹❳❺✪⑩❼❻❾❽✵❶❸❿➁➀❚➂☞➃➁❽✵➄❸➅✵❷✵❹☞➆ In chapter 5 of my Tietgen prize thesis (Jørgensen 1999), I have addressed the philosophical question whether there is a fundamental difference between a robot interactinglinguistically with its environment, a simulation thereof, and reference as performed by humans.
1.3. Reference and Computational Linguistics ➉✰➊✔➋❤➌✙➍➎➌✙➏✡➋➑➐➓➒☞➔✺➌✜→➣➋✫↔↕➌❤➙❚➇❘➈✰➛➜→✥➊☞➋☞➔❱➝ ➋✻➌❭➟✵➠✜➏✝➒❅➞✺➟✣➐❭➐❤➒✣➔✔➡ ➔❳➟☎↔➤➡✣➋➥↔✵➏✡➟❩➦➥➝ ➞✔➧❜➠➑➦➨➏✹➊✫➋◆↔✡➋✔➡◆➩✏➫ ➏➯➊✻➒☞➡❜➏✼➟❅➞✫➫ ➏✟➟✔➲✍➲➯➩✺➍❭➲✘➝ ➔☞➳❜➒❅➌➑➦✔➒☞➞✔➋➨➲✼➟☎↔✵➏❚➊☞➋❅➳✏↔✼➋✫➋☞➔❜➟✣➔☞➋❜➒✣➔✔➡◆➐❤➟✔➵✏➝ ➔☞➳➨➏❚➊☞➋❅➳✏↔✡➋☞➋☞➔❜➟✣➔☞➋❜➒☞→€➒✻➍ There are two red blocks in the model, one bigger than the other, so despite the indefinite, this is a distinguishing description. SHRDLU is capable of handling cases where more than one object fits the indefinite description. (See query 19 below).
⑨☞❿✟➀❚❻✫➺✵⑩❳➻➽➼☞➂❾⑩☞➾✵➀❚❻☎➚❼❶❸➃✟➆ ➉✟➊☞➋✏↔✡➋❜➒✵↔✼➋➨➏❚➊✏↔✡➋☞➋❅➦✺➍✣↔✍➒✣➐❜➝ ➔❭➏❚➊☞➋❜➌✺➞➑➋☞➔☞➋☎➭☞➌➎➟❜➏❚➊☞➋❅➦✏↔✍➟✣➳✏↔✍➒✣➐Ô➡☞➟✣➋✔➌❳➔✔➟✻➏✹➊➥➝ The definite is taken to indicate a unique description, in the Russellian manner. (Winograd notes that “in more general language use, definiteness is often used to convey new information”, Winograd 1972: 156). When the NP does not refer to a unique object, SHRDLU replies that the information is insufficient, rather than just picking any pyramid. It will need more information. This points towards the fact that referring is often a cooperative endeavour, to be discussed in Chapter 1.3.2.
⑨✔Ö➽❶➤×☞➃❾❻❾❽✵➄❸➅✵❷✵❹❭Ø❈➼✵❶❸❷✵➼❾❶❸➺❜➻Ù❻✵➄✍➄❸➂✵➀✟➻➽➼☞❻✵×❜➻Ú➼☞➂❅➅✟×☞➂❅➾☞➅✰❺❱❻✵➀❚➂❾➼☞➅✟➄❸➃✟❶✍×☞❿❑❻✵×☞➃➁⑩✣❺✫➻✥❶ ➻✥❶✍×✔➻✘➅◆➻➽➼☞➂❾❽☞➅✵Û✵➆ ➉✟➊☞➋❜➌✜➍✔➌✜➏❚➋☞➐✧➊✔➒☞➌❤➏✡➟❜➲Ù➝ ➳✏➠☞↔✡➋❜➟✵➠✜➏✟→✥➊✔➒✔➏✩➝ ➌❳➐❭➋✔➒✣➔➑➏➯➩✺➍❼➙❪➟✣➔✫➋✔➛✩➒✏➔✔➡◆➙Ü➝ ➏✡➛✸➭☞➒✫➌❭→✥➋✏➫ ➔☞➳❜➡✣➋✔➡☎➠➎➞✜➏✘➝ ➔✺➏❚➋✔➡❅➟✵➠✜➏✪➏✼➟◆➧✺➋☞➋☞➦❱➠➎➌❅➝ As Winograd indicates, SHRDLU has to resolve pronominal and one-anaphora. This calls attention to several important facts: Language understanding involves both the reference relation from expressions to non-linguistic entities (in this case, data structures representing the objects of the blockworld), and the coreference relation holding between linguistic expressions. And while it is possible to study coreference without getting involved with external reference, as it is indeed often done in linguistics, it is hardly possible to study or simulate reference to external entities without dealing with at least 1.3. Reference and Computational Linguistics ⑨☞Ø❤➼☞❻✔➻✩➃✵➅✵➂☞➺❜➻➽➼☞➂❾❽☞➅✵Û❼❷✫➅✟×✔➻Ù❻✵❶✍×✣å ç✪⑨☞Ø❤➼☞❻✔➻✥❶❸➺❜➻Ú➼☞➂❾⑩☞➾✵➀❪❻✵➚❅❶❸➃❑➺✣❺✪⑩✵⑩☞➅✟➀✹➻Ù➂✫➃❃❽☞➾✣å ➋✏↔☎➏✹➊✫➋❳➌è➍➑➌è➏✹➋✫➐é↔✡➋✔➌➑➦✻➟✣➔✔➡✣➋✔➡❜➏✡➟◆➙✹➏❚➊☞➋❅➦✺➍✣↔✍➒✣➐❜➝ ➡☞➛➣➩➑➍❳➌➎➒✔➍✏➝ ➏➜➠➑➔✔➡✣➋✏↔✍➌✜➏✡➒✣➔✻➡❼→✥➊➥➝ ➞✔➊❼➦➑➍✣↔✍➒✣➐❜➝ ➏✝➒✫➌➎➌☞➠✔➐❭➋✔➌❤➏❚➊✔➒✔➏✝→✥➋❅➐❭➋✔➒✏➔❭➏❚➊☞➋❜➟✣➔✫➋❭ê✍➠➎➌è➏➯➐❭➋☞➔✺➏✘➝ “The pyramid” is still a uniquely identifying description, but the search space is no longer the entire model — SHRDLU is aware that a certain pyramid has recently been in focus. This kind of restriction of search space is studied in the discourse modelling we will look at in 1.3.2.
⑧èì✰⑨☞Ø❤➼✵❶❸❷✵➼❅❷✣❺✪❽☞➂❾❶ ➻❪➻Ú❶✍×☞❿❾➅✟×❳➻➽➼☞➂❜➻Ù❻☎❽✵➄ ❒✥➷➴➬€➷✟➪Ï❰✵Ð✥➬ ➍❩➏❚➊☞➋❑➌✜➍➑➌è➏❚➋☞➐í→€➟✵➠☞➫ ➳❾➳✏↔✡➋☞➋☞➔❑➞☞➠➑➩☞➋✻➛✸➭✰➩➥➠✜➏➜➌☞➝ ➒✔➌➎➧✜➋✺➡❳➙✹→➣➊✫➝ ➞➎➊❤➞✔➠➎➩✔➋✺➛✩➝ ➏☎➊✺➒✔➌✭➏➤➟❳➩✔➋❭➐❘➟✏↔➤➋❤➌➎➦✔➋➎➞✔➝ ➉✟➊☞➋❱➠➎➌➑➋❜➟✔➲✟➌☞➠➑➩✻➌✜➏✘➝ ➏✘➠✜➏❚➋❅➔✔➟✵➠✔➔✔➌ ➒✏➔✔➡◆➦✏↔➤➟✏➔✔➟✵➠➑➔✔➌❅➝ ➔❜➒✣➔✔➌✺→✥➋✏↔➤➌❳➐❤➒✏➧➎➋✔➌❤➏❚➊☞➋❜➡✵➝ ➌➎➞✺➟✵➠☞↔✍➌➑➋❅➐❤➟✵↔✡➋❼➔✻➒✔➏✘➠☞↔➤➒☎➫ Though presented as a natural language understanding programme, when it comes to referring expressions, SHRDLU also represented advances in natural language generation. Some of its basic strategies for this are given below. Contemporary algorithms for the generation of referring expressions ⑧✺ð✵⑨✵❶❸➺❜➻Ú➼☞➂☎➀❚➂❅❻❾➄ ❻☎➀❚❿✵➂❾❽✵➄❸➅✵❷✵❹❅❽☞➂✵➼✵❶✍×☞➃❑❻❾⑩☞➾✵➀❪❻✵➚❅❶ ⑨✵⑩✣❺✫➻✩❻❅➺✵➚❭❻✵➄✍➄✏➅✟×☞➂❼➅✟×✔➻✘➅❱➻Ú➼☞➂❅❿✟➀❚➂✫➂✵×❼❷✏❺✪❽☞➂❼Ø❤➼✵❶❸❷✵➼❅➺✣❺✪⑩✵⑩☞➅✟➀✡➻✘➺❅❻❾⑩☞➾✵➀❚❻✵➚❅❶❸➃✟➆ ➫➑➟✣➔✫➋✔➛✥➐❜➠➎➌✜➏♠➞✺➟✣➔✺➏✘↔➤➒✫➌✜➏♠→✭➝ ➏❚➊❜➏❚➊☞➋❱➋✔➒✵↔❪➫ ➋✏↔♠➦☞➊✏↔✍➒☞➌➑➋◆➙❪➒ö➫ ➟☞➞✔➧✜➛✸➭✏➌➎➟❩➏✹➊➥➝ ➌❅➐❭➋✔➒✣➔✔➌❅➙❪➒◆➌➑➐❤➒✵➫ 19. shows the treatment of an indefinite without singular reference, 20. another handling of one- ⑨☞Ø❤➼☞➾❅➃✟❶❸➃❑➾☞➅✰❺❱➃✵➅❱➻Ú➼☞❻✔➻✕å 1.3. Reference and Computational Linguistics ➙❪➉✟➊✔➒✔➏✡➛♦↔✼➋✺➲❚➋✏↔✍➌❤➏✡➟❜➏✹➊✫➋❳➒✫➞✜➏✘➝ ➔➨➏❚➊☞➋❜➒✣➔✔➌✺→✥➋✏↔ The mentioned action was: moving a pyramid “to get rid of it”. “That” refers back to the proposition of a former utterance. Such sentential anaphora represent a hard problem that is still mostly ignored ⑨☞❻❅ú➽➺✔➻✘➂☞➂✵⑩✵➄❸➂☞û▲❶❸➺❅❻❅➺✔➻✘❻☞❷✵❹❭Ø❤➼✵❶ ❷☎➼❼❷☞➅✟×✔➻✘❻✵❶✍×☞➺❜➻ÙØ✭➅❾❿✟➀❚➂☞➂☎×❼❷✣❺✪❽☞➂✫➺❼❻☎×☞➃❑❻❾⑩☞➾✵➀❪❻✵➚❅❶ ➉✟➊☞➋❜ü✵➠➎➟✔➏❚➋❅➐❤➒✵↔✡➧✜➌❅➝ ➞➎➒✔➏❚➋➨➏❚➊✔➒✔➏✟→✥➋❳➒☎↔✼➋❜➡✣➋✺➲✘➝ ➔☞➳❜➒❩➔✫➋✔→õ→€➟☎↔➤➡ ⑨☞❷☞❻☎➄➤➄✫➻Ú➼☞➂❾❽✵❶❸❿✵❿✵➂☞➺✔➻♦❽✵➄ ➅✵❷☎❹❊ú➽➺✣❺✪⑩☞➂☎➀✕❽✵➄❸➅✵❷✵❹➎û✙➆ ➵✻➋❜➟✣➩➑êý➋✔➞è➏✡➌❳➔✔➒✣➐❭➋✔➌ SHRDLU is capable of extending its vocabulary, including acquiring proper names.
The range of phenomena covered by SHRDLU is impressive, and those concerning reference are just a few of them. However, the principles governing the way referring NP's are built, are quite simple. Some things are unique, such as the table, the hand (which is used to manipulate the object), the user, SHRDLU itself. These can be referred to by using their “names”: “the table”, “the hand”, “you”, and “I”. Similarly with “superblock”, whose christening we witnessed in query 40. If the intended referent is not unique, the noun phrase must be built.
Apart from the unique entities, all objects in SHRDLU's little world belong to one of three categories: block, ball, or pyramid. A check of dimensions reveal some blocks to be cubes. If this classification is not enough to uniquely identify the object, the properties colour and size will be added. If this is still not enough, supported entities will be treated as a property (linguistically realized as “which supports” + the name of the object). If there are no supported entities, spatial relations to other entities will be used (realized as “which is to the right of” + name of object). “This still may not characterize the object uniquely in some situations, but the system assumes that it does” (Winograd 1972: 167). This generation strategy works well — but mainly because it operates in a microworld where the number of basic classes and properties is so limited.
Finally, another aspect of SHRDLU should be mentioned: planning. In artificial intelligence,
planning means a piece of self-programming consisting of making and following a strategy to achieve a goal. For instance, the system can be asked to place a red box on top of another red box. Maybe this
1.3. Reference and Computational Linguistics cannot be done without moving a little pyramid away from one of the red boxes. The programme is not explicitly asked to move the pyramid, but must make a plan in order to obtain the goal of having a red box on top of another. This is worth mentioning, because linguistic activity can be analysed as a planning problem. It is quite compatible with Speech Act theory to view communication as goal- oriented behaviour. In chapters to come, we shall see the idea of planning being put to use in natural Discourse Modelling, and Models of Referring A program such as SHRDLU transforms an input description into a searchable format and matches this against possible referents, in SHRDLU's case all the objects in the blockworld. Basically, the same method can be used to search for referents in other domains, for instance entities mentioned in the previous discourse. A lot of computational work on referring has consisted in restricting the search space for possible referents, especially by modelling the structure of the discourse and thereby provide a clue as to whether a previously mentioned entity is a likely antecedent.
In this area, the work of Barbara Grosz has been important, especially by introducing a formal notion of focus (Grosz 1981). Focus means the interlocutors' concentration on a small portion
of their (mutual) knowledge and perspective, a little part of their “shared reality”. (Of course, the things that are highlighted in the course of a dialogue do not have to be real — speaker and hearer can focus on the same, imaginary entities). Focusing is something the speaker and hearer do — things do not come into focus all by themselves, and focus is shifted all the time, indicated by linguistic clues. Foci, just like beliefs, are not necessarily shared by speaker and hearer, but they will be speaking under the assumption that they do maintain the same focus. Focus does not just mean concentrating on the same entities, but on using the same perspective on the entities (similar to the modes of presentation in the philosophical tradition).
Grosz applies this to the analysis of definite descriptions in task-oriented dialogues. There seems to be a two-way relation between focusing and definite descriptions. The choice of description is influenced by what is currently in focus, and descriptions influence focus (by bringing entities into focus, providing a perspective on them, etc.). When the focus shifts, the discourse can be segmented into subunits, focus spaces. When looking for a possible referent for a definite description, we no
longer have to search the entire universe, but can restrict the search to the relevant focus space. In a computational system, the entities will be represented in a knowledge base, and focus spaces will be 1.3. Reference and Computational Linguistics segments of the base. A hierarchical ordering of the spaces will then model the structure of the Grosz and Sidner (Grosz and Sidner 1986) later integrated this into a larger theory of discourse structure. This includes interaction between modules of linguistic structure (the sequence
of utterances), intentional structure (the purposes in the conversation), and attentional state, which
represents the focus. Where focusing, as mentioned above, was something the individual participants did, attentional state is focus abstracted to a property of the discourse. The attentional state is a set of focus spaces (focus structure), plus rules for adding and deleting spaces. All this form part of the cognitive state, which also includes knowledge and belief.
Other later work in this tradition includes the theory of centering (Grosz, Joshi & Weinstein
1995), a computational theory of anaphora, assuming a ranked list of potential antecedents. It claims that for each utterance, there is exactly one centre of attention, which is the entity most likely to be pronominalised. Centering theory has become a very popular approach in computational linguistics (see for instance Cristea et al. 1999). In recent years, the central notion of a focus stack has been challenged by a cache model (Walker 1996). The cache is a limited capacity working memory separate
from long-term memory. This separation makes the information in the cache “almost instantaneously accessible”. It is an ongoing discussion whether the cache model really is more efficient and psychologically plausible than the focus stack (Grosz & Gordon 1999). For other computational work, independent but comparable to Grosz', see for instance Webber (Webber 1983, 1991).
Grosz' work deals with restricting the search space, but still assumes that the description must match all the attributes of the intended referent. Realizing that this is often not the case in real-life conversations, Bradley Goodman concentrates on referent identification failures (Goodman 1986).
When people speak, their interpretation of each other's utterances sometimes go wrong because they do not have the same background, beliefs, and goals. Reference problems are just one instance of miscommunication. This is not just interesting from a theoretical perspective — to improve the robustness of natural language systems, we will need a strategy for repairing reference failures.
Goodman views the referring process as one of negotiation. Negotiation can take place
among speakers, but Goodman focuses on self-negotiation, the way a hearer seeks to repair a failed reference. A description of an object can be insufficient for a number of reasons. It can be imprecise, confused, ambiguous, or it can even be misleading because it is overly specific, or because of confusion having to do with focus. In the repair, the hearer can apply many kinds of knowledge: linguistic knowledge, perception, knowledge about the discourse, and information gained from a trial 1.3. Reference and Computational Linguistics and error process. (In order to find the referent for “the salty-tasting macaroni”, the hearer would have to taste his way to one). The actual repair process is one of relaxation: in order to find a potential
candidate, the speaker must relax the demands given in the referring expression.
We will illustrate this with an example of relaxation using linguistic knowledge. “It is implicitly clear that the structure of a noun phrase can affect its meaning in many ways (.) Since there is no one-to-one mapping between a noun phrase's structure and its meaning, it is the hearer's job to determine how the structural information is being used” (Goodman 1986: 284). One of the ways structure can affect meaning, is seen in the case of the relative clause. In “the long blue tube that has two outlets on the side”, the information in the relative clause is used to confirm that we're dealing with the proper referent. Goodman's “claim is that the speaker would use the relative clause version to emphasize the information in the relative clause” (ibid.), as opposed to the prepositional version “the long blue tube with two outlets on the side”. This means that information in a relative clause is less likely to be erroneous, and consequently a less likely candidate for relaxation, than information from a prepositional phrase. But the kind of relaxation order resulting from this “may be modified by other

Source: http://www.stigw.dk/compref/files/chap1.pdf