Unbound Anaphoric Pronouns: E-Type, Dynamic, and Structured-Propositions Approaches

Unbound anaphoric pronouns or ‘E-type pronouns’ have presented notorious problems for semantic theory, leading to the development of dynamic semantics, where the primary function of a sentence is not considered that of expressing a proposition that may act as the object of propositional attitudes, but rather that of changing the current information state. The older, ‘E-type’ account of unbound anaphora leaves the traditional notion of proposition intact and takes the unbound anaphor to be replaced by a full NP whose semantics is assumed to be known (e.g. a definite description). In this paper, I argue that there are serious problems with any version of the E-type account as well as the (original form of the) dynamic account. I will explore a new account based on structured propositions, which can be considered a conservative extension of a traditional proposition-based semantics, but which at the same time incorporates some crucial insights of the dynamic account. Unbound anaphoric pronouns, as in (1a and b), relate to an indefinite noun phrase as linguistic antecedent, but are not in a position to naturally act as a variable bound by the antecedent:1 (1) a. Someone broke in. He stole the silver. b. If someone breaks in, he will steal the silver. c. Someone might break in. He might steal the silver. On any reasonable logical analysis of the sentences in (1) (respecting minimal syntactic conditions), someone cannot be analysed as an existential quantifier binding the variable that he may stand for. Unbound anaphoric pronouns have presented major challenges for semantic theory, giving rise to new developments of dynamic semantic approaches, which imply a major revision of the notion of sentence meaning in the traditional sense. A sentence on the dynamic semantic view does not express an independent proposition, but rather serves to change the previous information state in 200 FRIEDERIKE MOLTMANN certain ways so that it will be only the entire discourse that will have truth conditions. There is an alternative to the dynamic semantic approach which is theoretically much less involving and is in fact entirely compatible with any traditional view of sentence meaning. This is what, following Evans (1985), is called the E-type approach to unbound anaphoric pronouns. The E-type approach, essentially, aims at reducing the problem of the semantic relationship between antecedent and unbound anaphoric pronoun to that of replacing the pronoun by a full NP whose semantics is taken to be well-known. The pronoun then is not interpreted directly, but rather first replaced by a nonpronominal NP whose content is retrieved from the previous discourse context (generally a definite description). The dynamic semantic approach takes unbound anaphoric pronouns to be interpreted as they are, by assimilating them to variables of formal languages. The appropriate interpretation of the pronoun is then achieved by exploiting a certain notion of context: the pronoun can be treated as if it was a variable bound by the antecedent because of the role the preceding sentential or discourse context plays in the interpretation of a sentence. The E-type approach has enjoyed renewed interest because it appears to be a solution to the problem of unbound anaphoric pronouns that does not require a reconception of meaning beyond the treatment of those pronouns themselves. It is one of the aims of this paper to examine the viability of the E-type approach in its various versions – with an overall negative conclusion. The various difficulties an E-type account faces result from the aim of solving the problem of unbound anaphora in a purely formal way, by replacing the pronoun by a full NP whose semantics is taken to be wellknown. The problems concern the identification of the descriptive content of the replacement, certain purely semantic conditions involving antecedent and pronoun, and the choice of an appropriate determiner for the replacing NP. The second goal of the paper is to evaluate the dynamic semantic approach to unbound anaphoric pronouns. This approach, which treats unbound anaphora as variables or as variable-like, appears more adequate as an account of the various kinds of occurrences of unbound anaphoric pronouns once it is extended in a certain way to modal and attitude contexts. Despite the seeming superiority of the dynamic semantic approach for the treatment of unbound anaphoric pronouns, however, the overall reconception of meaning UNBOUND ANAPHORIC PRONOUNS 201 raises serious problems regarding embedded sentences. Plausible as it may seem for independent assertions, it leads to serious problems when propositional anaphora and truth conditions of individual sentences are taken into account – problems that would never arise on an E-type account, which leaves the traditional notion of propositional content intact. A third goal of the paper, therefore, is to outline a new account of unbound anaphora that preserves the insights of the dynamic account as regards the variable-like status of anaphora and the role of the discourse context for their semantic evaluation, but at the same time does not lead to the problems of the dynamic account, by assigning a central status to the notion of proposition as the meaning of individual sentences and the object of propositional attitudes. This account is based on structured propositions and introduces a notion of a bipartite propositional content of sentences (in an utterance context). A bipartite propositional content consists of a proposition and a background, which, itself construed as a set of structured propositions, possibly provides the truth conditional completion of the proposition, which in turn may contain variablelike objects for the representation of unbound anaphora. 1. THE CONTEXTS FOR UNBOUND ANAPHORIC PRONOUNS 1.1. Extensional Contexts There are three main contexts in which unbound anaphoric pronouns occur. The first context is one in which the sentence containing the pronoun stands in a conjunctive relationship to the sentences containing the antecedent, a relationship that obtains both when one sentence merely follows the other, as in (1a), and when the sentences are conjoined by an explicit conjunction, as in (2): (2) Someone broke into the house, and he stole the silver. Because of the kind of examples standardly used, such sentences can also be called ‘conjunctive donkey-sentences’. In conjunctive donkey-sentences, the antecedent may either be specific, as is possible in (1a) or nonspecific, as more plausibly in (2). The second context for unbound anaphoric pronouns is one in which the pronoun occurs in the consequent and its antecedent in 202 FRIEDERIKE MOLTMANN the antecedent of a conditional, as in (1b) and (3), with the adverb of quantification usually: (3) If a farmer owns a donkey, he usually beats it. Again, because of the kind of examples standardly used, such sentences can be called ‘conditional donkey-sentences’. In a third context, the pronoun occurs in the scope of a quantifier and the antecedent in the quantifier restriction: (4) a. Every farmer who owns a donkey beats it. b. Most farmers who own a donkey beat it. c. Someone who owns a donkey beats it. These sentences thus are ‘quantificational donkey-sentences’. What is characteristic about all three contexts for unbound anaphoric pronouns is that the pronoun cannot be taken as a variable bound by the existential quantifier the antecedent stands for – given any plausible analysis of the relevant sentences and standard assumptions about existential quantification and about variable binding. 1.2. Intensional Subordination There is a fourth type of context in which unbound anaphoric pronouns may occur, namely contexts of what, following Roberts (1987, 1989), is called modal subordination. The phenomenon in question actually is not restricted to modals, but occurs in the same way with conditionals, temporal operators, and attitude verbs. Since what these four contexts (modal, conditional, temporal, and attitudinal contexts) share is constitute an intensional context, the phenomenon in its most general form is better called intensional subordination. Once intensional subordination constructions are ‘completed’ in the appropriate way, the pronoun in such constructions is in fact related to its antecedent in just the same way as in contexts of ordinary conditionals and conjunctions. Intensional subordination is characterized by a pronoun in the scope of an intensional operator being related to an NP in a preceding clause as antecedent that occurs in the scope of a similar intensional operator. A familiar example is (5), where it occurs in the scope of the modal would and is related to a wolf as antecedent which occurs in the scope of the modal might: UNBOUND ANAPHORIC PRONOUNS 203 (5) A wolf might come in. It would eat you first. There are two quite different types of intensional subordination: conditional intensional subordination and conjunctive intensional subordination. The first type is exemplified by (5) because here the second sentence is to be understood as a conditional of the following sort (cf. Roberts 1987, 1989):2,3 (6) If a wolf came in, it would eat you first. Obviously the pronoun in (6) can get a wolf as antecedent because it occurs in what is to be understood as a conditional donkey-sentence. I will disregard this kind of intensional subordination henceforth since it arguably involves simply a modal taking an implicit argument (an antecedent) coming from the preceding discourse, without thus presenting a specific semantic challenge as such. With conjunctive intensional subordination, the clause required for the evaluation of the pronoun stands in a conjunctive relation to the clause containing the pronoun: (7) a. John must write a paper. He must hand it in tomorrow. b. John might buy a car. He might buy it soon. (8) a. John believes that somebody broke into the house. He believes that he is a relative. b. John believes that someone broke into the house. Mary believes that he stole the silver. These are cases of conjunctive intensional subordination because they are to be understood as if a single intensional operator applied to a conjunction. That is, (7a) is equivalent to (9a), and similarly (8a) to (9b):4 (9) a. John must write a paper and hand it in tomorrow. b. John believes that somebody broke into the house and that he is a relative. (8b) is a case of socalled intentional identity describing the attitudinal states of two different agents involving the same possibly only conceived object. It is also a case of conjunctive modal subordination because the second sentence is understood as ‘Mary believes that someone broke into the house and stole the silver’. 204 FRIEDERIKE MOLTMANN It is quite clear that the three sorts of contexts for unbound anaphoric pronouns together define one phenomenon of unbound anaphora, requiring a unified theory rather than separate accounts. First, the sentence-sequencing and conjunctive cases should not be separated from the conditional cases. Otherwise it would lead to difficulties when trying to classify the quantifier restriction case, where, depending on the quantifier, we get both conjunctive (some, no) and conditional (every) relationships (and the same quantificational construction should better have a unified semantic analysis). Also the intensional subordination cases should not be set aside from the others because once analysed properly, intensional subordination simply divides into conjunctive and conditional donkeysentence constructions, and once an analysis is developed for one of those, there is no reason why it should not apply to the relevant intensional subordination sentences as well. One might argue that intentional identity cases should be treated in a special way in that here the embedded sentences involve some sort of nonexistent object and therefore do not express general propositions. However, the general properties of unbound anaphora that we will identify will hold for intentional identity cases in just the same way as for the others, and thus the same account should apply to them as well. 1.3. The Readings of Unbound Anaphoric Pronouns It is a well-known fact that there need not be a unique entity satisfying the descriptive conditions associated with the antecedent of an unbound anaphoric pronoun – in short the antecedent conditions of the pronoun. These are standard examples: (10) a. Everyone who has a dog has to register it. b. If someone has a dog, he has to register it. (11) a. Everyone who has a dime should put it in the meter. b. If someone has a dime, he should put it in the meter. (10a and b) may be true even if some of the relevant people have more than one dog, and (11a and b) may be true even if some of the people have more than one dime. What (10a) and (10b) claim is that everyone should register every dog he has (as long as he has at least one dog), and what (11a) and (11b) claim is that everyone should put some dime in the meter (as long as he has a dime). UNBOUND ANAPHORIC PRONOUNS 205 In the absence of uniqueness, as we have just seen, an unbound anaphoric pronoun may display either of two sorts of readings: an existential (‘weak’) reading and a universal (‘strong’) reading.5 The possibility of uniqueness not being satisfied is obvious in the case of relative clauses and conditionals. It is less obvious in the case of conjunctions. In fact, Evans (1985) argues that in conjunctions as in (12), uniqueness is generally implied: (12) John has a dog. He feeds it daily. However, even in conjunctive constructions, a uniqueness condition may clearly be violated, for example in (13a), and, most obviously, in conjunctive intensional subordination contexts, as in (13b): (13) a. John had a dime. He put it in the meter. b. Someone might come in. He might want to sit down. (13a) can be true even if John has more than one dime. It simply claims that John put one of them in the meter. Similarly, (13b) does not imply that only one person might come in. Somewhat different cases exhibiting failure of uniqueness are the familiar examples in (14): (14) a. If a bishop meets another bishop, he blesses him. (attributed to Kamp) b. Everyone who bought a sage plant, bought two others along with it. (cf. Heim 1982) In (14a), the pronouns he and him cannot obviously be replaced by descriptions obtained from the information given by the antecedent. Similarly, no description satisfying the uniqueness condition can be obtained from the relative clause in (14b) to replace it. The Uniqueness Problem is a notorious problem for E-type theories on which the pronoun is to be replaced by a definite description, and it has been a chief motivation for the development of the alternative, dynamic semantic theories. 2. THE E-TYPE APPROACH 2.1. Types of E-type Accounts Le me now turn to a number of other important semantic properties of unbound anaphoric pronouns and the problems the varieties 206 FRIEDERIKE MOLTMANN of E-type accounts face that try to account for those properties as well as the Uniqueness Problem. There are different versions of the E-type account. On the original account proposed by Evans (1985), an unbound anaphoric pronoun is to be replaced by a definite description, which is retrieved in some way from the antecedent clause. The relevant sentence is then interpreted by applying a familiar Russelian semantics to the replacing definite NP. Thus, the second sentence of (1a) will receive its meaning by interpreting (15):6 (15) The person who broke in stole the silver. Later versions of the E-type account deviate from Evans’s version or elaborate it further in at least one of three respects: [1] the way the pronoun relates to the antecedent, [2] the formal or contextual identification of the descriptive content of the replacement, [3] the nature of the replacement of the pronoun. These three respects correspond to three general difficulties for any E-type analysis: getting the connection between pronoun and antecedent right, identifying the descriptive content of the replacement, and choosing an appropriate determiner for the replacement. In addition to that, there are two truly semantic conditions on unbound anaphoric pronouns that are hard to account for within an E-type analysis, namely what I will call the Common-Source Condition and the Same-Value Condition. Let us now see what those difficulties are by examining the variations of the E-type account in the three respects. 2.2. The antecedent-relatedness of unbound anaphoric pronouns Unbound anaphora require an explicit antecedent (cf. Heim 1982, 1990; Kadmon 1987). In this respect they differ from definite descriptions, both (more or less) complete ones, as seen in the contrast between (16a) and (16b) (as opposed to (16c)), and incomplete ones (even when their descriptive content is just as impoverished as the one of definite pronouns seems to be), as seen in (17): (16) a. John is married. His wife is French. b. ??John is married. She is French. c. John married someone. She is French. UNBOUND ANAPHORIC PRONOUNS 207 (17) a. Mary was raped. But the man was never found. b. ??Mary was raped. But he was never found. c. Someone raped Mary. He was never found. The antecedent moreover, needs to be of the right kind (definite or indefinite) and be in the right semantic context, as the classic pair of examples below illustrates: (18) a. ?? It is not the case that no man1is walking in the park. He1 is in a hurry. b. A man1 is walking in the park. He1 is in a hurry. This antecedent-relatedness of unbound anaphora is not captured by all versions of the E-type account, for example not by an account on which the antecedent is only part of a pragmatic context that supplies the replacement of the pronoun. One such account, proposed by Cooper (1979) and Heim and Kratzer (1998) says that an unbound anaphoric pronoun has an underlying syntactic form of the sort [D [R]NP]DP, where D is a variable for a determiner and R a variable for a restriction that needs to be supplied either from the linguistic or the nonlinguistic context. Another version proposed by van der Does (1986) says that the pronoun denotes a context-dependent generalized quantifier of the form D(X) where both the determiner D and its restriction X are contextually provided.7 What a more adequate E-type account needs to say is that an unbound anaphoric pronoun is to be replaced by a description on the basis of some syntactic relation of the pronoun to an NP antecedent (and perhaps other parts of the antecedent clause). The strongest version of such a syntactic-antecedent account takes it in (12) to be syntactically related to the NP a dog and to be replaced by a description whose content corresponds to syntactically identifiable parts of the antecedent clause, such as the description ‘the N′ which VP’ in a syntactic context ‘a N′ VP’ (cf. Evans 1985, Heim 1990). Weaker versions of the syntactic-antecedent account allow for a weaker descriptive content, including perhaps not all of the predicate’s information (cf. Neale 1990) or allowing additional contextual information to complete the description (cf. Ludlow and Neale 1991). The strong version, as we will see next, is hardly tenable. 208 FRIEDERIKE MOLTMANN 2.3. Deviations from Antecedent Conditions and the Discoursedrivenness of Unbound Anaphoric Pronouns Often the replacement of the pronoun is not or not entirely syntactically identifiable from the antecedent clause, and thus not all the information provided by the antecedent clause should be part of the description replacing the pronoun. Such deviations from antecedent conditions, as I will call them, can occur both across utterances of independent sentences and in cases of modal subordination The first case, as in (19), due to Strawson (1952), involves correction or contradiction, which takes place, typically, across utterances of different speakers: (19) A: A man fell over the bridge. B: He did not fall. He jumped. Here he refers to a person of which only part of the conditions expressed by the antecedent clause hold – only a movement ‘over the bridge’. It is quite easy to see that deviations from antecedent conditions may involve any part of the antecedent conditions. Deviation from antecedent conditions also occurs with intentional identity cases. One such case, discussed by Edelberg (1985), is this. Suppose X and Y died and detectives A and B (falsely) believe that both were murdered. B believes that X and Y were murdered by the same person, but A does not. Moreover, A believes that X is the mayor, but B does not believe that. In this scenario, the following will be true: (20) A believes that someone killed the mayor and B believes that he killed Y. Here, he cannot be replaced by the one who killed the mayor, but would have to be replaced by the one who killed X. There are some limits, though, as to the extent of deviation. Edelberg (1985) points out that in general the content of the belief of the first described agent is taken over as part of the belief content of the second agent – but not conversely. Consider (21): (21) A believes that someone killed X and B believes that he killed Y. UNBOUND ANAPHORIC PRONOUNS 209 (21) implies that B believes that the person that killed Y also killed X. The converse does not hold – that is, A need not believe that the person that killed X also killed Y. This asymmetry between the reported belief of A and the reported belief of B is a problem for any account that deals with intentional identity by means of existential quantification over intentional objects (as pointed out by Edelberg 1985). The asymmetry shows instead that the evaluation of a pronoun depends, in an important sense, on the context of the discourse that precedes the pronoun: at least some minimal part of the antecedent conditions needs to be taken over by the replacement of the pronoun.8 This Discourse-drivenness of unbound anaphora, as I will call it, is captured by the E-type account, which always involves some form of copying of at least some of the antecedent conditions onto what is to replace the pronoun. The problem, however, is that no formal rule of copying can be given that could identify the replacement of the pronoun. At the same time, though, the antecedent-relatedness of unbound anaphora does indicate a formal relationship between the pronoun and the antecedent. This conflict between a formal requirement for obtaining a replacement and the lack at the same time of any formal condition on fully identifying replacement certainly does not help the case of an E-type account. 2.4. The Regress Problem A rather different kind of problem with identifying the descriptive content of the pronoun’s replacement is analogous to the one posed by so-called Bach–Peters sentences, as in (22a). Any replacement of the pronouns in (22a) by a description would lead to an infinitive regress, as in (22b), or else lead to the wrong interpretation: (22) a. Every pilot who shot at it hit the MIG that chased him. b. Every pilot who shot at the MIG that chased every pilot ... hit the MIG that chased him. The same problem arises with donkey-sentences, as was noted by Martin Stokhof (p.c.): (23) a. If a farmer owns a donkey, he beats it. b. If a farmer owns a donkey, the farmer who owns the donkey ... beats the donkey he owns. 210 FRIEDERIKE MOLTMANN If he in (23a) is replaced by the farmer who owns it, then we have a description with a donkey-pronoun and hence not yet an interpretable full NP, and if he is replaced by the farmer who owns the donkey the farmer owns, we will not get the right interpretation, since the farmer alone is not a complete description. If he is replaced by the farmer who owns a donkey, then we have a problem with the replacement of it: if it is replaced by the donkey he owns, we still have a donkey-pronoun. But if it is replaced by the donkey a farmer owns, we get the wrong interpretation for the entire sentence. As in the case of Bach–Peters sentences, this points at the status of unbound anaphora as variable-like rather than descriptive pronouns (though, as we will see, there are also situation-based solutions available).9 2.5. The Common Source Condition Another general problem for the E-type account – as replacing the pronoun by an NP with a known semantics – is that there are purely semantic conditions governing the evaluation of the pronoun in relation to its antecedent. One such condition, noted and/or discussed by Asher (1987), Dekker and van Rooy (1998), Groenendijk et al. (1996b), and Zimmermann (1998), is what I will call the Common Source Condition. This condition manifests itself in sentences with intentional identity as in (24a and b): (24) a. A believes that someone killed X. B believes that he killed Y. b. John said that someone broke into the apartment. Mary said that he stole the silver. (24a) is acceptable only if A and B have a common source for their belief (e.g. having been presented with the same piece of evidence), or else if there is a communicative link between A’s and B’s belief (cf. Asher 1987). Similarly, for the report about John’s and Mary’s utterances in (24b) to be acceptable John and Mary must have been exposed jointly to the same evidence or have talked about some evidence to one another. The Common-Source Condition is fully general: it holds whenever an anaphoric pronoun takes an antecedent from the attitude context of a different agent. It can be observed even across utterances of different speakers: UNBOUND ANAPHORIC PRONOUNS 211 (25) a. A: Someone killed Mary. b. B: ?? He is insane / He must be insane. As Groenendijk et al. (1996b) observe, the first sentence in (25b) is inappropriate if B does not express his thoughts on the basis of the same direct evidence as A (and hence A and B act as a single agent). In this case, rather, the epistemic modal must has to be used. The Common Source Condition may manifest itself in various forms; what always matters, though, is that the agents seem to act as a single agent with respect to their beliefs or speech acts or whatever reported propositional attitude. There is a fundamental difficulty for any E-type account to explain the Common-Source Condition. If crossattitudinal anaphora are handled as at a purely formal level, by replacing them with appropriate full NPs, there is no reason to expect there to be any content-related condition between the attitude reports. An E-type account could only stipulate such a condition. 2.6. The Same Value Condition and Covariation Unbound anaphoric pronouns are subject to another condition that can hardly be conceived as a purely formal condition on the antecedent or the replacement of the pronoun. This condition, which I will call the Same Value Condition, says that it is not sufficient for the pronoun to be provided with descriptive conditions coming from the environment of the antecedent, but that it also has to have only semantic values shared by the antecedent. The Same Value Condition manifests itself in cases such as the following in which the antecedent NP contains a pronoun acting as a variable: (26) a. Many women saw portraits that resembled them. Many men saw them too. b. Last year, John met a student who would become his first assistant. Bill met him then too. The pronouns in the second sentences in (26) cannot have a reading on which the pronoun in question would stand for a description containing a variable bound by the subject. Thus, (26a) cannot mean ‘many women saw portraits that resembled them and many men saw portraits that resembled them (the men)’, and (26b) cannot mean that John and Bill both met their future assistants last year.10 212 FRIEDERIKE MOLTMANN There is a second variant of the Same-Value Condition and that is that the definite NP, were it to replace an unbound anaphora, has to have a rigid interpretation (cf. Evans 1985), or better, has to be interpreted as covariant with its antecedent (Soames 1989, Neale 1990) – that is, the replacing definite NP has to have the same value as its antecedent at a given context of evaluation in intensional contexts. This is illustrated by the coherence of the following examples: (27) a. John owns a donkey and it likes carrots. But it might have been that it did not like carrots. b. John hired a very good assistant. But he soon won’t be his assistant anymore. Let me call this property of unbound anaphoric pronouns in modal contexts covariation. Thus, the Same Value Condition prohibits antecedent and pronoun to take different values across changes of contexts induced by variable assignments and by an intensional operator. It is hard to account for a purely semantic condition such as the Same Value Condition within an E-type account other than by supplementing it by that very same condition. Covariation only obtains for modal and temporal contexts, not for contexts of propositional attitudes. Thus, no rigidity effect can be observed in (28): (28) Someone broke into the house. Mary believes that he stole the silver. The natural reading for he in the second sentence is not a de re reading, as covariation would have it, but a de dicto reading (Mary believes that whoever broke into the house stole the silver). The reason obviously is that Mary shares the speaker’s assumption, namely that someone broke into the house. If unbound anaphora can take into account such implicit information, then clearly the second sentence of (28) is simply a case of intensional subordination rather than a violation of covariation. 2.7. Problems with the Choice of the Determiner We have seen that unbound anaphoric pronouns may fail to denote a unique entity, as is necessary to satisfy the Russellian semanUNBOUND ANAPHORIC PRONOUNS 213 tics of any replacing definite descriptions. This obvious failure of uniqueness as well as the possibility of weak and strong readings have led to various proposals of how to modify Evans’ original E-type account. The proposals fall into two kinds. One kind takes the pronoun to be responsible for the weak and strong readings in the absence of uniqueness, positing two other sorts of replacements of the pronoun besides a singular description. The other proposal modifies the evaluation of the antecedent so as to retain the uniqueness condition, relativising the values of the antecedent to situations or events. The general problem of those proposals, we will see, is that the proposed modifications of the semantics of sentences with unbound anaphoric pronouns always lead to problems in some cases or other that ultimately should fall under the same account. 2.7.1. Modifying the Replacement of the Pronoun One proposal of the first sort allows singular unbound anaphoric pronouns to stand for groups (or sums) rather than individuals, where groups are taken to be the kind of objects that definite plurals such as the donkeys refer to, given the most common view of the semantics of plurals (cf. Link 1983 and others). Let me call this the group-referential (E-type) account. This account has been proposed for both the weak and the strong reading by Lappin and Frances (1995) and Krifka (1996) and for only the strong reading by Chierchia (1995). On the group-referential E-type account, the strong reading of a donkey-sentence would be represented as in (29a) or as in (29b), where f is a function mapping farmers to the groups of donkeys they own: (29) a. Every farmer who owns a donkey beats the donkeys he owns. b. Every farmer x who owns a donkey beats f (x). To account for the weak reading, Lappin and Frances (1995) allow the function f to, alternatively, act as a selection function, mapping a farmer to some group consisting of donkeys that the farmer owns (rather than necessarily the maximal group). The obvious advantage of this proposal is that different occurrences of unbound anaphoric pronouns – those meeting the uniqueness condition, those displaying a universal reading, and those displaying an existential reading – are given a unified treatment. 214 FRIEDERIKE MOLTMANN However, the proposal faces also some rather criticial problems, as discussed in great detail by Kanazawa (2001). The use of group-valued functions generates readings of singular pronouns analogous to those of plural pronouns in places where such readings are entirely unavailable. For example, with an appropriate predicate such as weigh or gather the group-referential E-type account would predict collective readings to be possible: (30) a. Every farmer who owns a donkey weighed it. b. Every farmer who owns a donkey gathered it in the yard. Weighed in (30a) should have a collective reading on which every donkey-owning farmer measured the ‘collective’ weight of the donkeys he owns. But (30a) only allows the reading on which the weight of any individual donkey a given farmer owns is measured. Similarly, one would expect (30b) with an obligatorily collective predicate to be acceptable, which it isn’t. A way of rescuing the proposal might be by stipulating that group-referring singular pronouns have to be obligatorily interpreted distributively. But even then problems arise, for example with (31a) and (31b): (31) a. Every man who has a dog complained about the doctor that examined it. b. Every man who has a daughter told her teacher that she is talented. (31a) allows for a universal reading; hence it in the relative clause would be interpreted as referring to the entire group of dogs of a given man. Now if at least one man has more than one dog, the sentence would imply that all the dogs were examined by the same doctor. But the sentence does not imply that. Also (31b) has a universal reading. Then she would have to refer to the group of daughters a single man has. This, however, implies that each man told the teacher (or the teachers) of his daughters this: ‘they are talented’. But (31b) allows each man to have uttered only sentences of the form ‘she is talented’ for each daughter. Thus, she in (31b) cannot generally refer to the entire group of daughters a given man has.11 The other proposal that modifies the replacement of the pronoun allows the pronoun to be replaced by a quantifier other than a definite description. Thus, for the strong reading, the pronoun would be replaced by a universal quantifier and for the weak reading by an UNBOUND ANAPHORIC PRONOUNS 215 existential one. Let me call this the quantificational (E-type) account. For the universal reading, such an account has been proposed by Neale (1990), and in its full generality, by van der Does (1996).12 On the quantificational view of E-type pronouns, (10a) would be interpreted as equivalent to (32a) and (11a) to (32b): (32) a. Every man who has a dog has to register every dog he has. b. Every man who has a dime should put some dime he has in the meter. There are two serious problems with the quantificational E-type account. First, E-type pronouns do not display any of the scope interactions with other quantifiers or operators that ordinary quantifiers do. The universal quantifiers E-type pronouns are supposed to stand for can, for example, never take narrow scope with respect to negation: (33) Every farmer who has a donkey does not beat it. (33) disallows a reading on which the sentence is true just in case one of the donkeys of any given farmer fails to be beaten. Also the supposed quantifiers do not interact in scope with any other quantifiers: (34) a. Every man who has a dog bought exactly one leash for it. b. Every man who has a dog told a neighbour about it. Given a universal reading with some men owning more than one dog, it is impossible to interpret it in (34a) with narrow scope (so that every man bought exactly one leash for every dog he owns). (34a) can mean only that each man bought exactly one leash for each of his dogs. (34b), similarly, excludes a reading on which every man must have told the same neighbour about each of his dogs. Another problem with the quantificational E-type account is that like the group-referential account, it yields wrong results with relative clauses and clauses embedded under attitude verbs as in (35): (35) Every man who has a daughter told the professor she wanted to study with that she was very talented. Given a universal reading (35) would be interpreted as equivalent to (36), which is not an available reading (given that at least one of the men has more than one daughter): 216 FRIEDERIKE MOLTMANN (36) Every man who has a daughter told the professor every daughter wanted to study with that every daughter he has is very talented. This type of problem also occurs within conditionals, as was pointed out by Barker (1987). Thus, in (37), the first it does not allow for a replacement by a universal quantifier, and the second it does not allow one by either a universal or an existential quantifier: (37) If a theory is classical, then if it is consistent, it is (usually) trivial. Let me call the problem posed by (35) and (37), i.e. the problem that the pronoun acts like a singular term yet cannot stand just for a single object, Barker’s problem. Given the problems with modifying the replacement of the pronoun discussed in this section, we can more generally say that the different readings of sentences with unbound anaphoric pronouns cannot be traced to the interpretation of the pronoun itself, but rather must somehow be traced to the interpretation of the antecedent. This leads to the second way of modifying the E-type account to solve the uniqueness problem, the use of situations. 2.7.2. Using Situations Using situations or events is a popular and often fruitful strategy in semantic analysis and it is no surprise to find the strategy taken in various efforts to rescue the E-type account of unbound anaphoric pronouns. The idea is that a singular definite description, in order to ensure uniqueness, is relativized to a situation or event (cf. Berman 1987; Kadmon 1987; Heim 1990; Ludlow 1994; Elbourne 2001). Let me call this the situation-based E-type-account. Because it seems at first sight so promising, it is appropriate to discuss the possibilities this account permits at some greater length. A situation-based E-type-account generally takes situations to be introduced by an implicit situation quantifier that is associated with the context in which the antecedent of the pronoun occurs. Let us see how this works first with a conditional such as (38): (38) If a farmer owns a donkey, he beats it. On the situation-based view (38) expresses universal quantification over situations and is equivalent to (39): UNBOUND ANAPHORIC PRONOUNS 217 (39) Every minimal situation s in which a farmer owns a donkey can be extended to a situation s ′ in which the farmer in s beats the donkey he has in s. The minimality condition is necessary to guarantee the uniqueness of the referent of the description. It is necessary also to account for adverbs of quantification like usually as in (40a), which are taken to range only over minimal situations described by the restriction of the adverb of quantification, i.e. the antecedent in (40a): (40) a. If a farmer owns a donkey, he usually beats it. b. For most situations s such that s is a minimal situation in which a farmer owns a donkey, there is an extension s ′ of s such that the farmer in s beats the donkey he owns in s. Quantification over situations seems fairly well motivated for certain indicative conditionals, especially those with adverbs of quantification.13 The situation-based E-type account also allows for ways to avoid the Regress Problem. First, the problem would be avoided if it is not required that all of the antecedent conditions make up the replacing description. Thus in a minimal situation s in which a farmer owns a donkey in (40a), the farmer in s and the donkey in s will already satisfy the uniqueness condition.14 Moreover, even suitably chosen full descriptions would do, namely the farmer who owns a donkey and the donkey that the farmer who owns a donkey owns. However, the situation-based account would have to posit situations for all contexts in which unbound anaphora can occur (since the Uniqueness Problem is general), which is hard to motivate. First, the situation-based account needs to be extended to unbound anaphora with an antecedent occurring in the relative clause of a quantified NP. Implicit quantification over situations must then somehow go along with the quantification expressed by that NP. Thus (41a) would have the analysis in (41b), on a universal reading: (41) a. Every farmer who has a donkey beats it. b. For every farmer x: every minimal situation s such that x has a donkey in s can be extended to a situation s ′ in which x beats the donkey he has in s. But why should (41a) be about situations in addition to being about farmers and donkeys? There does not seem to be any motivation 218 FRIEDERIKE MOLTMANN for introducing implicit situation quantifiers for quantificational donkey-sentences other than the Uniqueness Problem itself arising within the E-type account. Besides that, technical problems arise when trying to accommodate the weak reading of an unbound anaphoric pronoun, as in (42a), which cannot be analysed as in (42b): (42) a. Everyone who had a dime put it in the meter. b. For every person x: if there is a minimal situation s such that x has a dime in s, then s can be extended to a situation s ′ such that x puts the dime in s in the meter in s ′. In (42b), an existential quantifier ranging over situations occurs in the antecedent of a conditional which would have to bind the variable s in the consequent – thus, we have the familiar ‘donkey-problem’ in another form. Other cases where situations are hard to motivate are subjunctive conditionals as well as indicative conditionals with an ‘epistemic use’, relating to the world as such, rather than describing regularities among situations. These conditionals say something about a counterfactual state of the entire world or about inferences one would draw about it on the basis of new beliefs. Thus, it is hard to see how conditionals such as (43a) and (43b) can be taken as quantifying over situations, since there is neither partiality nor any spatio-temporal limitation involved: (43) a. If Mary had a son, she would have sent him to a good school. b. If Mary has a son, she will have sent him to a good school. (43a) and (43b) say something about how the world would be if Mary had a son, not something about a particular situation, there does not seem to be a reason to take them to quantify over situations in which Mary has a son. Another problem with the situation-based view, discussed in Heim (1990) and Ludlow (1994), is that situations do not provide sufficient information to guarantee the uniqueness of the referent, as in (14a) and (14b), where the pronouns in the consequent cannot be replaced by a definite description even when relativized to a minimal situation described by the antecedent. To account for such cases, one might take situations to be very fine-grained. For (14a) such a situation may UNBOUND ANAPHORIC PRONOUNS 219 include only the information that x meets y without including the information that y meets x or it may involve different thematic roles for different thematic roles for he and him (cf. Ludlow 1994). This way, however, not only do situations loose their intuitive content, but also cases such as (14b) are still not accounted for. An alternative pursued in Heim (1990) is to have the conditional in (14a) quantify over minimal situations containing one bishop, claiming that any such situation s when extended to any situation s ′ containing another bishop so that the two bishops met in s ′ can be extended to a situation in which the bishop in s blesses the other bishop in s ′. But this account does not work for examples such as (44): (44) If a book very similar to another book is published at the same time, it is generally immediately compared to the other book. Here any situation in which the antecedent is true, must already contain at least two books. Another problem with the situation-based E-type account arises with nonpersistent quantifiers – that is, quantifiers like no, exactly two, at most two, and few, which do not yield the same truth value under extensions of the model or domain. On the situation-based view as formulated so far, the following examples will not get the interpretations they in fact have: (45) a. If a farmer has exactly one donkey, he beats it. b. If a farmer has no donkey, he is poor. (46) a. Every farmer who has exactly one donkey beats it. b. Every farmer who has no donkey has nothing to do. (45a) would be true if any minimal situation in which a farmer owns exactly one donkey can be extended to a situation in which the farmer beats the donkey. But for any farmer who owns n donkeys there will be n minimal situations in which he owns exactly one donkey. This means that (45a) expresses the proposition that every farmer whatsoever beats every donkey he owns, which is clearly wrong. Also, for (45b), the proposal leads to a disaster. The sentence would express the proposition that every farmer – regardless of whether he has a donkey or not – is poor. (46a) and (46a) would be misinterpreted, too. The quantified NP in (46a) would range over all farmers that have one or more donkeys, not only those with exactly one donkey, 220 FRIEDERIKE MOLTMANN and the NP in (46b) would range over all farmers, whether they have a donkey or not. The proposal actually makes it impossible to talk about farmers who own a limited number of donkeys, since when minimal situations are invoked, such farmers would not be distinguished from farmers who own an unlimited number of donkeys. There are two potential ways of modifying the account so as to accommodate sentences with nonpersistent quantifiers. The first one is to modify the interpretation of nonpersistent quantifiers, as has been proposed by Kratzer (1989) in the context of a situationsemantic treatment of conditionals. She gives the following rules for the interpretation of nonpersistent quantifiers in a situation, where ws is the world that the situation s belongs to: (47) Persistent interpretation of quantifiers For a quantifier Q, [Qx, Fx: Gx] = true in a situation s iff for Q-many x-alternatives g′ of g such that [Fx] ′ = true in ws , [Fx] ′ = true in s and [Gx] ′ = true in s. That is, when a quantifier in a sentence S is interpreted persistently in a situation s, then for S to be true in s, the entities in the entire world have to be counted that satisfy the restriction and the scope of the quantifier, not just those in s. (47), however, leads to problems when applied to E-type pronouns. If quantifiers are systematically interpreted persistently, then also the definite description replacing an E-type pronoun should be interpretable that way – since a definite description, on a Russellian account, simply denotes a particular nonpersistent quantifier. But if (47) also applies to definite descriptions, the very idea of the situation-based E-type account is missed – namely, that of making the description satisfy the uniqueness condition when relativized to a situation. One might try to somehow exempt definite descriptions generally from (47). But this gives wrong results in other cases – for example, when an overt definite description occurs in the antecedent of a conditional or in a relative clause, as in (48): (48) a. If a student finds the solution to the problem, he will write it down. b. If a student finds a solution to the problem, he will write it down. (48a), unlike (48b), presupposes that there is exactly one solution in the world to the problem in question. Thus, the uniqueness condition of UNBOUND ANAPHORIC PRONOUNS 221 overt descriptions may have to be satisfied relative to the world rather than a minimal situation, requiring the persistent interpretation. Alternatively, one might take definite NPs to be interpreted persistently only in certain contexts – namely only when occurring in the antecedent of a conditional or the restriction of a quantifier, but not in the consequent or in the quantifier scope. But also this would be wrong. For example, (47) should not apply to the definite NP replacing a pronoun in a conjunctive donkey-construction when it occurs in the antecedent of a conditional, as in (49): (49) If John owns a donkey and beats it, Mary will be upset. The exemption from (47) in the relevant contexts, thus, would have to be limited to descriptions replacing pronouns, rather than applying to descriptions in general. But then the replacement of the pronoun would not be interpreted like any other NP of the same form (and thus a reduction of the semantics of unbound anaphora to the semantics of ordinary full NPs would not be achieved). Recall also from Section 2.2. that definite NPs may act just like donkeypronouns, in which case they are clearly not interpreted by (47). There is another possible way of rescuing the situation-based view of E-type pronouns, and that is to modify the notion of situation. The proposal would be that a (nonpersistent) quantifier plays a fundamentally different role in a situation than an indefinite or referential NP. A quantifier would contribute not an individual as part of a situation, but a set of properties (a generalized quantifier); only indefinite and referential NPs would contribute an individual. Thus, the situations described by the antecedent of (45a) would be something like sequences of the form , where a is a farmer and [exactly one donkey] the set of sets containing just one donkey. The problem with this is that certain quantified NPs behave just like indefinites in their support of unbound anaphoric pronouns. (48a) and (48b) are examples that are still accounted for since the consequent has to be true with respect to an extension of a minimal situation satisfying the antecedent, which means a situation to which a donkey could have been added. However, a case like (50) is a real problem: (50) If a couple has exactly three children and they receive only 200 dollar a month child support for them, they may not be satisfied. 222 FRIEDERIKE MOLTMANN If exactly two children in (50) contributes a property of properties to a situation, rather than individuals, then there is no guarantee that there will be an appropriate object in the minimal situation satisfying the antecedent that would be the referent of them. For if them as an E-type pronoun is replaced by the children they have, a minimal situation could just contain one or two children to act as the referent of that description relativized to the situation. Nothing will ensure that only the total of the three children can be the referent of them.15,16