Binding and Coreference: Views from Child Language

  • Cornelia Hamann
Part of the Studies in Theoretical Psycholinguistics book series (SITP, volume 41)


The chapter reviews work on a central topic in acquisition from the perspective of generative grammar: the Binding principles that dictate how pronouns and reflexives behave. The core issue is the “Pronoun Interpretation Problem (PIP)”: do children actually know Principle B of binding and their knowledge is masked in performance, or is there a real problem with pronoun interpretation that may require integrating syntax and pragmatics? Hamann provides a summary of recent theoretical and empirical work on binding that makes it clear that there is more involved in interpretation of pronouns than the simple binding principles. The cross –linguistic asymmetries are reviewed, since pronominal clitics in Romance languages are found to be better understood than non-clitic pronouns, but not in the case of clitic climbing (ECM) environments. Yet recent results on successful performance of children acquiring German belie any simple account of the PIP. Various theorists have proposed coreference rules that require consideration of pragmatics in one way or another, and the interaction of the principles with discourse antecedents. Others stress the possibility of an underspecification of the features of the pronoun in acquisition. Recent work considering the roles of referentiality and topic-hood is explored.


Simple Sentence Scalar Implicature Romance Language Romance Child Open Proposition 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

The Interaction of Theoretical and Empirical Research

The interpretative properties of pronominal systems have been the focus of attention in theoretical and empirical investigations for several decades now. Research has focused on the description of such systems in different languages aiming to establish the universals and the differences. At the same time, the interpretation of pronouns served as one of the most quoted arguments for the poverty of the stimulus, which led to a fruitful interaction of theoretical developments and empirical research on the acquisition of pronominal systems.

Soon after the formulation of the Binding Theory in the Lectures on Government and Binding (Chomsky 1981) constraining the interpretation of anaphors/reflexives (Principle A), pronouns (Principle B), and of lexical noun phrases (Principle C), research on young children’s comprehension of these elements was intensified. Jakubowicz (1984, 1989), Wexler and Chien (1985), and Crain and McKee (1985) showed that English speaking children almost always do well when interpreting pronouns and lexical nominal expressions in sentences like He washed Luke Skywalker or reflexives in sentences such as Luke Skywalker washed himself. However, in sentences such as Luke Skywalker washed him children interpreted pronouns correctly in only about half of the cases allowing the pronoun to refer to Luke Skywalker in the other half. Since then it has been widely accepted that there is a difficulty with Principle B, which was called the “Delay of Principle B Effect – DPBE”.

Children’s difficulty with pronouns as opposed to reflexives and referential expressions posed a serious problem since the structural constraints in all three cases involve essentially the same mechanisms, c-command and a locality constraint. How could children show knowledge of the constraints in sentences with reflexives and at the same time not know them in a sentence with a pronoun? The phenomenon was all the more puzzling as English children do well with pronouns in production, so that it seemed problematic to assume that the feature specification of pronouns is not mastered.

When Chien and Wexler (1990) reported on their Experiment 4, the problem appeared to be solved: with the picture version of a Truth Value Judgment Task, Experiment 4 showed that children only rarely interpreted sentences such as Every bear is touching her as every bear touching herself, whereas they were only about 50% successful with sentences such as Mama Bear is touching her. This result indicated that children indeed know Principle B: in case of a “misapplication” a quantified antecedent allows only a bound variable reading of the pronoun. Since this reading was not assigned, the principle had been applied correctly.

For acquisition, the result meant that Principle B is in fact acquired at the same time as the other two principles and what remained to be shown was which factors mask its application. Chien and Wexler (1990) and others suggested that a pragmatic principle (disallowing a pronoun to corefer with an antecedent in most contexts) has not yet been acquired, whereas Grodzinsky and Reinhart (1993) pointed to processing limitations aiming for a unified account of the difficulties shown by children and patients with agrammatic aphasia. The result of Chien and Wexler’s (1990) Experiment 4 has since been replicated in other languages, so that the label DPBE seems inappropriate. Arguments for abandoning the term have also been supplied in recent research focusing on the asymmetry in comprehension and production found in Dutch and English (Hendriks and Spenader 2006; de Villiers et al. 2006) or on cross-linguistic research with aphasic patients. Therefore the term DPBE has been replaced by the term “Pronoun Interpretation Problem – PIP”, see also Baauw and Cuetos (2003).

During the first decade of intensive research on binding, theoretical research was stimulated by the possibilities emerging from acquisition results, namely that there may be two mechanisms establishing the reference of anaphoric elements, bound variable and coreference, and by the division of labor between syntactic and pragmatic constraints suggested by this. On the empirical level, a pragmatic or processing explanation was challenged by cross-linguistic results. Using the same experimental paradigm for both languages, McKee (1992) corroborated Chien and Wexler’s result for English, but showed that Italian children of the same age do not show a PIP. The same holds for French and Spanish, see Jakubowicz (1989), Hamann et al. (1997), Hamann (2002), Baauw et al. (1997), Baauw (2000, 2002); McKee (1992) argued that the pragmatics of pronouns is the same cross-linguistically and that processing limitations should be equal in children of the same age. Therefore the early mastery of pronoun interpretation in Romance languages must be due to the fact that they are clitics. Since Romance clitics occur high in the clause and are referentially deficient in ways that Germanic pronouns are not, structural/derivational accounts of the good performance of Romance children were suggested (McKee 1992; Baauw 2000; Baauw and Cuetos 2003). However, pragmatic explanations have also been proposed (Avrutin and Wexler 1992; Hamann 2002; Hendriks et al. 2007).

Many cross-linguistic investigations of pronoun interpretation also addressed ECM-constructions as in The girl sees her dance. These constructions are of theoretical interest because different versions of the Binding Principles make different predictions. Under certain formulations of “local domain” the pronoun falls under principle B in the Standard Theory, whereas the Reflexivity Framework predicts that Principle B does not apply here, and coreference is ruled out through a general condition on chain formation. An experiment run by Philip and Coopmans (1996) showed not much difference in simple and ECM constructions in English but a clear difference in Dutch. Interestingly, also Romance children (and agrammatics) show a PIP in ECM constructions (Hamann et al. 1997; Baauw et al. 1997; Baauw and Cuetos 2003). Again, two basically different lines of argumentation have been employed in order to explain this effect. Researchers focusing on the syntactic side are Philip and Coopmans (1996), Baauw (2000), Baauw and Cuetos (2003), whereas Hamann (2002) favors a pragmatic account that can be uniformly applied to the ­different contexts.

So the interesting questions concerning acquisition of pronominal systems are: In which circumstances can a production/comprehension asymmetry be observed? What masks children’s knowledge in the case of Germanic pronouns? Why does the PIP not show up in languages that have clitic pronouns of the Romance type? Why is the PIP stronger in ECM contexts than in simple sentences in some languages? Why does the PIP show up only in ECM contexts in Romance languages? Why can the PIP be observed also in agrammatic patients in languages where children show it? Another question always accompanying the empirical research is which of the current versions of the binding theory the results support.

Recently, studies have focused on problems raised by the earlier results and their different explanations. If it plays a role that Germanic pronouns are ambiguous in that weak and strong pronouns coincide in form, then it is expected that in ­languages that separate the forms, the strong forms will show the PIP whereas the weak or clitic forms will not. Varlakosta (2000) and Baauw and Cuetos (2003) address this problem with different results. Moreover, in languages having a ­pronoun system very similar to English or Dutch, it is expected that children will show the PIP, but the surprising results of two recent studies on German (Ruigendijk 2008; Hamann and Ruigendijk 2009) show that German children behave like French or Spanish children in simple and ECM contexts.

One way of dealing with the problems of binding theory is integrating syntax and pragmatics into one rule system as was proposed by Burzio (1998) and has been further developed in Optimality Theory, see Fischer (2004). Focusing on the asymmetry in comprehension and production found in Dutch and English, where production is more or less intact, Hendriks and Spenader (2006) and de Villiers et al. (2006) suggest such accounts for child language. Hendriks et al. (2007) couch their analysis in a bidirectional Optimality framework using a constraint ranking for the hearer and a possibly different one for the speaker (see Jäger 2002). The idea is that in comprehension children neglect the corrective speaker’s perspective when matching a form to a meaning, while only the speaker’s perspective matters and no correction in ranking is needed in production. This account remains speculative as to why children cannot do bidirectional ranking. Since Theory of Mind is long acquired, as de Villiers et al. (2006) point out, an appeal to processing limitations comes to mind. Optimality Theory accounts thus allow integrating a rather classic syntactic analysis with well discussed pragmatic notions and processing ideas in order to explain asymmetries in comprehension and production.1

In other recent studies, the results obtained in earlier experiments, the basis for much theorizing, were criticized for methodological reasons. Especially the asymmetry found for quantified and lexical antecedents has been called into question in a careful analysis by Elbourne (2005). His analysis aims to show that this asymmetry is an artifact of the experimental method. Therefore, Elbourne (2005) ­concludes, there is no evidence in English that Principle B is acquired. Conroy et al. (2009) demonstrate the opposite: If the factors singled out as problematic by Elbourne (2005) are properly controlled for, then children show knowledge of Principle B in simple sentences. Other studies pointing to the production/­comprehension problem and using controlled context conditions also contribute to this discussion (see de Villiers et al. 2006). Finally, a recent study by Verbuk and Roeper (2010) concentrates on the question of how children can assign the relevant referentiality features to the pronominal elements they find in their language given that not all languages have a simple distinction into pronouns and reflexives and given that there is ambiguous input in English. This approach is in the tradition of accounting for cross-linguistic variation through parameters that differ with respect to the feature specification of certain lexical items. In this case the task is to select the specific lexical items that are reflexives/anaphors and pronouns and to determine their domain.

Aim and Structure of the Article

This article concentrates on the developments and problems of the theoretical ­framework and on questions arising from the cross-linguistic acquisition results on pronoun interpretation in different constructions. Section 2 is focused on theoretical considerations: the Standard Binding theory (Sect. 2.1) and its problems (Sect. 2.2), the distinction of reference assignment through a bound variable configuration and through coreference (Sect. 2.3), the typology of anaphors and the cross-linguistic variation in the types of anaphors (Sect. 2.4) as well as newer theoretical ­developments (Sect. 2.5). The article then addresses the acquisition results in detail, ­discussing the classical results on the asymmetries of pronouns and reflexives or of pronouns with lexical and quantified antecedents as well as the accounts given at the time in Sect. 3. Section 4 addresses the findings on ECM constructions and their explanations. More cross-linguistic results will be examined in Sect. 5 dealing with Romance languages, but also with new results from Modern Greek and German. Section 6 introduces the recent criticism as to the methodological problems in earlier experiments and shows how this critical work has already influenced approaches to acquisition. The conclusion gives a summary and outlines a possibility to integrate recent findings on acquisition with findings on agrammatic patients.

2 Theoretical Background

The Standard Binding Theory

The basic observation about pronominal elements is that they are different from other nominal elements such as names or definite descriptions (lexical DPs) in that they are referentially deficient. They are anaphoric elements receiving their ultimate interpretation through their antecedent. The interpretation of such elements therefore turns on the identification of their antecedent. Here the feature specification of the pronominal element (person, number, gender, case) plays a decisive role. Additionally, the traditional distinction of pronominal elements into reflexives and pronouns captures two fundamental possibilities of antecedent identification: the antecedent can be within the clause or it can be a salient element in the discourse. The interpretations of reflexives, pronouns and referring expressions are shown in (1a,b,c) using the conventions of formal semantics and the GB framework: coindexed terms are interpreted as referring to the same individual and contraindexed terms usually refer to different individuals since it holds that if the index of A is different from the index of B, then neither can be the antecedent of the other.2



Johni saw himselfi/*himselfj/*herselfi


Johni saw himj/*himi


Johni /hei saw Johnj/*i

If the examples in (1a,b,c) look straight forward, the examples in (2a,b,c,d) show that it is not easy to determine in which cases a pronoun (and a referential expression) can or cannot corefer with a possible antecedent. Reinhart (1983) showed that notions of linearity are irrelevant and introduced a structural relationship, the notion of c-command.



Johni says that he i is tired


*Hei says that Johni is tired


When hei was arrested, Johni was with his wife


The people who know himi well say that Johni is very intelligent

C-command is defined as given in (3), and the kind of referential identity achieved through coindexation in c-command relations is called (syntactic) binding; see (4) for the definition.


x c-commands y iff x does not contain y and the first branching node dominating x also dominates y


x binds y iff x c-commands and is coindexed with y

In the examples (1a,b) a certain complementarity of reflexives (called anaphors in GB-theory) and pronouns appears, also noticeable in (5a,b,c). It also appears that anaphors have an antecedent within the clause, whereas pronouns do not. Investigating examples like (5a,b,c), it emerges that the relevant notion is not the clause, but a local domain that nevertheless allows some distance between the coindexed elements (see the ungrammaticality of (5d)).



John hated Maryi’s pictures of herselfi


Johni hated Mary’s pictures of himi


*John hated Maryi’s pictures of heri


*Johni hated pictures of himi

Given the definitions in (3) and (4) and the fact that locality plays a role, possible antecedents for the different types of nominal expressions as illustrated in (1), (2) and (5) have been constrained in the Standard Binding Theory (Chomsky 1981) spelled out in (6).



Principle A:

An anaphor is bound in a local domain.


Principle B:

A pronoun is free in a local domain.


Principle C:

An R-expression is free.

The notion of the “local domain of x” is usually made precise as the governing category of x.3 See Chomsky (1981), Manzini and Wexler (1987), Reuland and Everaert (2000) for a discussion.

The principles given in (6) successfully describe “recurrent patterns in the ­languages of the world” (Reuland and Everaert 2000: 641) and capture the basic complementarity of pronouns and reflexives. So they provide a good enough ­working model of the referential properties of pronominal elements, even though it was clear from the beginning that there were problems with these purely syntactic ­constraints on interpretation.

Problems with the Standard Binding Theory

Already Ross (1982) pointed out that in English sentences like (7) the pronoun and the reflexive are not in complementary distribution but are both licit in a situation where the gun is near James Bond. Postulating that Principle B constrains only semantic co-arguments of a predicate, not adjuncts, solves this problem. This has consequences for sentences like (8), however, where Max and the pronoun are not semantic co-arguments. So the co-indexation in (8) and in other Exceptional Case Marking (ECM) constructions must be ruled out by a constraint different from Principle B.


James Bond noticed the gun near him/himself.


* Maxi heard himi criticised.

More problems emerge in the discussion of so-called picture noun anaphors (Pollard and Sag 1992; Reinhart and Reuland 1993) which freely take non-local antecedents as exemplified in (9). This type of anaphor appears to be exempt from the binding constraints, so that Reinhart and Reuland (1993) call the type of ­anaphor in (9) “logophoric” because it has to do with point of view or perspective. The same observation about the impact of point of view and emphasis has been made for other contexts in which anaphors are free in their local domain as in (10). Reuland and Everaert (2000: 643) and the authors quoted there provide a detailed discussion of this problem.


Johni remembered that Aunt Sally had kept a picture of himselfi in the attic.


There were three linguists in the room apart from myself.

Another possibility to deal with the questions raised by the examples (7), (8), (9) and (10) is outlined in Levinson (2000) and Verbuk and Roeper (2010). In their accounts, scalar implicatures and discourse context regulate the possible interpretations (see also Sect. 2.3). Similar mechanisms relying on a markedness scale for pronominal expressions are built into the constraint rankings suggested in OT ­treatments of binding.

Other problematic contexts are the ones in (11a,b) and (12) where pronouns have the same referent as a local antecedent. Such contexts have been extensively ­discussed by Evans (1980), Grodzinsky and Reinhart (1993), by Heim (1993) and Reinhart (2004), who develop an explicit, rigorous formal semantic/pragmatic ­system for such cases.



If everyone hates Oscari, then Oscari hates himi/*himselfi


I dreamt I was Mel Gibson and kissed me/*myself.


A: Is that speaker Zelda? – B: She must be. She praises her to the sky.

Binding vs. Coreference

Since the examples (11a,b) and (12) exemplify contexts where pronouns can have the same referents as their antecedents, one way of dealing with the problem is to assume that there are two mechanisms resulting in a mapping to the same referent, coindexation (which in a c-command configuration means binding) or coreference. Coreference can arise because in semantics it is the index of a nominal expression to which an individual is assigned as its interpretation. So, departing from the assumption of disjoint reference in such cases, the same individual could be accidentally assigned as referent to two different indices. Note that Principle B is not violated in such cases, since different indices have been marked on the pronoun and the antecedent.

These two different mechanisms had been discussed already by Sag (1976), Williams (1977), Evans (1980), and Higginbotham (1983) in connection with possessive pronouns in cases of VP-ellipsis like (13). In (13) the pronoun can refer to Al or to somebody else outside the sentence. Even if one disregards a discourse referent, the sentence is ambiguous due to the ellipsis in the second conjunct. It has the reading in (13a) where Bill also loves Al’s sister (strict reading) and the reading in (13b) where Bill loves his own sister (sloppy reading). In (13b) the pronouns are coindexed with an antecedent leading to a binding relation for each pronoun. In (13a) the referent of the pronoun has been fixed and happens to be the same in both cases – a mechanism which is also necessary in case the referent is a person salient in the discourse. Clearly, the interpretation (13a) involves a form of coreference (or linking – Higginbotham 1983).


Al loves his sister and Bill does too


Ali loves hisi sister and Billj loves hisi sister too


Ali loves hisi sister and Billj loves hisj sister too

Under the assumption that there are two mechanisms, rules are required that allow coreference in some contexts, force coreference precisely in the contexts of examples (11a,b) or (12) and rule it out in run of the mill contexts. Such rules restricting contexts are not constraining syntax but are pragmatic in nature. Their introduction therefore marks a break with the standard binding theory in assigning part of the work to syntax and part of the work to pragmatics, an approach which has been taken further in accounts integrating syntactic and pragmatic constraints into one (Optimality) constraint hierarchy (Burzio 1998; Fischer 2004) or in claims that all the work should be left to pragmatics, see Levinson (1987, 2000).

Chien and Wexler (1990) introduced one such pragmatic rule, their Rule P, given as (14). Rule P does not attempt to define the contexts, but is nevertheless a pragmatic rule.


Contraindexed NPs are noncoreferential unless the context explicitly forces coreference.

Another rule which can be interpreted as pragmatic in origin, namely as a special case of Grice’s Quantity maxim is the coreference rule proposed by Grodzinsky and Reinhart (1993), here paraphrased (and simplified) as (15). The rule essentially says that a pronoun cannot be used if replacing the pronoun with a reflexive leads to the same interpretation.


Rule I: Intrasentential Coreference

α cannot corefer with β if an indistinguishable interpretation can be generated by replacing α with a variable bound by the trace of β.

Rule I and the variant introduced by Heim (1993) (for discussion see Hamann 2002) allows coreference in (11a,b). If the pronoun were replaced by a reflexive in (11a), then a different interpretation would ensue, namely “If everyone hates Oscar then Oscar hates himself”. Whereas (11a) can be used to explain the rule of universal instantiation, the latter cannot – because the inference now is about self-hating, not about hating Oscar. The same reasoning applies to (11b), the dream is about Mel Gibson kissing the speaker, not about Mel Gibson kissing himself. Heim (1993) identifies these contexts as those where “structured meaning matters” so that the use of a pronoun is a licit expression of coreference. In contrast, in a run of the mill context as in (1a), coreference of the pronoun and the antecedent is excluded because a coreferential reading could not be distinguished from the reading obtained by replacing the pronoun with a reflexive.

Verbuk and Roeper (2010) also demonstrate how coreference and disjoint reference readings can be derived. They follow Levinson (2000) in using the Horn scale  <  reflexive, pronoun  >  where the reflexive is encoding the stronger dependency. This Horn scale allows deriving Rule I as a special scalar implicature. Once a child has acquired the properties of the different elements of the pronominal system found in her ambient language and discovered a certain complementarity, then the Horn scale can be established, identifying the reflexive as the element stronger in referential dependency and, at the same time, Principle B can be assumed to be operative in the language. In normal Principle B contexts the Q-maxim allows to calculate that the stronger term, the reflexive, should have been used if coreference had been intended. Since the pronoun has been used, it can be concluded that disjoint reference is meant. In the cases where pronouns allow coreference, this implicature has to be cancelled. Verbuk and Roeper (2010) following Ward (1983), point out that an open proposition must be computed in order to allow cancellation. In the case of (11a) this is x hates Oscar. Because of the open proposition the Q-maxim is ­cancelled and instead an I-implicature based on informativeness is calculated. Verbuk and Roeper (2010) formulate a Discourse Condition on Exceptional Coreference Contexts given in (16). In this condition the open proposition is the crucial factor in forcing coreference in contexts like (11a,b), whereas in contexts like (7) or (9) and (10) coreference is possible, but not forced because no open proposition has to be computed. In such cases, the context may fix the referent as part of the common ground for speaker and hearer or it may not have fixed the referent. In the former situation, using the reflexive would change the meaning and thus coreference is ­possible, whereas in the latter case, disjoint reference is forced by the Q-implicatures since the stronger anaphor has not been used (see Verbuk and Roeper 2010: 58).


Discourse Condition on Exceptional Coreference Contexts

A pronoun and a non-quantificational antecedent that c-commands it must corefer if the Common Ground contains an Open Proposition that fixes the referent of the pronoun to the same referent as that of the c-commanding antecedent.

So far, nothing has been said about the mechanism allowing coreference in (12). For such identity debate contexts, Heim (1993) points to the fact that the identification of the referent is mediated by the sense (intention) of the definite description. So we have the name, Zelda, which gives access to the identification of the individual we know as Zelda, and we have “the speaker”, which identifies Zelda in one of her roles. In this particular example there is also a deictic component (that) identifying the speaker in the situation through a pointing gesture. Heim (1993) uses the term “guise” for this phenomenon and allows coreference of contraindexed pronouns and antecedents if the pronoun and the antecedent are identified via different guises, here “the speaker” and “Zelda”. A modification of the coreference rule such that it takes into account guises will explain (12). Note that “guise” often involves the well-known “intension” involved in such famous examples as (17). Here coreference is asserted and this is not a tautology because the same referent is presented in two guises. “Guises” also arise, however, through direct reference assignment in the case of names or through deixis, the pointing gesture used to identify referents.


The morning star is the evening star

Summing up, we have identified two contexts where accidental coreference is the rule: One is a context where structured meaning matters and the other is a context where different guises are under debate. Presumably, these are the contexts which “force coreference” in Rule P from Chien and Wexler (1990). Note here that the structured meaning cases in (11a,b) in Heim’s (1993) model do not involve ­different guises, but can be derived straightforwardly from Rule I. In Verbuk and Roeper’s account, the identity contexts involve an open proposition, a question under discussion, present in the situational context, and so (16) forces coreference also in such a case.

The problem cases discussed so far led to a distinction of the cases of bound anaphora, which fall under binding theory, and cases of coreference, which can be allowed or excluded by pragmatic rules. Clearly, the interpretation of pronominal elements is not only regulated by the structural binding principles: for pronominals the identification of coreference and disjoint reference is subject to pragmatic computations.

Reflexivity and the Typology of Anaphors

More problems emerged when cross-linguistic investigation showed that many ­languages do not have clean-cut pronominal systems: pronouns on one side and anaphors on the other (Zribi-Hertz 1989). Some languages, e.g. Dutch and German have a tri-partite system with pronouns and two distinct kinds of anaphors, zich and zichself (Reinhart and Reuland 1993). In other languages, e.g. Frisian (Reuland 1994), Old and Middle English (Faltz 1985; Van Gelderen 2000), elements that are clearly pronouns do double duty as reflexives. They can be used in contexts where Dutch uses the pronoun or zich. Still other languages (Norwegian, Icelandic) have elements that are clearly anaphoric but are not necessarily locally bound, note especially the extensive discussion of the Icelandic long-distance anaphor sig, first described by Thráinsson (1976). Some of these problems can be addressed by ­postulating a parameterization. This can be made explicit by parameterizing the domain of an anaphor as suggested by Manzini and Wexler (1987), or through the feature specification of certain lexical elements in binding “parameters” as ­suggested by Elbourne (2005) for Frisian and older stages of English and further ­developed by Verbuk and Roeper (2010).

The Reflexivity framework proposed by Reinhart and Reuland (1993), and developed in Reuland and Everaert (2000), Reuland (2001, 2011) separates coreference from the cases of bound anaphora and integrates the idea of markedness in a typology of anaphors. In these accounts two components are responsible for the distri­bution of pronouns and anaphors: reflexivization, as the marked option, is defined over predicates and the configurational effects normally derived from the Binding Principles are obtained via chain formation. The reformulation of the binding principles in terms of reflexivity and reflexive marking as in (19) in turn relies on a typology of anaphoric expressions replacing the simple distinction into pronouns and anaphors. Nominal expressions are partitioned into three or four classes by the properties [SELF] and [R], where the property [SELF] describes the ability to reflexivize the predicate and the property [R] captures referential independence (Reinhart and Reuland 1993). In earlier versions of the Reflexivity framework only the first three types of pronominal elements given in (18) are listed, in Reuland (2008) we additionally find the type PRON/SELF relevant for Modern Greek.








(Dutch zich)



Reflexivizing function



R(eferential independence)




Condition A: a reflexive-marked syntactic predicate must be interpreted reflexively.

Condition B: a reflexive semantic predicate must be reflexive-marked.

Reflexive marking can be achieved in the lexicon (inherently reflexive predicates like shave and wash), by adding a -self suffix as in the case of English, or by ­choosing the lexically reflexive marked pronoun as in the case of zich in Dutch and se in French, see (20) for a definition. The definitions of syntactic and semantic ­predicates are given in (21), here following Reuland and Everaert (2000).



A predicate is reflexive iff at least two of its arguments are coindexed


A predicate (formed of P) is reflexive marked iff either P is lexically reflexive with respect to an indexed argument, or one of P’s indexed arguments is a SELF-anaphor.



The syntactic predicate of (a head) P is P, all its syntactic arguments, and an external argument of P (subject). The syntactic argument of P are the projections assigned theta-role or Case by P.


The semantic predicate of P is P and its arguments at the relevant semantic level.

The conditions (19) and the definitions (20) and (21) talk explicitly only about coindexing and its interpretation. However, Condition B has much the same effect as the standard Principle B. It works only by exclusion and does not apply if the semantic argument of a predicate is not the pronoun but a clause as in (8). Therefore Condition B is restricted to semantic predicates, whereas Condition A concerns syntactic predicates. In order to rule out (8), the other ingredient of this theory, chain formation is appealed to. (22) gives the definition of an A-chain used by Reinhart and Reuland (1993), see also the modifications in Reuland and Everaert (2000).


Any sequence of coindexation that is headed by an A-position and satisfies antecedent government is an A-chain (Reinhart and Reuland 1993: 693,4).

This definition subsumes trace-tailed coindexations (movement chains) and chains arising through coindexation for purposes of interpretation/reflexivization. In fact, it integrates both types of chains and treats a sequence of indices arising through a combination of movement and interpretative coindexation as an A-chain.4 The crucial condition on such general chains is formulated in (23).


General Condition on A-chains

A maximal A-chain (α1,....,αn) contains exactly one link – α1- that is both  +  referential [+R] and Case-marked. (Reinhart and Reuland 1993: 696)

Expressions that are [+R] carry a “full specification for ϕ-features and structural case” (Reinhart and Reuland 1993: 697). That this is only a necessary condition for referentiality not a sufficient one, however, can be demonstrated by English himself and him which have the same ϕ- and case features. The pronoun is [+R], however, the anaphor is [−R]. It is generally held that [−R] expressions are “referentially defective…which entails that they cannot be used as demonstratives, referring to some entity in the world” (Reinhart and Reuland 1993: 658).

Coming back to example (8) and the problem of how to exclude it if Condition B only applies to semantic predicates, we see that it is the classification of English him as [+R] together with the Chain Condition that disallows coindexation. Coindexation of him and Max would violate the Chain condition since two elements would be specified as [+R]. Note that the general conditions on chain formation include the configurational information so important for the standard binding theory, namely that one of the coindexed elements c-commands the other one. In this manner a major ingredient of the binding theory can be reduced to general properties of the computational system.

Binding Theory, Minimalism and Other Recent Developments

The modifications of the binding theory discussed so far have introduced a pragmatic component and have derived the basic structural constraint operative in the principles from more general constraints on chains. More radical changes are necessary, however, if the new approach to the nature of syntactic computations outlined in the Minimalist Program (Chomsky 1995) is to be applied. Without going into the details here, it is argued that the Binding Principles are not well-formedness conditions that apply at S-Structure/Spell-Out but are interface conditions holding at the LF interface. So Chomsky (1995: 211) proposes a “very simple interpretive version of binding theory”. With D being the “relevant” local domain, the new formulation is given in (24):



If α is an anaphor, interpret it as coreferential with a c-commanding phrase in D.


If α is a pronominal, interpret it as disjoint from every c-commanding phrase in D.


If α is an r-expression, interpret it as disjoint from every c-commanding phrase

It is clear that these principles, even though they are now interface conditions, do not integrate some of the insights discussed here. The possibility of accidental coreference is ignored together with the typology of anaphors obvious in languages that do not have a bipartite system like English. Another major change is that in the computational system indices can be abandoned, which is a welcome consequence for Chomsky. Indices are semantic in nature in that they encode the assignment of referents, and removing them from the computational system is another step towards the “autonomy of syntax”, a desirable aim, at least in Chomsky’s program.

The unfortunate consequence is that the (referential) relations they express must be replaced “by a structural account of the relation they annotate” Chomsky (1995: 217, footnote 53). It is at this point that former simplicity is replaced by not very intuitive, complex construction. Reuland and Everaert (2000: 666) propose to ­construct the relation stepwise through the existing morpho-syntactic relations: “an object anaphor is linked to the verb and the verb’s inflectional system by structural Case, the verb is linked to its functional system, and the functional V-system is linked to the subject.” Another conceivable approach to indices closer connected to exploring the “relation they annotate”, see above, would be to define Agree relations via the features [+/− reflexivising] and [+/− referential] which hold in the usual structural configurations. See Reuland (2011) for developments of such approaches. But see also Szabolcsi (2001) and Beghelli and Stowell (1997) who discuss the referentiality feature and a Referentiality Phrase in an attempt to eliminate the ­operation of quantifier raising. This treatment suggests that such a feature is necessary for other reasons anyway and exploring it for binding relations might turn out to be quite straightforward.

Apart from these difficulties and the remaining problems with accidental coreference, types of anaphors etc., shifting the binding conditions to the interpretative interface has a certain appeal.5 It enables a new division of labor: conditions can apply at the lexical level, at the syntactic level and at the interfaces. So the necessary pragmatic rules can be easily conceived as interface conditions at the next level: the semantic-pragmatic interface. This shift also forces a new look at the interaction of conditions working at different levels.

Research on the interaction of different types of constraints with the notion of economy of derivation (and processing) have led to the assumption that it is cheaper to be able to treat a phenomenon, here encoding a dependency, in the narrow syntax than having to use an interface rule which brings into play other cognitive domains. Reuland (2001) proposes that derivation in narrow syntax is more economical than derivation at logical syntax, which in turn is more economical than having to integrate contextual factors from discourse. Extending this to situational context, which under this account is even harder to integrate into the derivation, yields the hierarchy (25) for referential identification (adapted from Grillo 2008: 48).




Narrow Syntax

feature checking (chain formation)

Semantics (logical syntax)

bound variable

Discourse context


Non-linguistic context


The effect of the economy hierarchy is that certain types of dependencies are possible only if no cheaper derivation is available. So bound variable chains, which involve cross-modular operations, are more costly than chains established through feature checking at narrow syntax, and establishing coreference, which necessarily involves the discourse level is more costly still. Note that this hierarchy, Burzio’s (1998) constraint on Referential Economy and the Horn scale employed by Verbuk and Roeper (2010) derive very similar effects. In the adult grammar a sentence with a pronoun cannot (in a normal context) indicate coreference since in that case ­syntactic binding, i.e. a reflexive, would be preferred because it is the cheaper option. For explaining acquisition facts, it has to be assumed that the hierarchy is not in place yet. In particular, narrow syntax is not the cheapest option for the child, perhaps because full automatization or the step from the particular to the general is not achieved yet. This means that the use of a pronoun does not necessarily imply disjoint reference since the use of a reflexive might not be cheaper for the child.6 Such approaches to the economy of derivation can be tied in remarkably well with the idea of different classes of features instrumental in defining constraints on chain formation in narrow syntax, especially recent refinements of Relativized Minimality (Rizzi 1990, 2004). Since reflexivity could be derived by head-movement (see Reuland 2011) and Romance clitic pronouns are involved in both types of chains (Hamann 2002) such restrictions, resting on a distinction of argumental, quantificational, modifying and topic-features might be interacting with the hierarchy sketched above.

As to the acquisition of the Binding Principles, theoretical advances developed from a detailed examination of problem cases have influenced empirical research not only in hypothesis formation but also in experimental designs and, naturally, in the interpretation of the results. Especially the distinction of binding and ­coreference and the development of the reflexivity framework have proved fruitful for acquisition research.

3 Children’s Problems with Pronouns

The Pronoun/Reflexive Asymmetry

Wexler and Chien (1985) and Chien and Wexler (1990) showed that English ­children give adult responses to sentences containing reflexives in 80–90% of the cases between the age of 5;6–6;6. Performance on sentences with pronouns were much worse, however, showing close to chance performance in many cases. This asymmetry has been replicated for English and other languages in many studies (Deutsch et al. 1986; Grimshaw and Rosen 1990; McDaniel et al. 1990; Sigurjónsdóttir and Hyams 1992; Avrutin and Wexler 1992). Only Kaufmann’s (1988) results indicated that children are about as good with pronouns as with reflexives.

Chien and Wexler (1990) give an overview over a series of experiments showing that English children respect locality constraints and know c-command. Interestingly, their experiments demonstrate a slow acquisition of anaphors and Principle A: looking at the experiments 1–3, only 13% local readings are assigned to the anaphors at age 2;6–3;0 and there is a steady increase till at age 6;0–6;6 children choose the local antecedent in 89.4% of the cases. In the same experiments, the interpretation of pronouns was only 64% adult like at the age of 6;0–6;6, giving rise to the misnomer DPBE as discussed above.

Possible explanations were that children may not have yet determined the local domain for pronouns (see McKee 1992) or, advocated by Chien and Wexler themselves, that they have not mastered a pragmatic principle that governs coreference and so is relevant for pronouns, not for reflexives. This lead to a different set of experiments probing for the effects of accidental coreference.

The Quantified Antecedent/Simple Antecedent Asymmetry

As has been sketched in the introduction important issues are related to Chien and Wexler’s (1990) finding that there is an asymmetry in children’s mastery of contexts with lexical antecedents and with quantified antecedents: the claim that pronominal reference is regulated by two mechanisms, syntactic binding and pragmatic coreference, and the claim that the binding principles are all three mastered at the same time. So we will give some space to the original experiment.

In their Experiment 4, Chien and Wexler (1990) tested sentences with reflexives and pronouns paired either with a lexical antecedent or with a quantified antecedent as given in (26a,b,c,d). The distinction of (26c) and (26d) is especially important because quantified antecedents do not allow accidental coreference. Quantifiers are non-referential elements and so cannot enter into a coreference relation with ­pronouns. In the case of coindexation the quantifier automatically binds the pronoun on the semantic level so that reflexive interpretations always indicate violations of Principle B.

Chien and Wexler (1990) used the picture version of a Truth Value Judgment Task, also called Picture Verification Task. Children were shown a picture either depicting the situation where, e.g., the subject(s) touch(es) someone else or her/themselves, and they heard a sentence which described the picture correctly (match condition) or incorrectly (non-match condition) uttered by a puppet who is looking at the picture with the child. The child’s task is to tell whether the puppet is right or wrong about the situation in the picture.



Mama Bear is touching herself



Every bear is touching herself

quantifier- reflexive


Mama Bear is touching her



Every bear is touching her

quantifier –pronoun.

Chien and Wexler (1990) obtained the following results. At around 5 years of age, children interpret quantifiers correctly and at that time they do not have ­problems with the name-reflexive nor with the quantifier-reflexive conditions, even though there always is a slightly lower success rate for non-match conditions. For the name-pronoun conditions, it turned out that in the match condition even 2–3 year-olds interpret the pronoun as referring to a person outside the clause in 90% of the cases. The difficulty lies in the non-match conditions where the pronoun is ­interpreted as coreferring with the subject in 70% of the cases at 4 years of age, in 50% of the cases at age 5–6 and at 25% at 6–7. In contrast to these results on name-pronoun sentences, children gave perfect judgments for the quantifier-pronoun condition in the matching cases and gave correct “no”-responses to the non-match condition about 84% of the time between 5 and 6 years of age. Concentrating on the fifth year of life we get a contrast of 50% correct responses in the non-match condition for pronouns with lexical antecedents, 84% correct responses in the non-match conditions for pronouns with quantified antecedents and nearly perfect responses for reflexives.

These results and their explanation via a pragmatic rule (see 2.3, (14,15)) that has not yet been acquired by children fit in very well with the theoretical developments of the time and were corroborated in several experiments for English with different experimental paradigms. Among these studies are Philip and Coopmans (1996) also using a Picture Verification Task, McDaniel et al. (1990) and McDaniel and Maxfield (1992) using a Grammaticality Judgment Task, and studies using the story version of the TVJT such as Avrutin and Thornton (1994), Matsuoka (1997) and Thornton and Wexler (1999). Thornton (1990) and Boster (1991) used a TVJT and distinguished a coreference context from a bound variable context through the use of constructions with Wh-movement as in (27). They thus avoided the problem of a possibly late acquisition of quantifiers.


Bert and Huckleberry Hound scratched them.

I know who scratched them – Bert and Huckleberry Hound.

Boster (1991) in her Experiment 1 found 96% correct interpretation of the Wh-contexts vs. 62% correct interpretation of simple contexts for children 4;6–6;0 years old, and Thornton (1990) reports 92% correct interpretation of the Wh-contexts vs. 51% correct interpretation of the simple contexts by children between 3;7 and 4;8 years of age.

In contrast to these studies, some experiments did not find any asymmetry in the interpretation of pronouns with quantified and lexical antecedents. Studies on English that find about equal performance in both contexts are Kaufmann (1988) (90% and 87% correct for name-pronoun and quantifier-pronoun respectively), Lombardi and Sarma (1989) (45% and 51%), Boster (1991) Experiment 2 (62% vs. 58%), Utakis (1995) (63% and 60%). Note that Kaufmann (1988) finds a reasonably good performance in both contexts whereas the other studies show about chance performance.

Conroy et al. (2009) provide a very useful table summing up these different results according to experimental paradigm. Their Table 2 also gives information about results on languages other than English. Interestingly, Avrutin and Wexler (1992) find an asymmetry for Russian (48% and 83% correct) in contexts with Wh-binding, whereas they do not find it with the quantifier corresponding to every (48% vs. 59%). Likewise Hestvik and Philip (1999) do not find an asymmetry for Norwegian (91% and 97% correct), whereas Philip and Coopmans (1996) establish a small but significant difference for Dutch (36% vs. 53% correct). Though Dutch children show about chance performance in the quantified antecedent condition, their performance is significantly worse in simple contexts. Note again, that Kaufmann (1988) and Hestvik and Philip (1999) basically find mastery of pronoun interpretation, whereas the other two studies find chance performance or below chance.

It comes to mind that in experiments where performance is good on the simple cases, any asymmetry might be masked by a ceiling effect. In particular, this concerns the experiments on clitic languages: an asymmetry is not predicted since Jakubowicz (1989) and McKee (1992) established that Romance children do not have a PIP.

For completeness of the discussion on the quantifier asymmetry, however, some comments on experiments with French and Spanish children are given here. Hamann et al. (1997) and Baauw et al. (1997) used exactly the same kind of material for French and Spanish as Philip and Coopmans (1996) had used for Dutch and English. Hamann et al. found that 4 year-olds were 78% correct in simple non-matching contexts vs. 70% correct in non-matching contexts with quantified antecedents; 5 year-olds were 100% vs. 88% correct and 6–7 year-olds were 100% vs. 94% ­correct. Compared to overall almost-adult performance, they consistently find more problems in quantified non-matching contexts, but no significant differences. Baauw et al. (1997) report 90% mastery on both conditions for children with a mean age of 5;6, which almost exactly replicates the results on French. Since McKee (1992) also reports 90% adult performance in simple contexts, the results on French and Spanish were predicted by any account that sees the reason for the early mastery in Italian in some special property of clitic pronouns. Many accounts on clitics have in effect claimed that clitic pronouns do not allow coreference – which exempts them from the PIP.7

Recently, the experiments and results showing the quantifier /simple asymmetry have been criticized (see Elbourne 2005; Conroy et al. 2009, and Sects. 6.1 and 6.2 here) so that it is an open question whether there is a PIP or not. Nevertheless, I would like to outline the accounts offered.

Explaining the PIP

Whereas the reflexive/pronoun asymmetry showed that there is a PIP, the quantified antecedent asymmetry indicated that children have problems with pronouns only in contexts where accidental coreference is possible, presumably because they have not yet acquired the rule that excludes coreference in such contexts (Chien and Wexler 1990, here (14)). An alternative explanation claims that children’s processing limitations do not allow a full computation of the relevant constraint (Grodzinsky and Reinhart 1993, (15)).

The account proposed in Grodzinsky and Reinhart (1993) assumes that children know Principle B and the Coreference Rule I, but that their processing capacity does not suffice to compute the two interpretations and compare them. A processing account of the PIP is attractive because the difficulties are not only observed in language acquisition but also in language impairment and loss (see Grodzinsky and Reinhart 1993; Avrutin 2004; Baauw and Cuetos 2003 on agrammatic aphasia, and van der Lely and Stollwerck 1997 on SLI).

A different account of the PIP was developed by Thornton and Wexler (1999). After Heim (1993) had elaborated Reinhart’s Rule I and introduced guises in order to explain the identity debate contexts, Avrutin (1994), Hamann (2002) and Thornton and Wexler (1999) employed this notion to explain the PIP. Hamann (2002) following Grimshaw and Rosen (1990) and very similar to Avrutin (1994, 1999), assumes that children can anchor strong pronouns in the situational context, hence might allow deictic identification through pointing as a guise for the pronoun.8 This differs from the guise of the possible antecedent DP and so allows coreference. Thornton and Wexler (1999) develop Heim’s proposal into the stronger claim that children overextend guise creation. Since different guises make coreference readings possible, this explains the PIP. In their interpretation of Heim (1993, 1998), Thornton and Wexler (1999) make the generalization that different guises are created in cases like (12) as well as cases like (11a,b), whereas only the former are the case where Heim (1993) explicitly discusses different guises. So, they argue, what children do not know is which contexts restrict guise creation.

Recently, Verbuk and Roeper (2010) have criticized both the above explanations. They argue that a processing account relying on the derivation and comparison of two representations cannot hold, since children compute two representations and compare them in many cases where scalar implicatures are involved as e.g. the choice of some over all. They refer to an experiment by Papafragou (2002) who showed that 5;3-year-old children correctly compute scalar implicatures based on the  <  all, some  >  scale 77.5% of the time. So, Verbuk and Roeper conclude, children are well able to compare two interpretations at the age at which they still show the PIP. The argument is relativized by Verbuk and Roeper by also pointing out that in general children perform poorly in tasks where the implicature arises through an underinformative statement. This is the case for computing the implicature on exclusivity for the use of or as shown by Chierchia et al. (2001) and Gualmini et al. (2001). Chance performance was also found for implicatures of modal verbs and quantifiers by Noveck (2001). Likewise, Papafragou (2002) finds that 5-year-old children compute the non-completion implicature for aspectual verbs like start and begin only one third of the time. On the other hand, the same study showed that these children performed above chance if the sentence contained the term half. Papafragou and Musolino (2003) present similar findings: 5-year-olds perform below chance on  <  some, all  >  and on  <  start, finish>, but do pretty well on a scale based on numerals. So children may do poorly on some scales, but they can handle other scales pretty well. When Papafragou and Musolino extend their study such that they first train children to pay attention to what is the expectation in the given context and also provide more context so that this expectation is salient for the child, it turns out that children do much better in rejecting the less informative statement and that they give the right reasons for doing so.

The conclusion that can be drawn from these experiments is that children can do some implicatures pretty well, especially when they are given enough context to realize that the speaker’s statement has not met the expectation. So what it boils down to is that children have to compute an open proposition (the question under discussion) for implicatures9 and that this might be the difficulty. In any case, the argument put forward by Verbuk and Roeper needs corroboration by more empirical research. At this point, it is impossible to decide whether children cannot compute the alternative for processing reasons, whether they cannot do so because they ­cannot construct the open proposition, or whether they cannot do so because they lack the speaker’s perspective, the possibility offered by Hendriks and Spenader (2006).

Regarding extended guise creation, Verbuk and Roeper (2010) show that ­children give random answers in critical contexts like (11a,b) and (12). Though some of the experimental conditions may be problematic for other reasons,10 the experiment shows that children have difficulties in such contexts, which Verbuk and Roeper interpret as showing that children certainly do not overextend guise creation, rather they seem to have difficulties creating the elaborate guises necessary in such contexts.11

Apart from such criticism of specific accounts, it remains relatively uncontroversial that the manifestations of the PIP observed in different experiments must be related to the coreference relation. However, another asymmetry, that between ­simple sentences and ECM sentences with pronouns, indicated that other theoretical proposals should be explored in so far as additional factors may be involved.

4 Reflexivity and the Effects of the Chain Condition

Following Reinhart and Reuland’s (1993) different perspective on the binding ­principles, which turns on the fact that different languages may have different sets of anaphoric expressions and as a consequence different syntactic contrasts, acquisition research reoriented to the idea that “the acquisition of pronominal anaphora can be accounted for purely in terms of the acquisition of specific lexical features”, as stated in Philip and Coopmans (1996). This kind of approach is already suggested by Chien and Wexler (1990) for the developmental profile of Principle A, which suggests a slow acquisition of the lexical properties of reflexives (see Sect. 3.1 here), and is pursued very successfully in the paper by Verbuk and Roeper (2010). At the same time it is clear that decisive data on how different sets of anaphoric expressions can be acquired will best come from comparative studies.

Reflexivity and Types of Verbs: The Case of Dutch

Focusing on Dutch, which has a tripartite anaphoric system, see (18), Sigurjónsdóttir and Coopmans (1996) report that 5-to-6 year old children show adult performance in simple principle B sentences only 17% of the time, whereas they interpret the anaphor zichself correctly about 80% of the time. The reflexive/pronoun asymmetry is therefore unusually pronounced in Dutch.

Using the reflexivity framework, the above authors assumed that Dutch children have problems identifying pronouns as referentially independent so that the Chain Condition will not constrain pronoun use as it does for adults. To provide evidence that the Chain Condition is not applied in an adult-like manner by Dutch children, Sigurjónsdóttir and Coopmans (1996) compared inherently reflexive verbs and transitive verbs. This contrast can provide evidence because in this framework only the Chain Condition can rule out the use of a pronoun with inherently reflexive verbs as in (28b), whereas for regular transitive verbs both Condition B, as formulated in (19), and the Chain Condition (23) apply.



Fredi schaamde zichi

Fred shamed self

‘Fred was ashamed (of himself)’


*Fredi schaamde hemi

Fred shamed him

Fred was ashamed

By Condition A in (19), an inherently reflexive predicate must be interpreted reflexively which is the case in both (28a) and (28b) indicated by coindexation. The verb is inherently reflexive which implies that it is reflexive marked, so Condition B is fulfilled in (28b). (28b) is ruled out by the Chain Condition, however.

When Sigurjónsdóttir and Coopmans (1996) found that children do better with transitive verbs like aaien ‘stroke’ than with verbs like wassen ‘wash’ that also have an inherently reflexive reading, they concluded that children establish a chain in cases like (28b). This implies that in child grammar the pronoun is not specified as referential: only elements specified as [−R] allow a chain.

The Chain Condition and the Asymmetry in ECM and Simple Sentences

Pursuing this line of argumentation other contexts were investigated, in particular ECM cases, see (8). Recall that in such constructions the pronoun is not a semantic co-argument of the verb, so that Condition B does not apply and only the Chain Condition rules out coindexation. Accidental coreference can apply as always in cases of contraindexation.

The experiments typically were Picture Verification Tasks, which considerably narrowed the range of testable ECM constructions. So the pictures for testing ­children’s performance in ECM cases usually include a mirror where the protagonists can do both, see each other and themselves, and the embedding verb is see.12 Typical test sentences are given in (29a,b).



The girl sees her dance.


The mom sees her blow bubbles.

Using such materials, Philip and Coopmans (1996) tested English and Dutch children and found a very clear ECM effect for Dutch, though not for English.

Table 1 shows the results of the conditions in which adults would answer “No”. The data indicate that Dutch children have persistent and exacerbated problems in ECM-constructions till the age of 8 years, an age when they reach chance performance in simple cases.
Table 1

Correct adult no-responses to simple and ECM sentences with pronouns in Dutch and English (from Philip and Coopmans 1996)







6 year olds (%)

4–6 year olds (%)

7 year olds (%)

8 year olds (%)

The girl dries her





The girl sees her dance





As an explanation Philip and Coopmans (1996) propose that there is a feature underspecification for pronouns that allows coindexation without violating the Chain Condition. They suggest that the underspecified feature is the Case feature, arguing that in the Dutch case system structural and non-structural case are “overtly indistinguishable”. If Dutch pronouns are underspecified in this manner, this explains the below chance performance in ECM cases and the difference found for inherently reflexive and transitive verbs.

If problems with Rule I (or P) are responsible for chance performance in simple contexts in English, then Dutch children face a double difficulty: problems with Rule I and underspecification of pronouns. They can wrongly categorize the pronoun as [−R], which allows establishing a chain, similar to reflexives. They can also occasionally categorize the pronoun as [+R], in which case they might allow coreference because of Rule I problems. This double difficulty may explain their below chance performance in simple contexts (see Table 1) and the persistence of the problem.

Note that Verbuk and Roeper (2010) in a very similar manner explain the chance performance of English children not directly through a problem with the application of Rule I but through an underspecification of the referentiality feature on pronouns, which effectively blocks the derivation of implicatures such as Rule I. If some kind of chain condition can be integrated into their account, this underspecification could also explain the bad performance in ECM sentences. The OT-account proposed by Hendriks and Spenader (2006), however, does not predict different performance in ECM and simple sentences. It is quite possible, of course, that some constraint could be added to the relevant hierarchy, which, properly ranked, gives the desired result. Such speculation is beyond this paper, however.

5 More Asymmetries: Languages with Clitic Pronouns

One of the most striking asymmetries established in cross-linguistic research on binding concerns the contrast in children’s performance on pronoun interpretation in English and Dutch on the one hand and in Romance languages on the other hand. For research on binding this means in particular that any account of the PIP for English must also explain why the effect does not occur in languages with clitic pronouns. So different suggestions have been made in order to pin-point the specific property of clitics that must be responsible for their early mastery.

Formal Properties of Pronominal Clitics

In the Romance languages pronominal clitics differ from lexical nominal expressions and from full pronominals in that clitics cannot be used in isolation, cannot receive focal stress, cannot be conjoined, cannot be modified, or be separated from the verb (see Kayne 1975). Pronominal clitics fill a special head position in the highest part of the clausal functional structure.13 So one of their striking properties is that they are nominal arguments14 occurring in a head position of the verbal functional domain.

There are different approaches to account for this property. In base insertion accounts (Borer 1984; Sportiche 1996) complement clitics fill a dedicated head (Sportiche’s “clitic voice”) licensing a pro inserted in complement position so that the chain connecting the argumental pro and the clitic head captures their mixed status. Alternatively, complement clitics are assumed to be generated in complement position and move to a functional head high in the clausal functional structure (Belletti 1999; Burzio 1986; Kayne 1975, 1991; Rizzi 1978, 1982). In French and other languages with participle agreement, a complement clitic must be identifiable as a DP (an argument) at least till the Agr Participle Phrase, where it passes through the specifier. During the last links of the chain, however, only the head of this DP moves, see Belletti (1999). She also argues that Romance complement clitics have a strong case feature that needs checking and so forces overt movement.

The properties that clitics cannot be used in isolation and cannot receive focal stress go together with the observation that clitics cannot be used deictically with only a pointing gesture and no further discourse anchorage. So it has often been observed that clitics totally depend on discourse and their antecedent to receive an interpretation. Hamann (2002) argued that clitics must take on the guise of their DP antecedent so that similarities to Principle C might be observed. Note that in Principle C sentences disjoint reference readings can be suspended under certain conversational conditions as in the following dialogue: A:Have you met the director? Does your sister know her? – B. with a smile: Oh, yes, my sister knows the director quite well. In fact, she is the director. Hamann (2002: 106) discusses similar examples for clitics and also accepts arguments from Cardinaletti and Starke (2000) who demonstrate that clitics can be used demonstratively/deictically or with stress in specially constructed contexts. Hamann (2002) therefore concludes that accidental coreference for clitics is not categorically excluded, it is “just harder to get” than for English style pronouns (Hamann 2002: 103).

In any case, a clitic or a weak pronoun seems to be referentially deficient in a sense that full pronouns are not and so they need more context to define their guise or referent than a full pronoun or a full DP. This referential deficiency and strong identification of the clitic with its discourse antecedent has been captured by some accounts in the claim that clitics are always “bound” (see Baauw et al. 1997), an intuitive term including D-linking. Making precise this intuition, Delfitto (2002) has provided an account where clitics are variables bound by the λ-operator. For “binding” to a discourse antecedent, the left dislocation structures used for topicalization in Romance languages serve as a model. The clitic is bound by an empty topic as in (30a,b), see Baauw and Cuetos (2003: 233) who analyze clitic constructions as “hidden clitic-left dislocation structures”. Since empty topics must be discourse identified, the clitic can never introduce new information in the form of a new referent and thus deictic use (in the sense of pointing) is excluded.



[topic ec]i Juan lai vio en casa

eci John heri sees in the house


[λy (Juan vio y en casa)] (ec)

Note that in this account clitics cannot enter into accidental coreference relations because they are always bound variables. However, this idea turns on the possibility of using clitics in contexts like (11a), repeated in French as (31). If exceptional coreference is possible for a clitic in such contexts, then any account that strictly excludes coreference needs to be extended in ways which can explain (11a), see Hamann (2002) for an attempt. Unfortunately, judgments are not unanimous. Most, but not all, native speakers of French accept (31), and a version with strong pronouns does not immediately come to mind. Some Spanish speakers seem to accept the Spanish version of this sentence whereas others reject it, see Baauw and Cuetos (2003).


Tout le monde aime Oscar. Marie l’aime, Chantal l’aime et Oscar l’aime.

Everybody loves Oscar. Marie loves him, Chantal loves him and Oscar loves him.

Experiments on Binding in Clitic Languages

Given their referential deficiency and their high position in the functional structure of the clause, Romance clitics manifest many differences from English pronouns. So it should not have been very surprising that only very young children in Romance ­languages showed a “delay of principle B” whereas 5-to-6 year-olds show mastery.

The earliest binding experiments, Jakubowicz (1984, 1989), showed in an Act-Out and a Picture Matching task that French children’s performance is good already around the age of 4 years: At age 3;6–4;0 children performed 78% adult-like in (32b) and at age 4;7–5;0 children are 98% correct in (32a), 78% correct in (32b) and 75% correct in (32c). Jakubowicz (1989) also tested children’s production of subject and object clitics and found a substantial delay of object clitics in production with respect to subject clitics (see also Hamann et al. 1996). Subsequently, McKee (1992) showed near perfect performance on both reflexives and clitic pronouns by Italian children at the age of 4 years.



age group





Nounours dit que Kiki se brosse




Teddy bear says that Kiki brushes himself


Nounours dit que Kiki le peigne




Teddy bear says that Kiki brushes him


chaque Schroumpfette veut que


Marie la brosse




Every Smurfette wants that Marie brushes her

More recently, in an experiment testing 99 French speaking children between the ages of 4;0 and 6;0 both in production and in comprehension Zesiger et al. (2010) found that children have more difficulty with complement clitics than with reflexive clitics both in comprehension and production. For the comprehension experiment a Picture Verification Task with photographs was used (TVJT using pictures) testing complement clitics and reflexive clitics in simple sentences as given in (33a,b).



Papa le couvre

Papa him covers

‘Daddy covers him’


Papa se couvre.

Papa self covers

‘Daddy covers himself’

In the match conditions children were perfect for both reflexives and complement clitics. In the binding mismatch conditions, however, the authors found a small but significant difference throughout the age range under investigation. The children rejected mismatches in condition A contexts at ceiling (99.3%) already at age 4;0, but rejected such mismatches in condition B contexts between 87.5% at age 4;0 and 93.8% at age 6;0. Since the difference is so small compared to studies investigating the same contexts in English or Dutch, the authors conclude that their results confirm the sharp dissociation between languages with clitics and without clitics. They offer an analysis where complement clitics, but not reflexives, cross the subject chain during derivation. This is a factor that can explain the larger difference found for production and therefore might also be responsible for the small one in comprehension (Zesiger et al. 2010, but see also Sects. 5.1 and 5.3 and the discussion about rudimentary guise creation in footnote 19).

Accounting for the Absence of a PIP

Explanations for the good performance on clitic pronouns and the asymmetry found for clitic and non-clitic languages are basically twofold, structural or pragmatic, where the pragmatic observations are often derived from structural or lexical properties of clitics.

McKee (1992) referred to the higher position of the clitic and proposed that this high position made it clear for the child that the “local domain” for the pronominal clitic must be the clause, the IP, whereas English children could for a time assume the VP to be the “local domain”. This account has been criticized for many reasons, the most important one being that it must assume the VP external subject hypothesis. Otherwise the subject trace in the VP would provide a potential binder for the pronoun and English children should do well. See Avrutin and Wexler (1992) for a discussion, also Hamann (2002) or Baauw and Cuetos (2003) who argue that in the case of English agrammatics this would imply that they lose grammatical knowledge. This criticism does not mean that other structural accounts might not fare better, see Baauw and Cuetos (2003), who assume that clitics are bound variables and can therefore not show accidental coreference.

Avrutin and Wexler (1992) and Avrutin (1994) pursued the pragmatic road outlined in Chien and Wexler (1990). They point out that clitics are referentially deficient and therefore do not (usually) corefer. Avrutin (1994) formally derives coreference through the existence of two guises, see also Heim (1993), and argues that pronouns can introduce extra guises only if they can refer deictically. The idea simply is that deixis in its sense of pointing always introduces the guise of the entity or person pointed at – the physical presence. This possibility is also involved in the identity debate in example (12). The identity question is about whether the situationally identified referent of the pronoun is the person named Zelda. Situational, deictic reference therefore often provides one of the guises in such examples. As also discussed in Sect. 3.3, the idea then is that children might assume that such deictic reference is always available, even in contexts where this is not the case. This would allow the creation of two different guises, deixis for the pronoun and the NP-description for the possible antecedent, and thus allow coreference just as in the identity debate about Zelda. Whereas Romance strong pronouns allow deictic reference, Romance clitics do not – unless we consider the contexts discussed in Sect. 5.1. As a consequence, clitics cannot enter into accidental coreference relations, a statement which should again be taken with the proviso that examples like (31) are acceptable for many, but not all speakers.

This proposal is problematic in that it predicts that any pronominal form that cannot be used deictically will be exempt from coreference. Cross-linguistically, this means that Dutch and German children should perform better on weak pronouns in their reduced forms than on strong pronouns. Turned around, it also means that children from a clitic language should perform worse on structurally similar sentences with strong pronouns.

Cardinaletti and Starke (1995) suggested another view on the problem in proposing the deficiency hierarchy: clitic  <  weak pronoun  <  strong pronoun. That deficient pronouns do not allow coreference is derived through the assumption that they lack a referential restriction. Romance children, they argue, are confronted with two clearly different forms, clitics and strong pronouns occupying different positions and having different properties. Since children never misplace clitics in production, they clearly know that clitics are heads, and hence are deficient in lacking the outer DP shells. English pronouns, however, are ambiguous in form between weak and strong uses. Cardinaletti and Starke (1995, 2000) also introduce an Avoid Structure Principle requiring that the deficient form be chosen whenever this is possible. This means that strong pronouns are licensed only through focus or some other licensing context. Transferred to English this implies that in a sentence like (34a) in the absence of stress, the adult will interpret the pronoun as weak and so not allow coreference. The child, however, will have difficulties resolving the ambiguity and will sometimes allow coreference. The clear prediction is that children should do better in sentences with it, which Cardinaletti and Starke (1995) identify as the only English weak pronoun. So (34b) should show better pronoun resolution than (34a).



John sees him


The snake saw it.

The predicted asymmetries of weak and strong pronouns have been systematically investigated only in very few studies and the results are rather contradictory. A pilot experiment conducted by Cardinaletti and Starke (1995) confirmed their prediction. However, Baauw and Cuetos (2003: 229) report on an experiment ­showing that the performance of Dutch children is the same on weak and strong pronouns, namely around chance. Equal performance was also found in a pilot experiment for German by Hamann and Ruigendijk (2009), with the interesting twist, that pronouns in complement position (ambiguous) and in the so called Wackernagel position next to the complementizer or V2-verb (unambiguously weak) were both fully mastered, see Sect. 5.5. Baauw et al. (1997) report that Italian children allow coreference with strong pronouns. However, Varlakosta (2000) and Varlakosto and Dullaart (2001) show that Greek children allow coreference neither with clitics nor with strong pronouns in contexts where adults do not allow it.

More recently, another explanation of the clitic/non-clitic asymmetry has been suggested by Verbuk and Roeper (2010). As in Cardinaletti and Starke (1995), the crucial factor is seen in ambiguities. Cardinaletti and Starke assumed that a lexical ambiguity keeps the child from identifying the contexts where strong pronouns are excluded leading a child from a Germanic language to overdo coreference. Verbuk and Roeper (2010) point out that English children faced with sentences like (7), (9) and (10) need some time and more evidence to decide on the referential properties of pronouns and the special status of adjuncts.

I repeat (7) here for convenience. In clitic languages only strong pronouns can be used in such contexts.15 Therefore no lexical ambiguity interferes and children from a clitic language are much faster in determining the referential properties of pronominal elements. See Reuland (2001) for similar arguments.


James Bond noticed the gun near him/himself.

Clitics and the PIP in ECM-Constructions

Neither the structural account of McKee (1992) nor the different versions of the deficiency account discussed so far can offer an explanation for another result which was established for French by Hamann et al. (1997) and for Spanish by Baauw et al. (1997) (fully documented in Baauw 2000, 2002). This is the observation that French and Spanish children show a PIP in ECM or clitic-climbing constructions.

Using the same materials and experimental paradigm as Philip and Coopmans (1996) Hamann et al. (1997) ran the classical binding tasks as given in (35) and (36) in French. They extended the experiment to cases resembling ECM and involving clitic climbing, given in (37).16 They found that in ECM cases with pronouns performance is only about 60% adult like.



La fille la seche

nearly adult

the girl is drying her off


La fille se seche

fully adult

the girl is drying herself off



Chaque fille la seche

fully adult

every/each girl is drying her off



La fille la voit danser

60% adult like

the girl sees her dance


La fille se voit danser

nearly adult

the girl sees herself dance

Practically the same percentages were found for Spanish (Baauw et al. 1997; Baauw 2000, 2002). Baauw and Cuetos (2003) find a special problem with the ECM construction also for Spanish agrammatics, who rejected coreference in 79% of the simple sentences with clitic pronouns, but only in 21% of the ECM/clitic-climbing cases.

Following Philip and Coopman’s explanation for this effect in Dutch, Baauw and colleagues assume that only Reinhart and Reuland’s (1993) Chain Condition is operative in contexts such as (37) so that chance performance in this context can be explained by the assumption that children sometimes classify complement clitics as [−R].17 Since referentiality depends on the presence of the ϕ-features and case, several of these features have been discussed as the source of deficiency.

Cardinaletti and Starke (1995) suggested that clitic and weak pronouns are unspecified for the feature [human], which they have to pick up from the antecedent to become referential. Baauw et al. (1997) follow this reasoning in speculating that it is the [human] feature that remains unspecified and so leads children to sometimes analyze clitics as [−R]. Baauw and Cuetos (2003), following suggestions by Reuland (2001), argue that the number feature may be underspecified.

There is one basic problem with these accounts addressed in Hamann (2002) and in Baauw and Cuetos (2003). Hamann (2002) pointed out that it is surprising that this kind of underspecification only occurs in clitic climbing cases but not in simple sentences. Note that this problem does not concern the Dutch case where difficulties that might be due to such an underspecification are evident also in the simple case. It is acute only in clitic languages where practically no errors are found in the simple case. Hamann (2002) suggested that children get ample evidence that clitics move and never are tails of argumental chains so that underspecification of referentiality should be excluded. Baauw and Cuetos (2003) assume that children mostly assign the correct [+R] specification, in which case accidental coreference is excluded for clitics and the Chain Condition is applied faultlessly.

What remains to be shown is which factors make children erase a feature in clitic-climbing contexts but only rarely or not at all in simple cases. The fact that agrammatics also show these problems and that Italian SLI children have production problems in clitic climbing cases points to a processing account involving the syntactic complexity of the ECM construction.18 Aiming for a unified account, Hamann (2002) suggested a pragmatic account of the ECM/clitic climbing cases. As discussed in Sect. 5.1, she assumes that coreference for clitics is not strictly excluded, but hard to get: strongly constructed contexts allow deictic use and the exceptional coreference context (31) may also allow the use of a clitic. Deictic reference for clitics in normal contexts is excluded because in such contexts their relationship to a salient discourse topic amounts to discourse binding. Creating coreference with the help of a deictic guise is therefore practically impossible. This explains the good performance of children in simple contexts.19

For ECM/clitic climbing cases, however, Hamann (2002) points out that two guises are made available and prominent by the construction. In (37) there is the guise defined by the discourse (there is a girl and a mom), which is “mom”, and there is the guise offered by the embedded clause, “the dancer” which in turn is different from the guise “the girl”. Similar to Avrutin (1994), Hamann (2002) then argues that these different guises allow accidental coreference for the child whereas it is excluded by Rule I for the adult.20

The Case of German

Interestingly enough, German may be a test case for some of these predictions, but had not been investigated up to very recently. Given that German pronouns are ambiguous as to a strong and weak paradigm and do not exhibit properties of Romance-type clitics, the null hypothesis is that German children should not behave much differently from Dutch or English children. This is indeed the explicit prediction found in Hamann (2002) and Hendriks et al. (2007). So Ruigendijk (2008) started from the hypothesis that German children should show non-adult interpretations in simple sentences with pronouns, aggravated in ECM constructions. Using a Picture Choice Task, Ruigendijk found that German children between 4 and 6 years of age are 94% correct on simple sentences with reflexives and 95.1% correct with pronouns, mirroring the perfect performance of French or Spanish children. Like Romance children, German children are also 95.5% adult like in ECM cases with reflexives, but only 77.3% adult like in ECM cases with pronouns.

Ruigendijk (2008) explores this very surprising result by pursuing two questions: (a) is there a property of German pronouns which could establish a parallel to clitic languages and (b) is there a property of German which distinguishes it from Dutch and English. Taking up the discussion in Hamann (2002), Ruigendijk speculates that a property reminiscent of clitics is that German weak pronouns can occur in the so called Wackernagel position which is a position high in the left periphery of the clause; see (38a,b) for the high position and (39a,b) for the low position. As to a difference between Dutch and English, following Reuland (2001), she points to the differences in (40a,b,c), which also establish another parallel to clitic languages: in German no ambiguities as to referentiality of pronouns can arise.



…, dass ihn/’n der Junge gesehen hat

that him/’m the boy seen has

that the boy saw him


Sieht ihn/’n der Junge?

Sees him/’m the boy?

Does he boy see him?



…, dass der Junge ihn/?’n gesehen hat

that the boy him/?’m seen has

that the boy saw him


Sieht der Junge ihn/?’n ?

Sees the boy him/?’m

Does the boy see him



Der Manni legt das Buch neben sichi/*ihni


Der mani legt het boek naast zichi/hemi neer


The mani puts the book next to himselfi/himi.

An experiment conducted with a Picture Verification task by Hamann and Ruigendijk (2009) systematically contrasted simple and ECM sentences with the pronoun in complement position with sentences where the pronoun is in the Wackernagel position. This experiment corroborated that German children show early mastery of pronouns (98%) and reflexives (98%) in simple sentences and performed significantly worse in ECM constructions. However, there was no effect of the position of the pronoun since 4–5-year-old children performed at chance in both positions. The youngest children showed a significant effect, unexpectedly performing worse on the ECM sentences with the pronoun in the high position. The latter result demonstrates that the high position of pronouns available in German does not aid children.

In order to keep track of some of the findings discussed here, Table 2 gives an overview over experiments conducted in different languages examining crucial constructions.
Table 2

Cross-linguistic results on pronoun interpretation in principle B environments, Chain Condition contexts, and comparing strong and weak positions or forms (percentage of adult responses of children between 4 and 6 years of age)



Lexical antecedent (%)

Quantified antecedent (%)

Chain condition



Sigurjónsdóttir and Coopmans (1996)



Philip and Coopmans (1996)



ECM: 10%

Baauw and Cuetos (2003)

50% vs. 50%


Chien and Wexler (1990)




McKee (1992)



Conroy et al. (2009)





Avrutin and Wexler (1992)



48% vs. 66% (not signif.)


Hestvik and Philip (1999/2000)







Hamann and Ruigendijk (2009)



95% vs. 95%

Low 60%


High 30%


McKee (1992)



Jakubowicz (1989)



Hamann et al. (1997)



ECM: 51%


Baauw et al. (1997)



ECM: 64%

Mean age: 5;6


Varlakosta (2000)


87% vs. 95%

6 Recent Developments

Do Children Really Know Principle B?

Grimshaw and Rosen (1990) very early pointed out that the generally low performance on simple pronoun sentences could be due to the experimental set up. They observed that the poverty or absence of a discourse context in the Picture Verification task could lead children to override the syntactic requirements or ignore the prosodic information in order to be able to anchor the pronoun to an antecedent at all. They also pointed out that such a coreferential reading is always possible if the pronoun is stressed, i.e. strong, and hence contrastive or deictic, so that overriding syntax by pragmatics is motivated by sentences well in the child’s experience. In the same vein Elbourne (2005) sets out to demonstrate that in effect the asymmetry found in quantified and simple contexts is an artifact of the experimental method. This criticism concerns Chien and Wexler’s (1990) Experiment 4 but also experiments using the story variant of the TVJT so the argument is not only that there is practically no context but also that there is the wrong kind of context.

Elbourne (2005) shows that in most cases the experimental set up did not give highest salience to a possible antecedent in the discourse, or even provided more salience for the local antecedent in the simple condition whereas no such salience was given to the local antecedent in the quantified condition. He concludes that children assign reference to pronouns following Salience or Relevance. Conroy et al. (2009) going through the same experiments in a very similar manner, additionally point out that in the crucial non-match conditions often the Condition of Plausible Denial is not fulfilled (see Crain and Thornton 1998). Elbourne (2005) also analyzes the Picture Verification variants, especially the pictures for the quantifier-pronoun conditions in Chien and Wexler’s (1990) Experiment 4. He points out that the non-match condition for (26d) is an image of three little bears and Goldilocks where the three bears are much smaller than Goldilocks and are hard to identify as little female bears.21 If they are not immediately identifiable as female, then it would be easy to say “No” for the child in this situation – for the wrong reason.

Note that this criticism does not immediately carry over to the Picture Verification task used by Philip and Coopmans (1996) and also Hamann and Ruigendijk (2009). All the participants are easily recognizable as female (mothers, grandmothers, girls etc.) and the only difference in size is conventionalized so that one can distinguish the adults from the children. Still, Conroy et al. (2009) dismiss all Picture Verification tasks on the grounds that “we know relatively little about how the specifics of a static picture create a context that can guide a child’s interpretation of a pronoun” (ms, p. 27). This claim is forgetting that the pictures are presented together with verbal lead-ins often in a game situation introducing the participants, the action and normally mentioning the antecedent of the pronoun last suggesting the last mentioned DP as the antecedent.

The discouraging conclusion Elbourne (2005) reaches in his discussion is that there probably is no quantified/simple antecedent asymmetry: Some of the earlier studies did not find this asymmetry (see Sect. 3.2) and it might have been an experimental artifact in the other ones. This leads to the far-reaching conclusion that there is no evidence for the mastery of Principle B. Taken together with the evidence from older stages of English and modern Frisian (see Sect. 2.4) this leads Elbourne (2005) to speak of a parameter for pronouns.

Do Children Have a Problem with Pronouns at All?

Several recent studies have taken up the challenge posed by Elbourne’s (2005) conclusion and his criticism of experimental methods. As to the first point, it is here that the observed asymmetry in production and comprehension becomes relevant as several studies have pointed out. If children are good at pronoun production, then they clearly master Principle B.

This was first pointed out by Bloom et al. (1994), who studied the spontaneous production of English children and compared it to known results on comprehension. The problem with this study is, however, that it is based mostly on the occurrence of the speaker oriented pronouns I, me, my, for which identification of the referent through a salient discourse antecedent is irrelevant. De Villiers et al. (2006) provide an experimental comparison using third person pronouns and find that production of pronouns is significantly better than comprehension even when the antecedent is made relevant in the context. Similar results were obtained for Dutch by Hendricks et al. (2006).

To complicate matters, cross-linguistic research does not unanimously show better production of pronouns. In French the asymmetry rather goes in the opposite direction. Comprehension precedes production as numerous studies on the production and comprehension of complement clitics have shown, see e.g. Jakubowicz (1989) and Zesiger et al. (2010). Note that the production problem here is not so much the use of a reflexive but the omission of the complement clitic or the use of a lexical DP, which is usually attributed to the complexity of the clitic-construction. In a study of Hebrew (which has the PIP) and German (which does not), Ruigendijk and Friedman (2009) concentrate on reflexive contexts to establish whether children will use pronouns in such contexts. These, they argue, provide the real test for non-mastery of Principle B and the production/comprehension asymmetry. They find no asymmetry, neither in German nor in Hebrew.

Most of the work done on this asymmetry shows that controlling context conditions seems to make the gap between production and comprehension smaller (see de Villiers et al. 2006 or Spenader et al. 2009), which brings us back to Elbourne’s second contribution to the discussion of the PIP, his criticism of the early experimental conditions. In experiments by de Villiers et al. (2006) and Spenader et al. (2009) the context setting lead-ins are manipulated in controlled ways targeting factors like relevance and topic-hood. De Villiers et al. (2006) contrasted the paradigm in (41) with the paradigm used by Jakubowicz (1984) and Chien and Wexler (1990) given in (42).


Here is a mom and a girl. The mom dries her.


Mama Bear says that Baby Bear washes her.

In (42) the antecedent is highly relevant and de Villiers et al. (2006) obtain better results in comprehension and production in the conditions using paradigm (42) than in those using the paradigm exemplified in (41). They also suggest that clearly identifying the topic through mentioning only one antecedent not two should likewise help children in interpreting pronouns. Such an experiment has now been conducted by Spenader et al. (2009) confirming the prediction. It can be concluded that given the right kind of context children can interpret pronouns quite well, even though a difference in production remains. It is also evident that systematic experimental research is needed to pin down the exact context conditions which help English and Dutch children.

Conroy et al. (2009) is a contribution to this kind of systematic research. When all the context setting factors are properly controlled, they argue, English children show early mastery of simple sentences and prefer what they call a “deictic” interpretation defined as “the interpretations in which a pronoun lacks an intrasentential antecedent” (p. 7). In Conroy et al.’s (2009) experiments no asymmetry with quantified and simple sentences can be observed when both readings are made equally salient and there is a proper Condition of Plausible Denial. When they deliberately violate these conditions in their third experiment, they promptly obtain the asymmetry. Interestingly however, de Villiers et al. (2006) also tested pronoun conditions with quantified antecedents and found that they were better mastered than conditions with simple antecedents across paradigms. This shows that Conroy et al.’s findings as well as Elbourne’s claims might have to be reevaluated given more systematic research on the context conditions that facilitate interpretation for children.

If we follow Conroy et al.’s reasoning for the moment, controlling the context in specific ways makes the quantifier asymmetry disappear because children do well in the name condition. However, this also makes the PIP disappear. This in itself is an important result, especially as Elbourne (2005) and Conroy et al. (2009) speculate that the asymmetry and the PIP are only expected under those theories of binding which differentiate coreference and variable binding (Reinhart 2004, 2006; Heim 1993). They also point out that the asymmetry is unexpected in accounts deriving Principle C as a special case of Principle B as Reinhart and Heim do, given that children do well on the interpretation of lexical DPs.

However, there are several things left unexplained in Conroy et al.’s article: Why don’t adults need the same kind of elaborate context in order to exclude local antecedents in simple sentences with pronouns? Why do the same kind of experiments also find an asymmetry concerning the interpretation of reflexives and pronouns? Why do these experiments show a development in the interpretation of reflexives and pronouns (Chien and Wexler 1990; Jakubowicz 1989; Philip and Coopmans 1996)? Why is there a construction, the ECM construction, which is even more difficult for children than simple sentences with pronouns? Why can the same experimental material using a paradigm very much like (41) produce different results in different languages showing early mastery in Romance languages and in German, but not in Dutch and English (see Sects. 3.2 and 5.4)?

Conroy et al. (2009) answer the last question (with respect to clitic languages only) by following Baauw and Cuetos (2003) in pointing out that Romance clitics do not allow accidental coreference. So – despite their earlier speculation to the contrary – this notion must play a role in the interpretation of pronouns. Since the theoretical considerations we have outlined in Sect. 2.3. overwhelmingly indicate the need for a division of labor of syntactic and pragmatic constraints, it seems to have been a very fruitful scientific accident that the PIP emerged in certain experiments alongside with an asymmetry concerning reflexives and quantified antecedents and it may, after all, not have been an accident, as de Villiers et al.’s (2006) results seem to indicate.

Though the older experiments might not have optimally tapped into children’s grammatical knowledge, they did highlight an area of difficulty for children and raised the question of what is special about the discourse requirements for pronouns in some languages, but not in others. The new results reopen the discussion, however, and make innovative approaches possible.

How to Set a “Pronoun Parameter”

Verbuk and Roeper (2010) pursue the idea that lexical, syntactic and pragmatic knowledge interact during development. They follow Elbourne (2005) in his assumption that Principle B could be parameterized and must be acquired. The English child’s task is to establish that pronouns are in syntactic opposition to reflexives even though there is conflicting evidence.

Given the subset principle, the assumption is that children start with fine grained lexical distinctions of pronouns and reflexives including features such as Point of View, physical, intentional, etc. At this stage parameters do not apply and pronouns are ambiguous between pronouns and reflexives, which Verbuk and Roeper (2010) call the Middle English or Frisian stage. So children will be guided by pragmatic relevance (as shown by Elbourne) and disjoint reference and coreference readings appear at chance level in both, simple principle B contexts and exceptional coreference contexts (see (11) and (12)).

In the second stage children construct the scale  <  reflexive, pronoun>, which enables them to instantiate principle B. In order to arrive at this scale, children have to generalize from all the possible features distinguishing pronouns and reflexives and come to realize that pronouns and reflexives, though identical in their features in most respects, crucially differ with respect to referential dependence: “reflexives differ from pronouns in terms of being necessarily referentially dependent”, p. 58. Once a rule has been established that reflexives are used instead of coreferential objects, a contrast can be established for pronouns through exposure to specific contexts. Crucially, the restriction on reflexive objects is established by reference to the syntactic category of VP. Hence, adjuncts do not fall under this restriction.

Having ordered the elements in a scale of referential dependency, children will compute disjoint reference in simple Principle B sentences but not yet in exceptional coreference contexts. In order to differentiate between these contexts, the child needs to compute an open proposition. As was discussed in Sect. 2.3, a Q-implicature can be computed as soon as a scale of referential dependency is established (provided the child has the processing capacity to compare two ­meanings, see also Sect. 3.3). For exceptional coreference this implicature has to be cancelled. Cancellation is forced by the open proposition implicit in the context (for (11a) this open proposition is x hates Oscar). Since exceptional coreference contexts are rare in the input and the child is confused by ambiguous contexts such as (7), (9) and (10), the prediction is that adult performance will be delayed. The child will assign disjoint reference in accordance with principle B and a Q-implicature such as Rule I. In order to perform adult-like, the child has to realize that an open proposition has to be extracted from the context so that informativeness implicatures can overrule the quantity implicature.

Since Verbuk and Roeper (2010) assume that Principle B is learned, they have to offer an explanation for the production/comprehension asymmetry, which is usually taken to show early mastery of the grammatical principle. They assume that once the scale is established, production will be unproblematic. However, an earlier child grammar, the Frisian grammar, will be activated in comprehension in order to accommodate the task sentence as a true utterance by the speaker.

Though the account does not specifically refer to the PIP with respect to reflexives, it outlines a story where reflexives have to be firmly established before pronouns are categorized in comparison. This can predict the observed delay of pronoun interpretation with respect to reflexives. As was outlined in Sect. 5.3, the account also predicts an asymmetry in comparison to Romance languages and given the discussion in Sect. 5.5, the same asymmetry for German. However, like most of the other accounts based on the pragmatics of pronouns, it might run into difficulties with respect to ECM sentences – unless some kind of chain condition can be ­integrated (see also Sect. 4.2).

7 Conclusion

In this article I have given an overview of more than two decades of research focusing on relevant and controversial results on the acquisition of pronominal reference. I have tried to highlight the interaction of theoretical advances and empirical results, especially with respect to the notion of coreference and the notion of reflexivity. Summing up the facts and phenomena is not as easy now, however, as it might have been 5 or 8 years ago.

So let me start with what was accepted knowledge at the turn of the millennium. The important observations then were that children acquiring English or other Germanic (and Slavic) languages did well in sentences with reflexives and even in pronoun sentences with quantified potential antecedents, but at the same time had difficulties with pronouns in simple sentences. In contrast, children from Romance languages did show no such pronoun problem. These observations indicated the importance of cross-linguistic investigation, since language specific constructions can often serve as test cases for the division of labor of syntax and pragmatics (see Avrutin and Wexler 1992; Sigurjonsdottir and Coopman 1994; Hestvik and Philip 1997). So any explanation advanced for the PIP needed to account for the better performance on reflexives, on pronouns with quantified antecedents, and on clitics. Explanations usually pointed to the possibility of coreference in certain ­configurations but not in others and suggested that the pragmatic conditions for excluding coreference are not yet acquired. Other accounts claimed that computing the pragmatics overtaxed the processing capacities.

Further studies added other observations needing an explanation. One is the fact that agrammatics show a PIP in those languages or constructions where children show a PIP. Another observation is that a PIP often surfaces in ECM constructions even when there is no such difficulty in simple sentences with pronouns. And yet another one is the asymmetry in production and comprehension in certain languages.

Trying to account for all of these phenomena is not trivial. Studies integrating results on agrammatism usually suggest a processing explanation of the PIP roughly along the lines of Grodzinsky and Reinhart (1993): processing breaks down because a coreference condition has to be computed. Including the ECM-constructions needed additional assumptions in most accounts, namely that pronouns can be underspecified as to the referentiality feature, an assumption that needs to be motivated if the underspecification manifests itself only in ECM-constructions.

This state of affairs was called into question by Elbourne (2005), who examined the experimental methods of previous experiments. Elbourne (2005) concluded that there is indeed a big problem with pronouns that does not go away with quantified antecedents. In contrast, de Villiers et al. (2006) showed that – given the right kind of discourse context – English children get better and Conroy et al. (2009) claimed that they master simple pronoun sentences from early on. Though problems remain with these accounts (see Sects. 6.2 and 6.3), they might have helped to identify the sort of discourse conditions that enable children to compute the intended reading. An example of such discourse conditions is the “question under discussion” or open proposition which – when properly introduced – facilitates the computation of implicatures (Gualmini and Meroni 2009). Another crucial factor for pronouns is the presence of a relevant antecedent (de Villiers et al. 2006) or the introduction and control of an unambiguous discourse topic as shown by Spenader et al. (2009). From this perspective, these recent studies seem to indicate that the original pragmatic explanation of the PIP was on the right track, and more studies on the ways in which discourse can help with pronoun resolution are necessary in order to advance our understanding of the syntax-discourse interface.

Another merit of the recent studies is to take very seriously the cross-linguistic variation found in the properties and types of anaphora in different languages. This reopens the developmental perspective by indicating that the referential properties of language specific reflexive and pronominal elements must be acquired. However, if this kind of feature acquisition, which is made difficult in some cases by ambiguous input, is the reason for the PIP in children, then it is at first sight difficult to explain why agrammatics show the same problems.

Let me conclude with some speculation as to a possible solution to this dilemma inspired by Grillo’s (2008) account of agrammatism. Grillo assumes that due to the time course of processing, features that occur later in the derivation are most ­vulnerable. This mostly concerns features of the Left Periphery, the syntactic side of the syntax-discourse interface, and thus does not seem to apply to referentiality, which is traditionally defined via the ϕ-features and case. So one possibility would be to investigate not referentiality but topic-hood, which plays a role in some accounts. However, a referentiality feature might well be involved as some accounts of quantification postulate a referentiality feature and a Referential Phrase as a functional projection high in the clause, see Beghelli and Stowell (1997) and the discussion in Sect. 2.5. If this kind of referentiality feature is involved in pronoun resolution and is indeed a feature of the Left Periphery it might be vulnerable in agrammatism, especially in constructions involving a high processing load as in ECM-constructions. For children, this feature is hard to assign to specific elements if there is ambiguous evidence as in English, and it is vulnerable in complex constructions. Given all the currents and countercurrents in older and recent research and the surprising new results for English and other languages, many questions remain unanswered or have been reopened inviting theoretical discussion as well as further empirical research with as wide a variety of methods as possible. In my opinion, the most promising avenues of research are the further investigation of discourse conditions and the examination of referentiality and topic-hood in ­connection with the clitic/weak/strong distinctions for pronouns.


  1. 1.

     de Villiers et al. (2006) point to a principled problem inherent in the fexibility: whenever constraints are added to accommodate data this demonstrates a certain arbitrariness of the approach, see the ease with which Hendriks and Spenader (2006) change Fischer’s (2004) theoretically motivated constraint system in order to make it predictive of child data.

  2. 2.

     See Heim (1993), Hamann (2002) on the relationship of indices and the variables of formal semantics.

  3. 3.

     This is usually defined as the minimal category containing x, a governor of x and a SUBJECT (accessible to x).

  4. 4.

     See Chomsky (1981) and many others for arguments that these two types of coindexations do not have the same properties.

  5. 5.

     Heim and Kratzer (1998: 127) do not follow this view and explicitly treat the binding conditions as purely syntactic. Verbuk and Roeper (2010) also argue that they are syntactic because they refer to syntactic entities: the local domain or the minimal clause.

  6. 6.

     However, even if the approaches arrive at similar outcomes for the simple cases, differences in predictions might arise for cases like (11) and (12); see Verbuk and Roeper (2010) and their discussion of Reinhart (2004).

  7. 7.

     This is the account that Hestvik and Philip (1999) also offer for Norwegian since Norwegian pronouns move like clitics at LF.

  8. 8.

     Note that this deictic guise or interpretation of the pronoun always involves pointing and is radically different from what Conroy et al. call the deictic interpretation of a pronoun.

  9. 9.

     See also Gualmini and Meroni (2009), who suggest that the difficulty can be located in accommodating a “Question under discussion” for the derivation of an underinformative implicature.

  10. 10.

     In one of the stories Bill, Mary and Jane want to draw somebody. They get paper and crayons and then Bill draws Mary and Mary draws Bill. The story ends with the sentences: “Nobody drew Jane. So Jane drew her”. Here the following problem may arise: If nobody drew Jane, then a child may find it confusing that Jane drew her – nobody did. The problem is that the quantifier in the intended interpretation ranges only over the two protagonists mentioned in the last part of the story. The child might interpret it as ranging over all the protagonists including Jane.

  11. 11.

     Note here, that under a strict reading of Heim (1993), guise creation is not the crucial point in (11a), so that the criticism may be valid for Thornton and Wexler’s (1999) version but does not touch the original account. Note also that, Avrutin (1994, 1999) and Hamann (1997, 2002) may escape the criticism because they do not assume that children will always be able to create the guises required in complicated contexts, they refer specifically to the simple possibility of deictic anchoring which can provide the guise for the pronoun. Such non-adult deictic anchorage is evident in other areas of child speech as well, e.g. tense.

  12. 12.

     A careful discussion of the problems of this experimental set-up can be found in Hamann (2002) who ran a control experiment showing that French children have no trouble in determining who can see whom in a mirror and do very well when there is a mirror and the simple Priniple B sentence (i). So it can be excluded that using mirror images added a conceptual difficulty for children. See also Coopmans and Philip (2000) and Baauw and Cuetos (2003) on this point.

  13. 13.

    (i) La maman la voit.

    The mom her sees

    13  The possible exception are French subject clitics, for which different analyses have been proposed.

  14. 14.

     Clitic chains show A-chain properties, see Rizzi (1986).

  15. 15.
    1.  (i)

      Jean a mis le ballon derrière lui (*le)/ J’ai mis le ballon derrière moi (*me) John placed the ball behind HIM (*m)/ I put the ball behind ME (*me - clitic)

    2.  (ii)

      Il y’avait sept linguistes dans la salle sans (*me) compter moi-meme

      there were seven linguists in the class without counting (*me-clitic) myself.

  16. 16.

     Such sentences show “clitic climbing” (reminiscent of Postal’s (1974) raise subject-to-object) because the object clitic from the lower clause climbs to the higher clause.

  17. 17.

     Note that technical difficulties might arise since it must be explained how these [−R] elements can end up in high functional positions in the clause, establishing a chain with pro or trace in complement position.

  18. 18.

     Baauw and Cuetos (2003) demonstrate that children do not have difficulties with other types of embeddings with non-overt subjects such as control constructions. ECM constructions, however, are arguably more complex than these as their old name “Raise-Subject-to-Object” (Postal 1974) suggests.

  19. 19.

     Note that the assumption that coreference is not categorically excluded for clitics could also explain the results found by Zesiger et al. (2010): French children show better performance on reflexives than on clitic pronouns. The difference is not remarkable but significant and could ­represent the few cases where French children allow coreference.

  20. 20.

     This account leads Hamann (2002) to predict that German children should do better on unambiguously weak pronouns in the Wackernagel position than on pronouns in complement position. The data of Hamann and Ruigendijk 2009 do not confirm the prediction, see Sect. 5.5.

  21. 21.

     Many seminar discussions on this point have shown that even adults have difficulties identifying the little bows the girl bears wear.


  1. Avrutin, S. 1994. Psycholinguistic investigations in the theory of reference. Doctorial diss., Massachusetts Institute of Technology, Cambridge, MA.Google Scholar
  2. Avrutin, S. 1999. Development of the syntax-discourse interface. Dordrecht: Kluwer.Google Scholar
  3. Avrutin, S. 2004. Optionality in child and aphasic speech. Lingue e Linguaggio 1: 67–93.Google Scholar
  4. Avrutin, S., and K. Wexler. 1992. Development of Principle B in Russian: Coindexation at LF and coreference. Language Acquisition 2: 259–306.CrossRefGoogle Scholar
  5. Avrutin, S., and R. Thornton. 1994. Distributivity and binding in child grammar. Linguistic Inquiry, 25 (1): 167–171.CrossRefGoogle Scholar
  6. Baauw, S. 2000. Grammatical features and the acquisition of reference. A comparative study of Dutch and Spanish. Doctoral diss., Utrecht University, Utrecht, the Netherlands.Google Scholar
  7. Baauw, S. 2002. Grammatical features and the acquisition of reference. A comparative study of Dutch and Spanish’ GLOT International 6, 2/3: 65–71.CrossRefGoogle Scholar
  8. Baauw, S., and F. Cuetos. 2003. The interpretation of pronouns in Spanish language acquisition and breakdown: Evidence for the “Principle B Delay” as a Non-Unitary Phenomenon. Language Acquisition 11: 219–275.CrossRefGoogle Scholar
  9. Baauw, S., L. Escobar, and W. Philip. 1997. A delay of Principle B-Effect in Spanish speaking children: The role of lexical feature acquisition. In GALA 1997, ed. A. Sorace, C. Heycock, and R. Shillock, 16–21. Edinburgh: HCRC.Google Scholar
  10. Beghelli, F., and T. Stowell. 1997. Distributivity and negation: the syntax of each and every. In Ways of scope taking, ed. A. Szabolsci, 71–109. Dordrecht: Kluwer.Google Scholar
  11. Belletti, A. 1999. Italian/Romance Clitics: Structure and derivation. In Clitics in the languages of Europe, ed. H. van Riemsdijk, 543–579. Berlin: Mouton – de Gruyter.CrossRefGoogle Scholar
  12. Bloom, P., A. Barss, J. Nicol, and L. Conway, L. 1994. Children’s knowledge of binding and ­coreference: Evidence from spontaneous speech. Language 70: 53–71.CrossRefGoogle Scholar
  13. Borer, H. 1984. Parametric syntax. Dordrecht: Foris.Google Scholar
  14. Boster, C. T. 1991. Children’s failure to obey Principle B: Syntactic problem or lexical error? Ms., University of Connecticut, Storrs.Google Scholar
  15. Burzio, L. 1986. Italian syntax. Dordrecht: Reidel.Google Scholar
  16. Burzio, L. 1998. Anaphora and soft constraints. In Is the best good enough. Optimality and ­competition in syntax, ed. P. Barbosa, D. Fox, P. Hagstrom, M. McGinnis, and D. Pesetsky, 93–113. Cambridge, MA: MIT Press.Google Scholar
  17. Cardinaletti, A., and M. Starke. 1995. The tripartition of pronouns and its acquisition: Principle B puzzles are ambiguity problems. In NELS 25, ed. J. Beckman, 1–12. Philadelphia: University of Pennysylvania.Google Scholar
  18. Cardinaletti, A., and M. Starke. 2000. An overview of the grammar of clitics. In The acquisition of scrambling and cliticization, ed. S. Powers and C. Hamann, 165–186. Dordrecht: Kluwer.Google Scholar
  19. Chien, Y.-C., and K. Wexler. 1990. Children’s knowledge of locality conditions in binding as ­evidence for the modularity of syntax and pragmatics. Language Acquisition 1: 225–295.CrossRefGoogle Scholar
  20. Chierchia, G., S. Crain, M.T. Guasti, A. Gualmini, and L. Meroni. 2001. The acquisition of ­disjunction: evidence for a grammatical view of scalar implicature. In Proceedings of BUCLD 25, ed. A. Do, L. Dominguez, and A. Johansen, 157–168. Somerville: Cascadilla Press.Google Scholar
  21. Chomsky, N. 1981. Lectures on government and binding. Dordrecht: Foris.Google Scholar
  22. Chomsky, N. 1995. The minimalist program. Cambridge, MA: MIT Press.Google Scholar
  23. Conroy, A., E. Takahashi, J. Lidz, and C. Phillips. 2009. Equal treatment for all antecedents: How children succeed with Principle B. Linguistic Inquiry 40: 446–486.CrossRefGoogle Scholar
  24. Coopmans, P., and W. Philip. 2000. Notes on the January experiment. Ms., Utrecht University, Utrecht.Google Scholar
  25. Crain, S., and C. McKee. 1985. Acquisition of structural restrictions on anaphora. In NELS 16, ed. S. Berman, J.W. Choe, and J. McDonough, 94–110. Amherst: University of Massachusetts, GLSA.Google Scholar
  26. Crain, S., and R. Thornton. 1998. Investigations in universal grammar: A guide to experiments on the acquisition of syntax and semantics. Cambridge, MA: MIT Press.Google Scholar
  27. De Villiers, J., J. Cahillane, and E. Altreuter. 2006. What can production reveal about principle B? In Proceedings of the Inaugural Conference on Generative Approaches to Language Acquisition-North America (GALANA), ed. K.U. Deen, J. Nomura, B. Schulz, and B. Schwartz, 89–100. Honolulu: University of Connecticut Occasional Papers in Linguistics 4.Google Scholar
  28. Delfitto, D. 2002. On the semantics of pronominal clitics and some of its consequences. Catalan Journal of Linguistics 1: 41–69.Google Scholar
  29. Deutsch, W., C. Koster, and J. Koster. 1986. What can we learn from children’s errors in understanding anaphora? Linguistics 24: 203–225.CrossRefGoogle Scholar
  30. Elbourne, P. 2005. On the acquisition of Principle B. Linguistic Inquiry 36: 333–365.CrossRefGoogle Scholar
  31. Evans, G. 1980. Pronouns. Linguistic Inquiry 11: 337–362.Google Scholar
  32. Faltz, L.M. 1985. Reflexivization: A study in universal syntax. New York: Garland.Google Scholar
  33. Fischer, S. 2004. Optimal binding. Natural Language & Linguistic Theory 22: 481–526.CrossRefGoogle Scholar
  34. Grillo, A. 2008. Generalized minimality. Syntactic underspecification in Broca’s aphasia. Doctoral diss., Utrecht, the Netherlands and University of Siena, Sienna, Italy.Google Scholar
  35. Grimshaw, J., and S. Rosen. 1990. Knowledge and obedience: The developmental status of the binding theory. Linguistic Inquiry 21: 187–222.Google Scholar
  36. Grodzinsky, Y., and T. Reinhart. 1993. The innateness of binding and coreference. Linguistic Inquiry 24: 69–102.Google Scholar
  37. Gualmini, A., and L. Meroni. 2009. Scalar implicatures in child language: Cost and compliance. Presentation at GALA, Lisbon, September 2009.Google Scholar
  38. Gualmini, A., S. Crain, L. Meroni, G. Chierchia, and M.-T. Guasti. 2001. At the semantics/­pragmatics interface in child language. In Proceedings of Semantics and Linguistic Theory XI. Ithaca: CLC Publications, Department of Linguistics, Cornell University.Google Scholar
  39. Hamann, C. 1997. From syntax to discourse. Children’s use of pronominal clitics, null subjects, infinitives and operators. Habilitation thesis, University of Tübingen, Tübingen, Germany.Google Scholar
  40. Hamann, C. 2002. From syntax to discourse. Dordrecht: Kluwer.CrossRefGoogle Scholar
  41. Hamann, C., and E. Ruigendijk. 2009. The German pronoun puzzle. Presentation given at the 2nd NWLK (North West Linguistics Colloquium), Bremen.Google Scholar
  42. Hamann, C., L. Rizzi, and U. Frauenfelder. 1996. The acquisition of subject and object clitics in French. In Generative perspectives on language acquisition, ed. H. Clahsen, 309–334. Amsterdam: Benjamins.Google Scholar
  43. Hamann, C., O. Kowalski, and W. Philip. 1997. The French ‘Delay of Principle B’ Effect. In BUCLD 21, ed. E. Hughes, M. Hughes, and A. Greenhill, 205–219. Somerville: Cascadilla Press.Google Scholar
  44. Heim, I. 1993. Anaphora and semantic interpretation. SfS-Report-07-93. Tübingen: University of Tübingen.Google Scholar
  45. Heim, I. 1998. Anaphora and semantic interpretation. In The Interpretative Tract, ed. U. Sauerland and O. Percus, MIT Working Papers in Linguistics, vol. 25, 205–246. Cambridge, MA: MIT, Department of Linguistics and Philosophy, MITWPL.Google Scholar
  46. Heim, I., and A. Kratzer. 1998. Semantics in generative grammar. Oxford: Blackwell.Google Scholar
  47. Hendriks, P., and J. Spenader. 2006. When production precedes comprehension: an optimization approach to the acquisition of pronouns. Language Acquisition 13: 319–348.CrossRefGoogle Scholar
  48. Hendriks, P., Siekman I., Smits E.-J., and J. Spenader. 2007. Pronouns in competition: Predicting acquisition delays cross-linguistically. In ZAS Papers in Linguistics, vol. 48 (Intersentential Pronominal Reference in Child and Adult Language. Proceedings of the Conference on Intersentential Pronominal Reference in Child and Adult Language), eds. D. Bittner and N. Gagarina, 75–101.Google Scholar
  49. Hestvik, A., and W. Phillip. 1997. Reflexivity, anti-subject orientation and language acquisition. Proceedings of NELS 27: 171–185.CrossRefGoogle Scholar
  50. Hestvik, A., and W. Philip. 1999/2000. Binding and coreference in Norwegian child language. Language Acquisition 8: 171–235.CrossRefGoogle Scholar
  51. Higginbotham, J. 1983. Logical form, binding, and nominals. Linguistic Inquiry 14: 395–420.Google Scholar
  52. Jäger, G. 2002. Some notes on the formal properties of bidirectional optimality theory. Journal of Logic, Language and Information 11: 427–451.CrossRefGoogle Scholar
  53. Jakubowicz, C. 1984. On markedness and binding principles. In Proceedings of NELS 14, ed. C. Jones and P. Sells, 154–182. Amherst: University of Massachusetts, GLSA.Google Scholar
  54. Jakubowicz, C. 1989. Linguistic theory and language acquisition facts: Reformulation, maturation or invariance of binding principles. Paper presented at Knowledge and Language, Groningen, May 1989.Google Scholar
  55. Kaufmann, D. 1988. Grammatical and cognitive interactions in the study of children’s knowledge of binding theory and reference relations. Doctoral diss., Temple University, Philadelphia.Google Scholar
  56. Kayne, R. 1975. French syntax. Cambridge, MA: MIT Press.Google Scholar
  57. Kayne, R. 1991. Romance clitics, verb movement and PRO. Linguistic Inquiry 22: 647–686.Google Scholar
  58. Levinson, S. 1987. Pragmatics and the grammar of anaphora. Journal of Linguistics 23: 379–434.CrossRefGoogle Scholar
  59. Levinson, S. 2000. Presumptive meanings: The theory of generalized conversational implicature. Cambridge, MA: MIT Press.Google Scholar
  60. Lombardi, L., and J. Sarma. 1989. Against the bound variable hypothesis of the acquisition of Principle B. Paper presented at the annual meeting of the Linguistic Society of America, Washington, DC.Google Scholar
  61. Manzini, R., and K. Wexler. 1987. Parameters, binding theory, and learnability. Linguistic Inquiry 18: 413–444.Google Scholar
  62. Matsuoka, K. 1997. Binding conditions in young children’s grammar: Interpretation of pronouns inside conjoined NPs. Language Acquisition 6: 37–48.CrossRefGoogle Scholar
  63. McDaniel, D., and T. Maxfield. 1992. Principle B and contrastive stress. Language Acquisition 2: 337–358.CrossRefGoogle Scholar
  64. McDaniel, D., H. Cairns, and J. Hsu. 1990. Binding principles in the grammars of young children. Language Acquisition 1: 121–139.CrossRefGoogle Scholar
  65. McKee, C. 1992. A comparison of pronouns and anaphors in Italian and English acquisition. Language Acquisition 2: 21–54.CrossRefGoogle Scholar
  66. Noveck, I.A. 2001. When children are more logical than adults: experimental investigations of scalar implicature. Cognition 78: 165–188.CrossRefGoogle Scholar
  67. Papafragou, A. 2002. Scalar implicatures in language acquisition: Some evidence from Modern Greek. In Proceedings from the 38th Annual Meeting of the Chicago Linguistics Society. Chicago: University of Chicago Press.Google Scholar
  68. Papafragou, A., and J. Musolino. 2003. Scalar implicatures: Experiments at the semantics-pragmatics interface. Cognition 86: 253–282.CrossRefGoogle Scholar
  69. Philip, W., and P. Coopmans. 1996. The role of lexical feature acquisition in the development of pronominal anaphora. In Amsterdam series on child language development, vol. 5, ed. W. Philip and F. Wijnen. Amsterdam: Institute of General Linguistics at the University of Amsterdam.Google Scholar
  70. Pollard, C., and I. Sag. 1992. Anaphors in English and the scope of the binding theory. Linguistic Inquiry 23: 261–305.Google Scholar
  71. Postal, P. 1974. On raising. Cambridge, MA: MIT Press.Google Scholar
  72. Reinhart, T. 1983. Anaphora and semantic interpretation. London: Croom Helm.Google Scholar
  73. Reinhart, T. 2004. Processing or pragmatics? Explaining the coreference delay. In The processing and acquisition of reference, ed. T. Gibson and N. Perlmutter. Cambridge, MA: MIT Press.Google Scholar
  74. Reinhart, T. 2006. Interface strategies. Cambridge, MA: MIT Press.Google Scholar
  75. Reinhart, T., and E. Reuland. 1993. Reflexivity. Linguistic Inquiry 24: 675–720.Google Scholar
  76. Reuland, E. 1994. Commentary: The non-homogeneity of Condition B and related issues. In Syntactic theory and first language acquisition: Cross-linguistic perspectives, Binding, dependencies and learnability, vol. 2, ed. B. Lust, G. Hermon, and J. Kornfilt, 227–246. Hillsdale: Erlbaum.Google Scholar
  77. Reuland, E. 2001. Primitives of binding. Linguistic Inquiry 32: 439–492.CrossRefGoogle Scholar
  78. Reuland, E. 2008. Anaphoric dependencies: How are they encoded? Towards a derivation-based typology. Ms., OTS, Utrecht University, Utrecht, the Netherlands.Google Scholar
  79. Reuland, E. 2011. Anaphora and Language Design. Cambridge, MA: MIT Press.Google Scholar
  80. Reuland, E., and M. Everaert. 2000. Deconstructing binding. In Contemporary syntactic theory, ed. M. Baltin and C. Collins, 634–670. Oxford: Blackwell.Google Scholar
  81. Rizzi, L. 1978. A restructuring rule in Italian syntax. In Recent transformational studies in European language, ed. S.J. Keyser, 113–158. Cambridge, MA: MIT Press.Google Scholar
  82. Rizzi, L. 1982. Issues in Italian syntax. Dordrecht: Foris.Google Scholar
  83. Rizzi, L. 1986. On the status of subject clitics in romance. In Studies in romance linguistics, ed. O. Jaeggli and C. Silva-Corvalan, 391–419. Dordrecht: Foris.Google Scholar
  84. Rizzi, L. 1990. Relativized minimality. Cambridge, MA: MIT Press.Google Scholar
  85. Rizzi, L. 2004. Locality and the left periphery. In Structure and beyond, ed. A. Belletti, 223–251. New York: Oxford University Press.Google Scholar
  86. Ross, J.R. 1982. Pronoun Deleting Processes in German. Paper presented at the annual meeting of the Linguistic Society of America, San Diego, California.Google Scholar
  87. Ruigendijk, E. 2008. Reference assignment in German Preschool Children. In Proceedings of GALA 2007, ed. A. Gavarró and J. Freitas, 370–380. Cambridge: Cambridge Scholars.Google Scholar
  88. Ruigendijk, E., N. Friedmann, R. Novogrodsky, and N. Balaban. 2009. Symmetry in comprehension and production of pronouns: A comparison of German and Hebrew. Presented in GALA 2009, Lisbon, Portugal.CrossRefGoogle Scholar
  89. Sag, I. 1976. Deletion and logical form. Doctoral diss., University of Maryland, College Park.Google Scholar
  90. Sigurjónsdóttir, S., and P. Coopmans. 1996. The acquisition of anaphoric relations in Dutch. In Amsterdam Series on Child Language Development, ASCLD 5. Amsterdam: Instituut Algemene Taalwetenschap 68.Google Scholar
  91. Sigurjónsdóttir, S., and N. Hyams. 1992. Reflexivization and logophoricity: Evidence from the acquisition of Icelandic. Language Acquisition 2: 359–413.CrossRefGoogle Scholar
  92. Spenader, J., E.-J. Smits, and P. Hendriks. 2009. Coherent discourse solves the pronoun interpretation problem. Journal of Child Language 36: 23–52.CrossRefGoogle Scholar
  93. Sportiche, D. 1996. Clitic constructions. In Phrase structure and the Lexicon, ed. J.J. Rooryck and L. Zaring, 213–276. Dordrecht: Kluwer.Google Scholar
  94. Szabolcsi, A. 2001. The syntax of scope. In The handbook of contemporary syntactic theory, ed. M. Baltin and C. Collins, 607–633. Oxford: Blackwell.CrossRefGoogle Scholar
  95. Thornton, R. 1990. Adventures in long-distance moving: The acquisition of complex wh-questions. Doctoral diss., University of Connecticut, Storrs.Google Scholar
  96. Thornton, R., and K. Wexler. 1999. Principle B, VP-ellipsis, and interpretation in child grammar. Cambridge, MA: MIT Press.Google Scholar
  97. Thráinsson, H. 1976. Reflexives and subjunctives in Icelandic. In Proceedings of the Sixth Annual Meeting of NELS, L’Association linguistique de Montreal, Université de Montreal, Montreal, Quebec, 225–239.Google Scholar
  98. Utakis, S. 1995. Quantification and definiteness in child grammar. Doctoral diss., CUNY, New York.Google Scholar
  99. Van der Lely, H., and L. Stollwerck. 1997. Binding theory and specifically language impaired children. Cognition 62: 245–290.CrossRefGoogle Scholar
  100. Van Gelderen, E. 2000. A history of English reflexive pronouns: Person, self, and interpretability. Amsterdam: Benjamins.Google Scholar
  101. Varlakosta, S. 2000. Lack of clitic-pronoun distinctions in the acquisition of Principle B in child Greek. In Proceedings of the 24th Annual Boston University Conference on Language Development, ed. S.C. Howell, S. Fish, and T. Keith-Lucas, 738–748. Somerville: Cascadilla Press.Google Scholar
  102. Varlakosto, S., and J. Dullaart. 2001. The acquisition of pronominal reference by Greek-Dutch bilingual children: Evidence for early grammar differentiation and autonomous development in bilingual first language acquisition. In Proceedings of the 25th Annual Boston University Conference on Language Development, ed. A.H.-J. Do, L. Dominguez, and A. Johansen, 780–791. Somerville: Cascadilla Press.Google Scholar
  103. Verbuk, A., and T. Roeper. 2010. How pragmatics and syntax make Principle B acquirable. Language Acquisition 17: 51–65.CrossRefGoogle Scholar
  104. Ward, G. 1983. On Nonreflexive Pronouns in Reflexive Environments. Penn review of Linguistics 7, 12–19.Google Scholar
  105. Wexler, K., and Y-C. Chien. 1985. The development of lexical anaphors and pronouns. Papers and Reports on Child Language Development (PRCLD), Stanford University, 138–149.Google Scholar
  106. William, E. 1977. Discourse and logical form. Linguistic Inquiry 8: 101–139.Google Scholar
  107. Zesiger, P., L. Chillier-Zesiger, M. Arabatzi, L. Baranzini, S. Cronel-Ohayon, J. Franck, H.-U. Frauenfelder, C. Hamann, and L. Rizzi. 2010. The acquisition of pronouns by French children. A parallel study of production and comprehension. Applied Psycholinguistics 31: 571–603.CrossRefGoogle Scholar
  108. Zribi-Hertz, A. 1989. Anaphor binding and narrative point of view: English reflexive pronouns in sentence and discourse. Language 65: 695–727.CrossRefGoogle Scholar

Copyright information

© Springer Science+Business Media B.V. 2011

Authors and Affiliations

  1. 1.Fakultät III, Institut für Fremdsprachenphilologien, Seminar für Anglistik/AmerikanistikCarl von Ossietzky Universität OldenburgOldenburgGermany

Personalised recommendations