Keenan (1992) proves there is no generalized quantifier that expresses the meaning the same two books. And indeed, previous analyses of adjectives like same have been heavily pragmatic (Dowty 1985, Beck 2000) or else deliberately non-compositional, either building syntactically discontinuous higher-order quantifiers (Keenan 1992, van Eijck 2003), or relying on side calculations (Stump 1982, Moltmann 1992). Building on insights of Carlson (1987), I propose the first strictly compositional semantic account of same. New data, including especially NP-internal uses such as two men with the same name, suggests that same is a quantificational element taking scope over nominals. Given LIFT as a basic type-shifting operator, I show that this proposal follows naturally from the fact that same is an adjective. Independently-motivated assumptions extend the analysis to standard examples such as Anna and Bill read the same book via a mechanism I call PARASITIC SCOPE, in which the scope of same depends on the scope of some other scope-taking element in the sentence. Although I initially express the main analysis within a movement-based framework with Quantifier Raising in the style of Heim and Kratzer (1998), I go on to implement the analysis within a continuation-based, variable-free, directly compositional combinatory-categorial grammar, without Quantifier Movement or any use of Logical Form distinct from surface structure. The empirical payoff for dealing in continuations is that a simple generalization accounts for the first time for cases in which same distributes over objects other than NP denotations, as in the relevant interpretation of John hit and killed the same man.
This paper starts with the assumption that the expression saw everyone, as in John saw everyone, is a constituent. It can be coordinated, clefted, focussed, questioned, pronominalized, you name it. In general, if you have a constituent containing an NP, and you replace that NP with a quantificational NP like everyone, then the result is still a full-fledged constituent. But everyone can take scope outside of a verb phrase. As a result, the semantics of everyone must somehow gain access to material that can properly contain the constituent saw everyone. Despite decades of research on quantification, I will suggest that no existing analysis (including my own published accounts to date) provides an adequate characterization of both the local and the long-distance aspects of the syntax and the semantics of quantificational expressions. The main goal of this paper, then, is to propose an explicit new account that tries to do better.
I
provide evidence in support of Poesio's (1994) claim that
there is a productive, systematic class of definite descriptions whose
use does not appear to require either familiarity or uniqueness:
(1) I hope the cafe is located on the corner of a busy intersection.
Clearly, (1) can be used in a situation in which neither the
speaker nor the listener has any previous acquaintance with a
specific intersection or corner, nor is there an implication that the
intersection in question has only one corner.
Using the technique of function composition (independently motivated in various combinatory categorial grammars), the relational head nominal in (1) can combine with the definite determiner before combining with its genitive argument. If the uniqueness presupposition associated with the definite determiner applies to the first element it combines with semantically, rather than with the first element it would normally combine with syntactically, we correctly predict that weak interpretations emerge only in the presence of a relational nominal.
[This is an attempt to explain the applications of continuations in natural language semantics to computer scientists who already know what continuations are. Applications discussed include quantification, focus, coordination, and misplaced modifiers (e.g., an occasional sailor). Linguists looking for an introduction to linguistic applications of continuations are better off trying my NALS paper.]
Computer scientists, logicians and functional programmers have studied continuations in laboratory settings for years. As a result of that work, continuations are now accepted as an indispensable tool for reasoning about control, order of evaluation, classical versus intuitionistic proof, and more. But all of the applications just mentioned concern artificial languages; what about natural languages, the languages spoken by humans in their daily life? Do natural languages get by without any of the marvelous control operators provided by continuations, or can we find continuations in the wild? This paper argues yes: that an adequate and complete analysis of natural language must recognize and rely on continuations. In support of this claim, I identify four independent linguistic phenomena for which a simple CPS-based description provides an insightful analysis.
We present a general theory of scope and binding in which both crossover and superiority violations are ruled out by one key assumption: that natural language expressions are uniformly evaluated (processed) from left to right. Our theory is an extension of Shan's (2002 SALT) account of multiple-wh questions, combining continuations (see Barker's 2002 NALS) and dynamic type-shifting. Like other continuation-based analyses, but unlike most other treatments of crossover or superiority, our new analysis does not postulate a level of Logical Form or other representation distinct from surface syntax. One advantage of using continuations is that order-of-evaluation for programming languages is standardly modeled using continuations; that work provides us with a natural and independently-motivated characterization of what it means to evaluate expressions from left to right. We provide a combinatory categorial grammar that models the syntax and the semantics of quantifier scope and wh-question formation. It allows quantificational binding but not crossover, in-situ wh but not superiority violations.
Standard wisdom (Stalnaker 1979, van der Sandt 1992) holds that assertions
are felicitous only if they add new information to the common ground.
But if an utterance of (1) is true, then both speaker and addressee must
have
already possessed all the knowledge they need to conclude that Mary is a
doctor before the sentence was uttered.
(1)
It is clear that Mary is a doctor.
We suggest that an utterance of (1) adds new information about the state
of the discourse itself, rather than the facts under discussion (i.e.
proving whether Mary is a doctor). More technically, asserting (1)
constrains the vague standard for an absolute use of the gradable
adjective clear; the standard must be reset in such a way that only those
worlds in which Mary is a doctor are considered viable alternatives. We
extend the Stalnaker/Heim model of context update to explicitly deal with
the vague standards for gradable adjectives (Kyburg and Morreau 2000,
Barker 2002 [the dynamics of vagueness paper]).
Our conclusion is that there are some expression types
whose semantics guarantee that they can only be used to negotiate vague
standards, and that have no normal descriptive content at all.
In a series of papers beginning in the late 1980's, Jacobson develops a novel and provocative theory of binding that does entirely without movement or variables. She provides many theoretical and empirical arguments in favor of a variable-free approach. However, she does not discuss quantifier scope in any detail, and therefore does not provide an account of binding out of DP. This remark evaluates some of the strengths and the weaknesses of the variable-free program in the course of extending the fragment in Jacobson (1999) to handle a more complete range of binding constructions, including quantificational possessors and so-called inverse linking. In particular, I will show in detail how to incorporate Hendriks' Flexible Types approach to quantifier scope with excellent results, making good on a major promisory note in Jacobson's variable-free program. The characteristic property of Jacobson's variable-free framework is that all binding relationships must be strictly local: since there is no LF movement and (obviously) no variables, the dependence of one value on another can only be established through a chain of local function/argument relationships. It follows that all constraints on binding, must also be strictly local, in the sense that they can be sensitive only to syntactic information associated with immediate subconstituents. I propose a suitably local constraint that accounts for weak crossover.
This paper proposes that the meanings of some natural language expressions should be thought of as functions on their own continuations. Continuations are a well-established technique in the theory of programming language semantics; in brief, a continuation is the entire default future of a computation. I show how a continuation-based grammar can unify several aspects of natural language quantification in a new way: merely stating the truth conditions for quantificational expressions in terms of continuations automatically accounts for scope displacement and scope ambiguity. To prove this claim, I exhibit a simple finite context-free grammar with a strictly compositional semantics in which quantificational NPs are interpreted in-situ but take semantic scope over larger constituents. There is no Quantifier Raising (nor any use of a level of Logical Form distinct from overt syntax), no Cooper Storage (or other similar mechanisms used in many recent HPSG, Categorial, or Type-logical treatments), and no need for type-shifting (as in Hendriks' Flexible Types account). Continuations also provide a natural account of generalized coordination that does not require either type-shifting or type-polymorphism. Compositionality issues are discussed in some detail.
Many people have studied how vague predicates depend on context for their interpretation, but few have studied in detail how a use of a vague predicate affects the context against which other expressions get evaluated. In recent years, however, various dynamic theories of context update have made considerable progress in describing and explaining complex phenomena such as presupposition and anaphora. In this paper I will argue that taking an explicitly dynamic perspective on vagueness can lead to new insights into the nature of vagueness in general and the semantics of gradable adjectives in particular.
In order to develop an explicit theory of what a use of a vague expression does, I propose that expressions containing gradable adjectives make a specific kind of contribution to contextual information. Accepting an utterance of, say, Feynman is tall can simultaneously affect mutual assumptions concerning of Feynman's height and of what counts as tall, i.e., it can operate both at a descriptive and at a metalinguistic level. An extension of the basic analysis predicts that update with a use of a measure phrase or a comparative is guaranteed to result in no sharpening of the relevant vague standard. In addition, the dynamic approach naturally leads to a reasonable account of higher-order vagueness in which modifiers such as clearly come out as vagueness quantifiers. A second extension of the basic analysis leads to the first detailed semantics for infinitival-taking adjectives. The somewhat surprising conclusion is that when certain adjectives such as stupid occur with an infinitival complement, the resulting construction has no update effect apart from its presuppositions and its effect on vague standards. Empirical support for this claim comes from the otherwise unexplained fact that infinitival uses of such adjectives cannot be embedded beneath a control predicate (*Feynman wanted to be [stupid to dance like that]). In each case, the update effect of all of these expression types follows directly from stating their truth conditions in a dynamic framework.
2000 Theoretical Linguistics 26:211--227:
Standard wisdom holds that in order for a use of a definite description to be felicitous, its referent must be `familiar', either because it was mentioned in previous discourse or because it is otherwise salient in the non-linguistic context. I show that a grammatically-defined class of demonstrably definite NPs containing definite possessors (e.g., that man's daughter) are routinely able to describe novel (unfamiliar) entities. I propose to account for this systematic class of exceptions to the standard familiarity requirement by providing a refined notion of familiarity, one that determines when a use of a possessed NP as a whole will count as familiar based (in part) on whether its possessor phrase counts as familiar.
This paper seeks to explain a previously unnoticed semantic phenomenon illustrated by the contrast between (1) I am the oldest of my enemies, which presupposes that the speaker is one of his or her own enemies, versus (2) I am the oldest of my siblings, which for most speakers does not presuppose that the speaker is his or her own sibling. I argue that the behavior of (2) is the result of temporarily enlarging the extension of the predicate (sibling) in order to meet an otherwise unsatisfiable presupposition. I propose that this temporary accommodation occurs only with lexical and complex predicates whose denotations are of a certain mathematical class that I call quasi-equivalence relations. I also briefly discuss a number of closely related construction types, and draw out the functional motivation for this unusual type of grammaticized accommodation.
Judging from the way people talk, the number of entities in a given situation that get recognized as distinct individuals seems to depend on the specific communicative goals of a particular discourse. Admitting that semantic interpretation needs to provide a limited degree of ontological variability can resolve some empirical difficulties associated with Krifka's (1990) treatment of examples like 4000 ships passed through the lock last year.
This paper offers an explanation for a little-known but striking phenomenon first discussed by Jackendoff (1968b) that I will call ANTI-UNIQUENESS: partitives are incompatible with the definite determiner (*I met the one of John's friends), unless the partitive first receives additional modification (I met the [[one of John's friends] that he traveled with from Mexico]). I argue that an independently needed refinement of the semantic analyses of the partitive of Ladusaw (1982) and Hoeksema (1984) automatically predicts these anti-uniqueness facts. More specifically, I propose that partitivity is always proper partitivity. This will guarantee that any property denoted by a partitive will have at least two entities in its extension, and cannot uniquely identify an individual; thus partitives are anti-unique. In addition, this paper makes a new case for analyzing double genitives as partitives. A number of syntactic and semantic arguments will show that despite appearances, so-called double genitives (a friend of John's) have less in common with a superficially quite similar type of simple genitive (a friend of John) than with standard partitives (one of John's friends). If double genitives are indeed a type of partitive, this explains why they also exhibit anti-uniqueness effects: *I met [the friend of John's] is bad but I met the [[friend of John's] that he traveled with from Mexico]} is perfectly fine.
This paper offers a detailed analysis of the English suffix -ee (employee, escapee, refugee, etc.) in 1500 naturally-occurring tokens of some 500 word types. The data suggest that formation of nouns in -ee is moderately but genuinely productive, and that analyses based on the syntactic argument structure of the stem verb are unsatisfactory. Instead, formation of -ee nouns systematically adheres to three essentially semantic constraints: first, the referent of an -ee noun must be sentient; second, the denotation of an -ee noun must be episodically linked (as defined below) to the denotation of its stem; and third, a use of an - ee noun entails a relative lack of volitional control on the part of its referent. I argue that these semantic constraints taken together amount to a special-purpose thematic role that actively constrains productive use of derivational morphology.
Most studies of the so-called Proportion Problem seek to understand how lexical and structural properties of sentences containing adverbial quantifiers give rise to various proportional readings. This paper explores a related but distinct problem: given a use of a particular sentence in context, why do only some of the expected proportional readings seem to be available? That is, why do some sentences allow an asymmetric reading when other structurally similar sentences seem to require a symmetric reading? Potential factors suggested in the literature include the distribution of donkey pronouns, certain uniqueness implications, and focus structures. I argue here that a use of an adverbial quantifier presupposes HOMOGENEITY: all individual situations that get lumped into a single case for the purposes of evaluating the quantification must agree on whether they satisfy the nuclear scope. For instance, in order for a token of Usually, if a farmer owns a donkey, he beats it to be felicitous when construed under a farmer-dominant asymmetric reading, the context must be consistent with the proposition that each farmer either beats all or none of his donkeys. Thus proportional sentences are indeed systematically ambiguous, but only some readings will be felicitous in a given context.
This book is on sale at Amazon.
What do possessives mean? More specifically, how does the meaning of a possessive depend upon the meaning of its parts, and how does that meaning contribute to the interpretation of the expression in which the possessive is embedded? This dissertation develops a formal model-theoretic account of the semantics of possessives in English on which possessives are descriptions on a par with definite and indefinite descriptions.
Agreement facts show that possessives in English are not syntactically ambiguous; however, they are systematically ambiguous with respect to their descriptive content. If the possessee phrase denotes a two-place relation, then the possessive expresses that relation directly, giving rise to LEXICAL possession (John's child); but if the possessee phrase is not relational, the possessive expresses a contextually determined relation, giving rise to EXTRINSIC possession (John's stick).
Evidence from donkey anaphora and negative polarity licensing shows that a quantificational possessor phrase must raise in logical form to take scope over its host determiner phrase.
The central problem addressed in this dissertation is what I call the perspective paradox. Consider the sentence (1) ``Most students' dogs bark constantly''. There is a strong intuition that the quantification in (1) ranges either over students (their dogs tend to bark a lot) or over dogs (the ones owned by students bark a lot), depending on your perspective on the situation. How can we account for this intuition, and why is there no detectable effect on truth conditions?
I propose an unselective binding analysis on which the quantifier denoted by most binds the possessor variable and the possessee variable independently. In general, different proportional readings correspond to different ways of grouping instances that are relevant for a quantification into cases. The two perspectives on the quantification in (1) correspond to distinguishing cases solely on the basis of the student involved versus distinguishing cases on the basis of student/dog pairs. But since the use of a possessive presupposes that there is a unique set of dogs for each student, the two partitions of instances into cases are equivalent, leading to identical truth conditions.
Let us suppose that thematic roles, or something very much like them, are needed to describe lexical and semantic patterns in the behavior of verbal predicates. But what about nouns? Is there evidence independent of verbal constructions motivating a system of nominal thematic relations? Beginning with the analysis of relational nouns in Barker (1991), we go on to suggest that the general problem of argument selection does in fact motivate a set of quintessentially nominal thematic proto-roles which we call Proto-Part and Proto-Whole. These nominal proto-roles are parallel to but distinct from the verbal proto-roles of Proto-Agent and Proto-Patient proposed by Dowty (1991) to account for argument selection in verbal predicates. The resulting theory of nominal argument selection is properly semantic, that is, it distinguishes between argument positions purely on the basis of the semantic entailments of the predicates involved. Furthermore, it is non-dynamic, that is, it describes static patterns of lexicalization without hypothesizing any productive grammatical process that maps semantic arguments to syntactic arguments during the derivation of a sentence.
[Some illustrations are missing from this version; see the journal.]
What do terms such as the committee, the league, and the group of women denote? Pre-theoretically, group terms have a dual personality. On the one hand, the committee corresponds to an entity as ideosyncratic in its properties as any other object; for instance, two otherwise identical committees can vary with respect to the purpose for which they were formed. Call this aspect the group-as-individual. On the other hand, the identity of a group is at least partially determined by the properties of its members; for instance, a committee will be a committee of women just in case each of its members is a woman. Call this aspect the group-as-set. Elaborating on suggestions in Link (1984) and Lasersohn (1988), I propose that group terms in English denote atomic individuals, that is, entities lacking internal structure. In particular, it is not possible to determine the membership of a group by examining the denotation of a group term. The proposed account correctly predicts that group terms systematically behave differently semantically (as well as syntactically) from plurals such as the men and conjunctions such as John and Bill. Thus the atomic analysis advocated here stands in sharp contrast to previous proposals, including Bennet (1974), Link (1984), and Landman (1989), in which group terms are considered of a piece semantically with plurals and conjunctions. Additional arguments come from the use of names of groups as rigid designators, from the parallel between group nouns and measure nouns, and from the distribution of group terms across two dialects of English.
[Some illustrations are missing from this version; see the journal.]
Since Langacker (1969) introduced the `command' relation into the literature of syntax, such relations, which we can refer to generally as COMMAND RELATIONS, have increased greatly in importance. Yet the literature contains no systematic formal treatment of the general topic of command relations (notwithstanding a few articles focussed on the definitions of particular command relations, e.g., Aoun and Sportiche (1982), Saito (1984), Richardson and Chametzky (1985)). This paper sets out an explicit general definition that aims to capture the essential properties common to all command relations, and to allow specific relations to be distinguished in terms of different settings of a single parameter. We prove a number of theorems, some about command relations as a class, and some about particular command relations.
Extrametricality is a formal tool for prosodic analysis. It amounts to temporary invisibility for part of a form for the purposes of rule application: first you cover up the extrametrical material, then you apply the rule, then you uncover the extrametrical material again and continue. An appeal to extrametricality can significantly increase the elegance of a particular grammar. More importantly, extrametricality currently plays a crucial role in various areas of prosodic research; for instance, the typology of stress systems (e.g., Hayes (1981), Prince (1983)), as well as the prosodic morphology of McCarthy and Prince (1986, 1988) both depend on extrametricality. Its value to phonological theory in the long run, however, will depend on the degree to which it can be constrained by universal principles. Turkish is interesting in this regard since Poser (1984) suggests that Turkish word stress motivates a weakening in the theory of extrametricality. My main goal will be to show that Turkish supports a highly restrictive version of extrametricality. More specifically, I seek to maintain the following general constraints: extrametrical material must consist of some independently motivated prosodic or morphological constituent; extrametrical material must be peripheral; and extrametricality never persists for more than one cycle.