While Pintzuk's analysis of Old English V2 yields an economical description of many relevant facts of the language, it faces two significant problems. First, it is not clear how to make the analysis consistent with the fact that, CP-recursion environments apart, Old English texts do not freely exhibit subordinate clauses with non-subject topics and V2 word order. Kemenade (this volume) states that V2 order with non-subjects in first position occurs only in limited types of subordinate clause in the texts, and Pintzuk's observations (personal communication) confirm this finding. Pintzuk's analysis, however, does not predict this limitation. Second, the special clitic movement rule needed by Pintzuk to account for the placement of pronouns between topic and verb in V2 clauses has no counterpart elsewhere among the Germanic languages and does not have clear theoretical justification. In this section, therefore, we will propose a modification of Pintzuk's approach which preserves its essential claim -- that Old English is an IP-V2 language -- while mitigating these two difficulties. Suppose that, while the tensed verb in an Old English V2 sentence moves to I-zero, the topic moves not to Spec,IP but to Spec,CP. In that case, the clitic pronoun can move straightforwardly to the CP/IP boundary and the correct word order will result, without Pintzuk's special clitic inversion rule. The result of changing the landing site of the topic is that V2 in Old English seems to become a hybrid between the CP-V2 and the IP-V2 types. The tensed verb moves as in an IP-V2 language, while the topic moves as in a CP-V2 language. We will see below, however, that the other IP-V2 languages are more like Old English they first appear to be and that the analysis of the IP-V2 phenomenon itself must be changed. Once this change is made, Old English again falls together with the other IP-V2 languages.
Clearly, our proposal permits a standard treatment of Old English pronouns as Germanic-type clitics, which Pintzuk's account does not;[10] but to make the proposal viable, we must explain why both C-zero and Spec,IP can, and indeed must, remain empty in Old English main clauses. If either contained phonetically overt material, Old English would not exhibit V2 word order in sentences with non-subject topics. We begin with the question of C-zero. We have already seen (section 3.3) that Old English reserves the C-zero position for verbs with special morphosyntactic features, suggesting that ordinary indicative verbs do not belong in that position, at least on the surface. In line with recent proposals regarding economy (Chomsky 1993, 1995), we can say that the ordinary indicative tensed verb in Old English carries only a weak feature driving its movement to C-zero and so moves there only at LF. In questions and the other environments discussed above, on the other hand, the feature driving movement to C-zero will be strong, making movement visible on the surface. In either case, movement to C-zero will occur by LF; and, therefore, the topic will have to be in Spec,CP in order to be properly licensed. Cross-linguistically, topicalized constituents are always the leftmost elements of their clauses; and it seems reasonable to assume that they hold this position because they are the surface ``subjects'' of the clause's topmost predication level (Heycock 1994). Given the phrase structure we are using, this requirement implies that topics in matrix clauses must occupy Spec,CP, making our requirement on topics analogous to the wh-criterion of Rizzi 1990. In subordinate clauses with topicalization, this leftmost position seems to vary between Spec,IP and Spec,CP depending on the language, dialect and sentence type involved (see immediately below, also section 6).
We move on now to explaining why Spec,IP is empty, a more difficult question. The simplest treatment would be to say that the features driving movement of the subject to Spec,IP, which are strong in modern English, are weak in Old English. By economy then, this feature assignment would force movement of the subject to occur only at LF, leaving Spec,IP empty on the surface. However, while this solution is adequate for simple sentences like those in (3) above, it fails when more cases are considered. To start with, it predicts that in INFL-medial subordinate clauses the verb should come first, since Spec,IP will be empty; but the fact is, of course, that Old English subordinate clauses are predominantly subject-initial, not verb-first. There is no reason to question the standard line that these subjects are in Spec,IP. A second and extremely important consideration is that, as described by Kemenade (this volume), Old English subordinate clauses do sometimes exhibit V2 word order with a topicalized non-subject outside CP-recursion environments. Such subordinate V2 order occurs in clauses where the nominative subject is absent or is licensed to appear in a position other than Spec,IP, as in passive sentences or in sentences with experiencer dative ``subjects.'' In such cases there is often no nominative NP, and agreement on the verb is the default third person singular. Otherwise, the nominative NP which agrees with the verb remains inside VP, while some other constituent, often but not always an experiencer dative, appears in Spec,IP. These possibilities are illustrated in the following examples from Kemenade (this volume):
Significantly, it turns out that in Icelandic and Yiddish, both IP-V2 languages, [11] examples of subordinate clause topicalization outside CP-recursion contexts are similarly limited to (or are more acceptable in) contexts where the subject is missing or appears in a VP-internal or a postposed position (Magnússon 1990, Sigurdhsson 1990, Thráinsson 1994, Beatrice Santorini (personal communication)). The facts thus suggest very strongly that in the IP-V2 type of language, Spec,IP ordinarily hosts a subject and can only be a topic position in sentences where the subject is either absent or licensed to surface elsewhere. This marked use of Spec,IP in subordinate clauses must be made to harmonize with the emptiness of the position in main clauses.
It is difficult under current assumptions to account for the occurrence of topics in Spec,IP (or Spec,AGR-S) because that is the locus where Spec-head agreement between the finite verb and the subject is checked. Even if the subject in a certain sentence type does not appear there on the surface, checking theory of a Minimalist sort requires an expletive in Spec,IP, which forms a chain with the subject and is replaced by it at LF. The expletive may be phonetically null, but even so it will prevent a fronted topic from landing in Spec,IP. Thus, the presence of the empty expletive forces us to say that the topic lands in a higher specifier than Spec,IP, somewhere between I-zero (or AGR-S) and C-zero, as in Cardinaletti & Roberts 1991. If we follow this line, however, we will have no simple explanation for the complementarity that Kemenade has found in subordinate clauses between the appearance of an overt subject in Spec,IP and the possibility of a fronted topic. To avoid this consequence, we propose that the checking mechanism for subject-verb agreement in sentences with empty expletives be changed so as to free up Spec,IP as a landing site for topics. What we have in mind is similar to proposals that have been made many times to the effect that it is the agreement morpheme on the tensed verb rather than an expletive in Spec,IP that serves as the binder of a postposed or VP-internal subject (cf. den Besten 1985).
Such a subject forms a chain not with an empty expletive but directly with the agreement morpheme on the finite verb, with which it is coindexed. The index on the agreement morpheme can be taken to reflect incorporation of the empty expletive into the feature complex of I-zero. Such incorporation satisfies (that is, checks off) the agreement features in I-zero and also the nominative case feature. The chain between agreement and the subject thereby case-licenses the subject, rendering expletive replacement unnecessary. If we assume that incorporation frees up Spec,IP or that it occurs without movement through Spec,IP, topics will be able to move to that position, as desired. The proposed treatment applies in the obvious way to impersonal sentences with empty expletives, where no nominative case noun phrase occurs. In these sentences, there is, of course, no chain; but the empty expletive will check off agreement and nominative case by incorporation. The existence of such sentences (illustrated by the Old English and the German examples of (11) below) demonstrates, in any case, that expletive replacement cannot be the general mechanism by which the existence of expletives is reconciled with the Principle of Full Interpretation (Chomsky 1986 b) and suggests that we may be better off without the device.[12]
Our analysis of Spec,IP in subordinate clauses now allows us to understand why the position is empty in main clauses in Old English. V2 languages are universally defined by a requirement that topics be in a Spec-head relationship with the finite verb. While much discussed, this striking requirement has never been reduced to anything more fundamental. We assume that Old English, as a V2 language, is subject, like the others, to this ``V2 constraint.'' However, under our analysis the V2 constraint cannot apply directly to the topic and verb themselves in Old English, because these are not in a Spec-head relationship on the surface. The topic is necessarily in Spec,CP because that is the specifier of the highest projection of its clause; and the verb, due to its feature content, raises only as far as I-zero. To establish the required relationship, therefore, the topic must move through Spec,IP on its way to Spec,CP, leaving its trace to fulfill the V2 constraint. That traces may serve this function is shown by the following German and Yiddish examples of extraction out of complementizer-less subordinate clauses, where the verb in C-zero is in a Spec-head relationship with the trace of the extracted topic:[13]
Because Spec,IP is the intermediate landing site for the topic in an Old English main clause, it can never contain anything but the trace of the topic and so will always appear empty.
If the topic is to move through Spec,IP, that position must not be needed to establish a Spec-head relationship between the subject and the agreement morpheme. We must assume, therefore, that, as in subordinate clauses in Old English, matrix sentences with non-subject topics contain incorporated empty expletives to check off the agreement features of I-zero and to chain-license the subject in a lower position (Spec,VP or, assuming a split INFL, Spec,TP). Since the licensing of topics in main and subordinate clauses is identical, we are now without a simple syntactic explanation for the greatly reduced range of topicalizations in subordinate as opposed to matrix clauses; but this difference may, indeed, not be a syntactic fact. The difference between main and subordinate clauses seems to reflect discourse-based information-structure considerations. In matrix clauses, topicalization is often highly favored, or even required, by the discourse context; and in order for the needed topicalized sentences to surface in Old English, empty expletive chains must used across a wide range of cases. In (non-CP-recursion) subordinate clauses, by contrast, topicalization has very weak discourse motivation and so expletive chains are used only where information-structure considerations favor them -- that is, in the classical environments in which subjects prefer not to appear in Spec,IP. The use of the expletive chain then frees up Spec,IP for a non-subject to appear as topic even if the topicalization is only weakly motivated. The correctness of attributing the differences in the range and frequency of main and subordinate clause topicalization to discourse considerations cannot be demonstrated in the current state of our knowledge of discourse structure; but it is consistent with the facts as we know them, including the uncertainty of native speaker judgments on the acceptability of subordinate clause topicalization in living IP-V2 languages like Icelandic and Yiddish. In both of these languages, speakers differ on whether the full range of topicalization is available in subordinate clauses; and text corpora show only cases like those found in Old English, for which the expletive chain analysis is plausible (Eiríkur Rögnvaldsson, personal communication; Beatrice Santorini, personal communication). The variability in judgments and the difference between what is judged acceptable and what actually occurs in connected discourse point to the plausibility of a discourse constraint (and speakers' differential sensitivity to discourse factors in giving judgments) as the source of the main/subordinate difference in the range of topicalization.
To summarize a complex discussion, we have provided in our modified version of Pintzuk's analysis a treatment of Old English V2 with the following virtues: