Towards a definition of SUBJECT in binding domains and subject-oriented anaphors

This paper is concerned with the notion of SUBJECT and with providing a definition couched in Minimalist terms. The question of subjecthood has long been an area of serious debate. In current versions of Minimalism, subjects do not have primitive status and can only be defined in derived terms. However, subjects and the broader theoretical notion of SUBJECT remain important in linguistic description. This paper develops a definition of subjecthood in terms of set-theoretic notions of functional dependency: when a feature, say f, determines the value of some other feature, say uf. This notion is used to describe various phenomena where subjecthood has been invoked, namely binding domains and subject-oriented anaphors.


THE PROBLEM OF SUBJECTS
This paper is concerned with the notion of SUBJECT and with providing a definition couched in Minimalist terms.SUBJECTs are indispensable for the Binding Theory where they define domains for anaphors and pronouns (Chomsky 1981).
(1) a. i. Principle A: An anaphor must be bound within its governing category.
ii. Principle B: A pronominal must be free within its governing category.
iii.Principle C: An R-expression must be free (Chomsky 1981).
b. B is a governing category for A if and only if B is the minimal category containing A, a governor of A, and a SUBJECT accessible to A (Chomsky 1981).
The following pair of examples shows that anaphors must be bound, and that pronouns must be free, within a domain delimited by a subject (2) a.The twins said that || Sub he liked *each other/them b.He said that || Sub the twins liked each other/*them The paradigm can be extended to binding within DPs.A possessor defines a binding domain.
When the possessor is present (3a) then the anaphor must be bound within the DP -and the pronoun must be free within the DP.When the possessor is absent (3a), then the anaphor must be bound in the domain defined by the clausal subject -and the pronoun must be free in this domain.
( Thus, the possessor counts as a SUBJECT as far as the Binding Theory is concerned.

The difficulty of defining subjecthood
The problem with using SUBJECT as a primitive of the Binding Theory is that it is difficult to define adequately -a problem that stems from the difficulty of defining subjecthood more generally. 2Although subjecthood is an essential descriptive device in linguistics, it is not clear what it is derived from at a theoretical level.Over the years, various prototypical, non-exclusive properties of subjects have been proposed.None are either necessary or sufficient.The following is a non-exhaustive list that illustrates the extent of the probem.Subjects may: (4) a. be involved in predication b. be agents c. determine agreement on a predicate d. be located in Spec TP e. have Nominative case f. be linked to EPP phenomena g. be the highest argument of a VP (i.e.there is only one of them, it will typically precede other arguments etc) h. be antecedents for subject-oriented anaphors (SOAs) (e.g.Maling (1984)).
However, none of these diagnostics appear to be necessary or sufficient -there are putative counter examples to all of them.
One of the oldest notions of subjecthood was its link to predication.However, not all predication structures are domains for Binding Theory.Example (5) contains a small clause predication structure.However the anaphor can be bound by the clausal subject i.e. the subject of the small clause predicate does not appear to be a domain for binding.
(5) Dr. Robert Bruce Banner i considered [ SC the Incredible Hulk (to be) a clone of himself i /*him i ] 3 Similarly, subjects also cannot be defined purely in semantic terms.Although subjects are often agents, there are examples where they are not.In many Bantu languages, a semantic object may occur in subject position and determine agreement.Also, in the following English passive sentence, the subject is a theme.
(6) A cat was seen The agreement diagnostic raises the question of agreeing objects in languages with object agreement (e.g.many Bantu languages) and the DPs associated with postpositions etc.The agree-ment diagnostic also suggests that in the following example, the DP a cat is the subject, raising questions about the status of the expletive there. 4  (7) There is a cat at the door A related diagnostic is that the subject be located in Spec TP (i.e. the EPP holds).However, this is also problematic as the previous example demonstrates: an expletive is in Spec TP (an indicator of subjecthood), whereas agreement is determined by the indefinite DP (also an indicator of subjecthood). 5  The Case diagnostic can also lead to confusing results.In some languages (e.g.Korean), there can be more than one nominative DP in a clause.In languages with 'quirky' case (e.g.Icelandic) a DP (which otherwise conforms with other properties of subjects) may be marked with Dative or a default case other than Nominative.In addition, the possessor in (3) has Genetive case, not Nominative.
Another property is that the subject is the highest argument of a VP.However, if Nominative Case defines subjecthood, then this cannot be true in multiple Nominative constructions (e.g.

Korean)
. More often than not, the subjecthood of the highest argument is stipulated (e.g. in the argument list (HPSG,LFG) -in other words it is a theory-internal assumption.
A related issue is the Extended Projection Principle (EPP) which ensures that every clause has a subject.However, there is as yet no consensus on what the EPP is or even if it exists (Boeckx 2000a, Martin 1999) and it holds little explanatory value.At best, EPP ensures that the highest argument will move to subject position.
The final property I will discuss here is the fact that in some languages (e.g.Icelandic, Dutch), there are a subset of anaphors which can only be bound by subjects.While this has been used as a diagnostic for subjecthood (e.g.Maling (1984), Zaenen et al. (1985)) it is unclear what actually determines this binding behaviour or why subjects should be the sole antecedents for some anaphors but not others.The upshot is that this phenomenon is a diagnostic, and is defined circularly: a subject-oriented anaphor is bound by a subject -a subject can be an antecedent for a subject-oriented anaphor.
To summarize, although SUBJECT is important for the Binding Theory, it is not clear how SUBJECT is related to subjecthood more generally, or even what subjecthood reduces to at a theoretical level.In some frameworks, such as LFG, Relational Grammar and HPSG, subjecthood is stipulated.In the frameworks of P&P and the Minimalist Program, on the other hand, there has been a sustained attempt to sidestep the problem of subjecthood by deriving it from more fundamental properties.In this venture, I think that these frameworks have been largely successful although significant problems remain.For instance many properties of subjects are derived from an interaction of locality constraints on movement (thus the DP that moves to Spec TP will always be the highest DP in the VP etc).Similarly, Nominative case is regarded as a reflex of Tense (Pesetsky and Torrego 2001).This paper continues the tradition by proposing a theoretical basis for SUBJECT flowing from relational theory (Codd 1970).

BASIC ASSUMPTIONS
The following paper requires four main assumptions.The proposal is broadly couched within the Minimalist Program ( (Chomsky 1995b) and subsequent works) although some of the conclusions may diverge from some of the later versions of this framework.

Agree
Agree is asymmetric (Chomsky 1995b:277-279).Pairs of uninterpretable and interpretable features are mediated by AGREE a pairwise relationship between a PROBE and a GOAL where uninterpretable features on the probe are valued by the equivalent interpretable features on the goal (Chomsky 2000) yielding an ordered pair (GOAL, PROBE).In other words, the value of the goal e.g.φ determines the value of the probe uφ.In particular, I assume that Nominative case is a manifestation of uT on nominals checked by the corresponding T feature on the tense head (Pesetsky and Torrego 2001).

Anaphors
Since a large part of this paper will be a discussion of anaphors, I will outline some basic assumptions here.There is no equivalent of binding within the Minimalist Program (Chomsky 1995b) so, as a starting point, I assume traditional Binding Theory (Chomsky 1981) and specifically, principles A and B and the notion of governing category (1).
Concerning the feature specification of anaphors themselves, anaphors are traditionally specified as [+ANAPHORIC], a feature taken to be mnemonic for the referential defectiveness of the anaphor (Chomsky 1981, Reinhart and Reuland 1991, Thráinsson 1991).I take anaphors to lack (a subset of) appropriate φ features; φ features, and ultimately referentiality, is supplied by the antecedent mediated by a command relation. 6

Phrase structure
Concerning phrase structure, I start from the proposition that A merged with B yields a partially ordered set {A,{A,B}} (Chomsky 1995a, Devlin 1993, Halmos 1960).In particular, I assume that such a structure is unambiguous and can represent a single relationship.(Armstrong 1974, Beeri et al. 1977, Sagiv et al. 1981).
Intuitively, this means that if X selects a complement Y, then X functionally determines Y i.e.X → Y. Similarly, if Z agrees with W, then the morphological form of W (e.g.3SG) is determined by the properties of Z (i.e.3SG) i.e.Z → W. I take it as a fundamental assumption that phrase structure can be expressed in terms of functional dependencies. 9It is important to note that functional dependencies and their properties, including transitivity, are not contentious within Relational Theory.The only novelty about the current approach is that I propose to apply Functional dependency to syntactic relationships.
To summarize, this section has proposed that syntactic relations such as selection and AGREE can be represented by functional dependencies, a basic relationship derived from Relational Theory and Set Theory.In addition to their mathematical grounding independently of linguistic theory, functional dependencies are not actually an assumption per se -rather functional dependencies are a natural consequence of a set-theoretic approach to phrase structure (Chomsky 1995a).functional dependencies must therefore be regarded as a deep property of linguistic theory. 10

ARGUMENTS IN FAVOUR OF A FUNCTIONAL DEPENDENCY APPROACH TO ANAPHORIC DOMAINS
There are four main arguments for a functional dependency approach to anaphoric domains.
The first argument, set out in the previous section, is theoretical -functional dependencies follow from standard assumptions about phrase structure and syntactic relationships; if the starting point of Chomsky (1995a) is true, then functional dependencies are necessarily inherent in linguistic structures.The second argument is based on the fact that functional dependencies provide a way of distinguishing subjects from non-subjects -a distinction that has remained important in linguistics despite its resistance to formalization.The third argument for FDs is that they allow the integration of possessive and clausal subjects under a single banner.

Functional dependencies distinguish subjects from non-subjects
The usefulness of functional dependencies is that they can be used to distinguish subjects from non-subjects.At the heart of this idea is the notion that agreement can instantiate a functional dependency as explained in the previous section.Consider the relationships present in the following LF representation.I will assume a vP shell structure where light verbs introduce verbal arguments and where V-v raising takes place (Larson 1988).The subject DP has moved from Spec vP to Spec TP.Any DP which functionally determines some feature can be regarded as a SUBJECT.Typically, such features will be formal features.
(11) a. SUBJECT: A DP which functionally determines a φ feature is a SUBJECT.
b. Anaphoric Domain: The minimal domain containing an anaphor, a potential binder and a DP which functionally determines the anaphor.
It has long been known that agreement determines binding domains.For instance, (Leland and Kornfilt 1981) explored a Turkish dialect with agreeing and non-agreeing infinitives.Nonagreeing infinitives could not bind anaphors in their complement; agreeing ones could.This insight was incorporated into the Binding Theory of Chomsky (1981).Hitherto there has not been any reason why this should be the case; the present paper provides the theoretical framework to formalize this insight.

Possessive DPs as subjects
This paper began with the problem of defining SUBJECT as it pertains to binding domains.
Having provided an elegant definition of SUBJECT and binding domain in (11), I will now demonstrate how this definition fares with respect to the data.
(12) a. || Sub The twins expected that || Sub [I] would help *each other/them In situations where the SUBJECT is also the clausal subject, the data are easily explained.Since the clausal subject will always agree with T, the clausal subject will always functionally determine φ features of T and will consequently always be a SUBJECT.Clausal subjects will thus always be a domain for binding.
In section (1) it was shown that possessors are SUBJECTs.In (13a), there is no possessor/subject and the reciprocal can be bound by the sentential subject, they.In (13b), in contrast, a possessor/subject is present within the DP and induces a domain; the reciprocal cannot be bound by the sentential subject since it now lies outside the anaphoric domain.These examples show that the anaphoric domain is defined by the presence of an overt SUBJECT.1995:184-185) These data are puzzling from a traditional perspective.First there is the problem of why a possessor NP should count as a subject at all since this is not an intuitive idea (i.e. the possessive NP is optional, not necessarily agentive, does not determine agreement in English etc).Second, if one requires a subject position or field to determine a domain, then there is clearly a position available whether it is filled or not.Thus, in contexts where there is no possessor NP, it is not necessarily obvious that there should also be no anaphoric domain. 11 The definition of SUBJECT in terms of functional dependencies (11) immediately make the prediction if possessors are anaphoric domains then the non-clausal possessive pronoun should functionally determine its complement in the same way that a clausal DP subject functionally determines agreement on T. ( However, in English it is not immediately clear that the English possessive functionally determines its complement.There is a range of research which shows that DPs parallel the architecture of clauses (Szabolcsi 1983;1994).But making the claim that Spec NP is analogous to Spec TP does not really provide any deep explanations; why should N and T be analogous since at a feature level they are quite different? 12What is it about the relationship between DP and T and N respectively that makes the DP a subject?
The functional dependency proposal makes a clear prediction: the DP in Spec TP and Spec NP, should functionally determine features on T and N respectively.The crucial evidence for func-tional dependency comes from Hungarian where overt agreement occurs between a possessive and its complement.In ( 16), subject marking -m occurs in both clausal and possessive contexts.The possessive pronoun determines agreement morphology on N in the same way that DP subjects determine agreement morphology on T. This shows that SUBJECTs in both these contexts can be unified by the fact that both functionally determine their complements.This is a very important result; it confirms that the central characteristic underlying binding domains is functional dependency -in this case, expressed by means of agreement.

SUBJECT ORIENTED ANAPHORA
The next section deals with another issue in binding that is not predicted at all by Standard Binding theory: subject-oriented anaphora.The example in (18a) has an antecedent that is neither local, nor the closest c-commanding antecedent, namely John.In addition, it is possible for a local antecedent, namely Bill to bind the anaphor.Importantly, both possible antecedents are subjects of their respective clauses; the non-subject, namely Sam, cannot be an antecedent, contrary to what is predicted by Principle A. The configuration schematically represented in (18b).The data are similar to SOAs in many other languages including Icelandic, Norwegian and Danish. 13  Long-distance anaphors have several characteristics in common (Cole and Hermon 2005, Koster and Reuland 1991, Pica 1986;1991) What these correlations suggest is that SOAs are subject to strong cross-linguistic principles and that there must be some syntactic operation which can distinguish subjects from non-subjects.
In this paper, I have demonstrated that functional dependencies can make this distinction and I would like to propose that functional dependencies are responsible for SOA phenomena.

SOAs are not logophors
It might be claimed that SOAs are simply logophors.Logophors, seem to be determined by discourse and prominence factors rather than structural configurations.In all these examples, the logophoric element is licensed by antecedents which are not necessarily represented in the syntactic structure and are not necessarily subjects.
There are several reasons to claim that SOAs are distinct from logophoric phenomena.First, example ( 17) is an SOA that is obligatorily local.With the exception of its local character, it conforms with the properties in (19).Since logophors are characteristically non-local, SOAs cannot all be logophors.Another reason to exclude logophors from this category is that logophors can operate at arbitrary distances from the antecedent, may not necessarily have an antecedent at all and need not be in a c-command relationship with the antecedent.In contrast, SOAs must be bound by a c-commanding antecedent, an antecedent is obligatory and the antecedent must be within a domain typically defined by the tensed clause (Koster and Reuland 1991) (again, abstracting away from the local character of Dutch zich).Koster and Reuland (1991) suggest that there are three domains for binding (a) the local domain for himself type anaphors (b) a medium-range domain for SOAs and (c) a larger domain for logophors.All these properties suggest that SOAs are distinct from logophors.Finally, SOAs are cross-linguistically morphologically simplex (as opposed to local anaphors like himself ).There is no such restriction on logophors, which can be complex.
For these reasons, I do not think that lumping SOAs together with logophors is the right move.Doing so would obscure strong cross-linguistic correlations.Consequently, I will continue to treat SOAs as a distinct set of anaphoric possibilities.By excluding logophors it is also possible to make the claim that the strong subject-oriented character of SOAs must be derived from some deeper principle of grammar.

Previous analyses of SOAs
It has been proposed that SOAs are derived by head movement (Huang andTang 1991, Pica 1986).SOAs are always monomorphemic and are thus consistent with head status.It has been proposed that such head can adjoin to the subject.
However, this analysis cannot be correct for all SOAs, especially those found in the Germanic languages.The central criticism of the approach stems from generalization (19v).In languages with no agreement (e.g.Chinese), SOAs are subject to a 'blocking' effect.Long distance antecedents are only possible if the long-distance antecedent agrees with the possible antecedents beneath it.This has been used as a diagnostic for a head-movement analysis of these anaphors (Cole and Hermon 2005, Cole et al. 1993, Huang and Tang 1991, Huang 1996, Pica 1986).
( This movement also places the anaphor within the domain of the higher subject John and so ziji can also be bound by John.Importantly, however, since the anaphor is already bound by Bill, the higher subject must match the features of the lower subject.
This analysis is not available for the Germanic languages since the blocking effect is not visible and these languages also have morphological reflexes of subject-verb agreement.
( 'Jón i says that you love him i ' (Sigurðsson 1990:309) Another argument against the universal validity of the head-movement analysis is that SOAs can occur in islands in Icelandic (Thráinsson 1991:57).The examples in (28a,b) show that an anaphor can occur in a context where WH extraction is not possible.This militates against an analysis that involves movement of the anaphor.These arguments show that not all SOAs reduce to head-movement.
( Traditionally, anaphors are bound by a command relation: c-command. 15Although it is usually assumed that c-command is the sole command relation available to narrow syntax, if the results of section 3 are correct, then there must be another command relation based on functional dependencies.Recall that functional dependencies follow from standard assumptions about phrase structure and syntactic relationships.These assumptions lead to the conclusions that functional dependencies are inherent in syntactic representations.Thus, it would be very surprising if narrow syntax did not make use of them.So I would like to propose that there is a typological distinction between those anaphors that are bound by C-command and those that are bound by SUBJECT defined in terms of Functional Dependency. (

SOAs zich etc
Importantly, however, the notion of SUBJECT is central.If one assumes the existence of SUB-JECT for the purposes of the standard Binding Theory then one must also accept it for the purposes of SOAs.All I have done in this paper is to provide a formal account of what a SUBJECT is.
SOAs sensitive to Functional Dependency will only be bound by the subject and never a nonsubject.They may or may not be local -modulo minimality constraints on intervening antecedents. 16  C-command anaphors however will always be bound by a c-commanding antecedent. 17If the reflexive is lexically specified as being functional dependency-bound, then the matrix subject Petur is a possible antecedent.The grammatical object Jens cannot be a possible antecedent.The PRO 'subject' of the embedded clause is not an appropriate antecedent -since PRO probably does not functionally determine agreement features on T, PRO does not Functional Dependency T or the complement of T (although it is functionally determined by T since it receives null case).It is likely that infinitival clauses differ from language to language in this respect.It has been argued that two types of infinitives exist based on independent time reference or lack of it (Stowell 1982).In addition, some languages have overt inflection in infinitives (e.g.Portuguese).These intriguing issues await for future research.

EVIDENCE FOR FUNCTIONAL DEPENDENCIES
In addition to the conceptual argument for functional dependencies, the treatment of functional dependencies in binding domains, and the discussion about SOAs, in this section, I will provide some additional arguments in favour of the analysis I have proposed.

Object agreement
A prediction of the functional dependency approach is that if the agreement between a DP and T constitutes a SUBJECT and ultimately licenses SOAs, then languages with object agreement should allow the object to bind SOAs.The reason for this is that the agreeing object will act as an intervener between an agreeing subject and a SOA -in effect, the agreeing object will act as a kind of SUBJECT. ( Agr Agr

×
First note that the functional dependency approach does not make predictions about the availability of discourse logophors.As I have done throughout this paper, I will focus exclusively on SOAs (but see section 5.3.2 for a brief discussion).
An initial survey provides circumstantial evidence in favour of this prediction.Languages with object-agreement do not have exclusively SOAs.However, this argument is incomplete because many object-agreement languages, such as Mohawk, do not have anaphoric NPs a at all (Baker 1996;2003) and consequently cannot shed light on the distribution of SOAs.Similarly, many Bantu languages have reciprocal markers on the verb itself which affects the way binding operates in those languages.Nevertheless, there exist languages such as Georgian, Hungarian and Basque that do have object agreement and also have anaphoric NPs (Amiridze (2006), Everaert

Italian agreement and binding
More direct evidence for this position comes from Italian. 18Italian is a very useful language because it does not have object agreement and thus licenses SOAs like proprio.This lexeme is a subject-oriented anaphor when it acts as a LDA (Giorgi 1991:186); when it is bound locally within its clause, then it can be bound either by subject or the object.This dual character of proprio makes it ideal for exploring the interaction of agreement and binding.In fact, it can be shown that the antecedent of proprio is, in part, determined by agreement -and thus by functional dependency.
Example (34a) shows an object with a PP complement containing a reflexive.Only the grammatical subject is an appropriate antecedent for the reflexive.The preposition is underlined for ease of reference.In contrast an adjectival complement (34b)(underlined) does allow the object as a suitable antecedent for the reflexive.REFL's corso class 'Intended: Gianni saw the professor contents who followed his class' (Giorgi 1991:188) The data is supported by evidence that DP complements (35) (underlined) pattern with adjectives and not PPs.The general structure of these examples is illustrated in (36).Although the structures are identical, only for AP and NP complements does the NP small-clause 'subject' functionally determine the complement as evidenced by overt agreement.The dotted line informally illustrates the functional dependency.The broad generalization with these data is that in contexts where the XP complement agrees with the small clause SUBJECT (i.e.APs and DPs but not PPs) then a reflexive is licensed.
In other words, for SCs with PPs, the DP is not a true SUBJECT at all because it does not functionally determine PP.This is strong evidence for a functional dependency approach.

Icelandic anaphors
The following section examines putative counter evidence to the main proposal in this paper.It will be shown that the proposal makes a prediction about the nature of Icelandic quirky subjects which turns out to be true.The Icelandic data will thus ultimately support the analysis.
The proposal thus far is as follows: a SOA is bound by a SUBJECT defined by functional dependency (11).If a DP functionally determines its sister then it can be an antecedent for an SOA.The presence of the functional dependency is often indicated by agreement where the DP determines some feature on its sister.As far as the counter-evidence goes, first note, that Icelandic has SOAs (Maling 1984) (37).In these examples, the reflexive possessive sinni can only be bound by the subject.
The Nominative DP cannot bind a SOA.
(  (Maling p.c.)In example (38), the subject henni is marked with dative case.It is this quirky subject which is the antecedent of the SOA sig.The nominative DP uplýsingarnar cannot bind the anaphor.
The verb has plural agreement with the nominative DP.This constitutes counter-evidence to the proposal.This means that the Icelandic data are an important test-case for the current proposal.

Icelandic dative agreement
This is potentially problematic for the approach to SOAs proposed int this paper.I have claimed that SOAs are sensitive to functional dependencies.The data in (38) clearly show that it is the nominative DP which determines agreement on the verb and yet, the quirky dative-case-marked DP is the antecedent of the SOA.If the current proposal is to be sustained, then there must be agreement between the quirky DP and the verb.Note that agreement need only be with a single feature in order to constitute a functional dependency (see definition of Functional Dependency 9).
(40) Prediction of the functional dependency approach to SOAs: Icelandic quirky dative-case-marked DPs must functionally determine the verb if they are to serve as antecedents for SOAs.
A closer look at the Icelandic data confirm this prediction.First it should be noted that agreement between the verb and the nominative 'object' does not display the clear-cut paradigm associated with prototypical subject-verb agreement (Boeckx 2000b, Sigurðsson 1996, Taraldsen 1995): 'the facts get murky' (Boeckx 2000b:357) (Boeckx 2000b:360) Example (41a) shows that a 3PL nominative object triggers full agreement on the verb.However (41b), shows that when the nominative object is 1PL then agreement fails.In fact, the sentence is ineffable.This contrasts with the clear-cut and consistent instances of agreement between a nominative subject and the verb in canonical finite clauses.The data suggest that agreement with the nominative object agrees with the verb in terms of number features.But agreement in terms of person is subject to additional constraints.
The problem is further illustrated with raising contexts.When there is more than one dative DP in the clause, then agreement with the Nominative DP is not as clear cut.It appears that the dative DP of the embedded clause can determine agreement on the matrix raising verb to some extent.Thus quirky dative subjects are not inert for agreement.Although there is no actual morphological spellout of this agreement, it is proposed by Boeckx (2000b) that the agreement between the dative subject and the verb cannot be morphologically realized because of the complex relationship between T and nominative case checking.
By contrast, when the nominative object is first or second person, then the resulting sentences are predicted to be completely ungrammatical.This is born out by the facts.In (46a,b,c) the examples are all ungrammatical because the nominative object is not 3rd person.This blocks person agreement between the dative subject and the verb.
(  (Sigurðsson 1996) in Boeckx (2000b:367) The analysis of Boeckx (2000b) demonstrates that the quirky DP functionally determines person features on the verb.This is sufficient to construe the quirky DP as a SUBJECT in terms of the definition in (11).Consequently, it is predicted that the quirky DP can function as an antecedent for SOAs, as demonstrated by (38).Thus, although the Icelandic data initially seemed problematic for the proposal, they ultimately follow from it.

Features responsible for SOA binding
There is a question that has been implicit throughout this paper: exactly what features involved in functional dependency are responsible for licensing SOAs?The definition of SUBJECT in (11) does not shed light on this matter.However, to the extent that the analysis of quirky subjects in Icelandic is correct (section 5.3.1), the Icelandic data suggest that the feature responsible is PERSON.
(47) The antecedent of an SOA must be a DP which functionally determines the anaphor with regard to (at least) person features.This makes intuitive sense if one considers the PERSON specification as a speaker perspective (Boeckx 2000b, Levinson 2000).First person identifies the speaker, I; second person identifies the animate hearer; third person is for the rest.First person must always be a self-concious agent capable of locution (i.e.+animate).Second person is less agentive, being consistent with passive listeners, but nevertheless must always be an entity capable of comprehension (i.e.+animate). 19  Some researchers (e.g.Hellan (1991)) have posited the notion of 'perspective command' to account for the distributions of SOAs and logophors.For instance, Sigurðsson (1990) shows that Icelandic SOAs invoke "reference to a secondary ego from this ego's point of view" (Sigurðsson 1990:328).Hitherto there was no obvious way of defining this notion formally.However using functional dependency command to define this notion makes a prediction: the antecedents of SOAs should be obligatorily animate -at least in Icelandic given the analysis in section 5.More suggestive evidence comes from Afrikaans, a West-Germanic language that, unlike its close cousins Dutch, German and the North-Germanic languages, does not have a subjectoriented zich pronoun.Interestingly, in colloquial Afrikaans (especially the Orange-River dialects), weather verbs can be introduced with a masculine pronoun hy.(Fourie 1985:110: Riemvasmaak Afrikaans) This strongly suggests that the animacy feature associated with person is not present for agreement in the syntax -otherwise hy would refer to an animate referent. 20This is consistent with the complete absence of any kind of subject-verb agreement in Afrikaans.
In fact, this sheds light on why so many languages with SOAs also have logophors (the table in example -33 on page 21).Logophors are also oriented towards speaker perspective (Hellan 1991).Thus, there is a similarity between SOAs and logophors generally.
The key difference is that logophors are subject to a pragmatic construal of (speaker) perspective, whereas syntactic SOAs are subject to the grammaticalization of that perspective, namely person features. 21

CONCLUSION
This paper provides a definition of SUBJECT in terms of functional dependencies in the spirit of the Minimalist Program.I have argued that functional dependency is an important theoretical device that follows directly from standard assumptions about phrase structure and syntactic relationships.Narrow Syntax utilizes this tool to determine domains of anaphors and derive the subject-orientation of some types of anaphors.
In addition to the theoretical argument, have provided a variety of arguments for functional dependency from a variety of areas.First, functional dependencies were motivated on conceptual grounds and it was shown that they follow from basic assumptions about phrase structure, agreement etc.It was then demonstrated that functional dependencies could be used to define domains for local anaphora.The argument for functional dependencies was then extended to SOAs where it was shown that functional dependencies provide a means of accounting for (i.e. the EPP, a putatively central feature of subjects is not active inside English DPs). 13There are, however, differences between the SOA phenomena in these languages.See section 4.2. 14A higher subject can only bind a LDA if the lower subject agrees in person features (see also Cole and Hermon (2005), Huang and Tang (1991)).
15 But see 'Hellan (1991) for other possibilities. 16In fact, given my argument that agreement can constitute a Functional Dependency, this claim is consistent with the assumption that anaphors have uninterpretable φ features which must be checked.The difference between AGREE (technically defined in terms of C-command) and and FD is that only the subject's features functionally determine the anaphor, whereas, it is conceivable that any intervening DP could AGREE with it.
17 Parameterization implies the existence of a feature with the value [+functional dependency-BOUND] and another feature with the value [+C-BOUND].
18 Czech also has similar constructions (Toman 1991). 19Abstracting away from instances when non-human and non-animate objects are imbued with the human-like qualities of comprehension e.g. in fairy-tale contexts.
20 This is merely a claim about what kinds of features are available in the syntactic computation; I am not implying that animacy may not have a semantic role in Afrikaans.
21 An interesting question for future research would be to ascertain whether there is a diachronic grammaticalization cline between discourse logophors and syntactically bound SOAs.
) a.The twins liked [|| Sub John's pictures of *each other/them] b.The twins took [|| Sub ∅ pictures of each other/*them] dependencies in this tree, some of which are informally represented by arrows.Since selection is an instantiation of functional dependency, T will functionally determine v and by the Transitivity rule (9c) everything contained in v. Similarly, the topmost v will functionally determine the lower verb shells and everything contained in them.Now consider the DPs within the verb shell.V and each of the light verbs each select a DP argument and assign Θ-roles.Thus, each of the argument DPs is functionally determined by a verbal head.Within the verb shell, none of the DPs functionally determine any other element.This situation changes dramatically when one considers the status of the DP in Spec TP.T assigns Case to the subject DP and consequently functionally determines it.However, the DP also determines agreement on T. Thus the subject DP functionally determines T and, by the Transitivity rule (9c), everything contained within it.Thus, functional dependency exposes an asymmetry between subject DPs and other DPs.Subject DPs are functional determiners; non-subjects are functionally determined and do not functionally determine any other element in the representation.It is this asymmetry which I propose underpins the notion of SUBJECT.

( 30 )
Principle A: An anaphoric NP must be A1: c-command bound in a binding domain δ or [Local himself -type] A2: Functional Dependency bound [SOA: zich type] Domain: The domain δ is the first Functionally Determining DP.[see(11]This accounts for typical SOA examples.The following is from Icelandic.
the professor with the students who attended his class'(Giorgi 1991 3.1.It is worth pointing out that this restriction would remain puzzling under traditional Binding Theory.This prediction is confirmed by Maling (p.c.): a local reflexive (i.e. a not exclusively subjectoriented reflexive) in Icelandic may have an inanimate antecedent (48), but a long-distance reflexive must have an animate antecedent.covered/veiled all the suroundings in its thick smoke'(Maling p.c.) (Koster and Reuland 1991)always be long-distance anaphora, a notion contradicted by the Dutch data above.Although my analysis is applicable to SOAs generally, in the following section, I will concentrate on long-distance anaphors more specifically.Typically the domain for these longdistance SOAs is the minimal tensed clause (see(Koster and Reuland 1991)and references in that book).These SOAs do not seem to obey Principle A.
It will be argued that the central device underpinning this phenomenon is SUBJECT defined in terms of functional dependency.The discussion is adapted from (De Vos (2006)a) and (De Vos (2006)b).Subject-oriented anaphora (SOA) is a generic term that I will use to describe anaphoric phenomena that exclusively have a subject as an antecedent.The following is a Dutch example of a local, subject-oriented anaphor.The anaphor zich can only be bound by the clausal subject Jan.×Many other languages have SOAs that are bound by long-distance antecedents.In fact, .
He i thinks:) Maria was always so nasty.When Olaf j came, she would certainly tell himself i/ * j to leave.
a. [That we hang a picture of himself i on every wall] is one of the president ′ s i most outrageous demands [English:(Reinhart and Reuland 1991:317)] The local subject wo differs in features from the long-distance subject ni.The the SOA can only be bound by the local subject because the two subjects do not agree.This has been used as an indicator that the SOA ziji must adjoin to the local subject before it can be bound by the long-distance subject.
REFL taoyan hateMary MaryIn the schema in (26) the anaphor ziji adjoins to the local subject where it can be bound by Bill. .
It turns out that in the presence of a quirky dative-case-marked subject, blocks person agreement between the nominative DP and the verb.If the quirky DP induces minimality effects, then it must be the case that quirky DPs agree with verbs in person features.If the quirky DP had inert person features then no minimality effect would be apparent.The following schema applies.
The following example from the Afrikaans variety from Riemvasmaak is representative.