LINGUIST List 17.3393

Mon Nov 20 2006

Review: Semantics, Syntax: Elbourne (2006)

Editor for this issue: Laura Welcher <lauralinguistlist.org>


Directory         1.    Emar Maier, Situations and Individuals


Message 1: Situations and Individuals
Date: 20-Nov-2006
From: Emar Maier <emar.maiergmail.com>
Subject: Situations and Individuals


Announced at http://linguistlist.org/issues/17/17-1045.html

AUTHOR: Elbourne, Paul D. TITLE: Situations and Individuals SERIES: Current Studies in Linguistics PUBLISHER: MIT Press YEAR: 2006

REVIEWER: Emar Maier, Department of Philosophy, University of Amsterdam and Radboud University Nijmegen, The Netherlands

This book provides a unified analysis of third person pronouns, definite descriptions and proper names, in effect treating them all as descriptions. The bulk of the book is concerned with the analysis of pronouns as definite descriptions, which takes its cue from the analysis of so-called donkey pronouns as abbreviated descriptions:

(1) Every man who owns a donkey, beats it

This 'it' is called a donkey (or E/D-type) pronoun because it is intuitively bound but not c-commanded by 'a donkey'. The position defended by Elbourne (henceforth, E) is that this 'it' means something like 'the donkey', which, with the help of a finegrained situation semantics, picks out exactly those donkeys made available by the quantifier's restrictor. He goes on to show that this semantics extends to all the individual denoting expressions.

SUMMARY

Chapter 1 (pp.1-40) addresses the problem posed by donkey sentences like (1), viz. binding without c-command. The literature provides three solutions: the D-type approach sketched above, dynamic semantics, and variable-free semantics. A prima facie problem with the first is the uniqueness presupposition of definite descriptions, i.e. does (1) really imply that each man only has one donkey? Heim's (1990) solution uses Kratzer's situation semantics: every minimal situation s in which a man owns a donkey can be extended to a minimal situation s'>s where the unique man beats the unique donkey. This makes sense because the minimal situations involved really do contain only one donkey. Three serious problems for the D-type analysis remain:

- Indistinguishable Participants:

(2) If a bishop meets a bishop he blesses him

D-type analysis: for every minimal s of a bishop meeting a bishop, there is a minimal s'>s where the bishop blesses the bishop. This s' thus contains two bishops, neither of which is unique so the pronouns/definite descriptions would not be licensed.

- Formal Link:

(3) Every {man who has a wife/*married man} is sitting next to her

Since, semantically, there is no difference between these two, how come the second one is out?

- Pronominal Ambiguity: Not every pronoun is treated as a definite description, most are still treated as variables.

But the competition also suffers from a number of problems. Dynamic semantics, which addresses the problem of pronominal ambiguity by radically changing the logic of variable binding, has trouble with a variety of pronoun uses that are straightforwardly analyzed by D-type theories (e.g. paycheck sentences). Another alternative, Jacobson's (2000) variable-free semantics, also solves the pronominal ambiguity issue, but still suffers from the indistinguishable participants and formal link problems.

Chapter 2 (pp.40-92) fleshes out a new kind of D-type analysis, involving NP deletion. The idea is that (i) a donkey pronoun has the syntax and semantics of the determiner 'the', and (ii) it is followed by an phonologically deleted NP at LF. The LF of (1) thus ends in '...beats [it donkey]', meaning '...beats the donkey'. With respect to the three problems for Heim's analysis, we see that the formal link problem is solved by the fact that NP deletion requires a previous mention of the same NP; the other two problems are addressed in chapters 3 and 4. E further introduces a new set of data, involving elliptical continuations of donkey sentences:

(4) In this town, every farmer who owns a donkey beats it, and the priest does too [*sloppy, strict]

Heim would incorrectly derive a sloppy reading (priest beats his own donkey), while the NP deletion approach correctly predicts only the strict reading. In the final sections E discusses a variety of other data that a D-type theory can handle (but the competition can't): Bach-Peters and paycheck sentences, and modal subordination. Finally, E addresses some arguably problematic donkey phenomena. For instance, how to get weak readings, i.e. how to prevent (5) from meaning that everyone put all their dimes in the parking meter, as the current proposal seems to predict?

(5) Everyone who had a dime put it in the parking meter

Chapter 3 (pp.93-136) picks up the issue of pronominal ambiguity, so the first task is to extend the NP deletion approach to bound and referential uses of pronouns. In Heim & Kratzer, a regular pronoun is adorned with an index, a natural number which is mapped to an individual by the assignment function. Binding occurs when an index is lambda bound; (direct) reference when it is free. The new unification consists in analyzing indices as phonologically null NPs, so that a pronoun with a regular NP as argument (remember pronouns are determiners) is a D-type pronoun, while with an index NP, it's a regular pronoun.

On to overt definite descriptions. There are two schools of thought with respect to the semantics of these. The Fregean claims that the uniqueness requirement is a presupposition, the Russellian puts it in the truth conditions. An argument for the Russellian approach is the different scope possibilities, but the Fregean can do similar things if we allow explicit quantification over world variables. A second argument pro Russell involves embedded non-denoting definite descriptions that do not produce presupposition failure. Here, Karttunen's (1974) solution is adopted to save the Fregean account. All in all the Fregean analysis comes out the winner. But it needs to be amended in view of examples of bound descriptions. E settles on the option of adding an index as a second argument to the overt definite determiner. The semantics is as follows: || [[the 1] farmer] ||^g = the unique x such that x is a farmer and x=g(1). We single out a special index 0 to handle D-type uses: || [[the 0] farmer] ||^g = the unique x such that x is a farmer. In the final section E proposes to add a second NP argument to pronoun-determiners as well, thus completing the unification of definites and pronouns: it ~> [[it 3] donkey]. Supporting evidence comes from Sauerland's (2002) analysis of focussed bound pronouns and the intuitive analysis of resumptive pronouns as spelled-out traces.

Chapter 4 (pp.137-157) tackles the problem of indistinguishable participants, cf. Kamp's bishops, (2). Dynamic semantics correctly predicts the bishop sentence (2) to mean: for all x, all y: if x and y are bishops and x meets y, then x blesses y. Note already that, since meeting is a symmetric relationship, this entails that we see each bishop blessing his colleague and being blessed in return. A D-type account has trouble since pronouns are definite description requiring unique bishops. After showing why previous D-type solutions fail, E points out a neglected contrast between (2) and (6):

(6) *If a bishop and a bishop meet, he blesses him

Dynamic semantics cannot distinguish these sentences, but E's new D-type solution not only assigns the right truth conditions to (2), it also explains the infelicity of (6). His solution hinges on a subtle situation semantic interpretation of (2) that manages to distinguish the two bishops, or rather the two minimal situations in which they occur. The first situation contains just a bishop x, the second a bishop y meeting x. With the two bishops distinguished in this way, it is not hard to postulate contextually salient, deleted descriptions to pick them out in the consequent. Appealing to the syntactic Coordinate Structure Constraint E shows that the LF of (6) does not allow the bishops to be distinguished in this way.

Chapter 5 (pp.138-168) uses the Japanese pronoun 'kare' (he) in an argument for D-type, against dynamic and variable-free semantics. The observation is that 'kare' can be used as a donkey or referential pronoun but not as a bound variable, which would be problematic for these fully unified analyses. E's solution is to treat 'kare' as a traditional definite description, 'the unique male individual', without an index argument. Thanks to our situation semantics we get co-variation in donkey sentences, but because of the lack of index we do not get real variable binding. However, this theory overgenerates, i.e. it predicts co-varying readings in simple quantificational binding sentences, where 'kare' is actually out. To solve this, E appeals to Reinhart's Rule I as a kind of blocking mechanism: when a bound variable would be grammatical, use a regular pronoun (null in Japanese) instead of 'kare'.

Chapter 6 (pp.169-184) extends the unification to proper names, the last category of referring expression eligible for unification with pronouns and descriptions. Treating names as descriptions is already an established, though minority, view. E adapts the analysis of Burge (1973) and others that says that 'Alfred' is really a noun referring to the property of being called Alfred, as is apparent from sentences like 'There are a lot of Alfreds at MIT'. The unmodified occurrence is then analyzed as modified by a silent definite determiner. Add an index and the unification is complete: Alfred ~> [[THE 2]Alfred], interpreted as 'the x such that x is called Alfred and x=g(2)'. E shows how this analysis is immune to Kripke's famous objections, pointing out the referentiality of the index determined by the assignment. But, if names are like definite description and pronouns, we'd expect not just referential readings, but also bound and donkey uses. According to E, these exist, though their occurrence is obviously highly constrained, e.g. by Principle C, which might explain why they haven't been noticed before (see below). The crucial example of a donkey proper name is (7):

(7) Every woman who has a husband called John and a lover called Gerontius takes only Gerontius to the Rare Names Convention

Such donkey uses of names are extremely puzzling on the standard view of proper names as directly referential terms, yet they are straightforwardly analyzed (with index 0) on the current theory.

EVALUATION

Because of its always clear structure and style, never straying too far from the main point (the unification of type e expressions), I found the book very accessible. This is not to say that E skimped on technicalities; where necessary, there is enough formal rigor. And every formula is supplemented with explanations and/or easy to read diagrams. My main criticism would have to be a rather superficial matter of layout, probably to be blamed on the editor: why endnotes, numbered per chapter? I found it pretty disruptive to keep turning to the end and then look for the right note 7. That said, I also encountered a few points of substantial disagreement, which I discuss below.

First off, as is almost inevitable when reviewing a book like this, I found some judgments questionable. A case in point is E's set of examples of donkey ellipsis (p. 69), which distinguish Heim's D-type analysis from his own. I, for one, am not convinced that (4) [above] really lacks a sloppy reading, and if it does, I think that's better explained as a pragmatic artifact of the choice of words (priests don't usually have donkeys). This idea is almost suggested by E's own (very specific class of) counterexamples: cases where the subject of the elided clause is included in the class of subjects of the target clause do permit sloppy readings. E gives a rather ad hoc, pragmatic explanation (involving, what I would call local accommodation) for them, but I think the sloppy reading is in fact generally available and it is easier to appeal to pragmatic factors to explain why sometimes it is *dis*preferred. For instance, it seems to me (and some colleagues, one native) that sloppy readings are available for (8):

(8a) Every child that owns a donkey loves it, and so does the biggest donkey farmer in town

(8b) Every male farmer who owns a donkey beats it, but farmer Mary doesn't

The judgment is crucial: E predicts that the sloppy reading is totally out, so if my judgments are correct, the phenomenon of donkey ellipsis, presented as a knock-down argument against Heim, backfires. My next point is more theoretical. It concerns the data and analysis of 'sore' and 'kare'. A good part of the introduction is devoted to Kurafuji's (1999) arguments and data that show that Japanese has two kinds of pronouns: a null pronoun which behaves like English 'he', and an overt one, 'sore', that can be bound or referential but not donkey. This may well be the case, but I don't see that this requires a bunch of Japanese glosses, because in all examples the English demonstrative pronoun 'that' patterns exactly with 'sore'. Moreover, it is actually quite uncontroversial to assume that 'sore' is indeed a (distal/medial) demonstrative (and this is even very briefly hinted at by E). The basic observation would remain, but much clearer and more general: distal demonstrative pronouns are like third person pronouns in being bindable or referential, but they differ in not allowing donkey readings. Given this, I think the explanation of chapter 5's 'kare' as a proximal demonstrative deserves more attention than E gives it. Admittedly, E does mention such an account, which he quickly tries to dismiss with an example of an English distal demonstrative (cf. p.162 and n.3 , p.215). I'm not claiming a ready-made solution, but I do feel that the demonstrative analysis deserves a little more attention here.

Another theoretical remark concerns the semantics of proper names. E successfully refutes two out of the three Kripkean arguments against descriptivism, but the first one, I think, still applies to his new descriptivist proposal. This is Kripke's contrast between 'John is called John' (contingent truth) and 'The person called John is called John' (necessary, analytic truth). As far as I can tell, E's analysis of the former turns it into a necessary truth on a par with the latter, because the property of being called John ends up as part of the asserted contribution of a proper name 'John'. Note that this is obscured by the fact that, in a sense, E's names are also rigid: ''the whole descriptive content [of 'Socrates'] will be 'entity identical with Socrates and called ''Socrates''''' (p.173). In other words, the index determines a rigid component (viz. being equal to the actual Socrates), but in addition the name contributes something to the proposition, viz. that this individual is called 'Socrates'. Thus, 'Socrates is called Socrates' comes out equivalent to, say, 'Socrates is Socrates'.

I finish with two objections of a methodological nature. The first involves the general semantic framework. Cast in the Heim & Kratzer (1998) syntax-semantics framework and, in particular, building heavily on Kratzer's (1989) situation semantics and Heim's (1990) D-type analysis, E's semantic reduction is carried out 'East Coast fashion'. I feel that in some places the discussion might have benefited from a somewhat wider view of the diverse field of semantics. I am no impartial judge myself, but I'd recommend a closer look at the presupposition-as-anaphora theory in DRT, which has a lot to say about donkey anaphora, definite descriptions, pronouns and also proper names. A case in point is Geurts' (1997) analysis of proper names as presuppositions, i.e. just like definite descriptions and pronouns. Indeed, so close are the parallels between the main claims of that paper and this book that Geurts even came up with the same kind of evidence: Geurts' example of a bound proper name below, for instance, mirrors E's D-type example (7) to a T:

(9) If a child is christened Bambi and Disney Inc. hear about it, they will sue Bambi's parents

And there are other missed opportunities of comparison between the DRT framework and the East Coast way. The rather vague discussions of fixes that involve contexts being pragmatically enriched, call to mind alternative, perhaps cleaner, formulations involving accommodation of presuppositions (e.g. the explanation of the apparent counterexamples to E's prediction that elided donkeys have only strict readings, or the rather complicated Karttunen analyses of non-denoting descriptions (p.108)). I think the choice of identifying dynamic semantics with basic DPL is unfortunate for the purpose of this book, and a more thorough look at DRT and presupposition theory would have been more illuminating than the comparison with variable-free semantics.

Another roughly methodological worry is that E sometimes seems to claim more than he proves. A first case in point is his overall claim of formulating a unified semantics of the expressions used to refer to entities in natural language. In practice he covers third person pronouns, definite descriptions and proper names. He completely fails to mention indexical pronouns ('I', 'you') and demonstrative NPs ('this', 'that man over there'), which are, in a sense, the paradigm of referring expressions. I don't see immediately that E's D-type analysis extends naturally to these, but if it does, a few words pointing it out would have certainly been in place. A second example of E overstating his achievements occurs when he uses Reinhart's Rule I to overcome a prima facie problematic prediction of his analysis of 'kare'. E tries to turn this reliance on a pragmatic rule into a virtue by boldly presenting it as evidence for Rule I, rather than a weakness of his own theory, fixed by an arguably ad hoc application of another. A final example of this I found in E's reaction against the seemingly problematic weak readings: his D-type analysis of (5) would predict that everybody put all their dimes in the meter, which is clearly against our judgments. He boldly claims to deflect this obvious argument against his theory by pointing out that weak readings also occur with overt descriptions instead of pronouns. In fact, I don't see how this saves his theory at all. If anything it makes things worse, for it undermines his theory of all definites, not just pronouns. True, it shows that NP-deletion isn't the culprit, but surely the book defends more than just NP-deletion? What about the unification of definites using a finegrained situation semantics?

Finally, a remark about the Bugblatter Beast of Traal, which assumes that if you can't see it, it can't see you, thus seemingly symmetricizing an asymmetric relation. E likens this to the inverse trick of asymmetricizing the bishops meeting each other in (2). However, it's not at all clear that the BBT really makes the not-seeing relation symmetric, for that would require that it also assumes for instance that if it can't see you, you can't see it. Further research is required to find out if that is indeed the case.

Conclusion: A clear and precise exposition of a new D-type approach to pronouns, definite descriptions and proper names. A must read for anyone interested in donkey anaphora, and recommended for anyone in the East Coast school of semantics.

REFERENCES

Burge, T. 1973. Reference and proper names. Journal of Philosophy. Geurts, B. 1997. Good news for the description theory of names. Journal of Semantics.

Heim, I. 1990. E-type pronouns and donkey anaphora. Linguistics and Philosophy.

Heim, I. and A. Kratzer. 1998. Semantics in Generative Grammar.

Jacobson, P. 1999. Towards a variable-free semantics. Linguistics and Philosophy.

Karttunen, L. 1974. Presupposition and linguistic context. Theoretical Linguistics.

Kratzer, A. 1989. An investigation of the lumps of thought. Linguistics and Philosophy.

Kurafuji, T. 1999. Japanese Pronouns in Dynamic Semantics. Rutgers Dissertation.

ABOUT THE REVIEWER

Emar Maier is to defend his Ph.D. thesis 'Belief in Context: towards a unified semantics of de re and de se attitude reports' at the Radboud University Nijmegen, The Netherlands, in November 2006. Besides the semantics and pragmatics of attitude reports and quotation his research interests include proper names, indexicals and direct reference/presupposition in general. He's currently teaching at the University of Amsterdam.