2. A TYPOLOGY OF ANAPHORS: THE BINDING CONDITIONS

Following Chomsky (Reference Chomsky1981, Reference Chomsky1986), it has been standardly assumed that there are two types of anaphors, syntactically speaking, one which must have a syntactically local antecedent, which includes reflexives and reciprocals, and the other which cannot have a local antecedent, including pronouns.

1. (1)

   1. a. Binding Condition A

      A reflexive must be bound in its minimal governing category (i.e. a reflexive must have a local antecedent).

   2. b. Binding Condition B

      A pronoun must be free in its minimal governing category (i.e. a pronoun cannot have a local antecedent).

The Binding Conditions entail the prediction that there will be no more fine-grained typology of reflexives, since all reflexives are predicted to behave alike, i.e. be locally bound (and all pronouns must be locally free). This works for pronouns and some of the Scandinavian reflexives.

Unless otherwise indicated, all sentences and judgements given in this paper come from the author's own fieldwork notes. Two reviewers of this paper requested that I comment on some of the methodological issues associated with collecting grammaticality judgements on anaphora. I include these comments in the appendix. Note that indices are used in (2) and (3) to show coreference and disjointness, but that bolding will be used henceforth to indicate coreference only. R is used to gloss reflexives, and self is used in the translations, in order to inhibit any tendency to assume that the reflexives under investigation have the same distribution as English anaphors.

There is syntactic complementarity in all of the Scandinavian languages between reflexives and pronouns as shown in (2) and (3), just as there is in English, at least in the local domain (and ignoring anaphors in PPs).

1. (2)

2. (3)

However, as is well-known, there is a problem with saying that reflexives must have syntactically local antecedents, since this claim is actually not true for many languages, including most of the Scandinavian languages. All of the Scandinavian languages allow a reflexive to take as its antecedent an NP that is not a coargument, that is to say that the antecedent is not the subject of the predicate that the reflexive is an object of.

The example in (4a), which contains a non-finite complement clause, is grammatical with a nonlocal antecedent (and a local antecedent, which would be instantiated here with Maria, but this is not the focus of this paper, and will be ignored henceforth) in all of the Scandinavian languages, while (4b), which contains a finite complement clause, is grammatical with an LDR only in the Insular Scandinavian languages and is ungrammatical in the Mainland Scandinavian languages. (Coreference is indicated by the anaphor and its antecedent set in boldface.)

1. (4)

Examples like those in (4a) have been typically accounted for by re-examining the definition of ‘local’ or, in technical terms, the ‘minimal governing category’. The conclusion has been that the minimal governing category can actually vary among anaphors, with the result that different reflexives can easily have different binding domains, although each reflexive will have a single overall domain in which it must find its antecedent.

As we can see in the examples above, all of the Scandinavian languages allow ‘middle-distance’ binding of reflexives (over a non-finite clause boundary), but only the Insular Scandinavian languages allow ‘long-distance’ binding (over a finite clause boundary) as in (4b). So the idea of a ‘minimal governing category’ (ignoring picture-NPs and the like) creates a hierarchy of binding domains. (As pointed out in Strahan (Reference Strahan, Svenonius, Bentzen, Jacobsen, Johannessen, Heycock, Parrott, Strahan and Vangsnes2009a:119), this often includes the paragraph domain at the end of this hierarchy, but here we will restrict ourselves to the sentential level.)

1. (5) Local/coargument < non-finite < finite

This hierarchy also applies to non-Scandinavian languages. It was called the Subset Principle by Manzini & Wexler (Reference Manzini and Wexler1987), as each domain is a subset of the next larger domain. Adherence to this hierarchy entails that any anaphor which must be bound in the coargument domain be necessarily bound in both the non-finite and the finite domains. This hierarchy is also used by Dalrymple (Reference Dalrymple1993) in her analysis of anaphora in Norwegian and Marathi.

At this point, we have covered much of the basic data concerning LDRs. In addition, LDR are said to require a grammatical subject as antecedent, which is certainly the case for all of the examples presented here so far. Before turning to the analysis by Dalrymple (Reference Dalrymple1993), I will first give a brief introduction to Lexical-Functional Grammar.

3. LDR IN LEXICAL-FUNCTIONAL GRAMMAR

Lexical-Functional Grammar (LFG) is a constraint-based, model-theoretic grammatical framework. It contains many different structures, such as information-structure, semantic-structure and discourse-structure, along with two syntactic structures: the constituent-structure (normally referred to as the c-structure) and the functional-structure (or f-structure). The c-structure contains annotated phrase structure trees, see (6b), while the f-structure contains attribute–value matrices, also referred to as feature matrices, as illustrated in (6 a). Note that only words are represented in the c-structure, such that TENSE is not an overt element there. It is in the f-structure that much of the work of syntax occurs, such as unbounded dependencies and binding, and where our focus will be in this paper. All constraints on all structures must be satisfied simultaneously, there is no sense in which ‘the c-structure satisfies its constraints, then passes the output to the f-structure’ or some such. The lines linking the f-structure elements with the c-structure elements in (6) are for illustrative purposes only.

1. (6)

Dalrymple (Reference Dalrymple1993) suggested a typology of LDRs based on Norwegian and Marathi, using the architecture of Lexical-Functional Grammar. Her analysis is built around the main idea that all anaphoric elements can be defined with respect to (i) the grammatical function of the antecedent or the grammatical functions which cannot be an antecedent, (ii) a domain within which an antecedent must or must not be found relative to the reflexive, and (iii) the presence or absence of other grammatical functions along the path from the reflexive to the antecedent. In addition, although they apply to elements in f-structure, the constraints which are built of (i), (ii) and (iii) are ultimately semantic, since they concern the equivalent semantic projections of two syntactic elements. This is summarised in the generalised binding constraint in (7).

1. (7) Generalised binding constraint

   ((DomainPath↑) AntecedentFunction)_σ = ↑_σ

   [off-path constraint]

The equation in (7) states (reading from right to left) that the semantic representation (_σ) of an anaphor (↑) is the same as (=) the semantic representation (_σ) of an antecedent (which may be restricted to some grammatical function) that is found by following some path (which may be restricted to include or exclude certain grammatical functions such as a SUBJ(ect), a PRED(icate) or TENSE), and which may be restricted to include or exclude certain other grammatical functions within the f-structures described by the DomainPath (i.e. the ‘off-path constraint’).

As mentioned just above, this regular expression applies to the f-structure, not the c-structure. In LFG, binding is not considered a structural phenomenon, rather a functional one. Furthermore, the generalised binding constraint in LFG uses ‘inside-out functional uncertainty’. This means that it describes a path from the reflexive to the antecedent (i.e., from the most embedded to the least embedded element) using functional uncertainty, which uses regular expressions rather than fixed paths in describing long-distance dependencies.

For example, the binding constraint for Norwegian sin third person reflexive possessive, as illustrated in (8) below, states that the anaphor must be bound by a subject within the minimal finite domain (Dalrymple Reference Dalrymple1993:35), i.e., it may be bound over a nonfinite clause boundary but not over a finite clause boundary. This domain is indicated by the use of an ‘off-path’ constraint, which in this case disallows the path from going through (although it may go to) an f-structure containing TENSE (Dalrymple Reference Dalrymple2001:284). TENSE does not lie directly on the path from the reflexive to the anaphor, hence it is ‘off-path’. The notation GF stands for ‘(any) Grammatical Function (found in f-structure)’, with the Kleene star indicating that ‘zero or more’ GFs may be used.

1. (8) ((GF* ↑) subj)_σ = ↑_σ

   −(→tense)

We can apply this constraint to the sentence in (4a) above, repeated here as (9). The f-structure of this sentence as given in (10).

1. (9)

2. (10)

The PRED value of the main f-structure (i.e. the main clause) is be ‘ask’, which takes a subject, object and non-finite complement clause. (I will not go into the issue of control here, and just assume that the subject of the nonfinite complement clause is functionally controlled. Nothing hinges on this.) The tense of this f-structure is ‘past’. All of the subcategorized-for grammatical functions of the main predicate are included in this f-structure – the subject's PRED value is ‘Jon’, the object's is ‘Maria’, and the XCOMP nonfinite complement clause has its own predicate – the verb vaska ‘wash’ which takes a subject and an object of its own. The subject of this f-structure is controlled by the object of the outermost f-structure, namely ‘Maria’, while the object in this f-structure is ‘self's car’. Sin must be bound to a subject that cannot be outside of an f-structure containing a valued TENSE, thus it can be bound to Jon here, but it could not be bound to an antecedent beyond this point (for example in ‘Hans said that Jon asked Mary to wash self's car’, sin could not be bound to Hans, since it would require the path to go through the f-structure containing a valued TENSE). Notice that a link to Jon is indicated by the thick dashed line, although since Maria also fits the requirement of being a subject within the minimal finite domain, Maria is also a potential antecedent for sin here. (As mentioned earlier, I will not be talking about locally-bound reflexives in this paper.)

Dalrymple identifies four domains which the Domain Path may refer to, in which an anaphor must be bound or free. These are the coargument domain, identified by the presence of a PREDicate, the minimal complete nucleus, which contains a SUBJect, a non-finite domain, identified as being within the f-structure containing a valued TENSE function, and the Root Sentence, for Icelandic-style (Strahan Reference Strahan, Svenonius, Bentzen, Jacobsen, Johannessen, Heycock, Parrott, Strahan and Vangsnes2009a) LDR which is the f-structure of the entire sentence. Notice that these are exactly the domains identified above which obey the Subset Principle, with the addition of the minimal complete nucleus.

Dalrymple (Reference Dalrymple1993:7) states that ‘[c]ombinations of requirements statable in terms of these concepts [i.e. PRED, SUBJ and TENSE] form the complete set of universally-available constraints on the distribution of anaphoric elements’. In this paper I will show that reference to other f-structure elements, including COMP (complement clause), ADJ (adjunct clauses) and REL (relative clauses), as well as PERSON, MOOD and ANIMACY values, is necessary in a complete description of nonlocal reflexives in the Scandinavian languages. Of particular interest is the fact that a neat hierarchy of binding domains fulfilling the Subset Principle cannot be constructed, even ignoring pronouns (like hann ‘he’), complex reflexives (like sjálfan sig ‘oneself, herself, himself, themselves’) and complex pronouns (like han sjølv ‘him self’). Nonetheless, most of the linguistic correlates of LDRs in Scandinavian are syntactic in nature and are f-structural elements. This means that Dalrymple's basic analysis can be maintained (although there is some evidence that the binding constraints for LDR should use outside-in rather than, or in addition to, inside-out functional uncertainty, Strahan Reference Strahan, Butt and King2009b), but is extended to allow other f-structural elements to be included in the binding constraints. Thus, I suggest that here a syntactic typology of long-distance reflexives that is highly flexible, reflecting the wide variation found in the Scandinavian languages.

In the next section I give examples which seem to suggest that the Subset Principle can be maintained, and even extended, in an account of LDR, but then follow this with more data showing that this position is actually untenable. I then show that off-path constraints that refer to more than just SUBJ, TENSE and PRED are also required, and finally that an outside-in functional uncertainty binding constraint is also needed, not just inside-out equations.

4. LDR DOMAIN CONSTRAINTS IN SCANDINAVIAN

I begin this section with some data that supports and seems to suggest a natural extension of the Subset Principle. The hierarchy we have built up so far is given in (5) above, repeated in (11).

1. (11) Local/coargument < non-finite < finite

In fact, we can augment this hierarchy with data from Icelandic. Icelandic has grammatical mood – subjunctive and indicative. The standard variety of Icelandic as reported in the literature is that LDR is acceptable with the subjunctive mood but not with the indicative:

1. (12)

However, as reported by Sigurðsson (Reference Sigurðsson1986), some speakers do accept LDR with the indicative mood. According to Sigurðsson, these speakers do not accept LDR if the verb is negated, such that the subject is no longer aware of the contents of the proposition.

1. (13)

Moshagen & Trosterud (Reference Moshagen and Trosterud1990) showed that this is also the case for the Smøla dialect of Norwegian. (Notice that this means that some Norwegian dialects have ‘long-distance’ reflexives, as well as ‘middle-distance’ ones.)

1. (14)

We can thus augment the hierarchy of binding domains to include this information, where POV represents ‘point-of-view’, where the final domain in this hierarchy (‘finite indicative non-point-of-view’) does not allow LDR.

1. (15) local < non-finite < finite subj. < finite indic. POV < finite indic. non-POV

The inclusion of perspective/point-of-view in this hierarchy can be implemented theoretically in at least two ways. Either some syntax-based POV-status function could be included in the f-structure or reference to the POV-holder in the finite domain could be implemented via a constraint on the discourse-structure combining with the syntactic finiteness constraint. It seems to me that the choice between these two options depends on whether LDR is viewed as being necessarily purely syntactic (option one), or whether the distinction between LDR and logophoricity is seen as being fuzzy, with the POV constraint being a composite of syntactic and discoursal features (option two). This complex matter was explored in Strahan (Reference Strahan, Butt and King2009b),where it was concluded that LDR in Icelandic and Faroese is best viewed as a composite syntactic and discoursal phenomenon. I will assume here that perspective is a discoursal constraint, and further assume that the accounts given in Sells (Reference Sells1987) and Oshima (Reference Oshima2007) apply here. (I will not explore this topic further here but see comments in Section 6 below.)

Further data from Icelandic shows that LDR is only possible out of complement clauses:

1. (16)

This implies the hierarchy in (17). Given that both finite indicate non-POV clauses and adjunct clauses disallow LDR, they are included together at the end of the hierarchy, rather than ordered.

1. (17)

This hierarchy defines syntactic domains of increasing size, where each subsequent domain includes all of the preceding ones. (While I accept that this is pushing things a little, the subjunctive mood can be thought of as being contained within the indicative, in the sense that sentences in Icelandic normally have the main verb in the indicative mood, but that embedded clauses may then be in the subjunctive; see e.g. example (12) above. On the other hand, since the purpose of this paper is to show that the idea of this kind of subset domain for LDR is inherently flawed, I am not too concerned with arguing for this reasoning.)

For the languages considered so far, and if we believe the literature – for many other languages too, this is accurate. However, let us look in more detail at the Scandinavian languages. Does LDR in all of these languages fit this hierarchy of binding domains? As we shall see, the answer is ‘no’.

Let us begin with the last piece of data, which shows that LDRs cannot be bound out of an adjunct clause. Actually, they can. Binding past an inanimate relativised noun is acceptable in all Norwegian dialects, not just in some, according to Lødrup (Reference Lødrup2006). The relativised nouns are underlined in the following sentences:

1. (18)

This data is particularly problematic if we want to keep the hierarchy that we have been developing. Recall that, excepting some dialects, Norwegian has middle-distance (i.e. non-finite) LDR, but not long-distance (finite) LDR. We could therefore slot relative clauses between non-finite and finite complement clauses in our hierarchy, which is now becoming a hierarchy specifically about binding domains of LDR, and is no longer a subset hierarchy.

1. (19) local < non-fin. < rel cl. < fin. Subjnct. compl. < fin.indic.compl POV < fin.non-POV, ADJ

This would then predict that Icelandic and Faroese, which both have finite LDR, should also have relative clause LDR. However, what we find is that such sentences, illustrated in (20), are completely out in Icelandic, although Faroese accepts the construction, as is shown in (21) and (22). The percentages in the annotations in (21) and (22) indicate the positive responses (‘yes’, rather than ‘unsure’ or ‘no’) to the test sentences in the large-scale syntactic overview projects in these language, where over 600 speakers gave judgements on each of these sentences (Thráinsson et al. Reference Thráinsson, Rögnvaldsson, Sigurjónsdóttir, Jónsson, Svavarsdóttir and Blöndal2005–2007; Thráinsson et al. Reference Thráinsson, Jónsson and Eyþórsson2008–2009).

1. (20)

2. (21)

3. (22)

The Subset Principle does not account for LDR out of relative clauses at all, and even the hierarchy of clause types that we were developing here runs into problems when all of the Scandinavian languages are taken into account. Norwegian and Icelandic require different orderings of the clause-types, while further evidence from Faroese which we will see in the next section suggests that it is possible for clause type to not even matter.

With the suggested hierarchy appearing increasingly untenable, let us look at Vestjysk Danish (VJD). Recall that in Standard Danish, nonfinite LDR is acceptable, but finite LDR is not (Vikner Reference Vikner1985). However, fieldwork conducted in 2008 suggests that, in Vestjysk Danish, nonfinite LDR si is completely unacceptable and finite LDR is completely acceptable! All of the seven speakers consulted on sentence (23a) rejected it completely and did not suggest a version of the sentence which kept a nonfinite complement clause. On the other hand, all three of the speakers consulted on sentence (23b) completely accepted it. (There is some debate about whether si in VJD is a reflexive or a pronoun since it can apparently take an antecedent which is not within the same sentence (Hagedorn & Jørgensen Reference Hagedorn and Jørgensen2008; Strahan Reference Strahan2011a), but similar arguments could be applied to Faroese seg, see Section 6 below. One reviewer commented that if we treat si here as a pronoun, then the examples in (23) are straightforwardly accounted for. However, given that a pronoun such as hans ‘his’ is acceptable in both (23a) and (23b), treating si as a pronoun would still require a solution to our original problem.)

1. (23)

As just mentioned, Standard Danish has the reverse judgements:

1. (24)

Based on the evidence presented here, it appears not that there is a hierarchy of binding domains, but merely that different LDR domains are possible. Each of the binding domains identified here can be referred to by an element in the f-structure, meaning that the essence of Dalrymple's (Reference Dalrymple1993) analysis can be maintained, but with the addition of further grammatical functions permitted along the Domain Path, such as REL, ADJ (on-path constraints) and certain values of MOOD (off-path constraint).

5. NON-DOMAIN CONSTRAINTS

In addition to the question of the binding domain, there are other constraints on types of LDRs that are orthogonal to the domain hierarchy.

Recall from above that in Norwegian, a reflexive may be bound across an inanimate relativised noun, but not an animate one. This applies to all three of the reflexives, as shown in (25).

1. (25)

In VJD, only si, and not sæ or sæ sjøl may be bound out of a relative clause. In addition, the relativised noun need not be inanimate, but it must be non-specific, or general. The seven speakers who judged this sentence all accepted it easily:

1. (26)

This actually also appears to be the case for Norwegian, where Norwegian speakers prefer sentences like (26) over sentences like (25). Thus, it is not just the animacy of the relativised noun that is important, but also the ‘generalness’ of it, where ‘general’ nouns include ‘no-one’ and ‘anyone’. (Sells (Reference Sells1987) reports that a logophoric pronoun may be bound across an animate relativised noun in Tuburi, but not in Mundang, Gokana nor Ewe, so this is not an entirely novel proposal.)

A common ‘feature’ of LDRs described in the literature is that they must have a grammatical subject as the antecedent, as we see for Icelandic and Norwegian in (27). Since sig/seg and sinn/sitt are third person reflexives, they cannot take the matrix subject as their antecedent in (27) or (28). In Icelandic and Norwegian, this results in an ungrammatical sentence (although some speakers in the Fosen region of Norway judged this sentence acceptable with the indirect object Maria as antecedent, given an appropriate context, Strahan Reference Strahan2011b).

1. (27)

However, in Faroese, this is not necessarily the case, with many speakers accepting an indirect object as antecedent for a nonlocal reflexive. Strahan (Reference Strahan2010a) argues that this is because all reflexive binding in Faroese is discourse-binding, and thus, given an appropriate context, the grammatical function of the antecedent is irrelevant. Thirty per cent of the 40 speakers consulted accepted (i.e. nobody rejected or was ‘unsure about’) the object reflexive and 43% accepted the possessive reflexive in (28).

1. (28)

Faroese therefore appears to be like Marathi (Dalrymple Reference Dalrymple1993), in allowing any grammatical function (for some speakers) to act as the antecedent for an LDR.

Another constraint on LDR which is orthogonal to the hierarchy of clause types occurs in Faroese, in (29), where the presence of a non-third person pronoun renders LDR completely unacceptable for many speakers (but obviously not all, as the preceding example shows). In fact, there is other evidence that this is actually a ‘no second person’ constraint, rather than a ‘no non-third person’ constraint, but this matter has not yet been addressed in a systematic way and so I prefer to not commit myself on this issue at this time. Only 27% of speakers accepted (29b).

1. (29)

Another factor that is related to perspective-holding ability is the animacy of this antecedent. It has been noted for Icelandic, illustrated in (30), and I have found that it is also the case for Faroese, illustrated in (31), that the antecedent of an LDR must be animate (Thráinsson Reference Thráinsson2007). Note that the animacy restriction in these examples is on the LDR, it is not a selectional restriction on the subject of krefjast/krevja ‘demanded’, since (30b) is fine with a pronoun það ‘it’ referring to the ‘the problem’.

1. (30)

2. (31)

Thus, the ANIMACY of both a relativised noun and the antecedent are relevant to LDR binding, as well as the GENERALNESS (which is an invented but plausible attribute for nouns like ‘anyone’ and ‘no-one’) of a relativised noun. These attributes can be included in the binding constraints as complex on-path constraints.

6. NON-SYNTACTIC CONSTRAINT

In Faroese, context appears to be all-important in determining the antecedent of a reflexive (Strahan Reference Strahan2010a). Notice that the context in (32b) is enough to allow the reflexive to have an extrasentential antecedent, although the context in (32a) is not. Of the 24 speakers who judged these sentences, 11 judged (32b) completely acceptable with the intended reading, three judged it marginal, while 10 judged it completely unacceptable. The pronouns hann ‘he’ in (32b) are normally interpreted by speakers as also referring to Zakaris.

1. (32)

Finally, Icelandic is famous for allowing LDR if a proposition is embedded within a ‘perspectivising predicate’, even if it is not allowed without it.

1. (33)

These last two examples in (33) provide very strong evidence that a radically different approach to describing the distribution of reflexives is needed. This has already been suggested in Strahan (Reference Strahan, Butt and King2009b), namely that instead of rules defining the distribution of reflexives starting from the reflexive and defining the domain in which its antecedent must be found, the distribution of reflexives should be described from the outside-in. That is to say, at least one binding constraint in Icelandic (and also Faroese) should use outside-in functional uncertainty, rather than them all being inside-out functional uncertainty.

It seems to me, that, from the point-of-view of the reflexives in both (33a) and (33b), the outside world must look the same. In both cases they are within an adjunct clause which is in the subjunctive mood. There is nothing in the overt syntax, as we head outwards from the reflexive, which indicates that the reflexive must keep looking for an antecedent in (33b), while it cannot in (33a). In fact, the main indication in (33b) that LDR is possible is the presence of the verb segja ‘say’. The deeply embedded reflexive can take the subject of segja as its antecedent, but not any other NP.

1. (34)

I suggest that what is happening here is that the reflexive is binding to the perspective-holder of its domain. This is not a new suggestion. However, I interpret this as indicating that the perspective-holder, and thus the antecedent of the LDR, is calculated from the outside-in, not from the inside-out. A sketch of this approach is shown in (35), where P represents the ‘perspective-holder’.

1. (35)

A sentence begins with the speaker being the perspective-holder, or ‘centre of deixis’. Complements of certain predicates take the highest-ranked argument (defined in some way) of that predicate as their perspective-holder, thus the P of the complement clause changes to be hann. Adjuncts (in Icelandic) either change the perspective-holder to the speaker, or keep the current one. They cannot change the perspective-holder to be an argument of the higher predicate in the way that complements do. Thus the P of the nema adjunct clause cannot be the hún subject of the previous clause, and the reflexive sér can therefore not refer to this hún. I assume that the perspective-holder is calculated for reasons independent of binding purposes, e.g. for keeping track of what the sentence/clause is ‘about’, but that it is used for binding too. Thus, in at least Icelandic, Faroese and Fosen Norwegian, the perspective-holder is an important part of LDR.

This constraint could be formulated as in (36), which states that the reference of a reflexive (↑_σ) is the same as the reference of a perspective-holder (↑_p)_σ which holds the perspective for contiguous domains (P*). This equation is a kind of outside-in functional uncertainty.

1. (36) ↑_σ = (↑_p P*)_σ, where P = ‘perspective-holder of a domain’

This is the only non-syntactic constraint that applies to LDR.

7. SUMMARY, CONCLUSION AND FINAL REMARKS

In summary, even within the class of ‘non-local reflexives’, and within a close-knit language family such as the Scandinavian languages, the typology of reflexives does not merely follow a simple hierarchy of binding domains. There are several problems with such as approach. Firstly, different languages allow LDR in different sets of clause types. For example, Icelandic allows LDR out of finite and non-finite complement clauses but not out of relative clauses, while most Norwegian dialects do not allow LDR out of a finite complement clause but do out of (finite) relative clauses and out of non-finite complement clauses. Vestjysk Danish appears to allow LDR out of at least some finite complement clauses but not out of non-finite clauses, which is the reverse of Standard Danish. A simple implicational hierarchy cannot be used to describe this data.

Secondly, non-domain elements such as the presence of a second person pronoun can affect the acceptability of LDR. At least some speakers of Faroese allow LDR out of all clause types, including both finite complement clauses, relative clauses and other adjunct clauses such as ‘but’- and ‘if’-clauses, but the presence of a second person pronoun in the embedded clauses renders the LDR unacceptable. In Norwegian, the animacy and ‘generalness’ of a relativised noun plays a crucial role in whether or not LDR is acceptable. The grammatical function of the antecedent of an LDR need not always be subject in Faroese, contrary to the received facts of LDR, while the antecedent of an LDR must be animate in both Icelandic and Faroese.

Finally, I suggested that LDRs are always bound to the perspective-holder of the domain containing them, based in particular on Icelandic data showing that reflexives may in fact be bound out of non-complement clauses, if they are embedded within a perspectivising (or ‘logophoric’) predicate. Importantly, the reflexive may only refer to the subject of the perspectivising predicate, being the perspective-holder of the domain containing the reflexive. This data does not fit easily into a typology based on the Binding Conditions standardly accepted in Chomskyan syntax.

Based on this evidence, in this paper, I extended Dalrymple's (Reference Dalrymple1993) typology of anaphora that is set in the Lexical-Functional Grammar framework. Her typology defines anaphors in terms of three areas: the grammatical function of the antecedent, the grammatical functions along the path from the anaphor to the antecedent, and features contained in the f-structures along the path from the anaphor to the antecedent. The grammatical function of the antecedent can be either SUBJ or any GF, although I showed here that the ANIMACY of the antecedent is also relevant in at least Icelandic (Thráinsson Reference Thráinsson2007) and Faroese (Strahan Reference Strahan, Svenonius, Bentzen, Jacobsen, Johannessen, Heycock, Parrott, Strahan and Vangsnes2009a). Dalrymple used a hierarchy of clause types for the domain path, using the three grammatical functions of SUBJ, TENSE and PRED. Here I have shown that the domain path needs to also include ADJ and REL clauses as on-path constraints, as well as values of ANIMACY and possibly semantic generality for the relativised noun as off-paths constraints along this path. The off-path constraints also need to be able to refer to certain values of MOOD and PERSON, potentially embedded within another GF. Finally, there is evidence in Faroese, Icelandic and some varieties of Norwegian (i.e. Fosenmål) that LDRs refer to the perspective-holder of the domain containing the reflexive, and that this perspective-holder is necessarily calculated from the left, and thus uses outside-in functional uncertainty, which is the opposite of the other binding constraints.

While Darymple (Reference Dalrymple1993) restricted the binding constraints she suggested to referring to only three elements in the f-structure, it seems clear that this is not enough. However, the architecture of Lexical-Functional Grammar allows for the binding constraints to refer to any information in its syntactic functional-structures (f-structures), and it has the necessary tools for us to write constraints for each reflexive which will reflect its unique distributional pattern. Dalrymple (Reference Dalrymple1993) showed how this can be implemented for reflexives in Norwegian and Marathi, and by adding data from the other Scandinavian languages we have expanded the typology of LDR she suggested.

Finally, anaphors may be constrained by (37) – positive constraints (e.g. the antecedent must be a subject) and negative constraints (e.g. there cannot be a non-third person value nearby); and (38) – on-path constraints (e.g. the Domain Path must pass through a complement clause) and off-path constraints (e.g. the clause must contain a non-finite tense); and we can refer to (39) – both inside-out functional uncertainty (that is, the path from the reflexive to the antecedent can be constrained) and outside-in functional uncertainty (that is, the path from the antecedent to the reflexive defines the perspective-holder for each f-structure).

1. (37)

   1. a. positive constraints (antecedent must be a SUBJ)

      ((GF* ↑) SUBJ)σ = ↑_σ

   2. b. negative constraints (cannot be bound to a coargument; cannot have a second person pronoun within a GF that is itself contained in a GF on the path from anaphor to antecedent)

      ((GF ↑) GF)σ ≠ ↑σ; ((GF* ↑) SUBJ)σ = ↑σ

      → (GF PERSON) ≠ 2

2. (38)

   1. a. on-path constraints (must pass through COMP clauses only)

      ((COMP* ↑) GF)_σ = ↑_σ

   2. b. off-path constraints (the clause must contain tense)

      ((GF* ↑) SUBJ)_σ = ↑_σ

      → TENSE

3. (39)

   1. a. inside-out functional uncertainty (the path from the reflexive to the antecedent can be constrained)

      ((GF* ↑) SUBJ)_σ = ↑_σ

   2. b. outside-in functional uncertainty (the path from the antecedent to the reflexive must pass through perspective-holders in each (f-structure))

      ↑_σ = (↑_p P*)_σ, where P = ‘perspective-holder’

In conclusion, a typology of long-distance reflexives is describable in LFG terms by constraints upon attribute–value arrays found in f-structures, including positive and negative constraints, on- and off-path constraints, and both inside-out and outside-in functional uncertainty.