Plural pronominal anaphora in context

Author: Rick Nouwen
LOT Number: 84
ISBN: 90-76864-45-4
Pages: 203
Year: 2003
€31.00
Download this book as a free Open Access fulltext PDF

The starting point for this thesis was the observation that plural pronom­inal anaphoraisa very versatile phenomenon. It wouldbea gross over­simpli.cation to identify the relation between a plural pronoun and its antecedent asa simple coreference relation.In discourse, plural pronouns can have quanti.ed or quanti.cational, that is, not strictly referential an­tecedents. Gareth Evans’ famous example in (1), for instance, illustrated that some plural pronouns are neither bound by, nor corefer with their antecedents.
(1) Few senators admireKennedy; andthey are very junior.
The question was why the plural pronoun in (1) refers to the set of all senators that admire Kennedy. Apparently, the pronoun refers to a set constructedfrom sets thatplay a roleinthe .rst sentence, namelytothe intersection of the denotation of the restrictor (the senators) and the de­notation of the verb phrase (theKennedy-admirers). So, if (D(A))(B) is a quanti.cational sentence, where D corresponds to the meaning of a deter­miner, A to that of the restrictor and B to that of the VP, then A n B is a potential resolution of a subsequent (plural) pronoun. We called this set A n B the reference set.
Apart from reference to the reference set, however, it turned out that there are quite a few other types of plural pronominal anaphora. Apart from addressing examples like (1), this dissertation focused on two such other types of anaphora: (i) pronominal reference to the set-theoretical dif­ference between the restrictor denotation and the denotation of the VP,the so-called complement set A-B,as in example (2), and(ii) pronominal refer­ence to sets directly associated with the reference set, as in the example in (3). (Other forms of plural anaphora whichIdiscussed were non-maximal reference to the reference set and reference to the restrictor denotation, A.)
(2)
Few MPs attended the meeting. They went to the beachinstead.

(3)
Every student wrote a paper. Theyweren’t very good.

Additionally, there was a use of pronouns whichis strongly related to the form of anaphorain (3). The detailed relation between students and pa­pers, or, to put it slightly differently, the dependence of papers on students turned out to be accessible in discourse, too.
(4) Every student wrote a paper. Theysubmitted it to L&P.
The pronoun ‘it’ in (4) takes the inde.nite ‘a paper’ as its antecedent. This pronoun, however, is dependent on the subject pronoun ‘they’. The sec­ond sentencein(4) meansthat eachstudentwho wrotea paper submitted his or her paper to L&P. The students-paper pairs described by the .rst sentence are accessed in the second sentence.
Apparently then, there are several different types of plural anaphora. The most important goal of this thesis was to express this versatilityin a semantic theory. By far the most complete semantic theory of plural refer­encetodateisthat presentedinKampandReyle1993.Inthisbook,the discourse representation theoretic framework is extended to deal with plu­ral pronouns.Idiscussed two especially noteworthy aspectsof that theory:
(i) e-type anaphora, as in (1), is treated using a principle which is inde­pendent of other forms of discourse anaphora, and (ii) the accessibility of dependence information in discourse (as illustrated by (4)) is analysed us­ing a stipulated principle that allows the copying of representations under certain speci.c circumstances.
Plural anaphoraisthus accountedforbyDRTbyusingtwo principles which are assumed on top of DRT’s mechanism for ‘standard’ anaphora. Representations play a crucial role in these principles. Plural e-type ana­phora is analysed using an operation called abstraction. This operation allows representational material from a (representation of a) quanti.ca­tional sentence which is to be used to form a plural antecedent. The ac­cessibilityofdependence information, moreover, also boils downtoa con­strained license to re-use representational material.
In the literature, DRT’s emphasis on representations has lead to the question of whether the same empirical coverage could not be achieved without positing a level of representation. Moreover, due to the speci.c top-downarchitectureusedbyDRTto createthese representational struc­tures, the need for alternative accounts which were more faithful to the successful (bottom-up) Montague Grammar becameclear.With the inven­tion of dynamic predicate logic (Groenendijk and Stokhof 1991), an im­portant step was made to make such alternatives possible. This set off a programme of ‘dynamic semantics’ which had the non-representational and compositional interpretation of natural language expressions in dis­course as its goal. In dynamic semantics, the meaning of an expression is identi.ed as its context-change potential. ‘Context’, however, is not used hereas referringto some sortof representation. Contextisasimple model­theoretical object, namely a (set of) assignment function(s).
Given DRT’s successful account of plural reference,Ifound it interest­ing to see what a non-representational dynamic semantic theory which focuses on the same topic should look like. One speci.c aim of this disser­tationwas to discard the two DRT principles discussed aboveby moving to a dynamic semantic analysis.
The existence of examples like (2), however, formed a complicating fac­tor. BeforeIstarted to model plural pronominal reference,Itherefore .rst needed to investigate these examples.
The phenomenon illustratedby(2)was called complement anaphora. It has been (and still is) studied in detail in the psycholinguistic researchof Linda Moxey, Anthony Sanford and co-authors. Their experiments show thatsubjectstendtorefertothecomplementset followinganegative(that is, roughly, a downward monotone) quanti.cational expression. Within the semantic community, the reality of complement anaphora has been doubted several times. The likely root of this doubt is a generalisation made by Kamp and Reyle (1993). According to them, the operation of set­theoretical difference (like the complement set A - B)is not an admissible operation for antecedent-formation, in contrast to constructing the union of two sets.
There exists, however, an alternative interpretation of (2) which does justice to Kamp and Reyle’s generalisation, namely one in whichthe pro­noun does not refer to the complement set, but rather to the complete re­strictor set, A. The second sentence in (2) would then be paraphrasable by “generally speaking, the MPs went to the beachinstead.”
Moxey and Sanford argue against such an approach. Inchapter3of this thesis,Ifollowed their argumentation and showed that the idea that complement anaphora do not involve reference to the complement set is indeed very implausible. The existence of pronominal reference to the complement set, however, should not be seen as a clue that complement anaphora is a phenomenon comparable to cases of e-type anaphora as in (1). This became evident from two observations. First of all, whenever a pronoun can both take the reference set and the complement set as its antecedenttoyielda consistent interpretation,the resolutiontothe refer­ence set will ‘win’. In other words, complement anaphora is only possible once other forms of anaphora are excluded. Second, in contrast to refer­ence to the complement set, reference set reference is not constrained by the expression anteceding it. Complement anaphora can only occur once the antecedent guarantees that the complement set is non-empty. Such a condition does not applyto the reference set. Monotone decreasing expres­sions, whichdo not guarantee the non-emptiness of the reference set, are .ne antecedents for pronominal reference set reference.
I concluded from this that the reference set is made salient by aquan­ti.cational expression. An example like (1) is therefore an ordinary case of pronominal reference, for pronouns refer to individuals or groups of in­dividuals which are contextually highly salient. In order to refer to the complement set, its non-emptiness will .rst have to be inferred. This necessary inference step makes the complement set principally un.t for pronominal reference. Still, the experiments of Moxey and Sanford did show that apart from a preference for complement set reference follow­ing negative expressions, subjects also displayed a rhetorical preference for giving an explanation of the negativity of the antecedent. This effect makes the complement set a tempting antecedent. In conclusion, then, pronominal complement anaphora is a marked form of anaphora which owes its existence to a rhetorical preference of speakers.
This conclusion justi.edareductioninthe empirical coverageofthedy­namic semantics. A model of unmarked pronominal reference is a model thatdescribesthe introductionof salient setsin context. Pronominal refer­encetothe complementset, however,isinneedofan intervening inference process and, consequently, does not belongto such a model. Chapters 4,5 and 6 focused on the construction of a theory with this reduced empirical coverage.
Chapter 4 discussed three studies from the literature: the dynamic plu­ral predicate logic of van den Berg (1996b), the semantics for plural ana­phora based on parametrised sum individuals from Krifka (1996a) and the theoryof anaphoric information from Elworthy (1995). Thesestudieshave in common that they all address the dependent type of anaphora exempli­.ed in (4) and, moreover, use a structured notion of context to deal with examples involving dependence phenomena. The chapter showed that the proposals do not differ a lot and that the proposals of Krifka and van den Berg, in particular, are very similar with respect to important (formal) no­tions like dependence, independence and distributivity. These notions are responsible for the treatment of dependence phenomena and, as such, an elegant non-representational alternative to the DRT principles.
With respect to the data,I focused mainly on examples like (4) in chap­ter 4, and concluded that the theory of Elworthy is too strong. It generates a reading for (4) wherein each student submits a paper to L&P irrespec­tive of whether or not this is his or her own paper. The theories of van den Berg and Krifka, however, turned out to be too weak, since they can­not explain how, in the scope of distributivity, pronouns can refer to both sub-individualsinagroupandtotheoriginaltotalgroupitself.The exam­plein(5),for instance,isproblematic.Thesecond sentenceis ambiguous between a reading in which each student submits the two papers he or she wrote to L&P and (an odd) one wherein eachstudent submits all the written papers to L&P.
(5) Every student wrote exactly two papers. They eachsubmitted them to L&P.
Theexamplein(5)playedacrucialroleinchapter5.Thequestionwashow anaphora should be represented given that in examples like (5), the same syntactic antecedent (‘two papers’) antecedes two different resolutions of the pronoun, namelya dependent and an independent one. In theories like those of van den Berg and Krifka, there is one variable whichis associated with both readings. But only one of those can be realised at one time. I argued that the choice of the label whichis associated with an individual introduced into the context should be contextualised.
This conclusion is not new. It is related to the so-called problem of destructive assignment, which was noticed as being an important short­coming of dynamic predicate logic. The cause of the problem is the speci.c de.nition of existential quanti.cation as random assignment. That is, ex­istential quanti.cation over some variable x rewrites the value assigned to x previously. Consequently, languages like dynamic predicate logic allow values to be overwitten and information to be lost.
Asa solution forthe destructive assignment problem,I turned to the framework of incremental dynamics (van Eijck2001). In incremental dy­namics, the labellingof valuesis contextualised. Thisis realisedbyreplac­ing assignment functions (sets of variable-individual pairs) with ‘stacks’, sets of position-individual pairs. Since the position of an individual in a stack can change when this stackis combined with another stack, the in­dex which goes with an introduced individual is contextualised. This is because the position of a value is always relative to the stack(i.e. the con­text) it appears in.
Incremental dynamics is a variation on dynamic predicate logic. In chapter5,Ipresenteda variation on thedynamic plural predicate logicof van den Berg along the lines of incremental dynamics. This enabled us to devise a distributivity operator whichgives access to dependence informa­tion without overwriting ‘old’ antecedents. Using an underspeci.ed repre­sentationof theinterpretationof examples like (5), we can then show that a pronoun can have two plausible resolutions, completely in accordance with the intuitions given for these sentences.
In chapter 6, I returned to DRT and the principles it has to stipulate in order to show that our non-representational approach maydo without suchadditional principles. Ishowed that the semantics for distributivity which was developed in chapter 5 can be generalised to a semantics for quanti.cational determiners, which at the same time captures the intro­duction of the (maximal) reference set in context and the potential intro­duction of dependence information. Moreover, this approach was shown to give a straightforward account of the dynamics of monotone decreasing quanti.ers. Most importantly, however,all anaphoric effects were lexically governedand, consequently,no additional principles needed to be assumed to account for the range ofresolutions of plural pronouns.

 

The starting point for this thesis was the observation that plural pronom­inal anaphoraisa very versatile phenomenon. It wouldbea gross over­simpli.cation to identify the relation between a plural pronoun and its antecedent asa simple coreference relation.In discourse, plural pronouns can have quanti.ed or quanti.cational, that is, not strictly referential an­tecedents. Gareth Evans’ famous example in (1), for instance, illustrated that some plural pronouns are neither bound by, nor corefer with their antecedents.
(1) Few senators admireKennedy; andthey are very junior.
The question was why the plural pronoun in (1) refers to the set of all senators that admire Kennedy. Apparently, the pronoun refers to a set constructedfrom sets thatplay a roleinthe .rst sentence, namelytothe intersection of the denotation of the restrictor (the senators) and the de­notation of the verb phrase (theKennedy-admirers). So, if (D(A))(B) is a quanti.cational sentence, where D corresponds to the meaning of a deter­miner, A to that of the restrictor and B to that of the VP, then A n B is a potential resolution of a subsequent (plural) pronoun. We called this set A n B the reference set.
Apart from reference to the reference set, however, it turned out that there are quite a few other types of plural pronominal anaphora. Apart from addressing examples like (1), this dissertation focused on two such other types of anaphora: (i) pronominal reference to the set-theoretical dif­ference between the restrictor denotation and the denotation of the VP,the so-called complement set A-B,as in example (2), and(ii) pronominal refer­ence to sets directly associated with the reference set, as in the example in (3). (Other forms of plural anaphora whichIdiscussed were non-maximal reference to the reference set and reference to the restrictor denotation, A.)
(2)
Few MPs attended the meeting. They went to the beachinstead.

(3)
Every student wrote a paper. Theyweren’t very good.

Additionally, there was a use of pronouns whichis strongly related to the form of anaphorain (3). The detailed relation between students and pa­pers, or, to put it slightly differently, the dependence of papers on students turned out to be accessible in discourse, too.
(4) Every student wrote a paper. Theysubmitted it to L&P.
The pronoun ‘it’ in (4) takes the inde.nite ‘a paper’ as its antecedent. This pronoun, however, is dependent on the subject pronoun ‘they’. The sec­ond sentencein(4) meansthat eachstudentwho wrotea paper submitted his or her paper to L&P. The students-paper pairs described by the .rst sentence are accessed in the second sentence.
Apparently then, there are several different types of plural anaphora. The most important goal of this thesis was to express this versatilityin a semantic theory. By far the most complete semantic theory of plural refer­encetodateisthat presentedinKampandReyle1993.Inthisbook,the discourse representation theoretic framework is extended to deal with plu­ral pronouns.Idiscussed two especially noteworthy aspectsof that theory:
(i) e-type anaphora, as in (1), is treated using a principle which is inde­pendent of other forms of discourse anaphora, and (ii) the accessibility of dependence information in discourse (as illustrated by (4)) is analysed us­ing a stipulated principle that allows the copying of representations under certain speci.c circumstances.
Plural anaphoraisthus accountedforbyDRTbyusingtwo principles which are assumed on top of DRT’s mechanism for ‘standard’ anaphora. Representations play a crucial role in these principles. Plural e-type ana­phora is analysed using an operation called abstraction. This operation allows representational material from a (representation of a) quanti.ca­tional sentence which is to be used to form a plural antecedent. The ac­cessibilityofdependence information, moreover, also boils downtoa con­strained license to re-use representational material.
In the literature, DRT’s emphasis on representations has lead to the question of whether the same empirical coverage could not be achieved without positing a level of representation. Moreover, due to the speci.c top-downarchitectureusedbyDRTto createthese representational struc­tures, the need for alternative accounts which were more faithful to the successful (bottom-up) Montague Grammar becameclear.With the inven­tion of dynamic predicate logic (Groenendijk and Stokhof 1991), an im­portant step was made to make such alternatives possible. This set off a programme of ‘dynamic semantics’ which had the non-representational and compositional interpretation of natural language expressions in dis­course as its goal. In dynamic semantics, the meaning of an expression is identi.ed as its context-change potential. ‘Context’, however, is not used hereas referringto some sortof representation. Contextisasimple model­theoretical object, namely a (set of) assignment function(s).
Given DRT’s successful account of plural reference,Ifound it interest­ing to see what a non-representational dynamic semantic theory which focuses on the same topic should look like. One speci.c aim of this disser­tationwas to discard the two DRT principles discussed aboveby moving to a dynamic semantic analysis.
The existence of examples like (2), however, formed a complicating fac­tor. BeforeIstarted to model plural pronominal reference,Itherefore .rst needed to investigate these examples.
The phenomenon illustratedby(2)was called complement anaphora. It has been (and still is) studied in detail in the psycholinguistic researchof Linda Moxey, Anthony Sanford and co-authors. Their experiments show thatsubjectstendtorefertothecomplementset followinganegative(that is, roughly, a downward monotone) quanti.cational expression. Within the semantic community, the reality of complement anaphora has been doubted several times. The likely root of this doubt is a generalisation made by Kamp and Reyle (1993). According to them, the operation of set­theoretical difference (like the complement set A - B)is not an admissible operation for antecedent-formation, in contrast to constructing the union of two sets.
There exists, however, an alternative interpretation of (2) which does justice to Kamp and Reyle’s generalisation, namely one in whichthe pro­noun does not refer to the complement set, but rather to the complete re­strictor set, A. The second sentence in (2) would then be paraphrasable by “generally speaking, the MPs went to the beachinstead.”
Moxey and Sanford argue against such an approach. Inchapter3of this thesis,Ifollowed their argumentation and showed that the idea that complement anaphora do not involve reference to the complement set is indeed very implausible. The existence of pronominal reference to the complement set, however, should not be seen as a clue that complement anaphora is a phenomenon comparable to cases of e-type anaphora as in (1). This became evident from two observations. First of all, whenever a pronoun can both take the reference set and the complement set as its antecedenttoyielda consistent interpretation,the resolutiontothe refer­ence set will ‘win’. In other words, complement anaphora is only possible once other forms of anaphora are excluded. Second, in contrast to refer­ence to the complement set, reference set reference is not constrained by the expression anteceding it. Complement anaphora can only occur once the antecedent guarantees that the complement set is non-empty. Such a condition does not applyto the reference set. Monotone decreasing expres­sions, whichdo not guarantee the non-emptiness of the reference set, are .ne antecedents for pronominal reference set reference.
I concluded from this that the reference set is made salient by aquan­ti.cational expression. An example like (1) is therefore an ordinary case of pronominal reference, for pronouns refer to individuals or groups of in­dividuals which are contextually highly salient. In order to refer to the complement set, its non-emptiness will .rst have to be inferred. This necessary inference step makes the complement set principally un.t for pronominal reference. Still, the experiments of Moxey and Sanford did show that apart from a preference for complement set reference follow­ing negative expressions, subjects also displayed a rhetorical preference for giving an explanation of the negativity of the antecedent. This effect makes the complement set a tempting antecedent. In conclusion, then, pronominal complement anaphora is a marked form of anaphora which owes its existence to a rhetorical preference of speakers.
This conclusion justi.edareductioninthe empirical coverageofthedy­namic semantics. A model of unmarked pronominal reference is a model thatdescribesthe introductionof salient setsin context. Pronominal refer­encetothe complementset, however,isinneedofan intervening inference process and, consequently, does not belongto such a model. Chapters 4,5 and 6 focused on the construction of a theory with this reduced empirical coverage.
Chapter 4 discussed three studies from the literature: the dynamic plu­ral predicate logic of van den Berg (1996b), the semantics for plural ana­phora based on parametrised sum individuals from Krifka (1996a) and the theoryof anaphoric information from Elworthy (1995). Thesestudieshave in common that they all address the dependent type of anaphora exempli­.ed in (4) and, moreover, use a structured notion of context to deal with examples involving dependence phenomena. The chapter showed that the proposals do not differ a lot and that the proposals of Krifka and van den Berg, in particular, are very similar with respect to important (formal) no­tions like dependence, independence and distributivity. These notions are responsible for the treatment of dependence phenomena and, as such, an elegant non-representational alternative to the DRT principles.
With respect to the data,I focused mainly on examples like (4) in chap­ter 4, and concluded that the theory of Elworthy is too strong. It generates a reading for (4) wherein each student submits a paper to L&P irrespec­tive of whether or not this is his or her own paper. The theories of van den Berg and Krifka, however, turned out to be too weak, since they can­not explain how, in the scope of distributivity, pronouns can refer to both sub-individualsinagroupandtotheoriginaltotalgroupitself.The exam­plein(5),for instance,isproblematic.Thesecond sentenceis ambiguous between a reading in which each student submits the two papers he or she wrote to L&P and (an odd) one wherein eachstudent submits all the written papers to L&P.
(5) Every student wrote exactly two papers. They eachsubmitted them to L&P.
Theexamplein(5)playedacrucialroleinchapter5.Thequestionwashow anaphora should be represented given that in examples like (5), the same syntactic antecedent (‘two papers’) antecedes two different resolutions of the pronoun, namelya dependent and an independent one. In theories like those of van den Berg and Krifka, there is one variable whichis associated with both readings. But only one of those can be realised at one time. I argued that the choice of the label whichis associated with an individual introduced into the context should be contextualised.
This conclusion is not new. It is related to the so-called problem of destructive assignment, which was noticed as being an important short­coming of dynamic predicate logic. The cause of the problem is the speci.c de.nition of existential quanti.cation as random assignment. That is, ex­istential quanti.cation over some variable x rewrites the value assigned to x previously. Consequently, languages like dynamic predicate logic allow values to be overwitten and information to be lost.
Asa solution forthe destructive assignment problem,I turned to the framework of incremental dynamics (van Eijck2001). In incremental dy­namics, the labellingof valuesis contextualised. Thisis realisedbyreplac­ing assignment functions (sets of variable-individual pairs) with ‘stacks’, sets of position-individual pairs. Since the position of an individual in a stack can change when this stackis combined with another stack, the in­dex which goes with an introduced individual is contextualised. This is because the position of a value is always relative to the stack(i.e. the con­text) it appears in.
Incremental dynamics is a variation on dynamic predicate logic. In chapter5,Ipresenteda variation on thedynamic plural predicate logicof van den Berg along the lines of incremental dynamics. This enabled us to devise a distributivity operator whichgives access to dependence informa­tion without overwriting ‘old’ antecedents. Using an underspeci.ed repre­sentationof theinterpretationof examples like (5), we can then show that a pronoun can have two plausible resolutions, completely in accordance with the intuitions given for these sentences.
In chapter 6, I returned to DRT and the principles it has to stipulate in order to show that our non-representational approach maydo without suchadditional principles. Ishowed that the semantics for distributivity which was developed in chapter 5 can be generalised to a semantics for quanti.cational determiners, which at the same time captures the intro­duction of the (maximal) reference set in context and the potential intro­duction of dependence information. Moreover, this approach was shown to give a straightforward account of the dynamics of monotone decreasing quanti.ers. Most importantly, however,all anaphoric effects were lexically governedand, consequently,no additional principles needed to be assumed to account for the range ofresolutions of plural pronouns.

 

Categories