34
ANTECEDENT CONTAINMENT: A DEPENDENCY GRAMMAR SOLUTION IN TERMS OF CATENAE Timothy Osborne & Thomas Groß Abstract. The paper presents a new solution to an old problem, the problem of antecedent containment and the infinite regress that it gives rise to. The solution is couched in a dependency grammar (DG) framework and is possible by acknowl- edging DG catenae. The catena is a novel unit of syntactic analysis. Pronouns that appear to be contained within their antecedents actually take non-constituent catenae as their antecedents. These catenae quite naturally exclude the pronouns, so that antecedent containment never occurs. This account acknowledges surface configurations only. No appeals are made to derivational mechanisms or abstract representations. 1. The problem and its solution Data like the following have been examined in the literature on antecedent containment (e.g. Bouton 1970, Brody 1981, Williams 1982, Higginbotham 1983, Haı¨ k 1982–3, 1987, Hoeksema and Napoli 1990, Kennedy 2008). Examples (1–2) are taken from Haı¨ k (1987:505): (1) *[A picture of its i frame] i is on the table. (2) [The man next to his i dog] i burped. Both of these examples illustrate apparent antecedent containment, the pronoun being contained within its antecedent NP. 1 Yet only the indicated reading in (1) is unavailable, so apparently only (1) involves true antecedent containment. Antecedent containment is of course prohibited due to the infinite regress that it generates. What is the relevant difference between (1) and (2) that might explain this contrast? Why is apparent antecedent containment disallowed in (1) but allowed in (2)? This paper answers these questions. The following data further illustrate the problem: (3) a. *[A friend of his i brother] i stayed. b. *[A friend of Sam i Õs brother] i stayed. c. *[His i brotherÕs friend] i stayed. d. *[Sam i Õs brotherÕs friend] i stayed. 1 We assume NP (not DP) throughout this paper. Nothing hinges on this assumption. Studia Linguistica 66(2) 2012, pp. 94–127. Ó The authors 2012. Ó The Editorial Board of Studia Linguistica 2012. Published by Blackwell Publishing Ltd., 9600 Garsington Road, Oxford OX4 2DQ, UK, and 350 Main Street, Malden, MA 02148, USA

Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

Embed Size (px)

Citation preview

Page 1: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

ANTECEDENT CONTAINMENT:A DEPENDENCY GRAMMAR SOLUTION

IN TERMS OF CATENAE

Timothy Osborne & Thomas Groß

Abstract. The paper presents a new solution to an old problem, the problem ofantecedent containment and the infinite regress that it gives rise to. The solution iscouched in a dependency grammar (DG) framework and is possible by acknowl-edging DG catenae. The catena is a novel unit of syntactic analysis. Pronouns thatappear to be contained within their antecedents actually take non-constituentcatenae as their antecedents. These catenae quite naturally exclude the pronouns,so that antecedent containment never occurs. This account acknowledges surfaceconfigurations only. No appeals are made to derivational mechanisms or abstractrepresentations.

1. The problem and its solution

Data like the following have been examined in the literature onantecedent containment (e.g. Bouton 1970, Brody 1981, Williams 1982,Higginbotham 1983, Haık 1982–3, 1987, Hoeksema and Napoli 1990,Kennedy 2008). Examples (1–2) are taken from Haık (1987:505):

(1) *[A picture of itsi frame]i is on the table.

(2) [The man next to hisi dog]i burped.

Both of these examples illustrate apparent antecedent containment, thepronoun being contained within its antecedent NP.1 Yet only theindicated reading in (1) is unavailable, so apparently only (1) involvestrue antecedent containment. Antecedent containment is of courseprohibited due to the infinite regress that it generates. What is therelevant difference between (1) and (2) that might explain this contrast?Why is apparent antecedent containment disallowed in (1) but allowed in(2)? This paper answers these questions.The following data further illustrate the problem:

(3) a. *[A friend of hisi brother]i stayed.b. *[A friend of Sami�s brother]i stayed.c. *[Hisi brother�s friend]i stayed.d. *[Sami�s brother�s friend]i stayed.

1 We assume NP (not DP) throughout this paper. Nothing hinges on this assumption.

Studia Linguistica 66(2) 2012, pp. 94–127. � The authors 2012.� The Editorial Board of Studia Linguistica 2012. Published by Blackwell Publishing Ltd.,9600 Garsington Road, Oxford OX4 2DQ, UK, and 350 Main Street, Malden, MA 02148, USA

Page 2: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

(4) a. *[A supporter of heri mother]i arrived.b. *[A supporter of Susani�s mother]i arrived.c. *[Heri mother�s supporter]i arrived.d. *[Susani�s mother�s supporter]i arrived.

The indicated readings are all unavailable due to antecedent containment.The pronoun in the (a)- and (c)-sentences is contained within itsantecedent NP. Similarly, the names in the (b)- and (d)-sentences are alsocontained within the coindexed NP. The NPs in (3–4) must be comparedwith those in (5–6):

(5) a. [The actress studying heri lines]i works a lot.b. *[The actress studying Susani�s lines]i works a lot.

(6) a. [The students proud of theiri accomplishments]i celebrated.b. ??[The students proud of the studentsi� accomplishments]i celebrated.

The pronouns in the (a)-sentences can now be coindexed with the NPsthat contain them. The (b)-sentences, where the name is contained withinits antecedent, are, however, at best strongly marginal. The (b)-sentencesnow have the feel of traditional Condition C violations.The same sort of contrast occurs across complement and relative

clauses, e.g.

(7) a. *[The idea that Tom would mention iti]i was considered.b. [The idea thati Tom mentioned]i was considered.

(8) a. *[The attempt to repeat iti]i failed.b. [The attempt thati was repeated]i failed.

(9) a. *[The claim that nobody would understand iti]i scared everyone.b. [The claim thati nobody understood]i scared everyone.

The pronoun it in each of the complement clauses in the (a)-examplescannot be coindexed with the greater NP, whereas the relativepronouns in the (b)-examples are necessarily coindexed with thegreater NPs.The observation that immediately sheds light on these data is that the

pronouns in the bad NPs appear within arguments of those NPs, whereasthe acceptable NPs have the pronouns appearing within adjuncts. Inother words, the argument vs. adjunct distinction is the key to explainingthe contrast. While this insight is correct, it does not lead to an obvioussolution to the general problem, since it is not apparent how one mightformalize the insight in a principled manner. Furthermore, the followingexamples seem to contradict the insight:

(10) *[The man criticizing himi]i is here.

(11) *[The children teasing themi]i must go.

Antecedent containment 95

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 3: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

Within the greater NPs, the pronouns here are arguably inside adjuncts,yet the indicated readings are robustly unavailable. These sentences havethe feel of traditional Condition B violations.This paper will demonstrate that despite (10–11), the argument vs.

adjunct distinction is indeed key to explaining the data above and thatthis insight can be formalized in a dependency grammar (DG) frameworkthat views syntactic structure primarily in terms of catenae (as opposed toin terms of constituents). The catena is a syntactic unit associated withDG. By acknowledging catenae, one can define the domains in which trueantecedent containment occurs. The following formulations are thepillars of our account:

Catena

A word or a combination of words that is continuous with respectto dominance.

Predicate-argument structure

A predicate and those of its arguments with which it forms a catena.

Antecedent Containment Filter

A given catena cannot be coindexed with a predicate-argumentstructure containing it.

These three definitions represent the solution to the problem of antecedentcontainment illustrated in (1–11). An indexed expression (= pronoun,common noun, name, pro-form in general) may not be coindexed with apredicate-argument structure containing it. Predicate-argument struc-tures are always catenae, whereby a catena can easily include all thearguments of the given predicate at the same time that it excludes theadjuncts to that predicate.The problem of antecedent containment is of course not restricted to

data like (1–11). The majority of literature on the problem is associatedwith the antecedent contained deletion (ACD) of VP-ellipsis (e.g. Bouton1970, Sag 1976, May 1985, Baltin 1987, Larson & May 1990, Fiengo &May 1994, Hornstein 1994, Kennedy 1997, Johnson 2001, Wilder 2003,etc.). The antecedent to a VP-ellipsis is seen as containing the ellipsis,which gives rise to an infinite regress:

(12) Fred will read every book that Susan will read.

(13) Nolan doesn�t like the beers I do like.

The elided material is marked by the strikeouts, and the underline marksthe apparent antecedent to the ellipsis. Since the antecedent contains theellipsis itself, an infinite regress occurs. The preferred means ofovercoming this problem is quantifier raising (QR) (e.g. May 1985,Baltin 1987, Larson and May 1990, Fiengo & May 1994, Kennedy 1997,Johnson 2001, Harley 2002, Wilder 2003, etc.). The quantified NP is

96 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 4: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

raised to the left to a position at Logical Form where it is no longercontained within its antecedent. While Section 6 considers the QRapproach, we do not concentrate on ACD in this paper. The catena-based approach sees the problem of ACD as trivial, and the discussiontherefore focuses more on interesting data like those in (1–11).The paper is organized as follows: Some background information on

the antecedent containment problem is presented in Section 2. Section 3mentions some key aspects of our DG framework and the catena isintroduced and explained. Section 4 establishes the distinction betweenarguments and adjuncts in NPs, and Section 5 considers the extent towhich predicate-argument structures are catenae. Section 6 examines theACD associated with VP-ellipsis. Section 7 presents and explains theAntecedent Containment Filter, the principle that predicts whenantecedent containment does and does not occur. Section 8 addressessome peripheral but important issues, and Section 9 concludes the paper.The majority of data examined are from English. On occasion though,data from German are also examined to solidify the point at hand.

2. Some background

The following four subsections provide some background on the problemof antecedent containment. The first subsection briefly considers Chom-sky�s i-within-i Condition. While this condition is clearly insufficient in abroad sense, it succeeded at generating interest for the general problem ofantecedent containment. The next three subsections consider threepossible solutions to the problem: one by Haık (1987), one by Hoeksema& Napoli (1990), and one conceivable solution in terms of Late Merge(Lebeaux 1988, Chomsky 1993, 1995, Fox 2002). Each of these solutionsis rejected due to a significant shortcoming.

2.1 The i-within-i Condition

Chomsky (1981, 1986) proposed the i-within-i Condition as a means ofovercoming data that apparently contradicted his theory of binding:

(14) Fredi said that [a nasty rumor about himselfi] was circulating.

Chomsky was interested in accounting for the fact that the reflexivepronoun himself is not bound within the immediate clause that containsit, contrary to what a condition on the distribution of reflexives wouldseem to require (i.e. Condition A). In order to accomplish this goal, hestipulated that the local binding domain must contain an accessible‘‘subject’’. Thus since there is no accessible subject contained within thebracketed NP, the reflexive pronoun seeks its binder higher in thestructure.

Antecedent containment 97

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 5: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

The reflexive pronoun in (14) has two possible candidates to considerfor its binder. The first is the subject of the embedded clause, i.e. theentire NP in brackets. But this subject cannot be right, since the truebinder is obviously Fred in the matrix clause. Chomsky therefore neededto exclude the possibility that the embedded subject NP might be thebinder. The i-within-i Condition accomplished this task. A given proformcould not be contained within its antecedent, i.e. *[XP…Yi…]i. Given thiscondition, the next possible subject higher in the structure is the matrixsubject Fred. This was the right result. The i-within-i Condition thusenabled Chomsky to extend the relevant binding domain up into thematrix clause.The problems with Chomsky�s condition come from at least two

corners. The first is the actual failure of the condition to rectify therelevant condition on reflexives (see for instance Kuno 1987 and Johnson1987), e.g. [The children]i think that somebody found pictures of [eachother]i. The second problem for the condition is much more serious. Thei-within-i Condition makes the wrong prediction concerning the core datainvestigated in this paper, i.e. it incorrectly predicts NPs like [the mannext to hisi dog]i to be referentially circular and thus impossible. Thesedifficulties with the condition seem to have motivated some to quicklyreject the i-within-i Condition as empirically wrong.Despite these shortcomings, the i-within-i Condition survived, espe-

cially in numerous textbooks that chose to present Chomsky�s theory ofbinding. In this regard, the i-within-i Condition has perhaps generatedinterest in and helped focus investigations of the antecedent-containmentproblem in general.

2.2 Haık�s solution

Haık (1987) examines the antecedent containment problem associatedwith (1–2) directly. She addresses the problem in terms of binding. Hersolution is expressed as follows:

‘‘As a descriptive generalization, referential circularity is blocked if,in an NP of the form [NPi...xj], where xj is an empty category or apronominal to be constructed as referentially dependent on NPi,(i and j may be equal), the anaphoric element x is bound somehowinside NPi.’’ (Haık 1987:505)

In other words, a pronoun contained in its antecedent is possible if thatpronoun is bound by another phrasal constituent that also appears withinthat antecedent. Kennedy (2008) pursues an updated version of thissolution consistent with DPs.Haık (1987:505) illustrates her solution with the following structures

for the NPs in (1–2):

98 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 6: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

(15) NPi

Det N′

N PP

P NP

a. *a picture of itsi frame

PP

NPi

NPi

P NP

b. a man next to dog hisi

There is no phrasal constituent inside the greater NP in (15a) that couldbind its, so the NP is unacceptable due to referential circularity. In (15b)in contrast, the pronoun his is bound by the lower projection of a man,which, according to Haık�s solution, blocks referential circularity, theresult being perfectly acceptable.Consider how Haık�s solution might address a standard restrictive

relative clause. These clauses appear to involve two sources of referentialcircularity:

(16) NPi

NPi CP

C'

C IP

the artist thati Ruth prefers ti

The lower projection of the artist binds the relative pronoun that, whichovercomes the first source of apparent referential circularity, and therelative pronoun binds the trace, which overcomes the second apparentsource of referential circularity. In this manner, both dependentcategories are bound within the antecedent that contains them so thatthe infinite regress is overcome.While Haık�s solution seems to predict the data from the introduc-

tion, it is not apparent that it is applicable to other languages, aspointed out by Hoeksema and Napoli (1990). Hoeksema and Napoliview sentences like the following one from German as problematic forHaık�s solution. The pronoun appears within a pre-noun extendedparticiple phrase:

(17) [Der neben seinemi Hund stehende Mann]i hat gerulpst.the next.to his dog standing man has burped.�The man standing next to his dog burped.�

The pronoun seinem is contained within the greater NP der neben seinemHund stehende Mann. Thus according to Haık�s solution, the pronoun

Antecedent containment 99

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 7: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

seinem would have to be bound by the discontinuous expressionder...Mann. This does not seem possible. Perhaps one might view justthe noun Mann as c-commanding backwards and thus binding seinem.This assumption would, however, generate another problem. If just anoun (without its determiner) can bind a pronoun in (17), then we wouldexpect the same to hold in (15a), where just picture would be able to bindits. Viewed in this light, Haık�s solution is unconvincing.

2.3 Hoeksema and Napoli �s solution

Hoeksema and Napoli (1990) reject Haık�s account. They offer a similarsolution, though, insofar as the relevant pronoun must be bound withinthe greater NP. They assume, however, that the relevant binder is anoften covert operator instead of an overt NP. Their solution can beunderstood as follows:

(18) a. *[a picture of itsi frame]ib. a man (whoi is) next to hisi dog

The wh-operator, be it overt or covert, binds the pronoun his at LogicalForm. Since it is not possible to assume the presence of a covert operatorin (18a), the pronoun its is free within the greater NP and therefore seeksto adopt the greater NP itself as its antecedent, giving rise to referentialcircularity.The key insight underlying Hoeksema and Napoli�s solution is that the

acceptable instances of apparent antecedent containment often allow ananalysis in terms of a reduced relative clause, whereas the unacceptableNPs cannot do the same. The following NPs further illustrate thesolution:

(19) a poetess (whoi is) in love with heri topic

(20) the student (whoi is) upset about hisi grade

The relative pronoun, be it overt or covert, binds the pronoun in eachcase. The pronoun is therefore bound within its NP. No such analysis ispossible in the unacceptable cases:

(21) *the proof thati is of itsi existence

(22) *a friend whoi is of hisi father

Since these examples do not allow the presence of a wh-operator on thesurface, there is no motivation to assume the presence of such anoperator at LF.This solution to the problem of antecedent containment has a

shortcoming. This shortcoming is that it is possible to produce instanceslike (18b, 19–20) where a relative clause cannot appear on the surface:

100 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 8: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

(23) a. socks with holes in themb. *socks that are with holes in them

(24) a. children with ants in their pantsb. *children who are with ants in their pants

The (a)-NPs are fine even though the (b)-NPs illustrate that a relativepronoun cannot appear in surface syntax.This difficulty facing Hoeksema�s and Napoli�s solution is more evident

with extended participle and adjectival phrases in German. These phrasesdo not allow an analysis in terms covert operators:

(25) a. [der mit seinemi Los zufriedene Konig]ithe with his fate satisfied King�the king satisfied with his fate�

b. *[der, der mit seinemi Los zufriedene ist Konig]ithe who with his fate satisfied is kingIntended: �the king who is satisfied with his fate�

(26) a. [die ihreri Mutter helfende Tochter]ithe her mother helping daughter�the daughter helping her mother�

b. *[die, die ihreri Mutter helfende Tochter]ithe who her mother helping daughterIntended: �the daughter who is helping her mother�

The pre-noun modifiers cannot be manifest as relative clauses. Theattempt to do so results in robust grammaticality violations, as illustratedin the (b)-examples.The fact that analogous surface structures at times do not verify

Hoeksema and Napoli�s solution forces them to stipulate that therelevant level of syntactic representation in such cases is LF. Thewh-operator may not be possible on the surface, but it is always present atLF. We reject this reasoning. Stipulating the presence of some element atan unseen level of syntax or an unseen point in a derivation for whichthere is no surface analog robs the account of empirical validation. Inother words, it is impossible to falsify empirically the existence of covertelements at abstract levels of representation. We cannot be sure thatHoeksema and Napoli�s covert operators really exist.2 Our solutionbelow in terms of DG catenae does not rely on such covert elements.

2 We do not reject the existence of empty elements outright. For instance, one canplausibly argue for the existence of a covert relative pronoun in bare relative clauses, e.g.

(i) the people we avoid(ii) the people who we avoid

The fact that the appearance of the relative pronoun is always optional in such cases allowsone to plausibly argue that a null relative pronoun is present in cases like (i). Thisassumption is based on verifiable empiricism, though.

Antecedent containment 101

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 9: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

2.4 Late merge

One might also pursue a solution to the antecedent containment problemin terms of Late Merge (Lebeaux 1988, Chomsky 1993, Fox 2002), or interms of some notion akin to Late Merge. The basic motivation for LateMerge is seen with apparent exceptions to Condition C of the bindingtheory:

(27) a. ??Sam told heri that Susani should not give up.b. Sam told heri the news that Susani wanted to hear.

The reading in (27a) is unlikely due to a traditional Condition Cviolation. The similar structure in (27b), however, allows the indicatedreading. This contrast is explained in terms of Late Merge. The argumentthat Susan should not give up in (27a) is merged with the VP at a point inthe derivation before Condition C applies, whereas the relative clauseadjunct that Susan wanted to hear is merged late to the NP the news, at apoint in the derivation after Condition C applies.The same sort of analysis might be applied to the antecedent

containment problem illustrated by (1–2), repeated here again as (28–29):

(28) *[A picture of itsi frame]i is on the table.

(29) [The man next to hisi dog]i burped.

The argument PP of its frame in (28) would merge with a picture at apoint in the derivation before the relevant binding principle applies,whereas the adjunct next to his dog in (29) would be merged with the manat a point in the derivation after the binding principle has applied. Inother words, the Late Merge solution stipulates a specific ordering oftransformations and rule application in order to overcome the antecedentcontainment problem.Like the previous two solutions, the Late Merge solution is challenged

by extended participle phrases in German. Example (17) is repeated hereas (30):

(30) [der neben seinemi Hund stehende Mann]ithe next.to his dog standing man�the man standing next to his dog�

The relevant adjunct in this case, i.e. neben seinem Hund stehende, wouldhave to merge into the NP in a position that splits the determiner der andthe noun Mann. The basic architecture of constituency-based syntax andthe nature of Merge, which combines equi-level constituents, make thissort of late �insertion� seem unlikely. Viewed in this light, the Late Mergesolution is also unconvincing.The solution mentioned in the introduction and pursued below does

not suffer from such difficulties. The relevant syntactic units are catenae,and these catenae are clearly discernible in surface syntax.

102 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 10: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

3. The framework

Our account of antecedent containment is couched in the DG frameworkof Osborne (2005a, 2005b, 2006, 2007, 2008), Osborne & Groß (2009),Groß (2010), and Osborne et al (2011). This framework is consistent inimportant respects with a significant DG tradition (e.g. Tesniere 1959,Hays 1964, Robinson 1970, Kunze 1975, Matthews 1981, Hudson 1984,Sgall et al. 1986, Mel�cuk 1988, Schubert 1988, Starosta 1988, Hudson1990, Lobin 1993, Pickering & Barry 1993, Engel 1994, Jung 1995,Heringer 1996, Groß 1999, Eroms 2000, Kahane 2000, Tarvainen 2000,Agel et al. 2003, 2006).

3.1 Dependency grammar

The advantage that dependency-based syntax has over constituency-based syntax is its parsimony.3 The minimal structures that thedependency relation generates are transparent and can be understoodwith little background in DG. The dependency relation is a one-to-onerelation. For each word there is exactly one node in the structure, andvice versa. Thus a word is a node, and a node is a word. Dependencyorganizes the words of sentences hierarchically into tree structures likethe following one:

(31) matches

SUBJ OBJ

number number

The of

words

The number of words matches the number of nodes.

the of

nodes

The one-to-one relation is clearly visible. There are nine words and ninenodes. This one-to-one relation allows one to plug the words themselvesdirectly into the tree, as done here. The corresponding constituencystructure of (31) would contain approximately twice the number of nodesand edges.DG is verb centered. The finite verb is the root word of the entire tree.

A label for the dependency type can be assigned to each dependency edge,as illustrated in (31) for the subject NP and the object NP. Both subject

3 The term ‘‘dependency-based syntax’’ denotes all those frameworks that are based onthe dependency relation, e.g. Word Grammar, Meaning to Text Theory, Functional Gen-erative Description, Lexicase, Algebraic Syntax, Operator Grammar, etc. The term ‘‘con-stituency-based syntax’’ denotes all those frameworks based on the constituency relation,e.g. GB, MP, HPSG, LFG, etc.

Antecedent containment 103

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 11: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

and object can be dependents of the root verb if just a single finite verb ispresent, as in (31). As soon as an infinitive appears, however, subject andobject are no longer equi-level:

(32) willSUBJ

number matchOBJ

The of number

words the of

nodes

The number of words will match the number of nodes.

The subject the number of words and the object the number of nodes are nolonger equi-level, the subject appearing as a dependent of the finite verband therefore positioned higher in the hierarchy than the object. In thismanner, the special status of the subject is acknowledged.4

Dependency hierarchies are like constituency hierarchies insofar as thehierarchy assumed is the one that best matches the results of constituencytests. Standard constituency tests (e.g. topicalization, clefting, pseudoc-lefting, proform substitution, answer fragments, etc.) verify for the mostpart that the constituents of dependency structures (=complete subtrees)are as shown. For instance, the constituency tests will identify the subjectNP the number of words, the object NP the number of nodes, and the VPmatch the number of nodes in (32) as constituents. These wordcombinations are appropriately shown as complete subtrees, i.e. asconstituents.5

There is controversy concerning the status of certain dependencies. Forinstance, some DGs assume DP over NP. Most DGs, however, assumeNP (e.g. Schubert 1988; Engel 1994; Van Langendonck 1994; Heringer1996; Weber 1997; Tarvainen 2000; Groß 1999, 2003; Hellwig 2003;Hyvarinen 2003; Kahane 2003; Uzonyi 2003; Starosta 1988, 2003;

4 Following Tesniere, many DGs treat subjects much in the same manner as any otherkind of argument. However, DGs also have to acknowledge a dependency between thesubject nominal and the finite verb, e.g. the dependency connecting number to will in (32).The special status of the subject derives from the fact that subjects attach to the top node ofa predicate catena (see Section 5).

5 It is not common for DGs to acknowledge constituents, since the constituent is deemeda unit of constituency grammar, not of DG. DGs do acknowledge complete subtrees as aparticular unit of syntax, however (e.g. Tesniere 1959/69:14, Hays 1964:520, Hudson1984:92, Mel�cuk 1988:14, Groß 1999:69, Eroms 2000:86ff., and Hellwig 2003:603), althoughthe terminology varies. The advantage of acknowledging constituents in dependency-basedstructures is that doing so makes a comparison of constituent structure possible across thetwo structure types (i.e. DG constituents vs. constituency grammar constituents).

104 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 12: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

Mel�cuk 1988, 2003). We follow the majority view in this paper, althoughthe matter does not bear on the issue of antecedent containment.These few brief remarks about our DG will have to suffice. The

heavy use of dependency trees below should help bridge gaps inunderstanding.

3.2 The catena

The fundamental unit of our DG is introduced in this section, i.e. thecatena (Latin for �chain�, plural catenae). O�Grady (1998) was the first toemploy the catena concept; he posited the catena as a basis for hisaccount of the syntax of idioms. Osborne (2005a, 2007:32f.), Groß andOsborne (2009), and Groß (2010) develop the concept further, demon-strating that the catena provides insights into a number of phenomena,not just idioms. O�Grady, Osborne, and Groß employed the term‘‘chain’’, however, not the term ‘‘catena’’. Osborne et al. (in press)introduce the term ‘‘catena’’ in place of ‘‘chain’’ to avoid confusion withpre-existing chain concepts, the chain of derivational constituency-basedtheories being a different notion almost entirely.The catena is defined as follows:

Catena

A word or a combination of words that is continuous with respectto dominance

The catena is illustrated using the following tree:

(33) Do A

they B understand C

view E

thisD of F

things G

Do they understand this view of things?

The capital letters serve to abbreviate the words. There are 34 distinctcatenae in (33), all of which are listed here: A, B, C, D, E, F, G, AB, AC,CE, DE, EF, FG, ABC, ACE, CDE, CEF, DEF, EFG, ABCE, ACDE,ACEF, CDEF, CEFG, DEFG, ABCDE, ABCEF, ACDEF, ACEFG,CDEFG, ABCDEF, ABCEFG, ACDEFG, ABCDEFG. Each of theseword combinations is continuous with respect to the vertical axis, i.e.with respect to dominance. Notice that some of the catenae have morethan one ‘‘branch’’, e.g. ABC, DEF, DEFG. Notice also that some ofthese catenae are not continuous with respect to the horizontal axis, i.e.with respect to precedence, e.g. AC, CEF, ACE, etc.

Antecedent containment 105

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 13: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

The DG catena is a much more flexible unit of syntax than theconstituency grammar constituent. A strictly binary branching constit-uency tree of (33) would contain 14 constituents. There are thereforealmost three times more catenae in (33) than constituency grammarconstituents. Despite this much greater number, the catena is not anall inclusive unit. There are, namely, altogether 127 (=27–1) distinctword combinations in (33),6 which means there are 93 (=127–34)word combinations in (33) that fail to qualify as catenae, e.g. BC,CD, BCF, ACEG. Viewed in this light, the catena is also an exclusiveunit.

4. Arguments vs. adjuncts

Most if not all theories of grammar acknowledge the distinction betweenarguments and adjuncts, and DG has traditionally shown a particularlystrong interest in the distinction, starting with Tesniere (1959) andextending through decades of work on the valence of predicates (� fullverbs). The concept of verb valence is directly linked to Tesniere�s work,since he introduced the valence metaphor (from chemistry) directly tolinguistics. Tesniere distinguished between actants (� arguments) andcirconstants (� adjuncts), the former being directly associated with coreclausal semantic content and the latter with additional, circumstantialinformation. Most modern valence theory now posits a three-waydistinction: in addition to the argument-adjunct distinction, one discernsbetween obligatory and optional arguments. While our system alsoassumes the three-way distinction, the early two-way division (betweenjust arguments and adjuncts) is sufficient for the analysis of antecedentcontainment.The introduction mentioned that the argument vs. adjunct distinction

is the key to identifying when antecedent containment does and does notoccur. In this regard, the discussion now examines the distinctionbetween arguments and adjuncts within NPs. One can note in this areathat the distinction is widely acknowledged for clauses, but less effort hasbeen devoted to the distinction within NPs. For the antecedentcontainment phenomenon, however, the NP is the pertinent domain forthe analysis. In other words, what is needed is a reliable means ofdistinguishing between arguments and adjuncts within NPs. We employthe relative clause diagnostic to this end.The relative clause diagnostic proceeds from the assumption that

relative clauses are adjuncts (not arguments) and that if an expressionwithin an NP can alternately appear as the predicate of a relative clause,

6 The number of distinct word combinations is calculated using the formula 2n-1, wheren = the number of words.

106 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 14: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

that expression is an adjunct. This diagnostic identifies many post-nounPPs as adjuncts:

(34) a. the musicians from Texasb. the musicians who are from Texas

(35) a. a concert on Tuesdayb. a concert that is on Tuesday

(36) a. our discussion about politicsb. our discussion that was about politics

The acceptability of the relative clauses in the (b)-examples identifies thepost-noun PPs in the (a)-examples as adjuncts. The PPs in (34–36) mustbe compared with of-PPs. The preposition of typically introducesarguments:

(37) a. the description of Fredb. ??the description that was of Fred

(38) a. half of the milkb. *half that was of the milk

(39) a. the syntax of idiomsb. *the syntax that is of idioms

The attempts in the (b)-examples to render the of-PPs in the (a)-examplesas relative clauses fail. Based on these data, one can conclude that of-PPs(within NPs) are arguments.While the relative clause diagnostic identifies many post-noun PPs as

adjuncts (i.e. many of the PPs introduced by a preposition other than of),other post-noun PPs are arguments:

(40) a. the wait for Fredb. *the wait that was for Fred

(41) a. satisfaction with the possibilitiesb. *satisfaction that is with the possibilities

(42) a. reliance on the boysb. *reliance that is on the boys

The (b)-examples identify the PPs in the (a)-examples as arguments. Thediagnostic is also applicable to pre-noun expressions. It identifiesattributive adjectives as adjuncts:

(43) a. this old carb. this car that is old

(44) a. a funny storyb. a story that is funny

Antecedent containment 107

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 15: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

(45) a. many difficult choicesb. many choices that are difficult

The diagnostic sees determiners as arguments (or as argument-like):

(46) a. that long manuscriptb. *long manuscript that is that

(47) a. my freshly-cut flowersb. *freshly-cut flowers that are my

(48) a. an unusual analysisb. *unusual analysis that is a(n)

While it is not clear from a semantic point of view how one canview many determiners as arguments, their behavior with respect tothe relative clause diagnostic suggests that they are at least argument-like.The relative clause diagnostic is also applicable to pre-noun extended

participle phrases in German (of the sort mentioned in the previoussections). The fact that these phrases can be rendered as relative clausesidentifies them as adjuncts:

(49) a. der auf der Bank schlafende Touristthe on the bench sleeping tourist

b. der Tourist, der auf der Bank schlaftthe tourist who on the bench sleeps

(50) a. die, dass sie bleiben werden, versprechenden Kinderthe that they stay will promising children

b. die Kinder, die versprechen, dass sie bleiben werdenthe children who promise that they stay will

(51) a. die die Elements lesenden Studenten

the the Elements reading studentsb. die Studenten, die die Elements lesen

the students who the Elements read

Even though the participles in the (a)-NPs must change form (participlefi finite verb) in order to appear in a relative clause, the pattern isconsistent with that of attributive adjectives. Thus the stance that pre-noun participle phrases in German are adjuncts is motivated. These pre-noun phrases are crucial to the analysis of the antecedent containmentphenomenon in general, as demonstrated above and as will be furtherestablished below.There are cases where the relative clause diagnostic struggles to provide

a clear judgment, for example with certain determiner-like expressions,e.g. the two men fi ?the men who are two. Despite such difficulties, thediagnostic is reliable for the data examined in this paper.

108 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 16: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

5. Predicate-argument structures

A catena-based theory of syntax can straightforwardly acknowledgepredicate-argument structures in the syntax. Predicates and their argu-ments are always catenae in continuous structures. As an illustration,consider the following periphrastic verb combinations:

(52) visits will

She him She visit

him

a. She visits him. b. She will visit him.

will is

She have He being

visited visited

him

c. She will have visited him. d. He is being visited.

will will

He be He have

visited been

visited

e. He will be visited. f. He will have been visited.

will have been being visited

have

will

have

been

being

visited

.

Will Will

visit he

been

visited

He

g. He

she

him

h. Will she visit him? i. Will he have been visited?

Antecedent containment 109

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 17: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

The matrix predicate in these cases consists of the lexical verb visit and upto four function verbs. The verbs together form a catena in each case.These catenae are in italics. Note that the passive-active dichotomy doesnot alter the catena status of these predicates. Note also that subject-verbinversion does not affect their status as catenae.Sentences (52a–i) illustrate the catena status of matrix predicates.

The arguments of predicate-argument structures are also always catenaein continuous structures. This fact is illustrated well using presentparticiples.

(53) children children

the laughing the laughing

at

them

a. the laughing children b. the children laughing at them

(54) people people

the eating the trying

rice to

eat

rice

a. the people eating rice b. the people trying to eat rice

Following a similar convention employed by Eroms (2000:85f.), adjunctsare identified with a dependency edge pointing away from the adjunct(Eroms� half arrow points toward the adjunct). The adjunct predicate isin italics in each case. The fact that the arrow points away from theadjunct (toward the governor of the adjunct) is motivated by thebehavior of adjuncts. A defining trait of adjuncts is that they are notselected by their governor, but rather they select their governor. In otherwords, semantic selection is running opposite ‘‘normal’’. Placing thearrow so that it points up the hierarchy is intended to signify this state ofaffairs.The argument of laughing (the...children) is a catena in (53a) and the

arguments of laughing (the children and at them) are catenae in (53b).Similarly, the arguments of eating (the people and rice) are catenae in(54a), and the arguments of trying (the people and to eat rice) are catenaein (54b). Consider next the following example from German involving anextended pre-noun participle phrase:

110 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 18: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

(55) kinder

die herunterlügenden

uns Blaue

das vomHimmel

die uns das Blaue vom Himmel herunterlügenden Kinder the us the blue from.the sky down.lying children‘the children blatantly lying to us’

The participle predicate herunterlugenden takes three arguments (i.e.die...Kinder, uns, and the idiomatic das Blaue vom Himmel), each of whichis a catena.Examples (52–55) show verbal predicates. Of course other categories

can also be predicates, e.g. adjectives, common nouns, and certainprepositions. The arguments of common nouns are of particular interestto the antecedent containment phenomenon, as mentioned in theintroduction and considered further in the previous section:

(56) friend father desire

a of the of a to

mine Fred stay

a. a friend of mine b. the father of fred c. a desire to stay

claim possibility

the that the that

departed might

she it work

d. the claim that she departed e. the possibility that it might work

The common noun predicates are in italics. In each example, the post-dependent (=the dependent that follows) of the common noun is anargument of that noun. Each of these arguments is a catena.Perhaps the most striking insight about predicate-argument structures

is that subordinators can be viewed as two-place predicates, whereby thefirst argument of these predicates encompasses the main clause, and thesecond argument encompasses the subordinate clause.

Antecedent containment 111

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 19: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

(57) left helped

He before They because

did are

she they friendly

a. He left before she did. b. They helped because they are friendly.

The adjunct predicates are again in italics. The predicate before takes theargument catenae he left and she did, and the predicate because takes theargument catenae they helped and they are friendly.It must be emphasized that while predicates and arguments are always

catenae in continuous structures, they can be non-catenae in discontin-uous structures, e.g.

(58) do

happy you think

How isg

Fred

How happy do you think Fred is?

The embedded predicate in this case is a non-catena. That is, is andhappy form the matrix predicate of the embedded clause, but due to thediscontinuity, i.e. to the fronting of how happy, this predicate does notform a catena. We address such discontinuities in terms of rising(Osborne 2005:236ff., 2007:34ff., Groß & Osborne 2009, Osborne et al.in press). The discontinuous catena is said to have risen. The dasheddependency edge marks the risen catena and the ‘‘g’’ subscript marksthe governor of that risen catena. We do not explore rising here.Instead, we merely wish to emphasize that this aspect of discontinuitiesis not a problem for the DG understanding of predicate-argumentstructures, since such cases are accounted for in terms of the DG theoryof discontinuities – see Groß and Osborne (2009) for a detaileddiscussion.

6. Antecedent contained deletion

The catena-based approach to the syntax of natural language is not

challenged by putative instances of antecedent contained deletion (ACD)(e.g. Bouton 1970, Sag 1976, May 1985, Baltin 1987, Larson &May 1990,Fiengo & May 1994, Hornstein 1994, Johnson 2001, Wilder 2003,Kennedy 1997, 2008, etc.). The VP-ellipsis corresponds to a catena in

112 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 20: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

cases of putative ACD. Sentences (11–12) from the introduction arereproduced here as (59–60):

(59) will

Fred read

book

every that

Susan readg

every book that Susan will read.

will

Fred will read

the beers I do

(60) doesn’t

Nolan like

beers

the do I

Nolan doesn’t likelike.

The struck through word in each case identifies the antecedent to theellipsis, which is underlined. Thus just read is the antecedent to the ellipsisin (59) and just like is the antecedent to the ellipsis in (60). By definition,an individual word is a catena. The point, then, is that a catena-basedapproach sees no reason to assume that the entire VP is the antecedent tothe ellipsis. The catena-based approach can view just the verb as theantecedent in (59–60). As a catena, this antecedent is a concrete andexclusive unit of syntax. The infinite regress problem associated withACD never arises.Individual words are of course also always constituents in constituency

grammars. Therefore in light of (59–60), a constituency grammar mightpursue a similar approach, whereby one abandons the notion that theentire VP must be the antecedent to the ellipsis, allowing instead just theverb to be the antecedent. Indeed, Wilder (2003) suggests that such anapproach might be feasible, although he does not pursue the possibility,preferring instead the QR approach. Such a constituency grammarapproach would be, however, immediately challenged by cases like thefollowing ones:

Antecedent containment 113

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 21: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

every book she does to read.

(61) wants

He to

read

book

every does

she to

read

He wants to readwants

g visit that

everyone will visit Mary that you do willg visit Mary.

(62) expect

I will

everyone

Mary do

you will

visit

Mary

I expectexpect

The underlined antecedents in these cases are clearly non-constituents inconstituency grammars (and in DGs as well). They are, however, catenae.Thus the catena-based approach has no problem with such sentences.There continues to be no reason to assume that the antecedentencompasses the entire VP.Instances of pseudogapping, which is arguably a particular manifes-

tation of VP-ellipsis (Jayaseelan 1990, Lasnik 1995, 1999, Baltin 2003,Johnson 2001, 2009), challenge most constituency grammar approachesand support most our catena-based account:

to me, or

(63) Would

you want

to

say

that to

me

Would you want to say thatwould I to you?

114 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 22: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

Susan some money, but

(64) will

I give

Susan money

some

I will giveI won’t Fred.

The underlined antecedents to the ellipses in these cases are clearly non-constituents, but they are of course catenae. The key trait of pseudogappingis that a remnant of the elided VP remains. The result is that elidedmaterialoften fails to qualify as a constituent. A widespread approach to thisproblem is to stipulate that the remnant has been raised out of the VP, theresult being that a constituent VP is actually elided, contrary to appearances(e.g.Kuno 1981, Jayaseelan 1990,Lasnik 1995, Johnson 2001,Wilder 2003).It should be apparent that the catena-based approach can address these non-constituent gaps without appealing to movement. The elided materialcorresponds to a catena, and as such, it is a well-defined unit of syntax.The catena-requirement on VP-ellipsis is easily tested. The following

example is taken from Osborne (2005a:281). The example itself is closelysimilar to an example discussed byRoss (1970:52) in the context of gapping:

(65) will

He want

to

try

to

begin

to

write

play

a

He will want to try to begin to write a play, but a. he won’t a novel. b. he won’t to write a novel. c. he won’t to begin to write a novel. d. he won’t to try to begin to write a novel. e. *he won’t to try a novel. f. *he won’t to try to write a novel. g. *he won’t to try to begin a novel. h. *he won’t to begin a novel.

Antecedent containment 115

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 23: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

The (at least somewhat) acceptable (65a–d) each have the ellipsiscorresponding to a catena, whereas in the ungrammatical (65e–h), theellipsis fails to correspond to a catena.The examples in this section suggest that VP-ellipses correspond to

catenae. A more detailed discussion would firmly establish the point. Thefocus of this paper is not VP-ellipsis, however, but rather it is the problemof antecedent containment. It suffices to acknowledge that the infiniteregress of ACD does not arise in a catena-based theory of VP-ellipsis.

7. The Antecedent Containment Filter

The discussion now returns to the antecedent containment probleminvolving NPs described in the introduction. In order to address the data,the relevant catenae need to be identified within which a given proformcan and cannot find its antecedent. These catenae are called predicate-argument structures:

Predicate-argument structure

A predicate and those of its arguments with which it forms a catena

Given this notion, the filter that predicts the antecedent containment datais as follows:

Antecedent Containment Filter

A given catena cannot be coindexed with a predicate-argumentstructure containing it.

The predicate-argument structure of a given predicate excludes theadjuncts to that predicate. The relevant predicate-argument structureappears in italics and its root word receives the index. Examples (1–2)from the introduction receive the following analyses:

(66) picturei (67) mani

a of the

frame

next

to

itsi dog

hisi

*a picturei of itsi frame the mani next to hisi dog

The predicate-argument structure for the predicate picture in (66)therefore encompasses the entire NP a picture of its frame. Since thisNP includes the pronoun its, the Antecedent Containment Filter isviolated and the indicated reading is predictably impossible. Thepredicate-argument structure of the predicate man in (67), in contrast,does not include the adjunct next to his dog, so the pronoun his is not

116 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 24: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

coindexed with a predicate-argument structure containing it, so theindicated reading is predictably quite possible. Note that both predicate-argument structures, i.e. a picture of its frame and the man, are catenae.Examples (6a–b) from the introduction, reproduced here as (68a–b),

receive the following analyses:7

ideai ideai

thati

would

Tom mentiong

i

(68)

the that the

would

Tom mention

it

a. *the ideai that Tom would mention iti b. the ideai thati Tom would mention7

The predicate-argument structure for the predicate idea in (68a) is theentire NP because the clause that Tom would mention it is an argumentclause, not an adjunct clause. The indicated reading in (68a) is thereforeblocked. The predicate-argument structure for the predicate idea in (68b),in contrast, excludes the relative clause because relative clauses areadjuncts, not arguments. The indicated reading in (68b) is thereforepredictably possible.Our analysis is most convincing when confronted with pre-noun

pronouns, such as those in extended participle phrases in German:

Kinderi

die versprechenden

i helfen

die dass siei helfen würden versprechenden Kinderi

the that they help would promising children

(69)

dass

würden

sie

‘the children promising that they would help’

The predicate-argument structure for the predicate Kinder excludes theadjunct participle phrase dass sie helfen wurden versprechenden. Thus thepronoun sie can take the catena die…Kinder as its antecedent, whichmeans the Antecedent Containment Filter is not violated and theindicated reading is natural.The following NP contains two pronouns, each of them in an adjunct:

7 As mentioned in Section 5, the dashed dependency edge indicates rising and the ‘‘g’’subscript marks the governor of the risen catena.

Antecedent containment 117

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 25: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

Fussgängeri

der stolze mit

i

i

der auf seineni Hut stolze Fussgängeri mit seinemi Hund

(70)

auf Hund

Hut seinem

seinen

the of his hat proud pedestrian with his dog‘the pedestrian with his dog who is proud of his hat’

Despite being flanked on both sides by adjuncts, the noun Fussgangerforms a catena with its determiner der. The predicate-argument structurefor the predicate Fussganger is therefore just der…Fussganger.Examine next the predicate-argument structure of the following

example:

Vateri

der zufriedene Freundes

meines

i

der mit seinemi Los zufriedene Vateri meines Freundes

der mit seinemi Los zufriedene Vateri seinesi Freundes of.his

(71)

mitLos

seinem

a. the with his fate satisfied father of.my friend ‘the father of my friend who is satisfied with his fate’

b. *

The predicate-argument structure for Vater is der…Vater meines/seinesFreundes. Both the pre-dependent of Vater, i.e. der, and the post-dependent of Vater, i.e. meines/seines Freundes, are included in thispredicate-argument structure. Despite this fact, the predicate-argumentstructure is a catena, so the indicated readings are consistent with theAntecedent Containment Filter, the pre-dependent adjunct constituentmit seinem Los zufriedene being outside of the relevant predicate-argument structure. Clause (71a) is fine because there is no antecedentcontainment violation within the greater predicate-argument structure.Clause (71b) is bad, in contrast, because the pronoun seines is coindexedwith the predicate argument structure that contains it.The Antecedent Containment Filter is the current DG�s solution to the

problem of antecedent containment. Antecedent containment of coursenever actually occurs. What looks like antecedent containment isaddressed in terms of catenae. The antecedents to pronouns are catenae,whereby these catenae may or may not qualify as constituents.

118 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 26: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

8. Remaining issues

The following four sections briefly consider peripheral but impor-tant aspects of our account in terms of the Antecedent Containment Filter.

8.1 Condition B effects

Examples (9–10) from the introduction involving participle phrasesappear to challenge the Antecedent Containment Filter. Examples (9–10)are reproduced here as (72–73):

man childreni

criticizing teasing

themi

the mani criticizing himi the childreni teasing themi

(72)

the

him

a.*

i (73)

the

i

. a. *

At first glance, these examples appear to contradict the current account.The indicated readings are impossible despite the fact that the pronounappears in an adjunct. Appearances are deceptive in these cases, though.The participles are predicates too, which means they have their ownpredicate-argument structures. The predicate criticizing in (72a), forinstance, takes the arguments the man and him, so its predicate-argumentstructure is the entire NP the man criticizing him.What is occurring in these examples is a traditional Condition B effect.

Within the predicate-argument structures of the participles, the pronoun iscoindexedwithaco-argument.Of courseapronouncannotbecoindexedwithits co-argument. A reflexive, however, can be coindexed with a co-argument:

mani childreni

the the teasing

i i

(72)

himself

b. the mani criticizing himselfi

(73)

criticizing

themselves

b. the childreni teasing themselvesi

Given the reflexives, Condition B is satisfied. The reflexive in each case iscoindexed with a co-argument. Note that the Antecedent ContainmentFilter has no problem with (72b, 73b), since the reflexives appear insideadjuncts. The point, then, is that the traditional binding conditions – i.e.Conditions A, B, and C effects – are (at least partially) independent of theAntecedent Containment Filter.8

8 Major problems plague the traditional binding theory, as has been thoroughly discussedand demonstrated in the literature on binding for two and a half decades now. In thisregard, we view the binding theories developed by Pollard and Sag (1992, 1994) andReinhart and Reuland (1993) as more promising. Our terminology, i.e. ‘‘Condition A/B/Ceffect’’, is motivated entirely by the desire for efficient exposition.

Antecedent containment 119

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 27: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

8.2 Predicate NPs

The behavior of predicate NPs complicates the analysis. Referentialcircularity does not occur if the greater NP is a predicate NP:

(74) is (75) is

Fred friend She supporter

a of a of

brother mother

his her

Fredi is [a friend of hisi brother]i. Shei is [a supporter of heri mother]i

Data like these are discussed in a number of places (e.g. Williams 1983,Hornstein 1984:113, Hoeksema & Napoli 1990:406ff.). These sentencesare perfect despite the fact that the pronoun appears to be containedwithin its antecedent predicate-argument structure.Appearances here are misleading, though. The bracketed NPs

actually fail to qualify as complete predicate-argument structures; theyare, rather, only partial predicate-argument structures. The matrixpredicates in these cases necessarily include the copula. Thus in orderfor the pronouns to be coindexed with complete predicate-argumentstructures, they would have to be coindexed with the entire sentences.Such a coindexation is prohibited on two counts, the AntecedentContainment Filter being the one and the mismatch in category(copular verb vs. pronoun) the other. Coindexed catenae should be ofthe same category.Given these insights, the peculiar behavior of predicate NPs does not

challenge the current account of antecedent containment. Predicate NPsdo not qualify as (complete) predicate-argument structures, which meansthe Antecedent Containment Filter is inapplicable.

8.3 Circularity

The apparent circularity of reference in the following example is closelyassociated with the antecedent containment problem:

(76) ??[Hisi wife]j saw [herj husband]i.

Higgenbotham (1983), Haık (1987), and Kennedy (2008) take theindicated reading to be fully unavailable. The current theory, however,does not predict the indicated reading to be impossible. Since there is noantecedent containment, the Antecedent Containment Filter actually saysnothing about (76). If Higginbotham, Haık, and Kennedy are correct andthe circularity of reference in (76) is completely impossible, then thecurrent account of antecedent containment is lacking, since it leaves some

120 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 28: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

apparently closely related data unexplained. We think, however, that thereading indicated in (76) is not wholly impossible and that independentconsiderations account for the strong marginality of the indicatedreading. We therefore agree with Hoeksema and Napoli (1990:422), whoobserve that with the correct context and intonation, the readingindicated in (76) is possible:

(76) A: Whose wife saw her husband?B: (pointing to a man) HIS wife saw her husband.

Notice further that the reading in (76) does not have the feel of trueantecedent containment. Compare (76) with (1), repeated here as (77):

(77) *[A picture of itsi frame]i is on the table.

The indicted reading in (77) is robustly unavailable; in fact it isnonsensical. The same cannot be said about (76). The indicated readingin (76) is far from nonsensical, for it is quite plausible for a wife to likeher husband. Second, observe that if a proper noun replaces the firstpronoun, the availability of the reading increases, whereby an addendumcan force the indicated reading:

(78) [Sami�s wife]j saw [herj husband]i, i.e. she saw Sam.

This fact demonstrates that the circularity of the indices alone is not thesource of the strong marginality of (76).Third, note that pronouns in general often reluctantly seek their

antecedent in the following context:

(79) ?Hisi wife likes Sami.

Informants consistently view the reading in (79) as unlikely. Pronounsprefer to function anaphorically (= looking backward) rather thancataphorically (= looking forward). The indicated reading becomes evenmore unlikely if a common noun replaces the proper noun:

(80) ??Hisi wife likes [the plumber]i.

Furthermore, the following reading seems completely unavailable:

(81) *Sami�s wife likes [the plumber]i.

In this case, the indicated reading is unavailable even though there is nocircularity of reference and no pronoun is involved.Based on these additional considerations, the conclusion we draw is

that the strong marginality of the reading indicated in (76) is not directlyrelated to the antecedent containment issue. Some principle of bindingthat is at least partially independent of the Antecedent ContainmentFilter must be at work in such cases.

Antecedent containment 121

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 29: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

8.4 Reflexives

Due to a complication, a discussion of the data involving reflexives hasbeen postponed until now. As things stand, the Antecedent ContainmentFilter applies to reflexives too; it prevents a reflexive from beingcontained in its antecedent predicate-argument structure. The data,however, are unclear in this area:

(82) a. ?We met [a supporter of himselfi]i.b. We met someone who supports himself.

(83) a. ?I avoid [admirers of themselvesi]i.b. I avoid people who admire themselves.

(84) a. ?Do you know [the tormenter of himselfi]i.b. Do you know the guy who torments himself.

The Antecedent Containment Filter predicts the (a)-sentences to be badbecause the reflexive pronoun in each case is contained in its antecedentpredicate-argument structure. The (a)-sentences seem, however, at leastsomewhat possible. While they are undoubtedly awkward – as thecontrast with the fully acceptable (b)-sentences demonstrates – theynevertheless do not seem on par with instances of antecedent contain-ment involving pronouns. Furthermore, we can certainly assign aninterpretation to the (a)-sentences; they make sense.The marginality of the (a)-sentences places them into a grey area. The

Antecedent Containment Filter predicts them to be bad, which they arenot entirely. The fact that they are marginal, though, suggests thatantecedent containment is involved to some extent. A perhaps tellinginsight about (82–84) is that the common noun predicates haveincorporated their subject arguments (a supporter = someone whosupports, an admirer = someone who admires, a tormenter = someonewho torments). Examples where the relevant predicates do not incorpo-rate their subject arguments are much worse, e.g.

(85) *[A picture of itselfi]i is on the table.

(86) *[The boss of himselfi]i missed the meeting.

(87) *[The top of itselfi]i is in sight.

These sentences are nonsensical. One is hardly in a position to assignthem an interpretation at all. For instance, it is difficult to imagine what a�picture of itself� would look like. These data are therefore consistent withthe Antecedent Containment Filter.Returning to (82–84), what seems to be occurring with (82a, 83a, 84a) is

that the reflexives are coindexed with the incorporated arguments, notwith the greater predicate-argument structures. DG has a means ofaccommodating this possibility. One acknowledges that the incorporated

122 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 30: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

arguments are manifest in the morphology as the -er suffix. One thenextends the dependencies into the words in such a manner that both inter-and intra-word dependencies are acknowledged (see Osborne 2005a:286ff.and Groß 2010). The resulting analyses of (82a, 83a) are as follows:

i i

themselvesi

(88) met avoid

We -er

a support admir

of of

himself

a . ?

We -ers

i

(89)

We met a support-er of himself. a . ?We avoid admir -ers of themselves.

The dotted edges mark intra-word dependencies. In each case the reflexiveis coindexed just with the -er suffix, which means there is no violation ofthe Antecedent Containment Filter. The -er suffix has the status of anargument, not of a predicate, so there is no predicate-argument structurewith which the reflexive is coindexed. The behavior of reflexives withrespect to the Antecedent Containment Filter is therefore predictable.

9. Conclusion

This paper has demonstrated that a catena-based theory of syntax has noproblem with structures involving putative antecedent containment.Referring expressions that at first blush seem to appear within theirantecedents actually take non-constituent antecedents that do notcontain them. The overall account is possible due to the DG catena, amuch more flexible unit of syntax than the constituent. By acknowledgingcatenae, the account easily overcomes the problem of antecedentcontainment, and it accomplishes this task while acknowledging surfaceconfigurations only.The catenae important for the account of putative antecedent contain-

ment are identified and defined in terms of predicate-argument structures.Predicates and their arguments always qualify as catenae in continuousstructures. The three key definitions are given one last time here:

Catena

A word or a combination of words that is continuous with respectto dominance

Predicate-argument structure

A predicate and those of its arguments with which it forms a catena

Antecedent Containment Filter

A given catena cannot be coindexed with a predicate-argumentstructure containing it.

Antecedent containment 123

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 31: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

These definitions formalize the key insight mentioned in the introduction.This insight is that apparent counterexamples to the general prohibitionon antecedent containment always involve adjuncts. These adjuncts areoutside of the antecedent predicate-argument structures. In this manner,antecedent containment never occurs.

References

gel, V., Eichinger, L., Eroms, H.-W., Hellwig, P., Heringer, H. J., & Lobin,H. 2003. Dependency and valency: An international handbook of contemporaryresearch, Vol. 1. Berlin: Walter de Gruyter.

gel, V., Eichinger, L., Eroms, H.-W., Hellwig, P., Heringer, H. J., & Lobin,H. 2006. Dependency and valency: An international handbook of contemporaryresearch, Vol. 2. Berlin: Walter de Gruyter.

Baltin, M. 1987. Do antecedent-contained deletions exist? Linguistic Inquiry 18/4, 579–595.

Baltin, M. 2003. The interaction of ellipsis and binding: Implications for thesequencing of Principle A. Natural Language and Linguistic Theory 21, 215–246.

Bouton, L., 1970. Antecedent-contained pro-forms. Proceedings of Sixth Re-gional Meeting of the Chicago Linguistic Society. ed. M. Cambell, 154–167.Chicago: The University of Chicago Press.

Brody, M. 1981. On circular readings. Linguistic Research 1, 95–105.Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: ForisPublications.

Chomsky, N. 1986. Knowledge of language: Its nature, origin and use. New York:Praeger.

Chomsky, N. 1993. A minimalist program for linguistic theory. The view fromBuilding 20: Essays in linguistics in honor of Sylvain Bromberger, eds. K. Hale &S. Keyser, 1–52. Cambridge, MA: MIT Press.

Chomsky, N. 1995. The Minimalist Program. Cambridge, MA: MIT Press.Engel, U. 1994. Syntax der deutschen Gegenwartssprache, 3rd revised edition.Erich Berlin: Schmidt.

Eroms, H.-W. 2000. Syntax der deutschen Sprache. Berlin: Walter de Gruyter.Fox, D. 2002. Antecedent-contained deletion and the copy theory of movement.Linguistic Inquiry 33/1, 63–96.

Fiengo, R. & May, R. 1994. Indices and identity. Cambridge: MIT Press.Groß, T. 1999. Theoretical foundations of dependency syntax. Munich: Iudicium.Groß, T. 2003. Dependency grammar�s limits – and ways of extending them.Dependency and valency: An international handbook of contemporary research,Vol. 1. eds. V. Agel et al., 331–351. Berlin: Walter de Gruyter.

Groß, T. & Osborne, T. 2009. Toward a practical DG theory of discontinuities.SKY Journal of Linguistics 22, 43–90.

Groß, T. 2010. Chains in syntax and morphology. Proceedings of the 24th PacificAsia conference on language, information and computation at Tohoku University,eds. O. Ryo, K. Ishikawa, H. Uemoto, K. Yoshimoto & Y. Harada, 143–152.Tokyo: Waseda University.

Haık, I. 1982–3. Indirect binding and referential circularity. Linguistic Review 2,313–330.

Haık, I. 1987. Bound VPs that need to be. Linguistics and Philosophy 10, 503–530.Harley, H. 2002. ACO, ACD, and QR of DPs. Linguistic Inquiry 33/4, 659–664.

124 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 32: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

Hays, D. 1964. Dependency theory: A formalism and some observations. Lan-guage 40, 511–525.

Hellwig, P. 2003. Dependency unification grammar. Dependency and valency: Aninternational handbook of contemporary research, Vol. 1. eds. V. Agel et al.,593–635. Berlin: Walter de Gruyter.

Heringer, H. J. 1996. Deutsche Syntax Dependentiell. Tubingen: Staufenberg.Higginbotham, J. 1983. Logical Form, binding and nominals. Linguistic Inquiry14/3, 395–420.

Higgenbotham, J. & May, R. 1981. Questions, quantifiers, and crossing. TheLinguistic Review 1/1, 41–80.

Hoeksema, J. & Napoli, D. J. 1990. A condition on circular catenae: a restate-ment of i-within-i. Journal of Linguistics 26/2, 403–424.

Hornstein, N. 1984. Logic as grammar. Cambridge, MA: MIT Press.Hornstein, N. 1994. An argument for Minimalism: The case of antecedent-contained deletion. Linguistic Inquiry 25/3, 455–480.

Hudson, R. 1984. Word Grammar. New York: Basil Blackwell.Hudson, R. 1990. An English word grammar. Oxford: Basil Blackwell.Hyvarinen, I. 2003. Der verbale Valenztrager. Dependency and valency: Aninternational handbook of contemporary research, Vol. 1. eds. V. Agel et al.,738–763. Berlin: Walter de Gruyter.

Jayaseelan, K. 1990. Incomplete VP deletion and gapping. Linguistic Analysis20, 64–81.

Johnson, K. 1987. Against the notion �SUBJECT�. Linguistic Inquiry 18, 354–61.Johnson, K. 2001. What VP ellipsis can do and what it can�t, but not why. Thehandbook of contemporary syntactic theory. ed. M. Baltin & C. Collins, 439–474. Oxford: Blackwell Publishers.

Johnson, K. 2009. Gapping is not VP-ellipsis. Linguistic Inquiry 40, 289–328.Jung, W.-Y. 1995. Syntaktische Relationen im Rahmen der Dependenzgrammatik.Hamburg: Buske.

Kahane, S. (ed.) 2000. Les grammaires de dependance (dependency grammars),Traitement automatique des langues 41. Paris: Hermes.

Kahane, S. 2003. The meaning-text theory. Dependency and valency: An inter-national handbook of contemporary research Vol. 1. eds. V. Agel et al., 546–569.Berlin: Walter de Gruyter.

Kennedy, C. 1997. Antecedent-contained deletion and the syntax of quantifica-tion. Linguistic Inquiry 28/4, 662–688.

Kennedy, C. 2008. Argument contained ellipsis. Topics in ellipsis, ed. K. John-son, 95–131. Cambridge: Cambridge University Press.

Kuno, S. 1981. The syntax of comparative clauses. Papers from the seventeenthregional meeting: Chicago Linguistics Society, eds. R. Hendrick, C. Masek &M.F. Miller, 136–155. Chicago Linguistic Society.

Kuno, S. 1987. Functional syntax: Anaphora, discourse and empathy. Chicago:University of Chicago Press.

Kunze, J. 1975. Abhangigkeitsgrammatik, Studia Grammatica 12. Berlin:Akademie Verlag.

Larson, R. & May, R. 1990. Antecedent containment or vacuous movement:reply to Baltin. Linguistic Inquiry 21, 103–122.

Lasnik, H. 1995. A note on pseudogapping. MIT Working Papers in Linguistics27.

Lasnik, H. 1999. Pseudogapping puzzles. Fragments: Studies in ellipsis andgapping, eds. E. Elabbas & S. Lappin, 141–174. Oxford: Oxford UniversityPress [Reprinted in Minimalist investigations in linguistic theory, HowardLasnik, 55–82. London: Routledge, 2003.]

Antecedent containment 125

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 33: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

Lebeaux, D. 1988. Language acquisition and the form of the grammar.University of Massachusetts, Amherst. PhD diss.

Lobin, H. 1993. Koordinationssyantax als prozedurales Phanomen. Studien zurdeutschen Grammatik 46. Tubingen: Narr.

Matthews, P. 1981. Syntax. Cambridge: Cambridge University Press.May, R. 1985. Logical form: Its structure and derivation. Cambridge, MA: MITPress.

Mel�cuk, I. 1988. Dependency syntax: Theory and practice. Albany: StateUniversity of New York Press.

Mel�cuk, I. 2003. Levels of dependency description: concepts and problems.Dependency and valency: An international handbook of contemporary research,Vol. 1. eds. V. Agel et al., 188–229. Berlin: Walter de Gruyter.

O�Grady, W. 1998. The syntax of idioms. Natural Language and LinguisticTheory 16, 79–312.

Osborne, T. 2005a. Beyond the constituent: A dependency grammar analysis ofcatenae. Folia Linguistica 39/3–4, 251–297.

Osborne, T. 2005b. Coherence: a dependency grammar analysis. Sky Journal ofLinguistics 18, 223–286.

Osborne, T. 2006. Shared material and grammar: a dependency grammar theoryof non- gapping coordination. Zeitschrift fur Sprachwissenschaft 25, 39–93.

Osborne, T. 2007. The weight of predicates: a dependency grammar analysis ofpredicate weight in German. Journal of Germanic Linguistics 19/1, 23–72.

Osborne, T. 2008. Major constituents and two dependency grammar constraintson sharing in coordination. Linguistics 46/6, 1109–1166.

Osborne, T., Putnam, M. & Groß, T. Catenae: Introducing a novel unit ofsyntactic analysis. Syntax (in press).

Pickering, M. & Barry, G. 1993. Dependency Categorial Grammar and coor-dination. Linguistics 31, 855–902.

Robinson, J. 1970. Dependency structures and transformational rules. Language46, 259–285.

Pollard, C. & Sag, I. 1992. Reflexives in English and the scope of bindingtheory. Linguistic Inquiry 23, 261–303.

Pollard, C. & Sag, I. 1994. Head-driven phrase structure grammar. University ofChicago Press.

Reinhart, T. & Reuland, E. 1993. Reflexivity. Linguistic Inquiry 24, 657–720.Ross, J. 1970. Gapping and the order of constituents. Progress in linguistics: Acollection of papers, eds. M. Bierwisch & K. Heidolph, 249–259. The Hague:Mouton.

Sag, I. 1976. Deletion and logical form. MIT. PhD diss.Schubert, K. 1988. Metataxis: Contrastive dependency syntax for machinetranslation. Dordrecht: Foris.

Sgall, P., Hajicova, E. & Panevova, J. 1986. The meaning of the sentence in itssemantic and pragmatic aspects. Dordrecht: D. Reidel.

Starosta, S. 1988. The case for lexicase: An outline of lexicase grammaticaltheory. New York: Pinter Publishers.

Starosta, S. 2003. Lexicase grammar. Dependency and valency: An internationalhandbook of contemporary research, Vol. 1. eds. V. Agel et al., 526–545. Berlin:Walter de Gruyter.

Tarvainen, K. 2000. Einfuhrung in die Dependenzgrammatik. Tubingen:Niemeyer.

Tesniere, L. 1959. Elements de syntaxe structurale. Paris: Klincksieck.Tesniere, L. 1969. Elements de syntaxe structurale, 2nd edition. Paris:Klincksieck.

126 Timothy Osborne & Thomas Groß

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.

Page 34: Antecedent Containment: A Dependency Grammar Solution in Terms of Catenae

Uzonyi, P. 2003. Dependenzstruktur und Konstituenzstruktur. Dependency andvalency: An international handbook of contemporary research, Vol. 1. eds.V. Agel et al., 230–246. Berlin: Walter de Gruyter.

Van Langendonck, W. 1994. Determiners as heads? Cognitive Linguistics 5, 243–259.

Weber, H. 1997. Dependenzgrammatik: Ein Arbeitsbuch. Tubingen: Narr.Wilder, C., 2003. Antecedent containment and ellipsis. The interfaces: Derivingand interpreting omitted structures, eds. K. Schwabe & S. Winkler, 79–119.Amsterdam: John Benjamins.

Williams, E. 1982. The NP cycle. Linguistic Inquiry 13/2, 277–295.Williams, E. 1983. Semantic and syntactic categories. Linguistics and Philosophy6, 423–446.

Received September 12, 2010Accepted November 30, 2011

Timothy Osborne615 6th Str. Apt 110Kirkland, WA 98033

[email protected]

Thomas GroßEspaul Yayoi 501Azanakahara 38-1

Yayoi-cho, Toyohashi-shiAichi-Ken441-8106

[email protected]

Antecedent containment 127

� The authors 2012. Studia Linguistica � The Editorial Board of Studia Linguistica 2012.