“Language Change”
GERMANIC REFLEXIVES AND THE IMPLEMENTATION
OF BINDING CONDITIONS
1.0.In this paper, I will be concerned with the problem of explaining certain changes in the distribution of reflexive and nonreflexive pronouns in the Germanic languages. I will show that there are systematic tendencies in evidence in those changes which suggest that aspects of the distribution reconstructable for the parent language were inherently unstable, and I will argue this instability in turn can be considered to follow from a somewhat modified version of the Theory of Binding developed in Chomsky (1979). That theory characterizes as marked precisely those aspects of Germanic pronominal syntax which prove to be liable to change, since they all involve violations of a single condition on binding in universal grammar. The changes in question may therefore be regarded as individual manifestations of the implementation of that condition.
It will be seen, however, that in many of the languages considered, the condition does not appear to have been suddenly implemented in its entirety. Its effects, rather than appearing abruptly across the board in all environments where they are expected, appear first in some subset of those environments, and only later in others. Thus, the introduction of the condition seems in many cases to have taken place gradually by way of the introduction of a series of less general conditions which it entails.
2.0. We may begin by considering the distribution of the two forms in Gothic—the earliest extensively documented of the Germanic languages. The bulk of the extant corpus of Gothic consists of a translation of portions of the Bible from New Testament Greek—a translation which is on the whole rather literal, but which exhibits numerous instances of systematic failure to imitate the Greek model, motivated apparently by the desire to avoid producing ungrammatical constructions.1 Evidence from this sort of systematic deviation allows us to arrive at a fairly clear picture of conditions on anaphor binding in Gothic.
New Testament Greek, like Gothic, distinguished between reflexives (e.g. έαυτον ‘himself’) and nonreflexive pronouns (e.g. αὐτον ‘him’), but the distribution of these forms in the Greek text differs systematically from that of the corresponding Gothic forms (sik ‘self’, ina ‘him’) in a number of environments. In both languages, reflexives were used in simple clauses to corefer with the subjects of those clauses.2 Nonreflexive pronouns were understood as disjoint in reference with tautoclausal subjects. These facts are illustrated in (1). [+R] designates a reflexive form, and [-R] designates a nonreflexive form.
As demonstrated by the glosses on these examples, this distribution is shared by English as well.3
Example (2a) shows that in both Gothic and Greek (as well as English), nonreflexive pronouns which are accusative subjects of infinitives were also understood as disjoint in reference with the subject of the higher clause. Reflexives were used in that position to corefer with the higher subject, as illustrated by (2b).4
The distribution of reflexive and nonreflexive forms in Gothic, however, differs from that found in Greek (and English) in the following ways. First, in Gothic, but not in Greek, reflexive pronouns in nonsubject positions in infinitive complements could be bound to the subject of higher clause.5 This is illustrated by the examples in (3). As demonstrated by the glosses, English patterns with Greek in this respect.
The same difference between languages is found in participial relative clauses, as demonstrated by the examples in (4).
Second, the two languages differed with respect to their treatment of possessive modifiers on NPs. In New Testament Greek, the genitive form of the nonreflexive pronoun αὐτον was normally used to indicate both possession by the subject and possession by some NP other than the subject. The genitive of the reflexive έαυτον appears to have been used as a possessive only emphatically. In Gothic, on the other hand, nonreflexive possessives were used just where possession by some NP other than the subject was intended. In order to indicate possession by the subject, the reflexive form sein- was used. The usual Gothic/Greek correspondences are illustrated in (5).7
Example (5e) shows that subjects of NPs in Gothic could also serve as antecedents for reflexive possessives.
As illustrated by the English glosses on these examples, English once again patterns with Greek. That is, in English, as in Greek, distinctly reflexive possessives are not used systematically in cases of possession by the subject. (The reflexive possessive his own is used only emphatically.) Correspondingly, as in Greek, nonreflexive pronouns occurring as possessives are not understood as necessarily disjoint in reference with the subject. Gothic differs from Greek and English in both respects.
Finally, Gothic differs from Greek in that, with certain prepositions, the nonreflexive pronoun is used in Greek in prepositional object position to refer back to the subject, while the reflexive is used in the corresponding Gothic constructions. This is demonstrated by the examples in (6):
Once again, as illustrated by the English glosses, English patterns like Greek.8
2.1.We can summarize the similarities and differences discussed above by means of the following chart. [+] in a column means that, in the position in question, reflexives are used routinely to refer to the subject of the matrix clause and nonreflexive pronouns are interpreted as being disjoint in reference with the subject of the matrix clause. [-] means that, in the position in question, (nonemphatic) reflexives are not used to refer back to the matrix subject, and nonreflexive pronouns are not understood as necessarily disjoint in reference with the matrix subject.
Thus, there are at least four different positions which are normally inaccessible to binding in Greek and English, but accessible to it in Gothic.9 Of course, we would not wish to propose that Gothic differs from Greek and English with respect to four independent conditions, each specifying a separate construction. The simplest hypothesis would hold rather that the four cases in which Gothic differs from the other languages are actually just individual consequences of its failure to observe a single, general condition which is observed in those languages. In fact, most of the differences which we have noted between Gothic and Greek/English would follow directly from the assumption that Gothic, but not the other languages failed to observe the condition on anaphors and pronouns attributed by Chomsky (1979) to Universal Grammar. This condition is given in (8).
(8) a. An anaphor [e.g. a reflexive] cannot be free in its minimal governing category. [That is, it must be coindexed with a c-commanding element within its m.g.c.]
b. A pronoun must be free in its minimal governing category.
The minimal governing category of an NP can be taken, for the present purposes, to be the minimal NP or S within which it is assigned case.10 An object NP, for example, is assigned its case by virtue of being governed by the verb which is the head of VP. Its minimal governing category, therefore, is the S dominating VP. The subject of a tensed clause received its nominative case by virtue of being c-commanded by tense. Its minimal governing category is the S dominating it and tense. The minimal governing category for the accusative subject of an infinitive, on the other hand, is the higher S, since it receives its case from the verb of the higher S.11 The minimal governing category for a possessor NP is the NP dominating it, since it receives its genitive case by virtue of its membership in that NP.12 Now, if we assume that condition (8) is observed in Greek and English, but not in Gothic, then the differences reflected in (7c), (7d), and (7e) can be accounted for rather simply. In each case, Gothic allows a reflexive to be bound outside of its minimal governing category, while the other two languages do not; and, in each case, Greek and English allow nonreflexive pronouns to be interpreted as preferential with subject NPs outside their minimal governing categories, while Gothic apparently does not.13 we may explain this by claiming that Gothic, Greek, and English all share the unbracketed portion of condition (8f), but that only the grammars of English and Greek incorporate the bracketed portion.14
(8’) a. An anaphor cannot be free [in its minimal governing category].
b. A pronoun must be free [in its minimal governing category].
Still unexplained by this proposal would be the difference reflected in (7f), involving the position of prepositional object. It has been seen that this position, at least with locative prepositions, is opaque for the purpose of disjoint reference assignment or in Greek and English, but apparently transparent to DR in Gothic. This difference could be accounted for under the proposal being considered if we added PP to the list of phrase types which may constitute governing categories. (However, cf. note 8.)
Now a grammar of Gothic containing the truncated versions of (8a) and (8b) as its only conditions on binding and disjoint reference would, in fact, be insufficiently constrained. In Gothic, as in Greek and English, nonreflexive pronouns in nonsubject positions in TENSED embedded clauses are not interpreted as necessarily disjoint in reference with the subject of the higher clause. This is illustrated by (9).
It appears moreover that, as in Greek and English, it was not possible for a reflexive in such a position to be bound to the higher clause subject. There are, at least, no instances in the corpus where this occurs.15
In English and Greek, the failure of the lower clause pronoun and the higher clause subject to be assigned mandatory disjoint reference in such constructions would follow from the fact that, by subcondition (8b), a pronoun is required to be free only within its minimal governing category. However, this cannot be the reason for the failure of DR in the Gothic examples in (9) if, as we have suggested, the Gothic version of (8b) did not refer to minimal governing categories. In English and Greek, the opacity of tensed clauses needs no special mention, since it follows from the fact that any NP or S meeting the definition of a minimal governing category is opaque, but in Gothic there is an asymmetry between tensed clauses on the one hand and untensed clauses and NPs on the other. The former, but not the latter are opaque. It appears therefore that the minimal set of binding conditions posited for Gothic will have to contain, in addition to the truncated versions of (8a) and (8b), a principle which, like the Propositional Island Condition proposed by Chomsky in earlier work, singles out tensed clauses in particular as opaque domains.16
For the sake of orderly exposition, I will assume in the following that the syntax of reflexive and nonreflex- ive forms in the Germanic parent language was essentially like that found in Gothic. At the end of the paper, I will return to show that this assumption is justified.
2.2.If, as Chomsky proposes, condition (8) has the status of a principle of Universal Grammar, then two predictions fall out concerning the patterns of distribution of reflexive and nonreflexive forms which we might expect to find crosslinguistically. First, languages like English and Greek, in which NPs, participial relative constructions, and infinitive complements are opaque, should be commonplace, while languages like Gothic, in which they are transparent, should be rare. Second, if a language observes (8), then we should expect to find its effects in all relevant constructions, and if a language does not observe it, we should not expect to find its effects anywhere. Thus, for example, if a language exhibits a systematic distinction between reflexive and nonreflexive possessives, we should expect to find that it also allows reflexive binding and DR to operate into nonsubject positions in infinitive complements, and conversely, if NPs are opaque to reflexive binding and DR in some language, then infinitive complements should be opaque in that language as well. (Of course, if this turned out generally not to be the case, then the claim that the opacity of NPs and the opacity of infinitive complements are to be accounted for by a single condition would be called into question.)
I have insufficient data to assess the validity of the first prediction. Languages like Gothic do not appear to be particularly rare within the Indo-European family, where they include at least Icelandic, Swedish, Danish, Latin, Russian, and Hindi,17 but I have no idea about their relative frequency outside of Indo-European. The second prediction appears to be borne out by most of the available evidence. In Greek and English, where infinitive complements are opaque to reflexive binding, the original systematic reflexive/nonreflexive possessive distinction has been lost. On the other hand, languages in which infinitive complements are transparent generally distinguish systematically between reflexive and nonre- flexive possessives. Compare the following examples from Russian and Latin.18
As will be seen below, however, this correlation is not always found.
2.3.There are also some indirect diachronic predictions inherent in the claim that (8) is a universal principle. In particular, it follows from that claim that binding in languages like Gothic, insofar as it does not observe (8), is marked, and if we make the not unreasonable assumption that marked grammars are unstable and therefore liable to change over time, we may expect to find that such languages will tend diachronically to come into conformity with the pattern exhibited by languages like English and Greek.19 This is the diachronic counterpart of the first synchronic prediction. The second also has a diachronic counterpart. We have hypothesized that all of the observed differences between languages like Gothic and those like Greek and English with respect to binding and disjoint reference result from the fact that the former fail to observe a single unmarked principle. If that principle is subsequently incorporated into the grammar, we should expect to see its effects in all relevant environments roughly simultcineously.
In the following sections, I will present evidence showing that at least the first of these predictions appears to be borne out in several Germanic languages. For each of them, I will show that at an early stage the distribution of reflexive and nonreflexive pronouns resembled that found in Biblical Gothic, but that later stages reflect the predicted development in the direction of a distribution more like that found in English. I will also consider the relative chronology of changes in individual environments, to see to what extent these conform with the second prediction.
3.0.First, let us consider some evidence for change in the pronominal syntax of Gothic. For Gothic, unlike most of the Germanic languages, we do not have documents spanning a long period of development. Thus, evidence for syntactic change in Gothic is sparse—but, I propose, not altogether lacking. First of all, in the Gothic Bible there are sporadic instances in which, contrary to the pattern in evidence in the great majority of like constructions, a nonreflexive pronoun—rather than a reflexive—appearing in nonsubject position in an untensed clause corefers with the subject of the matrix clause. In (12) I have listed the five examples of this type appearing in the corpus.
The prevailing pattern found in all constructions of this type, except for these five instances, would lead us to expect a reflexive form in the underlined positions [compare examples (3) and (4)].
Similarly, in the three examples in (13), the translator used nonreflexive possessive pronouns where (under the usual interpretation of the passages in question) co-reference with the subject is intended. These examples are once again rare exceptions to the overwhelmingly attested use of reflexive possessives where reference to the subject is intended, and of nonreflexive possessives where reference to some NP other than the subject is intended.
The failure of the translator to meet our expectations in (12) and (13) admits of a few possible explanations. For example, we may be dealing here with simple lapses, in which, under the influence of the nonreflexive form in the Greek model, he carelessly used a nonreflexive form where he should have used a reflexive. (Note that in a number of these examples a great deal of material intervenes between the higher subject and the underlined form.) Or it may be that an exegetical error is involved. The Greek text in all such instances, of course, would have used a nonreflexive pronoun whether coreference with the higher subject was intended or not, and the model constructions would be ambiguous except for context. Finally, however, a more principled explanation is available for the “errors” in question. In all of these instances, a nonreflexive form appears where the grammar seems to dictate that a reflexive should appear (given that the translator shared our understanding of the passages), but where the proposed universal principle (8) predicts that reflexives should be excluded. Thus, we might suggest that examples (12) and (13) in fact reflect incipient syntactic change in the direction of a less marked binding situation. This explanation, in addition to not invoking accident, finds support in the developments to be discussed below.
3.1.Next, let us turn to the Gothic of the Commentary on the Gospel of John, or Skeiveins,—the only major Gothic document besides the Gospel translation. Since the Skeiveins quotes from the Bible translation extensively, it is to be regarded as a later composition.20
The commentary is a relatively short text, and therefore provides comparatively little basis for generalization about the language it represents. Significantly, however, there are two participial relative constructions in the corpus which contain pronominal forms coreferring with the matrix subject, and in both, those forms are nonreflexive pronouns—as in the exceptional Biblical Gothic examples in (12)—rather than reflexives—as in the normal Biblical Gothic examples in (4). The two examples in question are given in (14).
These examples indicate that participial relative constructions have become opaque to disjoint reference by this stage in Gothic—a development whose beginnings, as we have noted, may be reflected in the Biblical Gothic examples in (12), and one which is consistent with the predictions of our hypothesis concerning likely changes in Gothic binding conditions. On the basis of these examples, it appears as if we could claim that the Gothic of the Commentary has departed from Biblical Gothic by instituting the full version of condition (8). Unfortunately, this claim is not borne out by the remainder of the evidence provided by the Skeireins. First of all, as illustrated by (15), Skeireins Gothic continued to maintain the systematic reflexive/nonreflexive possessive distinction.
Moreover, nonsubject positions in untensed clauses other than those involved in participial relative constructions appear to have remained bindable to clause-external subjects, as illustrated by the absolute participial construction in (16) :
In this example, the two reflexives in the participial phrase refer back to the (reconstructed) matrix subject Iohannes.21
Thus, while the scant data from the Skeireins in part support our hypothesis, by showing that a change in distribution apparently did take place in one construction in which it was expected to take place, they also pose some problems for that hypothesis, since parallel changes appear not be have occurred in other environments where the hypothesis predicts that they should have. If Skeireins Gothic did in fact differ from Biblical Gothic in that it had implemented the bracketed portion of the conditions in (8), then we should expect that changes would have taken place in all relevant environments, and that reflexives should not be possible in (15b) and (16). Rather than rejecting the hypothesis in favor of a weaker one on the basis of the few Skeireins examples, however, I will continue to regard it as a potentially useful one, pending examination of data from the other dialects. It will be seen in the following discussion that similar problems arise in the history of German.
3.2.The syntax of reflexive and nonreflexive pronouns in German has, from the period of first attestation, resembled that found in Biblical Gothic in the following respects: first, in simple sentences, nonreflexive (accusative) pronouns are understood as disjoint in reference with the subject.22 in order to indicate coreference with the subject in such instances, it has at all stages been necessary to use a reflexive. This is illustrated by the following examples from the Old High German Tatian.23 (The Tatian is a translation from Latin. I have indicated in the examples taken from it the nature of the Latin pronouns which served as models for the underlined German forms.)
Second, as in Gothic, nonreflexive forms in tensed clauses, whether in subject or nonsubject position, are not interpreted as disjoint in reference with the subject of the higher clause. Nor do reflexives occur in tensed clauses with higher clause subjects as antecedents. These facts are illustrated by the examples in (18). (The same restrictions hold for Modern German as well.)
Third, in Old High German, as in Gothic, nonreflexive pronouns in the position of accusative subject of an untensed clause appear to have been interpreted as necéssarily disjoint in reference with the subject of the higher clause. When coreference was intended, a reflexive was used. These facts are illustrated in (19).
This is also true in Modern German for the small group of verbs (e.g. lassen ‘cause/permit’, sehen ‘see’) which can still take infinitive complements with lexical subjects.
However, the distribution of forms in Old High German (and Modern German) also differs in some ways from that found in Gothic. First of all, as the examples in (20) show, the reflexive/nonreflexive possessive distinction has been wholly levelled. Sîn (Modern German sein ), originally a strictly reflexive form (as evidenced by the fact that its cognates in all other dialects are reflexive), had lost its reflexive function by Old High German times. It could be used either to indicate coreference with the subject (e.g. in translation of the Latin suus ) or to indicate disjoint reference (e.g. in translation of the Latin eius).
We are justified in claiming, on the basis of the comparative evidence, that at some point in the prehistory of Old High German the distinction was observed? that sîn was exclusively reflexive, and that possession by nonsubjects was expressed by means of the genitive of the nonreflexive pronoun paradigm. The development between that stage and the stage represented by attested Old High German, then, involved the generalization of the reflexive form at the expense of the nonreflexive. Now, it is perhaps possible to view this development as the result of an accidental syncretism local to German—of the type apparently represented by the loss of the dative reflexive (cf. note 22). However, the hypothesis which we have developed allows for a more principled explanation of this change. As reflexive binding became restricted to minimal governing categories, we would expect that reflexive possessive forms should lose their obligatory anaphoric interpretation, therefore becoming subject to generalization into environments where disjoint reference was intended. An explanation based on such changes in binding conditions has the considerable advantage of allowing us to relate the loss of the reflexive/nonreflexive distinction in German to the remarkable instability of that seemingly quite useful distinction elsewhere in Germanic. We have already observed some variation with respect to it in Biblical Gothic, where nonreflexive possessives sometimes occurred in place of expected reflexive possessives. Evidence of a similar sort from other dialects will be introduced below. (Interestingly, the same development is found in Romance as well, where the forms descended from the reflexive possessive suus have become the general third person possessive forms.) It should also be noted that the feminine/plural possessive ihr ‘her, their,’ historically a nonreflexive, is now also used in German to refer back to subject antecedents.
German has also, from the earliest period of attestation, differed from Biblical Gothic in that reflexive pronouns are blocked from appearing in participial relative constructions when their antecedents are external to those constructions. Correspondingly, nonreflexive pronouns in such constructions admit an interpretation of coreference with the higher subject. This is illustrated by the Old High German example in (21a) and the Modern German example in (21b).
On the basis of this restriction and the disappearance of distinctly reflexive possessive forms, we could propose that German, like English, differs from Biblical Gothic in that at some point in its prehistory it incorporated the minimal governing category condition into its grammar. In fact, though, such a claim would lead to incorrect predictions about the treatment of nonsubject positions in untensed clauses other than those found in participial relative constructions. As illustrated by example (22a) (Old High German) and examples (22b-d) (Middle High German), reflexives in such positions could in fact be bound to the subject of the higher clause.
We will return to these constructions shortly.
The similarities and differences between Gothic and German discussed above are summarized in the following chart:26
Once again, our hypothesis is partially supported by these results. The distribution of reflexive forms in early German differs from Gothic and resembles Modern English in some predicted ways. In particular, NPs and participial relative clauses have become opaque for binding and DR. Again, however, the data are somewhat problematic for the hypothesis. If the failure of such processes to operate across the boundaries of these constructions were due to the implementation of the minimal governing category condition, then we should find that infinitive complements had also become opaque for them. The examples in (22) indicate that this is not the case.
There is evidence, though, that the grammar of German has in fact tended in historical times toward realization of the predicted state of affairs with respect to such constructions. Thus, for example, in Modern German, unlike Middle High German, a nonsubject reflexive in the complement of bitten ‘ask’ cannot have the subject of bitten as its antecedent. In (22e), the literal Modern German translation of the Middle High German example (22d), the reflexive pronoun is disallowed.
With certain verbs in Modern German (e.g. lassen ‘cause, permit,’ sehen ‘see’), a nonsubject reflexive in the complement can still be bound to the matrix subject. This is possible, however, only when the reflexive is complement- initial or the object of a preposition.27
A comparison of (24b) with (22b) demonstrates that this restriction did not exist in earlier stages of German. Thus, the accessibility of complements of verbs like lassen to reflexive binding is more limited in Modern German than it was at earlier stages of the language. Note also that the reflexive is not required in (24a). That is, matrix controlled DR does not have access to the underlined position.
Such changes are consistent with the predictions of our hypothesis, but we are still faced with the problem that they did not occur at the same time as the other changes considered. Should the hypothesis be abandoned in view of this lack of simultaneity? As an alternative, we might propose that the parent language was marked with respect to several less general conditions, rather than a single condition of greater generality. Thus, for example, the fact that nonsubject positions in Gothic infinitive complements were bindable by the matrix subject could be attributed to nonobservance of the Specified Subject Condition of Chomsky (1980), while the obligatory disjoint reference assigned to nonreflexive possessives in Gothic but not Greek or English could be explained by claiming that, in the unmarked case represented by the latter languages, assignment of disjoint reference is subject to the A-over-A condition. According to this approach Gothic and like languages will be marked with respect to both constructions and both may be expected to change over time, but not necessarily simultaneously since two separate conditions are involved.
The disadvantage of this solution is that it characterizes as accidental the strikingly high degree of cooccurrence between the two marked binding situations. There is, according to this view, no particular reason why Gothic should differ from Greek and English with respect to both. We might equally well expect to find languages in which there is a systematic distinction between the two types of possessives, but in which infinitive complements are opaque to reflexive binding and DR. Indeed, there should be more such languages than languages like Gothic, since the former are marked with respect to one less universal condition. I know of no languages of this type. Although pre-Old High German appears to have been a language in which the opposite situation is represented, it also appears atypical. Numerous other languages besides Gothic, including Icelandic, Danish, Swedish, Latin, Russian, and Hindi, differ from English and Greek with respect to both constructions.28 Examples from Latin and Russian were cited above. Examples from Icelandic, Danish, and Swedish will be introduced below.
What is needed therefore, is an analysis which will predict that if infinitive complements in a language are transparent for binding, NPs are also likely to be transparent and the converse, but which will be compatible with a diachronic situation in which one remains transparent while the other becomes opaque. Perhaps the resolution of this apparent paradox is to be found in the manner in which the conditions are implemented. In particular, we might assume that languages like Gothic are in fact marked with respect to a single condition, resulting in the transparency of a number of constructions which ought to be opaque; the language learner, however, rather than correcting this situation by wholly implementing the unmarked condition, might attempt instead to effect local repairs, singling out individual constructions and incorporating into his or her grammar only as much of the condition as pertains to those constructions—the particular constructions chosen perhaps having to do with their relative order of acquisition. Those constructions not singled out in this way will remain marked, and therefore subject to later change. Under this not implausible view of syntactic change, we can reconcile the generally parallel treatment of infinitive complements and NPs by binding processes with asymmetrical changes of the type observed in the history of German.
3.3.Let us turn next to the history of reflexive binding in English, which is closely related to German within the Germanic language family. The evidence for the reconstruction of this history is sparse, since English and immediately related languages lack the reflexive pronouns of the Germanic *sik paradigm. If those languages did in fact possess such forms at some point, their loss must have occurred in prehistoric times, at least prior to the exodus of the English tribes from the continent. Throughout the Old English period, historically nonreflexive pronouns were used in both reflexive and nonreflexive roles. The new reflexive pronouns of the self type became established as normal (nonemphatic) reflexive pronouns only in Middle English. Significantly, though, since the time of their systematic incorporation into the grammar, they seem to have been subject to the same (presumably unmarked) conditions on distribution which prevail in Modern English.
Also significant is the fate of the one Germanic reflexive form which did survive into attested English—the reflexive possessive adjective sīn. This form is cognate with Gothic sein and Old High German sîn, and unlike the latter it remained reflexive, occurring only where co- referential with the matrix subject, as in (25a).
Example (25b) shows, though, that even at this stage the maintenance of the form was unnecessary, since nonreflexives in the position of possessive modifier were not interpreted as necessarily disjoint in reference with the subject.
The reflexive possessive was lost altogether by the classical Old English period. Its disappearance, not to be connected with the apparent loss of the reflexive pronominal forms, which must have taken place at least several centuries earlier, is consistent with our hypothesis. Thus, the hypothesis finds support in at least one bit of direct evidence from the history of reflexives in English.
3.4.Next, let us turn to the North Germanic languages. In Old Icelandic, the earliest extensively documented of these, the situation with respect to reflexive binding and disjoint reference is essentially like that found in Gothic. Within simple sentences, coreference with the subject had to be expressed by means of the reflexive. Nonreflexive pronouns were interpreted as disjoint in reference with the subject. This is illustrated by (26a) and (26b):
As shown by (26c) and (26d), the same was true of accusative subjects of infinitives:
Subject and nonsubject positions in tensed embedded clauses were opaque to these processes—or at least to DR—as illustrated by (27).29
On the other hand, as in Gothic, nonsubject positions in TENSELESS complements were transparent (28a, b), as were the positions of subject of NP (28c) and object of locative PP (28d):30
However, there are, as in Gothic, sporadic instances of constructions of the first type in which nonreflexive pronouns are used, rather than reflexives, to corefer with the matrix subject. The following is such a case:
Again, it is possible to interpret this marginal variation as incipient syntactic change, in the direction predicted by our hypothesis.
The modern West Scandinavian languages, to which Old Icelandic is immediately ancestral, do not seem to differ from it substantially in terms of pronominal syntax. See Beito (1970:246) for examples from New Norwegian, Lockwood (1964:117 ff.) for examples from Faroese, and Thráinsson (1976b) for examples from Modern Icelandic. The value for the present study of the evidence provided by West Scandinavian lies in the fact that it supports the claim that the Gothic type of binding situation was the original one in Germanic. I will discuss this issue at greater length below.
3.4.1. Evidence from North Germanic supporting the main contention of this paper comes from languages of the eastern group of that branch: specifically, Danish and Dano-Norwegian. The references I have consulted on those languages present a relatively sketchy picture of developments in pronominal syntax, and a number of questions remain to be answered regarding dialect and relative chronology, but the distinction of reflexives and nonreflexive pronouns in both possessive modifier position and nonsubject position in untensed clauses has undergone changes consistent with the predictions of our hypothesis. First, let us consider possessive constructions. As illustrated by (30a) and (30b), Old Danish continued the Germanic distinction between reflexive and nonreflexive possessives.
However, it is possible at a very early date to find instances in which the nonreflexive pronoun is used as a possessive where coreference with the subject is intended:
According to Falk and Torp (1900:134) this became general in later stages of the language:
Falk and Torp give the impression that the replacement of the reflexive by the nonreflexive possessive was normal during the period in question, and it seems to have persisted for a considerable time. It is not, however, a feature of Modern Standard Danish, as illustrated by (32a).
In Modern Danish, the effects of the replacement survive only in the plural, where the originally nonreflexive deres is used to express possession by both subjects and nonsubjects.31 (This is not true of Standard Dano- Norwegian, where sin is still used in reflexive meaning even in the plural.) In the Jutish dialect, on the other hand, the distinction has been levelled in favor of the nonreflexive pronoun in both singular and plural:
Thus, the neutralization of the opposition, which, as we saw earlier, took place in English and German, has also occurred in some varieties of Danish, providing further support for our hypothesis. The apparent countertendency in Standard Danish reflected in the reinstatement of sin in the singular is of course problematic for the hypothesis, but it could possibly turn out to be an artifact of dialect conflict.
Corresponding to this development in possessives in Danish, we also find the expected change in nonsubject position in tenseless complements. Falk and Torp (1900: 132) note this change, citing the example in (33a) from the literature (mid-nineteenth century) and presenting sentences like (33b) and (33c) as representative of contemporary usage in Danish and Dano-Norwegian.
They also admit the reflexive in such constructions, though, as in (33d) and (33e), so it appears that DR only, not reflexive binding, was constrained from applying to the position in question.32
Spore (1965:162 ff.) still characterizes the use of the nonreflexive pronoun in constructions like (33a)-(33c) as normal in Danish where reference to the higher subject is intended. However, my informant, a native of Copenhagen, rejects the nonreflexive in similar sentences, which again suggests the existence of dialect differences:
Iversen (1918:33 ff.) reports for the Tromsø dialect of Dano-Norwegian facts which are similar to those presented by Falk and Torp. He states that around the turn of the century it was not uncommon for the nonreflexive pronoun to be used in nonsubject position in infinitive complements to corefer with the higher subject (although the reflexive was also possible):
Although the reflexive had been extensively reintroduced in such constructions by the time of Iversen’s study, at the expense of the nonreflexive, the latter was still generally used in at least some of them.
In summary, we have observed that in Danish nonreflexive pronouns coreferring with the matrix subject began appearing in two environments where DR assignment had previously prevented their occurrence—subject position within NPs and nonsubject position within infinitive complements. These changes took place during overlapping, though apparently not simultaneous periods, and they were carried through to different degrees in different dialects. In Standard Danish, as spoken by my informant, they are reflected only in the use of deres as the sole possessive form in the plural.
While the details of these developments merit further investigation, the facts outlined above are sufficient to show that the predictions of our hypothesis are borne out in North Germanic as they were in East Germanic (represented by Gothic) and West Germanic. In Danish, as in the languages considered earlier, two positions which should be opaque to binding processes according to (8) have in fact tended to become opaque. Moreover, the rough chronological correspondence between the two changes provides some support for the claim that a single general principle was responsible for both.
3.4.2. In Swedish, the other major language of the Eastern North Germanic group, the distribution of the two forms appears to be much the same as in Gothic and the West North Germanic languages. As the following examples from Edmondson and Lindau (1974) show, reflexives must be used in the positions of subject of an NP (35a), subject or nonsubject of a tenseless complement (35b,c), and object of a locative preposition (35d) when coreference with the matrix subject is intended.
4.0.We claimed earlier that the Germanic parent language must have been essentially like Gothic in that NPs and tenseless (but not tensed) clauses were transparent to disjoint reference and reflexive binding. The comparative evidence introduced in the last several sections leaves little room to doubt the correctness of this claim. The Gothic facts are duplicated in North Germanic by Old Icelandic and its descendents, by Swedish, and by some varieties of Danish. In West Germanic, the original reflexive/nonreflexive possessive distinction is reflected vestigially in Old English, and the transparency of infinitive complements is still found in earlier stages of German.
4.1.We have also seen that various Germanic languages have tended to move away from this original situation. In Biblical Gothic, nonreflexive pronouns occasionally appear in place of expected reflexives in both of the abovementioned positions, as well as nonsubject positions within participial relative clauses. In Skeireins Gothic, nonreflexive pronouns appear exclusively in place of expected reflexives in the latter. In German, participial relative constructions have also become opaque to reflexive binding and DR, and the reflexive/nonreflexive distinction has been wholly neutralized in the case of possessives. Moreover, during the history of German, the ability of reflexives in nonsubject position in infinitive complements to corefer with the higher subject has become increasingly restricted. In English, the reflexive possessive sin, the only surviving Germanic reflexive form, first became optional (indicating that NPs were no longer transparent to DR) and then was lost from the language. By the time the -self reflexives became established, both NPs and tenseless clause were opaque to reflexive binding. In Danish, too, reflexives have tended to be replaced by nonreflexives in both NPs and tenseless clauses where coreference with a subject external to those constructions is intended.
In some of the languages investigated, it appears that NPs and tenseless clauses were not affected by the observed changes at the same time? nonetheless, in those languages in which one environment was affected, the other was eventually affected as well. Unless we wish to dismiss this correspondence as accidental, we must assume that the changes in both are to be assigned some common cause, such as the implementation of principle (8).
The correspondence also argues against traditional accounts of the restriction on reflexives in infinitive complements, which attribute it to the fact that it eliminates the potential for ambiguity existing when both the higher subject and the lower subject are grammatically possible antecedents for a lower clause reflexive. No such explanation is available for the elimination of the reflexive possessive, since that change creates, rather than eliminates ambiguity.
It should also perhaps be mentioned that the changes discussed here do not seem to be connected with the general typological shift from SOV to SVO word order which has been claimed to be responsible for several other systematic changes in the syntax of the Germanic languages, since they have taken place in German, which preserves SOV order except in root clauses, but not in Icelandic, in which the word order shift has been entirely carried through.
4.2.In Chomsky’s Government Binding model, the class of anaphors is held to include not only phonologically realized forms such as reflexives, but certain phonologically null elements as well: specifically, the traces left by NP movement. Bounding conditions on raising to subject, passive, and other movements whose targets are argument positions are claimed to derive from the anaphoric nature of these traces. If this claim is true, then we should expect that the changes in conditions on anaphors in Germanic should also be reflected in radical changes in the operation of NP movement. I know of no evidence suggesting that this is the case. This apparent asymmetry raises interesting questions about the nature of the relationship between bounding conditions on movements and bounding conditions on binding of lexical anaphors.
NOTES
1.Although the translation dates from the fourth century, it is represented only in sixth century manuscripts. Citations in this paper are taken from the edition of Streitberg (1971), whose version of the Greek original I have also consulted.
2.In Gothic, unlike Greek and English, distinct reflexives exist only in the third person. In the first and second persons, the personal pronouns are used in both reflexive and nonreflexive functions. Robertson (1914:687 ff.) claimed that personal pronouns in all persons could also be used reflexively in New Testament Greek. However, virtually all of his examples involve pronouns which are either the objects of locative prepositions or possessive modifiers of NPs. My own examination of the distribution of personal pronouns and reflexive pronouns in the Greek text leads me to believe that he drew the wrong conclusion from these occurrences. It appears that personal pronouns were in fact uniformly interpreted as disjoint in reference with a tautoclausal subject EXCEPT when they occurred in one of these positions. An explanation of the opacity of these two constructions to disjoint reference assignments will be presented below. In other cases of intended coreference with a clause-mate subject, overtly reflexive pronouns are used almost exclusively. The only two counterexamples of which I am aware are Mat 6:19 and Joh 2:24. Robertson noted both of these, but proposed that the pronoun in the latter could in fact be an abbreviated form of the reflexive, indistinguishable from the corresponding personal pronoun because the manuscripts did not indicate accents and breathings.
3.In Gothic and Greek, unlike English, only subjects could serve as antecedents for reflexives. In these languages, therefore, it will not suffice to define potential antecedents solely in terms of c-command, as can be done in English.
4.In (2b), the Gothic version translates the Greek reflexive with the emphatic reflexive sik silban, to be discussed at greater length in note 7. The Gothic reflexive also appears in the position in question independent of the Greek, however, as in (a):
5.Correspondingly, nonreflexive pronouns in these positions appear to have been interpreted consistently as disjoint in reference with the matrix subject. The few exceptions will be discussed below. It should be noted, incidentally, that the evidence for obligatory disjoint reference assignment in Gothic—and all historical languages—is necessarily indirect. The only way to argue that a nonreflexive pronoun in some position HAD to be interpreted as disjoint in reference with the subject is to show that reflexives were uniformly used in that position whenever coreference was intended.
6.The claim that the infinitive complements in such examples do in fact have PRO subjects can be justified in the following way: As observed in note 3, reflexives in Gothic required subject antecedents. In examples like (a), however, there is no overt subject antecedent. The reflexive seinccmma is understood to corefer with the dative qenai.
Unless we wish to abandon the well-established generalization about subject control, we must posit a PRO subject, controlled by qenai, as the real antecedent of the reflexive. Similar examples motivate the PRO subjects posited for participial relative constructions, to be discussed below.
7.Whenever the Greek uses αντοu to corefer with the subject, the Gothic translates it as sein-. This is also the usual translation for the emphatic reflexive ιδιos ‘one’s own’. However, the emphatic reflexive possessive έαντοu is routinely translated as seins silbins ‘self’s own’, probably because it is bimorphemic as well as emphatic.
8.Compare also Joh 17:13, Col 2:15, Eph 2:16. In Greek, the reflexive is occasionally used in the position of object of a locative preposition to refer back to the subject, indicating that this position was accessible to reflexive binding. However, the fact that nonreflexive pronouns could also be used indicates that it was inaccessible to disjoint reference assignment (cf. also note 2). The facts seem to be the same for English. Many speakers accept the reflexive in (a), but for most, the nonreflexive is preferred.
(a) Hei put the glass beside [himi/?himselfi]
Thus, while not opaque to reflexive binding, locative prepositional phrases in English do appear to be opaque to DR. Note that only locative PPs are so restricted: “case” PPs are not:
(b) Hei bought it for [himselfi/*himi]
9.I assume that the linking of emphatic reflexive possessives like his own, ἰδιτs, ἑαντοῦ to their antecedents is not effected by core grammar binding processes, but by discourse-level processes. Even if this is not accepted, it remains the case that Gothic differs from English and Greek with respect to disjoint reference in all of these environments. It should also be noted that all anaphors do not behave alike in English. Thus, for example, unlike the reflexive, the reciprocal eaoh other occurs freely in possessive position.
10.For a more technical definition, see Chomsky (1979).
11.Chomsky proposes that infinitive complements with lexical subjects have undergone ͞S deletion, allowing the higher verb to govern the subject. When an untensed complement fails to undergo ͞S-deletion, its subject can only be PRO, since only PRO can occur in an ungoverned position. The subject of an infinitive is not governed by any element in its own clause, and ͞S-boundaries are absolute boundaries to governance from outside.
12.According to Chomsky, adnominal modifiers are not assigned genitive case by the head noun, since nouns do not assign case. Rather, case is inserted in these constructions by a special rule [NP NP-X̄] → [NP NP-x̄gen-X̄]
13.Except as noted in note 9; in Greek and English, emphatic reflexive possessives (ἑαντοῦ, ἰδιτ, his own) can be bound outside of their governing NPs. Another apparent exception is found in English constructions like (a), in which either a reflexive or a nonreflexive may be used to corefer with the matrix subject, across the boundary of an NP.
(a) Hei gave me[NP a picture [PP of himselfi/himi ]]
The admissibility of this type of alternation, however, may be accounted for by assuming an optional rebracketing convention of the form [Np ... [pp]]→ [np] [pp], whose application puts the reflexive in an accessible position. Such a restructuring rule has already been proposed to account for extractions involving constructions of this type.
14.By contrast, under Chomsky’s earlier models (e.g. Chomsky 1977, 1980), it would have to be assumed that Gothic differed from English and Greek with respect to more than a single condition. In those models, the differences in (7c) and (7d) could be attributed to the fact that English and Greek, but not Gothic, observe the Specified Subject Condition. However, this would not explain the difference reflected in (7e), where the SSC is not applicable. Instead, it would probably be necessary to account for this difference by claiming that reflexive binding and disjoint reference assignment in Greek and English, but not Gothic, were sensitive to the A-over-A condition. The single-condition account made available by the Government Binding model thus involves an empirical claim not made by the two-condition account required in the earlier models. In particular, it predicts that in a given language either the position of nonsubject of an infinitive complement and the position of subject of an NP should both be inaccessible to binding processes, or, if the condition is not observed, they should both be accessible. On the other hand, if two separate conditions are involved, then relaxation of one should have no necessary effect on the other. We will return to this question below.
15.However, in the Skeiveins, the other major Gothic text, we find the following, in which a reflexive possessive modifier in the subject of a tensed clause has the higher clause subject as its antecedent.
The reflexive here admits of a few possible explanations. It will be noted, for instance, that it is in a position prohibited according to principle (8a) but not prohibited by either the Nominative Island or the Specified Subject Condition of the earlier binding model presented in Chomsky (1980). Thus, the example could be construed as evidence in favor of the earlier model. Alternatively, this may be taken as a logophoric use of the reflexive, of the type discussed in note 29. In any case, since the example is unique, and since it occurs in a text of uncertain provenance, it doesn’t appear possible to determine what significance is to be attached to it.
16.In fact, I have argued elsewhere (Harbert 1980) that such a principle should be included in the inventory of binding conditions in universal grammar. The condition in (8), of course, would duplicate the effects of a tensed-clause principle of this type. Therefore, a model of universal grammar containing both would be less than optimally elegant. However, it is justified by the existence of numerous languages in which tensed clauses remain opaque to binding even though governing categories of other types are transparent.
17.Other possible candidates, listed by Edmondson (1978:643) are Gã, Japanese, and Okinawan. Examples from Icelandic, Swedish, Latin, and Russian will be given below. Hindi examples are presented in Harbert (1980) and the references given there.
18.Faltz (1977:158) observes an asymmetry in Russian between tensed and untensed clauses similar to that found in Gothic. The latter, but not the former are transparent to reflexive binding.
19.The question arises, of course, concerning how Gothic (and Germanic) acquired a marked distribution of reflexives in the first place. An interesting suggestion is advanced by Faltz (1977:250 ff.), who proposes that the Germanic reflexives and their cognates in other Indo-European languages originated as reportative subordinate reflexives, which were used in indirect discourse and which were restricted to occurring in clauses not containing their antecedents. Only subsequently did they develop into true reflexives, subject to conditions on binding. The reader is referred to Faltz1 discussion. Compare also note 29.
20.The damaged and fragmentary manuscript in which the Skeireins has been transmitted was painstakingly edited by Bennett (1960), whose edition I cite here.
21.Once again, this is the sole example of its kind in the corpus.
22.As in Gothic, this applies only to third person pronouns. In the first and second persons, for which no distinct reflexive forms existed, the personal pronoun was used in both reflexive and nonreflexive functions. Moreover, the dative third person reflexive was lost prehistorically in German, and the corresponding personal pronoun was used in Old High German in reflexive as well as nonre-flexive functions.
Only later in the history of the language was a new dative reflexive introduced, formally identical with the accusative. Of course, cases like those pose an interesting problem for the theory: How is disjoint reference assignment to be constrained from applying to just those pronouns for which no corresponding reflexive exists? For an interesting discussion of this problem, see Thrainsson (1976).
23.The Old High German Tatian translation is preserved in a 9th century manuscript. Citations here come from the edition of Sievers (1892), whose Latin parallel text I have also consulted.
24.Corresponding to the underlined pronoun in this example, the Latin model has the reflexive se, in what is probably to be considered a logophoric use (cf. note 29).
25.However, elsewhere the translator failed to imitate a Latin reflexive in a comparable construction:
26.Locative prepositional phrases in German appear to have been transparent to both DR and reflexive binding at all stages.
27.These facts were reported by Reis (1973). A reflexive in a position other than that of prepositional object can be bound a-cross the accusative subject of the complement if that subject is the dummy neuter pronoun es :
28.The same facts would also argue against a hypothesis holding that binding and historical changes in binding are governed by a hierarchy of individual constructions ranked according to relative accessibility, rather than by a single condition referring to a generalized configuration of which these individual constructions are special cases. Edmondson (1978:644) has suggested the possible existence of such a hierarchy. If we assume that it does exist as part of universal grammar, we could propose that nonsubject positions in infinitive complements rank higher on it (and are therefore more accessible to binding) than possessive modifiers. The cutoff point for binding in a given language would either fall higher on the hierarchy than either position, as in English, or lower than either position, as in Gothic, or between the two positions. In the latter case, represented by Old High German, binding would treat the two constructions differently. We could account for the apparent uni- directionality of changes in binding by stipulating that the position of the cutoff point, if it changes at all, must move UP the hierarchy. Such an account would have an apparent advantage over the single-condition account, in that it would predict the possibility of situations like the one found in Old High German. This is accomplished, however, only at the cost of introducing considerable apparatus, and it is not clear that it is even a desirable result, in view of the apparent infrequency of the Old High German situation relative to either that represented in English or that represented in Gothic.
29.It is a well-known fact about Old Icelandic and several other North Germanic languages that reflexives can be used in tensed complements under certain circumstances to refer back to the subject of the higher clause. These are traditionally referred to as indirect reflexives:
The most extensive statement of the conditions under which this can happen is to be found in Thrainsson’s discussion of Modern Icelandic (Thrainsson 1976). He demonstrates that these conditions must be formulated in terms of semantics or discourse, rather than in syntactic terms; in particular, reflexives of the type in question occur predominantly (although not exclusively) in indirect discourse. Clements (1977) notes a similar use of the reflexive in other languages and characterizes it as a discourse-level device for designating the person whose speech or thoughts are being reported. He uses the term LOGOPHORIC for this function, and suggests that it is accidental that the logophoric function is expressed by means of reflexive forms in these languages. In other languages, it is assumed by personal pronouns, while some Niger-Congo languages have distinct logophoric pronominal forms. Faltz (1977:253 ff.) also notes the parallel between the logophoric (“reportative”) forms in the Niger-Congo languages and indirect reflexives of the type illustrated in (a)-(e). The arguments presented by Thrainsson and Clements suggest strongly that the linking of these logophoric reflexives to their antecedents should be considered to be effected by processes at the level of discourse, rather than by core grammar binding. Cf. also Köster (1978:586 ff.).
30.Of course, in Old Icelandic and the other languages discussed here, a reflexive in an infinitive complement can corefer with the subject of the complement as well as the matrix subject. Examples have been omitted here due to considerations of space. The reflexive possessive could also be used in Old Icelandic, under conditions of emphasis, to refer to a nonsubject antecedent. This is true in Modern Icelandic, Faroese, and Danish as well.
31.Falk and Torp attribute the adoption of deres as the general plural possessive to the influence of the German ihr. This explanation is insufficient for a few reasons. First, the introduction of deres seems to have coincided with, and is clearly to be related to, the spread of nonreflexive possessives to orginally reflexive contexts in the singular—a change which cannot be explained as the result of German influence. (If anything, the model of German sein would have favored the extension of the reflexive sin to nonreflexive contexts.) Second, a possessive adjective would constitute a rather odd borrowing unless it had some particular system internal motivation, such as might be provided by increased opacity on the part of NPs.
32.Or perhaps both DR and reflexive binding were blocked, and these reflexives were associated with their antecedents by the type of discourse-level linking discussed in note 29.
REFERENCES
Bartsch, K. (ed.) 1870. Der Nibelunge Nôt. Leipzig: Brockhaus.
Beito, Olav T. 1970. Nynorsk Grammatik. Oslo: Det Norske Samlaget.
Bennett, W. H. 1960. The Gothic Commentary on the Gospel of John. New York: Modern Language Association. [Reprinted in 1966.]
Chomsky, N. 1977. On WH-Movement. Formal syntax, ed. by P. Culicover, T. Wascow, and A. Akmajian, 71-132. New York: Academic Press.
——. 1981. Principles and parameters in syntactic theory. Explanations in linguistics, ed. by N. Hornstein and D. Lightfoot, 32-75. New York: Longman.
——. 1980. On binding. Linguistic Inquiry 11:1-46.
Clements, G. N. 1975. The logophoric pronoun in Ewe: Its role in discourse. Journal of West African Linguistics 10:141-78.
Edmondson, J. A. 1978. Ergative languages, accessibility hierarchies governing reflexives and questions of formal analysis. Valence semantic case and grammatical relations, ed. by W. Abraham, 633-60. Amsterdam: John Benjamins.
——, and M. Lindau. 1974. A study of Swedish pronominalization. UCLA Papers in Syntax 5:18-48.
Falk, H., and A. Torp. 1900. Dansk-Norskens Syntax. Kristiania: H. Aschehoug.
Faltz, L. 1977. Reflexivization: A study in universal grammar. Dissertation, University of California, Berkeley.
Harbert, W. 1980. In defense of tense. Cornell Working Papers in Linguistics 1:66-92.
Iversen, R. 1918. Syntaksen i Tromsø Bymaal. Kristiania:Bymaals-Lagets Forlag.
Jónsson, Finnur (ed.) 1908. Brennu-Njálssaga. Halle: Max Niemeyer.
Keller, A. V. (ed.) 1957. Konrads von Wurzburg trojaner Krieg. Publicationen des litterarischen Vereins in Stuttgart, vol. 44.
Klaeber, F. 1950. Beowulf and the fight at Finnsburg. 3rd ed. Lexington, MA: D. C. Heath and Co.
Koster, J. 1978. Conditions, empty nodes, and markedness. Linguistic Inquiry 9:551-93.
Leitzmann, A. (1963). Wolfram von Eschenbach. 6th ed. Tübingen: Max Niemeyer.
Lockwood, W. B. 1964. An introduction to modern Faroese. Copenhagen: Munksgaard.
Novum Testamentum Latine. 1971. 11th ed. Stuttgart: Württem-burgische Bibelanstalt.
Nygaard, M. 1905. Norrtfn Syntax. Kristiania: H. Aschehoug.
Ranke, F. 1967. Altnordisches Elementarbuch. (3rd ed. revised by D. Hoffmann.) Berlin: Walter de Gruyter.
Reis, M. 1973. Is there a rule of Subject-To-Object Raising in German? CLS 9:519-29.
Robertson, A. T. 1914. A grammar of the Greek New Testament in the light of historical research. New York: George H. Doran.
Sievers, E. (ed.) 1892. Tatian. Lateinisch und altdeutsch. 2nd ed. Paderborn: F. Schöningh.
Spore, Palle. 1965. La Langue Danoise. Copenhagen: Akademisk Forlag.
Streitberg, W. (ed.) 1971. Die gotische Bibel. 6th ed. Darmstadt: Wissenschaftliche Buchgesellschaft.
Thrainsson, H. 1976a. Some arguments against the interpretive theory of pronouns and reflexives. Harvard Studies in Syntax and Semantics 2:573-624.
——. 1976b. Reflexives and subjunctives in Icelandic. Papers from the Sixth Annual Meeting, North Eastern Linguistics Society, 225-40.
Timberlake, A. 1980. Reference conditions on Russian reflexivization. Language 56:777-96.
We use cookies to analyze our traffic. Please decide if you are willing to accept cookies from our website. You can change this setting anytime in Privacy Settings.