The links of causal chains

This paper is about the Causal Theory of Names, as out-lined by Kripke in Naming and Necessity . The paper argues that causal chains which connect users in command of a name N with those present at the baptismal event in which N was introduced are branches of networks of ‘ N - labelled ’ entity representations in the minds of past and present users of N. These networks of N -labelled entity representations are special cases of networks that result in general from the use of referring expressions. Such networks are an important part of the fabric that holds a speech community together and point towards a view of language as a social practice. The theory of networks and chains is developed within MSDRT ( ‘ Mental State Discourse Representation Theory ’ ), an extension of DRT designed for the description of utterance contents, propositional attitudes, mental states and the ways in which mental states change in the course of verbal communication. The last section of the paper explores the view of languages as social practices somewhat further in the light of the network theory developed in the sections leading up to it.


| ORIENTATION
This paper is about just one part of Saul Kripke's many lasting contributions to logic and philosophy.But it is about a central part, about how names work in language and thought, about their uses in overt and silent speech and writingand, just as importantly, about how they do not work but had been assumed to work by practically everybody before Naming and Necessity turned the world of logic, language, metaphysics, and epistemology right-side up (Kripke, 1980).
I was part of that world, and my world also was made anew by Naming and Necessity.And I also could witness, from that point onwards, the snowballing literature that came in its wake.Over the years, that literature has increased to the point that it is difficult to maintain an overview, and I am well aware that I have much less of an overview than many others.But in the part of it that I have observed, the predominant focus has been on the logical and semantic implications of what Naming and Necessity has to say about proper names: that names are rigid designators and exactly what that entails for the meaning and use of sentences containing names and for the general semantics and logic of languages containing such sentences.
What Naming and Necessity has to say about proper names, however, also has an important social dimension.It is a familiar observation that the use of language has a double function.We use language to tell others the things that we want them to know, or to get them to do certain things, and so on.But the utterances that competent speakers produce for these various purposes also serve as displays of how language is to be used, as samples of how it works.This second aspect is especially plain and prominent when utterance recipients are not fully competent speakers, schoolchildren with a foreign background, for example, who struggle to learn the local language from the lessons they attend while trying to figure out at the same time what they are being told about the subjects they are taughthistory, geography, mathematics, biology, and the like.But utterances by competent speakers do not only play their role as samples of proper use when the recipients are not yet fully competent users of the language (according to the informal and not very well-defined criteria for linguistic competence that people willy-nilly go by).They also can serve effectively as samples of language use when their recipients are what we consider competent.There just is no practically applicable notion of complete language competence, and certainly not for human languages today, with their steadily growing special vocabularies to keep apace with cultural diversification and scientific progress.For pretty much all of us, language acquisition never stops.
That utterances serve as samples of how their language functions is a tired triviality, which bears no repeating unless followed by something more specific.More specific, and of genuine linguistic and philosophical interest, are the particular mechanisms that make it possible for utterances to serve their recipients as examples of proper language use while at the same time getting the messages across that the speaker or author wants to communicate.Such mechanisms are of interest in their own right and merit closer investigation on that account.But their investigation can also gain us a better understanding of the essential intersubjective dimension of human languages, as tightly structured practices shared by speech communities that can change in the course of time but that nevertheless have the remarkable degree of stability without which verbal communication would be impossible.It is because of their stability that languages can be fruitfully studied as abstract, user-independent systems in the spirit of for instance (Morris, 1946), (Carnap, 1947) or (Montague, 1970b) and why so much progress has been made in the past 50 years by treating them that way.The mechanisms I have in mind have the potential to explain the stability of language, while accounting at the same time for their capacity to change.
For the most part, serious investigation of such mechanisms has been getting under way only in recent years. 1 But one such mechanism, the causal theory of names that is outlined in Naming and Necessity, has been known to us for more than half a century.Recall the central 1 For some remarks on a couple of recent developments in this direction, see section 5.
THE LINKS OF CAUSAL CHAINS ideas of the causal chain theory: Proper names do not refer via unique satisfaction of some descriptive content, but because of how they originate by some form of "baptismal" ceremony and then spread through the language community.That is, a name N starts out when one or more persons introduce it as "label" for some entity r, thus providing those who are in on the act with a convenient means to refer to r.From this initial core of N-users, the command of N can then be passed to others: a person S who is in command of N as name of r can pass her command to some other speaker H who did not yet have it, either by simply using N in utterances that say something about its referent r or by introducing H explicitly to the use of N. 2 Either way, H can pick up the name from what he is told, or hears or reads, and then start using N himself to refer to r.
It is through such transfers that members of the speech community at large are inducted into the subcommunity of those who are in command of N as name for r.In other words, at any time after the introduction of N as a name for r this subcommunity is composed of (i) members of the "baptismal core," who were a party to the introduction of N as name of r (assuming that any of those are still alive); and (ii) speakers who are connected to one or more people in the core by some "causal chain" of individual transfers of the command of N. Because of these transfers, the subcommunity changes all the timeand with it, indirectly, the speech community as a whole.But at the same time the transfer mechanism guarantees the stability that matters: the reference of N is in principle always preserved.
In this paper, I want to look more closely at the transfer mechanism that is responsible for the links of causal chains.But for what I want to say we need a formal framework.The one I will be using is MSDRT (short for 'mental state discourse representation theory', an extension of DRT (discourse representation theory).Because I do not assume that readers of the present issue of Theoria are familiar with either framework, I will provide introductions to both DRT and MSDRT, stripping them down to the minimum that I will need for the remarks on causal chain links that follow.Section 2 will be devoted to DRT and section 3 to MSDRT.Section 4 applies MSDRT to causal chains and their links.The central notion developed in that section will be that of a network of linked entity representations.Causal chains will be branches within such networks.Some speculative remarks on the generalisation of the approach of section 4 will follow in the concluding section 5.

| DISCOURSE REPRESENTATION THEORY (DRT)
The original motivation for DRT were certain problems in the semantics of tense and aspect.A proper account of the temporal and aspectual properties of tenses must have something to say about how certain tenses relate what is described in the sentences containing them to what is described in earlier sentences. 3But DRT owed its early visibility to what it has to say about the so-called donkey problem, a problem from the medieval scholastic literature that had gained prominence with twentieth-century philosophers and linguists through its discussion in (Geach, 1962).One example of the problemonly one is needed hereis the pronoun it in the 2 For discussion of different ways in which one speaker can introduce another one to the use of a name, see (Kamp, 2015).

3
The first tenses that led to this conclusion were the French past tense forms passé simple and imparfait.These are both past tenses, but they differ in that the passé simple tends to express temporal progression: an event described by a passé simple is typically understood to have occurred after the event described by a preceding sentence, usually the immediately preceding one.In contrast, the imparfait normally conveys simultaneity between the state of affairs described by its sentence and the event or state described by a preceding sentence.A similar contrast to that between passé simple and imparfait can be observed for the simple past and the past progressive of English; compare for instance the two-sentence discourses "John turned around.The man behind him pulled his gun." and "John turned around.The man behind him was pulling his gun."For discussion see e.g.(Kamp et al., 2011).consequent of the conditional sentence (1), which is anaphoric to the indefinite NP a donkey in the conditional's antecedent. 1 ð Þ If bill owns a donkey, he beats it: Geach observes that (1) has truth conditions roughly captured by the paraphrase that Bill beats every donkey that he owns.Sentences like (1) present a difficulty for truth-conditional natural language semantics along the lines of Montague, 4 the widely accepted standard for formal semantics in the late 1970s.More precisely, if we follow Geach in assuming that (1) has the "universal" truth conditions given by the paraphrase above, then such semantics is confronted with a dilemma: either the truth conditions come out wrong or the pronoun becomes a "dangling" variable and the sentence fails to get a proper interpretation altogether. 5 The account that DRT offers for sentences like (1) is based on the following assumptions.Its point of departure is that the donkey pronoun problem of sentences like (1) is at bottom a discourse phenomenon, much like the interpretation of tense forms in multi-sentence discourses of the sort described in footnote 3. (And by the way, the tense forms mentioned there also manifest their anaphoric properties in conditionals and other complex sentences).To deal with Geach's problem, it is therefore recommendable to start with a donkey discourse, such as (2). 2 ð Þ Bill owns a donkey: He beats it: According to DRT the interpretation of this sentence takes the form of first constructing a semantic representation/logical form6 for the first sentence and then making use of that representation in constructing a semantic representation for the second sentence.The logical form for the two-sentence discourse in (2) is then obtained by merging the representations for its two sentences.
Formally the representation of the first sentence is that in (3).
(3) is an example of a so-called discourse representation structure or DRS.DRSs are DRT's logical forms.A DRS consists of two parts, (i) its universe (in (3) this is the part above the horizontal bar in the middle), which is a set of discourse referents (or drefs); and (ii) its condition set (the part below the bar), a set of DRS conditions.Discourse referents represent entities, DRS conditions are typically open formulas which may contain drefs occurring in the universe of the 4 See in particular (Montague, 1970a), (Montague, 1970b).

5
There has been much debate over the question if (1) really does have the truth conditions Geach assumed.For present purposes, this matters little because other readings that have been suggested present similar difficulties for a Montagovian semantics.And in any case, there remains the question how to deal with "donkey discourses," such as (2) below.
DRS (as well as, often, drefs not occurring there) in argument positions of predicates.Just as formulas of predicate logic can be regarded as formulas belonging to particular languages of predicate logic (determined by their non-logical vocabularies), so DRSs can be regarded as belonging to particular DRS languages.(For instance, a DRS language that includes (3) among its formulas will have the 1-place predicates Bill', donkey', and the 2-place predicate own' in its vocabulary).
DRT comes with an explicit model-theoretic semantics for its DRS languages.The models for these languages are like models for predicate logic.For instance, a model M for a DRS language that contains (3) as one of its "formulas" will assign to Bill' and donkey' subsets of its Universe U M and to own' a 2-place relation-in-extension, that is, a set of ordered pairs of members of U M .(On the assumption that a name is the name of one and no more than one individual, the extension of Bill' in M will be a singleton.If you do not like this as an account of proper names, do not worry.The point of this paper is to do better).
In the model-theoretic semantics for DRS languages, the drefs in the universe of DRSs are interpreted existentially.This existential interpretation results from the general principle that a DRS K = < U,C > is true in a model M iff there exists a verifying embedding of K in M -a function f from U into the Universe U M of M such that any condition γ in C is true in M under the assignment that f provides for the drefs in γ. (Example: (3) is true in a model M if there exists a function f that maps b to an individual b in U M and y to an individual y such that b is in the extension in M of Bill', y is in the extension in M of donkey', and the pair <b,y> belongs to the extension in M of own').
The construction of a DRS for the second sentence of (2) makes use of (3); (3) serves as context for its interpretation (just as the first sentence of the two sentence discourses in footnote (3) provides the context for the interpretation of the sentence following it).The use that the DRS construction for the second sentence of (2) makes of ( 3) is determined by an interpretation rule for anaphoric pronouns, according to which personal pronouns can be construed as anaphoric to drefs in the universe of the context DRS.Using this rule, it is possible to construct (4.a) as representation of the second sentence, in which the dref x, introduced to represent the pronoun he, is identified with its "anaphoric antecedent" b in the universe of (3); and the dref z, introduced to represent the pronoun it, is identified with y.After its completion, (4.a) can be merged with the representation (3) of the first sentence.The result, shown in (4.b), is the logical form for the two sentences of (2) together.
To deal with donkey sentences like (1), conditionals have to be treated as built from (i) an antecedent which describes a situation (or a set of possible worlds) and (ii) a consequent, which gives additional information about the situation (or the worlds in the set) described by the antecedent.In this case, the semantic representation (i.e., the DRS) of the antecedent provides a context for the interpretation of the consequent in the same way that the DRS for the first sentence of (2) provides a context for the interpretation of the second sentence of (2).Note well that to make this explicit three further stipulations are needed: i.The DRS for a conditional like (2) must be a representation in which the DRSs for antecedent and consequent are connected by an operator that represents their conditional relationship.DRT uses the sign ')' for this purpose; it forms DRS conditions K ) K 0 from DRSs K and K 0 , their antecedents and consequents.Thus, the DRS for (2) takes the form shown in (5), with empty universe and a condition set whose only member is a conditional DRS condition of the form K ) K 0 : ii.To construct the DRS in (5) for the sentence in (1), a rule is needed that converts the part of (1) that expresses conditionalitythat is, the combination of if in the subordinate clause of (2) and the comma that separates it from the main clauseinto the operator ) that connects the DRS for the if-clause with the DRS for the main clause.Furthermore, it must be stipulated that the DRS that stands to the left of ) can serve as interpretation context for the DRS on the right.And finally, iii. the definition of truth for DRSs in models must be extended with the following clause for DRS Conditions of the form K ) K 0 : (6) An embedding function f verifies DRS condition K ) K 0 in model M iff for every extension g of f to the universe of K 7 that verifies K there exists an extension h of g to the universe of K 0 which verifies K 0 . 8DRT contends that these stipulations are natural; they capture the intuitive semantics of the words and grammatical constructions to which they apply and do this in a conceptually transparent manner.In fact, most of the work that goes into DRT-based treatments of natural language fragments consists in finding and correctly stating the DRS construction rules for the different words and grammatical constructions of the fragment; in this regard DRT is no different from other approaches to the semantics of natural language that conform to the standards of precision and explicitness set by the work of Montague.
Some DRS construction rules relate the DRS under construction to the context DRS that has been constructed from the preceding part of the discourse.(Our only example of such a rule so far is the one for anaphoric pronouns that we needed to construct the DRSs for (1) and ( 2), but for present purposes this example is all we need).It is rules of this kind that make DRT into a genuine theory of discourse semantics and not just a semantic theory for single sentences in isolation.But for this very reason they present a special challenge: 7 That is, a function g that coincides with f on the domain of f and whose domain is the extension of the domain of f with the universe of K.

8
Note that the occurrences of the drefs b and y in the consequent DRS of (5) are "free" in this DRS in that these drefs do not occur in its Universe.When clause ( 6) is applied to the )-condition of (5) as part of applying the verification definition for DRSs in models to DRS (5), then these drefs are bound "from the outside": the functions g spoken of in (6) already assign values to b and y, which are then passed on to their extensions h that must verify the consequent DRS.There are also DRSs K with occurrences of drefs that are nowhere bound within K in that they occur neither in the Universe of K itself nor in the Universe of any of its sub-DRSs.Such DRSs are like open formulas of predicate logic in that they can be evaluated for truth or falsity only relative to an assignment: a function from the unbound drefs of K to entities in the model that is determined independently (e.g., by the context in which the content represented by K is being used).
The structures they relate to each otherthe partly constructed representation for the current sentence and the already constructed representation that serves as contextmust have forms that make it possible to state these rules in such a way that they produce the right results whenever they are applied.It was the attempt to meet this challenge that led to the representation format of DRSs.And it was thought to be an appealing feature of DRT that it had succeeded to meet this challenge by adopting a representation format for its DRSs that enables them to do both things at the same time: correctly capture the content of the bits of language for which they are constructed as logical forms and serve as discourse contexts for the construction of DRSs for further bits.
It is tempting to see this feature of DRTthat the same structures serve as content representations for what has been interpreted already and as contexts for the interpretation of what comes nextas evidence that DRT captures some important aspects of how the information that human interpreters obtain from what they read or hear is mentally represented.Whether this evidence is to be taken seriously, or how seriously, has been a bone of contention from the time when DRT became accessible to a wider public.Some advocates of DRT have been attracted to the approach because they see it as one that can tell us something about how language is processed by humans.(I was one of those advocates, and I still am up to a point, albeit more guardedly).But others who have been sympathetic to DRT accounts of linguistic phenomena such as nominal and temporal anaphora and who have adopted versions of the theory in their own work (see Partee (1984), (Geurts, 1999)) have been sceptical of a mentalistic take on DRT.As a consequence, the status of DRT within the wider context of natural language semantics and cognitive science has remained somewhat ambivalent.No such ambivalence is possible for MSDRT, to which we turn in the next section.As indicated by its name, MSDRT is explicitly about the mental states of human agents.
Irrespective of whether the form of DRSs has psychological import, however, DRSs are semantic representations; they are not just convenient, formal terms that the theory uses to describe semantic values of natural language expressions in models, in the way that Lambda Calculus terms are used in Montague grammar.As shown by the examples we have discussed, the form of context DRSs plays its part in DRT's account of pronominal anaphora.That DRS form is essential here is demonstrated by Partee's marble example: 7 ð Þ a: One of the marbles is not in the bag: It is under the sofa: b: Nine of the marbles are in the bag: It is under the sofa: The first sentence of (7.a) and the first sentence of (7.b) have the same propositional content: they determine the same truth values in every model.Nevertheless, the pronoun it in the second sentence of (7.a) is a perfectly good way to refer to the missing marble, whereas it is marked or impossible in (7.b). 9 DRT is thus a genuine logical form framework for doing natural language semantics: It assigns formulas of some formally defined representation language (one of the possible DRS languages) to expressions from the chosen natural language fragment, and the role played by those representations goes beyond that of denoting semantic values.This general architectural feature of DRT is inherited by MSDRT, but with the difference that MSDRT's representations are mental representations. 109 It is clear that this difference must have to do with the difference in form of the first sentences of (7.a) and (7.b) because that is the only difference between (7.a) and (7.b).In DRT, this difference is captured by the DRSs for the two first sentences.The DRS for the first sentence of (7.a) has a dref for the missing marble in its Universe, which is accessible to the pronoun it in the second sentence.The DRS for the first sentence of (7.b) does not have such a dref.(This follows from details about DRS construction that have not been discussed here, including a precisely formulated rule for the interpretation of anaphoric pronouns.For details, see for example (Kamp & Reyle, 1993)).10 For more information about DRT: An excellent introduction, by Beaver, Geurts, and Maier, can be found in the Stanford Encyclopedia of Philosophy.A detailed introduction to DRT from the early 1990s is (Kamp & Reyle, 1993), which presents DRT The original motivation of MSDRT was the conviction that a more fine-grained semantics is needed for attitude attributions than the possible world-based accounts that have dominated formal semantics since the 1960s.All these world-based accounts are variants of the proposal made by Hintikka for the semantics of belief and knowledge reports in (Hintikka, 1962).Hintikka's proposal is appealing for its remarkable formal simplicity: The belief state, say, of agent A (at some given time t) is characterised by the set Bel(A, t) of all those possible worlds that are compatible with the totality of beliefs that A holds at t.And a belief report of the form 'A believes that ϕ' is true if the proposition expressed by ϕ is entailed by the totality of A's beliefs at t; formally: if the set ([ϕ]) of worlds in which ϕ is true is a superset of Bel(A, t).Knowledge attributions are handled likewise on the basis of the set Know(A, t), consisting of all worlds that are compatible with everything that A knows at t. Later proposals in this general spirit have been more complex.An example is the proposal that attributions of the form 'A wants it to be the case that ϕ' are true iff worlds in which ϕ holds are (other things being equal) preferred to worlds in which ϕ does not hold (Heim, 1992).But a problem with all such proposals is that they strip semantic content down to something that abstracts from all aspects of the form of the attribution.That is, to put the matter more precisely, they abstract away from the form of the syntactic complement to the attitudinal predicate (such as believes or wants), which describes what it is that is attributed as belief or desire.And that leads to problems when this form is of the essence to the message that the attribution is meant to convey, as it often is.For instance, many of the attitude attributions we make are made as part of an effort to explain why the attributee did the thing or does the things they did or are doing.But agents do the things for which attitude attributions offer explanations on the basis of bits of practical reasoning from their beliefs and desires, and in those bits of reasoning they have to rely on the ways in which they have represented the contents of those beliefs and desires.And attributors try to remain faithful to what they take those content representations to be in how they phrase their descriptions of the contents of the beliefs, desires, and so forth that they attribute in their attributions.The forms of those descriptions matter and a credible semantics for such attitude attributions needs to be sensitive to the forms in which attitude contents are described.The possible worlds approach is incapable of that.
MSDRT was developed as an alternative approach that can be sensitive to the form of attitude attributions.One of its ingredients is the logical form architecture that it inherits from DRT.But it is only one.The logical forms that MSDRT proposes for attitude attributions consist of two levels: a first level that offers a description of the relevant parts of the attributee's mental state and a second level at which this description is embedded as one of the arguments of attitude attributing predicate Att.The predications involving Att are DRS conditions that can occur in the condition sets of DRSs together with DRS conditions like those we encountered in section 2.
For our use in this paper, it is first and foremost the first of these two stages that matters, but certain aspects of the second stage are relevant as well.
treatments of a fairly broad range of linguistic phenomena, including plurals and tense and aspect.As is plain from this book, most of the hard work in DRT goes into the crafting of DRS construction rules for grammatical constructions and vocabulary items from the natural language fragments to which the framework is being applied.In this regard, DRT is like other formal semantics frameworks.An awkward feature of (Kamp & Reyle, 1993)a complaint that has been voiced repeatedly from the time when the book appeared and one that I would now endorseis the top-down method it adopts for DRS construction, in which sentence DRSs are constructed by starting at the top of the parse tree for a sentence and then stepwise breaking down this tree all the way to its leaves, constructing the DRS in the course of it.(There was a reason for wanting to construct DRSs top-down that had to do with the treatment of anaphoric pronouns, but that reason disappeared when anaphora came to be treated as a species of presupposition in the wake of Van Der Sandt's proposal in (van der Sandt, 1992).Kamp et al. (2011) provides an updated survey of DRT, which covers the treatment of presuppositions as well as a number of other developments from the time after the appearance of (Kamp & Reyle, 1993), including a discussion of MSDRT (if not under this name).But there is little in this handbook article about the details of DRS construction.

| MSDRT, stage I
In the introduction above attention was drawn to attitude attributions that aim to make sense of people's behaviour.Such attributions typically consist of combinations of belief and desire attributions in the spirit of belief-desire accounts 11 of human action. 12They describe the mental states of their attributees as involving attitudes of different "modes"beliefs, desires, and intentions, but also doubts, expectations, queries, and more.The mental state descriptions of MSDRT capture this in being sets whose elements are descriptions of attitudes of these various modes.More specifically, we assume that each such element is a pair consisting of (i) a mode indicator -BEL for "belief," DES for "desire," INT for "intention," and so on (the choice will depend on the way the framework is being applied)and (ii) a DRS, as specification of the propositional content of the attitude.
But propositional attitude constituents are not the only constituents of MSDRT's mental state descriptions.A crucial further ingredient of its mental state descriptions are entity representations, or ERs, as I will refer to them from now on.ERs are representations of thingsindividuals, as the term is used in model theoryrather than propositional contents.The motivation for including them in the mental state descriptions that MSDRT provides is twofold.On the one hand, they prove to be needed in dealing with the semantic contributions made by names and other definite noun phrase occurrences within the scope of attitudinal predicates.More often than not, the force of such NPs is that the contents of the attributed attitudes are singular propositions, which are "directly" about the referents of those NPs.(More on singular propositions below).On the other hand, there are also more general, less specifically linguistic reasons for wanting to include ERs in our mental state descriptions.There is a strong pretheoretic intuition that all of us go through life equipped with large inventories of representations of all manner of thingspeople, places, animals, trees, buildings, works of art, utensils, events of various sorts, and even our own thoughts; and entities of all those kinds can either be currently existing or be things of the past.Most of our thoughts are about things for which we have ERs.And because those thoughts are about entities for which we have those representations, they too have singular content.
The reason that our ERs afford us singular thoughts about the entities they represent is that ERs are connected to those entities by causal links.ERs "directly refer" to the entities they represent, to borrow a term from the philosophy of logic and language, because they are linked to them in this manner.That makes it possible for thoughts that involve ERs to be "directly" about the entities they represent.ERs can thus be regarded as the mental counterparts of directly referring terms in language.I will therefore also speak of ERs as referring to the entities they represent and to those entities as their referents.
But what is it for a thought to "involve" an ER?The answer given by MSDRT is that ERs can play the part of arguments in representations of propositional content.More specifically, because propositional content representations are assumed to take the form of DRSs, ERs can act as arguments in DRS conditions.The way MSDRT has chosen to implement this, staying as close to DRT as possible in this respect, is to assume that each ER comes with a distinguished discourse referent.Occurring as an argument in a DRS predication then takes the form of this distinguished dref filling the relevant argument slot of the given DRS predicate.
To see more concretely what MSDRT's mental state descriptions are like, consider the example in (8).This is the description of an agent A who believes that his friend Julia is in Paris and who would like her to be in Berlin (where he is himself).The description assumes that A 11 For one reference from a large and complex field, see (Davidson, 1963).12 In fact, many of the attitude attributions we make extend over several sentences, and often there are anaphoric and other presuppositional relations between the sentences of such attributions.This is a further reason for wanting an account of the semantics of such reports that builds on DRT.
has ERs for Julia, Paris, and Berlin and has the specified belief and desire about the entities represented by these ERs.
No explanation is needed for the two propositional attitude constituents of ( 8).Their content representations are DRSs whose form is familiar from what we have seen in section 2. But note that the universes of these DRSs are empty.That is because the drefs occurring in their conditions are imported from the ERs of which they are the distinguished drefs.With regard to the ERs of (8), for now just this: As can be seen from the ERs in (8), ERs have a good deal of structure, sub-divided into three components.A couple of hints about these components: (i) The first component consists of (a) the mode indicator ENT, which tells us that we are dealing with an entity representation (and not some type of propositional attitude constituent); and (b) the distinguished discourse referent of the ER.(ii) The second component contains descriptive information about the referent, which is essential to the way in which the agent conceives of the referent.(iii) The third component is the anchor set of the ER.The anchors that are the elements of these anchor sets function as records of causal relations that connect the ER and the agent whose ER it is with the entity represented by the ER.Anchors will play a central role in the story this paper has to tell about causal chains, and will require careful discussion.But that will have to wait till section 3.3.
I will refer to the mental state descriptions of MSDRT as MSDs.

| MSDRT, stage II
The second stage of MSDRT deals with the logical forms for attitude attributions.In these logical forms MSDs play a decisive part, but they only occur in special positions, as argument terms of the attitude attributing predicate Att.Formally, Att predications are DRS conditions.That makes it possible to combine attitude reports with other kinds of information within one and the same DRS.It also makes MSDRT into a proper extension of DRT.More precisely, like DRT, MSDRT can be thought of as making use of a family of different representation languages; and for each of these MSDRT languages there is a DRT language of the kind described in section 2 of which the MSDRT language is a proper extension.An example of a DRS of such an MSDRT language is the one in (10), 13 the logical form for the attitude attribution ( 9), made by a speaker S who is addressing some person H. 9 ð Þ Fred believes that Julia is in Paris: He would like her to be in Berlin: Some comments on how (10) is to be read: (1) The predicate Att in (10) has three argument slots: (i) for the attributee, (ii) for an MSD, and (iii) for a set of links which provide referents for ERs from the MSD in the second slot.(2) The links in this set are pairs of drefs < j,j 0 > and so forth, where the first dref j is the distinguished dref of an ER in the MSD and the second dref j 0 is external to the Att predication.The meaning of the link is that the value of j 0 (as it gets determined in the course of a model-theoretic evaluation of ( 10)) is the referent of the ER.So the propositional content DRSs that contain occurrences of jthe belief DRS and the desire DRS in (10)evaluate to singular propositions about this value.Note what this entails for the case when (10) is true.(10) entails that Fred, the individual represented by the dref f , is in a mental state that contains an ER with distinguished dref j that is anchored to some individual j.A verifying embedding g of (10) in a model M will have to assign to j 0 that individual j from U M to which this ER is anchored.
The claim made in the last couple of sentences follows from the model-theoretic semantics for MSDRT.That semantics is fairly complex, and there is no justification for presenting it here because the second stage of MSDRT will play no direct role in what follows.(For formal details, see (Kamp, 2003) and the forthcoming (Kamp, 2021b); and for informal discussion of DRSs like (10), see (Kamp & Bende-Farkas, 2019).
13 Throughout this section, I ignore, purely for reasons of expository convenience, the times at which agents are in their respective mental states.For instance, in (10) nothing is said about the time at which Fred is supposed to be in the mental state that (9) attributes to him.In the fuller versions of MSDRT that are used in the references given in this section, the Att predicate has an additional argument slot for a state of affairs s, which consists in the attributee being in the mental state that the Att-Condition ascribes to him.The DRS to which the Att-Condition belongs can then locate s in time, for instance, as simultaneous with the represented utterance of (9).
DRSs like (10) only give a glimpse of the expressive power of MSDRT.Part of that power derives from the fact that the outer drefs from anchoring links like < j, j 0 > in (10) can be bound in a variety of ways, including by universal quantification.I will not go into further details here, but readers who have some antecedent experience with DRT may want to play around with possible options.
Another source of the expressive power of MSDRT is the possibility of embedding Att-Conditions within other such conditions.So far, we have discussed (10) as the logical form of the utterance in ( 9) and its Att-Conditions as the semantic representation of the complex of ERs and attitudes that (9) attributes to Fred.But it is also possible to use Att-Conditions as constituents of content DRSs of attitude constituents of MDSs.For instance, if we make the reasonable assumption that the speaker S of ( 9) believes what (9) expresses, then the relevant part of her mental state could presumably be described by the MDS in (11), consisting of ERs for Fred, Julia, Paris, and Berlin together with a belief whose content representation consists just of the Att-Condition in (10): In (11) the anchoring links are between the distinguished drefs of the ERs in the content DRS of S's belief and the distinguished drefs of S's own ERs.This guarantees that the propositional contents of the attitudes ascribed to Fred by the MSD that occurs as argument to the Att-Condition denote the propositions that S wants to attribute to Fred: singular propositions about the referents of her ERs for Julia, Paris, and Berlin.A similar point can be made about what presumably happens to the mental state of H as a result of his interpretation of S's words in (9).Let us assume that H too has ERs for the Fred, Julia, Paris, and Berlin that S is referring to in (9); that these ERs are labelled with the names Fred, Julia, Paris, and Berlin, respectively; and that he uses these ERs in his interpretation of (9).If he furthermore constructs his interpretation according to the rules of the language, obtaining a representing DRS for (9) that is in essence like (10), and if he also believes what this DRS tells him, then the relevant part of his resulting mental state may be assumed to look like (11).The contents represented by the attributed belief and desire in this part will now be tied to the referents of his ERs for Julia, Paris, and Berlin.But if these ERs are properly coordinated with the corresponding ERs of S, in the sense that they refer to the same referents, then the belief and desire attributions to Fred by H and S will have the same propositional content; and with that the propositional contents that S and H associate with (9) as a whole will be the same as well. 14 This concludes the high-speed introduction to MSDRT, except for the discussion of the internal structure of ERs in the next two subsections.To repeat once more what I have emphasised at various points: The formal foundations of the framework -DRS construction rules, syntax, and model-theoretic semantics for MSDRT's DRS-languageshave been passed over but can be recovered from the documents that were mentioned along the way.14 Often agents will have several ERs labelled by the same name.For instance, both S and H may have more than one Julia-labelled ER: they each know more than one person with the name Julia.If that is so, H should use the right Julia-labelled ER when he interprets (9)that ER which is coreferential with the Julia-labelled ER that S made use of when choosing the name Julia as part of her utterance of (9).Interpretational mishaps, when the interpreter uses a wrong N-labelled ER to interpret the speaker's use of N, are in principle always possible and happen not infrequently.It is of course also possible that, although H has one or more Julia-labelled ERs, none of those is coreferential with the ER that S has used.For discussion of various such cases, see (Kamp, 2015), and also footnote 24 in section 3.5 and section 4.There is an extensive literature on entity representations.In much of it, such representations go by the names of "file card" (Perry, 1980) (Heim, 1982(Heim, , 1988) ) or "mental file" (Recanati, 2012).I expect that many readers will be familiar with a good part of this literature and that those who have been wondering to what extent the notion of an Entity Representation as it is used in MSDRT is a misappropriation, barely concealed by a new (rather uninspired) label.There is some justification for this suspicion.The use of files and the file cards in the philosophy of language has been known to me from before the first formulation of DRT, first through the work of Perry and then through the file change semantics that Heim developed in her doctoral dissertation.And at the time when file change semantics and DRT were made accessible to a wider public, there was a discussion about the relations between Heim's file cards and DRT's discourse referents.Of the various notions of entity representations that can be found in the literature and the roles they play in the different theories that make use of them, the one to which the ERs of MSDRT bear the closest resemblance are the mental files defined and used in the work of Recanati.(See in particular (Recanati, 2012), which discusses many examples and puzzles from the philosophy of language and mind literature to which his concept of a mental file can be fruitfully applied).The reason for not sticking with the file card/mental file terminology and adopting the label "entity representation" was to avoid unwanted identifications with notions from the file literature that are not developed in the formally explicit way that ERs are treated in MSDRT.(The formal explicitness of MSDRT is also found in Maier's Attitude Description Theory, see e.g.(Maier, 2016).

| Entity representations and their anchors
(ii) K descr is a DRS (which contains certain descriptive information about the represented entity).(iii) K anch is a set of anchor-DRSs. 16  No further comments are needed for the first ER-component.As regards the descriptive information in the second component, just this: the guiding intuition is that information in this component is restricted to what the agent treats as non-contingent information about the referent because she conceives of that referent by virtue of her ER for it.It is not easy to be more precise on this point, for one thing because agents often waver in their conceptions of entitieswhat the agent treats as non-contingent information about the referent at one time or in one context need not be the same as what she treats as non-contingent information about the referent at other times or in other contexts. 17In this paper, I make no assumptions about what information goes into the second components of ERs except that this information includes (i) "sortal"' information, which specifies what sort of entity the ER's referent is (a person or a city or an artefact and so on), and (ii) information about the name or names that the agent assumes the referent goes by.What else may or may not go into the second components of ERs is of no consequence for the remainder if this paper.
What does matter crucially for present purposes are the third components of ERs, their anchor sets.The remainder of this section is about them.
For a start, it is important to keep firmly in mind that the anchors that make up the anchor sets of ERs are internal anchors in the sense of footnote 16; they are records of causal relations between the ER and its referent.
Second, the anchors that can be found in anchor sets come in different types reflecting the different causal relations to which they bear witness.One of those types are the vicarious anchors.These are the anchor types that are of principal interest in this paper as the pivots of our reconstruction of causal chains.However, the idea behind the notion of anchors as constituents of entity representations is more easily explained by first looking at another type, that of the perceptual anchors.
One situation in which an agent can form a new ER is when she perceives something that she takes to be something that she does not know.Such an ER comes equipped with an anchor set containing a single anchor, which records the perception that results in the formation of the ER.Such anchors, which act as records of perceptions of the entities represented by such ERs, are called perceptual anchors. 18But such anchors are not only formed as part of newly constructed ERs.Perceptual events can also be the records of perceptual recognition events in which the agent perceives an entity that she takes to be the referent of an ER she already has.This is one of the reasons why MSDRT assumes that ERs have anchor sets, which can consist of several anchors rather than just a single one. 1916 For those readers who are familiar with some of the existing publications in which MSDRT is used: The anchors that occur in anchor sets are often referred to as internal anchors in those publications, whereas the term external anchor is used to refer to what in the present paper is described as "the entity represented by the ER" or as the ER's "referent."In the present paper, the terms internal and external will only be used in the discussion of anchors in this section.17 Contingent information about the referent of an ER may be assumed to take the form of belief constituents of the agent's mental state, whose content specifying DRSs have occurrences of the distinguished dref of the ER.For some discussion of MSDRT's distinction between treating information as contingent and treating it as non-contingent, see (Kamp, 2021), Sn. 4.2.The idea that what we take as essential to entities of which we have a conception can vary with the context in which we think about the entity seems to have been an important ingredient in the so-called cluster theory of names.See Naming and Necessity, (Cumming, 2019).18 Sometimes our perceptions are non-veridical.We can be the victims of fata morganas and other kinds of perceptual illusions.Because agents cannot distinguish such "fake" perceptions from true ones, it may be assumed that the fake perceptions lead to the formation of ERs with perceptual anchors just as real perceptions do.But in such cases the perceptual anchor of the ER is a false witness, the ER has no referent and attitude constituents of the agent's mental state whose content representations have occurrences of the ER's distinguished dref fail to have a well-defined propositional content.Important as this aspect of MSDRT may be, I have decided to set it aside in this paper to reduce expositional overload.

19
A consequence of this way of treating recognition in MSDRT is that the anchor sets of ERs for entities that one interacts with on a regular basis and that one thus encounters again and again would grow to very large proportions.From a psychological point of view, this seems unrealistic.The problem will be addressed in a forthcoming paper (Kamp, 2021a).Another problem connected with the use of ERs in recognition is the information that must be associated with the ER in order that it can play its part in the recognition process.For one much-discussed type of case, consider facial recognition of people.This is something that most of us are remarkably good at.(And that algorithms are now getting very good at too, although I do not know if that has any relevance for what I want to say).In

THE LINKS OF CAUSAL CHAINS
In applications of MSDRT it is often important to define anchor sets explicitly, and for that we need explicit descriptions of the anchors they contain.To this end, MSDRT has been making use of DRSs from some specialised anchor representation language.(13) is a simple example of such a DRS.
It identifies the anchor of an agent who is currently looking at an entity, represented by the distinguished dref x of the ER to whose anchor set (13) belongs.(Non-visual perceptions will of course require other perception predicates than see').After the perception is over, but the ER continues to be part of the agent's mental state, (13) will be adjusted to an anchor form that expresses that a visual perception of the entity has occurred in the past, possibly with some specification of when that was; but we do not need to go into the details of that here.
So much for perceptual anchors.As I said, more important for the purposes of this paper are, as I said, vicarious anchors, the topic of the next section.

| Vicarious anchoring and vicarious anchors
One of MSDRT's central claims is that ERs can be linked to their referents via vicarious anchors.Agents can achieve direct referential access to entities just on the strength of the linguistic references made to them by others; this was one of the lessons I first learned from Naming and Necessity.But the implementation of the message in MSDRT looks somewhat different from the way I understood it at the time, and it also is somewhat more general.I should add that the messagethat direct access to entities can be obtained from the references made to them by othersis not one that can be inferred from any other assumptions made in MSDRT.But although it could be separated from the rest of it, MSDRT would be a very different enterprise without it.
As a first example of vicarious anchoring and vicarious anchors, consider the miniconversation in ( 14). ( 14) S: I just reread a short story by Gogol last night.Amazingly good, really!H: Ah yes, interesting!
This exchange is about as innocuous as they come.But it could plausibly occur in a number of different situations, and it is the difference between some of those situations on which we focus.One difference has to do with H.The name Gogol will either be familiar to H when the conversation starts, or it will not be.In the first case, a "Gogol"-labelled ER for Gogol will be a constituent of H's mental state; in the second case, it will not.A second question has to do with what S thinks about H. S herself will of course have a "Gogol"-labelled ER for Gogol; otherwise, she could not form the thought expressed by her words and express that thought the way she does in (14).But she may also have an idea of whether H has such an ER as well; and if she is using the name "Gogol" in the way she uses it in ( 14), she better should assume that he does because "standard" uses of proper names like this one carry a presupposition that the recipient is familiar with the namein our terms: that he has an ER labelled with the name for the entity order that my ER for you can do its job when I see you or a picture of you and see that it is you, my ER must have some kind of mental image associated with it (one that I will also be able to call up before my mind when you are not there).What precisely is associated with the ER that makes such recognitions possible and how it is associated with the ER are questions about which MSDRT has had nothing to say so far.
that the speaker is referring to by using the name.Only when H has such an ER will he be in a position to interpret A's use of Gogol in the way such name uses are meant to be interpreted.If H does not have a Gogol-labelled ER for Gogol, then he must accommodate the presupposition of S's use of Gogol by adopting one afresh, as if he had had one all along. 20 But if H does not have a Gogol-labelled ER for Gogol and needs to accommodate one, what should that ER be like if it is to play the part in his interpretation of (14) that would have been played by a Gogol-labelled ER for Gogol if the name had been familiar to H and he had had such an ER from the start?The answer to this question is one of Naming and Necessity's important messages: H can take over S's use of Gogol, as name for the individual that S has just used Gogol to refer to, from the use that S has just made of it in the utterance that reaches him.In the terms of MSDRT: H can mark the ER he forms as one that refers to whatever it is that S referred to with her use of Gogol.Such marks are called vicarious anchors. 21 By using the distinguished dref of his new vicariously anchored ER for Gogol, H is in a position to form a representation of S's statement S that correctly captures its propositional contenta representation that expresses the proposition about Gogol that S read one of his short stories the night before.That is secured by the vicarious anchor of his ER, which renders that ER coreferential with the one relied on by S. But the vicarious anchor of H's ER also accomplishes something else: it establishes a Gogol-related correlation between H and S. As we noted, it is one of the assumptions of MSDRT that to use a name N in the manner that S uses Gogol in ( 14) you must have an N-labelled ER that represents the referent that you are using N to refer to.Thus, if S made a proper use of Gogol in ( 14), then she had a Gogol-labelled ER ER S for the Gogol that she was referring to on this occasion.And when H interpreted S's use of Gogol by accommodating an ER ER H for the individual referred to by her, and that is represented by the Gogol-labelled ER ER S that her use of Gogol was based on, then the vicarious anchor of ER H can be seen as creating a link between their ERs ER S and ER H .We will represent the link as the ordered pair < ER S ,ER H > and see it as standing for a coordination relation between S and H to the effect that their ERs ER S and ER H represent the same individual.
When the recipient already has an N-labelled ER ER H for the referent that the speaker is referring to by means of N and uses it to interpret the speaker's use of N, then no vicarious anchor is needed to establish coreference between ER H and the ER ER S on which her use was based.But the use H makes of ER H in his interpretation of N also has a coordinating effect in this case.It will not normally establish the coordination between their ERs, but it will often strengthen the coordination, for instance by making it more direct.To do justice to these coordination effects, MSDRT also assumes that in such cases the use that H makes of his ER ER H to interpret S's use of N involves the adding of a vicarious anchor to the anchor set of ER H , thereby also creating a link between ER H and the ER ER S on which the speaker's use of N was based.Thus, in these cases a link < ER S , ER H > is created as well, as a signal for the reinforcement of the coordination relation due to the given use and interpretation of N between S and H, and more generally within the N-using community.
To summarise the main points of this section: 20 Sometimes people will use names (in the "standard way," see (Kamp, 2015)) when they do not believe the name is familiar to their addressee but want to impress or intimidate them.Or they really have no idea whether their addressees are familiar with the names they are using but simply do not care.Related to cases of this last sort are those where the names that interpreters are confronted with are used in conversations they overhear but that are not intended for them.See (Evans & Altham, 1973).

21
Whether just one such exposure suffices to give H a proper command of Gogol has been a point of long-standing debate.For an early discussion see (Evans & Altham, 1973).To relate the point to ( 14), suppose that H has never heard of Gogol.Perhaps the most natural situations for some of us to acquire what feels like a genuine command of a name are our encounters with unfamiliar names in texts.Suppose I read in a textbook: "The first proof that combustion involved the binding of oxygen was given by Lavoisier."I never heard of Lavoisier before, and so one of the things I learn from this sentence is that there was a person by this name who had something to do with chemistry.(If I am a conscientious student, I will make a note of the name as well as the information that the passage provides about oxygen).In this case too I may feel I need to learn more before I can be confident in using the name myself.But it seems quite plausible that just on the strength of my first encounter with the name Lavoisier I set up a labelled ER for its referent.
(15) (i) When a speaker uses a name in the "standard" way exemplified by the use of N by S in ( 14), then she must have an N-labelled ER on the basis of which she does this; and by basing her use of N on this ER, the entity she refers to by her use is the one represented by this ER.
(ii) A member of the speaker's addressee who interprets her use of N must either use an N-labelled ER that he has or else accommodate an N-labelled ER as representation for the entity that the speaker has referred to with her use of N. In either case, a vicarious anchor is added to the anchor set of the ER as a witness of the ER's use in the interpretation of N on the given occasion of its use.
The vicarious anchor enforces coreference between the interpreter's ER ER H and the speaker's ER ER S .Furthermore it adds a new link < ER S ,ER H > to the coordination between S and H with regard to their use of N.
The links < ER S , ER H > will be the pivots in our reconstruction of the causal chain theory.This is the topic of section 4. In section 3.5, the last part of the present section, we look at a couple of examples of how utterance interpretation affects the mental state of the interpreter. 22

| The effect of utterance interpretation on mental states
In this section, we look in some detail at how according to MSDRT utterance production and interpretation are related to the mental states of source and interpreter.We will be looking at just one sentence, a streamlined version of S's utterance in (15):

22
The term 'vicarious' is familiar from the literature on file cards and mental files, though its meaning varies between the different theories that make use of it.One prominent use of 'vicarious' is made in (Recanati 2012).Recanati's work on mental files is in my judgment the most substantial single contribution to this topic from the past two decades.I believe it is agreed between Recanati and myself that MSDRT has much in common with his views, and that this is so in particular for the proposals of the present paper.But there are also differences.In one part the differences are terminological.In particular, for the 'vicarious' of the vicarious anchors of MSDRT Recanati uses 'deferential'.A more substantial difference, even if in the end it may turn out to be mostly a matter of focus or emphasis, is the following.Recanati distinguishes between regular files and indexed files.Regular files are much like the ERs of MSDRT, but indexed files are not.Rather, an indexed file of an agent A is a file/ER that A attributes to some other agent B. When A assumes that an indexed file she attributes to B properly represents an entity d, then she will have her own file for d and there will be a link between this regular file and the indexed file, which testifies to and/or secures their coreference (by virtue of both referring to d).But it is also possible for A not to assume that the indexed file she attributes to B properly represents; in that case she won't have a corresponding regular file.Cases of this second kind are important for the analysis of many well-known puzzles in the literature on reference and attitude attribution, like that of the sociologist who doesn't believe in witches but who is aware that witches are believed to exist by the community she is investigating, and who in her report about the community somewhere writes about two of its members that Hob once claimed that a witch killed Cob's cow and that then Nob added to that that she had also maimed Bob's mare; or like that of the mother who talks to four year old Johnny about Santa: He believes that Santa exists, she doesn't, but she knows that he does.Most importantly in the context of this paper, indexed files play an important role in the formation of causal chain links that has not been mentioned in our discussion.When H interprets S's use of the name N, concluding that she has used N to make a proper reference to some entity d, then that involves, according to the account I have offered, his conclusion that S must have an ER for this entity.That amounts to H attributing an indexed file to S that is linked to his own regular file for d (the N-labeled ER that he has been using in his interpretation of S's use of N).Furthermore, in face to face communication a move of this kind can be expected also on the part of S: If H reacts to her utterance in the manner that is typical of someone who takes himself to have understood what the speaker is saying, then she will attribute an indexed file to him for the entity she has used N to refer to and link that indexed file to her regular file for this entity.(These are the first steps on the ladder towards referential common ground).Within MSDRT indexed files can, to the best of my understanding, be modeled as mental state attributions involving MSDs that consist of a single ER.If A makes such an attribution to B, then this will take the form of an Att-predication with such an MSD in second position and a dref representing B in first position.(In the cases I have thought about, this dref will be the distinguished dref of an ER that A has for B, but other possibilities exist as well and may well be important too).When A assumes that the ER she attributes to B has a proper referent d, then she will have her own ER for d and the ER she attributes to B will be linked to her ER via a link <x,x 0 > in the third argument position of the Att-predicate, where x is the distinguished dref of the attributed ER and x 0 the distinguished dref of her own ER for d.If A does not assume that the attributed ER properly refers, then there won't be such a link in Att's third position.
(16) S: I read a short story by Gogol.
We consider two scenarios in which S uses and H interprets ( 16).In the first, H does not have an ER for Gogol; in the second, he does.( 18) gives the MDS for the relevant part of the mental state of S, assuming that S herself believes the information that is expressed in ( 16) and that she has a Gogol-labelled ER for Gogol that puts her in a position to use the name Gogol in ( 16). ( 18) describes the relevant part of H's mental state in the first of our two scenarios just before his interpretation of ( 16) gets under way.Note that the only ER shown in ( 18) is an ER for the speaker S. The fact that (18) does not display an ER for Gogol is to be understood in this case as an indication that the mental state partially described by ( 18) does not have such an ER.
When H is in the mental state described in (18), then his interpretation of ( 16) requires the accommodation of a vicariously anchored ER for Gogol.The result of this accommodation and the construction of H's representation of the content of ( 16), which, we assume, H accepts as true as well and thus adopts as a belief, leads to the mental state described in (19).
23 Notation: 1.The symbol "g S " that has been chosen here for the distinguished dref of S's ER for Gogol is motivated by wanting distinct symbols for this dref and the distinguished dref of H's ER for Gogol in (20) below.(The subscripts S and H are chosen solely for mnemonic reasons -S means that we are dealing with a dref in an MSD for S, and H likewise that we are dealing with a dref in an MSD for H.They have no theoretical import).2. The content DRS of the belief in ( 17) belongs to a DRS language that extends beyond the description of DRS languages in section 2. As noted in section 2, the original motivation for DRT was to account for certain properties of tense forms in French and English.Versions of DRT in which these motivating ideas received an early implementation can be found in (Kamp, 1981) see also the English version (Kamp, 2017) and in (Kamp & Reyle, 1993).The intuitive meaning of the DRS in ( 17) is that there was an event e of S reading some short story by Gogol at some time in the past of the time n at which S is in the mental state that (17) describes.

THE LINKS OF CAUSAL CHAINS
Unfamiliar in ( 19) is the notation used for the vicarious anchor of H's accommodated ER for Gogol.Like for perceptual anchors, some suitable form is needed to represent such anchors.The form adopted here is shown in (20).
In ( 20), e is the act of reference that gives rise to H's accommodated ER; s is the source of the reference (the speaker or text); γ is the referring expression used by s; and α is the distinguished dref of the ER of which ( 20) is an anchor.(Thus, in the vicarious anchor of the ER for Gogol in ( 19), s is the dref s H , γ is the expression Gogol, and α is the distinguished dref g H ).
We now consider the case where H does have a Gogol-labelled ER for Gogol before S says (16) to him.In this case, H's mental state before he starts his interpretation of ( 16) can be described as in ( 21).
According to MSDRT, the result of processing ( 16) is the one shown in ( 22).
The one aspect of ( 22) that requires attention is the anchor set of its Gogol-labelled ER.This set results from adding to the anchor or anchors that were part of the ER before interpretation started the same vicarious anchor that figures as the unique anchor of the accommodated ER in (19).The reason I gave in the last section for insisting that in this case a vicarious anchor is introduced into the anchor set too (and not only when a new ER is accommodated) was that also when the interpreter uses a previously adopted ER, the communication reinforces the coordination between him and S.But a further justification could be given as well.When H uses an ER he already has, this means that he is understanding the name that S is using as used for someone familiar to him; he knows who Gogol isthe famous author of the Dead Soulsand understands that S is saying something about that author.In a sense, that also is a form of recognition, in certain ways quite like what happens when you recognise someone you meet in the flesh.In both cases, the experience strengthens your relationship with the referent.Thus, it is natural that just as we assumed that visual recognition of an entity r for which one has an ER adds a perceptual anchor to the ER's anchor set, so you add a vicarious anchor to the anchor set when you establish contact with r by hearing or reading their name.
In fact, anchor sets are quite often mixtures of anchors of either type.Take for instance the case of Victor, who sees a woman at a party with whom he is much taken and tries to find out more about her.When someone tells him her name, he will add a vicarious anchor to his thus far purely perceptually anchored ER for her (as well as the condition "Named(x,N)," where x is the distinguished dref of the ER and N the name he has just been told).Or conversely, take Pierre, who grows up in Paris and learns about London from the uses that his family makes of the name Londres, thus obtaining a Londres-labelled ER for London with a (possibly quite large) set of vicarious anchors.When he then finally gets to London and recognises it as the place for which he thus far only had an ER whose anchor set was purely vicarious, from then on that set will be boosted with a growing number of perceptual anchors. 24 It is time to summarise the main points of this and the last section on vicarious anchors and their reference-fixing and coordinating roles.Vicarious anchors have the following two functions: • Vicarious anchors secure proper reference for the ERs of which they are part, by themselves or in cooperation with other anchors with which they cohabit in their anchor sets.This is so in particular for the vicarious anchors of ERs that are accommodated in response to unfamiliar names and that are the sole members of their anchor sets.• A vicarious anchor links its ER to the relevant ER or information source of the reference that the anchor records.These links are the constituents of the intersubjective networks that stabilise the use of names between language users and within the language communities to which they belong.• Furthermore, as I read what Naming and Necessity has to say about causal chains, MSDRT differs from that in assuming that every interpretation of a reference by a proper name produces a link between interpreter and source, irrespective of whether the name is new or familiar to the interpreter.In the latter case, a link is established between language users both of whom already have command of the name.

24
Of course, it is also possibleeven if the possibility is a somewhat outlandish onethat Pierre does not recognise London as the referent of his vicariously anchored ER for London when he finally gets there.Instead, he will form a new ER for London, with multiple perceptual anchors and, as in the story of (Kripke, 1979a), also vicarious anchors that connect him with people who have been referring to the place where he is now by using the name London in his presence.So now Pierre has two ERs for London, both of which are labelled, but with different names, and he lacks the resources to put one and one together.(What we are to say about belief attributions that can be made to Pierre in which the names London and Londres are used is a difficult matter, as we have learned from Kripke's paper, one of the harder nuts that theories of the use of names in attitude attributions should be able to crack.But it is not one of the nuts to be cracked in this paper).I should add that cases of recognition failure, where you are confronted with an entity for which you have an ER but do not realise this, are only one way in which we can be making a suboptimal use of our ERs.Another form of suboptimality, and in this case we can speak of an outright mistake, is when there is too much recognition: you think that whom you see on the other side of the street is Julia, adding a perceptual anchor to your ER for Julia (as well as, potentially more seriously, associating new descriptive information with this ER that derives from your present encounter).According to the account presented here, this renders your ER incoherent in that different anchors in its anchor set connect it to different entities.Trying to refer to two things at the same time by virtue of being connected to both is hardly better than failing to refer because of not being connected to anything.Still, mishaps of this kind often remain without serious consequences.When they are harmless and when they cause serious havoc is yet another topic for further investigation.Recall the brief discussion of this issue in footnote 14.It is tempting to add an MSDRT analysis of the predicaments of Kripke's Pierre at this point, but it is one that I resist, as it would lead us away from the main story.From what has been said in this paper so far, however, it is probably not too hard to see how at least part of such an analysis would go: Pierre has a Londres-labelled and a London-labelled ER for London.The distinguished drefs of those ERs may occur in the content representations of various attitudes of his, without any logical constraints on their mutual distribution.In particular it is possible for Pierre to hold beliefs with the content specifications "beautiful"(l f )' and "ugly"(l e )', where l f is the distinguished dref of his Londres-labelled ER and l e the distinguished dref of his London-labelled ER.The propositional contents determined by these specifications are mutually exclusive because the two ERs are coreferential.But to realise that they are incompatible one has to be aware of their coreferentiality, and Pierre is not aware of that.Everyone who has read (Kripke, 1979a) knows there is more to be said about this problem, but this is how the first part of a MSDRT-based analysis would go.
• Moreover, vicarious anchoring is not restricted to the interpretation of reference events involving proper names but can result from the interpretation of other referring noun phrases as well.
Let me give one illustration of this last point.When I was a teenager and my mother told me about the apartment in Charlottenburg where she spent the first 9 years of her life, she must have had, by any of the criteria I can think of, a multiply anchored ER for that apartment.And through what she was telling me about the apartment, she enabled me to form an ER of that apartment as well, one that I reused again and again when I listened to more of her stories about it or indulged in my own fantasies about it.But my ER only had vicarious anchors, and it was without a name.(If I had known the address of the apartment, that might have played a role much like that of a name, but I never knew it, and now regret I never asked).The NPs my mother used to refer to the apartment were descriptions like our apartment, descriptions she used to refer to the entity represented by her ER.Such uses of definite descriptions are common; and other noun phrase types, for example, demonstratives and pronouns, can also be used to refer to entities represented by one's ERs.Vicarious anchors, MSDRT assumes, result from the interpretations of any such referential acts.

| ER NETWORKS
The main gist of the story will now be clear: Vicarious anchors establish links between ERs of different language users, ER to which the anchor belongs and the ER on which the speaker relied in producing the reference that the anchor records.It is such links that form the chains which link any speaker who has command of a name N as name for a referent r to someone who participated in the "baptismal event" in which N was conferred as name upon r.But there is more to say.
The causal chains that link N-users with baptism participants are normally linear substructures of ER networks that are much bigger, and that also tend to have a much more complex structure.We start with a look at the links from which such networks are built. 25 Vicarious anchors, I said towards the end of the last section, arise not only in response to referential uses of names but also when sources refer through the use of other types of definite noun phrases. 26When the interpretations of referring acts involving these other types of noun phrases give rise to new ERs, these ERs will not be named (i.e., not be N-labelled for any name N).The vicarious anchors of such ERs will nevertheless establish links < ER S , ER H > between the new ER ER H of the interpreter and the ER ER S of the speaker who made the reference.In such cases, ER S can be either named or unnamed: S may not have a name for the entity she is referring to, but it is also possible that her ER is N-labelled but that she prefers to refer to the entity it represents by using some other type of noun phrase, for instance, because she thinks that this will make it easier for H to understand which entity she is referring to.And the opposite situation, where ER H is named but ER S is not, is possible as well, for instance, when S does not have a name for the entity she is looking at while making her reference, and where H, who can also see the entity, recognises it as the one represented by an N-labelled ER he has for it.The most general notion of an intersubjective coreference network is that which consists of all such links < ER S , ER H > (Kamp, 2021).In this paper, however, we restrict attention to the sub-networks that consist of links < ER S , ER H > where ER S and ER H are both N-labelled for 25 For a discussion of ER networks that partly overlaps with the one in this section, see (Kamp, 2021).26 (Kamp & Bende-Farkas, 2019) argues that this is so even when speakers make epistemically specific uses of indefinites, and their interpreters take them to be doing that.For definitions of epistemically specific and other types of specific uses of indefinites, see (Farkas, 2002).
the same N and that are the result of interpretations by H of references made by S through the use of N.
Before we move on with the discussion of N-labelled networks and causal chains, there is something I want to make fully explicit, although it may have become clear enough from the discussions up to this point.It is about the identity of names.I am assuming that the identity of a name is merely a matter of morphophonemic form.If your friend is called Julia and my friend is called Julia, then your friend's name and my friend's name are the same name, which your friend and my friend share.The name Julia is ambiguous between your friend and my friend and the countless other people who are also called Julia.When you say to me "I just talked to Julia," I may misunderstand you by taking you to be talking about my friend, whereas you were talking about your friend.For me your statement is ambiguous: among my Julia-labelled ERs, there are at least two, the one for my girlfriend and the one for your girlfriend, both of whom are plausible candidates for my interpretation of the use you are making of Julia.On this conception of the identity of names, all names are potentially ambiguous; quite a few names are ambiguous; and some are ambiguous to a very high degree. 27 Where there is ambiguity, there is always a risk of misinterpretation.The misinterpretations caused by the ambiguity of a name N take the form of S using N to refer to the entity represented by one of her N-labelled ERs and H taking her to be referring to some other entity, represented by some N-labelled ER ER H that he has.When this happens, the damage is twofold.On the one hand, the interpretation renders ER H incoherent in that its anchor set now contains divergent anchors: the one or ones that link it with the entity for which ER H was originally introduced and the new vicarious anchor that links it to the entity that S has referred to (see footnote [24]).On the other hand, H's misinterpretation also produces a deficient link < ER S , ER H > ; there is not any r such that ER S and ER H are both unequivocal representations of r.Such deficient links are a threat to the networks to which they are added: it ought to be the case that any two ERs occurring in the network that are connected by a chain of links are unequivocally coreferential, but for networks with deficient links this is not so; and when the ERs from the deficient link are both N-labelled, this may cause upheavals in the subsequent use of N. (For some discussion, see [Evans & Altham, 1973]). 28 To conclude this discussion of link deficiency, note that the use of a name N can never lead to a deficient link when the interpreter does not have an N-labelled ER to interpret the speaker's use of N.That is, in those cases when referring by means of N has the effect of introducing the addressee to the given use of N , there is no room for misinterpretation and therefore also no risk of it.The causal chains that, on my understanding, Naming and Necessity is speaking ofthose consisting of links < ER S , ER H > where both ER S and ER H are N-labelled and ER H was formed at the same time as < ER S , ER H >are therefore immune from link deficiency.Because it is causal chains of this kind that are the ultimate target of our reconstruction, link deficiency will be ignored in the remainder of the paper.
By setting deficient links aside, we finesse much of the structural complexity of ER networks.But there is another source of complexity to network structure that we will not be able to do without.Each link < ER S , ER H > is created at some particular time by the interpretation of some act of reference.These "time stamps" of links impose a temporal order on them.Because this temporal order is important for the definition of causal chains, we make it part of the relational structure that will serve as the formal basis for our further considerations.That is, we define ER networks as sets of triples < ER S , ER H ,t > , where t is a moment of time -I assume that time is like the real number structure ℛ, but this is of no great importance here 27 For some discussion, see (Kamp, 2015).

28
For the most part, misinterpretations of names do not seem to do much damage.But to my knowledge, the mechanisms that immunise name uses against occasional misunderstandings of this kind are not very well understood and have not yet been seriously explored.
and ER S and ER H are linked by a vicarious anchor in the anchor set of ER H that becomes part of this anchor set at time t.Let us assume moreover that we are looking at the uses of a name N for an entity r in a speech community SC over some period of time T (some finite interval of ℛ) and that T includes the baptism event in which N was introduced as name for r.Note that, for certain names, for instance the biblical names Moses and Jonah, but also probably even more ancient names like Gilgamesh or Djoser, T has to extend quite far into the past.(I am taking it for granted that these are all names of real people).
The ER networks defined in ( 23) are those whose ER links are the result of communication events in the course of T in which agents from SC add vicarious anchors to their N-labelled ERs for r as part of their interpretations of tokens of N.Among these events we also include those where a member of SC encounters a reference to r by a token of N in a text.For such cases, I am making the somewhat simplifying assumption that the link triple < ER S , ER H , t > has as first constituent the ER that enabled the author S of the text to use N to refer to r; the second constituent is (obviously) the ER ER H of the reader H to which the new vicarious anchor is added; and t is the time at which H is reading this occurrence of N. 29  Here is the definition of the notion of an N-labelled ER network.
(23) (Definition of N-labelled ER networks) The N-labelled ER network for N as name for r within the community SC over the period of time T, NW(SC,T,N,r), is the set of all triples < ER S ,ER H ,t > , where t in T, S, and H belong to SC, ER S and ER H are both N-labelled ER S representing r, and ER H contains a vicarious anchor that is introduced into it at t as a record for the given use by S of N to refer to r.
When the members of SC behave throughout T in accordance with the rules of the language, including those for the processing of referential uses of proper names, then every member S of SC with an N-labelled ER ER S that represents r will be connected by a "causal" chain of elements of this network to some member of SC that was present at the baptismal event which conferred N upon r.Formally: (24) (Definition of causal chains) A causal chain of the network NW(SC,T,N,r) that accounts for the command of S at t of N as name for r is a finite sequence < cl 1 , …, cl nÀ1 ,cl n > of some "chain length" n, where (i) for each (iv) t n < t; and (v) S 0 was participant in the baptism in which r received the name N.
Definition ( 24) is the formal reconstruction I offer for the chains of Naming and Necessity's causal chain account of names.Put in less technical terms: 1.A causal chain accounting for the command at time t within T by a speaker S who belongs to the speech community SC of the name N as name for r is a linear substructure of NW(SC,T,N, r) that is a causal chain according to definition (24) above.2. In a community SC in which everyone abides by the production and interpretation rules governing the referential use of names, it will be the case for any member S of SC who at 29 The reason for wanting to include such links was already mentioned in footnote 21 of section 3.4: in literate societies like ours, encounters with names in texts are often the main source of name transfers, especially to members of the speech community who are born a long time after the name was introduced.Our reliance on texts to acquire command of old names becomes crucially important when their referents no longer exist.They are our remaining lifelines to those referents.(Recall the remark about Lavoisier in footnote 21).I am speaking here of "S's N-labelled ER for r," suggesting that there is just one such.But is that justified?Not when we allow for cases like Peter of Kripe's "A Puzzle about Belief," who has two Paderewski-labelled ERs for the scion of early twentieth-century Polish destiny.Should the Paderewski-labelled ERs of someone like Peter be excluded from links in causal chains that account for his command of Paderewski (as name of this Paderewski)?I have no definite view of the matter and leave it as food for the Paderewski experts.

32
I am making the assumption here that no two vicarious anchors can be formed and added to their ERs by the same agent at exactly the same time.
33 Note that such chains need not be the same as the "earliest exposures" chain described in the last paragraph.To see this, suppose for instance that S acquires the name Wolfgang Amadeus Mozart through countless mentions of it by otherspeople she talks to, radio announcers, and so onwho also got the name from such mentions by yet others, and then lays her hands on Von Nissen's Mozart biography, written by a man who did not personally know Mozart himself but who intimately knew Mozart's wife Constanze.(Von Nissen was her second husband).Because Constanze knew Mozart's father, undoubtedly a participant in Mozart's baptism, this gives a chain of length 3. It is a reasonable assumption that S's earlier exposures to Mozart's name all connect her to Mozart's baptism by chains that are much longer.
these two types of chainor any other type for that mattershould deserve preferential treatment in explanations of the command that speakers have of names.And in fact, can we think of reasons for wanting to disregard any chains from CH(S,N, r) in an account of S's command of N? Hardly.For one thing, full command of a name often requires, we noted, more than a single exposure to it.From that point of view, the more chains, the better.And a further consideration in this connection: even after S has achieved what might be considered full command of N, additional exposures to uses of N by othersand arguably also S's own uses of N-may have the effect of reinforcing her command of N and strengthening her connection to r.I want to suggest therefore that we see CH(S,N, r) as the most natural way of accounting for S's command of N: Her command gets under way the moment the set CH(S,N, r) becomes non-empty.And as time goes on and CH(S,N, r) grows bigger, it too will grow through the new chains connecting S with r that result from her further exposures to uses of N, but also, and perhaps even more so, through the new information about r that she acquires from those exposures. 34 But nevertheless, the chain sets CH(S,N,r) for various combinations of agents S, names N, and referents r may differ considerably in their internal structure.What structural properties are of interest to you will depend on your dispositions, scientific, philosophical, or otherwise.
Here is one such property, which might have some interest from a historiographic perspective.Some of the chain sets CH(S,N, r) that connect speakers S today with persons and other sorts of entities from antiquity have a structure that is reminiscent of the shape of an hour glass.The narrow hole in the middle of the glass is a single text in which N occurs.This text was written at a time when the referent r of N no longer existed.But let us assume that the author, who could not have known r directly, was nevertheless multiply connected with r via one or more people who participated in the baptism of r with N. The text then went dormant for many years, and r completely disappeared from collective memory.When eventually the text was rediscovered, at which time it was the only extant source of N, the name N, and r with it, returned to public awareness, with growing numbers of people using N as name for r.From that point on.speakers with a command of N will be connected with r by many chains, but all those chains pass through the one text that has been rediscovered.Because of that, any such speaker will feel her connection with r to be a precarious one, no matter how many chains there are in her chain set CH(S,N, r).Finding another text with occurrences of N that can be confidently identified as coreferential with the occurrences of N in the rediscovered text would make a big difference if the texts could be shown to be independent of each other, in the sense that neither author had been relying on the other. 35 So much for the structure of chain sets that connect speakers with the origins of names for entities from a distant past.There are other kinds of questions about ER networks that can be asked as well.We can, instead of looking at such networks from the perspective of some particular speaker S, also look at them from that of a given name N, either as name of some particular referent r or as a name that is used in the speech community as the name of multiple referents.And lastly, we can look at networks from the perspective of an entity r that is represented by ERs occurring in networks for different names (all of which have been used, by some people, at some times within T, to refer to r).Let me end this section with a couple of questions that could be asked from these perspectives.
The first question has to do with the ambiguity of names.The network we need to look at in this case is the union of the networks NW(SC,T,N, r) for a fixed name N, but different entities r for which N has been used as a name within SC in the course of T. Suppose we subdivide T into k equal portions T 1 ,…, T k .For each period T i (i = 1,…,k), there is a certain set Ref(N,T i ) of entities for which N has been used as a name within SC during T i .These sets give us a 34 Compare the discussion of multiple exposures to uses of names in (Evans & Altham, 1973).

35
To be precise, suppose that N occurs both in text T 1 from author A 1 and in text T 2 from author A 2 .We say that A 1 relies in her use of N in T 1 on author A 2 iff every chain in CH(A 1 ,N, r) contains a link of the form < ER A2 ,ER B ,t > (for some B or other).T 1 and T 2 were independent with regard to their occurrences of N iff neither A 1 relied in her use of N in T 1 on A 2 nor A 2 in T2 on A 1 .window on the "popularity" of N during T i ; by juxtaposing these sets for the different T i , we get something like a popularity profile for the name N over the entire period T.
When we look at ER networks from the perspective of different names for the same entity (as opposed to different entities that have been referred to by the same name), other questions come to mind: Focus on some entity r and consider the union of the networks NW(SC,T,N, r) for the different names N that are or have been in use for r).Assume that T starts when r comes into being (e.g., when r is born, in case r is a person), and assume again that T is subdivided into k equal portions T 1 ,…,T k .For each period T i , we can now ask who in SC knew r under some name or other during T i .In this way, we get some kind of "popularity profile" for r.
There are many more such questions that can be asked about networks and chains.But the three I have mentioned should have given enough of the general flavour, and I will not embroider any further.There is, however, a further question I want to raise at this point.It is a kind of second order question: Some questions about network or chain structure might be of some interest from some perspective.But is there any philosophical substance there might be to them?I suspect that the prospects are not good, and if I have included a couple of such questions here, it has been with the awareness that they will seem oddly placed in an essay that is intended as a tribute to Saul Kripke.If there is any justification at all for mentioning such questions here, it is that I see this as a good way of bringing out to what extent and in what way the causal theory of names is a social construct.That causal chains are social constructs may be obvious enough, irrespective of any formal reconstruction of them.But the reconstruction I have proposed shows how inseparable this social dimension is from an issue that is usually treated as belonging somewhere quite else: How is linguistic meaning determined by linguistic form?In the next and final section, we will have a quick look at the implications of this inseparability for the methodology of natural language semantics.

| NAMES AND THE SOCIAL DIMENSION OF MEANING
Central to our reconstruction of the causal theory of names is the way coreference links are established between entity representations of different agents.According to the story told in section 3.4, these links arise as a kind of secondary effect in the course of utterance interpretation: utterance recipients need them for the semantic representations they construct.But the very connections that interpreters must assume to get their interpretations right create the linguistic cohesion within the speech communities to which they belong.Intersubjective linking is an integral part of utterance interpretation.
If this is correctand I hope that the arguments of this paper have made the case that it isthen the role of the interpreter, who can acquire the use of names by the very same procedures that enable him to understand the messages conveyed by sentences containing names, must be at the centre of an account of the truth-conditional content of such sentences.And that poses a problem for a conception of natural language semantics that has been dominant since the beginnings of formal semantics more than 50 years ago: The principal task for natural language semantics is to formulate accounts of the truth conditions of sentences (and perhaps also larger units, like dialogues or texts), and these truth conditions should be treated as properties that natural languages have qua autonomous systems, which are what they are independently of how and by whom they are used.In semantic theories that conform to this methodology, truth conditions must be described without any reference to their users. 3636 This does not mean that truth conditions must be treated as independent of the context of use.For instance, as early as in the work of Montague, the starting point and paradigm for the conception of formal semantics as a science of natural languages treated as userindependent systems, there is a clear awareness of the context dependence of indexical expressions like I, you, now, here, tomorrow (Montague, 1970b), (Kaplan, 1989).What the conception excludes are references to the mental states of interpreter and speaker.
This methodological principle has been remarkably successful; most of the progress in formal semantics of natural languages over the past half century has been made in work that has been guided by it.There is nothing surprising about this.Human languages show a remarkable degree of community-wide uniformity and stability, also and in particular in how the forms of their expressions determine their meanings.If it wasn't for such uniformity and stability, our languages could not be the effective toolboxes they are for expressing and communicating the often complex information that we want to make explicit and get across to others.Semantic theories that describe the form-meaning relation of human languages without explicit reference to the mental states of users work as well as they do because the user can normally rely on this uniformity and stability.
But where do the uniformity and stability of the form-meaning relations of human languages come from?What brings them about and maintains them?That is a question a general theory of linguistic meaning should be allowed to ask as well.And here too we can find the beginning of an answer in Naming and Necessity.The reconstruction of the causal chain account I have outlined in this paper helps, I hope, to make the nature of this answer a little clearer: Interpreters of name tokens follow a strategy that aligns their command of those names with that of the producers of those tokens.It is a tempting thought that this strategy, which seems to cater so well to the needs of a community that must be able to rely on the uniformity and stability of its language, is operative not only in connection with proper names but more generally.But is that so?That it isto some extent at leastis something that we can also learn from Naming and Necessity.Natural kind termswords like tiger, lemon, gold, water, carbon dioxyderefer to kinds in much the same way that proper names like Julia or Paris refer to their bearers.There are important differences between the two cases.Kind nouns differ from proper names in that they are more commonly used to make statements about members of the kinds they refer to than about those kinds qua kinds. 37But the similaritieshow proper names and kind terms are acquired and how their reference is stabilised by their useare striking nonetheless. 38oes the mechanism that is operative in our use of proper names and natural kind terms cover yet other domains as well?As things stand, I do not dare to say.But I think there can be no doubt that different mechanisms are at work elsewhere, which are like this one in that they too create uniformity and stability and enable language users to "learn while doing."Here is one example: relative gradable adjectives such as tall or expensive.In order that the positive form of such an adjective A can be truly predicated of an entity r that is an instance of a comparison class defined by a noun Nthat is: in order for the statement "r is an A N" to be true, given that r is an Nthe degree to which r satisfies A must exceed a certain "threshold degree" for A and this comparison class (Kennedy, 2007).This entails that to understand what a speaker who is making a statement of this form is saying the interpreter must know what this threshold is.For instance, suppose you say ( 26 (25) John is a tall basketball player 37 Another important source of the view of kind terms is (Putnam, 1975).Putnam discusses at length how the meaning/use of kind terms is explained to those who do not yet know them.One way to do this is by making an "ostensive" use of the term.For example, to explain the meaning of lemon to someone one can say "This (here) is a lemon" while pointing at some lemon, as an instance of the kind that lemon refers to.Note well, however, that the ostensive use of kind terms is not the only use that enables novices to learn their meanings.Natural kind terms can also be inferred from their "standard" uses.An example of the standard use of a kind term is that of lemon in the statement "This lemon is spoiled," whose purpose is to inform the addressee about some particular property of some particular lemon.Ostensive uses of natural kind terms are like introductory uses of proper names, as when I say to you: "This is Julia," gesturing towards my girlfriend, who is standing next to me.Such uses of proper names are clearly different from the standard uses on which I have concentrated in the present paper.Ostensive uses of kind terms differ from their standard uses analogously.For discussion of standard and introductory uses of proper names, see (Kamp, 2015).
But what determines such thresholds?It has long been assumed that thresholds for gradable adjectives are at best loosely determined.In fact, the positive forms of gradable adjectives have been treated as paradigms of vagueness, and for an adjective like tall its vagueness is threshold vagueness, vagueness about where the thresholds for different comparison classes are situated along the adjective's scale.
Over the past decade, attention has turned increasingly to the use-based mechanisms that calibrate the likely positions of these thresholds within speech communities, making use of the fact that interpreters of statements like (25) sometimes have independent information about how tall the subject is.(If I know that John's height is six feet seven inches, then I can infer from (25) that you assume a standard for tallness among basketball players that is less than six feet seven inches).Currently, there are several models on the table for how inferences of this kind can calibrate the thresholds that are assumed within a speech community (Potts, n.d.), (Lassiter & Goodman, 2013), (Qing & Franke, 2014).These models all treat semantic information as probability distributions and assume a radically different framework for doing natural language semantics than the one I have tacitly presupposed in this paper.That in itself is a prima facie reason to expect that the process we have focused on in this paper and the processes of adjective use described by those Bayesian models have little in common apart from the fact that they both make coordination and calibration inevitable concomitants of interpretation.When we look at these Bayesian models more closely, we find that expectation confirmed.
The second example has to do with the emergence of "non-literal" uses of predicate words (nouns, verbs, adjectives, prepositions).Well-worn examples are shark in "My lawyer is a shark," bark for what officers do when they shout instructions at their men, or idioms like bite the bullet.It can be assumed that the first instances of such uses were like new live metaphors, which contributed to the sentences containing them not only the criteria for predicate satisfaction that remain after they have become conventionalized items of the lexicon, but also the surplus value of a surprising analogy with the pre-existing "literal" use of the word or phrase.But when conventionalization sets in, the surplus loses its surprise value and gradually fadesand often it fades away completely and the connection with the literal use is lost altogether.(Who today is still aware that our current use of bite the bullet as a way to refer to decisions to go ahead with something that one knows is going to be painful or unpleasant originated as a simile with a person who has to undergo surgery without anaesthetics and who is given a lead bullet to bite on to cope with the pain as well as he can?) Three observations about the emergence and spread of new word meanings: Processes that start when a word is first used with a new meaning, which can be understood because of a salient relationship with its established ("literal") meaning, but where the new meaning then gradually becomes self-supporting and a separate item in the lexicon, are exceedingly common and probably the most important force in the extensions and refinements of the vocabularies of human languages in the course of their histories.
At least the early stages of these processes look like instances of the general pattern of "learning while doing" and community-wide coordination: Interpreters can catch on to the new meaning because of its ties to the old meaning they are familiar with, helped in this by the cultural context they share with the speaker or author and what the speaker or author is likely to want to say with her utterance.Thus the new meaning gets transferred to them and with the usual calibrating effects.The later development stages of non-literal meanings seem different.The new meaning has now become emancipated from the literal meaning, and the word as denotation of its non-literal meaning is now one of the many lexical predicates of the language, with its own conventionally fixed application criteria.The general question how predicate words can be learned and their application criteria kept stable is one of the central aims of the study of linguistic meaning from the perspective we are considering.But a detailed account of what happens in the early stages of non-literal meaning development can, I believe, be an important step towards a better understanding of this more general problem.
In the psychological and psycholinguistic literature detailed descriptions can be found of the mechanisms that lead from imaginative uses that can be made of words that endow them with new meanings to the emancipation of those meanings as independent predicate meanings in their own right (Gentner & Grudin, 1985).Much of what is relevant to the study of these mechanisms from the perspective on language discussed in this section does not seem to be very well understood.However, in recent years more formal methods have been brought to bear on the study of these mechanisms, especially in computational linguistics.Here too new research paradigms have been getting under way in which there is a strong emphasis on the social dimension of meaning.
The two mechanisms I have touched upon in this sectionthreshold determination for relative adjectives and the birth and growth to adulthood of non-literal meaningsare, I repeat, like the mechanism that governs the use, acquisition, and community-wide stability of proper names, which has been the central focus of this paper: All three work in the way that should be expected for mechanisms that are part of a highly complex and conventionalized social practice such as the human use of language.The recent explorations of these mechanisms that I mentioned in passing in this section are in the spirit of this perspective, and I am confident that the trend they have been setting will steadily gain in scope and importance.But if detailed work of this kind is a recent phenomenon, the motivation for it has long been there for all of us to see.And perhaps the most prominent place where it has been clearly visible for all this time is Naming and Necessity.What Naming and Necessity taught us about the semantics and pragmatics of proper names was a radical innovation at the time, and that was understood pretty well from the start.What could not be seen then, but what we can see today, is that it was also, in its social implications, prophetic.
Some things about the structure of ERs have been noted in passing.But what we need is a formal definition of MSDRT's concept of entity representations. 1512 ð Þ Definition of Entity Representation: An entity representation (ER) is a triple < ENT,α ½ , K descr , K anch > , where (i.a) ENT is a mode indicator, indicating that the mental state constituent to which it belongs is an entity representation, as distinct from the various modes of the possible propositional attitude constituents of MSDs; (i.b) α is a discourse referent (the distinguished dref of the ER). 15 ) to me.If I am to understand what you are telling me about John's height, I must know what threshold you are assuming for tallness among basketball players.